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 void ctl_shutdown(void);
 428 static int ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td);
 429 static int ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td);
 430 static void ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio);
 431 static void ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
 432                               struct ctl_ooa *ooa_hdr,
 433                               struct ctl_ooa_entry *kern_entries);
 434 static int ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
 435                      struct thread *td);
 436 static int ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
 437                          struct ctl_be_lun *be_lun);
 438 static int ctl_free_lun(struct ctl_lun *lun);
 439 static void ctl_create_lun(struct ctl_be_lun *be_lun);
 440
 441 static int ctl_do_mode_select(union ctl_io *io);
 442 static int ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun,
 443                            uint64_t res_key, uint64_t sa_res_key,
 444                            uint8_t type, uint32_t residx,
 445                            struct ctl_scsiio *ctsio,
 446                            struct scsi_per_res_out *cdb,
 447                            struct scsi_per_res_out_parms* param);
 448 static void ctl_pro_preempt_other(struct ctl_lun *lun,
 449                                   union ctl_ha_msg *msg);
 450 static void ctl_hndl_per_res_out_on_other_sc(union ctl_io *io);
 451 static int ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len);
 452 static int ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len);
 453 static int ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len);
 454 static int ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len);
 455 static int ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len);
 456 static int ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio,
 457                                          int alloc_len);
 458 static int ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio,
 459                                          int alloc_len);
 460 static int ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len);
 461 static int ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len);
 462 static int ctl_inquiry_evpd(struct ctl_scsiio *ctsio);
 463 static int ctl_inquiry_std(struct ctl_scsiio *ctsio);
 464 static int ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len);
 465 static ctl_action ctl_extent_check(union ctl_io *io1, union ctl_io *io2,
 466     bool seq);
 467 static ctl_action ctl_extent_check_seq(union ctl_io *io1, union ctl_io *io2);
 468 static ctl_action ctl_check_for_blockage(struct ctl_lun *lun,
 469     union ctl_io *pending_io, union ctl_io *ooa_io);
 470 static ctl_action ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
 471                                 union ctl_io *starting_io);
 472 static int ctl_check_blocked(struct ctl_lun *lun);
 473 static int ctl_scsiio_lun_check(struct ctl_lun *lun,
 474                                 const struct ctl_cmd_entry *entry,
 475                                 struct ctl_scsiio *ctsio);
 476 static void ctl_failover_lun(union ctl_io *io);
 477 static int ctl_scsiio_precheck(struct ctl_softc *ctl_softc,
 478                                struct ctl_scsiio *ctsio);
 479 static int ctl_scsiio(struct ctl_scsiio *ctsio);
 480
 481 static int ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io);
 482 static int ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
 483                             ctl_ua_type ua_type);
 484 static int ctl_do_lun_reset(struct ctl_lun *lun, union ctl_io *io,
 485                          ctl_ua_type ua_type);
 486 static int ctl_lun_reset(struct ctl_softc *ctl_softc, union ctl_io *io);
 487 static int ctl_abort_task(union ctl_io *io);
 488 static int ctl_abort_task_set(union ctl_io *io);
 489 static int ctl_query_task(union ctl_io *io, int task_set);
 490 static int ctl_i_t_nexus_reset(union ctl_io *io);
 491 static int ctl_query_async_event(union ctl_io *io);
 492 static void ctl_run_task(union ctl_io *io);
 493 #ifdef CTL_IO_DELAY
 494 static void ctl_datamove_timer_wakeup(void *arg);
 495 static void ctl_done_timer_wakeup(void *arg);
 496 #endif /* CTL_IO_DELAY */
 497
 498 static void ctl_send_datamove_done(union ctl_io *io, int have_lock);
 499 static void ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq);
 500 static int ctl_datamove_remote_dm_write_cb(union ctl_io *io);
 501 static void ctl_datamove_remote_write(union ctl_io *io);
 502 static int ctl_datamove_remote_dm_read_cb(union ctl_io *io);
 503 static void ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq);
 504 static int ctl_datamove_remote_sgl_setup(union ctl_io *io);
 505 static int ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
 506                                     ctl_ha_dt_cb callback);
 507 static void ctl_datamove_remote_read(union ctl_io *io);
 508 static void ctl_datamove_remote(union ctl_io *io);
 509 static void ctl_process_done(union ctl_io *io);
 510 static void ctl_lun_thread(void *arg);
 511 static void ctl_thresh_thread(void *arg);
 512 static void ctl_work_thread(void *arg);
 513 static void ctl_enqueue_incoming(union ctl_io *io);
 514 static void ctl_enqueue_rtr(union ctl_io *io);
 515 static void ctl_enqueue_done(union ctl_io *io);
 516 static void ctl_enqueue_isc(union ctl_io *io);
 517 static const struct ctl_cmd_entry *
 518     ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa);
 519 static const struct ctl_cmd_entry *
 520     ctl_validate_command(struct ctl_scsiio *ctsio);
 521 static int ctl_cmd_applicable(uint8_t lun_type,
 522     const struct ctl_cmd_entry *entry);
 523
 524 static uint64_t ctl_get_prkey(struct ctl_lun *lun, uint32_t residx);
 525 static void ctl_clr_prkey(struct ctl_lun *lun, uint32_t residx);
 526 static void ctl_alloc_prkey(struct ctl_lun *lun, uint32_t residx);
 527 static void ctl_set_prkey(struct ctl_lun *lun, uint32_t residx, uint64_t key);
 528
 529 /*
 530  * Load the serialization table.  This isn't very pretty, but is probably
 531  * the easiest way to do it.
 532  */
 533 #include "ctl_ser_table.c"
 534
 535 /*
 536  * We only need to define open, close and ioctl routines for this driver.
 537  */
 538 static struct cdevsw ctl_cdevsw = {
 539         .d_version =    D_VERSION,
 540         .d_flags =      0,
 541         .d_open =       ctl_open,
 542         .d_close =      ctl_close,
 543         .d_ioctl =      ctl_ioctl,
 544         .d_name =       "ctl",
 545 };
 546
 547
 548 MALLOC_DEFINE(M_CTL, "ctlmem", "Memory used for CTL");
 549
 550 static int ctl_module_event_handler(module_t, int /*modeventtype_t*/, void *);
 551
 552 static moduledata_t ctl_moduledata = {
 553         "ctl",
 554         ctl_module_event_handler,
 555         NULL
 556 };
 557
 558 DECLARE_MODULE(ctl, ctl_moduledata, SI_SUB_CONFIGURE, SI_ORDER_THIRD);
 559 MODULE_VERSION(ctl, 1);
 560
 561 static struct ctl_frontend ha_frontend =
 562 {
 563         .name = "ha",
 564 };
 565
 566 static void
 567 ctl_ha_datamove(union ctl_io *io)
 568 {
 569         struct ctl_lun *lun = CTL_LUN(io);
 570         struct ctl_sg_entry *sgl;
 571         union ctl_ha_msg msg;
 572         uint32_t sg_entries_sent;
 573         int do_sg_copy, i, j;
 574
 575         memset(&msg.dt, 0, sizeof(msg.dt));
 576         msg.hdr.msg_type = CTL_MSG_DATAMOVE;
 577         msg.hdr.original_sc = io->io_hdr.original_sc;
 578         msg.hdr.serializing_sc = io;
 579         msg.hdr.nexus = io->io_hdr.nexus;
 580         msg.hdr.status = io->io_hdr.status;
 581         msg.dt.flags = io->io_hdr.flags;
 582
 583         /*
 584          * We convert everything into a S/G list here.  We can't
 585          * pass by reference, only by value between controllers.
 586          * So we can't pass a pointer to the S/G list, only as many
 587          * S/G entries as we can fit in here.  If it's possible for
 588          * us to get more than CTL_HA_MAX_SG_ENTRIES S/G entries,
 589          * then we need to break this up into multiple transfers.
 590          */
 591         if (io->scsiio.kern_sg_entries == 0) {
 592                 msg.dt.kern_sg_entries = 1;
 593 #if 0
 594                 if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
 595                         msg.dt.sg_list[0].addr = io->scsiio.kern_data_ptr;
 596                 } else {
 597                         /* XXX KDM use busdma here! */
 598                         msg.dt.sg_list[0].addr =
 599                             (void *)vtophys(io->scsiio.kern_data_ptr);
 600                 }
 601 #else
 602                 KASSERT((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0,
 603                     ("HA does not support BUS_ADDR"));
 604                 msg.dt.sg_list[0].addr = io->scsiio.kern_data_ptr;
 605 #endif
 606                 msg.dt.sg_list[0].len = io->scsiio.kern_data_len;
 607                 do_sg_copy = 0;
 608         } else {
 609                 msg.dt.kern_sg_entries = io->scsiio.kern_sg_entries;
 610                 do_sg_copy = 1;
 611         }
 612
 613         msg.dt.kern_data_len = io->scsiio.kern_data_len;
 614         msg.dt.kern_total_len = io->scsiio.kern_total_len;
 615         msg.dt.kern_data_resid = io->scsiio.kern_data_resid;
 616         msg.dt.kern_rel_offset = io->scsiio.kern_rel_offset;
 617         msg.dt.sg_sequence = 0;
 618
 619         /*
 620          * Loop until we've sent all of the S/G entries.  On the
 621          * other end, we'll recompose these S/G entries into one
 622          * contiguous list before processing.
 623          */
 624         for (sg_entries_sent = 0; sg_entries_sent < msg.dt.kern_sg_entries;
 625             msg.dt.sg_sequence++) {
 626                 msg.dt.cur_sg_entries = MIN((sizeof(msg.dt.sg_list) /
 627                     sizeof(msg.dt.sg_list[0])),
 628                     msg.dt.kern_sg_entries - sg_entries_sent);
 629                 if (do_sg_copy != 0) {
 630                         sgl = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr;
 631                         for (i = sg_entries_sent, j = 0;
 632                              i < msg.dt.cur_sg_entries; i++, j++) {
 633 #if 0
 634                                 if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
 635                                         msg.dt.sg_list[j].addr = sgl[i].addr;
 636                                 } else {
 637                                         /* XXX KDM use busdma here! */
 638                                         msg.dt.sg_list[j].addr =
 639                                             (void *)vtophys(sgl[i].addr);
 640                                 }
 641 #else
 642                                 KASSERT((io->io_hdr.flags &
 643                                     CTL_FLAG_BUS_ADDR) == 0,
 644                                     ("HA does not support BUS_ADDR"));
 645                                 msg.dt.sg_list[j].addr = sgl[i].addr;
 646 #endif
 647                                 msg.dt.sg_list[j].len = sgl[i].len;
 648                         }
 649                 }
 650
 651                 sg_entries_sent += msg.dt.cur_sg_entries;
 652                 msg.dt.sg_last = (sg_entries_sent >= msg.dt.kern_sg_entries);
 653                 if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
 654                     sizeof(msg.dt) - sizeof(msg.dt.sg_list) +
 655                     sizeof(struct ctl_sg_entry) * msg.dt.cur_sg_entries,
 656                     M_WAITOK) > CTL_HA_STATUS_SUCCESS) {
 657                         io->io_hdr.port_status = 31341;
 658                         io->scsiio.be_move_done(io);
 659                         return;
 660                 }
 661                 msg.dt.sent_sg_entries = sg_entries_sent;
 662         }
 663
 664         /*
 665          * Officially handover the request from us to peer.
 666          * If failover has just happened, then we must return error.
 667          * If failover happen just after, then it is not our problem.
 668          */
 669         if (lun)
 670                 mtx_lock(&lun->lun_lock);
 671         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
 672                 if (lun)
 673                         mtx_unlock(&lun->lun_lock);
 674                 io->io_hdr.port_status = 31342;
 675                 io->scsiio.be_move_done(io);
 676                 return;
 677         }
 678         io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
 679         io->io_hdr.flags |= CTL_FLAG_DMA_INPROG;
 680         if (lun)
 681                 mtx_unlock(&lun->lun_lock);
 682 }
 683
 684 static void
 685 ctl_ha_done(union ctl_io *io)
 686 {
 687         union ctl_ha_msg msg;
 688
 689         if (io->io_hdr.io_type == CTL_IO_SCSI) {
 690                 memset(&msg, 0, sizeof(msg));
 691                 msg.hdr.msg_type = CTL_MSG_FINISH_IO;
 692                 msg.hdr.original_sc = io->io_hdr.original_sc;
 693                 msg.hdr.nexus = io->io_hdr.nexus;
 694                 msg.hdr.status = io->io_hdr.status;
 695                 msg.scsi.scsi_status = io->scsiio.scsi_status;
 696                 msg.scsi.tag_num = io->scsiio.tag_num;
 697                 msg.scsi.tag_type = io->scsiio.tag_type;
 698                 msg.scsi.sense_len = io->scsiio.sense_len;
 699                 msg.scsi.sense_residual = io->scsiio.sense_residual;
 700                 msg.scsi.residual = io->scsiio.residual;
 701                 memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
 702                     io->scsiio.sense_len);
 703                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
 704                     sizeof(msg.scsi) - sizeof(msg.scsi.sense_data) +
 705                     msg.scsi.sense_len, M_WAITOK);
 706         }
 707         ctl_free_io(io);
 708 }
 709
 710 static void
 711 ctl_isc_handler_finish_xfer(struct ctl_softc *ctl_softc,
 712                             union ctl_ha_msg *msg_info)
 713 {
 714         struct ctl_scsiio *ctsio;
 715
 716         if (msg_info->hdr.original_sc == NULL) {
 717                 printf("%s: original_sc == NULL!\n", __func__);
 718                 /* XXX KDM now what? */
 719                 return;
 720         }
 721
 722         ctsio = &msg_info->hdr.original_sc->scsiio;
 723         ctsio->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
 724         ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
 725         ctsio->io_hdr.status = msg_info->hdr.status;
 726         ctsio->scsi_status = msg_info->scsi.scsi_status;
 727         ctsio->sense_len = msg_info->scsi.sense_len;
 728         ctsio->sense_residual = msg_info->scsi.sense_residual;
 729         ctsio->residual = msg_info->scsi.residual;
 730         memcpy(&ctsio->sense_data, &msg_info->scsi.sense_data,
 731                msg_info->scsi.sense_len);
 732         ctl_enqueue_isc((union ctl_io *)ctsio);
 733 }
 734
 735 static void
 736 ctl_isc_handler_finish_ser_only(struct ctl_softc *ctl_softc,
 737                                 union ctl_ha_msg *msg_info)
 738 {
 739         struct ctl_scsiio *ctsio;
 740
 741         if (msg_info->hdr.serializing_sc == NULL) {
 742                 printf("%s: serializing_sc == NULL!\n", __func__);
 743                 /* XXX KDM now what? */
 744                 return;
 745         }
 746
 747         ctsio = &msg_info->hdr.serializing_sc->scsiio;
 748         ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
 749         ctl_enqueue_isc((union ctl_io *)ctsio);
 750 }
 751
 752 void
 753 ctl_isc_announce_lun(struct ctl_lun *lun)
 754 {
 755         struct ctl_softc *softc = lun->ctl_softc;
 756         union ctl_ha_msg *msg;
 757         struct ctl_ha_msg_lun_pr_key pr_key;
 758         int i, k;
 759
 760         if (softc->ha_link != CTL_HA_LINK_ONLINE)
 761                 return;
 762         mtx_lock(&lun->lun_lock);
 763         i = sizeof(msg->lun);
 764         if (lun->lun_devid)
 765                 i += lun->lun_devid->len;
 766         i += sizeof(pr_key) * lun->pr_key_count;
 767 alloc:
 768         mtx_unlock(&lun->lun_lock);
 769         msg = malloc(i, M_CTL, M_WAITOK);
 770         mtx_lock(&lun->lun_lock);
 771         k = sizeof(msg->lun);
 772         if (lun->lun_devid)
 773                 k += lun->lun_devid->len;
 774         k += sizeof(pr_key) * lun->pr_key_count;
 775         if (i < k) {
 776                 free(msg, M_CTL);
 777                 i = k;
 778                 goto alloc;
 779         }
 780         bzero(&msg->lun, sizeof(msg->lun));
 781         msg->hdr.msg_type = CTL_MSG_LUN_SYNC;
 782         msg->hdr.nexus.targ_lun = lun->lun;
 783         msg->hdr.nexus.targ_mapped_lun = lun->lun;
 784         msg->lun.flags = lun->flags;
 785         msg->lun.pr_generation = lun->pr_generation;
 786         msg->lun.pr_res_idx = lun->pr_res_idx;
 787         msg->lun.pr_res_type = lun->pr_res_type;
 788         msg->lun.pr_key_count = lun->pr_key_count;
 789         i = 0;
 790         if (lun->lun_devid) {
 791                 msg->lun.lun_devid_len = lun->lun_devid->len;
 792                 memcpy(&msg->lun.data[i], lun->lun_devid->data,
 793                     msg->lun.lun_devid_len);
 794                 i += msg->lun.lun_devid_len;
 795         }
 796         for (k = 0; k < CTL_MAX_INITIATORS; k++) {
 797                 if ((pr_key.pr_key = ctl_get_prkey(lun, k)) == 0)
 798                         continue;
 799                 pr_key.pr_iid = k;
 800                 memcpy(&msg->lun.data[i], &pr_key, sizeof(pr_key));
 801                 i += sizeof(pr_key);
 802         }
 803         mtx_unlock(&lun->lun_lock);
 804         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->port, sizeof(msg->port) + i,
 805             M_WAITOK);
 806         free(msg, M_CTL);
 807
 808         if (lun->flags & CTL_LUN_PRIMARY_SC) {
 809                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
 810                         ctl_isc_announce_mode(lun, -1,
 811                             lun->mode_pages.index[i].page_code & SMPH_PC_MASK,
 812                             lun->mode_pages.index[i].subpage);
 813                 }
 814         }
 815 }
 816
 817 void
 818 ctl_isc_announce_port(struct ctl_port *port)
 819 {
 820         struct ctl_softc *softc = port->ctl_softc;
 821         union ctl_ha_msg *msg;
 822         int i;
 823
 824         if (port->targ_port < softc->port_min ||
 825             port->targ_port >= softc->port_max ||
 826             softc->ha_link != CTL_HA_LINK_ONLINE)
 827                 return;
 828         i = sizeof(msg->port) + strlen(port->port_name) + 1;
 829         if (port->lun_map)
 830                 i += port->lun_map_size * sizeof(uint32_t);
 831         if (port->port_devid)
 832                 i += port->port_devid->len;
 833         if (port->target_devid)
 834                 i += port->target_devid->len;
 835         if (port->init_devid)
 836                 i += port->init_devid->len;
 837         msg = malloc(i, M_CTL, M_WAITOK);
 838         bzero(&msg->port, sizeof(msg->port));
 839         msg->hdr.msg_type = CTL_MSG_PORT_SYNC;
 840         msg->hdr.nexus.targ_port = port->targ_port;
 841         msg->port.port_type = port->port_type;
 842         msg->port.physical_port = port->physical_port;
 843         msg->port.virtual_port = port->virtual_port;
 844         msg->port.status = port->status;
 845         i = 0;
 846         msg->port.name_len = sprintf(&msg->port.data[i],
 847             "%d:%s", softc->ha_id, port->port_name) + 1;
 848         i += msg->port.name_len;
 849         if (port->lun_map) {
 850                 msg->port.lun_map_len = port->lun_map_size * sizeof(uint32_t);
 851                 memcpy(&msg->port.data[i], port->lun_map,
 852                     msg->port.lun_map_len);
 853                 i += msg->port.lun_map_len;
 854         }
 855         if (port->port_devid) {
 856                 msg->port.port_devid_len = port->port_devid->len;
 857                 memcpy(&msg->port.data[i], port->port_devid->data,
 858                     msg->port.port_devid_len);
 859                 i += msg->port.port_devid_len;
 860         }
 861         if (port->target_devid) {
 862                 msg->port.target_devid_len = port->target_devid->len;
 863                 memcpy(&msg->port.data[i], port->target_devid->data,
 864                     msg->port.target_devid_len);
 865                 i += msg->port.target_devid_len;
 866         }
 867         if (port->init_devid) {
 868                 msg->port.init_devid_len = port->init_devid->len;
 869                 memcpy(&msg->port.data[i], port->init_devid->data,
 870                     msg->port.init_devid_len);
 871                 i += msg->port.init_devid_len;
 872         }
 873         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->port, sizeof(msg->port) + i,
 874             M_WAITOK);
 875         free(msg, M_CTL);
 876 }
 877
 878 void
 879 ctl_isc_announce_iid(struct ctl_port *port, int iid)
 880 {
 881         struct ctl_softc *softc = port->ctl_softc;
 882         union ctl_ha_msg *msg;
 883         int i, l;
 884
 885         if (port->targ_port < softc->port_min ||
 886             port->targ_port >= softc->port_max ||
 887             softc->ha_link != CTL_HA_LINK_ONLINE)
 888                 return;
 889         mtx_lock(&softc->ctl_lock);
 890         i = sizeof(msg->iid);
 891         l = 0;
 892         if (port->wwpn_iid[iid].name)
 893                 l = strlen(port->wwpn_iid[iid].name) + 1;
 894         i += l;
 895         msg = malloc(i, M_CTL, M_NOWAIT);
 896         if (msg == NULL) {
 897                 mtx_unlock(&softc->ctl_lock);
 898                 return;
 899         }
 900         bzero(&msg->iid, sizeof(msg->iid));
 901         msg->hdr.msg_type = CTL_MSG_IID_SYNC;
 902         msg->hdr.nexus.targ_port = port->targ_port;
 903         msg->hdr.nexus.initid = iid;
 904         msg->iid.in_use = port->wwpn_iid[iid].in_use;
 905         msg->iid.name_len = l;
 906         msg->iid.wwpn = port->wwpn_iid[iid].wwpn;
 907         if (port->wwpn_iid[iid].name)
 908                 strlcpy(msg->iid.data, port->wwpn_iid[iid].name, l);
 909         mtx_unlock(&softc->ctl_lock);
 910         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->iid, i, M_NOWAIT);
 911         free(msg, M_CTL);
 912 }
 913
 914 void
 915 ctl_isc_announce_mode(struct ctl_lun *lun, uint32_t initidx,
 916     uint8_t page, uint8_t subpage)
 917 {
 918         struct ctl_softc *softc = lun->ctl_softc;
 919         union ctl_ha_msg msg;
 920         u_int i;
 921
 922         if (softc->ha_link != CTL_HA_LINK_ONLINE)
 923                 return;
 924         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
 925                 if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) ==
 926                     page && lun->mode_pages.index[i].subpage == subpage)
 927                         break;
 928         }
 929         if (i == CTL_NUM_MODE_PAGES)
 930                 return;
 931
 932         /* Don't try to replicate pages not present on this device. */
 933         if (lun->mode_pages.index[i].page_data == NULL)
 934                 return;
 935
 936         bzero(&msg.mode, sizeof(msg.mode));
 937         msg.hdr.msg_type = CTL_MSG_MODE_SYNC;
 938         msg.hdr.nexus.targ_port = initidx / CTL_MAX_INIT_PER_PORT;
 939         msg.hdr.nexus.initid = initidx % CTL_MAX_INIT_PER_PORT;
 940         msg.hdr.nexus.targ_lun = lun->lun;
 941         msg.hdr.nexus.targ_mapped_lun = lun->lun;
 942         msg.mode.page_code = page;
 943         msg.mode.subpage = subpage;
 944         msg.mode.page_len = lun->mode_pages.index[i].page_len;
 945         memcpy(msg.mode.data, lun->mode_pages.index[i].page_data,
 946             msg.mode.page_len);
 947         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg.mode, sizeof(msg.mode),
 948             M_WAITOK);
 949 }
 950
 951 static void
 952 ctl_isc_ha_link_up(struct ctl_softc *softc)
 953 {
 954         struct ctl_port *port;
 955         struct ctl_lun *lun;
 956         union ctl_ha_msg msg;
 957         int i;
 958
 959         /* Announce this node parameters to peer for validation. */
 960         msg.login.msg_type = CTL_MSG_LOGIN;
 961         msg.login.version = CTL_HA_VERSION;
 962         msg.login.ha_mode = softc->ha_mode;
 963         msg.login.ha_id = softc->ha_id;
 964         msg.login.max_luns = CTL_MAX_LUNS;
 965         msg.login.max_ports = CTL_MAX_PORTS;
 966         msg.login.max_init_per_port = CTL_MAX_INIT_PER_PORT;
 967         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg.login, sizeof(msg.login),
 968             M_WAITOK);
 969
 970         STAILQ_FOREACH(port, &softc->port_list, links) {
 971                 ctl_isc_announce_port(port);
 972                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
 973                         if (port->wwpn_iid[i].in_use)
 974                                 ctl_isc_announce_iid(port, i);
 975                 }
 976         }
 977         STAILQ_FOREACH(lun, &softc->lun_list, links)
 978                 ctl_isc_announce_lun(lun);
 979 }
 980
 981 static void
 982 ctl_isc_ha_link_down(struct ctl_softc *softc)
 983 {
 984         struct ctl_port *port;
 985         struct ctl_lun *lun;
 986         union ctl_io *io;
 987         int i;
 988
 989         mtx_lock(&softc->ctl_lock);
 990         STAILQ_FOREACH(lun, &softc->lun_list, links) {
 991                 mtx_lock(&lun->lun_lock);
 992                 if (lun->flags & CTL_LUN_PEER_SC_PRIMARY) {
 993                         lun->flags &= ~CTL_LUN_PEER_SC_PRIMARY;
 994                         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
 995                 }
 996                 mtx_unlock(&lun->lun_lock);
 997
 998                 mtx_unlock(&softc->ctl_lock);
 999                 io = ctl_alloc_io(softc->othersc_pool);
1000                 mtx_lock(&softc->ctl_lock);
1001                 ctl_zero_io(io);
1002                 io->io_hdr.msg_type = CTL_MSG_FAILOVER;
1003                 io->io_hdr.nexus.targ_mapped_lun = lun->lun;
1004                 ctl_enqueue_isc(io);
1005         }
1006
1007         STAILQ_FOREACH(port, &softc->port_list, links) {
1008                 if (port->targ_port >= softc->port_min &&
1009                     port->targ_port < softc->port_max)
1010                         continue;
1011                 port->status &= ~CTL_PORT_STATUS_ONLINE;
1012                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1013                         port->wwpn_iid[i].in_use = 0;
1014                         free(port->wwpn_iid[i].name, M_CTL);
1015                         port->wwpn_iid[i].name = NULL;
1016                 }
1017         }
1018         mtx_unlock(&softc->ctl_lock);
1019 }
1020
1021 static void
1022 ctl_isc_ua(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1023 {
1024         struct ctl_lun *lun;
1025         uint32_t iid = ctl_get_initindex(&msg->hdr.nexus);
1026
1027         mtx_lock(&softc->ctl_lock);
1028         if (msg->hdr.nexus.targ_mapped_lun >= CTL_MAX_LUNS ||
1029             (lun = softc->ctl_luns[msg->hdr.nexus.targ_mapped_lun]) == NULL) {
1030                 mtx_unlock(&softc->ctl_lock);
1031                 return;
1032         }
1033         mtx_lock(&lun->lun_lock);
1034         mtx_unlock(&softc->ctl_lock);
1035         if (msg->ua.ua_type == CTL_UA_THIN_PROV_THRES && msg->ua.ua_set)
1036                 memcpy(lun->ua_tpt_info, msg->ua.ua_info, 8);
1037         if (msg->ua.ua_all) {
1038                 if (msg->ua.ua_set)
1039                         ctl_est_ua_all(lun, iid, msg->ua.ua_type);
1040                 else
1041                         ctl_clr_ua_all(lun, iid, msg->ua.ua_type);
1042         } else {
1043                 if (msg->ua.ua_set)
1044                         ctl_est_ua(lun, iid, msg->ua.ua_type);
1045                 else
1046                         ctl_clr_ua(lun, iid, msg->ua.ua_type);
1047         }
1048         mtx_unlock(&lun->lun_lock);
1049 }
1050
1051 static void
1052 ctl_isc_lun_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1053 {
1054         struct ctl_lun *lun;
1055         struct ctl_ha_msg_lun_pr_key pr_key;
1056         int i, k;
1057         ctl_lun_flags oflags;
1058         uint32_t targ_lun;
1059
1060         targ_lun = msg->hdr.nexus.targ_mapped_lun;
1061         mtx_lock(&softc->ctl_lock);
1062         if (targ_lun >= CTL_MAX_LUNS ||
1063             (lun = softc->ctl_luns[targ_lun]) == NULL) {
1064                 mtx_unlock(&softc->ctl_lock);
1065                 return;
1066         }
1067         mtx_lock(&lun->lun_lock);
1068         mtx_unlock(&softc->ctl_lock);
1069         if (lun->flags & CTL_LUN_DISABLED) {
1070                 mtx_unlock(&lun->lun_lock);
1071                 return;
1072         }
1073         i = (lun->lun_devid != NULL) ? lun->lun_devid->len : 0;
1074         if (msg->lun.lun_devid_len != i || (i > 0 &&
1075             memcmp(&msg->lun.data[0], lun->lun_devid->data, i) != 0)) {
1076                 mtx_unlock(&lun->lun_lock);
1077                 printf("%s: Received conflicting HA LUN %d\n",
1078                     __func__, targ_lun);
1079                 return;
1080         } else {
1081                 /* Record whether peer is primary. */
1082                 oflags = lun->flags;
1083                 if ((msg->lun.flags & CTL_LUN_PRIMARY_SC) &&
1084                     (msg->lun.flags & CTL_LUN_DISABLED) == 0)
1085                         lun->flags |= CTL_LUN_PEER_SC_PRIMARY;
1086                 else
1087                         lun->flags &= ~CTL_LUN_PEER_SC_PRIMARY;
1088                 if (oflags != lun->flags)
1089                         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
1090
1091                 /* If peer is primary and we are not -- use data */
1092                 if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
1093                     (lun->flags & CTL_LUN_PEER_SC_PRIMARY)) {
1094                         lun->pr_generation = msg->lun.pr_generation;
1095                         lun->pr_res_idx = msg->lun.pr_res_idx;
1096                         lun->pr_res_type = msg->lun.pr_res_type;
1097                         lun->pr_key_count = msg->lun.pr_key_count;
1098                         for (k = 0; k < CTL_MAX_INITIATORS; k++)
1099                                 ctl_clr_prkey(lun, k);
1100                         for (k = 0; k < msg->lun.pr_key_count; k++) {
1101                                 memcpy(&pr_key, &msg->lun.data[i],
1102                                     sizeof(pr_key));
1103                                 ctl_alloc_prkey(lun, pr_key.pr_iid);
1104                                 ctl_set_prkey(lun, pr_key.pr_iid,
1105                                     pr_key.pr_key);
1106                                 i += sizeof(pr_key);
1107                         }
1108                 }
1109
1110                 mtx_unlock(&lun->lun_lock);
1111                 CTL_DEBUG_PRINT(("%s: Known LUN %d, peer is %s\n",
1112                     __func__, targ_lun,
1113                     (msg->lun.flags & CTL_LUN_PRIMARY_SC) ?
1114                     "primary" : "secondary"));
1115
1116                 /* If we are primary but peer doesn't know -- notify */
1117                 if ((lun->flags & CTL_LUN_PRIMARY_SC) &&
1118                     (msg->lun.flags & CTL_LUN_PEER_SC_PRIMARY) == 0)
1119                         ctl_isc_announce_lun(lun);
1120         }
1121 }
1122
1123 static void
1124 ctl_isc_port_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1125 {
1126         struct ctl_port *port;
1127         struct ctl_lun *lun;
1128         int i, new;
1129
1130         port = softc->ctl_ports[msg->hdr.nexus.targ_port];
1131         if (port == NULL) {
1132                 CTL_DEBUG_PRINT(("%s: New port %d\n", __func__,
1133                     msg->hdr.nexus.targ_port));
1134                 new = 1;
1135                 port = malloc(sizeof(*port), M_CTL, M_WAITOK | M_ZERO);
1136                 port->frontend = &ha_frontend;
1137                 port->targ_port = msg->hdr.nexus.targ_port;
1138                 port->fe_datamove = ctl_ha_datamove;
1139                 port->fe_done = ctl_ha_done;
1140         } else if (port->frontend == &ha_frontend) {
1141                 CTL_DEBUG_PRINT(("%s: Updated port %d\n", __func__,
1142                     msg->hdr.nexus.targ_port));
1143                 new = 0;
1144         } else {
1145                 printf("%s: Received conflicting HA port %d\n",
1146                     __func__, msg->hdr.nexus.targ_port);
1147                 return;
1148         }
1149         port->port_type = msg->port.port_type;
1150         port->physical_port = msg->port.physical_port;
1151         port->virtual_port = msg->port.virtual_port;
1152         port->status = msg->port.status;
1153         i = 0;
1154         free(port->port_name, M_CTL);
1155         port->port_name = strndup(&msg->port.data[i], msg->port.name_len,
1156             M_CTL);
1157         i += msg->port.name_len;
1158         if (msg->port.lun_map_len != 0) {
1159                 if (port->lun_map == NULL ||
1160                     port->lun_map_size * sizeof(uint32_t) <
1161                     msg->port.lun_map_len) {
1162                         port->lun_map_size = 0;
1163                         free(port->lun_map, M_CTL);
1164                         port->lun_map = malloc(msg->port.lun_map_len,
1165                             M_CTL, M_WAITOK);
1166                 }
1167                 memcpy(port->lun_map, &msg->port.data[i], msg->port.lun_map_len);
1168                 port->lun_map_size = msg->port.lun_map_len / sizeof(uint32_t);
1169                 i += msg->port.lun_map_len;
1170         } else {
1171                 port->lun_map_size = 0;
1172                 free(port->lun_map, M_CTL);
1173                 port->lun_map = NULL;
1174         }
1175         if (msg->port.port_devid_len != 0) {
1176                 if (port->port_devid == NULL ||
1177                     port->port_devid->len < msg->port.port_devid_len) {
1178                         free(port->port_devid, M_CTL);
1179                         port->port_devid = malloc(sizeof(struct ctl_devid) +
1180                             msg->port.port_devid_len, M_CTL, M_WAITOK);
1181                 }
1182                 memcpy(port->port_devid->data, &msg->port.data[i],
1183                     msg->port.port_devid_len);
1184                 port->port_devid->len = msg->port.port_devid_len;
1185                 i += msg->port.port_devid_len;
1186         } else {
1187                 free(port->port_devid, M_CTL);
1188                 port->port_devid = NULL;
1189         }
1190         if (msg->port.target_devid_len != 0) {
1191                 if (port->target_devid == NULL ||
1192                     port->target_devid->len < msg->port.target_devid_len) {
1193                         free(port->target_devid, M_CTL);
1194                         port->target_devid = malloc(sizeof(struct ctl_devid) +
1195                             msg->port.target_devid_len, M_CTL, M_WAITOK);
1196                 }
1197                 memcpy(port->target_devid->data, &msg->port.data[i],
1198                     msg->port.target_devid_len);
1199                 port->target_devid->len = msg->port.target_devid_len;
1200                 i += msg->port.target_devid_len;
1201         } else {
1202                 free(port->target_devid, M_CTL);
1203                 port->target_devid = NULL;
1204         }
1205         if (msg->port.init_devid_len != 0) {
1206                 if (port->init_devid == NULL ||
1207                     port->init_devid->len < msg->port.init_devid_len) {
1208                         free(port->init_devid, M_CTL);
1209                         port->init_devid = malloc(sizeof(struct ctl_devid) +
1210                             msg->port.init_devid_len, M_CTL, M_WAITOK);
1211                 }
1212                 memcpy(port->init_devid->data, &msg->port.data[i],
1213                     msg->port.init_devid_len);
1214                 port->init_devid->len = msg->port.init_devid_len;
1215                 i += msg->port.init_devid_len;
1216         } else {
1217                 free(port->init_devid, M_CTL);
1218                 port->init_devid = NULL;
1219         }
1220         if (new) {
1221                 if (ctl_port_register(port) != 0) {
1222                         printf("%s: ctl_port_register() failed with error\n",
1223                             __func__);
1224                 }
1225         }
1226         mtx_lock(&softc->ctl_lock);
1227         STAILQ_FOREACH(lun, &softc->lun_list, links) {
1228                 if (ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
1229                         continue;
1230                 mtx_lock(&lun->lun_lock);
1231                 ctl_est_ua_all(lun, -1, CTL_UA_INQ_CHANGE);
1232                 mtx_unlock(&lun->lun_lock);
1233         }
1234         mtx_unlock(&softc->ctl_lock);
1235 }
1236
1237 static void
1238 ctl_isc_iid_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1239 {
1240         struct ctl_port *port;
1241         int iid;
1242
1243         port = softc->ctl_ports[msg->hdr.nexus.targ_port];
1244         if (port == NULL) {
1245                 printf("%s: Received IID for unknown port %d\n",
1246                     __func__, msg->hdr.nexus.targ_port);
1247                 return;
1248         }
1249         iid = msg->hdr.nexus.initid;
1250         port->wwpn_iid[iid].in_use = msg->iid.in_use;
1251         port->wwpn_iid[iid].wwpn = msg->iid.wwpn;
1252         free(port->wwpn_iid[iid].name, M_CTL);
1253         if (msg->iid.name_len) {
1254                 port->wwpn_iid[iid].name = strndup(&msg->iid.data[0],
1255                     msg->iid.name_len, M_CTL);
1256         } else
1257                 port->wwpn_iid[iid].name = NULL;
1258 }
1259
1260 static void
1261 ctl_isc_login(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1262 {
1263
1264         if (msg->login.version != CTL_HA_VERSION) {
1265                 printf("CTL HA peers have different versions %d != %d\n",
1266                     msg->login.version, CTL_HA_VERSION);
1267                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1268                 return;
1269         }
1270         if (msg->login.ha_mode != softc->ha_mode) {
1271                 printf("CTL HA peers have different ha_mode %d != %d\n",
1272                     msg->login.ha_mode, softc->ha_mode);
1273                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1274                 return;
1275         }
1276         if (msg->login.ha_id == softc->ha_id) {
1277                 printf("CTL HA peers have same ha_id %d\n", msg->login.ha_id);
1278                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1279                 return;
1280         }
1281         if (msg->login.max_luns != CTL_MAX_LUNS ||
1282             msg->login.max_ports != CTL_MAX_PORTS ||
1283             msg->login.max_init_per_port != CTL_MAX_INIT_PER_PORT) {
1284                 printf("CTL HA peers have different limits\n");
1285                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1286                 return;
1287         }
1288 }
1289
1290 static void
1291 ctl_isc_mode_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1292 {
1293         struct ctl_lun *lun;
1294         u_int i;
1295         uint32_t initidx, targ_lun;
1296
1297         targ_lun = msg->hdr.nexus.targ_mapped_lun;
1298         mtx_lock(&softc->ctl_lock);
1299         if (targ_lun >= CTL_MAX_LUNS ||
1300             (lun = softc->ctl_luns[targ_lun]) == NULL) {
1301                 mtx_unlock(&softc->ctl_lock);
1302                 return;
1303         }
1304         mtx_lock(&lun->lun_lock);
1305         mtx_unlock(&softc->ctl_lock);
1306         if (lun->flags & CTL_LUN_DISABLED) {
1307                 mtx_unlock(&lun->lun_lock);
1308                 return;
1309         }
1310         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
1311                 if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) ==
1312                     msg->mode.page_code &&
1313                     lun->mode_pages.index[i].subpage == msg->mode.subpage)
1314                         break;
1315         }
1316         if (i == CTL_NUM_MODE_PAGES) {
1317                 mtx_unlock(&lun->lun_lock);
1318                 return;
1319         }
1320         memcpy(lun->mode_pages.index[i].page_data, msg->mode.data,
1321             lun->mode_pages.index[i].page_len);
1322         initidx = ctl_get_initindex(&msg->hdr.nexus);
1323         if (initidx != -1)
1324                 ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
1325         mtx_unlock(&lun->lun_lock);
1326 }
1327
1328 /*
1329  * ISC (Inter Shelf Communication) event handler.  Events from the HA
1330  * subsystem come in here.
1331  */
1332 static void
1333 ctl_isc_event_handler(ctl_ha_channel channel, ctl_ha_event event, int param)
1334 {
1335         struct ctl_softc *softc = control_softc;
1336         union ctl_io *io;
1337         struct ctl_prio *presio;
1338         ctl_ha_status isc_status;
1339
1340         CTL_DEBUG_PRINT(("CTL: Isc Msg event %d\n", event));
1341         if (event == CTL_HA_EVT_MSG_RECV) {
1342                 union ctl_ha_msg *msg, msgbuf;
1343
1344                 if (param > sizeof(msgbuf))
1345                         msg = malloc(param, M_CTL, M_WAITOK);
1346                 else
1347                         msg = &msgbuf;
1348                 isc_status = ctl_ha_msg_recv(CTL_HA_CHAN_CTL, msg, param,
1349                     M_WAITOK);
1350                 if (isc_status != CTL_HA_STATUS_SUCCESS) {
1351                         printf("%s: Error receiving message: %d\n",
1352                             __func__, isc_status);
1353                         if (msg != &msgbuf)
1354                                 free(msg, M_CTL);
1355                         return;
1356                 }
1357
1358                 CTL_DEBUG_PRINT(("CTL: msg_type %d\n", msg->msg_type));
1359                 switch (msg->hdr.msg_type) {
1360                 case CTL_MSG_SERIALIZE:
1361                         io = ctl_alloc_io(softc->othersc_pool);
1362                         ctl_zero_io(io);
1363                         // populate ctsio from msg
1364                         io->io_hdr.io_type = CTL_IO_SCSI;
1365                         io->io_hdr.msg_type = CTL_MSG_SERIALIZE;
1366                         io->io_hdr.original_sc = msg->hdr.original_sc;
1367                         io->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC |
1368                                             CTL_FLAG_IO_ACTIVE;
1369                         /*
1370                          * If we're in serialization-only mode, we don't
1371                          * want to go through full done processing.  Thus
1372                          * the COPY flag.
1373                          *
1374                          * XXX KDM add another flag that is more specific.
1375                          */
1376                         if (softc->ha_mode != CTL_HA_MODE_XFER)
1377                                 io->io_hdr.flags |= CTL_FLAG_INT_COPY;
1378                         io->io_hdr.nexus = msg->hdr.nexus;
1379 #if 0
1380                         printf("port %u, iid %u, lun %u\n",
1381                                io->io_hdr.nexus.targ_port,
1382                                io->io_hdr.nexus.initid,
1383                                io->io_hdr.nexus.targ_lun);
1384 #endif
1385                         io->scsiio.tag_num = msg->scsi.tag_num;
1386                         io->scsiio.tag_type = msg->scsi.tag_type;
1387 #ifdef CTL_TIME_IO
1388                         io->io_hdr.start_time = time_uptime;
1389                         getbinuptime(&io->io_hdr.start_bt);
1390 #endif /* CTL_TIME_IO */
1391                         io->scsiio.cdb_len = msg->scsi.cdb_len;
1392                         memcpy(io->scsiio.cdb, msg->scsi.cdb,
1393                                CTL_MAX_CDBLEN);
1394                         if (softc->ha_mode == CTL_HA_MODE_XFER) {
1395                                 const struct ctl_cmd_entry *entry;
1396
1397                                 entry = ctl_get_cmd_entry(&io->scsiio, NULL);
1398                                 io->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
1399                                 io->io_hdr.flags |=
1400                                         entry->flags & CTL_FLAG_DATA_MASK;
1401                         }
1402                         ctl_enqueue_isc(io);
1403                         break;
1404
1405                 /* Performed on the Originating SC, XFER mode only */
1406                 case CTL_MSG_DATAMOVE: {
1407                         struct ctl_sg_entry *sgl;
1408                         int i, j;
1409
1410                         io = msg->hdr.original_sc;
1411                         if (io == NULL) {
1412                                 printf("%s: original_sc == NULL!\n", __func__);
1413                                 /* XXX KDM do something here */
1414                                 break;
1415                         }
1416                         io->io_hdr.msg_type = CTL_MSG_DATAMOVE;
1417                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1418                         /*
1419                          * Keep track of this, we need to send it back over
1420                          * when the datamove is complete.
1421                          */
1422                         io->io_hdr.serializing_sc = msg->hdr.serializing_sc;
1423                         if (msg->hdr.status == CTL_SUCCESS)
1424                                 io->io_hdr.status = msg->hdr.status;
1425
1426                         if (msg->dt.sg_sequence == 0) {
1427 #ifdef CTL_TIME_IO
1428                                 getbinuptime(&io->io_hdr.dma_start_bt);
1429 #endif
1430                                 i = msg->dt.kern_sg_entries +
1431                                     msg->dt.kern_data_len /
1432                                     CTL_HA_DATAMOVE_SEGMENT + 1;
1433                                 sgl = malloc(sizeof(*sgl) * i, M_CTL,
1434                                     M_WAITOK | M_ZERO);
1435                                 io->io_hdr.remote_sglist = sgl;
1436                                 io->io_hdr.local_sglist =
1437                                     &sgl[msg->dt.kern_sg_entries];
1438
1439                                 io->scsiio.kern_data_ptr = (uint8_t *)sgl;
1440
1441                                 io->scsiio.kern_sg_entries =
1442                                         msg->dt.kern_sg_entries;
1443                                 io->scsiio.rem_sg_entries =
1444                                         msg->dt.kern_sg_entries;
1445                                 io->scsiio.kern_data_len =
1446                                         msg->dt.kern_data_len;
1447                                 io->scsiio.kern_total_len =
1448                                         msg->dt.kern_total_len;
1449                                 io->scsiio.kern_data_resid =
1450                                         msg->dt.kern_data_resid;
1451                                 io->scsiio.kern_rel_offset =
1452                                         msg->dt.kern_rel_offset;
1453                                 io->io_hdr.flags &= ~CTL_FLAG_BUS_ADDR;
1454                                 io->io_hdr.flags |= msg->dt.flags &
1455                                     CTL_FLAG_BUS_ADDR;
1456                         } else
1457                                 sgl = (struct ctl_sg_entry *)
1458                                         io->scsiio.kern_data_ptr;
1459
1460                         for (i = msg->dt.sent_sg_entries, j = 0;
1461                              i < (msg->dt.sent_sg_entries +
1462                              msg->dt.cur_sg_entries); i++, j++) {
1463                                 sgl[i].addr = msg->dt.sg_list[j].addr;
1464                                 sgl[i].len = msg->dt.sg_list[j].len;
1465
1466 #if 0
1467                                 printf("%s: DATAMOVE: %p,%lu j=%d, i=%d\n",
1468                                     __func__, sgl[i].addr, sgl[i].len, j, i);
1469 #endif
1470                         }
1471
1472                         /*
1473                          * If this is the last piece of the I/O, we've got
1474                          * the full S/G list.  Queue processing in the thread.
1475                          * Otherwise wait for the next piece.
1476                          */
1477                         if (msg->dt.sg_last != 0)
1478                                 ctl_enqueue_isc(io);
1479                         break;
1480                 }
1481                 /* Performed on the Serializing (primary) SC, XFER mode only */
1482                 case CTL_MSG_DATAMOVE_DONE: {
1483                         if (msg->hdr.serializing_sc == NULL) {
1484                                 printf("%s: serializing_sc == NULL!\n",
1485                                        __func__);
1486                                 /* XXX KDM now what? */
1487                                 break;
1488                         }
1489                         /*
1490                          * We grab the sense information here in case
1491                          * there was a failure, so we can return status
1492                          * back to the initiator.
1493                          */
1494                         io = msg->hdr.serializing_sc;
1495                         io->io_hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
1496                         io->io_hdr.flags &= ~CTL_FLAG_DMA_INPROG;
1497                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1498                         io->io_hdr.port_status = msg->scsi.fetd_status;
1499                         io->scsiio.residual = msg->scsi.residual;
1500                         if (msg->hdr.status != CTL_STATUS_NONE) {
1501                                 io->io_hdr.status = msg->hdr.status;
1502                                 io->scsiio.scsi_status = msg->scsi.scsi_status;
1503                                 io->scsiio.sense_len = msg->scsi.sense_len;
1504                                 io->scsiio.sense_residual =msg->scsi.sense_residual;
1505                                 memcpy(&io->scsiio.sense_data,
1506                                     &msg->scsi.sense_data,
1507                                     msg->scsi.sense_len);
1508                                 if (msg->hdr.status == CTL_SUCCESS)
1509                                         io->io_hdr.flags |= CTL_FLAG_STATUS_SENT;
1510                         }
1511                         ctl_enqueue_isc(io);
1512                         break;
1513                 }
1514
1515                 /* Preformed on Originating SC, SER_ONLY mode */
1516                 case CTL_MSG_R2R:
1517                         io = msg->hdr.original_sc;
1518                         if (io == NULL) {
1519                                 printf("%s: original_sc == NULL!\n",
1520                                     __func__);
1521                                 break;
1522                         }
1523                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1524                         io->io_hdr.msg_type = CTL_MSG_R2R;
1525                         io->io_hdr.serializing_sc = msg->hdr.serializing_sc;
1526                         ctl_enqueue_isc(io);
1527                         break;
1528
1529                 /*
1530                  * Performed on Serializing(i.e. primary SC) SC in SER_ONLY
1531                  * mode.
1532                  * Performed on the Originating (i.e. secondary) SC in XFER
1533                  * mode
1534                  */
1535                 case CTL_MSG_FINISH_IO:
1536                         if (softc->ha_mode == CTL_HA_MODE_XFER)
1537                                 ctl_isc_handler_finish_xfer(softc, msg);
1538                         else
1539                                 ctl_isc_handler_finish_ser_only(softc, msg);
1540                         break;
1541
1542                 /* Preformed on Originating SC */
1543                 case CTL_MSG_BAD_JUJU:
1544                         io = msg->hdr.original_sc;
1545                         if (io == NULL) {
1546                                 printf("%s: Bad JUJU!, original_sc is NULL!\n",
1547                                        __func__);
1548                                 break;
1549                         }
1550                         ctl_copy_sense_data(msg, io);
1551                         /*
1552                          * IO should have already been cleaned up on other
1553                          * SC so clear this flag so we won't send a message
1554                          * back to finish the IO there.
1555                          */
1556                         io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
1557                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1558
1559                         /* io = msg->hdr.serializing_sc; */
1560                         io->io_hdr.msg_type = CTL_MSG_BAD_JUJU;
1561                         ctl_enqueue_isc(io);
1562                         break;
1563
1564                 /* Handle resets sent from the other side */
1565                 case CTL_MSG_MANAGE_TASKS: {
1566                         struct ctl_taskio *taskio;
1567                         taskio = (struct ctl_taskio *)ctl_alloc_io(
1568                             softc->othersc_pool);
1569                         ctl_zero_io((union ctl_io *)taskio);
1570                         taskio->io_hdr.io_type = CTL_IO_TASK;
1571                         taskio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
1572                         taskio->io_hdr.nexus = msg->hdr.nexus;
1573                         taskio->task_action = msg->task.task_action;
1574                         taskio->tag_num = msg->task.tag_num;
1575                         taskio->tag_type = msg->task.tag_type;
1576 #ifdef CTL_TIME_IO
1577                         taskio->io_hdr.start_time = time_uptime;
1578                         getbinuptime(&taskio->io_hdr.start_bt);
1579 #endif /* CTL_TIME_IO */
1580                         ctl_run_task((union ctl_io *)taskio);
1581                         break;
1582                 }
1583                 /* Persistent Reserve action which needs attention */
1584                 case CTL_MSG_PERS_ACTION:
1585                         presio = (struct ctl_prio *)ctl_alloc_io(
1586                             softc->othersc_pool);
1587                         ctl_zero_io((union ctl_io *)presio);
1588                         presio->io_hdr.msg_type = CTL_MSG_PERS_ACTION;
1589                         presio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
1590                         presio->io_hdr.nexus = msg->hdr.nexus;
1591                         presio->pr_msg = msg->pr;
1592                         ctl_enqueue_isc((union ctl_io *)presio);
1593                         break;
1594                 case CTL_MSG_UA:
1595                         ctl_isc_ua(softc, msg, param);
1596                         break;
1597                 case CTL_MSG_PORT_SYNC:
1598                         ctl_isc_port_sync(softc, msg, param);
1599                         break;
1600                 case CTL_MSG_LUN_SYNC:
1601                         ctl_isc_lun_sync(softc, msg, param);
1602                         break;
1603                 case CTL_MSG_IID_SYNC:
1604                         ctl_isc_iid_sync(softc, msg, param);
1605                         break;
1606                 case CTL_MSG_LOGIN:
1607                         ctl_isc_login(softc, msg, param);
1608                         break;
1609                 case CTL_MSG_MODE_SYNC:
1610                         ctl_isc_mode_sync(softc, msg, param);
1611                         break;
1612                 default:
1613                         printf("Received HA message of unknown type %d\n",
1614                             msg->hdr.msg_type);
1615                         ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1616                         break;
1617                 }
1618                 if (msg != &msgbuf)
1619                         free(msg, M_CTL);
1620         } else if (event == CTL_HA_EVT_LINK_CHANGE) {
1621                 printf("CTL: HA link status changed from %d to %d\n",
1622                     softc->ha_link, param);
1623                 if (param == softc->ha_link)
1624                         return;
1625                 if (softc->ha_link == CTL_HA_LINK_ONLINE) {
1626                         softc->ha_link = param;
1627                         ctl_isc_ha_link_down(softc);
1628                 } else {
1629                         softc->ha_link = param;
1630                         if (softc->ha_link == CTL_HA_LINK_ONLINE)
1631                                 ctl_isc_ha_link_up(softc);
1632                 }
1633                 return;
1634         } else {
1635                 printf("ctl_isc_event_handler: Unknown event %d\n", event);
1636                 return;
1637         }
1638 }
1639
1640 static void
1641 ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest)
1642 {
1643
1644         memcpy(&dest->scsiio.sense_data, &src->scsi.sense_data,
1645             src->scsi.sense_len);
1646         dest->scsiio.scsi_status = src->scsi.scsi_status;
1647         dest->scsiio.sense_len = src->scsi.sense_len;
1648         dest->io_hdr.status = src->hdr.status;
1649 }
1650
1651 static void
1652 ctl_copy_sense_data_back(union ctl_io *src, union ctl_ha_msg *dest)
1653 {
1654
1655         memcpy(&dest->scsi.sense_data, &src->scsiio.sense_data,
1656             src->scsiio.sense_len);
1657         dest->scsi.scsi_status = src->scsiio.scsi_status;
1658         dest->scsi.sense_len = src->scsiio.sense_len;
1659         dest->hdr.status = src->io_hdr.status;
1660 }
1661
1662 void
1663 ctl_est_ua(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua)
1664 {
1665         struct ctl_softc *softc = lun->ctl_softc;
1666         ctl_ua_type *pu;
1667
1668         if (initidx < softc->init_min || initidx >= softc->init_max)
1669                 return;
1670         mtx_assert(&lun->lun_lock, MA_OWNED);
1671         pu = lun->pending_ua[initidx / CTL_MAX_INIT_PER_PORT];
1672         if (pu == NULL)
1673                 return;
1674         pu[initidx % CTL_MAX_INIT_PER_PORT] |= ua;
1675 }
1676
1677 void
1678 ctl_est_ua_port(struct ctl_lun *lun, int port, uint32_t except, ctl_ua_type ua)
1679 {
1680         int i;
1681
1682         mtx_assert(&lun->lun_lock, MA_OWNED);
1683         if (lun->pending_ua[port] == NULL)
1684                 return;
1685         for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1686                 if (port * CTL_MAX_INIT_PER_PORT + i == except)
1687                         continue;
1688                 lun->pending_ua[port][i] |= ua;
1689         }
1690 }
1691
1692 void
1693 ctl_est_ua_all(struct ctl_lun *lun, uint32_t except, ctl_ua_type ua)
1694 {
1695         struct ctl_softc *softc = lun->ctl_softc;
1696         int i;
1697
1698         mtx_assert(&lun->lun_lock, MA_OWNED);
1699         for (i = softc->port_min; i < softc->port_max; i++)
1700                 ctl_est_ua_port(lun, i, except, ua);
1701 }
1702
1703 void
1704 ctl_clr_ua(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua)
1705 {
1706         struct ctl_softc *softc = lun->ctl_softc;
1707         ctl_ua_type *pu;
1708
1709         if (initidx < softc->init_min || initidx >= softc->init_max)
1710                 return;
1711         mtx_assert(&lun->lun_lock, MA_OWNED);
1712         pu = lun->pending_ua[initidx / CTL_MAX_INIT_PER_PORT];
1713         if (pu == NULL)
1714                 return;
1715         pu[initidx % CTL_MAX_INIT_PER_PORT] &= ~ua;
1716 }
1717
1718 void
1719 ctl_clr_ua_all(struct ctl_lun *lun, uint32_t except, ctl_ua_type ua)
1720 {
1721         struct ctl_softc *softc = lun->ctl_softc;
1722         int i, j;
1723
1724         mtx_assert(&lun->lun_lock, MA_OWNED);
1725         for (i = softc->port_min; i < softc->port_max; i++) {
1726                 if (lun->pending_ua[i] == NULL)
1727                         continue;
1728                 for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
1729                         if (i * CTL_MAX_INIT_PER_PORT + j == except)
1730                                 continue;
1731                         lun->pending_ua[i][j] &= ~ua;
1732                 }
1733         }
1734 }
1735
1736 void
1737 ctl_clr_ua_allluns(struct ctl_softc *ctl_softc, uint32_t initidx,
1738     ctl_ua_type ua_type)
1739 {
1740         struct ctl_lun *lun;
1741
1742         mtx_assert(&ctl_softc->ctl_lock, MA_OWNED);
1743         STAILQ_FOREACH(lun, &ctl_softc->lun_list, links) {
1744                 mtx_lock(&lun->lun_lock);
1745                 ctl_clr_ua(lun, initidx, ua_type);
1746                 mtx_unlock(&lun->lun_lock);
1747         }
1748 }
1749
1750 static int
1751 ctl_ha_role_sysctl(SYSCTL_HANDLER_ARGS)
1752 {
1753         struct ctl_softc *softc = (struct ctl_softc *)arg1;
1754         struct ctl_lun *lun;
1755         struct ctl_lun_req ireq;
1756         int error, value;
1757
1758         value = (softc->flags & CTL_FLAG_ACTIVE_SHELF) ? 0 : 1;
1759         error = sysctl_handle_int(oidp, &value, 0, req);
1760         if ((error != 0) || (req->newptr == NULL))
1761                 return (error);
1762
1763         mtx_lock(&softc->ctl_lock);
1764         if (value == 0)
1765                 softc->flags |= CTL_FLAG_ACTIVE_SHELF;
1766         else
1767                 softc->flags &= ~CTL_FLAG_ACTIVE_SHELF;
1768         STAILQ_FOREACH(lun, &softc->lun_list, links) {
1769                 mtx_unlock(&softc->ctl_lock);
1770                 bzero(&ireq, sizeof(ireq));
1771                 ireq.reqtype = CTL_LUNREQ_MODIFY;
1772                 ireq.reqdata.modify.lun_id = lun->lun;
1773                 lun->backend->ioctl(NULL, CTL_LUN_REQ, (caddr_t)&ireq, 0,
1774                     curthread);
1775                 if (ireq.status != CTL_LUN_OK) {
1776                         printf("%s: CTL_LUNREQ_MODIFY returned %d '%s'\n",
1777                             __func__, ireq.status, ireq.error_str);
1778                 }
1779                 mtx_lock(&softc->ctl_lock);
1780         }
1781         mtx_unlock(&softc->ctl_lock);
1782         return (0);
1783 }
1784
1785 static int
1786 ctl_init(void)
1787 {
1788         struct make_dev_args args;
1789         struct ctl_softc *softc;
1790         void *other_pool;
1791         int i, error;
1792
1793         softc = control_softc = malloc(sizeof(*control_softc), M_DEVBUF,
1794                                M_WAITOK | M_ZERO);
1795
1796         make_dev_args_init(&args);
1797         args.mda_devsw = &ctl_cdevsw;
1798         args.mda_uid = UID_ROOT;
1799         args.mda_gid = GID_OPERATOR;
1800         args.mda_mode = 0600;
1801         args.mda_si_drv1 = softc;
1802         error = make_dev_s(&args, &softc->dev, "cam/ctl");
1803         if (error != 0) {
1804                 free(softc, M_DEVBUF);
1805                 control_softc = NULL;
1806                 return (error);
1807         }
1808
1809         sysctl_ctx_init(&softc->sysctl_ctx);
1810         softc->sysctl_tree = SYSCTL_ADD_NODE(&softc->sysctl_ctx,
1811                 SYSCTL_STATIC_CHILDREN(_kern_cam), OID_AUTO, "ctl",
1812                 CTLFLAG_RD, 0, "CAM Target Layer");
1813
1814         if (softc->sysctl_tree == NULL) {
1815                 printf("%s: unable to allocate sysctl tree\n", __func__);
1816                 destroy_dev(softc->dev);
1817                 free(softc, M_DEVBUF);
1818                 control_softc = NULL;
1819                 return (ENOMEM);
1820         }
1821
1822         mtx_init(&softc->ctl_lock, "CTL mutex", NULL, MTX_DEF);
1823         softc->io_zone = uma_zcreate("CTL IO", sizeof(union ctl_io),
1824             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1825         softc->flags = 0;
1826
1827         SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1828             OID_AUTO, "ha_mode", CTLFLAG_RDTUN, (int *)&softc->ha_mode, 0,
1829             "HA mode (0 - act/stby, 1 - serialize only, 2 - xfer)");
1830
1831         /*
1832          * In Copan's HA scheme, the "master" and "slave" roles are
1833          * figured out through the slot the controller is in.  Although it
1834          * is an active/active system, someone has to be in charge.
1835          */
1836         SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1837             OID_AUTO, "ha_id", CTLFLAG_RDTUN, &softc->ha_id, 0,
1838             "HA head ID (0 - no HA)");
1839         if (softc->ha_id == 0 || softc->ha_id > NUM_HA_SHELVES) {
1840                 softc->flags |= CTL_FLAG_ACTIVE_SHELF;
1841                 softc->is_single = 1;
1842                 softc->port_cnt = CTL_MAX_PORTS;
1843                 softc->port_min = 0;
1844         } else {
1845                 softc->port_cnt = CTL_MAX_PORTS / NUM_HA_SHELVES;
1846                 softc->port_min = (softc->ha_id - 1) * softc->port_cnt;
1847         }
1848         softc->port_max = softc->port_min + softc->port_cnt;
1849         softc->init_min = softc->port_min * CTL_MAX_INIT_PER_PORT;
1850         softc->init_max = softc->port_max * CTL_MAX_INIT_PER_PORT;
1851
1852         SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1853             OID_AUTO, "ha_link", CTLFLAG_RD, (int *)&softc->ha_link, 0,
1854             "HA link state (0 - offline, 1 - unknown, 2 - online)");
1855
1856         STAILQ_INIT(&softc->lun_list);
1857         STAILQ_INIT(&softc->pending_lun_queue);
1858         STAILQ_INIT(&softc->fe_list);
1859         STAILQ_INIT(&softc->port_list);
1860         STAILQ_INIT(&softc->be_list);
1861         ctl_tpc_init(softc);
1862
1863         if (ctl_pool_create(softc, "othersc", CTL_POOL_ENTRIES_OTHER_SC,
1864                             &other_pool) != 0)
1865         {
1866                 printf("ctl: can't allocate %d entry other SC pool, "
1867                        "exiting\n", CTL_POOL_ENTRIES_OTHER_SC);
1868                 return (ENOMEM);
1869         }
1870         softc->othersc_pool = other_pool;
1871
1872         if (worker_threads <= 0)
1873                 worker_threads = max(1, mp_ncpus / 4);
1874         if (worker_threads > CTL_MAX_THREADS)
1875                 worker_threads = CTL_MAX_THREADS;
1876
1877         for (i = 0; i < worker_threads; i++) {
1878                 struct ctl_thread *thr = &softc->threads[i];
1879
1880                 mtx_init(&thr->queue_lock, "CTL queue mutex", NULL, MTX_DEF);
1881                 thr->ctl_softc = softc;
1882                 STAILQ_INIT(&thr->incoming_queue);
1883                 STAILQ_INIT(&thr->rtr_queue);
1884                 STAILQ_INIT(&thr->done_queue);
1885                 STAILQ_INIT(&thr->isc_queue);
1886
1887                 error = kproc_kthread_add(ctl_work_thread, thr,
1888                     &softc->ctl_proc, &thr->thread, 0, 0, "ctl", "work%d", i);
1889                 if (error != 0) {
1890                         printf("error creating CTL work thread!\n");
1891                         ctl_pool_free(other_pool);
1892                         return (error);
1893                 }
1894         }
1895         error = kproc_kthread_add(ctl_lun_thread, softc,
1896             &softc->ctl_proc, NULL, 0, 0, "ctl", "lun");
1897         if (error != 0) {
1898                 printf("error creating CTL lun thread!\n");
1899                 ctl_pool_free(other_pool);
1900                 return (error);
1901         }
1902         error = kproc_kthread_add(ctl_thresh_thread, softc,
1903             &softc->ctl_proc, NULL, 0, 0, "ctl", "thresh");
1904         if (error != 0) {
1905                 printf("error creating CTL threshold thread!\n");
1906                 ctl_pool_free(other_pool);
1907                 return (error);
1908         }
1909
1910         SYSCTL_ADD_PROC(&softc->sysctl_ctx,SYSCTL_CHILDREN(softc->sysctl_tree),
1911             OID_AUTO, "ha_role", CTLTYPE_INT | CTLFLAG_RWTUN,
1912             softc, 0, ctl_ha_role_sysctl, "I", "HA role for this head");
1913
1914         if (softc->is_single == 0) {
1915                 ctl_frontend_register(&ha_frontend);
1916                 if (ctl_ha_msg_init(softc) != CTL_HA_STATUS_SUCCESS) {
1917                         printf("ctl_init: ctl_ha_msg_init failed.\n");
1918                         softc->is_single = 1;
1919                 } else
1920                 if (ctl_ha_msg_register(CTL_HA_CHAN_CTL, ctl_isc_event_handler)
1921                     != CTL_HA_STATUS_SUCCESS) {
1922                         printf("ctl_init: ctl_ha_msg_register failed.\n");
1923                         softc->is_single = 1;
1924                 }
1925         }
1926         return (0);
1927 }
1928
1929 void
1930 ctl_shutdown(void)
1931 {
1932         struct ctl_softc *softc = control_softc;
1933         struct ctl_lun *lun, *next_lun;
1934
1935         if (softc->is_single == 0) {
1936                 ctl_ha_msg_shutdown(softc);
1937                 if (ctl_ha_msg_deregister(CTL_HA_CHAN_CTL)
1938                     != CTL_HA_STATUS_SUCCESS)
1939                         printf("%s: ctl_ha_msg_deregister failed.\n", __func__);
1940                 if (ctl_ha_msg_destroy(softc) != CTL_HA_STATUS_SUCCESS)
1941                         printf("%s: ctl_ha_msg_destroy failed.\n", __func__);
1942                 ctl_frontend_deregister(&ha_frontend);
1943         }
1944
1945         mtx_lock(&softc->ctl_lock);
1946
1947         STAILQ_FOREACH_SAFE(lun, &softc->lun_list, links, next_lun)
1948                 ctl_free_lun(lun);
1949
1950         mtx_unlock(&softc->ctl_lock);
1951
1952 #if 0
1953         ctl_shutdown_thread(softc->work_thread);
1954         mtx_destroy(&softc->queue_lock);
1955 #endif
1956
1957         ctl_tpc_shutdown(softc);
1958         uma_zdestroy(softc->io_zone);
1959         mtx_destroy(&softc->ctl_lock);
1960
1961         destroy_dev(softc->dev);
1962
1963         sysctl_ctx_free(&softc->sysctl_ctx);
1964
1965         free(softc, M_DEVBUF);
1966         control_softc = NULL;
1967 }
1968
1969 static int
1970 ctl_module_event_handler(module_t mod, int what, void *arg)
1971 {
1972
1973         switch (what) {
1974         case MOD_LOAD:
1975                 return (ctl_init());
1976         case MOD_UNLOAD:
1977                 return (EBUSY);
1978         default:
1979                 return (EOPNOTSUPP);
1980         }
1981 }
1982
1983 /*
1984  * XXX KDM should we do some access checks here?  Bump a reference count to
1985  * prevent a CTL module from being unloaded while someone has it open?
1986  */
1987 static int
1988 ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td)
1989 {
1990         return (0);
1991 }
1992
1993 static int
1994 ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td)
1995 {
1996         return (0);
1997 }
1998
1999 /*
2000  * Remove an initiator by port number and initiator ID.
2001  * Returns 0 for success, -1 for failure.
2002  */
2003 int
2004 ctl_remove_initiator(struct ctl_port *port, int iid)
2005 {
2006         struct ctl_softc *softc = port->ctl_softc;
2007
2008         mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
2009
2010         if (iid > CTL_MAX_INIT_PER_PORT) {
2011                 printf("%s: initiator ID %u > maximun %u!\n",
2012                        __func__, iid, CTL_MAX_INIT_PER_PORT);
2013                 return (-1);
2014         }
2015
2016         mtx_lock(&softc->ctl_lock);
2017         port->wwpn_iid[iid].in_use--;
2018         port->wwpn_iid[iid].last_use = time_uptime;
2019         mtx_unlock(&softc->ctl_lock);
2020         ctl_isc_announce_iid(port, iid);
2021
2022         return (0);
2023 }
2024
2025 /*
2026  * Add an initiator to the initiator map.
2027  * Returns iid for success, < 0 for failure.
2028  */
2029 int
2030 ctl_add_initiator(struct ctl_port *port, int iid, uint64_t wwpn, char *name)
2031 {
2032         struct ctl_softc *softc = port->ctl_softc;
2033         time_t best_time;
2034         int i, best;
2035
2036         mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
2037
2038         if (iid >= CTL_MAX_INIT_PER_PORT) {
2039                 printf("%s: WWPN %#jx initiator ID %u > maximum %u!\n",
2040                        __func__, wwpn, iid, CTL_MAX_INIT_PER_PORT);
2041                 free(name, M_CTL);
2042                 return (-1);
2043         }
2044
2045         mtx_lock(&softc->ctl_lock);
2046
2047         if (iid < 0 && (wwpn != 0 || name != NULL)) {
2048                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
2049                         if (wwpn != 0 && wwpn == port->wwpn_iid[i].wwpn) {
2050                                 iid = i;
2051                                 break;
2052                         }
2053                         if (name != NULL && port->wwpn_iid[i].name != NULL &&
2054                             strcmp(name, port->wwpn_iid[i].name) == 0) {
2055                                 iid = i;
2056                                 break;
2057                         }
2058                 }
2059         }
2060
2061         if (iid < 0) {
2062                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
2063                         if (port->wwpn_iid[i].in_use == 0 &&
2064                             port->wwpn_iid[i].wwpn == 0 &&
2065                             port->wwpn_iid[i].name == NULL) {
2066                                 iid = i;
2067                                 break;
2068                         }
2069                 }
2070         }
2071
2072         if (iid < 0) {
2073                 best = -1;
2074                 best_time = INT32_MAX;
2075                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
2076                         if (port->wwpn_iid[i].in_use == 0) {
2077                                 if (port->wwpn_iid[i].last_use < best_time) {
2078                                         best = i;
2079                                         best_time = port->wwpn_iid[i].last_use;
2080                                 }
2081                         }
2082                 }
2083                 iid = best;
2084         }
2085
2086         if (iid < 0) {
2087                 mtx_unlock(&softc->ctl_lock);
2088                 free(name, M_CTL);
2089                 return (-2);
2090         }
2091
2092         if (port->wwpn_iid[iid].in_use > 0 && (wwpn != 0 || name != NULL)) {
2093                 /*
2094                  * This is not an error yet.
2095                  */
2096                 if (wwpn != 0 && wwpn == port->wwpn_iid[iid].wwpn) {
2097 #if 0
2098                         printf("%s: port %d iid %u WWPN %#jx arrived"
2099                             " again\n", __func__, port->targ_port,
2100                             iid, (uintmax_t)wwpn);
2101 #endif
2102                         goto take;
2103                 }
2104                 if (name != NULL && port->wwpn_iid[iid].name != NULL &&
2105                     strcmp(name, port->wwpn_iid[iid].name) == 0) {
2106 #if 0
2107                         printf("%s: port %d iid %u name '%s' arrived"
2108                             " again\n", __func__, port->targ_port,
2109                             iid, name);
2110 #endif
2111                         goto take;
2112                 }
2113
2114                 /*
2115                  * This is an error, but what do we do about it?  The
2116                  * driver is telling us we have a new WWPN for this
2117                  * initiator ID, so we pretty much need to use it.
2118                  */
2119                 printf("%s: port %d iid %u WWPN %#jx '%s' arrived,"
2120                     " but WWPN %#jx '%s' is still at that address\n",
2121                     __func__, port->targ_port, iid, wwpn, name,
2122                     (uintmax_t)port->wwpn_iid[iid].wwpn,
2123                     port->wwpn_iid[iid].name);
2124
2125                 /*
2126                  * XXX KDM clear have_ca and ua_pending on each LUN for
2127                  * this initiator.
2128                  */
2129         }
2130 take:
2131         free(port->wwpn_iid[iid].name, M_CTL);
2132         port->wwpn_iid[iid].name = name;
2133         port->wwpn_iid[iid].wwpn = wwpn;
2134         port->wwpn_iid[iid].in_use++;
2135         mtx_unlock(&softc->ctl_lock);
2136         ctl_isc_announce_iid(port, iid);
2137
2138         return (iid);
2139 }
2140
2141 static int
2142 ctl_create_iid(struct ctl_port *port, int iid, uint8_t *buf)
2143 {
2144         int len;
2145
2146         switch (port->port_type) {
2147         case CTL_PORT_FC:
2148         {
2149                 struct scsi_transportid_fcp *id =
2150                     (struct scsi_transportid_fcp *)buf;
2151                 if (port->wwpn_iid[iid].wwpn == 0)
2152                         return (0);
2153                 memset(id, 0, sizeof(*id));
2154                 id->format_protocol = SCSI_PROTO_FC;
2155                 scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->n_port_name);
2156                 return (sizeof(*id));
2157         }
2158         case CTL_PORT_ISCSI:
2159         {
2160                 struct scsi_transportid_iscsi_port *id =
2161                     (struct scsi_transportid_iscsi_port *)buf;
2162                 if (port->wwpn_iid[iid].name == NULL)
2163                         return (0);
2164                 memset(id, 0, 256);
2165                 id->format_protocol = SCSI_TRN_ISCSI_FORMAT_PORT |
2166                     SCSI_PROTO_ISCSI;
2167                 len = strlcpy(id->iscsi_name, port->wwpn_iid[iid].name, 252) + 1;
2168                 len = roundup2(min(len, 252), 4);
2169                 scsi_ulto2b(len, id->additional_length);
2170                 return (sizeof(*id) + len);
2171         }
2172         case CTL_PORT_SAS:
2173         {
2174                 struct scsi_transportid_sas *id =
2175                     (struct scsi_transportid_sas *)buf;
2176                 if (port->wwpn_iid[iid].wwpn == 0)
2177                         return (0);
2178                 memset(id, 0, sizeof(*id));
2179                 id->format_protocol = SCSI_PROTO_SAS;
2180                 scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->sas_address);
2181                 return (sizeof(*id));
2182         }
2183         default:
2184         {
2185                 struct scsi_transportid_spi *id =
2186                     (struct scsi_transportid_spi *)buf;
2187                 memset(id, 0, sizeof(*id));
2188                 id->format_protocol = SCSI_PROTO_SPI;
2189                 scsi_ulto2b(iid, id->scsi_addr);
2190                 scsi_ulto2b(port->targ_port, id->rel_trgt_port_id);
2191                 return (sizeof(*id));
2192         }
2193         }
2194 }
2195
2196 /*
2197  * Serialize a command that went down the "wrong" side, and so was sent to
2198  * this controller for execution.  The logic is a little different than the
2199  * standard case in ctl_scsiio_precheck().  Errors in this case need to get
2200  * sent back to the other side, but in the success case, we execute the
2201  * command on this side (XFER mode) or tell the other side to execute it
2202  * (SER_ONLY mode).
2203  */
2204 static void
2205 ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio)
2206 {
2207         struct ctl_softc *softc = CTL_SOFTC(ctsio);
2208         struct ctl_port *port = CTL_PORT(ctsio);
2209         union ctl_ha_msg msg_info;
2210         struct ctl_lun *lun;
2211         const struct ctl_cmd_entry *entry;
2212         uint32_t targ_lun;
2213
2214         targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
2215
2216         /* Make sure that we know about this port. */
2217         if (port == NULL || (port->status & CTL_PORT_STATUS_ONLINE) == 0) {
2218                 ctl_set_internal_failure(ctsio, /*sks_valid*/ 0,
2219                                          /*retry_count*/ 1);
2220                 goto badjuju;
2221         }
2222
2223         /* Make sure that we know about this LUN. */
2224         mtx_lock(&softc->ctl_lock);
2225         if (targ_lun >= CTL_MAX_LUNS ||
2226             (lun = softc->ctl_luns[targ_lun]) == NULL) {
2227                 mtx_unlock(&softc->ctl_lock);
2228
2229                 /*
2230                  * The other node would not send this request to us unless
2231                  * received announce that we are primary node for this LUN.
2232                  * If this LUN does not exist now, it is probably result of
2233                  * a race, so respond to initiator in the most opaque way.
2234                  */
2235                 ctl_set_busy(ctsio);
2236                 goto badjuju;
2237         }
2238         mtx_lock(&lun->lun_lock);
2239         mtx_unlock(&softc->ctl_lock);
2240
2241         /*
2242          * If the LUN is invalid, pretend that it doesn't exist.
2243          * It will go away as soon as all pending I/Os completed.
2244          */
2245         if (lun->flags & CTL_LUN_DISABLED) {
2246                 mtx_unlock(&lun->lun_lock);
2247                 ctl_set_busy(ctsio);
2248                 goto badjuju;
2249         }
2250
2251         entry = ctl_get_cmd_entry(ctsio, NULL);
2252         if (ctl_scsiio_lun_check(lun, entry, ctsio) != 0) {
2253                 mtx_unlock(&lun->lun_lock);
2254                 goto badjuju;
2255         }
2256
2257         CTL_LUN(ctsio) = lun;
2258         CTL_BACKEND_LUN(ctsio) = lun->be_lun;
2259
2260         /*
2261          * Every I/O goes into the OOA queue for a
2262          * particular LUN, and stays there until completion.
2263          */
2264 #ifdef CTL_TIME_IO
2265         if (TAILQ_EMPTY(&lun->ooa_queue))
2266                 lun->idle_time += getsbinuptime() - lun->last_busy;
2267 #endif
2268         TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2269
2270         switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
2271                 (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr, ctl_ooaq,
2272                  ooa_links))) {
2273         case CTL_ACTION_BLOCK:
2274                 ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
2275                 TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
2276                                   blocked_links);
2277                 mtx_unlock(&lun->lun_lock);
2278                 break;
2279         case CTL_ACTION_PASS:
2280         case CTL_ACTION_SKIP:
2281                 if (softc->ha_mode == CTL_HA_MODE_XFER) {
2282                         ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
2283                         ctl_enqueue_rtr((union ctl_io *)ctsio);
2284                         mtx_unlock(&lun->lun_lock);
2285                 } else {
2286                         ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
2287                         mtx_unlock(&lun->lun_lock);
2288
2289                         /* send msg back to other side */
2290                         msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
2291                         msg_info.hdr.serializing_sc = (union ctl_io *)ctsio;
2292                         msg_info.hdr.msg_type = CTL_MSG_R2R;
2293                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
2294                             sizeof(msg_info.hdr), M_WAITOK);
2295                 }
2296                 break;
2297         case CTL_ACTION_OVERLAP:
2298                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2299                 mtx_unlock(&lun->lun_lock);
2300                 ctl_set_overlapped_cmd(ctsio);
2301                 goto badjuju;
2302         case CTL_ACTION_OVERLAP_TAG:
2303                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2304                 mtx_unlock(&lun->lun_lock);
2305                 ctl_set_overlapped_tag(ctsio, ctsio->tag_num);
2306                 goto badjuju;
2307         case CTL_ACTION_ERROR:
2308         default:
2309                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2310                 mtx_unlock(&lun->lun_lock);
2311
2312                 ctl_set_internal_failure(ctsio, /*sks_valid*/ 0,
2313                                          /*retry_count*/ 0);
2314 badjuju:
2315                 ctl_copy_sense_data_back((union ctl_io *)ctsio, &msg_info);
2316                 msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
2317                 msg_info.hdr.serializing_sc = NULL;
2318                 msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
2319                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
2320                     sizeof(msg_info.scsi), M_WAITOK);
2321                 ctl_free_io((union ctl_io *)ctsio);
2322                 break;
2323         }
2324 }
2325
2326 /*
2327  * Returns 0 for success, errno for failure.
2328  */
2329 static void
2330 ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
2331                    struct ctl_ooa *ooa_hdr, struct ctl_ooa_entry *kern_entries)
2332 {
2333         union ctl_io *io;
2334
2335         mtx_lock(&lun->lun_lock);
2336         for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); (io != NULL);
2337              (*cur_fill_num)++, io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2338              ooa_links)) {
2339                 struct ctl_ooa_entry *entry;
2340
2341                 /*
2342                  * If we've got more than we can fit, just count the
2343                  * remaining entries.
2344                  */
2345                 if (*cur_fill_num >= ooa_hdr->alloc_num)
2346                         continue;
2347
2348                 entry = &kern_entries[*cur_fill_num];
2349
2350                 entry->tag_num = io->scsiio.tag_num;
2351                 entry->lun_num = lun->lun;
2352 #ifdef CTL_TIME_IO
2353                 entry->start_bt = io->io_hdr.start_bt;
2354 #endif
2355                 bcopy(io->scsiio.cdb, entry->cdb, io->scsiio.cdb_len);
2356                 entry->cdb_len = io->scsiio.cdb_len;
2357                 if (io->io_hdr.flags & CTL_FLAG_BLOCKED)
2358                         entry->cmd_flags |= CTL_OOACMD_FLAG_BLOCKED;
2359
2360                 if (io->io_hdr.flags & CTL_FLAG_DMA_INPROG)
2361                         entry->cmd_flags |= CTL_OOACMD_FLAG_DMA;
2362
2363                 if (io->io_hdr.flags & CTL_FLAG_ABORT)
2364                         entry->cmd_flags |= CTL_OOACMD_FLAG_ABORT;
2365
2366                 if (io->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR)
2367                         entry->cmd_flags |= CTL_OOACMD_FLAG_RTR;
2368
2369                 if (io->io_hdr.flags & CTL_FLAG_DMA_QUEUED)
2370                         entry->cmd_flags |= CTL_OOACMD_FLAG_DMA_QUEUED;
2371         }
2372         mtx_unlock(&lun->lun_lock);
2373 }
2374
2375 static void *
2376 ctl_copyin_alloc(void *user_addr, unsigned int len, char *error_str,
2377                  size_t error_str_len)
2378 {
2379         void *kptr;
2380
2381         kptr = malloc(len, M_CTL, M_WAITOK | M_ZERO);
2382
2383         if (copyin(user_addr, kptr, len) != 0) {
2384                 snprintf(error_str, error_str_len, "Error copying %d bytes "
2385                          "from user address %p to kernel address %p", len,
2386                          user_addr, kptr);
2387                 free(kptr, M_CTL);
2388                 return (NULL);
2389         }
2390
2391         return (kptr);
2392 }
2393
2394 static void
2395 ctl_free_args(int num_args, struct ctl_be_arg *args)
2396 {
2397         int i;
2398
2399         if (args == NULL)
2400                 return;
2401
2402         for (i = 0; i < num_args; i++) {
2403                 free(args[i].kname, M_CTL);
2404                 free(args[i].kvalue, M_CTL);
2405         }
2406
2407         free(args, M_CTL);
2408 }
2409
2410 static struct ctl_be_arg *
2411 ctl_copyin_args(int num_args, struct ctl_be_arg *uargs,
2412                 char *error_str, size_t error_str_len)
2413 {
2414         struct ctl_be_arg *args;
2415         int i;
2416
2417         args = ctl_copyin_alloc(uargs, num_args * sizeof(*args),
2418                                 error_str, error_str_len);
2419
2420         if (args == NULL)
2421                 goto bailout;
2422
2423         for (i = 0; i < num_args; i++) {
2424                 args[i].kname = NULL;
2425                 args[i].kvalue = NULL;
2426         }
2427
2428         for (i = 0; i < num_args; i++) {
2429                 uint8_t *tmpptr;
2430
2431                 if (args[i].namelen == 0) {
2432                         snprintf(error_str, error_str_len, "Argument %d "
2433                                  "name length is zero", i);
2434                         goto bailout;
2435                 }
2436
2437                 args[i].kname = ctl_copyin_alloc(args[i].name,
2438                         args[i].namelen, error_str, error_str_len);
2439                 if (args[i].kname == NULL)
2440                         goto bailout;
2441
2442                 if (args[i].kname[args[i].namelen - 1] != '\0') {
2443                         snprintf(error_str, error_str_len, "Argument %d "
2444                                  "name is not NUL-terminated", i);
2445                         goto bailout;
2446                 }
2447
2448                 if (args[i].flags & CTL_BEARG_RD) {
2449                         if (args[i].vallen == 0) {
2450                                 snprintf(error_str, error_str_len, "Argument %d "
2451                                          "value length is zero", i);
2452                                 goto bailout;
2453                         }
2454
2455                         tmpptr = ctl_copyin_alloc(args[i].value,
2456                                 args[i].vallen, error_str, error_str_len);
2457                         if (tmpptr == NULL)
2458                                 goto bailout;
2459
2460                         if ((args[i].flags & CTL_BEARG_ASCII)
2461                          && (tmpptr[args[i].vallen - 1] != '\0')) {
2462                                 snprintf(error_str, error_str_len, "Argument "
2463                                     "%d value is not NUL-terminated", i);
2464                                 free(tmpptr, M_CTL);
2465                                 goto bailout;
2466                         }
2467                         args[i].kvalue = tmpptr;
2468                 } else {
2469                         args[i].kvalue = malloc(args[i].vallen,
2470                             M_CTL, M_WAITOK | M_ZERO);
2471                 }
2472         }
2473
2474         return (args);
2475 bailout:
2476
2477         ctl_free_args(num_args, args);
2478
2479         return (NULL);
2480 }
2481
2482 static void
2483 ctl_copyout_args(int num_args, struct ctl_be_arg *args)
2484 {
2485         int i;
2486
2487         for (i = 0; i < num_args; i++) {
2488                 if (args[i].flags & CTL_BEARG_WR)
2489                         copyout(args[i].kvalue, args[i].value, args[i].vallen);
2490         }
2491 }
2492
2493 /*
2494  * Escape characters that are illegal or not recommended in XML.
2495  */
2496 int
2497 ctl_sbuf_printf_esc(struct sbuf *sb, char *str, int size)
2498 {
2499         char *end = str + size;
2500         int retval;
2501
2502         retval = 0;
2503
2504         for (; *str && str < end; str++) {
2505                 switch (*str) {
2506                 case '&':
2507                         retval = sbuf_printf(sb, "&amp;");
2508                         break;
2509                 case '>':
2510                         retval = sbuf_printf(sb, "&gt;");
2511                         break;
2512                 case '<':
2513                         retval = sbuf_printf(sb, "&lt;");
2514                         break;
2515                 default:
2516                         retval = sbuf_putc(sb, *str);
2517                         break;
2518                 }
2519
2520                 if (retval != 0)
2521                         break;
2522
2523         }
2524
2525         return (retval);
2526 }
2527
2528 static void
2529 ctl_id_sbuf(struct ctl_devid *id, struct sbuf *sb)
2530 {
2531         struct scsi_vpd_id_descriptor *desc;
2532         int i;
2533
2534         if (id == NULL || id->len < 4)
2535                 return;
2536         desc = (struct scsi_vpd_id_descriptor *)id->data;
2537         switch (desc->id_type & SVPD_ID_TYPE_MASK) {
2538         case SVPD_ID_TYPE_T10:
2539                 sbuf_printf(sb, "t10.");
2540                 break;
2541         case SVPD_ID_TYPE_EUI64:
2542                 sbuf_printf(sb, "eui.");
2543                 break;
2544         case SVPD_ID_TYPE_NAA:
2545                 sbuf_printf(sb, "naa.");
2546                 break;
2547         case SVPD_ID_TYPE_SCSI_NAME:
2548                 break;
2549         }
2550         switch (desc->proto_codeset & SVPD_ID_CODESET_MASK) {
2551         case SVPD_ID_CODESET_BINARY:
2552                 for (i = 0; i < desc->length; i++)
2553                         sbuf_printf(sb, "%02x", desc->identifier[i]);
2554                 break;
2555         case SVPD_ID_CODESET_ASCII:
2556                 sbuf_printf(sb, "%.*s", (int)desc->length,
2557                     (char *)desc->identifier);
2558                 break;
2559         case SVPD_ID_CODESET_UTF8:
2560                 sbuf_printf(sb, "%s", (char *)desc->identifier);
2561                 break;
2562         }
2563 }
2564
2565 static int
2566 ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
2567           struct thread *td)
2568 {
2569         struct ctl_softc *softc = dev->si_drv1;
2570         struct ctl_port *port;
2571         struct ctl_lun *lun;
2572         int retval;
2573
2574         retval = 0;
2575
2576         switch (cmd) {
2577         case CTL_IO:
2578                 retval = ctl_ioctl_io(dev, cmd, addr, flag, td);
2579                 break;
2580         case CTL_ENABLE_PORT:
2581         case CTL_DISABLE_PORT:
2582         case CTL_SET_PORT_WWNS: {
2583                 struct ctl_port *port;
2584                 struct ctl_port_entry *entry;
2585
2586                 entry = (struct ctl_port_entry *)addr;
2587
2588                 mtx_lock(&softc->ctl_lock);
2589                 STAILQ_FOREACH(port, &softc->port_list, links) {
2590                         int action, done;
2591
2592                         if (port->targ_port < softc->port_min ||
2593                             port->targ_port >= softc->port_max)
2594                                 continue;
2595
2596                         action = 0;
2597                         done = 0;
2598                         if ((entry->port_type == CTL_PORT_NONE)
2599                          && (entry->targ_port == port->targ_port)) {
2600                                 /*
2601                                  * If the user only wants to enable or
2602                                  * disable or set WWNs on a specific port,
2603                                  * do the operation and we're done.
2604                                  */
2605                                 action = 1;
2606                                 done = 1;
2607                         } else if (entry->port_type & port->port_type) {
2608                                 /*
2609                                  * Compare the user's type mask with the
2610                                  * particular frontend type to see if we
2611                                  * have a match.
2612                                  */
2613                                 action = 1;
2614                                 done = 0;
2615
2616                                 /*
2617                                  * Make sure the user isn't trying to set
2618                                  * WWNs on multiple ports at the same time.
2619                                  */
2620                                 if (cmd == CTL_SET_PORT_WWNS) {
2621                                         printf("%s: Can't set WWNs on "
2622                                                "multiple ports\n", __func__);
2623                                         retval = EINVAL;
2624                                         break;
2625                                 }
2626                         }
2627                         if (action == 0)
2628                                 continue;
2629
2630                         /*
2631                          * XXX KDM we have to drop the lock here, because
2632                          * the online/offline operations can potentially
2633                          * block.  We need to reference count the frontends
2634                          * so they can't go away,
2635                          */
2636                         if (cmd == CTL_ENABLE_PORT) {
2637                                 mtx_unlock(&softc->ctl_lock);
2638                                 ctl_port_online(port);
2639                                 mtx_lock(&softc->ctl_lock);
2640                         } else if (cmd == CTL_DISABLE_PORT) {
2641                                 mtx_unlock(&softc->ctl_lock);
2642                                 ctl_port_offline(port);
2643                                 mtx_lock(&softc->ctl_lock);
2644                         } else if (cmd == CTL_SET_PORT_WWNS) {
2645                                 ctl_port_set_wwns(port,
2646                                     (entry->flags & CTL_PORT_WWNN_VALID) ?
2647                                     1 : 0, entry->wwnn,
2648                                     (entry->flags & CTL_PORT_WWPN_VALID) ?
2649                                     1 : 0, entry->wwpn);
2650                         }
2651                         if (done != 0)
2652                                 break;
2653                 }
2654                 mtx_unlock(&softc->ctl_lock);
2655                 break;
2656         }
2657         case CTL_GET_OOA: {
2658                 struct ctl_ooa *ooa_hdr;
2659                 struct ctl_ooa_entry *entries;
2660                 uint32_t cur_fill_num;
2661
2662                 ooa_hdr = (struct ctl_ooa *)addr;
2663
2664                 if ((ooa_hdr->alloc_len == 0)
2665                  || (ooa_hdr->alloc_num == 0)) {
2666                         printf("%s: CTL_GET_OOA: alloc len %u and alloc num %u "
2667                                "must be non-zero\n", __func__,
2668                                ooa_hdr->alloc_len, ooa_hdr->alloc_num);
2669                         retval = EINVAL;
2670                         break;
2671                 }
2672
2673                 if (ooa_hdr->alloc_len != (ooa_hdr->alloc_num *
2674                     sizeof(struct ctl_ooa_entry))) {
2675                         printf("%s: CTL_GET_OOA: alloc len %u must be alloc "
2676                                "num %d * sizeof(struct ctl_ooa_entry) %zd\n",
2677                                __func__, ooa_hdr->alloc_len,
2678                                ooa_hdr->alloc_num,sizeof(struct ctl_ooa_entry));
2679                         retval = EINVAL;
2680                         break;
2681                 }
2682
2683                 entries = malloc(ooa_hdr->alloc_len, M_CTL, M_WAITOK | M_ZERO);
2684                 if (entries == NULL) {
2685                         printf("%s: could not allocate %d bytes for OOA "
2686                                "dump\n", __func__, ooa_hdr->alloc_len);
2687                         retval = ENOMEM;
2688                         break;
2689                 }
2690
2691                 mtx_lock(&softc->ctl_lock);
2692                 if ((ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) == 0 &&
2693                     (ooa_hdr->lun_num >= CTL_MAX_LUNS ||
2694                      softc->ctl_luns[ooa_hdr->lun_num] == NULL)) {
2695                         mtx_unlock(&softc->ctl_lock);
2696                         free(entries, M_CTL);
2697                         printf("%s: CTL_GET_OOA: invalid LUN %ju\n",
2698                                __func__, (uintmax_t)ooa_hdr->lun_num);
2699                         retval = EINVAL;
2700                         break;
2701                 }
2702
2703                 cur_fill_num = 0;
2704
2705                 if (ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) {
2706                         STAILQ_FOREACH(lun, &softc->lun_list, links) {
2707                                 ctl_ioctl_fill_ooa(lun, &cur_fill_num,
2708                                     ooa_hdr, entries);
2709                         }
2710                 } else {
2711                         lun = softc->ctl_luns[ooa_hdr->lun_num];
2712                         ctl_ioctl_fill_ooa(lun, &cur_fill_num, ooa_hdr,
2713                             entries);
2714                 }
2715                 mtx_unlock(&softc->ctl_lock);
2716
2717                 ooa_hdr->fill_num = min(cur_fill_num, ooa_hdr->alloc_num);
2718                 ooa_hdr->fill_len = ooa_hdr->fill_num *
2719                         sizeof(struct ctl_ooa_entry);
2720                 retval = copyout(entries, ooa_hdr->entries, ooa_hdr->fill_len);
2721                 if (retval != 0) {
2722                         printf("%s: error copying out %d bytes for OOA dump\n",
2723                                __func__, ooa_hdr->fill_len);
2724                 }
2725
2726                 getbinuptime(&ooa_hdr->cur_bt);
2727
2728                 if (cur_fill_num > ooa_hdr->alloc_num) {
2729                         ooa_hdr->dropped_num = cur_fill_num -ooa_hdr->alloc_num;
2730                         ooa_hdr->status = CTL_OOA_NEED_MORE_SPACE;
2731                 } else {
2732                         ooa_hdr->dropped_num = 0;
2733                         ooa_hdr->status = CTL_OOA_OK;
2734                 }
2735
2736                 free(entries, M_CTL);
2737                 break;
2738         }
2739         case CTL_DELAY_IO: {
2740                 struct ctl_io_delay_info *delay_info;
2741
2742                 delay_info = (struct ctl_io_delay_info *)addr;
2743
2744 #ifdef CTL_IO_DELAY
2745                 mtx_lock(&softc->ctl_lock);
2746                 if (delay_info->lun_id >= CTL_MAX_LUNS ||
2747                     (lun = softc->ctl_luns[delay_info->lun_id]) == NULL) {
2748                         mtx_unlock(&softc->ctl_lock);
2749                         delay_info->status = CTL_DELAY_STATUS_INVALID_LUN;
2750                         break;
2751                 }
2752                 mtx_lock(&lun->lun_lock);
2753                 mtx_unlock(&softc->ctl_lock);
2754                 delay_info->status = CTL_DELAY_STATUS_OK;
2755                 switch (delay_info->delay_type) {
2756                 case CTL_DELAY_TYPE_CONT:
2757                 case CTL_DELAY_TYPE_ONESHOT:
2758                         break;
2759                 default:
2760                         delay_info->status = CTL_DELAY_STATUS_INVALID_TYPE;
2761                         break;
2762                 }
2763                 switch (delay_info->delay_loc) {
2764                 case CTL_DELAY_LOC_DATAMOVE:
2765                         lun->delay_info.datamove_type = delay_info->delay_type;
2766                         lun->delay_info.datamove_delay = delay_info->delay_secs;
2767                         break;
2768                 case CTL_DELAY_LOC_DONE:
2769                         lun->delay_info.done_type = delay_info->delay_type;
2770                         lun->delay_info.done_delay = delay_info->delay_secs;
2771                         break;
2772                 default:
2773                         delay_info->status = CTL_DELAY_STATUS_INVALID_LOC;
2774                         break;
2775                 }
2776                 mtx_unlock(&lun->lun_lock);
2777 #else
2778                 delay_info->status = CTL_DELAY_STATUS_NOT_IMPLEMENTED;
2779 #endif /* CTL_IO_DELAY */
2780                 break;
2781         }
2782 #ifdef CTL_LEGACY_STATS
2783         case CTL_GETSTATS: {
2784                 struct ctl_stats *stats = (struct ctl_stats *)addr;
2785                 int i;
2786
2787                 /*
2788                  * XXX KDM no locking here.  If the LUN list changes,
2789                  * things can blow up.
2790                  */
2791                 i = 0;
2792                 stats->status = CTL_SS_OK;
2793                 stats->fill_len = 0;
2794                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
2795                         if (stats->fill_len + sizeof(lun->legacy_stats) >
2796                             stats->alloc_len) {
2797                                 stats->status = CTL_SS_NEED_MORE_SPACE;
2798                                 break;
2799                         }
2800                         retval = copyout(&lun->legacy_stats, &stats->lun_stats[i++],
2801                                          sizeof(lun->legacy_stats));
2802                         if (retval != 0)
2803                                 break;
2804                         stats->fill_len += sizeof(lun->legacy_stats);
2805                 }
2806                 stats->num_luns = softc->num_luns;
2807                 stats->flags = CTL_STATS_FLAG_NONE;
2808 #ifdef CTL_TIME_IO
2809                 stats->flags |= CTL_STATS_FLAG_TIME_VALID;
2810 #endif
2811                 getnanouptime(&stats->timestamp);
2812                 break;
2813         }
2814 #endif /* CTL_LEGACY_STATS */
2815         case CTL_ERROR_INJECT: {
2816                 struct ctl_error_desc *err_desc, *new_err_desc;
2817
2818                 err_desc = (struct ctl_error_desc *)addr;
2819
2820                 new_err_desc = malloc(sizeof(*new_err_desc), M_CTL,
2821                                       M_WAITOK | M_ZERO);
2822                 bcopy(err_desc, new_err_desc, sizeof(*new_err_desc));
2823
2824                 mtx_lock(&softc->ctl_lock);
2825                 if (err_desc->lun_id >= CTL_MAX_LUNS ||
2826                     (lun = softc->ctl_luns[err_desc->lun_id]) == NULL) {
2827                         mtx_unlock(&softc->ctl_lock);
2828                         free(new_err_desc, M_CTL);
2829                         printf("%s: CTL_ERROR_INJECT: invalid LUN %ju\n",
2830                                __func__, (uintmax_t)err_desc->lun_id);
2831                         retval = EINVAL;
2832                         break;
2833                 }
2834                 mtx_lock(&lun->lun_lock);
2835                 mtx_unlock(&softc->ctl_lock);
2836
2837                 /*
2838                  * We could do some checking here to verify the validity
2839                  * of the request, but given the complexity of error
2840                  * injection requests, the checking logic would be fairly
2841                  * complex.
2842                  *
2843                  * For now, if the request is invalid, it just won't get
2844                  * executed and might get deleted.
2845                  */
2846                 STAILQ_INSERT_TAIL(&lun->error_list, new_err_desc, links);
2847
2848                 /*
2849                  * XXX KDM check to make sure the serial number is unique,
2850                  * in case we somehow manage to wrap.  That shouldn't
2851                  * happen for a very long time, but it's the right thing to
2852                  * do.
2853                  */
2854                 new_err_desc->serial = lun->error_serial;
2855                 err_desc->serial = lun->error_serial;
2856                 lun->error_serial++;
2857
2858                 mtx_unlock(&lun->lun_lock);
2859                 break;
2860         }
2861         case CTL_ERROR_INJECT_DELETE: {
2862                 struct ctl_error_desc *delete_desc, *desc, *desc2;
2863                 int delete_done;
2864
2865                 delete_desc = (struct ctl_error_desc *)addr;
2866                 delete_done = 0;
2867
2868                 mtx_lock(&softc->ctl_lock);
2869                 if (delete_desc->lun_id >= CTL_MAX_LUNS ||
2870                     (lun = softc->ctl_luns[delete_desc->lun_id]) == NULL) {
2871                         mtx_unlock(&softc->ctl_lock);
2872                         printf("%s: CTL_ERROR_INJECT_DELETE: invalid LUN %ju\n",
2873                                __func__, (uintmax_t)delete_desc->lun_id);
2874                         retval = EINVAL;
2875                         break;
2876                 }
2877                 mtx_lock(&lun->lun_lock);
2878                 mtx_unlock(&softc->ctl_lock);
2879                 STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
2880                         if (desc->serial != delete_desc->serial)
2881                                 continue;
2882
2883                         STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc,
2884                                       links);
2885                         free(desc, M_CTL);
2886                         delete_done = 1;
2887                 }
2888                 mtx_unlock(&lun->lun_lock);
2889                 if (delete_done == 0) {
2890                         printf("%s: CTL_ERROR_INJECT_DELETE: can't find "
2891                                "error serial %ju on LUN %u\n", __func__,
2892                                delete_desc->serial, delete_desc->lun_id);
2893                         retval = EINVAL;
2894                         break;
2895                 }
2896                 break;
2897         }
2898         case CTL_DUMP_STRUCTS: {
2899                 int j, k;
2900                 struct ctl_port *port;
2901                 struct ctl_frontend *fe;
2902
2903                 mtx_lock(&softc->ctl_lock);
2904                 printf("CTL Persistent Reservation information start:\n");
2905                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
2906                         mtx_lock(&lun->lun_lock);
2907                         if ((lun->flags & CTL_LUN_DISABLED) != 0) {
2908                                 mtx_unlock(&lun->lun_lock);
2909                                 continue;
2910                         }
2911
2912                         for (j = 0; j < CTL_MAX_PORTS; j++) {
2913                                 if (lun->pr_keys[j] == NULL)
2914                                         continue;
2915                                 for (k = 0; k < CTL_MAX_INIT_PER_PORT; k++){
2916                                         if (lun->pr_keys[j][k] == 0)
2917                                                 continue;
2918                                         printf("  LUN %ju port %d iid %d key "
2919                                                "%#jx\n", lun->lun, j, k,
2920                                                (uintmax_t)lun->pr_keys[j][k]);
2921                                 }
2922                         }
2923                         mtx_unlock(&lun->lun_lock);
2924                 }
2925                 printf("CTL Persistent Reservation information end\n");
2926                 printf("CTL Ports:\n");
2927                 STAILQ_FOREACH(port, &softc->port_list, links) {
2928                         printf("  Port %d '%s' Frontend '%s' Type %u pp %d vp %d WWNN "
2929                                "%#jx WWPN %#jx\n", port->targ_port, port->port_name,
2930                                port->frontend->name, port->port_type,
2931                                port->physical_port, port->virtual_port,
2932                                (uintmax_t)port->wwnn, (uintmax_t)port->wwpn);
2933                         for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
2934                                 if (port->wwpn_iid[j].in_use == 0 &&
2935                                     port->wwpn_iid[j].wwpn == 0 &&
2936                                     port->wwpn_iid[j].name == NULL)
2937                                         continue;
2938
2939                                 printf("    iid %u use %d WWPN %#jx '%s'\n",
2940                                     j, port->wwpn_iid[j].in_use,
2941                                     (uintmax_t)port->wwpn_iid[j].wwpn,
2942                                     port->wwpn_iid[j].name);
2943                         }
2944                 }
2945                 printf("CTL Port information end\n");
2946                 mtx_unlock(&softc->ctl_lock);
2947                 /*
2948                  * XXX KDM calling this without a lock.  We'd likely want
2949                  * to drop the lock before calling the frontend's dump
2950                  * routine anyway.
2951                  */
2952                 printf("CTL Frontends:\n");
2953                 STAILQ_FOREACH(fe, &softc->fe_list, links) {
2954                         printf("  Frontend '%s'\n", fe->name);
2955                         if (fe->fe_dump != NULL)
2956                                 fe->fe_dump();
2957                 }
2958                 printf("CTL Frontend information end\n");
2959                 break;
2960         }
2961         case CTL_LUN_REQ: {
2962                 struct ctl_lun_req *lun_req;
2963                 struct ctl_backend_driver *backend;
2964
2965                 lun_req = (struct ctl_lun_req *)addr;
2966
2967                 backend = ctl_backend_find(lun_req->backend);
2968                 if (backend == NULL) {
2969                         lun_req->status = CTL_LUN_ERROR;
2970                         snprintf(lun_req->error_str,
2971                                  sizeof(lun_req->error_str),
2972                                  "Backend \"%s\" not found.",
2973                                  lun_req->backend);
2974                         break;
2975                 }
2976                 if (lun_req->num_be_args > 0) {
2977                         lun_req->kern_be_args = ctl_copyin_args(
2978                                 lun_req->num_be_args,
2979                                 lun_req->be_args,
2980                                 lun_req->error_str,
2981                                 sizeof(lun_req->error_str));
2982                         if (lun_req->kern_be_args == NULL) {
2983                                 lun_req->status = CTL_LUN_ERROR;
2984                                 break;
2985                         }
2986                 }
2987
2988                 retval = backend->ioctl(dev, cmd, addr, flag, td);
2989
2990                 if (lun_req->num_be_args > 0) {
2991                         ctl_copyout_args(lun_req->num_be_args,
2992                                       lun_req->kern_be_args);
2993                         ctl_free_args(lun_req->num_be_args,
2994                                       lun_req->kern_be_args);
2995                 }
2996                 break;
2997         }
2998         case CTL_LUN_LIST: {
2999                 struct sbuf *sb;
3000                 struct ctl_lun_list *list;
3001                 struct ctl_option *opt;
3002
3003                 list = (struct ctl_lun_list *)addr;
3004
3005                 /*
3006                  * Allocate a fixed length sbuf here, based on the length
3007                  * of the user's buffer.  We could allocate an auto-extending
3008                  * buffer, and then tell the user how much larger our
3009                  * amount of data is than his buffer, but that presents
3010                  * some problems:
3011                  *
3012                  * 1.  The sbuf(9) routines use a blocking malloc, and so
3013                  *     we can't hold a lock while calling them with an
3014                  *     auto-extending buffer.
3015                  *
3016                  * 2.  There is not currently a LUN reference counting
3017                  *     mechanism, outside of outstanding transactions on
3018                  *     the LUN's OOA queue.  So a LUN could go away on us
3019                  *     while we're getting the LUN number, backend-specific
3020                  *     information, etc.  Thus, given the way things
3021                  *     currently work, we need to hold the CTL lock while
3022                  *     grabbing LUN information.
3023                  *
3024                  * So, from the user's standpoint, the best thing to do is
3025                  * allocate what he thinks is a reasonable buffer length,
3026                  * and then if he gets a CTL_LUN_LIST_NEED_MORE_SPACE error,
3027                  * double the buffer length and try again.  (And repeat
3028                  * that until he succeeds.)
3029                  */
3030                 sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3031                 if (sb == NULL) {
3032                         list->status = CTL_LUN_LIST_ERROR;
3033                         snprintf(list->error_str, sizeof(list->error_str),
3034                                  "Unable to allocate %d bytes for LUN list",
3035                                  list->alloc_len);
3036                         break;
3037                 }
3038
3039                 sbuf_printf(sb, "<ctllunlist>\n");
3040
3041                 mtx_lock(&softc->ctl_lock);
3042                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
3043                         mtx_lock(&lun->lun_lock);
3044                         retval = sbuf_printf(sb, "<lun id=\"%ju\">\n",
3045                                              (uintmax_t)lun->lun);
3046
3047                         /*
3048                          * Bail out as soon as we see that we've overfilled
3049                          * the buffer.
3050                          */
3051                         if (retval != 0)
3052                                 break;
3053
3054                         retval = sbuf_printf(sb, "\t<backend_type>%s"
3055                                              "</backend_type>\n",
3056                                              (lun->backend == NULL) ?  "none" :
3057                                              lun->backend->name);
3058
3059                         if (retval != 0)
3060                                 break;
3061
3062                         retval = sbuf_printf(sb, "\t<lun_type>%d</lun_type>\n",
3063                                              lun->be_lun->lun_type);
3064
3065                         if (retval != 0)
3066                                 break;
3067
3068                         if (lun->backend == NULL) {
3069                                 retval = sbuf_printf(sb, "</lun>\n");
3070                                 if (retval != 0)
3071                                         break;
3072                                 continue;
3073                         }
3074
3075                         retval = sbuf_printf(sb, "\t<size>%ju</size>\n",
3076                                              (lun->be_lun->maxlba > 0) ?
3077                                              lun->be_lun->maxlba + 1 : 0);
3078
3079                         if (retval != 0)
3080                                 break;
3081
3082                         retval = sbuf_printf(sb, "\t<blocksize>%u</blocksize>\n",
3083                                              lun->be_lun->blocksize);
3084
3085                         if (retval != 0)
3086                                 break;
3087
3088                         retval = sbuf_printf(sb, "\t<serial_number>");
3089
3090                         if (retval != 0)
3091                                 break;
3092
3093                         retval = ctl_sbuf_printf_esc(sb,
3094                             lun->be_lun->serial_num,
3095                             sizeof(lun->be_lun->serial_num));
3096
3097                         if (retval != 0)
3098                                 break;
3099
3100                         retval = sbuf_printf(sb, "</serial_number>\n");
3101
3102                         if (retval != 0)
3103                                 break;
3104
3105                         retval = sbuf_printf(sb, "\t<device_id>");
3106
3107                         if (retval != 0)
3108                                 break;
3109
3110                         retval = ctl_sbuf_printf_esc(sb,
3111                             lun->be_lun->device_id,
3112                             sizeof(lun->be_lun->device_id));
3113
3114                         if (retval != 0)
3115                                 break;
3116
3117                         retval = sbuf_printf(sb, "</device_id>\n");
3118
3119                         if (retval != 0)
3120                                 break;
3121
3122                         if (lun->backend->lun_info != NULL) {
3123                                 retval = lun->backend->lun_info(lun->be_lun->be_lun, sb);
3124                                 if (retval != 0)
3125                                         break;
3126                         }
3127                         STAILQ_FOREACH(opt, &lun->be_lun->options, links) {
3128                                 retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3129                                     opt->name, opt->value, opt->name);
3130                                 if (retval != 0)
3131                                         break;
3132                         }
3133
3134                         retval = sbuf_printf(sb, "</lun>\n");
3135
3136                         if (retval != 0)
3137                                 break;
3138                         mtx_unlock(&lun->lun_lock);
3139                 }
3140                 if (lun != NULL)
3141                         mtx_unlock(&lun->lun_lock);
3142                 mtx_unlock(&softc->ctl_lock);
3143
3144                 if ((retval != 0)
3145                  || ((retval = sbuf_printf(sb, "</ctllunlist>\n")) != 0)) {
3146                         retval = 0;
3147                         sbuf_delete(sb);
3148                         list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3149                         snprintf(list->error_str, sizeof(list->error_str),
3150                                  "Out of space, %d bytes is too small",
3151                                  list->alloc_len);
3152                         break;
3153                 }
3154
3155                 sbuf_finish(sb);
3156
3157                 retval = copyout(sbuf_data(sb), list->lun_xml,
3158                                  sbuf_len(sb) + 1);
3159
3160                 list->fill_len = sbuf_len(sb) + 1;
3161                 list->status = CTL_LUN_LIST_OK;
3162                 sbuf_delete(sb);
3163                 break;
3164         }
3165         case CTL_ISCSI: {
3166                 struct ctl_iscsi *ci;
3167                 struct ctl_frontend *fe;
3168
3169                 ci = (struct ctl_iscsi *)addr;
3170
3171                 fe = ctl_frontend_find("iscsi");
3172                 if (fe == NULL) {
3173                         ci->status = CTL_ISCSI_ERROR;
3174                         snprintf(ci->error_str, sizeof(ci->error_str),
3175                             "Frontend \"iscsi\" not found.");
3176                         break;
3177                 }
3178
3179                 retval = fe->ioctl(dev, cmd, addr, flag, td);
3180                 break;
3181         }
3182         case CTL_PORT_REQ: {
3183                 struct ctl_req *req;
3184                 struct ctl_frontend *fe;
3185
3186                 req = (struct ctl_req *)addr;
3187
3188                 fe = ctl_frontend_find(req->driver);
3189                 if (fe == NULL) {
3190                         req->status = CTL_LUN_ERROR;
3191                         snprintf(req->error_str, sizeof(req->error_str),
3192                             "Frontend \"%s\" not found.", req->driver);
3193                         break;
3194                 }
3195                 if (req->num_args > 0) {
3196                         req->kern_args = ctl_copyin_args(req->num_args,
3197                             req->args, req->error_str, sizeof(req->error_str));
3198                         if (req->kern_args == NULL) {
3199                                 req->status = CTL_LUN_ERROR;
3200                                 break;
3201                         }
3202                 }
3203
3204                 if (fe->ioctl)
3205                         retval = fe->ioctl(dev, cmd, addr, flag, td);
3206                 else
3207                         retval = ENODEV;
3208
3209                 if (req->num_args > 0) {
3210                         ctl_copyout_args(req->num_args, req->kern_args);
3211                         ctl_free_args(req->num_args, req->kern_args);
3212                 }
3213                 break;
3214         }
3215         case CTL_PORT_LIST: {
3216                 struct sbuf *sb;
3217                 struct ctl_port *port;
3218                 struct ctl_lun_list *list;
3219                 struct ctl_option *opt;
3220                 int j;
3221                 uint32_t plun;
3222
3223                 list = (struct ctl_lun_list *)addr;
3224
3225                 sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3226                 if (sb == NULL) {
3227                         list->status = CTL_LUN_LIST_ERROR;
3228                         snprintf(list->error_str, sizeof(list->error_str),
3229                                  "Unable to allocate %d bytes for LUN list",
3230                                  list->alloc_len);
3231                         break;
3232                 }
3233
3234                 sbuf_printf(sb, "<ctlportlist>\n");
3235
3236                 mtx_lock(&softc->ctl_lock);
3237                 STAILQ_FOREACH(port, &softc->port_list, links) {
3238                         retval = sbuf_printf(sb, "<targ_port id=\"%ju\">\n",
3239                                              (uintmax_t)port->targ_port);
3240
3241                         /*
3242                          * Bail out as soon as we see that we've overfilled
3243                          * the buffer.
3244                          */
3245                         if (retval != 0)
3246                                 break;
3247
3248                         retval = sbuf_printf(sb, "\t<frontend_type>%s"
3249                             "</frontend_type>\n", port->frontend->name);
3250                         if (retval != 0)
3251                                 break;
3252
3253                         retval = sbuf_printf(sb, "\t<port_type>%d</port_type>\n",
3254                                              port->port_type);
3255                         if (retval != 0)
3256                                 break;
3257
3258                         retval = sbuf_printf(sb, "\t<online>%s</online>\n",
3259                             (port->status & CTL_PORT_STATUS_ONLINE) ? "YES" : "NO");
3260                         if (retval != 0)
3261                                 break;
3262
3263                         retval = sbuf_printf(sb, "\t<port_name>%s</port_name>\n",
3264                             port->port_name);
3265                         if (retval != 0)
3266                                 break;
3267
3268                         retval = sbuf_printf(sb, "\t<physical_port>%d</physical_port>\n",
3269                             port->physical_port);
3270                         if (retval != 0)
3271                                 break;
3272
3273                         retval = sbuf_printf(sb, "\t<virtual_port>%d</virtual_port>\n",
3274                             port->virtual_port);
3275                         if (retval != 0)
3276                                 break;
3277
3278                         if (port->target_devid != NULL) {
3279                                 sbuf_printf(sb, "\t<target>");
3280                                 ctl_id_sbuf(port->target_devid, sb);
3281                                 sbuf_printf(sb, "</target>\n");
3282                         }
3283
3284                         if (port->port_devid != NULL) {
3285                                 sbuf_printf(sb, "\t<port>");
3286                                 ctl_id_sbuf(port->port_devid, sb);
3287                                 sbuf_printf(sb, "</port>\n");
3288                         }
3289
3290                         if (port->port_info != NULL) {
3291                                 retval = port->port_info(port->onoff_arg, sb);
3292                                 if (retval != 0)
3293                                         break;
3294                         }
3295                         STAILQ_FOREACH(opt, &port->options, links) {
3296                                 retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3297                                     opt->name, opt->value, opt->name);
3298                                 if (retval != 0)
3299                                         break;
3300                         }
3301
3302                         if (port->lun_map != NULL) {
3303                                 sbuf_printf(sb, "\t<lun_map>on</lun_map>\n");
3304                                 for (j = 0; j < port->lun_map_size; j++) {
3305                                         plun = ctl_lun_map_from_port(port, j);
3306                                         if (plun == UINT32_MAX)
3307                                                 continue;
3308                                         sbuf_printf(sb,
3309                                             "\t<lun id=\"%u\">%u</lun>\n",
3310                                             j, plun);
3311                                 }
3312                         }
3313
3314                         for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
3315                                 if (port->wwpn_iid[j].in_use == 0 ||
3316                                     (port->wwpn_iid[j].wwpn == 0 &&
3317                                      port->wwpn_iid[j].name == NULL))
3318                                         continue;
3319
3320                                 if (port->wwpn_iid[j].name != NULL)
3321                                         retval = sbuf_printf(sb,
3322                                             "\t<initiator id=\"%u\">%s</initiator>\n",
3323                                             j, port->wwpn_iid[j].name);
3324                                 else
3325                                         retval = sbuf_printf(sb,
3326                                             "\t<initiator id=\"%u\">naa.%08jx</initiator>\n",
3327                                             j, port->wwpn_iid[j].wwpn);
3328                                 if (retval != 0)
3329                                         break;
3330                         }
3331                         if (retval != 0)
3332                                 break;
3333
3334                         retval = sbuf_printf(sb, "</targ_port>\n");
3335                         if (retval != 0)
3336                                 break;
3337                 }
3338                 mtx_unlock(&softc->ctl_lock);
3339
3340                 if ((retval != 0)
3341                  || ((retval = sbuf_printf(sb, "</ctlportlist>\n")) != 0)) {
3342                         retval = 0;
3343                         sbuf_delete(sb);
3344                         list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3345                         snprintf(list->error_str, sizeof(list->error_str),
3346                                  "Out of space, %d bytes is too small",
3347                                  list->alloc_len);
3348                         break;
3349                 }
3350
3351                 sbuf_finish(sb);
3352
3353                 retval = copyout(sbuf_data(sb), list->lun_xml,
3354                                  sbuf_len(sb) + 1);
3355
3356                 list->fill_len = sbuf_len(sb) + 1;
3357                 list->status = CTL_LUN_LIST_OK;
3358                 sbuf_delete(sb);
3359                 break;
3360         }
3361         case CTL_LUN_MAP: {
3362                 struct ctl_lun_map *lm  = (struct ctl_lun_map *)addr;
3363                 struct ctl_port *port;
3364
3365                 mtx_lock(&softc->ctl_lock);
3366                 if (lm->port < softc->port_min ||
3367                     lm->port >= softc->port_max ||
3368                     (port = softc->ctl_ports[lm->port]) == NULL) {
3369                         mtx_unlock(&softc->ctl_lock);
3370                         return (ENXIO);
3371                 }
3372                 if (port->status & CTL_PORT_STATUS_ONLINE) {
3373                         STAILQ_FOREACH(lun, &softc->lun_list, links) {
3374                                 if (ctl_lun_map_to_port(port, lun->lun) ==
3375                                     UINT32_MAX)
3376                                         continue;
3377                                 mtx_lock(&lun->lun_lock);
3378                                 ctl_est_ua_port(lun, lm->port, -1,
3379                                     CTL_UA_LUN_CHANGE);
3380                                 mtx_unlock(&lun->lun_lock);
3381                         }
3382                 }
3383                 mtx_unlock(&softc->ctl_lock); // XXX: port_enable sleeps
3384                 if (lm->plun != UINT32_MAX) {
3385                         if (lm->lun == UINT32_MAX)
3386                                 retval = ctl_lun_map_unset(port, lm->plun);
3387                         else if (lm->lun < CTL_MAX_LUNS &&
3388                             softc->ctl_luns[lm->lun] != NULL)
3389                                 retval = ctl_lun_map_set(port, lm->plun, lm->lun);
3390                         else
3391                                 return (ENXIO);
3392                 } else {
3393                         if (lm->lun == UINT32_MAX)
3394                                 retval = ctl_lun_map_deinit(port);
3395                         else
3396                                 retval = ctl_lun_map_init(port);
3397                 }
3398                 if (port->status & CTL_PORT_STATUS_ONLINE)
3399                         ctl_isc_announce_port(port);
3400                 break;
3401         }
3402         case CTL_GET_LUN_STATS: {
3403                 struct ctl_get_io_stats *stats = (struct ctl_get_io_stats *)addr;
3404                 int i;
3405
3406                 /*
3407                  * XXX KDM no locking here.  If the LUN list changes,
3408                  * things can blow up.
3409                  */
3410                 i = 0;
3411                 stats->status = CTL_SS_OK;
3412                 stats->fill_len = 0;
3413                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
3414                         if (lun->lun < stats->first_item)
3415                                 continue;
3416                         if (stats->fill_len + sizeof(lun->stats) >
3417                             stats->alloc_len) {
3418                                 stats->status = CTL_SS_NEED_MORE_SPACE;
3419                                 break;
3420                         }
3421                         retval = copyout(&lun->stats, &stats->stats[i++],
3422                                          sizeof(lun->stats));
3423                         if (retval != 0)
3424                                 break;
3425                         stats->fill_len += sizeof(lun->stats);
3426                 }
3427                 stats->num_items = softc->num_luns;
3428                 stats->flags = CTL_STATS_FLAG_NONE;
3429 #ifdef CTL_TIME_IO
3430                 stats->flags |= CTL_STATS_FLAG_TIME_VALID;
3431 #endif
3432                 getnanouptime(&stats->timestamp);
3433                 break;
3434         }
3435         case CTL_GET_PORT_STATS: {
3436                 struct ctl_get_io_stats *stats = (struct ctl_get_io_stats *)addr;
3437                 int i;
3438
3439                 /*
3440                  * XXX KDM no locking here.  If the LUN list changes,
3441                  * things can blow up.
3442                  */
3443                 i = 0;
3444                 stats->status = CTL_SS_OK;
3445                 stats->fill_len = 0;
3446                 STAILQ_FOREACH(port, &softc->port_list, links) {
3447                         if (port->targ_port < stats->first_item)
3448                                 continue;
3449                         if (stats->fill_len + sizeof(port->stats) >
3450                             stats->alloc_len) {
3451                                 stats->status = CTL_SS_NEED_MORE_SPACE;
3452                                 break;
3453                         }
3454                         retval = copyout(&port->stats, &stats->stats[i++],
3455                                          sizeof(port->stats));
3456                         if (retval != 0)
3457                                 break;
3458                         stats->fill_len += sizeof(port->stats);
3459                 }
3460                 stats->num_items = softc->num_ports;
3461                 stats->flags = CTL_STATS_FLAG_NONE;
3462 #ifdef CTL_TIME_IO
3463                 stats->flags |= CTL_STATS_FLAG_TIME_VALID;
3464 #endif
3465                 getnanouptime(&stats->timestamp);
3466                 break;
3467         }
3468         default: {
3469                 /* XXX KDM should we fix this? */
3470 #if 0
3471                 struct ctl_backend_driver *backend;
3472                 unsigned int type;
3473                 int found;
3474
3475                 found = 0;
3476
3477                 /*
3478                  * We encode the backend type as the ioctl type for backend
3479                  * ioctls.  So parse it out here, and then search for a
3480                  * backend of this type.
3481                  */
3482                 type = _IOC_TYPE(cmd);
3483
3484                 STAILQ_FOREACH(backend, &softc->be_list, links) {
3485                         if (backend->type == type) {
3486                                 found = 1;
3487                                 break;
3488                         }
3489                 }
3490                 if (found == 0) {
3491                         printf("ctl: unknown ioctl command %#lx or backend "
3492                                "%d\n", cmd, type);
3493                         retval = EINVAL;
3494                         break;
3495                 }
3496                 retval = backend->ioctl(dev, cmd, addr, flag, td);
3497 #endif
3498                 retval = ENOTTY;
3499                 break;
3500         }
3501         }
3502         return (retval);
3503 }
3504
3505 uint32_t
3506 ctl_get_initindex(struct ctl_nexus *nexus)
3507 {
3508         return (nexus->initid + (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3509 }
3510
3511 int
3512 ctl_lun_map_init(struct ctl_port *port)
3513 {
3514         struct ctl_softc *softc = port->ctl_softc;
3515         struct ctl_lun *lun;
3516         int size = ctl_lun_map_size;
3517         uint32_t i;
3518
3519         if (port->lun_map == NULL || port->lun_map_size < size) {
3520                 port->lun_map_size = 0;
3521                 free(port->lun_map, M_CTL);
3522                 port->lun_map = malloc(size * sizeof(uint32_t),
3523                     M_CTL, M_NOWAIT);
3524         }
3525         if (port->lun_map == NULL)
3526                 return (ENOMEM);
3527         for (i = 0; i < size; i++)
3528                 port->lun_map[i] = UINT32_MAX;
3529         port->lun_map_size = size;
3530         if (port->status & CTL_PORT_STATUS_ONLINE) {
3531                 if (port->lun_disable != NULL) {
3532                         STAILQ_FOREACH(lun, &softc->lun_list, links)
3533                                 port->lun_disable(port->targ_lun_arg, lun->lun);
3534                 }
3535                 ctl_isc_announce_port(port);
3536         }
3537         return (0);
3538 }
3539
3540 int
3541 ctl_lun_map_deinit(struct ctl_port *port)
3542 {
3543         struct ctl_softc *softc = port->ctl_softc;
3544         struct ctl_lun *lun;
3545
3546         if (port->lun_map == NULL)
3547                 return (0);
3548         port->lun_map_size = 0;
3549         free(port->lun_map, M_CTL);
3550         port->lun_map = NULL;
3551         if (port->status & CTL_PORT_STATUS_ONLINE) {
3552                 if (port->lun_enable != NULL) {
3553                         STAILQ_FOREACH(lun, &softc->lun_list, links)
3554                                 port->lun_enable(port->targ_lun_arg, lun->lun);
3555                 }
3556                 ctl_isc_announce_port(port);
3557         }
3558         return (0);
3559 }
3560
3561 int
3562 ctl_lun_map_set(struct ctl_port *port, uint32_t plun, uint32_t glun)
3563 {
3564         int status;
3565         uint32_t old;
3566
3567         if (port->lun_map == NULL) {
3568                 status = ctl_lun_map_init(port);
3569                 if (status != 0)
3570                         return (status);
3571         }
3572         if (plun >= port->lun_map_size)
3573                 return (EINVAL);
3574         old = port->lun_map[plun];
3575         port->lun_map[plun] = glun;
3576         if ((port->status & CTL_PORT_STATUS_ONLINE) && old == UINT32_MAX) {
3577                 if (port->lun_enable != NULL)
3578                         port->lun_enable(port->targ_lun_arg, plun);
3579                 ctl_isc_announce_port(port);
3580         }
3581         return (0);
3582 }
3583
3584 int
3585 ctl_lun_map_unset(struct ctl_port *port, uint32_t plun)
3586 {
3587         uint32_t old;
3588
3589         if (port->lun_map == NULL || plun >= port->lun_map_size)
3590                 return (0);
3591         old = port->lun_map[plun];
3592         port->lun_map[plun] = UINT32_MAX;
3593         if ((port->status & CTL_PORT_STATUS_ONLINE) && old != UINT32_MAX) {
3594                 if (port->lun_disable != NULL)
3595                         port->lun_disable(port->targ_lun_arg, plun);
3596                 ctl_isc_announce_port(port);
3597         }
3598         return (0);
3599 }
3600
3601 uint32_t
3602 ctl_lun_map_from_port(struct ctl_port *port, uint32_t lun_id)
3603 {
3604
3605         if (port == NULL)
3606                 return (UINT32_MAX);
3607         if (port->lun_map == NULL)
3608                 return (lun_id);
3609         if (lun_id > port->lun_map_size)
3610                 return (UINT32_MAX);
3611         return (port->lun_map[lun_id]);
3612 }
3613
3614 uint32_t
3615 ctl_lun_map_to_port(struct ctl_port *port, uint32_t lun_id)
3616 {
3617         uint32_t i;
3618
3619         if (port == NULL)
3620                 return (UINT32_MAX);
3621         if (port->lun_map == NULL)
3622                 return (lun_id);
3623         for (i = 0; i < port->lun_map_size; i++) {
3624                 if (port->lun_map[i] == lun_id)
3625                         return (i);
3626         }
3627         return (UINT32_MAX);
3628 }
3629
3630 uint32_t
3631 ctl_decode_lun(uint64_t encoded)
3632 {
3633         uint8_t lun[8];
3634         uint32_t result = 0xffffffff;
3635
3636         be64enc(lun, encoded);
3637         switch (lun[0] & RPL_LUNDATA_ATYP_MASK) {
3638         case RPL_LUNDATA_ATYP_PERIPH:
3639                 if ((lun[0] & 0x3f) == 0 && lun[2] == 0 && lun[3] == 0 &&
3640                     lun[4] == 0 && lun[5] == 0 && lun[6] == 0 && lun[7] == 0)
3641                         result = lun[1];
3642                 break;
3643         case RPL_LUNDATA_ATYP_FLAT:
3644                 if (lun[2] == 0 && lun[3] == 0 && lun[4] == 0 && lun[5] == 0 &&
3645                     lun[6] == 0 && lun[7] == 0)
3646                         result = ((lun[0] & 0x3f) << 8) + lun[1];
3647                 break;
3648         case RPL_LUNDATA_ATYP_EXTLUN:
3649                 switch (lun[0] & RPL_LUNDATA_EXT_EAM_MASK) {
3650                 case 0x02:
3651                         switch (lun[0] & RPL_LUNDATA_EXT_LEN_MASK) {
3652                         case 0x00:
3653                                 result = lun[1];
3654                                 break;
3655                         case 0x10:
3656                                 result = (lun[1] << 16) + (lun[2] << 8) +
3657                                     lun[3];
3658                                 break;
3659                         case 0x20:
3660                                 if (lun[1] == 0 && lun[6] == 0 && lun[7] == 0)
3661                                         result = (lun[2] << 24) +
3662                                             (lun[3] << 16) + (lun[4] << 8) +
3663                                             lun[5];
3664                                 break;
3665                         }
3666                         break;
3667                 case RPL_LUNDATA_EXT_EAM_NOT_SPEC:
3668                         result = 0xffffffff;
3669                         break;
3670                 }
3671                 break;
3672         }
3673         return (result);
3674 }
3675
3676 uint64_t
3677 ctl_encode_lun(uint32_t decoded)
3678 {
3679         uint64_t l = decoded;
3680
3681         if (l <= 0xff)
3682                 return (((uint64_t)RPL_LUNDATA_ATYP_PERIPH << 56) | (l << 48));
3683         if (l <= 0x3fff)
3684                 return (((uint64_t)RPL_LUNDATA_ATYP_FLAT << 56) | (l << 48));
3685         if (l <= 0xffffff)
3686                 return (((uint64_t)(RPL_LUNDATA_ATYP_EXTLUN | 0x12) << 56) |
3687                     (l << 32));
3688         return ((((uint64_t)RPL_LUNDATA_ATYP_EXTLUN | 0x22) << 56) | (l << 16));
3689 }
3690
3691 int
3692 ctl_ffz(uint32_t *mask, uint32_t first, uint32_t last)
3693 {
3694         int i;
3695
3696         for (i = first; i < last; i++) {
3697                 if ((mask[i / 32] & (1 << (i % 32))) == 0)
3698                         return (i);
3699         }
3700         return (-1);
3701 }
3702
3703 int
3704 ctl_set_mask(uint32_t *mask, uint32_t bit)
3705 {
3706         uint32_t chunk, piece;
3707
3708         chunk = bit >> 5;
3709         piece = bit % (sizeof(uint32_t) * 8);
3710
3711         if ((mask[chunk] & (1 << piece)) != 0)
3712                 return (-1);
3713         else
3714                 mask[chunk] |= (1 << piece);
3715
3716         return (0);
3717 }
3718
3719 int
3720 ctl_clear_mask(uint32_t *mask, uint32_t bit)
3721 {
3722         uint32_t chunk, piece;
3723
3724         chunk = bit >> 5;
3725         piece = bit % (sizeof(uint32_t) * 8);
3726
3727         if ((mask[chunk] & (1 << piece)) == 0)
3728                 return (-1);
3729         else
3730                 mask[chunk] &= ~(1 << piece);
3731
3732         return (0);
3733 }
3734
3735 int
3736 ctl_is_set(uint32_t *mask, uint32_t bit)
3737 {
3738         uint32_t chunk, piece;
3739
3740         chunk = bit >> 5;
3741         piece = bit % (sizeof(uint32_t) * 8);
3742
3743         if ((mask[chunk] & (1 << piece)) == 0)
3744                 return (0);
3745         else
3746                 return (1);
3747 }
3748
3749 static uint64_t
3750 ctl_get_prkey(struct ctl_lun *lun, uint32_t residx)
3751 {
3752         uint64_t *t;
3753
3754         t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3755         if (t == NULL)
3756                 return (0);
3757         return (t[residx % CTL_MAX_INIT_PER_PORT]);
3758 }
3759
3760 static void
3761 ctl_clr_prkey(struct ctl_lun *lun, uint32_t residx)
3762 {
3763         uint64_t *t;
3764
3765         t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3766         if (t == NULL)
3767                 return;
3768         t[residx % CTL_MAX_INIT_PER_PORT] = 0;
3769 }
3770
3771 static void
3772 ctl_alloc_prkey(struct ctl_lun *lun, uint32_t residx)
3773 {
3774         uint64_t *p;
3775         u_int i;
3776
3777         i = residx/CTL_MAX_INIT_PER_PORT;
3778         if (lun->pr_keys[i] != NULL)
3779                 return;
3780         mtx_unlock(&lun->lun_lock);
3781         p = malloc(sizeof(uint64_t) * CTL_MAX_INIT_PER_PORT, M_CTL,
3782             M_WAITOK | M_ZERO);
3783         mtx_lock(&lun->lun_lock);
3784         if (lun->pr_keys[i] == NULL)
3785                 lun->pr_keys[i] = p;
3786         else
3787                 free(p, M_CTL);
3788 }
3789
3790 static void
3791 ctl_set_prkey(struct ctl_lun *lun, uint32_t residx, uint64_t key)
3792 {
3793         uint64_t *t;
3794
3795         t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3796         KASSERT(t != NULL, ("prkey %d is not allocated", residx));
3797         t[residx % CTL_MAX_INIT_PER_PORT] = key;
3798 }
3799
3800 /*
3801  * ctl_softc, pool_name, total_ctl_io are passed in.
3802  * npool is passed out.
3803  */
3804 int
3805 ctl_pool_create(struct ctl_softc *ctl_softc, const char *pool_name,
3806                 uint32_t total_ctl_io, void **npool)
3807 {
3808         struct ctl_io_pool *pool;
3809
3810         pool = (struct ctl_io_pool *)malloc(sizeof(*pool), M_CTL,
3811                                             M_NOWAIT | M_ZERO);
3812         if (pool == NULL)
3813                 return (ENOMEM);
3814
3815         snprintf(pool->name, sizeof(pool->name), "CTL IO %s", pool_name);
3816         pool->ctl_softc = ctl_softc;
3817 #ifdef IO_POOLS
3818         pool->zone = uma_zsecond_create(pool->name, NULL,
3819             NULL, NULL, NULL, ctl_softc->io_zone);
3820         /* uma_prealloc(pool->zone, total_ctl_io); */
3821 #else
3822         pool->zone = ctl_softc->io_zone;
3823 #endif
3824
3825         *npool = pool;
3826         return (0);
3827 }
3828
3829 void
3830 ctl_pool_free(struct ctl_io_pool *pool)
3831 {
3832
3833         if (pool == NULL)
3834                 return;
3835
3836 #ifdef IO_POOLS
3837         uma_zdestroy(pool->zone);
3838 #endif
3839         free(pool, M_CTL);
3840 }
3841
3842 union ctl_io *
3843 ctl_alloc_io(void *pool_ref)
3844 {
3845         struct ctl_io_pool *pool = (struct ctl_io_pool *)pool_ref;
3846         union ctl_io *io;
3847
3848         io = uma_zalloc(pool->zone, M_WAITOK);
3849         if (io != NULL) {
3850                 io->io_hdr.pool = pool_ref;
3851                 CTL_SOFTC(io) = pool->ctl_softc;
3852         }
3853         return (io);
3854 }
3855
3856 union ctl_io *
3857 ctl_alloc_io_nowait(void *pool_ref)
3858 {
3859         struct ctl_io_pool *pool = (struct ctl_io_pool *)pool_ref;
3860         union ctl_io *io;
3861
3862         io = uma_zalloc(pool->zone, M_NOWAIT);
3863         if (io != NULL) {
3864                 io->io_hdr.pool = pool_ref;
3865                 CTL_SOFTC(io) = pool->ctl_softc;
3866         }
3867         return (io);
3868 }
3869
3870 void
3871 ctl_free_io(union ctl_io *io)
3872 {
3873         struct ctl_io_pool *pool;
3874
3875         if (io == NULL)
3876                 return;
3877
3878         pool = (struct ctl_io_pool *)io->io_hdr.pool;
3879         uma_zfree(pool->zone, io);
3880 }
3881
3882 void
3883 ctl_zero_io(union ctl_io *io)
3884 {
3885         struct ctl_io_pool *pool;
3886
3887         if (io == NULL)
3888                 return;
3889
3890         /*
3891          * May need to preserve linked list pointers at some point too.
3892          */
3893         pool = io->io_hdr.pool;
3894         memset(io, 0, sizeof(*io));
3895         io->io_hdr.pool = pool;
3896         CTL_SOFTC(io) = pool->ctl_softc;
3897 }
3898
3899 int
3900 ctl_expand_number(const char *buf, uint64_t *num)
3901 {
3902         char *endptr;
3903         uint64_t number;
3904         unsigned shift;
3905
3906         number = strtoq(buf, &endptr, 0);
3907
3908         switch (tolower((unsigned char)*endptr)) {
3909         case 'e':
3910                 shift = 60;
3911                 break;
3912         case 'p':
3913                 shift = 50;
3914                 break;
3915         case 't':
3916                 shift = 40;
3917                 break;
3918         case 'g':
3919                 shift = 30;
3920                 break;
3921         case 'm':
3922                 shift = 20;
3923                 break;
3924         case 'k':
3925                 shift = 10;
3926                 break;
3927         case 'b':
3928         case '\0': /* No unit. */
3929                 *num = number;
3930                 return (0);
3931         default:
3932                 /* Unrecognized unit. */
3933                 return (-1);
3934         }
3935
3936         if ((number << shift) >> shift != number) {
3937                 /* Overflow */
3938                 return (-1);
3939         }
3940         *num = number << shift;
3941         return (0);
3942 }
3943
3944
3945 /*
3946  * This routine could be used in the future to load default and/or saved
3947  * mode page parameters for a particuar lun.
3948  */
3949 static int
3950 ctl_init_page_index(struct ctl_lun *lun)
3951 {
3952         int i, page_code;
3953         struct ctl_page_index *page_index;
3954         const char *value;
3955         uint64_t ival;
3956
3957         memcpy(&lun->mode_pages.index, page_index_template,
3958                sizeof(page_index_template));
3959
3960         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
3961
3962                 page_index = &lun->mode_pages.index[i];
3963                 if (lun->be_lun->lun_type == T_DIRECT &&
3964                     (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
3965                         continue;
3966                 if (lun->be_lun->lun_type == T_PROCESSOR &&
3967                     (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
3968                         continue;
3969                 if (lun->be_lun->lun_type == T_CDROM &&
3970                     (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
3971                         continue;
3972
3973                 page_code = page_index->page_code & SMPH_PC_MASK;
3974                 switch (page_code) {
3975                 case SMS_RW_ERROR_RECOVERY_PAGE: {
3976                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
3977                             ("subpage %#x for page %#x is incorrect!",
3978                             page_index->subpage, page_code));
3979                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_CURRENT],
3980                                &rw_er_page_default,
3981                                sizeof(rw_er_page_default));
3982                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_CHANGEABLE],
3983                                &rw_er_page_changeable,
3984                                sizeof(rw_er_page_changeable));
3985                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_DEFAULT],
3986                                &rw_er_page_default,
3987                                sizeof(rw_er_page_default));
3988                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_SAVED],
3989                                &rw_er_page_default,
3990                                sizeof(rw_er_page_default));
3991                         page_index->page_data =
3992                                 (uint8_t *)lun->mode_pages.rw_er_page;
3993                         break;
3994                 }
3995                 case SMS_FORMAT_DEVICE_PAGE: {
3996                         struct scsi_format_page *format_page;
3997
3998                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
3999                             ("subpage %#x for page %#x is incorrect!",
4000                             page_index->subpage, page_code));
4001
4002                         /*
4003                          * Sectors per track are set above.  Bytes per
4004                          * sector need to be set here on a per-LUN basis.
4005                          */
4006                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_CURRENT],
4007                                &format_page_default,
4008                                sizeof(format_page_default));
4009                         memcpy(&lun->mode_pages.format_page[
4010                                CTL_PAGE_CHANGEABLE], &format_page_changeable,
4011                                sizeof(format_page_changeable));
4012                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_DEFAULT],
4013                                &format_page_default,
4014                                sizeof(format_page_default));
4015                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_SAVED],
4016                                &format_page_default,
4017                                sizeof(format_page_default));
4018
4019                         format_page = &lun->mode_pages.format_page[
4020                                 CTL_PAGE_CURRENT];
4021                         scsi_ulto2b(lun->be_lun->blocksize,
4022                                     format_page->bytes_per_sector);
4023
4024                         format_page = &lun->mode_pages.format_page[
4025                                 CTL_PAGE_DEFAULT];
4026                         scsi_ulto2b(lun->be_lun->blocksize,
4027                                     format_page->bytes_per_sector);
4028
4029                         format_page = &lun->mode_pages.format_page[
4030                                 CTL_PAGE_SAVED];
4031                         scsi_ulto2b(lun->be_lun->blocksize,
4032                                     format_page->bytes_per_sector);
4033
4034                         page_index->page_data =
4035                                 (uint8_t *)lun->mode_pages.format_page;
4036                         break;
4037                 }
4038                 case SMS_RIGID_DISK_PAGE: {
4039                         struct scsi_rigid_disk_page *rigid_disk_page;
4040                         uint32_t sectors_per_cylinder;
4041                         uint64_t cylinders;
4042 #ifndef __XSCALE__
4043                         int shift;
4044 #endif /* !__XSCALE__ */
4045
4046                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4047                             ("subpage %#x for page %#x is incorrect!",
4048                             page_index->subpage, page_code));
4049
4050                         /*
4051                          * Rotation rate and sectors per track are set
4052                          * above.  We calculate the cylinders here based on
4053                          * capacity.  Due to the number of heads and
4054                          * sectors per track we're using, smaller arrays
4055                          * may turn out to have 0 cylinders.  Linux and
4056                          * FreeBSD don't pay attention to these mode pages
4057                          * to figure out capacity, but Solaris does.  It
4058                          * seems to deal with 0 cylinders just fine, and
4059                          * works out a fake geometry based on the capacity.
4060                          */
4061                         memcpy(&lun->mode_pages.rigid_disk_page[
4062                                CTL_PAGE_DEFAULT], &rigid_disk_page_default,
4063                                sizeof(rigid_disk_page_default));
4064                         memcpy(&lun->mode_pages.rigid_disk_page[
4065                                CTL_PAGE_CHANGEABLE],&rigid_disk_page_changeable,
4066                                sizeof(rigid_disk_page_changeable));
4067
4068                         sectors_per_cylinder = CTL_DEFAULT_SECTORS_PER_TRACK *
4069                                 CTL_DEFAULT_HEADS;
4070
4071                         /*
4072                          * The divide method here will be more accurate,
4073                          * probably, but results in floating point being
4074                          * used in the kernel on i386 (__udivdi3()).  On the
4075                          * XScale, though, __udivdi3() is implemented in
4076                          * software.
4077                          *
4078                          * The shift method for cylinder calculation is
4079                          * accurate if sectors_per_cylinder is a power of
4080                          * 2.  Otherwise it might be slightly off -- you
4081                          * might have a bit of a truncation problem.
4082                          */
4083 #ifdef  __XSCALE__
4084                         cylinders = (lun->be_lun->maxlba + 1) /
4085                                 sectors_per_cylinder;
4086 #else
4087                         for (shift = 31; shift > 0; shift--) {
4088                                 if (sectors_per_cylinder & (1 << shift))
4089                                         break;
4090                         }
4091                         cylinders = (lun->be_lun->maxlba + 1) >> shift;
4092 #endif
4093
4094                         /*
4095                          * We've basically got 3 bytes, or 24 bits for the
4096                          * cylinder size in the mode page.  If we're over,
4097                          * just round down to 2^24.
4098                          */
4099                         if (cylinders > 0xffffff)
4100                                 cylinders = 0xffffff;
4101
4102                         rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4103                                 CTL_PAGE_DEFAULT];
4104                         scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4105
4106                         if ((value = ctl_get_opt(&lun->be_lun->options,
4107                             "rpm")) != NULL) {
4108                                 scsi_ulto2b(strtol(value, NULL, 0),
4109                                      rigid_disk_page->rotation_rate);
4110                         }
4111
4112                         memcpy(&lun->mode_pages.rigid_disk_page[CTL_PAGE_CURRENT],
4113                                &lun->mode_pages.rigid_disk_page[CTL_PAGE_DEFAULT],
4114                                sizeof(rigid_disk_page_default));
4115                         memcpy(&lun->mode_pages.rigid_disk_page[CTL_PAGE_SAVED],
4116                                &lun->mode_pages.rigid_disk_page[CTL_PAGE_DEFAULT],
4117                                sizeof(rigid_disk_page_default));
4118
4119                         page_index->page_data =
4120                                 (uint8_t *)lun->mode_pages.rigid_disk_page;
4121                         break;
4122                 }
4123                 case SMS_VERIFY_ERROR_RECOVERY_PAGE: {
4124                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4125                             ("subpage %#x for page %#x is incorrect!",
4126                             page_index->subpage, page_code));
4127                         memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_CURRENT],
4128                                &verify_er_page_default,
4129                                sizeof(verify_er_page_default));
4130                         memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_CHANGEABLE],
4131                                &verify_er_page_changeable,
4132                                sizeof(verify_er_page_changeable));
4133                         memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_DEFAULT],
4134                                &verify_er_page_default,
4135                                sizeof(verify_er_page_default));
4136                         memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_SAVED],
4137                                &verify_er_page_default,
4138                                sizeof(verify_er_page_default));
4139                         page_index->page_data =
4140                                 (uint8_t *)lun->mode_pages.verify_er_page;
4141                         break;
4142                 }
4143                 case SMS_CACHING_PAGE: {
4144                         struct scsi_caching_page *caching_page;
4145
4146                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4147                             ("subpage %#x for page %#x is incorrect!",
4148                             page_index->subpage, page_code));
4149                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_DEFAULT],
4150                                &caching_page_default,
4151                                sizeof(caching_page_default));
4152                         memcpy(&lun->mode_pages.caching_page[
4153                                CTL_PAGE_CHANGEABLE], &caching_page_changeable,
4154                                sizeof(caching_page_changeable));
4155                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4156                                &caching_page_default,
4157                                sizeof(caching_page_default));
4158                         caching_page = &lun->mode_pages.caching_page[
4159                             CTL_PAGE_SAVED];
4160                         value = ctl_get_opt(&lun->be_lun->options, "writecache");
4161                         if (value != NULL && strcmp(value, "off") == 0)
4162                                 caching_page->flags1 &= ~SCP_WCE;
4163                         value = ctl_get_opt(&lun->be_lun->options, "readcache");
4164                         if (value != NULL && strcmp(value, "off") == 0)
4165                                 caching_page->flags1 |= SCP_RCD;
4166                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_CURRENT],
4167                                &lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4168                                sizeof(caching_page_default));
4169                         page_index->page_data =
4170                                 (uint8_t *)lun->mode_pages.caching_page;
4171                         break;
4172                 }
4173                 case SMS_CONTROL_MODE_PAGE: {
4174                         switch (page_index->subpage) {
4175                         case SMS_SUBPAGE_PAGE_0: {
4176                                 struct scsi_control_page *control_page;
4177
4178                                 memcpy(&lun->mode_pages.control_page[
4179                                     CTL_PAGE_DEFAULT],
4180                                        &control_page_default,
4181                                        sizeof(control_page_default));
4182                                 memcpy(&lun->mode_pages.control_page[
4183                                     CTL_PAGE_CHANGEABLE],
4184                                        &control_page_changeable,
4185                                        sizeof(control_page_changeable));
4186                                 memcpy(&lun->mode_pages.control_page[
4187                                     CTL_PAGE_SAVED],
4188                                        &control_page_default,
4189                                        sizeof(control_page_default));
4190                                 control_page = &lun->mode_pages.control_page[
4191                                     CTL_PAGE_SAVED];
4192                                 value = ctl_get_opt(&lun->be_lun->options,
4193                                     "reordering");
4194                                 if (value != NULL &&
4195                                     strcmp(value, "unrestricted") == 0) {
4196                                         control_page->queue_flags &=
4197                                             ~SCP_QUEUE_ALG_MASK;
4198                                         control_page->queue_flags |=
4199                                             SCP_QUEUE_ALG_UNRESTRICTED;
4200                                 }
4201                                 memcpy(&lun->mode_pages.control_page[
4202                                     CTL_PAGE_CURRENT],
4203                                        &lun->mode_pages.control_page[
4204                                     CTL_PAGE_SAVED],
4205                                        sizeof(control_page_default));
4206                                 page_index->page_data =
4207                                     (uint8_t *)lun->mode_pages.control_page;
4208                                 break;
4209                         }
4210                         case 0x01:
4211                                 memcpy(&lun->mode_pages.control_ext_page[
4212                                     CTL_PAGE_DEFAULT],
4213                                        &control_ext_page_default,
4214                                        sizeof(control_ext_page_default));
4215                                 memcpy(&lun->mode_pages.control_ext_page[
4216                                     CTL_PAGE_CHANGEABLE],
4217                                        &control_ext_page_changeable,
4218                                        sizeof(control_ext_page_changeable));
4219                                 memcpy(&lun->mode_pages.control_ext_page[
4220                                     CTL_PAGE_SAVED],
4221                                        &control_ext_page_default,
4222                                        sizeof(control_ext_page_default));
4223                                 memcpy(&lun->mode_pages.control_ext_page[
4224                                     CTL_PAGE_CURRENT],
4225                                        &lun->mode_pages.control_ext_page[
4226                                     CTL_PAGE_SAVED],
4227                                        sizeof(control_ext_page_default));
4228                                 page_index->page_data =
4229                                     (uint8_t *)lun->mode_pages.control_ext_page;
4230                                 break;
4231                         default:
4232                                 panic("subpage %#x for page %#x is incorrect!",
4233                                       page_index->subpage, page_code);
4234                         }
4235                         break;
4236                 }
4237                 case SMS_INFO_EXCEPTIONS_PAGE: {
4238                         switch (page_index->subpage) {
4239                         case SMS_SUBPAGE_PAGE_0:
4240                                 memcpy(&lun->mode_pages.ie_page[CTL_PAGE_CURRENT],
4241                                        &ie_page_default,
4242                                        sizeof(ie_page_default));
4243                                 memcpy(&lun->mode_pages.ie_page[
4244                                        CTL_PAGE_CHANGEABLE], &ie_page_changeable,
4245                                        sizeof(ie_page_changeable));
4246                                 memcpy(&lun->mode_pages.ie_page[CTL_PAGE_DEFAULT],
4247                                        &ie_page_default,
4248                                        sizeof(ie_page_default));
4249                                 memcpy(&lun->mode_pages.ie_page[CTL_PAGE_SAVED],
4250                                        &ie_page_default,
4251                                        sizeof(ie_page_default));
4252                                 page_index->page_data =
4253                                         (uint8_t *)lun->mode_pages.ie_page;
4254                                 break;
4255                         case 0x02: {
4256                                 struct ctl_logical_block_provisioning_page *page;
4257
4258                                 memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_DEFAULT],
4259                                        &lbp_page_default,
4260                                        sizeof(lbp_page_default));
4261                                 memcpy(&lun->mode_pages.lbp_page[
4262                                        CTL_PAGE_CHANGEABLE], &lbp_page_changeable,
4263                                        sizeof(lbp_page_changeable));
4264                                 memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_SAVED],
4265                                        &lbp_page_default,
4266                                        sizeof(lbp_page_default));
4267                                 page = &lun->mode_pages.lbp_page[CTL_PAGE_SAVED];
4268                                 value = ctl_get_opt(&lun->be_lun->options,
4269                                     "avail-threshold");
4270                                 if (value != NULL &&
4271                                     ctl_expand_number(value, &ival) == 0) {
4272                                         page->descr[0].flags |= SLBPPD_ENABLED |
4273                                             SLBPPD_ARMING_DEC;
4274                                         if (lun->be_lun->blocksize)
4275                                                 ival /= lun->be_lun->blocksize;
4276                                         else
4277                                                 ival /= 512;
4278                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4279                                             page->descr[0].count);
4280                                 }
4281                                 value = ctl_get_opt(&lun->be_lun->options,
4282                                     "used-threshold");
4283                                 if (value != NULL &&
4284                                     ctl_expand_number(value, &ival) == 0) {
4285                                         page->descr[1].flags |= SLBPPD_ENABLED |
4286                                             SLBPPD_ARMING_INC;
4287                                         if (lun->be_lun->blocksize)
4288                                                 ival /= lun->be_lun->blocksize;
4289                                         else
4290                                                 ival /= 512;
4291                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4292                                             page->descr[1].count);
4293                                 }
4294                                 value = ctl_get_opt(&lun->be_lun->options,
4295                                     "pool-avail-threshold");
4296                                 if (value != NULL &&
4297                                     ctl_expand_number(value, &ival) == 0) {
4298                                         page->descr[2].flags |= SLBPPD_ENABLED |
4299                                             SLBPPD_ARMING_DEC;
4300                                         if (lun->be_lun->blocksize)
4301                                                 ival /= lun->be_lun->blocksize;
4302                                         else
4303                                                 ival /= 512;
4304                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4305                                             page->descr[2].count);
4306                                 }
4307                                 value = ctl_get_opt(&lun->be_lun->options,
4308                                     "pool-used-threshold");
4309                                 if (value != NULL &&
4310                                     ctl_expand_number(value, &ival) == 0) {
4311                                         page->descr[3].flags |= SLBPPD_ENABLED |
4312                                             SLBPPD_ARMING_INC;
4313                                         if (lun->be_lun->blocksize)
4314                                                 ival /= lun->be_lun->blocksize;
4315                                         else
4316                                                 ival /= 512;
4317                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4318                                             page->descr[3].count);
4319                                 }
4320                                 memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_CURRENT],
4321                                        &lun->mode_pages.lbp_page[CTL_PAGE_SAVED],
4322                                        sizeof(lbp_page_default));
4323                                 page_index->page_data =
4324                                         (uint8_t *)lun->mode_pages.lbp_page;
4325                                 break;
4326                         }
4327                         default:
4328                                 panic("subpage %#x for page %#x is incorrect!",
4329                                       page_index->subpage, page_code);
4330                         }
4331                         break;
4332                 }
4333                 case SMS_CDDVD_CAPS_PAGE:{
4334                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4335                             ("subpage %#x for page %#x is incorrect!",
4336                             page_index->subpage, page_code));
4337                         memcpy(&lun->mode_pages.cddvd_page[CTL_PAGE_DEFAULT],
4338                                &cddvd_page_default,
4339                                sizeof(cddvd_page_default));
4340                         memcpy(&lun->mode_pages.cddvd_page[
4341                                CTL_PAGE_CHANGEABLE], &cddvd_page_changeable,
4342                                sizeof(cddvd_page_changeable));
4343                         memcpy(&lun->mode_pages.cddvd_page[CTL_PAGE_SAVED],
4344                                &cddvd_page_default,
4345                                sizeof(cddvd_page_default));
4346                         memcpy(&lun->mode_pages.cddvd_page[CTL_PAGE_CURRENT],
4347                                &lun->mode_pages.cddvd_page[CTL_PAGE_SAVED],
4348                                sizeof(cddvd_page_default));
4349                         page_index->page_data =
4350                                 (uint8_t *)lun->mode_pages.cddvd_page;
4351                         break;
4352                 }
4353                 default:
4354                         panic("invalid page code value %#x", page_code);
4355                 }
4356         }
4357
4358         return (CTL_RETVAL_COMPLETE);
4359 }
4360
4361 static int
4362 ctl_init_log_page_index(struct ctl_lun *lun)
4363 {
4364         struct ctl_page_index *page_index;
4365         int i, j, k, prev;
4366
4367         memcpy(&lun->log_pages.index, log_page_index_template,
4368                sizeof(log_page_index_template));
4369
4370         prev = -1;
4371         for (i = 0, j = 0, k = 0; i < CTL_NUM_LOG_PAGES; i++) {
4372
4373                 page_index = &lun->log_pages.index[i];
4374                 if (lun->be_lun->lun_type == T_DIRECT &&
4375                     (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
4376                         continue;
4377                 if (lun->be_lun->lun_type == T_PROCESSOR &&
4378                     (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
4379                         continue;
4380                 if (lun->be_lun->lun_type == T_CDROM &&
4381                     (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
4382                         continue;
4383
4384                 if (page_index->page_code == SLS_LOGICAL_BLOCK_PROVISIONING &&
4385                     lun->backend->lun_attr == NULL)
4386                         continue;
4387
4388                 if (page_index->page_code != prev) {
4389                         lun->log_pages.pages_page[j] = page_index->page_code;
4390                         prev = page_index->page_code;
4391                         j++;
4392                 }
4393                 lun->log_pages.subpages_page[k*2] = page_index->page_code;
4394                 lun->log_pages.subpages_page[k*2+1] = page_index->subpage;
4395                 k++;
4396         }
4397         lun->log_pages.index[0].page_data = &lun->log_pages.pages_page[0];
4398         lun->log_pages.index[0].page_len = j;
4399         lun->log_pages.index[1].page_data = &lun->log_pages.subpages_page[0];
4400         lun->log_pages.index[1].page_len = k * 2;
4401         lun->log_pages.index[2].page_data = &lun->log_pages.lbp_page[0];
4402         lun->log_pages.index[2].page_len = 12*CTL_NUM_LBP_PARAMS;
4403         lun->log_pages.index[3].page_data = (uint8_t *)&lun->log_pages.stat_page;
4404         lun->log_pages.index[3].page_len = sizeof(lun->log_pages.stat_page);
4405         lun->log_pages.index[4].page_data = (uint8_t *)&lun->log_pages.ie_page;
4406         lun->log_pages.index[4].page_len = sizeof(lun->log_pages.ie_page);
4407
4408         return (CTL_RETVAL_COMPLETE);
4409 }
4410
4411 static int
4412 hex2bin(const char *str, uint8_t *buf, int buf_size)
4413 {
4414         int i;
4415         u_char c;
4416
4417         memset(buf, 0, buf_size);
4418         while (isspace(str[0]))
4419                 str++;
4420         if (str[0] == '0' && (str[1] == 'x' || str[1] == 'X'))
4421                 str += 2;
4422         buf_size *= 2;
4423         for (i = 0; str[i] != 0 && i < buf_size; i++) {
4424                 while (str[i] == '-')   /* Skip dashes in UUIDs. */
4425                         str++;
4426                 c = str[i];
4427                 if (isdigit(c))
4428                         c -= '0';
4429                 else if (isalpha(c))
4430                         c -= isupper(c) ? 'A' - 10 : 'a' - 10;
4431                 else
4432                         break;
4433                 if (c >= 16)
4434                         break;
4435                 if ((i & 1) == 0)
4436                         buf[i / 2] |= (c << 4);
4437                 else
4438                         buf[i / 2] |= c;
4439         }
4440         return ((i + 1) / 2);
4441 }
4442
4443 /*
4444  * LUN allocation.
4445  *
4446  * Requirements:
4447  * - caller allocates and zeros LUN storage, or passes in a NULL LUN if he
4448  *   wants us to allocate the LUN and he can block.
4449  * - ctl_softc is always set
4450  * - be_lun is set if the LUN has a backend (needed for disk LUNs)
4451  *
4452  * Returns 0 for success, non-zero (errno) for failure.
4453  */
4454 static int
4455 ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *ctl_lun,
4456               struct ctl_be_lun *const be_lun)
4457 {
4458         struct ctl_lun *nlun, *lun;
4459         struct scsi_vpd_id_descriptor *desc;
4460         struct scsi_vpd_id_t10 *t10id;
4461         const char *eui, *naa, *scsiname, *uuid, *vendor, *value;
4462         int lun_number, lun_malloced;
4463         int devidlen, idlen1, idlen2 = 0, len;
4464
4465         if (be_lun == NULL)
4466                 return (EINVAL);
4467
4468         /*
4469          * We currently only support Direct Access or Processor LUN types.
4470          */
4471         switch (be_lun->lun_type) {
4472         case T_DIRECT:
4473         case T_PROCESSOR:
4474         case T_CDROM:
4475                 break;
4476         case T_SEQUENTIAL:
4477         case T_CHANGER:
4478         default:
4479                 be_lun->lun_config_status(be_lun->be_lun,
4480                                           CTL_LUN_CONFIG_FAILURE);
4481                 break;
4482         }
4483         if (ctl_lun == NULL) {
4484                 lun = malloc(sizeof(*lun), M_CTL, M_WAITOK);
4485                 lun_malloced = 1;
4486         } else {
4487                 lun_malloced = 0;
4488                 lun = ctl_lun;
4489         }
4490
4491         memset(lun, 0, sizeof(*lun));
4492         if (lun_malloced)
4493                 lun->flags = CTL_LUN_MALLOCED;
4494
4495         /* Generate LUN ID. */
4496         devidlen = max(CTL_DEVID_MIN_LEN,
4497             strnlen(be_lun->device_id, CTL_DEVID_LEN));
4498         idlen1 = sizeof(*t10id) + devidlen;
4499         len = sizeof(struct scsi_vpd_id_descriptor) + idlen1;
4500         scsiname = ctl_get_opt(&be_lun->options, "scsiname");
4501         if (scsiname != NULL) {
4502                 idlen2 = roundup2(strlen(scsiname) + 1, 4);
4503                 len += sizeof(struct scsi_vpd_id_descriptor) + idlen2;
4504         }
4505         eui = ctl_get_opt(&be_lun->options, "eui");
4506         if (eui != NULL) {
4507                 len += sizeof(struct scsi_vpd_id_descriptor) + 16;
4508         }
4509         naa = ctl_get_opt(&be_lun->options, "naa");
4510         if (naa != NULL) {
4511                 len += sizeof(struct scsi_vpd_id_descriptor) + 16;
4512         }
4513         uuid = ctl_get_opt(&be_lun->options, "uuid");
4514         if (uuid != NULL) {
4515                 len += sizeof(struct scsi_vpd_id_descriptor) + 18;
4516         }
4517         lun->lun_devid = malloc(sizeof(struct ctl_devid) + len,
4518             M_CTL, M_WAITOK | M_ZERO);
4519         desc = (struct scsi_vpd_id_descriptor *)lun->lun_devid->data;
4520         desc->proto_codeset = SVPD_ID_CODESET_ASCII;
4521         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_T10;
4522         desc->length = idlen1;
4523         t10id = (struct scsi_vpd_id_t10 *)&desc->identifier[0];
4524         memset(t10id->vendor, ' ', sizeof(t10id->vendor));
4525         if ((vendor = ctl_get_opt(&be_lun->options, "vendor")) == NULL) {
4526                 strncpy((char *)t10id->vendor, CTL_VENDOR, sizeof(t10id->vendor));
4527         } else {
4528                 strncpy(t10id->vendor, vendor,
4529                     min(sizeof(t10id->vendor), strlen(vendor)));
4530         }
4531         strncpy((char *)t10id->vendor_spec_id,
4532             (char *)be_lun->device_id, devidlen);
4533         if (scsiname != NULL) {
4534                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4535                     desc->length);
4536                 desc->proto_codeset = SVPD_ID_CODESET_UTF8;
4537                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4538                     SVPD_ID_TYPE_SCSI_NAME;
4539                 desc->length = idlen2;
4540                 strlcpy(desc->identifier, scsiname, idlen2);
4541         }
4542         if (eui != NULL) {
4543                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4544                     desc->length);
4545                 desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4546                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4547                     SVPD_ID_TYPE_EUI64;
4548                 desc->length = hex2bin(eui, desc->identifier, 16);
4549                 desc->length = desc->length > 12 ? 16 :
4550                     (desc->length > 8 ? 12 : 8);
4551                 len -= 16 - desc->length;
4552         }
4553         if (naa != NULL) {
4554                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4555                     desc->length);
4556                 desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4557                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4558                     SVPD_ID_TYPE_NAA;
4559                 desc->length = hex2bin(naa, desc->identifier, 16);
4560                 desc->length = desc->length > 8 ? 16 : 8;
4561                 len -= 16 - desc->length;
4562         }
4563         if (uuid != NULL) {
4564                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4565                     desc->length);
4566                 desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4567                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4568                     SVPD_ID_TYPE_UUID;
4569                 desc->identifier[0] = 0x10;
4570                 hex2bin(uuid, &desc->identifier[2], 16);
4571                 desc->length = 18;
4572         }
4573         lun->lun_devid->len = len;
4574
4575         mtx_lock(&ctl_softc->ctl_lock);
4576         /*
4577          * See if the caller requested a particular LUN number.  If so, see
4578          * if it is available.  Otherwise, allocate the first available LUN.
4579          */
4580         if (be_lun->flags & CTL_LUN_FLAG_ID_REQ) {
4581                 if ((be_lun->req_lun_id > (CTL_MAX_LUNS - 1))
4582                  || (ctl_is_set(ctl_softc->ctl_lun_mask, be_lun->req_lun_id))) {
4583                         mtx_unlock(&ctl_softc->ctl_lock);
4584                         if (be_lun->req_lun_id > (CTL_MAX_LUNS - 1)) {
4585                                 printf("ctl: requested LUN ID %d is higher "
4586                                        "than CTL_MAX_LUNS - 1 (%d)\n",
4587                                        be_lun->req_lun_id, CTL_MAX_LUNS - 1);
4588                         } else {
4589                                 /*
4590                                  * XXX KDM return an error, or just assign
4591                                  * another LUN ID in this case??
4592                                  */
4593                                 printf("ctl: requested LUN ID %d is already "
4594                                        "in use\n", be_lun->req_lun_id);
4595                         }
4596 fail:
4597                         free(lun->lun_devid, M_CTL);
4598                         if (lun->flags & CTL_LUN_MALLOCED)
4599                                 free(lun, M_CTL);
4600                         be_lun->lun_config_status(be_lun->be_lun,
4601                                                   CTL_LUN_CONFIG_FAILURE);
4602                         return (ENOSPC);
4603                 }
4604                 lun_number = be_lun->req_lun_id;
4605         } else {
4606                 lun_number = ctl_ffz(ctl_softc->ctl_lun_mask, 0, CTL_MAX_LUNS);
4607                 if (lun_number == -1) {
4608                         mtx_unlock(&ctl_softc->ctl_lock);
4609                         printf("ctl: can't allocate LUN, out of LUNs\n");
4610                         goto fail;
4611                 }
4612         }
4613         ctl_set_mask(ctl_softc->ctl_lun_mask, lun_number);
4614         mtx_unlock(&ctl_softc->ctl_lock);
4615
4616         mtx_init(&lun->lun_lock, "CTL LUN", NULL, MTX_DEF);
4617         lun->lun = lun_number;
4618         lun->be_lun = be_lun;
4619         /*
4620          * The processor LUN is always enabled.  Disk LUNs come on line
4621          * disabled, and must be enabled by the backend.
4622          */
4623         lun->flags |= CTL_LUN_DISABLED;
4624         lun->backend = be_lun->be;
4625         be_lun->ctl_lun = lun;
4626         be_lun->lun_id = lun_number;
4627         atomic_add_int(&be_lun->be->num_luns, 1);
4628         if (be_lun->flags & CTL_LUN_FLAG_EJECTED)
4629                 lun->flags |= CTL_LUN_EJECTED;
4630         if (be_lun->flags & CTL_LUN_FLAG_NO_MEDIA)
4631                 lun->flags |= CTL_LUN_NO_MEDIA;
4632         if (be_lun->flags & CTL_LUN_FLAG_STOPPED)
4633                 lun->flags |= CTL_LUN_STOPPED;
4634
4635         if (be_lun->flags & CTL_LUN_FLAG_PRIMARY)
4636                 lun->flags |= CTL_LUN_PRIMARY_SC;
4637
4638         value = ctl_get_opt(&be_lun->options, "removable");
4639         if (value != NULL) {
4640                 if (strcmp(value, "on") == 0)
4641                         lun->flags |= CTL_LUN_REMOVABLE;
4642         } else if (be_lun->lun_type == T_CDROM)
4643                 lun->flags |= CTL_LUN_REMOVABLE;
4644
4645         lun->ctl_softc = ctl_softc;
4646 #ifdef CTL_TIME_IO
4647         lun->last_busy = getsbinuptime();
4648 #endif
4649         TAILQ_INIT(&lun->ooa_queue);
4650         TAILQ_INIT(&lun->blocked_queue);
4651         STAILQ_INIT(&lun->error_list);
4652         lun->ie_reported = 1;
4653         callout_init_mtx(&lun->ie_callout, &lun->lun_lock, 0);
4654         ctl_tpc_lun_init(lun);
4655         if (lun->flags & CTL_LUN_REMOVABLE) {
4656                 lun->prevent = malloc((CTL_MAX_INITIATORS + 31) / 32 * 4,
4657                     M_CTL, M_WAITOK);
4658         }
4659
4660         /*
4661          * Initialize the mode and log page index.
4662          */
4663         ctl_init_page_index(lun);
4664         ctl_init_log_page_index(lun);
4665
4666         /* Setup statistics gathering */
4667 #ifdef CTL_LEGACY_STATS
4668         lun->legacy_stats.device_type = be_lun->lun_type;
4669         lun->legacy_stats.lun_number = lun_number;
4670         lun->legacy_stats.blocksize = be_lun->blocksize;
4671         if (be_lun->blocksize == 0)
4672                 lun->legacy_stats.flags = CTL_LUN_STATS_NO_BLOCKSIZE;
4673         for (len = 0; len < CTL_MAX_PORTS; len++)
4674                 lun->legacy_stats.ports[len].targ_port = len;
4675 #endif /* CTL_LEGACY_STATS */
4676         lun->stats.item = lun_number;
4677
4678         /*
4679          * Now, before we insert this lun on the lun list, set the lun
4680          * inventory changed UA for all other luns.
4681          */
4682         mtx_lock(&ctl_softc->ctl_lock);
4683         STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) {
4684                 mtx_lock(&nlun->lun_lock);
4685                 ctl_est_ua_all(nlun, -1, CTL_UA_LUN_CHANGE);
4686                 mtx_unlock(&nlun->lun_lock);
4687         }
4688         STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links);
4689         ctl_softc->ctl_luns[lun_number] = lun;
4690         ctl_softc->num_luns++;
4691         mtx_unlock(&ctl_softc->ctl_lock);
4692
4693         lun->be_lun->lun_config_status(lun->be_lun->be_lun, CTL_LUN_CONFIG_OK);
4694         return (0);
4695 }
4696
4697 /*
4698  * Delete a LUN.
4699  * Assumptions:
4700  * - LUN has already been marked invalid and any pending I/O has been taken
4701  *   care of.
4702  */
4703 static int
4704 ctl_free_lun(struct ctl_lun *lun)
4705 {
4706         struct ctl_softc *softc = lun->ctl_softc;
4707         struct ctl_lun *nlun;
4708         int i;
4709
4710         mtx_assert(&softc->ctl_lock, MA_OWNED);
4711
4712         STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links);
4713
4714         ctl_clear_mask(softc->ctl_lun_mask, lun->lun);
4715
4716         softc->ctl_luns[lun->lun] = NULL;
4717
4718         if (!TAILQ_EMPTY(&lun->ooa_queue))
4719                 panic("Freeing a LUN %p with outstanding I/O!!\n", lun);
4720
4721         softc->num_luns--;
4722
4723         /*
4724          * Tell the backend to free resources, if this LUN has a backend.
4725          */
4726         atomic_subtract_int(&lun->be_lun->be->num_luns, 1);
4727         lun->be_lun->lun_shutdown(lun->be_lun->be_lun);
4728
4729         lun->ie_reportcnt = UINT32_MAX;
4730         callout_drain(&lun->ie_callout);
4731
4732         ctl_tpc_lun_shutdown(lun);
4733         mtx_destroy(&lun->lun_lock);
4734         free(lun->lun_devid, M_CTL);
4735         for (i = 0; i < CTL_MAX_PORTS; i++)
4736                 free(lun->pending_ua[i], M_CTL);
4737         for (i = 0; i < CTL_MAX_PORTS; i++)
4738                 free(lun->pr_keys[i], M_CTL);
4739         free(lun->write_buffer, M_CTL);
4740         free(lun->prevent, M_CTL);
4741         if (lun->flags & CTL_LUN_MALLOCED)
4742                 free(lun, M_CTL);
4743
4744         STAILQ_FOREACH(nlun, &softc->lun_list, links) {
4745                 mtx_lock(&nlun->lun_lock);
4746                 ctl_est_ua_all(nlun, -1, CTL_UA_LUN_CHANGE);
4747                 mtx_unlock(&nlun->lun_lock);
4748         }
4749
4750         return (0);
4751 }
4752
4753 static void
4754 ctl_create_lun(struct ctl_be_lun *be_lun)
4755 {
4756
4757         /*
4758          * ctl_alloc_lun() should handle all potential failure cases.
4759          */
4760         ctl_alloc_lun(control_softc, NULL, be_lun);
4761 }
4762
4763 int
4764 ctl_add_lun(struct ctl_be_lun *be_lun)
4765 {
4766         struct ctl_softc *softc = control_softc;
4767
4768         mtx_lock(&softc->ctl_lock);
4769         STAILQ_INSERT_TAIL(&softc->pending_lun_queue, be_lun, links);
4770         mtx_unlock(&softc->ctl_lock);
4771         wakeup(&softc->pending_lun_queue);
4772
4773         return (0);
4774 }
4775
4776 int
4777 ctl_enable_lun(struct ctl_be_lun *be_lun)
4778 {
4779         struct ctl_softc *softc;
4780         struct ctl_port *port, *nport;
4781         struct ctl_lun *lun;
4782         int retval;
4783
4784         lun = (struct ctl_lun *)be_lun->ctl_lun;
4785         softc = lun->ctl_softc;
4786
4787         mtx_lock(&softc->ctl_lock);
4788         mtx_lock(&lun->lun_lock);
4789         if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4790                 /*
4791                  * eh?  Why did we get called if the LUN is already
4792                  * enabled?
4793                  */
4794                 mtx_unlock(&lun->lun_lock);
4795                 mtx_unlock(&softc->ctl_lock);
4796                 return (0);
4797         }
4798         lun->flags &= ~CTL_LUN_DISABLED;
4799         mtx_unlock(&lun->lun_lock);
4800
4801         STAILQ_FOREACH_SAFE(port, &softc->port_list, links, nport) {
4802                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0 ||
4803                     port->lun_map != NULL || port->lun_enable == NULL)
4804                         continue;
4805
4806                 /*
4807                  * Drop the lock while we call the FETD's enable routine.
4808                  * This can lead to a callback into CTL (at least in the
4809                  * case of the internal initiator frontend.
4810                  */
4811                 mtx_unlock(&softc->ctl_lock);
4812                 retval = port->lun_enable(port->targ_lun_arg, lun->lun);
4813                 mtx_lock(&softc->ctl_lock);
4814                 if (retval != 0) {
4815                         printf("%s: FETD %s port %d returned error "
4816                                "%d for lun_enable on lun %jd\n",
4817                                __func__, port->port_name, port->targ_port,
4818                                retval, (intmax_t)lun->lun);
4819                 }
4820         }
4821
4822         mtx_unlock(&softc->ctl_lock);
4823         ctl_isc_announce_lun(lun);
4824
4825         return (0);
4826 }
4827
4828 int
4829 ctl_disable_lun(struct ctl_be_lun *be_lun)
4830 {
4831         struct ctl_softc *softc;
4832         struct ctl_port *port;
4833         struct ctl_lun *lun;
4834         int retval;
4835
4836         lun = (struct ctl_lun *)be_lun->ctl_lun;
4837         softc = lun->ctl_softc;
4838
4839         mtx_lock(&softc->ctl_lock);
4840         mtx_lock(&lun->lun_lock);
4841         if (lun->flags & CTL_LUN_DISABLED) {
4842                 mtx_unlock(&lun->lun_lock);
4843                 mtx_unlock(&softc->ctl_lock);
4844                 return (0);
4845         }
4846         lun->flags |= CTL_LUN_DISABLED;
4847         mtx_unlock(&lun->lun_lock);
4848
4849         STAILQ_FOREACH(port, &softc->port_list, links) {
4850                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0 ||
4851                     port->lun_map != NULL || port->lun_disable == NULL)
4852                         continue;
4853
4854                 /*
4855                  * Drop the lock before we call the frontend's disable
4856                  * routine, to avoid lock order reversals.
4857                  *
4858                  * XXX KDM what happens if the frontend list changes while
4859                  * we're traversing it?  It's unlikely, but should be handled.
4860                  */
4861                 mtx_unlock(&softc->ctl_lock);
4862                 retval = port->lun_disable(port->targ_lun_arg, lun->lun);
4863                 mtx_lock(&softc->ctl_lock);
4864                 if (retval != 0) {
4865                         printf("%s: FETD %s port %d returned error "
4866                                "%d for lun_disable on lun %jd\n",
4867                                __func__, port->port_name, port->targ_port,
4868                                retval, (intmax_t)lun->lun);
4869                 }
4870         }
4871
4872         mtx_unlock(&softc->ctl_lock);
4873         ctl_isc_announce_lun(lun);
4874
4875         return (0);
4876 }
4877
4878 int
4879 ctl_start_lun(struct ctl_be_lun *be_lun)
4880 {
4881         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4882
4883         mtx_lock(&lun->lun_lock);
4884         lun->flags &= ~CTL_LUN_STOPPED;
4885         mtx_unlock(&lun->lun_lock);
4886         return (0);
4887 }
4888
4889 int
4890 ctl_stop_lun(struct ctl_be_lun *be_lun)
4891 {
4892         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4893
4894         mtx_lock(&lun->lun_lock);
4895         lun->flags |= CTL_LUN_STOPPED;
4896         mtx_unlock(&lun->lun_lock);
4897         return (0);
4898 }
4899
4900 int
4901 ctl_lun_no_media(struct ctl_be_lun *be_lun)
4902 {
4903         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4904
4905         mtx_lock(&lun->lun_lock);
4906         lun->flags |= CTL_LUN_NO_MEDIA;
4907         mtx_unlock(&lun->lun_lock);
4908         return (0);
4909 }
4910
4911 int
4912 ctl_lun_has_media(struct ctl_be_lun *be_lun)
4913 {
4914         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4915         union ctl_ha_msg msg;
4916
4917         mtx_lock(&lun->lun_lock);
4918         lun->flags &= ~(CTL_LUN_NO_MEDIA | CTL_LUN_EJECTED);
4919         if (lun->flags & CTL_LUN_REMOVABLE)
4920                 ctl_est_ua_all(lun, -1, CTL_UA_MEDIUM_CHANGE);
4921         mtx_unlock(&lun->lun_lock);
4922         if ((lun->flags & CTL_LUN_REMOVABLE) &&
4923             lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
4924                 bzero(&msg.ua, sizeof(msg.ua));
4925                 msg.hdr.msg_type = CTL_MSG_UA;
4926                 msg.hdr.nexus.initid = -1;
4927                 msg.hdr.nexus.targ_port = -1;
4928                 msg.hdr.nexus.targ_lun = lun->lun;
4929                 msg.hdr.nexus.targ_mapped_lun = lun->lun;
4930                 msg.ua.ua_all = 1;
4931                 msg.ua.ua_set = 1;
4932                 msg.ua.ua_type = CTL_UA_MEDIUM_CHANGE;
4933                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg.ua),
4934                     M_WAITOK);
4935         }
4936         return (0);
4937 }
4938
4939 int
4940 ctl_lun_ejected(struct ctl_be_lun *be_lun)
4941 {
4942         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4943
4944         mtx_lock(&lun->lun_lock);
4945         lun->flags |= CTL_LUN_EJECTED;
4946         mtx_unlock(&lun->lun_lock);
4947         return (0);
4948 }
4949
4950 int
4951 ctl_lun_primary(struct ctl_be_lun *be_lun)
4952 {
4953         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4954
4955         mtx_lock(&lun->lun_lock);
4956         lun->flags |= CTL_LUN_PRIMARY_SC;
4957         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
4958         mtx_unlock(&lun->lun_lock);
4959         ctl_isc_announce_lun(lun);
4960         return (0);
4961 }
4962
4963 int
4964 ctl_lun_secondary(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_PRIMARY_SC;
4970         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
4971         mtx_unlock(&lun->lun_lock);
4972         ctl_isc_announce_lun(lun);
4973         return (0);
4974 }
4975
4976 int
4977 ctl_invalidate_lun(struct ctl_be_lun *be_lun)
4978 {
4979         struct ctl_softc *softc;
4980         struct ctl_lun *lun;
4981
4982         lun = (struct ctl_lun *)be_lun->ctl_lun;
4983         softc = lun->ctl_softc;
4984
4985         mtx_lock(&lun->lun_lock);
4986
4987         /*
4988          * The LUN needs to be disabled before it can be marked invalid.
4989          */
4990         if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4991                 mtx_unlock(&lun->lun_lock);
4992                 return (-1);
4993         }
4994         /*
4995          * Mark the LUN invalid.
4996          */
4997         lun->flags |= CTL_LUN_INVALID;
4998
4999         /*
5000          * If there is nothing in the OOA queue, go ahead and free the LUN.
5001          * If we have something in the OOA queue, we'll free it when the
5002          * last I/O completes.
5003          */
5004         if (TAILQ_EMPTY(&lun->ooa_queue)) {
5005                 mtx_unlock(&lun->lun_lock);
5006                 mtx_lock(&softc->ctl_lock);
5007                 ctl_free_lun(lun);
5008                 mtx_unlock(&softc->ctl_lock);
5009         } else
5010                 mtx_unlock(&lun->lun_lock);
5011
5012         return (0);
5013 }
5014
5015 void
5016 ctl_lun_capacity_changed(struct ctl_be_lun *be_lun)
5017 {
5018         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
5019         union ctl_ha_msg msg;
5020
5021         mtx_lock(&lun->lun_lock);
5022         ctl_est_ua_all(lun, -1, CTL_UA_CAPACITY_CHANGE);
5023         mtx_unlock(&lun->lun_lock);
5024         if (lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
5025                 /* Send msg to other side. */
5026                 bzero(&msg.ua, sizeof(msg.ua));
5027                 msg.hdr.msg_type = CTL_MSG_UA;
5028                 msg.hdr.nexus.initid = -1;
5029                 msg.hdr.nexus.targ_port = -1;
5030                 msg.hdr.nexus.targ_lun = lun->lun;
5031                 msg.hdr.nexus.targ_mapped_lun = lun->lun;
5032                 msg.ua.ua_all = 1;
5033                 msg.ua.ua_set = 1;
5034                 msg.ua.ua_type = CTL_UA_CAPACITY_CHANGE;
5035                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg.ua),
5036                     M_WAITOK);
5037         }
5038 }
5039
5040 /*
5041  * Backend "memory move is complete" callback for requests that never
5042  * make it down to say RAIDCore's configuration code.
5043  */
5044 int
5045 ctl_config_move_done(union ctl_io *io)
5046 {
5047         int retval;
5048
5049         CTL_DEBUG_PRINT(("ctl_config_move_done\n"));
5050         KASSERT(io->io_hdr.io_type == CTL_IO_SCSI,
5051             ("Config I/O type isn't CTL_IO_SCSI (%d)!", io->io_hdr.io_type));
5052
5053         if ((io->io_hdr.port_status != 0) &&
5054             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5055              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5056                 /*
5057                  * For hardware error sense keys, the sense key
5058                  * specific value is defined to be a retry count,
5059                  * but we use it to pass back an internal FETD
5060                  * error code.  XXX KDM  Hopefully the FETD is only
5061                  * using 16 bits for an error code, since that's
5062                  * all the space we have in the sks field.
5063                  */
5064                 ctl_set_internal_failure(&io->scsiio,
5065                                          /*sks_valid*/ 1,
5066                                          /*retry_count*/
5067                                          io->io_hdr.port_status);
5068         }
5069
5070         if (ctl_debug & CTL_DEBUG_CDB_DATA)
5071                 ctl_data_print(io);
5072         if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN) ||
5073             ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
5074              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS) ||
5075             ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) {
5076                 /*
5077                  * XXX KDM just assuming a single pointer here, and not a
5078                  * S/G list.  If we start using S/G lists for config data,
5079                  * we'll need to know how to clean them up here as well.
5080                  */
5081                 if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5082                         free(io->scsiio.kern_data_ptr, M_CTL);
5083                 ctl_done(io);
5084                 retval = CTL_RETVAL_COMPLETE;
5085         } else {
5086                 /*
5087                  * XXX KDM now we need to continue data movement.  Some
5088                  * options:
5089                  * - call ctl_scsiio() again?  We don't do this for data
5090                  *   writes, because for those at least we know ahead of
5091                  *   time where the write will go and how long it is.  For
5092                  *   config writes, though, that information is largely
5093                  *   contained within the write itself, thus we need to
5094                  *   parse out the data again.
5095                  *
5096                  * - Call some other function once the data is in?
5097                  */
5098
5099                 /*
5100                  * XXX KDM call ctl_scsiio() again for now, and check flag
5101                  * bits to see whether we're allocated or not.
5102                  */
5103                 retval = ctl_scsiio(&io->scsiio);
5104         }
5105         return (retval);
5106 }
5107
5108 /*
5109  * This gets called by a backend driver when it is done with a
5110  * data_submit method.
5111  */
5112 void
5113 ctl_data_submit_done(union ctl_io *io)
5114 {
5115         /*
5116          * If the IO_CONT flag is set, we need to call the supplied
5117          * function to continue processing the I/O, instead of completing
5118          * the I/O just yet.
5119          *
5120          * If there is an error, though, we don't want to keep processing.
5121          * Instead, just send status back to the initiator.
5122          */
5123         if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5124             (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5125             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5126              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5127                 io->scsiio.io_cont(io);
5128                 return;
5129         }
5130         ctl_done(io);
5131 }
5132
5133 /*
5134  * This gets called by a backend driver when it is done with a
5135  * configuration write.
5136  */
5137 void
5138 ctl_config_write_done(union ctl_io *io)
5139 {
5140         uint8_t *buf;
5141
5142         /*
5143          * If the IO_CONT flag is set, we need to call the supplied
5144          * function to continue processing the I/O, instead of completing
5145          * the I/O just yet.
5146          *
5147          * If there is an error, though, we don't want to keep processing.
5148          * Instead, just send status back to the initiator.
5149          */
5150         if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5151             (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5152             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5153              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5154                 io->scsiio.io_cont(io);
5155                 return;
5156         }
5157         /*
5158          * Since a configuration write can be done for commands that actually
5159          * have data allocated, like write buffer, and commands that have
5160          * no data, like start/stop unit, we need to check here.
5161          */
5162         if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5163                 buf = io->scsiio.kern_data_ptr;
5164         else
5165                 buf = NULL;
5166         ctl_done(io);
5167         if (buf)
5168                 free(buf, M_CTL);
5169 }
5170
5171 void
5172 ctl_config_read_done(union ctl_io *io)
5173 {
5174         uint8_t *buf;
5175
5176         /*
5177          * If there is some error -- we are done, skip data transfer.
5178          */
5179         if ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0 ||
5180             ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
5181              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) {
5182                 if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5183                         buf = io->scsiio.kern_data_ptr;
5184                 else
5185                         buf = NULL;
5186                 ctl_done(io);
5187                 if (buf)
5188                         free(buf, M_CTL);
5189                 return;
5190         }
5191
5192         /*
5193          * If the IO_CONT flag is set, we need to call the supplied
5194          * function to continue processing the I/O, instead of completing
5195          * the I/O just yet.
5196          */
5197         if (io->io_hdr.flags & CTL_FLAG_IO_CONT) {
5198                 io->scsiio.io_cont(io);
5199                 return;
5200         }
5201
5202         ctl_datamove(io);
5203 }
5204
5205 /*
5206  * SCSI release command.
5207  */
5208 int
5209 ctl_scsi_release(struct ctl_scsiio *ctsio)
5210 {
5211         struct ctl_lun *lun = CTL_LUN(ctsio);
5212         uint32_t residx;
5213
5214         CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
5215
5216         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5217
5218         /*
5219          * XXX KDM right now, we only support LUN reservation.  We don't
5220          * support 3rd party reservations, or extent reservations, which
5221          * might actually need the parameter list.  If we've gotten this
5222          * far, we've got a LUN reservation.  Anything else got kicked out
5223          * above.  So, according to SPC, ignore the length.
5224          */
5225
5226         mtx_lock(&lun->lun_lock);
5227
5228         /*
5229          * According to SPC, it is not an error for an intiator to attempt
5230          * to release a reservation on a LUN that isn't reserved, or that
5231          * is reserved by another initiator.  The reservation can only be
5232          * released, though, by the initiator who made it or by one of
5233          * several reset type events.
5234          */
5235         if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == residx))
5236                         lun->flags &= ~CTL_LUN_RESERVED;
5237
5238         mtx_unlock(&lun->lun_lock);
5239
5240         ctl_set_success(ctsio);
5241         ctl_done((union ctl_io *)ctsio);
5242         return (CTL_RETVAL_COMPLETE);
5243 }
5244
5245 int
5246 ctl_scsi_reserve(struct ctl_scsiio *ctsio)
5247 {
5248         struct ctl_lun *lun = CTL_LUN(ctsio);
5249         uint32_t residx;
5250
5251         CTL_DEBUG_PRINT(("ctl_reserve\n"));
5252
5253         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5254
5255         /*
5256          * XXX KDM right now, we only support LUN reservation.  We don't
5257          * support 3rd party reservations, or extent reservations, which
5258          * might actually need the parameter list.  If we've gotten this
5259          * far, we've got a LUN reservation.  Anything else got kicked out
5260          * above.  So, according to SPC, ignore the length.
5261          */
5262
5263         mtx_lock(&lun->lun_lock);
5264         if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx != residx)) {
5265                 ctl_set_reservation_conflict(ctsio);
5266                 goto bailout;
5267         }
5268
5269         /* SPC-3 exceptions to SPC-2 RESERVE and RELEASE behavior. */
5270         if (lun->flags & CTL_LUN_PR_RESERVED) {
5271                 ctl_set_success(ctsio);
5272                 goto bailout;
5273         }
5274
5275         lun->flags |= CTL_LUN_RESERVED;
5276         lun->res_idx = residx;
5277         ctl_set_success(ctsio);
5278
5279 bailout:
5280         mtx_unlock(&lun->lun_lock);
5281         ctl_done((union ctl_io *)ctsio);
5282         return (CTL_RETVAL_COMPLETE);
5283 }
5284
5285 int
5286 ctl_start_stop(struct ctl_scsiio *ctsio)
5287 {
5288         struct ctl_lun *lun = CTL_LUN(ctsio);
5289         struct scsi_start_stop_unit *cdb;
5290         int retval;
5291
5292         CTL_DEBUG_PRINT(("ctl_start_stop\n"));
5293
5294         cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
5295
5296         if ((cdb->how & SSS_PC_MASK) == 0) {
5297                 if ((lun->flags & CTL_LUN_PR_RESERVED) &&
5298                     (cdb->how & SSS_START) == 0) {
5299                         uint32_t residx;
5300
5301                         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5302                         if (ctl_get_prkey(lun, residx) == 0 ||
5303                             (lun->pr_res_idx != residx && lun->pr_res_type < 4)) {
5304
5305                                 ctl_set_reservation_conflict(ctsio);
5306                                 ctl_done((union ctl_io *)ctsio);
5307                                 return (CTL_RETVAL_COMPLETE);
5308                         }
5309                 }
5310
5311                 if ((cdb->how & SSS_LOEJ) &&
5312                     (lun->flags & CTL_LUN_REMOVABLE) == 0) {
5313                         ctl_set_invalid_field(ctsio,
5314                                               /*sks_valid*/ 1,
5315                                               /*command*/ 1,
5316                                               /*field*/ 4,
5317                                               /*bit_valid*/ 1,
5318                                               /*bit*/ 1);
5319                         ctl_done((union ctl_io *)ctsio);
5320                         return (CTL_RETVAL_COMPLETE);
5321                 }
5322
5323                 if ((cdb->how & SSS_START) == 0 && (cdb->how & SSS_LOEJ) &&
5324                     lun->prevent_count > 0) {
5325                         /* "Medium removal prevented" */
5326                         ctl_set_sense(ctsio, /*current_error*/ 1,
5327                             /*sense_key*/(lun->flags & CTL_LUN_NO_MEDIA) ?
5328                              SSD_KEY_NOT_READY : SSD_KEY_ILLEGAL_REQUEST,
5329                             /*asc*/ 0x53, /*ascq*/ 0x02, SSD_ELEM_NONE);
5330                         ctl_done((union ctl_io *)ctsio);
5331                         return (CTL_RETVAL_COMPLETE);
5332                 }
5333         }
5334
5335         retval = lun->backend->config_write((union ctl_io *)ctsio);
5336         return (retval);
5337 }
5338
5339 int
5340 ctl_prevent_allow(struct ctl_scsiio *ctsio)
5341 {
5342         struct ctl_lun *lun = CTL_LUN(ctsio);
5343         struct scsi_prevent *cdb;
5344         int retval;
5345         uint32_t initidx;
5346
5347         CTL_DEBUG_PRINT(("ctl_prevent_allow\n"));
5348
5349         cdb = (struct scsi_prevent *)ctsio->cdb;
5350
5351         if ((lun->flags & CTL_LUN_REMOVABLE) == 0 || lun->prevent == NULL) {
5352                 ctl_set_invalid_opcode(ctsio);
5353                 ctl_done((union ctl_io *)ctsio);
5354                 return (CTL_RETVAL_COMPLETE);
5355         }
5356
5357         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5358         mtx_lock(&lun->lun_lock);
5359         if ((cdb->how & PR_PREVENT) &&
5360             ctl_is_set(lun->prevent, initidx) == 0) {
5361                 ctl_set_mask(lun->prevent, initidx);
5362                 lun->prevent_count++;
5363         } else if ((cdb->how & PR_PREVENT) == 0 &&
5364             ctl_is_set(lun->prevent, initidx)) {
5365                 ctl_clear_mask(lun->prevent, initidx);
5366                 lun->prevent_count--;
5367         }
5368         mtx_unlock(&lun->lun_lock);
5369         retval = lun->backend->config_write((union ctl_io *)ctsio);
5370         return (retval);
5371 }
5372
5373 /*
5374  * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
5375  * we don't really do anything with the LBA and length fields if the user
5376  * passes them in.  Instead we'll just flush out the cache for the entire
5377  * LUN.
5378  */
5379 int
5380 ctl_sync_cache(struct ctl_scsiio *ctsio)
5381 {
5382         struct ctl_lun *lun = CTL_LUN(ctsio);
5383         struct ctl_lba_len_flags *lbalen;
5384         uint64_t starting_lba;
5385         uint32_t block_count;
5386         int retval;
5387         uint8_t byte2;
5388
5389         CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
5390
5391         retval = 0;
5392
5393         switch (ctsio->cdb[0]) {
5394         case SYNCHRONIZE_CACHE: {
5395                 struct scsi_sync_cache *cdb;
5396                 cdb = (struct scsi_sync_cache *)ctsio->cdb;
5397
5398                 starting_lba = scsi_4btoul(cdb->begin_lba);
5399                 block_count = scsi_2btoul(cdb->lb_count);
5400                 byte2 = cdb->byte2;
5401                 break;
5402         }
5403         case SYNCHRONIZE_CACHE_16: {
5404                 struct scsi_sync_cache_16 *cdb;
5405                 cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5406
5407                 starting_lba = scsi_8btou64(cdb->begin_lba);
5408                 block_count = scsi_4btoul(cdb->lb_count);
5409                 byte2 = cdb->byte2;
5410                 break;
5411         }
5412         default:
5413                 ctl_set_invalid_opcode(ctsio);
5414                 ctl_done((union ctl_io *)ctsio);
5415                 goto bailout;
5416                 break; /* NOTREACHED */
5417         }
5418
5419         /*
5420          * We check the LBA and length, but don't do anything with them.
5421          * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5422          * get flushed.  This check will just help satisfy anyone who wants
5423          * to see an error for an out of range LBA.
5424          */
5425         if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5426                 ctl_set_lba_out_of_range(ctsio,
5427                     MAX(starting_lba, lun->be_lun->maxlba + 1));
5428                 ctl_done((union ctl_io *)ctsio);
5429                 goto bailout;
5430         }
5431
5432         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5433         lbalen->lba = starting_lba;
5434         lbalen->len = block_count;
5435         lbalen->flags = byte2;
5436         retval = lun->backend->config_write((union ctl_io *)ctsio);
5437
5438 bailout:
5439         return (retval);
5440 }
5441
5442 int
5443 ctl_format(struct ctl_scsiio *ctsio)
5444 {
5445         struct scsi_format *cdb;
5446         int length, defect_list_len;
5447
5448         CTL_DEBUG_PRINT(("ctl_format\n"));
5449
5450         cdb = (struct scsi_format *)ctsio->cdb;
5451
5452         length = 0;
5453         if (cdb->byte2 & SF_FMTDATA) {
5454                 if (cdb->byte2 & SF_LONGLIST)
5455                         length = sizeof(struct scsi_format_header_long);
5456                 else
5457                         length = sizeof(struct scsi_format_header_short);
5458         }
5459
5460         if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5461          && (length > 0)) {
5462                 ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5463                 ctsio->kern_data_len = length;
5464                 ctsio->kern_total_len = length;
5465                 ctsio->kern_data_resid = 0;
5466                 ctsio->kern_rel_offset = 0;
5467                 ctsio->kern_sg_entries = 0;
5468                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5469                 ctsio->be_move_done = ctl_config_move_done;
5470                 ctl_datamove((union ctl_io *)ctsio);
5471
5472                 return (CTL_RETVAL_COMPLETE);
5473         }
5474
5475         defect_list_len = 0;
5476
5477         if (cdb->byte2 & SF_FMTDATA) {
5478                 if (cdb->byte2 & SF_LONGLIST) {
5479                         struct scsi_format_header_long *header;
5480
5481                         header = (struct scsi_format_header_long *)
5482                                 ctsio->kern_data_ptr;
5483
5484                         defect_list_len = scsi_4btoul(header->defect_list_len);
5485                         if (defect_list_len != 0) {
5486                                 ctl_set_invalid_field(ctsio,
5487                                                       /*sks_valid*/ 1,
5488                                                       /*command*/ 0,
5489                                                       /*field*/ 2,
5490                                                       /*bit_valid*/ 0,
5491                                                       /*bit*/ 0);
5492                                 goto bailout;
5493                         }
5494                 } else {
5495                         struct scsi_format_header_short *header;
5496
5497                         header = (struct scsi_format_header_short *)
5498                                 ctsio->kern_data_ptr;
5499
5500                         defect_list_len = scsi_2btoul(header->defect_list_len);
5501                         if (defect_list_len != 0) {
5502                                 ctl_set_invalid_field(ctsio,
5503                                                       /*sks_valid*/ 1,
5504                                                       /*command*/ 0,
5505                                                       /*field*/ 2,
5506                                                       /*bit_valid*/ 0,
5507                                                       /*bit*/ 0);
5508                                 goto bailout;
5509                         }
5510                 }
5511         }
5512
5513         ctl_set_success(ctsio);
5514 bailout:
5515
5516         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5517                 free(ctsio->kern_data_ptr, M_CTL);
5518                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5519         }
5520
5521         ctl_done((union ctl_io *)ctsio);
5522         return (CTL_RETVAL_COMPLETE);
5523 }
5524
5525 int
5526 ctl_read_buffer(struct ctl_scsiio *ctsio)
5527 {
5528         struct ctl_lun *lun = CTL_LUN(ctsio);
5529         uint64_t buffer_offset;
5530         uint32_t len;
5531         uint8_t byte2;
5532         static uint8_t descr[4];
5533         static uint8_t echo_descr[4] = { 0 };
5534
5535         CTL_DEBUG_PRINT(("ctl_read_buffer\n"));
5536
5537         switch (ctsio->cdb[0]) {
5538         case READ_BUFFER: {
5539                 struct scsi_read_buffer *cdb;
5540
5541                 cdb = (struct scsi_read_buffer *)ctsio->cdb;
5542                 buffer_offset = scsi_3btoul(cdb->offset);
5543                 len = scsi_3btoul(cdb->length);
5544                 byte2 = cdb->byte2;
5545                 break;
5546         }
5547         case READ_BUFFER_16: {
5548                 struct scsi_read_buffer_16 *cdb;
5549
5550                 cdb = (struct scsi_read_buffer_16 *)ctsio->cdb;
5551                 buffer_offset = scsi_8btou64(cdb->offset);
5552                 len = scsi_4btoul(cdb->length);
5553                 byte2 = cdb->byte2;
5554                 break;
5555         }
5556         default: /* This shouldn't happen. */
5557                 ctl_set_invalid_opcode(ctsio);
5558                 ctl_done((union ctl_io *)ctsio);
5559                 return (CTL_RETVAL_COMPLETE);
5560         }
5561
5562         if (buffer_offset > CTL_WRITE_BUFFER_SIZE ||
5563             buffer_offset + len > CTL_WRITE_BUFFER_SIZE) {
5564                 ctl_set_invalid_field(ctsio,
5565                                       /*sks_valid*/ 1,
5566                                       /*command*/ 1,
5567                                       /*field*/ 6,
5568                                       /*bit_valid*/ 0,
5569                                       /*bit*/ 0);
5570                 ctl_done((union ctl_io *)ctsio);
5571                 return (CTL_RETVAL_COMPLETE);
5572         }
5573
5574         if ((byte2 & RWB_MODE) == RWB_MODE_DESCR) {
5575                 descr[0] = 0;
5576                 scsi_ulto3b(CTL_WRITE_BUFFER_SIZE, &descr[1]);
5577                 ctsio->kern_data_ptr = descr;
5578                 len = min(len, sizeof(descr));
5579         } else if ((byte2 & RWB_MODE) == RWB_MODE_ECHO_DESCR) {
5580                 ctsio->kern_data_ptr = echo_descr;
5581                 len = min(len, sizeof(echo_descr));
5582         } else {
5583                 if (lun->write_buffer == NULL) {
5584                         lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE,
5585                             M_CTL, M_WAITOK);
5586                 }
5587                 ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5588         }
5589         ctsio->kern_data_len = len;
5590         ctsio->kern_total_len = len;
5591         ctsio->kern_data_resid = 0;
5592         ctsio->kern_rel_offset = 0;
5593         ctsio->kern_sg_entries = 0;
5594         ctl_set_success(ctsio);
5595         ctsio->be_move_done = ctl_config_move_done;
5596         ctl_datamove((union ctl_io *)ctsio);
5597         return (CTL_RETVAL_COMPLETE);
5598 }
5599
5600 int
5601 ctl_write_buffer(struct ctl_scsiio *ctsio)
5602 {
5603         struct ctl_lun *lun = CTL_LUN(ctsio);
5604         struct scsi_write_buffer *cdb;
5605         int buffer_offset, len;
5606
5607         CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5608
5609         cdb = (struct scsi_write_buffer *)ctsio->cdb;
5610
5611         len = scsi_3btoul(cdb->length);
5612         buffer_offset = scsi_3btoul(cdb->offset);
5613
5614         if (buffer_offset + len > CTL_WRITE_BUFFER_SIZE) {
5615                 ctl_set_invalid_field(ctsio,
5616                                       /*sks_valid*/ 1,
5617                                       /*command*/ 1,
5618                                       /*field*/ 6,
5619                                       /*bit_valid*/ 0,
5620                                       /*bit*/ 0);
5621                 ctl_done((union ctl_io *)ctsio);
5622                 return (CTL_RETVAL_COMPLETE);
5623         }
5624
5625         /*
5626          * If we've got a kernel request that hasn't been malloced yet,
5627          * malloc it and tell the caller the data buffer is here.
5628          */
5629         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5630                 if (lun->write_buffer == NULL) {
5631                         lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE,
5632                             M_CTL, M_WAITOK);
5633                 }
5634                 ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5635                 ctsio->kern_data_len = len;
5636                 ctsio->kern_total_len = len;
5637                 ctsio->kern_data_resid = 0;
5638                 ctsio->kern_rel_offset = 0;
5639                 ctsio->kern_sg_entries = 0;
5640                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5641                 ctsio->be_move_done = ctl_config_move_done;
5642                 ctl_datamove((union ctl_io *)ctsio);
5643
5644                 return (CTL_RETVAL_COMPLETE);
5645         }
5646
5647         ctl_set_success(ctsio);
5648         ctl_done((union ctl_io *)ctsio);
5649         return (CTL_RETVAL_COMPLETE);
5650 }
5651
5652 int
5653 ctl_write_same(struct ctl_scsiio *ctsio)
5654 {
5655         struct ctl_lun *lun = CTL_LUN(ctsio);
5656         struct ctl_lba_len_flags *lbalen;
5657         uint64_t lba;
5658         uint32_t num_blocks;
5659         int len, retval;
5660         uint8_t byte2;
5661
5662         CTL_DEBUG_PRINT(("ctl_write_same\n"));
5663
5664         switch (ctsio->cdb[0]) {
5665         case WRITE_SAME_10: {
5666                 struct scsi_write_same_10 *cdb;
5667
5668                 cdb = (struct scsi_write_same_10 *)ctsio->cdb;
5669
5670                 lba = scsi_4btoul(cdb->addr);
5671                 num_blocks = scsi_2btoul(cdb->length);
5672                 byte2 = cdb->byte2;
5673                 break;
5674         }
5675         case WRITE_SAME_16: {
5676                 struct scsi_write_same_16 *cdb;
5677
5678                 cdb = (struct scsi_write_same_16 *)ctsio->cdb;
5679
5680                 lba = scsi_8btou64(cdb->addr);
5681                 num_blocks = scsi_4btoul(cdb->length);
5682                 byte2 = cdb->byte2;
5683                 break;
5684         }
5685         default:
5686                 /*
5687                  * We got a command we don't support.  This shouldn't
5688                  * happen, commands should be filtered out above us.
5689                  */
5690                 ctl_set_invalid_opcode(ctsio);
5691                 ctl_done((union ctl_io *)ctsio);
5692
5693                 return (CTL_RETVAL_COMPLETE);
5694                 break; /* NOTREACHED */
5695         }
5696
5697         /* ANCHOR flag can be used only together with UNMAP */
5698         if ((byte2 & SWS_UNMAP) == 0 && (byte2 & SWS_ANCHOR) != 0) {
5699                 ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
5700                     /*command*/ 1, /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 0);
5701                 ctl_done((union ctl_io *)ctsio);
5702                 return (CTL_RETVAL_COMPLETE);
5703         }
5704
5705         /*
5706          * The first check is to make sure we're in bounds, the second
5707          * check is to catch wrap-around problems.  If the lba + num blocks
5708          * is less than the lba, then we've wrapped around and the block
5709          * range is invalid anyway.
5710          */
5711         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5712          || ((lba + num_blocks) < lba)) {
5713                 ctl_set_lba_out_of_range(ctsio,
5714                     MAX(lba, lun->be_lun->maxlba + 1));
5715                 ctl_done((union ctl_io *)ctsio);
5716                 return (CTL_RETVAL_COMPLETE);
5717         }
5718
5719         /* Zero number of blocks means "to the last logical block" */
5720         if (num_blocks == 0) {
5721                 if ((lun->be_lun->maxlba + 1) - lba > UINT32_MAX) {
5722                         ctl_set_invalid_field(ctsio,
5723                                               /*sks_valid*/ 0,
5724                                               /*command*/ 1,
5725                                               /*field*/ 0,
5726                                               /*bit_valid*/ 0,
5727                                               /*bit*/ 0);
5728                         ctl_done((union ctl_io *)ctsio);
5729                         return (CTL_RETVAL_COMPLETE);
5730                 }
5731                 num_blocks = (lun->be_lun->maxlba + 1) - lba;
5732         }
5733
5734         len = lun->be_lun->blocksize;
5735
5736         /*
5737          * If we've got a kernel request that hasn't been malloced yet,
5738          * malloc it and tell the caller the data buffer is here.
5739          */
5740         if ((byte2 & SWS_NDOB) == 0 &&
5741             (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5742                 ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);
5743                 ctsio->kern_data_len = len;
5744                 ctsio->kern_total_len = len;
5745                 ctsio->kern_data_resid = 0;
5746                 ctsio->kern_rel_offset = 0;
5747                 ctsio->kern_sg_entries = 0;
5748                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5749                 ctsio->be_move_done = ctl_config_move_done;
5750                 ctl_datamove((union ctl_io *)ctsio);
5751
5752                 return (CTL_RETVAL_COMPLETE);
5753         }
5754
5755         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5756         lbalen->lba = lba;
5757         lbalen->len = num_blocks;
5758         lbalen->flags = byte2;
5759         retval = lun->backend->config_write((union ctl_io *)ctsio);
5760
5761         return (retval);
5762 }
5763
5764 int
5765 ctl_unmap(struct ctl_scsiio *ctsio)
5766 {
5767         struct ctl_lun *lun = CTL_LUN(ctsio);
5768         struct scsi_unmap *cdb;
5769         struct ctl_ptr_len_flags *ptrlen;
5770         struct scsi_unmap_header *hdr;
5771         struct scsi_unmap_desc *buf, *end, *endnz, *range;
5772         uint64_t lba;
5773         uint32_t num_blocks;
5774         int len, retval;
5775         uint8_t byte2;
5776
5777         CTL_DEBUG_PRINT(("ctl_unmap\n"));
5778
5779         cdb = (struct scsi_unmap *)ctsio->cdb;
5780         len = scsi_2btoul(cdb->length);
5781         byte2 = cdb->byte2;
5782
5783         /*
5784          * If we've got a kernel request that hasn't been malloced yet,
5785          * malloc it and tell the caller the data buffer is here.
5786          */
5787         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5788                 ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);
5789                 ctsio->kern_data_len = len;
5790                 ctsio->kern_total_len = len;
5791                 ctsio->kern_data_resid = 0;
5792                 ctsio->kern_rel_offset = 0;
5793                 ctsio->kern_sg_entries = 0;
5794                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5795                 ctsio->be_move_done = ctl_config_move_done;
5796                 ctl_datamove((union ctl_io *)ctsio);
5797
5798                 return (CTL_RETVAL_COMPLETE);
5799         }
5800
5801         len = ctsio->kern_total_len - ctsio->kern_data_resid;
5802         hdr = (struct scsi_unmap_header *)ctsio->kern_data_ptr;
5803         if (len < sizeof (*hdr) ||
5804             len < (scsi_2btoul(hdr->length) + sizeof(hdr->length)) ||
5805             len < (scsi_2btoul(hdr->desc_length) + sizeof (*hdr)) ||
5806             scsi_2btoul(hdr->desc_length) % sizeof(*buf) != 0) {
5807                 ctl_set_invalid_field(ctsio,
5808                                       /*sks_valid*/ 0,
5809                                       /*command*/ 0,
5810                                       /*field*/ 0,
5811                                       /*bit_valid*/ 0,
5812                                       /*bit*/ 0);
5813                 goto done;
5814         }
5815         len = scsi_2btoul(hdr->desc_length);
5816         buf = (struct scsi_unmap_desc *)(hdr + 1);
5817         end = buf + len / sizeof(*buf);
5818
5819         endnz = buf;
5820         for (range = buf; range < end; range++) {
5821                 lba = scsi_8btou64(range->lba);
5822                 num_blocks = scsi_4btoul(range->length);
5823                 if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5824                  || ((lba + num_blocks) < lba)) {
5825                         ctl_set_lba_out_of_range(ctsio,
5826                             MAX(lba, lun->be_lun->maxlba + 1));
5827                         ctl_done((union ctl_io *)ctsio);
5828                         return (CTL_RETVAL_COMPLETE);
5829                 }
5830                 if (num_blocks != 0)
5831                         endnz = range + 1;
5832         }
5833
5834         /*
5835          * Block backend can not handle zero last range.
5836          * Filter it out and return if there is nothing left.
5837          */
5838         len = (uint8_t *)endnz - (uint8_t *)buf;
5839         if (len == 0) {
5840                 ctl_set_success(ctsio);
5841                 goto done;
5842         }
5843
5844         mtx_lock(&lun->lun_lock);
5845         ptrlen = (struct ctl_ptr_len_flags *)
5846             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5847         ptrlen->ptr = (void *)buf;
5848         ptrlen->len = len;
5849         ptrlen->flags = byte2;
5850         ctl_check_blocked(lun);
5851         mtx_unlock(&lun->lun_lock);
5852
5853         retval = lun->backend->config_write((union ctl_io *)ctsio);
5854         return (retval);
5855
5856 done:
5857         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5858                 free(ctsio->kern_data_ptr, M_CTL);
5859                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5860         }
5861         ctl_done((union ctl_io *)ctsio);
5862         return (CTL_RETVAL_COMPLETE);
5863 }
5864
5865 int
5866 ctl_default_page_handler(struct ctl_scsiio *ctsio,
5867                          struct ctl_page_index *page_index, uint8_t *page_ptr)
5868 {
5869         struct ctl_lun *lun = CTL_LUN(ctsio);
5870         uint8_t *current_cp;
5871         int set_ua;
5872         uint32_t initidx;
5873
5874         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5875         set_ua = 0;
5876
5877         current_cp = (page_index->page_data + (page_index->page_len *
5878             CTL_PAGE_CURRENT));
5879
5880         mtx_lock(&lun->lun_lock);
5881         if (memcmp(current_cp, page_ptr, page_index->page_len)) {
5882                 memcpy(current_cp, page_ptr, page_index->page_len);
5883                 set_ua = 1;
5884         }
5885         if (set_ua != 0)
5886                 ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
5887         mtx_unlock(&lun->lun_lock);
5888         if (set_ua) {
5889                 ctl_isc_announce_mode(lun,
5890                     ctl_get_initindex(&ctsio->io_hdr.nexus),
5891                     page_index->page_code, page_index->subpage);
5892         }
5893         return (CTL_RETVAL_COMPLETE);
5894 }
5895
5896 static void
5897 ctl_ie_timer(void *arg)
5898 {
5899         struct ctl_lun *lun = arg;
5900         uint64_t t;
5901
5902         if (lun->ie_asc == 0)
5903                 return;
5904
5905         if (lun->MODE_IE.mrie == SIEP_MRIE_UA)
5906                 ctl_est_ua_all(lun, -1, CTL_UA_IE);
5907         else
5908                 lun->ie_reported = 0;
5909
5910         if (lun->ie_reportcnt < scsi_4btoul(lun->MODE_IE.report_count)) {
5911                 lun->ie_reportcnt++;
5912                 t = scsi_4btoul(lun->MODE_IE.interval_timer);
5913                 if (t == 0 || t == UINT32_MAX)
5914                         t = 3000;  /* 5 min */
5915                 callout_schedule(&lun->ie_callout, t * hz / 10);
5916         }
5917 }
5918
5919 int
5920 ctl_ie_page_handler(struct ctl_scsiio *ctsio,
5921                          struct ctl_page_index *page_index, uint8_t *page_ptr)
5922 {
5923         struct ctl_lun *lun = CTL_LUN(ctsio);
5924         struct scsi_info_exceptions_page *pg;
5925         uint64_t t;
5926
5927         (void)ctl_default_page_handler(ctsio, page_index, page_ptr);
5928
5929         pg = (struct scsi_info_exceptions_page *)page_ptr;
5930         mtx_lock(&lun->lun_lock);
5931         if (pg->info_flags & SIEP_FLAGS_TEST) {
5932                 lun->ie_asc = 0x5d;
5933                 lun->ie_ascq = 0xff;
5934                 if (pg->mrie == SIEP_MRIE_UA) {
5935                         ctl_est_ua_all(lun, -1, CTL_UA_IE);
5936                         lun->ie_reported = 1;
5937                 } else {
5938                         ctl_clr_ua_all(lun, -1, CTL_UA_IE);
5939                         lun->ie_reported = -1;
5940                 }
5941                 lun->ie_reportcnt = 1;
5942                 if (lun->ie_reportcnt < scsi_4btoul(pg->report_count)) {
5943                         lun->ie_reportcnt++;
5944                         t = scsi_4btoul(pg->interval_timer);
5945                         if (t == 0 || t == UINT32_MAX)
5946                                 t = 3000;  /* 5 min */
5947                         callout_reset(&lun->ie_callout, t * hz / 10,
5948                             ctl_ie_timer, lun);
5949                 }
5950         } else {
5951                 lun->ie_asc = 0;
5952                 lun->ie_ascq = 0;
5953                 lun->ie_reported = 1;
5954                 ctl_clr_ua_all(lun, -1, CTL_UA_IE);
5955                 lun->ie_reportcnt = UINT32_MAX;
5956                 callout_stop(&lun->ie_callout);
5957         }
5958         mtx_unlock(&lun->lun_lock);
5959         return (CTL_RETVAL_COMPLETE);
5960 }
5961
5962 static int
5963 ctl_do_mode_select(union ctl_io *io)
5964 {
5965         struct ctl_lun *lun = CTL_LUN(io);
5966         struct scsi_mode_page_header *page_header;
5967         struct ctl_page_index *page_index;
5968         struct ctl_scsiio *ctsio;
5969         int page_len, page_len_offset, page_len_size;
5970         union ctl_modepage_info *modepage_info;
5971         uint16_t *len_left, *len_used;
5972         int retval, i;
5973
5974         ctsio = &io->scsiio;
5975         page_index = NULL;
5976         page_len = 0;
5977
5978         modepage_info = (union ctl_modepage_info *)
5979                 ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
5980         len_left = &modepage_info->header.len_left;
5981         len_used = &modepage_info->header.len_used;
5982
5983 do_next_page:
5984
5985         page_header = (struct scsi_mode_page_header *)
5986                 (ctsio->kern_data_ptr + *len_used);
5987
5988         if (*len_left == 0) {
5989                 free(ctsio->kern_data_ptr, M_CTL);
5990                 ctl_set_success(ctsio);
5991                 ctl_done((union ctl_io *)ctsio);
5992                 return (CTL_RETVAL_COMPLETE);
5993         } else if (*len_left < sizeof(struct scsi_mode_page_header)) {
5994
5995                 free(ctsio->kern_data_ptr, M_CTL);
5996                 ctl_set_param_len_error(ctsio);
5997                 ctl_done((union ctl_io *)ctsio);
5998                 return (CTL_RETVAL_COMPLETE);
5999
6000         } else if ((page_header->page_code & SMPH_SPF)
6001                 && (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
6002
6003                 free(ctsio->kern_data_ptr, M_CTL);
6004                 ctl_set_param_len_error(ctsio);
6005                 ctl_done((union ctl_io *)ctsio);
6006                 return (CTL_RETVAL_COMPLETE);
6007         }
6008
6009
6010         /*
6011          * XXX KDM should we do something with the block descriptor?
6012          */
6013         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6014                 page_index = &lun->mode_pages.index[i];
6015                 if (lun->be_lun->lun_type == T_DIRECT &&
6016                     (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6017                         continue;
6018                 if (lun->be_lun->lun_type == T_PROCESSOR &&
6019                     (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6020                         continue;
6021                 if (lun->be_lun->lun_type == T_CDROM &&
6022                     (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6023                         continue;
6024
6025                 if ((page_index->page_code & SMPH_PC_MASK) !=
6026                     (page_header->page_code & SMPH_PC_MASK))
6027                         continue;
6028
6029                 /*
6030                  * If neither page has a subpage code, then we've got a
6031                  * match.
6032                  */
6033                 if (((page_index->page_code & SMPH_SPF) == 0)
6034                  && ((page_header->page_code & SMPH_SPF) == 0)) {
6035                         page_len = page_header->page_length;
6036                         break;
6037                 }
6038
6039                 /*
6040                  * If both pages have subpages, then the subpage numbers
6041                  * have to match.
6042                  */
6043                 if ((page_index->page_code & SMPH_SPF)
6044                   && (page_header->page_code & SMPH_SPF)) {
6045                         struct scsi_mode_page_header_sp *sph;
6046
6047                         sph = (struct scsi_mode_page_header_sp *)page_header;
6048                         if (page_index->subpage == sph->subpage) {
6049                                 page_len = scsi_2btoul(sph->page_length);
6050                                 break;
6051                         }
6052                 }
6053         }
6054
6055         /*
6056          * If we couldn't find the page, or if we don't have a mode select
6057          * handler for it, send back an error to the user.
6058          */
6059         if ((i >= CTL_NUM_MODE_PAGES)
6060          || (page_index->select_handler == NULL)) {
6061                 ctl_set_invalid_field(ctsio,
6062                                       /*sks_valid*/ 1,
6063                                       /*command*/ 0,
6064                                       /*field*/ *len_used,
6065                                       /*bit_valid*/ 0,
6066                                       /*bit*/ 0);
6067                 free(ctsio->kern_data_ptr, M_CTL);
6068                 ctl_done((union ctl_io *)ctsio);
6069                 return (CTL_RETVAL_COMPLETE);
6070         }
6071
6072         if (page_index->page_code & SMPH_SPF) {
6073                 page_len_offset = 2;
6074                 page_len_size = 2;
6075         } else {
6076                 page_len_size = 1;
6077                 page_len_offset = 1;
6078         }
6079
6080         /*
6081          * If the length the initiator gives us isn't the one we specify in
6082          * the mode page header, or if they didn't specify enough data in
6083          * the CDB to avoid truncating this page, kick out the request.
6084          */
6085         if (page_len != page_index->page_len - page_len_offset - page_len_size) {
6086                 ctl_set_invalid_field(ctsio,
6087                                       /*sks_valid*/ 1,
6088                                       /*command*/ 0,
6089                                       /*field*/ *len_used + page_len_offset,
6090                                       /*bit_valid*/ 0,
6091                                       /*bit*/ 0);
6092                 free(ctsio->kern_data_ptr, M_CTL);
6093                 ctl_done((union ctl_io *)ctsio);
6094                 return (CTL_RETVAL_COMPLETE);
6095         }
6096         if (*len_left < page_index->page_len) {
6097                 free(ctsio->kern_data_ptr, M_CTL);
6098                 ctl_set_param_len_error(ctsio);
6099                 ctl_done((union ctl_io *)ctsio);
6100                 return (CTL_RETVAL_COMPLETE);
6101         }
6102
6103         /*
6104          * Run through the mode page, checking to make sure that the bits
6105          * the user changed are actually legal for him to change.
6106          */
6107         for (i = 0; i < page_index->page_len; i++) {
6108                 uint8_t *user_byte, *change_mask, *current_byte;
6109                 int bad_bit;
6110                 int j;
6111
6112                 user_byte = (uint8_t *)page_header + i;
6113                 change_mask = page_index->page_data +
6114                               (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
6115                 current_byte = page_index->page_data +
6116                                (page_index->page_len * CTL_PAGE_CURRENT) + i;
6117
6118                 /*
6119                  * Check to see whether the user set any bits in this byte
6120                  * that he is not allowed to set.
6121                  */
6122                 if ((*user_byte & ~(*change_mask)) ==
6123                     (*current_byte & ~(*change_mask)))
6124                         continue;
6125
6126                 /*
6127                  * Go through bit by bit to determine which one is illegal.
6128                  */
6129                 bad_bit = 0;
6130                 for (j = 7; j >= 0; j--) {
6131                         if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
6132                             (((1 << i) & ~(*change_mask)) & *current_byte)) {
6133                                 bad_bit = i;
6134                                 break;
6135                         }
6136                 }
6137                 ctl_set_invalid_field(ctsio,
6138                                       /*sks_valid*/ 1,
6139                                       /*command*/ 0,
6140                                       /*field*/ *len_used + i,
6141                                       /*bit_valid*/ 1,
6142                                       /*bit*/ bad_bit);
6143                 free(ctsio->kern_data_ptr, M_CTL);
6144                 ctl_done((union ctl_io *)ctsio);
6145                 return (CTL_RETVAL_COMPLETE);
6146         }
6147
6148         /*
6149          * Decrement these before we call the page handler, since we may
6150          * end up getting called back one way or another before the handler
6151          * returns to this context.
6152          */
6153         *len_left -= page_index->page_len;
6154         *len_used += page_index->page_len;
6155
6156         retval = page_index->select_handler(ctsio, page_index,
6157                                             (uint8_t *)page_header);
6158
6159         /*
6160          * If the page handler returns CTL_RETVAL_QUEUED, then we need to
6161          * wait until this queued command completes to finish processing
6162          * the mode page.  If it returns anything other than
6163          * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
6164          * already set the sense information, freed the data pointer, and
6165          * completed the io for us.
6166          */
6167         if (retval != CTL_RETVAL_COMPLETE)
6168                 goto bailout_no_done;
6169
6170         /*
6171          * If the initiator sent us more than one page, parse the next one.
6172          */
6173         if (*len_left > 0)
6174                 goto do_next_page;
6175
6176         ctl_set_success(ctsio);
6177         free(ctsio->kern_data_ptr, M_CTL);
6178         ctl_done((union ctl_io *)ctsio);
6179
6180 bailout_no_done:
6181
6182         return (CTL_RETVAL_COMPLETE);
6183
6184 }
6185
6186 int
6187 ctl_mode_select(struct ctl_scsiio *ctsio)
6188 {
6189         struct ctl_lun *lun = CTL_LUN(ctsio);
6190         union ctl_modepage_info *modepage_info;
6191         int bd_len, i, header_size, param_len, pf, rtd, sp;
6192         uint32_t initidx;
6193
6194         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
6195         switch (ctsio->cdb[0]) {
6196         case MODE_SELECT_6: {
6197                 struct scsi_mode_select_6 *cdb;
6198
6199                 cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
6200
6201                 pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6202                 rtd = (cdb->byte2 & SMS_RTD) ? 1 : 0;
6203                 sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6204                 param_len = cdb->length;
6205                 header_size = sizeof(struct scsi_mode_header_6);
6206                 break;
6207         }
6208         case MODE_SELECT_10: {
6209                 struct scsi_mode_select_10 *cdb;
6210
6211                 cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
6212
6213                 pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6214                 rtd = (cdb->byte2 & SMS_RTD) ? 1 : 0;
6215                 sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6216                 param_len = scsi_2btoul(cdb->length);
6217                 header_size = sizeof(struct scsi_mode_header_10);
6218                 break;
6219         }
6220         default:
6221                 ctl_set_invalid_opcode(ctsio);
6222                 ctl_done((union ctl_io *)ctsio);
6223                 return (CTL_RETVAL_COMPLETE);
6224         }
6225
6226         if (rtd) {
6227                 if (param_len != 0) {
6228                         ctl_set_invalid_field(ctsio, /*sks_valid*/ 0,
6229                             /*command*/ 1, /*field*/ 0,
6230                             /*bit_valid*/ 0, /*bit*/ 0);
6231                         ctl_done((union ctl_io *)ctsio);
6232                         return (CTL_RETVAL_COMPLETE);
6233                 }
6234
6235                 /* Revert to defaults. */
6236                 ctl_init_page_index(lun);
6237                 mtx_lock(&lun->lun_lock);
6238                 ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
6239                 mtx_unlock(&lun->lun_lock);
6240                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6241                         ctl_isc_announce_mode(lun, -1,
6242                             lun->mode_pages.index[i].page_code & SMPH_PC_MASK,
6243                             lun->mode_pages.index[i].subpage);
6244                 }
6245                 ctl_set_success(ctsio);
6246                 ctl_done((union ctl_io *)ctsio);
6247                 return (CTL_RETVAL_COMPLETE);
6248         }
6249
6250         /*
6251          * From SPC-3:
6252          * "A parameter list length of zero indicates that the Data-Out Buffer
6253          * shall be empty. This condition shall not be considered as an error."
6254          */
6255         if (param_len == 0) {
6256                 ctl_set_success(ctsio);
6257                 ctl_done((union ctl_io *)ctsio);
6258                 return (CTL_RETVAL_COMPLETE);
6259         }
6260
6261         /*
6262          * Since we'll hit this the first time through, prior to
6263          * allocation, we don't need to free a data buffer here.
6264          */
6265         if (param_len < header_size) {
6266                 ctl_set_param_len_error(ctsio);
6267                 ctl_done((union ctl_io *)ctsio);
6268                 return (CTL_RETVAL_COMPLETE);
6269         }
6270
6271         /*
6272          * Allocate the data buffer and grab the user's data.  In theory,
6273          * we shouldn't have to sanity check the parameter list length here
6274          * because the maximum size is 64K.  We should be able to malloc
6275          * that much without too many problems.
6276          */
6277         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6278                 ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6279                 ctsio->kern_data_len = param_len;
6280                 ctsio->kern_total_len = param_len;
6281                 ctsio->kern_data_resid = 0;
6282                 ctsio->kern_rel_offset = 0;
6283                 ctsio->kern_sg_entries = 0;
6284                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6285                 ctsio->be_move_done = ctl_config_move_done;
6286                 ctl_datamove((union ctl_io *)ctsio);
6287
6288                 return (CTL_RETVAL_COMPLETE);
6289         }
6290
6291         switch (ctsio->cdb[0]) {
6292         case MODE_SELECT_6: {
6293                 struct scsi_mode_header_6 *mh6;
6294
6295                 mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6296                 bd_len = mh6->blk_desc_len;
6297                 break;
6298         }
6299         case MODE_SELECT_10: {
6300                 struct scsi_mode_header_10 *mh10;
6301
6302                 mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6303                 bd_len = scsi_2btoul(mh10->blk_desc_len);
6304                 break;
6305         }
6306         default:
6307                 panic("%s: Invalid CDB type %#x", __func__, ctsio->cdb[0]);
6308         }
6309
6310         if (param_len < (header_size + bd_len)) {
6311                 free(ctsio->kern_data_ptr, M_CTL);
6312                 ctl_set_param_len_error(ctsio);
6313                 ctl_done((union ctl_io *)ctsio);
6314                 return (CTL_RETVAL_COMPLETE);
6315         }
6316
6317         /*
6318          * Set the IO_CONT flag, so that if this I/O gets passed to
6319          * ctl_config_write_done(), it'll get passed back to
6320          * ctl_do_mode_select() for further processing, or completion if
6321          * we're all done.
6322          */
6323         ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6324         ctsio->io_cont = ctl_do_mode_select;
6325
6326         modepage_info = (union ctl_modepage_info *)
6327                 ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6328         memset(modepage_info, 0, sizeof(*modepage_info));
6329         modepage_info->header.len_left = param_len - header_size - bd_len;
6330         modepage_info->header.len_used = header_size + bd_len;
6331
6332         return (ctl_do_mode_select((union ctl_io *)ctsio));
6333 }
6334
6335 int
6336 ctl_mode_sense(struct ctl_scsiio *ctsio)
6337 {
6338         struct ctl_lun *lun = CTL_LUN(ctsio);
6339         int pc, page_code, dbd, llba, subpage;
6340         int alloc_len, page_len, header_len, total_len;
6341         struct scsi_mode_block_descr *block_desc;
6342         struct ctl_page_index *page_index;
6343
6344         dbd = 0;
6345         llba = 0;
6346         block_desc = NULL;
6347
6348         CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6349
6350         switch (ctsio->cdb[0]) {
6351         case MODE_SENSE_6: {
6352                 struct scsi_mode_sense_6 *cdb;
6353
6354                 cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6355
6356                 header_len = sizeof(struct scsi_mode_hdr_6);
6357                 if (cdb->byte2 & SMS_DBD)
6358                         dbd = 1;
6359                 else
6360                         header_len += sizeof(struct scsi_mode_block_descr);
6361
6362                 pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6363                 page_code = cdb->page & SMS_PAGE_CODE;
6364                 subpage = cdb->subpage;
6365                 alloc_len = cdb->length;
6366                 break;
6367         }
6368         case MODE_SENSE_10: {
6369                 struct scsi_mode_sense_10 *cdb;
6370
6371                 cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6372
6373                 header_len = sizeof(struct scsi_mode_hdr_10);
6374
6375                 if (cdb->byte2 & SMS_DBD)
6376                         dbd = 1;
6377                 else
6378                         header_len += sizeof(struct scsi_mode_block_descr);
6379                 if (cdb->byte2 & SMS10_LLBAA)
6380                         llba = 1;
6381                 pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6382                 page_code = cdb->page & SMS_PAGE_CODE;
6383                 subpage = cdb->subpage;
6384                 alloc_len = scsi_2btoul(cdb->length);
6385                 break;
6386         }
6387         default:
6388                 ctl_set_invalid_opcode(ctsio);
6389                 ctl_done((union ctl_io *)ctsio);
6390                 return (CTL_RETVAL_COMPLETE);
6391                 break; /* NOTREACHED */
6392         }
6393
6394         /*
6395          * We have to make a first pass through to calculate the size of
6396          * the pages that match the user's query.  Then we allocate enough
6397          * memory to hold it, and actually copy the data into the buffer.
6398          */
6399         switch (page_code) {
6400         case SMS_ALL_PAGES_PAGE: {
6401                 u_int i;
6402
6403                 page_len = 0;
6404
6405                 /*
6406                  * At the moment, values other than 0 and 0xff here are
6407                  * reserved according to SPC-3.
6408                  */
6409                 if ((subpage != SMS_SUBPAGE_PAGE_0)
6410                  && (subpage != SMS_SUBPAGE_ALL)) {
6411                         ctl_set_invalid_field(ctsio,
6412                                               /*sks_valid*/ 1,
6413                                               /*command*/ 1,
6414                                               /*field*/ 3,
6415                                               /*bit_valid*/ 0,
6416                                               /*bit*/ 0);
6417                         ctl_done((union ctl_io *)ctsio);
6418                         return (CTL_RETVAL_COMPLETE);
6419                 }
6420
6421                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6422                         page_index = &lun->mode_pages.index[i];
6423
6424                         /* Make sure the page is supported for this dev type */
6425                         if (lun->be_lun->lun_type == T_DIRECT &&
6426                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6427                                 continue;
6428                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6429                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6430                                 continue;
6431                         if (lun->be_lun->lun_type == T_CDROM &&
6432                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6433                                 continue;
6434
6435                         /*
6436                          * We don't use this subpage if the user didn't
6437                          * request all subpages.
6438                          */
6439                         if ((page_index->subpage != 0)
6440                          && (subpage == SMS_SUBPAGE_PAGE_0))
6441                                 continue;
6442
6443 #if 0
6444                         printf("found page %#x len %d\n",
6445                                page_index->page_code & SMPH_PC_MASK,
6446                                page_index->page_len);
6447 #endif
6448                         page_len += page_index->page_len;
6449                 }
6450                 break;
6451         }
6452         default: {
6453                 u_int i;
6454
6455                 page_len = 0;
6456
6457                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6458                         page_index = &lun->mode_pages.index[i];
6459
6460                         /* Make sure the page is supported for this dev type */
6461                         if (lun->be_lun->lun_type == T_DIRECT &&
6462                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6463                                 continue;
6464                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6465                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6466                                 continue;
6467                         if (lun->be_lun->lun_type == T_CDROM &&
6468                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6469                                 continue;
6470
6471                         /* Look for the right page code */
6472                         if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6473                                 continue;
6474
6475                         /* Look for the right subpage or the subpage wildcard*/
6476                         if ((page_index->subpage != subpage)
6477                          && (subpage != SMS_SUBPAGE_ALL))
6478                                 continue;
6479
6480 #if 0
6481                         printf("found page %#x len %d\n",
6482                                page_index->page_code & SMPH_PC_MASK,
6483                                page_index->page_len);
6484 #endif
6485
6486                         page_len += page_index->page_len;
6487                 }
6488
6489                 if (page_len == 0) {
6490                         ctl_set_invalid_field(ctsio,
6491                                               /*sks_valid*/ 1,
6492                                               /*command*/ 1,
6493                                               /*field*/ 2,
6494                                               /*bit_valid*/ 1,
6495                                               /*bit*/ 5);
6496                         ctl_done((union ctl_io *)ctsio);
6497                         return (CTL_RETVAL_COMPLETE);
6498                 }
6499                 break;
6500         }
6501         }
6502
6503         total_len = header_len + page_len;
6504 #if 0
6505         printf("header_len = %d, page_len = %d, total_len = %d\n",
6506                header_len, page_len, total_len);
6507 #endif
6508
6509         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6510         ctsio->kern_sg_entries = 0;
6511         ctsio->kern_data_resid = 0;
6512         ctsio->kern_rel_offset = 0;
6513         if (total_len < alloc_len) {
6514                 ctsio->residual = alloc_len - total_len;
6515                 ctsio->kern_data_len = total_len;
6516                 ctsio->kern_total_len = total_len;
6517         } else {
6518                 ctsio->residual = 0;
6519                 ctsio->kern_data_len = alloc_len;
6520                 ctsio->kern_total_len = alloc_len;
6521         }
6522
6523         switch (ctsio->cdb[0]) {
6524         case MODE_SENSE_6: {
6525                 struct scsi_mode_hdr_6 *header;
6526
6527                 header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
6528
6529                 header->datalen = MIN(total_len - 1, 254);
6530                 if (lun->be_lun->lun_type == T_DIRECT) {
6531                         header->dev_specific = 0x10; /* DPOFUA */
6532                         if ((lun->be_lun->flags & CTL_LUN_FLAG_READONLY) ||
6533                             (lun->MODE_CTRL.eca_and_aen & SCP_SWP) != 0)
6534                                 header->dev_specific |= 0x80; /* WP */
6535                 }
6536                 if (dbd)
6537                         header->block_descr_len = 0;
6538                 else
6539                         header->block_descr_len =
6540                                 sizeof(struct scsi_mode_block_descr);
6541                 block_desc = (struct scsi_mode_block_descr *)&header[1];
6542                 break;
6543         }
6544         case MODE_SENSE_10: {
6545                 struct scsi_mode_hdr_10 *header;
6546                 int datalen;
6547
6548                 header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
6549
6550                 datalen = MIN(total_len - 2, 65533);
6551                 scsi_ulto2b(datalen, header->datalen);
6552                 if (lun->be_lun->lun_type == T_DIRECT) {
6553                         header->dev_specific = 0x10; /* DPOFUA */
6554                         if ((lun->be_lun->flags & CTL_LUN_FLAG_READONLY) ||
6555                             (lun->MODE_CTRL.eca_and_aen & SCP_SWP) != 0)
6556                                 header->dev_specific |= 0x80; /* WP */
6557                 }
6558                 if (dbd)
6559                         scsi_ulto2b(0, header->block_descr_len);
6560                 else
6561                         scsi_ulto2b(sizeof(struct scsi_mode_block_descr),
6562                                     header->block_descr_len);
6563                 block_desc = (struct scsi_mode_block_descr *)&header[1];
6564                 break;
6565         }
6566         default:
6567                 panic("%s: Invalid CDB type %#x", __func__, ctsio->cdb[0]);
6568         }
6569
6570         /*
6571          * If we've got a disk, use its blocksize in the block
6572          * descriptor.  Otherwise, just set it to 0.
6573          */
6574         if (dbd == 0) {
6575                 if (lun->be_lun->lun_type == T_DIRECT)
6576                         scsi_ulto3b(lun->be_lun->blocksize,
6577                                     block_desc->block_len);
6578                 else
6579                         scsi_ulto3b(0, block_desc->block_len);
6580         }
6581
6582         switch (page_code) {
6583         case SMS_ALL_PAGES_PAGE: {
6584                 int i, data_used;
6585
6586                 data_used = header_len;
6587                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6588                         struct ctl_page_index *page_index;
6589
6590                         page_index = &lun->mode_pages.index[i];
6591                         if (lun->be_lun->lun_type == T_DIRECT &&
6592                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6593                                 continue;
6594                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6595                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6596                                 continue;
6597                         if (lun->be_lun->lun_type == T_CDROM &&
6598                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6599                                 continue;
6600
6601                         /*
6602                          * We don't use this subpage if the user didn't
6603                          * request all subpages.  We already checked (above)
6604                          * to make sure the user only specified a subpage
6605                          * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case.
6606                          */
6607                         if ((page_index->subpage != 0)
6608                          && (subpage == SMS_SUBPAGE_PAGE_0))
6609                                 continue;
6610
6611                         /*
6612                          * Call the handler, if it exists, to update the
6613                          * page to the latest values.
6614                          */
6615                         if (page_index->sense_handler != NULL)
6616                                 page_index->sense_handler(ctsio, page_index,pc);
6617
6618                         memcpy(ctsio->kern_data_ptr + data_used,
6619                                page_index->page_data +
6620                                (page_index->page_len * pc),
6621                                page_index->page_len);
6622                         data_used += page_index->page_len;
6623                 }
6624                 break;
6625         }
6626         default: {
6627                 int i, data_used;
6628
6629                 data_used = header_len;
6630
6631                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6632                         struct ctl_page_index *page_index;
6633
6634                         page_index = &lun->mode_pages.index[i];
6635
6636                         /* Look for the right page code */
6637                         if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6638                                 continue;
6639
6640                         /* Look for the right subpage or the subpage wildcard*/
6641                         if ((page_index->subpage != subpage)
6642                          && (subpage != SMS_SUBPAGE_ALL))
6643                                 continue;
6644
6645                         /* Make sure the page is supported for this dev type */
6646                         if (lun->be_lun->lun_type == T_DIRECT &&
6647                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6648                                 continue;
6649                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6650                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6651                                 continue;
6652                         if (lun->be_lun->lun_type == T_CDROM &&
6653                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6654                                 continue;
6655
6656                         /*
6657                          * Call the handler, if it exists, to update the
6658                          * page to the latest values.
6659                          */
6660                         if (page_index->sense_handler != NULL)
6661                                 page_index->sense_handler(ctsio, page_index,pc);
6662
6663                         memcpy(ctsio->kern_data_ptr + data_used,
6664                                page_index->page_data +
6665                                (page_index->page_len * pc),
6666                                page_index->page_len);
6667                         data_used += page_index->page_len;
6668                 }
6669                 break;
6670         }
6671         }
6672
6673         ctl_set_success(ctsio);
6674         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6675         ctsio->be_move_done = ctl_config_move_done;
6676         ctl_datamove((union ctl_io *)ctsio);
6677         return (CTL_RETVAL_COMPLETE);
6678 }
6679
6680 int
6681 ctl_lbp_log_sense_handler(struct ctl_scsiio *ctsio,
6682                                struct ctl_page_index *page_index,
6683                                int pc)
6684 {
6685         struct ctl_lun *lun = CTL_LUN(ctsio);
6686         struct scsi_log_param_header *phdr;
6687         uint8_t *data;
6688         uint64_t val;
6689
6690         data = page_index->page_data;
6691
6692         if (lun->backend->lun_attr != NULL &&
6693             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "blocksavail"))
6694              != UINT64_MAX) {
6695                 phdr = (struct scsi_log_param_header *)data;
6696                 scsi_ulto2b(0x0001, phdr->param_code);
6697                 phdr->param_control = SLP_LBIN | SLP_LP;
6698                 phdr->param_len = 8;
6699                 data = (uint8_t *)(phdr + 1);
6700                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6701                 data[4] = 0x02; /* per-pool */
6702                 data += phdr->param_len;
6703         }
6704
6705         if (lun->backend->lun_attr != NULL &&
6706             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "blocksused"))
6707              != UINT64_MAX) {
6708                 phdr = (struct scsi_log_param_header *)data;
6709                 scsi_ulto2b(0x0002, phdr->param_code);
6710                 phdr->param_control = SLP_LBIN | SLP_LP;
6711                 phdr->param_len = 8;
6712                 data = (uint8_t *)(phdr + 1);
6713                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6714                 data[4] = 0x01; /* per-LUN */
6715                 data += phdr->param_len;
6716         }
6717
6718         if (lun->backend->lun_attr != NULL &&
6719             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "poolblocksavail"))
6720              != UINT64_MAX) {
6721                 phdr = (struct scsi_log_param_header *)data;
6722                 scsi_ulto2b(0x00f1, phdr->param_code);
6723                 phdr->param_control = SLP_LBIN | SLP_LP;
6724                 phdr->param_len = 8;
6725                 data = (uint8_t *)(phdr + 1);
6726                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6727                 data[4] = 0x02; /* per-pool */
6728                 data += phdr->param_len;
6729         }
6730
6731         if (lun->backend->lun_attr != NULL &&
6732             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "poolblocksused"))
6733              != UINT64_MAX) {
6734                 phdr = (struct scsi_log_param_header *)data;
6735                 scsi_ulto2b(0x00f2, phdr->param_code);
6736                 phdr->param_control = SLP_LBIN | SLP_LP;
6737                 phdr->param_len = 8;
6738                 data = (uint8_t *)(phdr + 1);
6739                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6740                 data[4] = 0x02; /* per-pool */
6741                 data += phdr->param_len;
6742         }
6743
6744         page_index->page_len = data - page_index->page_data;
6745         return (0);
6746 }
6747
6748 int
6749 ctl_sap_log_sense_handler(struct ctl_scsiio *ctsio,
6750                                struct ctl_page_index *page_index,
6751                                int pc)
6752 {
6753         struct ctl_lun *lun = CTL_LUN(ctsio);
6754         struct stat_page *data;
6755         struct bintime *t;
6756
6757         data = (struct stat_page *)page_index->page_data;
6758
6759         scsi_ulto2b(SLP_SAP, data->sap.hdr.param_code);
6760         data->sap.hdr.param_control = SLP_LBIN;
6761         data->sap.hdr.param_len = sizeof(struct scsi_log_stat_and_perf) -
6762             sizeof(struct scsi_log_param_header);
6763         scsi_u64to8b(lun->stats.operations[CTL_STATS_READ],
6764             data->sap.read_num);
6765         scsi_u64to8b(lun->stats.operations[CTL_STATS_WRITE],
6766             data->sap.write_num);
6767         if (lun->be_lun->blocksize > 0) {
6768                 scsi_u64to8b(lun->stats.bytes[CTL_STATS_WRITE] /
6769                     lun->be_lun->blocksize, data->sap.recvieved_lba);
6770                 scsi_u64to8b(lun->stats.bytes[CTL_STATS_READ] /
6771                     lun->be_lun->blocksize, data->sap.transmitted_lba);
6772         }
6773         t = &lun->stats.time[CTL_STATS_READ];
6774         scsi_u64to8b((uint64_t)t->sec * 1000 + t->frac / (UINT64_MAX / 1000),
6775             data->sap.read_int);
6776         t = &lun->stats.time[CTL_STATS_WRITE];
6777         scsi_u64to8b((uint64_t)t->sec * 1000 + t->frac / (UINT64_MAX / 1000),
6778             data->sap.write_int);
6779         scsi_u64to8b(0, data->sap.weighted_num);
6780         scsi_u64to8b(0, data->sap.weighted_int);
6781         scsi_ulto2b(SLP_IT, data->it.hdr.param_code);
6782         data->it.hdr.param_control = SLP_LBIN;
6783         data->it.hdr.param_len = sizeof(struct scsi_log_idle_time) -
6784             sizeof(struct scsi_log_param_header);
6785 #ifdef CTL_TIME_IO
6786         scsi_u64to8b(lun->idle_time / SBT_1MS, data->it.idle_int);
6787 #endif
6788         scsi_ulto2b(SLP_TI, data->ti.hdr.param_code);
6789         data->it.hdr.param_control = SLP_LBIN;
6790         data->ti.hdr.param_len = sizeof(struct scsi_log_time_interval) -
6791             sizeof(struct scsi_log_param_header);
6792         scsi_ulto4b(3, data->ti.exponent);
6793         scsi_ulto4b(1, data->ti.integer);
6794         return (0);
6795 }
6796
6797 int
6798 ctl_ie_log_sense_handler(struct ctl_scsiio *ctsio,
6799                                struct ctl_page_index *page_index,
6800                                int pc)
6801 {
6802         struct ctl_lun *lun = CTL_LUN(ctsio);
6803         struct scsi_log_informational_exceptions *data;
6804
6805         data = (struct scsi_log_informational_exceptions *)page_index->page_data;
6806
6807         scsi_ulto2b(SLP_IE_GEN, data->hdr.param_code);
6808         data->hdr.param_control = SLP_LBIN;
6809         data->hdr.param_len = sizeof(struct scsi_log_informational_exceptions) -
6810             sizeof(struct scsi_log_param_header);
6811         data->ie_asc = lun->ie_asc;
6812         data->ie_ascq = lun->ie_ascq;
6813         data->temperature = 0xff;
6814         return (0);
6815 }
6816
6817 int
6818 ctl_log_sense(struct ctl_scsiio *ctsio)
6819 {
6820         struct ctl_lun *lun = CTL_LUN(ctsio);
6821         int i, pc, page_code, subpage;
6822         int alloc_len, total_len;
6823         struct ctl_page_index *page_index;
6824         struct scsi_log_sense *cdb;
6825         struct scsi_log_header *header;
6826
6827         CTL_DEBUG_PRINT(("ctl_log_sense\n"));
6828
6829         cdb = (struct scsi_log_sense *)ctsio->cdb;
6830         pc = (cdb->page & SLS_PAGE_CTRL_MASK) >> 6;
6831         page_code = cdb->page & SLS_PAGE_CODE;
6832         subpage = cdb->subpage;
6833         alloc_len = scsi_2btoul(cdb->length);
6834
6835         page_index = NULL;
6836         for (i = 0; i < CTL_NUM_LOG_PAGES; i++) {
6837                 page_index = &lun->log_pages.index[i];
6838
6839                 /* Look for the right page code */
6840                 if ((page_index->page_code & SL_PAGE_CODE) != page_code)
6841                         continue;
6842
6843                 /* Look for the right subpage or the subpage wildcard*/
6844                 if (page_index->subpage != subpage)
6845                         continue;
6846
6847                 break;
6848         }
6849         if (i >= CTL_NUM_LOG_PAGES) {
6850                 ctl_set_invalid_field(ctsio,
6851                                       /*sks_valid*/ 1,
6852                                       /*command*/ 1,
6853                                       /*field*/ 2,
6854                                       /*bit_valid*/ 0,
6855                                       /*bit*/ 0);
6856                 ctl_done((union ctl_io *)ctsio);
6857                 return (CTL_RETVAL_COMPLETE);
6858         }
6859
6860         total_len = sizeof(struct scsi_log_header) + page_index->page_len;
6861
6862         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6863         ctsio->kern_sg_entries = 0;
6864         ctsio->kern_data_resid = 0;
6865         ctsio->kern_rel_offset = 0;
6866         if (total_len < alloc_len) {
6867                 ctsio->residual = alloc_len - total_len;
6868                 ctsio->kern_data_len = total_len;
6869                 ctsio->kern_total_len = total_len;
6870         } else {
6871                 ctsio->residual = 0;
6872                 ctsio->kern_data_len = alloc_len;
6873                 ctsio->kern_total_len = alloc_len;
6874         }
6875
6876         header = (struct scsi_log_header *)ctsio->kern_data_ptr;
6877         header->page = page_index->page_code;
6878         if (page_index->page_code == SLS_LOGICAL_BLOCK_PROVISIONING)
6879                 header->page |= SL_DS;
6880         if (page_index->subpage) {
6881                 header->page |= SL_SPF;
6882                 header->subpage = page_index->subpage;
6883         }
6884         scsi_ulto2b(page_index->page_len, header->datalen);
6885
6886         /*
6887          * Call the handler, if it exists, to update the
6888          * page to the latest values.
6889          */
6890         if (page_index->sense_handler != NULL)
6891                 page_index->sense_handler(ctsio, page_index, pc);
6892
6893         memcpy(header + 1, page_index->page_data, page_index->page_len);
6894
6895         ctl_set_success(ctsio);
6896         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6897         ctsio->be_move_done = ctl_config_move_done;
6898         ctl_datamove((union ctl_io *)ctsio);
6899         return (CTL_RETVAL_COMPLETE);
6900 }
6901
6902 int
6903 ctl_read_capacity(struct ctl_scsiio *ctsio)
6904 {
6905         struct ctl_lun *lun = CTL_LUN(ctsio);
6906         struct scsi_read_capacity *cdb;
6907         struct scsi_read_capacity_data *data;
6908         uint32_t lba;
6909
6910         CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
6911
6912         cdb = (struct scsi_read_capacity *)ctsio->cdb;
6913
6914         lba = scsi_4btoul(cdb->addr);
6915         if (((cdb->pmi & SRC_PMI) == 0)
6916          && (lba != 0)) {
6917                 ctl_set_invalid_field(/*ctsio*/ ctsio,
6918                                       /*sks_valid*/ 1,
6919                                       /*command*/ 1,
6920                                       /*field*/ 2,
6921                                       /*bit_valid*/ 0,
6922                                       /*bit*/ 0);
6923                 ctl_done((union ctl_io *)ctsio);
6924                 return (CTL_RETVAL_COMPLETE);
6925         }
6926
6927         ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
6928         data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
6929         ctsio->residual = 0;
6930         ctsio->kern_data_len = sizeof(*data);
6931         ctsio->kern_total_len = sizeof(*data);
6932         ctsio->kern_data_resid = 0;
6933         ctsio->kern_rel_offset = 0;
6934         ctsio->kern_sg_entries = 0;
6935
6936         /*
6937          * If the maximum LBA is greater than 0xfffffffe, the user must
6938          * issue a SERVICE ACTION IN (16) command, with the read capacity
6939          * serivce action set.
6940          */
6941         if (lun->be_lun->maxlba > 0xfffffffe)
6942                 scsi_ulto4b(0xffffffff, data->addr);
6943         else
6944                 scsi_ulto4b(lun->be_lun->maxlba, data->addr);
6945
6946         /*
6947          * XXX KDM this may not be 512 bytes...
6948          */
6949         scsi_ulto4b(lun->be_lun->blocksize, data->length);
6950
6951         ctl_set_success(ctsio);
6952         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6953         ctsio->be_move_done = ctl_config_move_done;
6954         ctl_datamove((union ctl_io *)ctsio);
6955         return (CTL_RETVAL_COMPLETE);
6956 }
6957
6958 int
6959 ctl_read_capacity_16(struct ctl_scsiio *ctsio)
6960 {
6961         struct ctl_lun *lun = CTL_LUN(ctsio);
6962         struct scsi_read_capacity_16 *cdb;
6963         struct scsi_read_capacity_data_long *data;
6964         uint64_t lba;
6965         uint32_t alloc_len;
6966
6967         CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
6968
6969         cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
6970
6971         alloc_len = scsi_4btoul(cdb->alloc_len);
6972         lba = scsi_8btou64(cdb->addr);
6973
6974         if ((cdb->reladr & SRC16_PMI)
6975          && (lba != 0)) {
6976                 ctl_set_invalid_field(/*ctsio*/ ctsio,
6977                                       /*sks_valid*/ 1,
6978                                       /*command*/ 1,
6979                                       /*field*/ 2,
6980                                       /*bit_valid*/ 0,
6981                                       /*bit*/ 0);
6982                 ctl_done((union ctl_io *)ctsio);
6983                 return (CTL_RETVAL_COMPLETE);
6984         }
6985
6986         ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
6987         data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
6988
6989         if (sizeof(*data) < alloc_len) {
6990                 ctsio->residual = alloc_len - sizeof(*data);
6991                 ctsio->kern_data_len = sizeof(*data);
6992                 ctsio->kern_total_len = sizeof(*data);
6993         } else {
6994                 ctsio->residual = 0;
6995                 ctsio->kern_data_len = alloc_len;
6996                 ctsio->kern_total_len = alloc_len;
6997         }
6998         ctsio->kern_data_resid = 0;
6999         ctsio->kern_rel_offset = 0;
7000         ctsio->kern_sg_entries = 0;
7001
7002         scsi_u64to8b(lun->be_lun->maxlba, data->addr);
7003         /* XXX KDM this may not be 512 bytes... */
7004         scsi_ulto4b(lun->be_lun->blocksize, data->length);
7005         data->prot_lbppbe = lun->be_lun->pblockexp & SRC16_LBPPBE;
7006         scsi_ulto2b(lun->be_lun->pblockoff & SRC16_LALBA_A, data->lalba_lbp);
7007         if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
7008                 data->lalba_lbp[0] |= SRC16_LBPME | SRC16_LBPRZ;
7009
7010         ctl_set_success(ctsio);
7011         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7012         ctsio->be_move_done = ctl_config_move_done;
7013         ctl_datamove((union ctl_io *)ctsio);
7014         return (CTL_RETVAL_COMPLETE);
7015 }
7016
7017 int
7018 ctl_get_lba_status(struct ctl_scsiio *ctsio)
7019 {
7020         struct ctl_lun *lun = CTL_LUN(ctsio);
7021         struct scsi_get_lba_status *cdb;
7022         struct scsi_get_lba_status_data *data;
7023         struct ctl_lba_len_flags *lbalen;
7024         uint64_t lba;
7025         uint32_t alloc_len, total_len;
7026         int retval;
7027
7028         CTL_DEBUG_PRINT(("ctl_get_lba_status\n"));
7029
7030         cdb = (struct scsi_get_lba_status *)ctsio->cdb;
7031         lba = scsi_8btou64(cdb->addr);
7032         alloc_len = scsi_4btoul(cdb->alloc_len);
7033
7034         if (lba > lun->be_lun->maxlba) {
7035                 ctl_set_lba_out_of_range(ctsio, lba);
7036                 ctl_done((union ctl_io *)ctsio);
7037                 return (CTL_RETVAL_COMPLETE);
7038         }
7039
7040         total_len = sizeof(*data) + sizeof(data->descr[0]);
7041         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7042         data = (struct scsi_get_lba_status_data *)ctsio->kern_data_ptr;
7043
7044         if (total_len < alloc_len) {
7045                 ctsio->residual = alloc_len - total_len;
7046                 ctsio->kern_data_len = total_len;
7047                 ctsio->kern_total_len = total_len;
7048         } else {
7049                 ctsio->residual = 0;
7050                 ctsio->kern_data_len = alloc_len;
7051                 ctsio->kern_total_len = alloc_len;
7052         }
7053         ctsio->kern_data_resid = 0;
7054         ctsio->kern_rel_offset = 0;
7055         ctsio->kern_sg_entries = 0;
7056
7057         /* Fill dummy data in case backend can't tell anything. */
7058         scsi_ulto4b(4 + sizeof(data->descr[0]), data->length);
7059         scsi_u64to8b(lba, data->descr[0].addr);
7060         scsi_ulto4b(MIN(UINT32_MAX, lun->be_lun->maxlba + 1 - lba),
7061             data->descr[0].length);
7062         data->descr[0].status = 0; /* Mapped or unknown. */
7063
7064         ctl_set_success(ctsio);
7065         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7066         ctsio->be_move_done = ctl_config_move_done;
7067
7068         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
7069         lbalen->lba = lba;
7070         lbalen->len = total_len;
7071         lbalen->flags = 0;
7072         retval = lun->backend->config_read((union ctl_io *)ctsio);
7073         return (CTL_RETVAL_COMPLETE);
7074 }
7075
7076 int
7077 ctl_read_defect(struct ctl_scsiio *ctsio)
7078 {
7079         struct scsi_read_defect_data_10 *ccb10;
7080         struct scsi_read_defect_data_12 *ccb12;
7081         struct scsi_read_defect_data_hdr_10 *data10;
7082         struct scsi_read_defect_data_hdr_12 *data12;
7083         uint32_t alloc_len, data_len;
7084         uint8_t format;
7085
7086         CTL_DEBUG_PRINT(("ctl_read_defect\n"));
7087
7088         if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
7089                 ccb10 = (struct scsi_read_defect_data_10 *)&ctsio->cdb;
7090                 format = ccb10->format;
7091                 alloc_len = scsi_2btoul(ccb10->alloc_length);
7092                 data_len = sizeof(*data10);
7093         } else {
7094                 ccb12 = (struct scsi_read_defect_data_12 *)&ctsio->cdb;
7095                 format = ccb12->format;
7096                 alloc_len = scsi_4btoul(ccb12->alloc_length);
7097                 data_len = sizeof(*data12);
7098         }
7099         if (alloc_len == 0) {
7100                 ctl_set_success(ctsio);
7101                 ctl_done((union ctl_io *)ctsio);
7102                 return (CTL_RETVAL_COMPLETE);
7103         }
7104
7105         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
7106         if (data_len < alloc_len) {
7107                 ctsio->residual = alloc_len - data_len;
7108                 ctsio->kern_data_len = data_len;
7109                 ctsio->kern_total_len = data_len;
7110         } else {
7111                 ctsio->residual = 0;
7112                 ctsio->kern_data_len = alloc_len;
7113                 ctsio->kern_total_len = alloc_len;
7114         }
7115         ctsio->kern_data_resid = 0;
7116         ctsio->kern_rel_offset = 0;
7117         ctsio->kern_sg_entries = 0;
7118
7119         if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
7120                 data10 = (struct scsi_read_defect_data_hdr_10 *)
7121                     ctsio->kern_data_ptr;
7122                 data10->format = format;
7123                 scsi_ulto2b(0, data10->length);
7124         } else {
7125                 data12 = (struct scsi_read_defect_data_hdr_12 *)
7126                     ctsio->kern_data_ptr;
7127                 data12->format = format;
7128                 scsi_ulto2b(0, data12->generation);
7129                 scsi_ulto4b(0, data12->length);
7130         }
7131
7132         ctl_set_success(ctsio);
7133         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7134         ctsio->be_move_done = ctl_config_move_done;
7135         ctl_datamove((union ctl_io *)ctsio);
7136         return (CTL_RETVAL_COMPLETE);
7137 }
7138
7139 int
7140 ctl_report_tagret_port_groups(struct ctl_scsiio *ctsio)
7141 {
7142         struct ctl_softc *softc = CTL_SOFTC(ctsio);
7143         struct ctl_lun *lun = CTL_LUN(ctsio);
7144         struct scsi_maintenance_in *cdb;
7145         int retval;
7146         int alloc_len, ext, total_len = 0, g, pc, pg, ts, os;
7147         int num_ha_groups, num_target_ports, shared_group;
7148         struct ctl_port *port;
7149         struct scsi_target_group_data *rtg_ptr;
7150         struct scsi_target_group_data_extended *rtg_ext_ptr;
7151         struct scsi_target_port_group_descriptor *tpg_desc;
7152
7153         CTL_DEBUG_PRINT(("ctl_report_tagret_port_groups\n"));
7154
7155         cdb = (struct scsi_maintenance_in *)ctsio->cdb;
7156         retval = CTL_RETVAL_COMPLETE;
7157
7158         switch (cdb->byte2 & STG_PDF_MASK) {
7159         case STG_PDF_LENGTH:
7160                 ext = 0;
7161                 break;
7162         case STG_PDF_EXTENDED:
7163                 ext = 1;
7164                 break;
7165         default:
7166                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7167                                       /*sks_valid*/ 1,
7168                                       /*command*/ 1,
7169                                       /*field*/ 2,
7170                                       /*bit_valid*/ 1,
7171                                       /*bit*/ 5);
7172                 ctl_done((union ctl_io *)ctsio);
7173                 return(retval);
7174         }
7175
7176         num_target_ports = 0;
7177         shared_group = (softc->is_single != 0);
7178         mtx_lock(&softc->ctl_lock);
7179         STAILQ_FOREACH(port, &softc->port_list, links) {
7180                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7181                         continue;
7182                 if (ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
7183                         continue;
7184                 num_target_ports++;
7185                 if (port->status & CTL_PORT_STATUS_HA_SHARED)
7186                         shared_group = 1;
7187         }
7188         mtx_unlock(&softc->ctl_lock);
7189         num_ha_groups = (softc->is_single) ? 0 : NUM_HA_SHELVES;
7190
7191         if (ext)
7192                 total_len = sizeof(struct scsi_target_group_data_extended);
7193         else
7194                 total_len = sizeof(struct scsi_target_group_data);
7195         total_len += sizeof(struct scsi_target_port_group_descriptor) *
7196                 (shared_group + num_ha_groups) +
7197             sizeof(struct scsi_target_port_descriptor) * num_target_ports;
7198
7199         alloc_len = scsi_4btoul(cdb->length);
7200
7201         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7202
7203         ctsio->kern_sg_entries = 0;
7204
7205         if (total_len < alloc_len) {
7206                 ctsio->residual = alloc_len - total_len;
7207                 ctsio->kern_data_len = total_len;
7208                 ctsio->kern_total_len = total_len;
7209         } else {
7210                 ctsio->residual = 0;
7211                 ctsio->kern_data_len = alloc_len;
7212                 ctsio->kern_total_len = alloc_len;
7213         }
7214         ctsio->kern_data_resid = 0;
7215         ctsio->kern_rel_offset = 0;
7216
7217         if (ext) {
7218                 rtg_ext_ptr = (struct scsi_target_group_data_extended *)
7219                     ctsio->kern_data_ptr;
7220                 scsi_ulto4b(total_len - 4, rtg_ext_ptr->length);
7221                 rtg_ext_ptr->format_type = 0x10;
7222                 rtg_ext_ptr->implicit_transition_time = 0;
7223                 tpg_desc = &rtg_ext_ptr->groups[0];
7224         } else {
7225                 rtg_ptr = (struct scsi_target_group_data *)
7226                     ctsio->kern_data_ptr;
7227                 scsi_ulto4b(total_len - 4, rtg_ptr->length);
7228                 tpg_desc = &rtg_ptr->groups[0];
7229         }
7230
7231         mtx_lock(&softc->ctl_lock);
7232         pg = softc->port_min / softc->port_cnt;
7233         if (lun->flags & (CTL_LUN_PRIMARY_SC | CTL_LUN_PEER_SC_PRIMARY)) {
7234                 /* Some shelf is known to be primary. */
7235                 if (softc->ha_link == CTL_HA_LINK_OFFLINE)
7236                         os = TPG_ASYMMETRIC_ACCESS_UNAVAILABLE;
7237                 else if (softc->ha_link == CTL_HA_LINK_UNKNOWN)
7238                         os = TPG_ASYMMETRIC_ACCESS_TRANSITIONING;
7239                 else if (softc->ha_mode == CTL_HA_MODE_ACT_STBY)
7240                         os = TPG_ASYMMETRIC_ACCESS_STANDBY;
7241                 else
7242                         os = TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7243                 if (lun->flags & CTL_LUN_PRIMARY_SC) {
7244                         ts = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7245                 } else {
7246                         ts = os;
7247                         os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7248                 }
7249         } else {
7250                 /* No known primary shelf. */
7251                 if (softc->ha_link == CTL_HA_LINK_OFFLINE) {
7252                         ts = TPG_ASYMMETRIC_ACCESS_UNAVAILABLE;
7253                         os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7254                 } else if (softc->ha_link == CTL_HA_LINK_UNKNOWN) {
7255                         ts = TPG_ASYMMETRIC_ACCESS_TRANSITIONING;
7256                         os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7257                 } else {
7258                         ts = os = TPG_ASYMMETRIC_ACCESS_TRANSITIONING;
7259                 }
7260         }
7261         if (shared_group) {
7262                 tpg_desc->pref_state = ts;
7263                 tpg_desc->support = TPG_AO_SUP | TPG_AN_SUP | TPG_S_SUP |
7264                     TPG_U_SUP | TPG_T_SUP;
7265                 scsi_ulto2b(1, tpg_desc->target_port_group);
7266                 tpg_desc->status = TPG_IMPLICIT;
7267                 pc = 0;
7268                 STAILQ_FOREACH(port, &softc->port_list, links) {
7269                         if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7270                                 continue;
7271                         if (!softc->is_single &&
7272                             (port->status & CTL_PORT_STATUS_HA_SHARED) == 0)
7273                                 continue;
7274                         if (ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
7275                                 continue;
7276                         scsi_ulto2b(port->targ_port, tpg_desc->descriptors[pc].
7277                             relative_target_port_identifier);
7278                         pc++;
7279                 }
7280                 tpg_desc->target_port_count = pc;
7281                 tpg_desc = (struct scsi_target_port_group_descriptor *)
7282                     &tpg_desc->descriptors[pc];
7283         }
7284         for (g = 0; g < num_ha_groups; g++) {
7285                 tpg_desc->pref_state = (g == pg) ? ts : os;
7286                 tpg_desc->support = TPG_AO_SUP | TPG_AN_SUP | TPG_S_SUP |
7287                     TPG_U_SUP | TPG_T_SUP;
7288                 scsi_ulto2b(2 + g, tpg_desc->target_port_group);
7289                 tpg_desc->status = TPG_IMPLICIT;
7290                 pc = 0;
7291                 STAILQ_FOREACH(port, &softc->port_list, links) {
7292                         if (port->targ_port < g * softc->port_cnt ||
7293                             port->targ_port >= (g + 1) * softc->port_cnt)
7294                                 continue;
7295                         if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7296                                 continue;
7297                         if (port->status & CTL_PORT_STATUS_HA_SHARED)
7298                                 continue;
7299                         if (ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
7300                                 continue;
7301                         scsi_ulto2b(port->targ_port, tpg_desc->descriptors[pc].
7302                             relative_target_port_identifier);
7303                         pc++;
7304                 }
7305                 tpg_desc->target_port_count = pc;
7306                 tpg_desc = (struct scsi_target_port_group_descriptor *)
7307                     &tpg_desc->descriptors[pc];
7308         }
7309         mtx_unlock(&softc->ctl_lock);
7310
7311         ctl_set_success(ctsio);
7312         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7313         ctsio->be_move_done = ctl_config_move_done;
7314         ctl_datamove((union ctl_io *)ctsio);
7315         return(retval);
7316 }
7317
7318 int
7319 ctl_report_supported_opcodes(struct ctl_scsiio *ctsio)
7320 {
7321         struct ctl_lun *lun = CTL_LUN(ctsio);
7322         struct scsi_report_supported_opcodes *cdb;
7323         const struct ctl_cmd_entry *entry, *sentry;
7324         struct scsi_report_supported_opcodes_all *all;
7325         struct scsi_report_supported_opcodes_descr *descr;
7326         struct scsi_report_supported_opcodes_one *one;
7327         int retval;
7328         int alloc_len, total_len;
7329         int opcode, service_action, i, j, num;
7330
7331         CTL_DEBUG_PRINT(("ctl_report_supported_opcodes\n"));
7332
7333         cdb = (struct scsi_report_supported_opcodes *)ctsio->cdb;
7334         retval = CTL_RETVAL_COMPLETE;
7335
7336         opcode = cdb->requested_opcode;
7337         service_action = scsi_2btoul(cdb->requested_service_action);
7338         switch (cdb->options & RSO_OPTIONS_MASK) {
7339         case RSO_OPTIONS_ALL:
7340                 num = 0;
7341                 for (i = 0; i < 256; i++) {
7342                         entry = &ctl_cmd_table[i];
7343                         if (entry->flags & CTL_CMD_FLAG_SA5) {
7344                                 for (j = 0; j < 32; j++) {
7345                                         sentry = &((const struct ctl_cmd_entry *)
7346                                             entry->execute)[j];
7347                                         if (ctl_cmd_applicable(
7348                                             lun->be_lun->lun_type, sentry))
7349                                                 num++;
7350                                 }
7351                         } else {
7352                                 if (ctl_cmd_applicable(lun->be_lun->lun_type,
7353                                     entry))
7354                                         num++;
7355                         }
7356                 }
7357                 total_len = sizeof(struct scsi_report_supported_opcodes_all) +
7358                     num * sizeof(struct scsi_report_supported_opcodes_descr);
7359                 break;
7360         case RSO_OPTIONS_OC:
7361                 if (ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) {
7362                         ctl_set_invalid_field(/*ctsio*/ ctsio,
7363                                               /*sks_valid*/ 1,
7364                                               /*command*/ 1,
7365                                               /*field*/ 2,
7366                                               /*bit_valid*/ 1,
7367                                               /*bit*/ 2);
7368                         ctl_done((union ctl_io *)ctsio);
7369                         return (CTL_RETVAL_COMPLETE);
7370                 }
7371                 total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7372                 break;
7373         case RSO_OPTIONS_OC_SA:
7374                 if ((ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) == 0 ||
7375                     service_action >= 32) {
7376                         ctl_set_invalid_field(/*ctsio*/ ctsio,
7377                                               /*sks_valid*/ 1,
7378                                               /*command*/ 1,
7379                                               /*field*/ 2,
7380                                               /*bit_valid*/ 1,
7381                                               /*bit*/ 2);
7382                         ctl_done((union ctl_io *)ctsio);
7383                         return (CTL_RETVAL_COMPLETE);
7384                 }
7385                 /* FALLTHROUGH */
7386         case RSO_OPTIONS_OC_ASA:
7387                 total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7388                 break;
7389         default:
7390                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7391                                       /*sks_valid*/ 1,
7392                                       /*command*/ 1,
7393                                       /*field*/ 2,
7394                                       /*bit_valid*/ 1,
7395                                       /*bit*/ 2);
7396                 ctl_done((union ctl_io *)ctsio);
7397                 return (CTL_RETVAL_COMPLETE);
7398         }
7399
7400         alloc_len = scsi_4btoul(cdb->length);
7401
7402         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7403
7404         ctsio->kern_sg_entries = 0;
7405
7406         if (total_len < alloc_len) {
7407                 ctsio->residual = alloc_len - total_len;
7408                 ctsio->kern_data_len = total_len;
7409                 ctsio->kern_total_len = total_len;
7410         } else {
7411                 ctsio->residual = 0;
7412                 ctsio->kern_data_len = alloc_len;
7413                 ctsio->kern_total_len = alloc_len;
7414         }
7415         ctsio->kern_data_resid = 0;
7416         ctsio->kern_rel_offset = 0;
7417
7418         switch (cdb->options & RSO_OPTIONS_MASK) {
7419         case RSO_OPTIONS_ALL:
7420                 all = (struct scsi_report_supported_opcodes_all *)
7421                     ctsio->kern_data_ptr;
7422                 num = 0;
7423                 for (i = 0; i < 256; i++) {
7424                         entry = &ctl_cmd_table[i];
7425                         if (entry->flags & CTL_CMD_FLAG_SA5) {
7426                                 for (j = 0; j < 32; j++) {
7427                                         sentry = &((const struct ctl_cmd_entry *)
7428                                             entry->execute)[j];
7429                                         if (!ctl_cmd_applicable(
7430                                             lun->be_lun->lun_type, sentry))
7431                                                 continue;
7432                                         descr = &all->descr[num++];
7433                                         descr->opcode = i;
7434                                         scsi_ulto2b(j, descr->service_action);
7435                                         descr->flags = RSO_SERVACTV;
7436                                         scsi_ulto2b(sentry->length,
7437                                             descr->cdb_length);
7438                                 }
7439                         } else {
7440                                 if (!ctl_cmd_applicable(lun->be_lun->lun_type,
7441                                     entry))
7442                                         continue;
7443                                 descr = &all->descr[num++];
7444                                 descr->opcode = i;
7445                                 scsi_ulto2b(0, descr->service_action);
7446                                 descr->flags = 0;
7447                                 scsi_ulto2b(entry->length, descr->cdb_length);
7448                         }
7449                 }
7450                 scsi_ulto4b(
7451                     num * sizeof(struct scsi_report_supported_opcodes_descr),
7452                     all->length);
7453                 break;
7454         case RSO_OPTIONS_OC:
7455                 one = (struct scsi_report_supported_opcodes_one *)
7456                     ctsio->kern_data_ptr;
7457                 entry = &ctl_cmd_table[opcode];
7458                 goto fill_one;
7459         case RSO_OPTIONS_OC_SA:
7460                 one = (struct scsi_report_supported_opcodes_one *)
7461                     ctsio->kern_data_ptr;
7462                 entry = &ctl_cmd_table[opcode];
7463                 entry = &((const struct ctl_cmd_entry *)
7464                     entry->execute)[service_action];
7465 fill_one:
7466                 if (ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
7467                         one->support = 3;
7468                         scsi_ulto2b(entry->length, one->cdb_length);
7469                         one->cdb_usage[0] = opcode;
7470                         memcpy(&one->cdb_usage[1], entry->usage,
7471                             entry->length - 1);
7472                 } else
7473                         one->support = 1;
7474                 break;
7475         case RSO_OPTIONS_OC_ASA:
7476                 one = (struct scsi_report_supported_opcodes_one *)
7477                     ctsio->kern_data_ptr;
7478                 entry = &ctl_cmd_table[opcode];
7479                 if (entry->flags & CTL_CMD_FLAG_SA5) {
7480                         entry = &((const struct ctl_cmd_entry *)
7481                             entry->execute)[service_action];
7482                 } else if (service_action != 0) {
7483                         one->support = 1;
7484                         break;
7485                 }
7486                 goto fill_one;
7487         }
7488
7489         ctl_set_success(ctsio);
7490         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7491         ctsio->be_move_done = ctl_config_move_done;
7492         ctl_datamove((union ctl_io *)ctsio);
7493         return(retval);
7494 }
7495
7496 int
7497 ctl_report_supported_tmf(struct ctl_scsiio *ctsio)
7498 {
7499         struct scsi_report_supported_tmf *cdb;
7500         struct scsi_report_supported_tmf_ext_data *data;
7501         int retval;
7502         int alloc_len, total_len;
7503
7504         CTL_DEBUG_PRINT(("ctl_report_supported_tmf\n"));
7505
7506         cdb = (struct scsi_report_supported_tmf *)ctsio->cdb;
7507
7508         retval = CTL_RETVAL_COMPLETE;
7509
7510         if (cdb->options & RST_REPD)
7511                 total_len = sizeof(struct scsi_report_supported_tmf_ext_data);
7512         else
7513                 total_len = sizeof(struct scsi_report_supported_tmf_data);
7514         alloc_len = scsi_4btoul(cdb->length);
7515
7516         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7517
7518         ctsio->kern_sg_entries = 0;
7519
7520         if (total_len < alloc_len) {
7521                 ctsio->residual = alloc_len - total_len;
7522                 ctsio->kern_data_len = total_len;
7523                 ctsio->kern_total_len = total_len;
7524         } else {
7525                 ctsio->residual = 0;
7526                 ctsio->kern_data_len = alloc_len;
7527                 ctsio->kern_total_len = alloc_len;
7528         }
7529         ctsio->kern_data_resid = 0;
7530         ctsio->kern_rel_offset = 0;
7531
7532         data = (struct scsi_report_supported_tmf_ext_data *)ctsio->kern_data_ptr;
7533         data->byte1 |= RST_ATS | RST_ATSS | RST_CTSS | RST_LURS | RST_QTS |
7534             RST_TRS;
7535         data->byte2 |= RST_QAES | RST_QTSS | RST_ITNRS;
7536         data->length = total_len - 4;
7537
7538         ctl_set_success(ctsio);
7539         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7540         ctsio->be_move_done = ctl_config_move_done;
7541         ctl_datamove((union ctl_io *)ctsio);
7542         return (retval);
7543 }
7544
7545 int
7546 ctl_report_timestamp(struct ctl_scsiio *ctsio)
7547 {
7548         struct scsi_report_timestamp *cdb;
7549         struct scsi_report_timestamp_data *data;
7550         struct timeval tv;
7551         int64_t timestamp;
7552         int retval;
7553         int alloc_len, total_len;
7554
7555         CTL_DEBUG_PRINT(("ctl_report_timestamp\n"));
7556
7557         cdb = (struct scsi_report_timestamp *)ctsio->cdb;
7558
7559         retval = CTL_RETVAL_COMPLETE;
7560
7561         total_len = sizeof(struct scsi_report_timestamp_data);
7562         alloc_len = scsi_4btoul(cdb->length);
7563
7564         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7565
7566         ctsio->kern_sg_entries = 0;
7567
7568         if (total_len < alloc_len) {
7569                 ctsio->residual = alloc_len - total_len;
7570                 ctsio->kern_data_len = total_len;
7571                 ctsio->kern_total_len = total_len;
7572         } else {
7573                 ctsio->residual = 0;
7574                 ctsio->kern_data_len = alloc_len;
7575                 ctsio->kern_total_len = alloc_len;
7576         }
7577         ctsio->kern_data_resid = 0;
7578         ctsio->kern_rel_offset = 0;
7579
7580         data = (struct scsi_report_timestamp_data *)ctsio->kern_data_ptr;
7581         scsi_ulto2b(sizeof(*data) - 2, data->length);
7582         data->origin = RTS_ORIG_OUTSIDE;
7583         getmicrotime(&tv);
7584         timestamp = (int64_t)tv.tv_sec * 1000 + tv.tv_usec / 1000;
7585         scsi_ulto4b(timestamp >> 16, data->timestamp);
7586         scsi_ulto2b(timestamp & 0xffff, &data->timestamp[4]);
7587
7588         ctl_set_success(ctsio);
7589         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7590         ctsio->be_move_done = ctl_config_move_done;
7591         ctl_datamove((union ctl_io *)ctsio);
7592         return (retval);
7593 }
7594
7595 int
7596 ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7597 {
7598         struct ctl_softc *softc = CTL_SOFTC(ctsio);
7599         struct ctl_lun *lun = CTL_LUN(ctsio);
7600         struct scsi_per_res_in *cdb;
7601         int alloc_len, total_len = 0;
7602         /* struct scsi_per_res_in_rsrv in_data; */
7603         uint64_t key;
7604
7605         CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7606
7607         cdb = (struct scsi_per_res_in *)ctsio->cdb;
7608
7609         alloc_len = scsi_2btoul(cdb->length);
7610
7611 retry:
7612         mtx_lock(&lun->lun_lock);
7613         switch (cdb->action) {
7614         case SPRI_RK: /* read keys */
7615                 total_len = sizeof(struct scsi_per_res_in_keys) +
7616                         lun->pr_key_count *
7617                         sizeof(struct scsi_per_res_key);
7618                 break;
7619         case SPRI_RR: /* read reservation */
7620                 if (lun->flags & CTL_LUN_PR_RESERVED)
7621                         total_len = sizeof(struct scsi_per_res_in_rsrv);
7622                 else
7623                         total_len = sizeof(struct scsi_per_res_in_header);
7624                 break;
7625         case SPRI_RC: /* report capabilities */
7626                 total_len = sizeof(struct scsi_per_res_cap);
7627                 break;
7628         case SPRI_RS: /* read full status */
7629                 total_len = sizeof(struct scsi_per_res_in_header) +
7630                     (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7631                     lun->pr_key_count;
7632                 break;
7633         default:
7634                 panic("%s: Invalid PR type %#x", __func__, cdb->action);
7635         }
7636         mtx_unlock(&lun->lun_lock);
7637
7638         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7639
7640         if (total_len < alloc_len) {
7641                 ctsio->residual = alloc_len - total_len;
7642                 ctsio->kern_data_len = total_len;
7643                 ctsio->kern_total_len = total_len;
7644         } else {
7645                 ctsio->residual = 0;
7646                 ctsio->kern_data_len = alloc_len;
7647                 ctsio->kern_total_len = alloc_len;
7648         }
7649
7650         ctsio->kern_data_resid = 0;
7651         ctsio->kern_rel_offset = 0;
7652         ctsio->kern_sg_entries = 0;
7653
7654         mtx_lock(&lun->lun_lock);
7655         switch (cdb->action) {
7656         case SPRI_RK: { // read keys
7657         struct scsi_per_res_in_keys *res_keys;
7658                 int i, key_count;
7659
7660                 res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7661
7662                 /*
7663                  * We had to drop the lock to allocate our buffer, which
7664                  * leaves time for someone to come in with another
7665                  * persistent reservation.  (That is unlikely, though,
7666                  * since this should be the only persistent reservation
7667                  * command active right now.)
7668                  */
7669                 if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7670                     (lun->pr_key_count *
7671                      sizeof(struct scsi_per_res_key)))){
7672                         mtx_unlock(&lun->lun_lock);
7673                         free(ctsio->kern_data_ptr, M_CTL);
7674                         printf("%s: reservation length changed, retrying\n",
7675                                __func__);
7676                         goto retry;
7677                 }
7678
7679                 scsi_ulto4b(lun->pr_generation, res_keys->header.generation);
7680
7681                 scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7682                              lun->pr_key_count, res_keys->header.length);
7683
7684                 for (i = 0, key_count = 0; i < CTL_MAX_INITIATORS; i++) {
7685                         if ((key = ctl_get_prkey(lun, i)) == 0)
7686                                 continue;
7687
7688                         /*
7689                          * We used lun->pr_key_count to calculate the
7690                          * size to allocate.  If it turns out the number of
7691                          * initiators with the registered flag set is
7692                          * larger than that (i.e. they haven't been kept in
7693                          * sync), we've got a problem.
7694                          */
7695                         if (key_count >= lun->pr_key_count) {
7696                                 key_count++;
7697                                 continue;
7698                         }
7699                         scsi_u64to8b(key, res_keys->keys[key_count].key);
7700                         key_count++;
7701                 }
7702                 break;
7703         }
7704         case SPRI_RR: { // read reservation
7705                 struct scsi_per_res_in_rsrv *res;
7706                 int tmp_len, header_only;
7707
7708                 res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7709
7710                 scsi_ulto4b(lun->pr_generation, res->header.generation);
7711
7712                 if (lun->flags & CTL_LUN_PR_RESERVED)
7713                 {
7714                         tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7715                         scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7716                                     res->header.length);
7717                         header_only = 0;
7718                 } else {
7719                         tmp_len = sizeof(struct scsi_per_res_in_header);
7720                         scsi_ulto4b(0, res->header.length);
7721                         header_only = 1;
7722                 }
7723
7724                 /*
7725                  * We had to drop the lock to allocate our buffer, which
7726                  * leaves time for someone to come in with another
7727                  * persistent reservation.  (That is unlikely, though,
7728                  * since this should be the only persistent reservation
7729                  * command active right now.)
7730                  */
7731                 if (tmp_len != total_len) {
7732                         mtx_unlock(&lun->lun_lock);
7733                         free(ctsio->kern_data_ptr, M_CTL);
7734                         printf("%s: reservation status changed, retrying\n",
7735                                __func__);
7736                         goto retry;
7737                 }
7738
7739                 /*
7740                  * No reservation held, so we're done.
7741                  */
7742                 if (header_only != 0)
7743                         break;
7744
7745                 /*
7746                  * If the registration is an All Registrants type, the key
7747                  * is 0, since it doesn't really matter.
7748                  */
7749                 if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7750                         scsi_u64to8b(ctl_get_prkey(lun, lun->pr_res_idx),
7751                             res->data.reservation);
7752                 }
7753                 res->data.scopetype = lun->pr_res_type;
7754                 break;
7755         }
7756         case SPRI_RC:     //report capabilities
7757         {
7758                 struct scsi_per_res_cap *res_cap;
7759                 uint16_t type_mask;
7760
7761                 res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7762                 scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7763                 res_cap->flags1 = SPRI_CRH;
7764                 res_cap->flags2 = SPRI_TMV | SPRI_ALLOW_5;
7765                 type_mask = SPRI_TM_WR_EX_AR |
7766                             SPRI_TM_EX_AC_RO |
7767                             SPRI_TM_WR_EX_RO |
7768                             SPRI_TM_EX_AC |
7769                             SPRI_TM_WR_EX |
7770                             SPRI_TM_EX_AC_AR;
7771                 scsi_ulto2b(type_mask, res_cap->type_mask);
7772                 break;
7773         }
7774         case SPRI_RS: { // read full status
7775                 struct scsi_per_res_in_full *res_status;
7776                 struct scsi_per_res_in_full_desc *res_desc;
7777                 struct ctl_port *port;
7778                 int i, len;
7779
7780                 res_status = (struct scsi_per_res_in_full*)ctsio->kern_data_ptr;
7781
7782                 /*
7783                  * We had to drop the lock to allocate our buffer, which
7784                  * leaves time for someone to come in with another
7785                  * persistent reservation.  (That is unlikely, though,
7786                  * since this should be the only persistent reservation
7787                  * command active right now.)
7788                  */
7789                 if (total_len < (sizeof(struct scsi_per_res_in_header) +
7790                     (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7791                      lun->pr_key_count)){
7792                         mtx_unlock(&lun->lun_lock);
7793                         free(ctsio->kern_data_ptr, M_CTL);
7794                         printf("%s: reservation length changed, retrying\n",
7795                                __func__);
7796                         goto retry;
7797                 }
7798
7799                 scsi_ulto4b(lun->pr_generation, res_status->header.generation);
7800
7801                 res_desc = &res_status->desc[0];
7802                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
7803                         if ((key = ctl_get_prkey(lun, i)) == 0)
7804                                 continue;
7805
7806                         scsi_u64to8b(key, res_desc->res_key.key);
7807                         if ((lun->flags & CTL_LUN_PR_RESERVED) &&
7808                             (lun->pr_res_idx == i ||
7809                              lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS)) {
7810                                 res_desc->flags = SPRI_FULL_R_HOLDER;
7811                                 res_desc->scopetype = lun->pr_res_type;
7812                         }
7813                         scsi_ulto2b(i / CTL_MAX_INIT_PER_PORT,
7814                             res_desc->rel_trgt_port_id);
7815                         len = 0;
7816                         port = softc->ctl_ports[i / CTL_MAX_INIT_PER_PORT];
7817                         if (port != NULL)
7818                                 len = ctl_create_iid(port,
7819                                     i % CTL_MAX_INIT_PER_PORT,
7820                                     res_desc->transport_id);
7821                         scsi_ulto4b(len, res_desc->additional_length);
7822                         res_desc = (struct scsi_per_res_in_full_desc *)
7823                             &res_desc->transport_id[len];
7824                 }
7825                 scsi_ulto4b((uint8_t *)res_desc - (uint8_t *)&res_status->desc[0],
7826                     res_status->header.length);
7827                 break;
7828         }
7829         default:
7830                 panic("%s: Invalid PR type %#x", __func__, cdb->action);
7831         }
7832         mtx_unlock(&lun->lun_lock);
7833
7834         ctl_set_success(ctsio);
7835         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7836         ctsio->be_move_done = ctl_config_move_done;
7837         ctl_datamove((union ctl_io *)ctsio);
7838         return (CTL_RETVAL_COMPLETE);
7839 }
7840
7841 /*
7842  * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
7843  * it should return.
7844  */
7845 static int
7846 ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
7847                 uint64_t sa_res_key, uint8_t type, uint32_t residx,
7848                 struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
7849                 struct scsi_per_res_out_parms* param)
7850 {
7851         union ctl_ha_msg persis_io;
7852         int i;
7853
7854         mtx_lock(&lun->lun_lock);
7855         if (sa_res_key == 0) {
7856                 if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
7857                         /* validate scope and type */
7858                         if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7859                              SPR_LU_SCOPE) {
7860                                 mtx_unlock(&lun->lun_lock);
7861                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7862                                                       /*sks_valid*/ 1,
7863                                                       /*command*/ 1,
7864                                                       /*field*/ 2,
7865                                                       /*bit_valid*/ 1,
7866                                                       /*bit*/ 4);
7867                                 ctl_done((union ctl_io *)ctsio);
7868                                 return (1);
7869                         }
7870
7871                         if (type>8 || type==2 || type==4 || type==0) {
7872                                 mtx_unlock(&lun->lun_lock);
7873                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7874                                                       /*sks_valid*/ 1,
7875                                                       /*command*/ 1,
7876                                                       /*field*/ 2,
7877                                                       /*bit_valid*/ 1,
7878                                                       /*bit*/ 0);
7879                                 ctl_done((union ctl_io *)ctsio);
7880                                 return (1);
7881                         }
7882
7883                         /*
7884                          * Unregister everybody else and build UA for
7885                          * them
7886                          */
7887                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
7888                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
7889                                         continue;
7890
7891                                 ctl_clr_prkey(lun, i);
7892                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7893                         }
7894                         lun->pr_key_count = 1;
7895                         lun->pr_res_type = type;
7896                         if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
7897                             lun->pr_res_type != SPR_TYPE_EX_AC_AR)
7898                                 lun->pr_res_idx = residx;
7899                         lun->pr_generation++;
7900                         mtx_unlock(&lun->lun_lock);
7901
7902                         /* send msg to other side */
7903                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7904                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7905                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7906                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
7907                         persis_io.pr.pr_info.res_type = type;
7908                         memcpy(persis_io.pr.pr_info.sa_res_key,
7909                                param->serv_act_res_key,
7910                                sizeof(param->serv_act_res_key));
7911                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
7912                             sizeof(persis_io.pr), M_WAITOK);
7913                 } else {
7914                         /* not all registrants */
7915                         mtx_unlock(&lun->lun_lock);
7916                         free(ctsio->kern_data_ptr, M_CTL);
7917                         ctl_set_invalid_field(ctsio,
7918                                               /*sks_valid*/ 1,
7919                                               /*command*/ 0,
7920                                               /*field*/ 8,
7921                                               /*bit_valid*/ 0,
7922                                               /*bit*/ 0);
7923                         ctl_done((union ctl_io *)ctsio);
7924                         return (1);
7925                 }
7926         } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
7927                 || !(lun->flags & CTL_LUN_PR_RESERVED)) {
7928                 int found = 0;
7929
7930                 if (res_key == sa_res_key) {
7931                         /* special case */
7932                         /*
7933                          * The spec implies this is not good but doesn't
7934                          * say what to do. There are two choices either
7935                          * generate a res conflict or check condition
7936                          * with illegal field in parameter data. Since
7937                          * that is what is done when the sa_res_key is
7938                          * zero I'll take that approach since this has
7939                          * to do with the sa_res_key.
7940                          */
7941                         mtx_unlock(&lun->lun_lock);
7942                         free(ctsio->kern_data_ptr, M_CTL);
7943                         ctl_set_invalid_field(ctsio,
7944                                               /*sks_valid*/ 1,
7945                                               /*command*/ 0,
7946                                               /*field*/ 8,
7947                                               /*bit_valid*/ 0,
7948                                               /*bit*/ 0);
7949                         ctl_done((union ctl_io *)ctsio);
7950                         return (1);
7951                 }
7952
7953                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
7954                         if (ctl_get_prkey(lun, i) != sa_res_key)
7955                                 continue;
7956
7957                         found = 1;
7958                         ctl_clr_prkey(lun, i);
7959                         lun->pr_key_count--;
7960                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7961                 }
7962                 if (!found) {
7963                         mtx_unlock(&lun->lun_lock);
7964                         free(ctsio->kern_data_ptr, M_CTL);
7965                         ctl_set_reservation_conflict(ctsio);
7966                         ctl_done((union ctl_io *)ctsio);
7967                         return (CTL_RETVAL_COMPLETE);
7968                 }
7969                 lun->pr_generation++;
7970                 mtx_unlock(&lun->lun_lock);
7971
7972                 /* send msg to other side */
7973                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7974                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7975                 persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7976                 persis_io.pr.pr_info.residx = lun->pr_res_idx;
7977                 persis_io.pr.pr_info.res_type = type;
7978                 memcpy(persis_io.pr.pr_info.sa_res_key,
7979                        param->serv_act_res_key,
7980                        sizeof(param->serv_act_res_key));
7981                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
7982                     sizeof(persis_io.pr), M_WAITOK);
7983         } else {
7984                 /* Reserved but not all registrants */
7985                 /* sa_res_key is res holder */
7986                 if (sa_res_key == ctl_get_prkey(lun, lun->pr_res_idx)) {
7987                         /* validate scope and type */
7988                         if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7989                              SPR_LU_SCOPE) {
7990                                 mtx_unlock(&lun->lun_lock);
7991                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7992                                                       /*sks_valid*/ 1,
7993                                                       /*command*/ 1,
7994                                                       /*field*/ 2,
7995                                                       /*bit_valid*/ 1,
7996                                                       /*bit*/ 4);
7997                                 ctl_done((union ctl_io *)ctsio);
7998                                 return (1);
7999                         }
8000
8001                         if (type>8 || type==2 || type==4 || type==0) {
8002                                 mtx_unlock(&lun->lun_lock);
8003                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
8004                                                       /*sks_valid*/ 1,
8005                                                       /*command*/ 1,
8006                                                       /*field*/ 2,
8007                                                       /*bit_valid*/ 1,
8008                                                       /*bit*/ 0);
8009                                 ctl_done((union ctl_io *)ctsio);
8010                                 return (1);
8011                         }
8012
8013                         /*
8014                          * Do the following:
8015                          * if sa_res_key != res_key remove all
8016                          * registrants w/sa_res_key and generate UA
8017                          * for these registrants(Registrations
8018                          * Preempted) if it wasn't an exclusive
8019                          * reservation generate UA(Reservations
8020                          * Preempted) for all other registered nexuses
8021                          * if the type has changed. Establish the new
8022                          * reservation and holder. If res_key and
8023                          * sa_res_key are the same do the above
8024                          * except don't unregister the res holder.
8025                          */
8026
8027                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
8028                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
8029                                         continue;
8030
8031                                 if (sa_res_key == ctl_get_prkey(lun, i)) {
8032                                         ctl_clr_prkey(lun, i);
8033                                         lun->pr_key_count--;
8034                                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8035                                 } else if (type != lun->pr_res_type &&
8036                                     (lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8037                                      lun->pr_res_type == SPR_TYPE_EX_AC_RO)) {
8038                                         ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8039                                 }
8040                         }
8041                         lun->pr_res_type = type;
8042                         if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
8043                             lun->pr_res_type != SPR_TYPE_EX_AC_AR)
8044                                 lun->pr_res_idx = residx;
8045                         else
8046                                 lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8047                         lun->pr_generation++;
8048                         mtx_unlock(&lun->lun_lock);
8049
8050                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8051                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8052                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8053                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8054                         persis_io.pr.pr_info.res_type = type;
8055                         memcpy(persis_io.pr.pr_info.sa_res_key,
8056                                param->serv_act_res_key,
8057                                sizeof(param->serv_act_res_key));
8058                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8059                             sizeof(persis_io.pr), M_WAITOK);
8060                 } else {
8061                         /*
8062                          * sa_res_key is not the res holder just
8063                          * remove registrants
8064                          */
8065                         int found=0;
8066
8067                         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8068                                 if (sa_res_key != ctl_get_prkey(lun, i))
8069                                         continue;
8070
8071                                 found = 1;
8072                                 ctl_clr_prkey(lun, i);
8073                                 lun->pr_key_count--;
8074                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8075                         }
8076
8077                         if (!found) {
8078                                 mtx_unlock(&lun->lun_lock);
8079                                 free(ctsio->kern_data_ptr, M_CTL);
8080                                 ctl_set_reservation_conflict(ctsio);
8081                                 ctl_done((union ctl_io *)ctsio);
8082                                 return (1);
8083                         }
8084                         lun->pr_generation++;
8085                         mtx_unlock(&lun->lun_lock);
8086
8087                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8088                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8089                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8090                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8091                         persis_io.pr.pr_info.res_type = type;
8092                         memcpy(persis_io.pr.pr_info.sa_res_key,
8093                                param->serv_act_res_key,
8094                                sizeof(param->serv_act_res_key));
8095                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8096                             sizeof(persis_io.pr), M_WAITOK);
8097                 }
8098         }
8099         return (0);
8100 }
8101
8102 static void
8103 ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
8104 {
8105         uint64_t sa_res_key;
8106         int i;
8107
8108         sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
8109
8110         if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8111          || lun->pr_res_idx == CTL_PR_NO_RESERVATION
8112          || sa_res_key != ctl_get_prkey(lun, lun->pr_res_idx)) {
8113                 if (sa_res_key == 0) {
8114                         /*
8115                          * Unregister everybody else and build UA for
8116                          * them
8117                          */
8118                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
8119                                 if (i == msg->pr.pr_info.residx ||
8120                                     ctl_get_prkey(lun, i) == 0)
8121                                         continue;
8122
8123                                 ctl_clr_prkey(lun, i);
8124                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8125                         }
8126
8127                         lun->pr_key_count = 1;
8128                         lun->pr_res_type = msg->pr.pr_info.res_type;
8129                         if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
8130                             lun->pr_res_type != SPR_TYPE_EX_AC_AR)
8131                                 lun->pr_res_idx = msg->pr.pr_info.residx;
8132                 } else {
8133                         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8134                                 if (sa_res_key == ctl_get_prkey(lun, i))
8135                                         continue;
8136
8137                                 ctl_clr_prkey(lun, i);
8138                                 lun->pr_key_count--;
8139                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8140                         }
8141                 }
8142         } else {
8143                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8144                         if (i == msg->pr.pr_info.residx ||
8145                             ctl_get_prkey(lun, i) == 0)
8146                                 continue;
8147
8148                         if (sa_res_key == ctl_get_prkey(lun, i)) {
8149                                 ctl_clr_prkey(lun, i);
8150                                 lun->pr_key_count--;
8151                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8152                         } else if (msg->pr.pr_info.res_type != lun->pr_res_type
8153                             && (lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8154                              lun->pr_res_type == SPR_TYPE_EX_AC_RO)) {
8155                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8156                         }
8157                 }
8158                 lun->pr_res_type = msg->pr.pr_info.res_type;
8159                 if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
8160                     lun->pr_res_type != SPR_TYPE_EX_AC_AR)
8161                         lun->pr_res_idx = msg->pr.pr_info.residx;
8162                 else
8163                         lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8164         }
8165         lun->pr_generation++;
8166
8167 }
8168
8169
8170 int
8171 ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
8172 {
8173         struct ctl_softc *softc = CTL_SOFTC(ctsio);
8174         struct ctl_lun *lun = CTL_LUN(ctsio);
8175         int retval;
8176         u_int32_t param_len;
8177         struct scsi_per_res_out *cdb;
8178         struct scsi_per_res_out_parms* param;
8179         uint32_t residx;
8180         uint64_t res_key, sa_res_key, key;
8181         uint8_t type;
8182         union ctl_ha_msg persis_io;
8183         int    i;
8184
8185         CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n"));
8186
8187         cdb = (struct scsi_per_res_out *)ctsio->cdb;
8188         retval = CTL_RETVAL_COMPLETE;
8189
8190         /*
8191          * We only support whole-LUN scope.  The scope & type are ignored for
8192          * register, register and ignore existing key and clear.
8193          * We sometimes ignore scope and type on preempts too!!
8194          * Verify reservation type here as well.
8195          */
8196         type = cdb->scope_type & SPR_TYPE_MASK;
8197         if ((cdb->action == SPRO_RESERVE)
8198          || (cdb->action == SPRO_RELEASE)) {
8199                 if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
8200                         ctl_set_invalid_field(/*ctsio*/ ctsio,
8201                                               /*sks_valid*/ 1,
8202                                               /*command*/ 1,
8203                                               /*field*/ 2,
8204                                               /*bit_valid*/ 1,
8205                                               /*bit*/ 4);
8206                         ctl_done((union ctl_io *)ctsio);
8207                         return (CTL_RETVAL_COMPLETE);
8208                 }
8209
8210                 if (type>8 || type==2 || type==4 || type==0) {
8211                         ctl_set_invalid_field(/*ctsio*/ ctsio,
8212                                               /*sks_valid*/ 1,
8213                                               /*command*/ 1,
8214                                               /*field*/ 2,
8215                                               /*bit_valid*/ 1,
8216                                               /*bit*/ 0);
8217                         ctl_done((union ctl_io *)ctsio);
8218                         return (CTL_RETVAL_COMPLETE);
8219                 }
8220         }
8221
8222         param_len = scsi_4btoul(cdb->length);
8223
8224         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
8225                 ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
8226                 ctsio->kern_data_len = param_len;
8227                 ctsio->kern_total_len = param_len;
8228                 ctsio->kern_data_resid = 0;
8229                 ctsio->kern_rel_offset = 0;
8230                 ctsio->kern_sg_entries = 0;
8231                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
8232                 ctsio->be_move_done = ctl_config_move_done;
8233                 ctl_datamove((union ctl_io *)ctsio);
8234
8235                 return (CTL_RETVAL_COMPLETE);
8236         }
8237
8238         param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
8239
8240         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
8241         res_key = scsi_8btou64(param->res_key.key);
8242         sa_res_key = scsi_8btou64(param->serv_act_res_key);
8243
8244         /*
8245          * Validate the reservation key here except for SPRO_REG_IGNO
8246          * This must be done for all other service actions
8247          */
8248         if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
8249                 mtx_lock(&lun->lun_lock);
8250                 if ((key = ctl_get_prkey(lun, residx)) != 0) {
8251                         if (res_key != key) {
8252                                 /*
8253                                  * The current key passed in doesn't match
8254                                  * the one the initiator previously
8255                                  * registered.
8256                                  */
8257                                 mtx_unlock(&lun->lun_lock);
8258                                 free(ctsio->kern_data_ptr, M_CTL);
8259                                 ctl_set_reservation_conflict(ctsio);
8260                                 ctl_done((union ctl_io *)ctsio);
8261                                 return (CTL_RETVAL_COMPLETE);
8262                         }
8263                 } else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
8264                         /*
8265                          * We are not registered
8266                          */
8267                         mtx_unlock(&lun->lun_lock);
8268                         free(ctsio->kern_data_ptr, M_CTL);
8269                         ctl_set_reservation_conflict(ctsio);
8270                         ctl_done((union ctl_io *)ctsio);
8271                         return (CTL_RETVAL_COMPLETE);
8272                 } else if (res_key != 0) {
8273                         /*
8274                          * We are not registered and trying to register but
8275                          * the register key isn't zero.
8276                          */
8277                         mtx_unlock(&lun->lun_lock);
8278                         free(ctsio->kern_data_ptr, M_CTL);
8279                         ctl_set_reservation_conflict(ctsio);
8280                         ctl_done((union ctl_io *)ctsio);
8281                         return (CTL_RETVAL_COMPLETE);
8282                 }
8283                 mtx_unlock(&lun->lun_lock);
8284         }
8285
8286         switch (cdb->action & SPRO_ACTION_MASK) {
8287         case SPRO_REGISTER:
8288         case SPRO_REG_IGNO: {
8289
8290 #if 0
8291                 printf("Registration received\n");
8292 #endif
8293
8294                 /*
8295                  * We don't support any of these options, as we report in
8296                  * the read capabilities request (see
8297                  * ctl_persistent_reserve_in(), above).
8298                  */
8299                 if ((param->flags & SPR_SPEC_I_PT)
8300                  || (param->flags & SPR_ALL_TG_PT)
8301                  || (param->flags & SPR_APTPL)) {
8302                         int bit_ptr;
8303
8304                         if (param->flags & SPR_APTPL)
8305                                 bit_ptr = 0;
8306                         else if (param->flags & SPR_ALL_TG_PT)
8307                                 bit_ptr = 2;
8308                         else /* SPR_SPEC_I_PT */
8309                                 bit_ptr = 3;
8310
8311                         free(ctsio->kern_data_ptr, M_CTL);
8312                         ctl_set_invalid_field(ctsio,
8313                                               /*sks_valid*/ 1,
8314                                               /*command*/ 0,
8315                                               /*field*/ 20,
8316                                               /*bit_valid*/ 1,
8317                                               /*bit*/ bit_ptr);
8318                         ctl_done((union ctl_io *)ctsio);
8319                         return (CTL_RETVAL_COMPLETE);
8320                 }
8321
8322                 mtx_lock(&lun->lun_lock);
8323
8324                 /*
8325                  * The initiator wants to clear the
8326                  * key/unregister.
8327                  */
8328                 if (sa_res_key == 0) {
8329                         if ((res_key == 0
8330                           && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
8331                          || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
8332                           && ctl_get_prkey(lun, residx) == 0)) {
8333                                 mtx_unlock(&lun->lun_lock);
8334                                 goto done;
8335                         }
8336
8337                         ctl_clr_prkey(lun, residx);
8338                         lun->pr_key_count--;
8339
8340                         if (residx == lun->pr_res_idx) {
8341                                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8342                                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8343
8344                                 if ((lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8345                                      lun->pr_res_type == SPR_TYPE_EX_AC_RO) &&
8346                                     lun->pr_key_count) {
8347                                         /*
8348                                          * If the reservation is a registrants
8349                                          * only type we need to generate a UA
8350                                          * for other registered inits.  The
8351                                          * sense code should be RESERVATIONS
8352                                          * RELEASED
8353                                          */
8354
8355                                         for (i = softc->init_min; i < softc->init_max; i++){
8356                                                 if (ctl_get_prkey(lun, i) == 0)
8357                                                         continue;
8358                                                 ctl_est_ua(lun, i,
8359                                                     CTL_UA_RES_RELEASE);
8360                                         }
8361                                 }
8362                                 lun->pr_res_type = 0;
8363                         } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8364                                 if (lun->pr_key_count==0) {
8365                                         lun->flags &= ~CTL_LUN_PR_RESERVED;
8366                                         lun->pr_res_type = 0;
8367                                         lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8368                                 }
8369                         }
8370                         lun->pr_generation++;
8371                         mtx_unlock(&lun->lun_lock);
8372
8373                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8374                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8375                         persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
8376                         persis_io.pr.pr_info.residx = residx;
8377                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8378                             sizeof(persis_io.pr), M_WAITOK);
8379                 } else /* sa_res_key != 0 */ {
8380
8381                         /*
8382                          * If we aren't registered currently then increment
8383                          * the key count and set the registered flag.
8384                          */
8385                         ctl_alloc_prkey(lun, residx);
8386                         if (ctl_get_prkey(lun, residx) == 0)
8387                                 lun->pr_key_count++;
8388                         ctl_set_prkey(lun, residx, sa_res_key);
8389                         lun->pr_generation++;
8390                         mtx_unlock(&lun->lun_lock);
8391
8392                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8393                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8394                         persis_io.pr.pr_info.action = CTL_PR_REG_KEY;
8395                         persis_io.pr.pr_info.residx = residx;
8396                         memcpy(persis_io.pr.pr_info.sa_res_key,
8397                                param->serv_act_res_key,
8398                                sizeof(param->serv_act_res_key));
8399                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8400                             sizeof(persis_io.pr), M_WAITOK);
8401                 }
8402
8403                 break;
8404         }
8405         case SPRO_RESERVE:
8406 #if 0
8407                 printf("Reserve executed type %d\n", type);
8408 #endif
8409                 mtx_lock(&lun->lun_lock);
8410                 if (lun->flags & CTL_LUN_PR_RESERVED) {
8411                         /*
8412                          * if this isn't the reservation holder and it's
8413                          * not a "all registrants" type or if the type is
8414                          * different then we have a conflict
8415                          */
8416                         if ((lun->pr_res_idx != residx
8417                           && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8418                          || lun->pr_res_type != type) {
8419                                 mtx_unlock(&lun->lun_lock);
8420                                 free(ctsio->kern_data_ptr, M_CTL);
8421                                 ctl_set_reservation_conflict(ctsio);
8422                                 ctl_done((union ctl_io *)ctsio);
8423                                 return (CTL_RETVAL_COMPLETE);
8424                         }
8425                         mtx_unlock(&lun->lun_lock);
8426                 } else /* create a reservation */ {
8427                         /*
8428                          * If it's not an "all registrants" type record
8429                          * reservation holder
8430                          */
8431                         if (type != SPR_TYPE_WR_EX_AR
8432                          && type != SPR_TYPE_EX_AC_AR)
8433                                 lun->pr_res_idx = residx; /* Res holder */
8434                         else
8435                                 lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8436
8437                         lun->flags |= CTL_LUN_PR_RESERVED;
8438                         lun->pr_res_type = type;
8439
8440                         mtx_unlock(&lun->lun_lock);
8441
8442                         /* send msg to other side */
8443                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8444                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8445                         persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8446                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8447                         persis_io.pr.pr_info.res_type = type;
8448                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8449                             sizeof(persis_io.pr), M_WAITOK);
8450                 }
8451                 break;
8452
8453         case SPRO_RELEASE:
8454                 mtx_lock(&lun->lun_lock);
8455                 if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8456                         /* No reservation exists return good status */
8457                         mtx_unlock(&lun->lun_lock);
8458                         goto done;
8459                 }
8460                 /*
8461                  * Is this nexus a reservation holder?
8462                  */
8463                 if (lun->pr_res_idx != residx
8464                  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8465                         /*
8466                          * not a res holder return good status but
8467                          * do nothing
8468                          */
8469                         mtx_unlock(&lun->lun_lock);
8470                         goto done;
8471                 }
8472
8473                 if (lun->pr_res_type != type) {
8474                         mtx_unlock(&lun->lun_lock);
8475                         free(ctsio->kern_data_ptr, M_CTL);
8476                         ctl_set_illegal_pr_release(ctsio);
8477                         ctl_done((union ctl_io *)ctsio);
8478                         return (CTL_RETVAL_COMPLETE);
8479                 }
8480
8481                 /* okay to release */
8482                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8483                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8484                 lun->pr_res_type = 0;
8485
8486                 /*
8487                  * If this isn't an exclusive access reservation and NUAR
8488                  * is not set, generate UA for all other registrants.
8489                  */
8490                 if (type != SPR_TYPE_EX_AC && type != SPR_TYPE_WR_EX &&
8491                     (lun->MODE_CTRL.queue_flags & SCP_NUAR) == 0) {
8492                         for (i = softc->init_min; i < softc->init_max; i++) {
8493                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
8494                                         continue;
8495                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8496                         }
8497                 }
8498                 mtx_unlock(&lun->lun_lock);
8499
8500                 /* Send msg to other side */
8501                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8502                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8503                 persis_io.pr.pr_info.action = CTL_PR_RELEASE;
8504                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8505                      sizeof(persis_io.pr), M_WAITOK);
8506                 break;
8507
8508         case SPRO_CLEAR:
8509                 /* send msg to other side */
8510
8511                 mtx_lock(&lun->lun_lock);
8512                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8513                 lun->pr_res_type = 0;
8514                 lun->pr_key_count = 0;
8515                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8516
8517                 ctl_clr_prkey(lun, residx);
8518                 for (i = 0; i < CTL_MAX_INITIATORS; i++)
8519                         if (ctl_get_prkey(lun, i) != 0) {
8520                                 ctl_clr_prkey(lun, i);
8521                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8522                         }
8523                 lun->pr_generation++;
8524                 mtx_unlock(&lun->lun_lock);
8525
8526                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8527                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8528                 persis_io.pr.pr_info.action = CTL_PR_CLEAR;
8529                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8530                      sizeof(persis_io.pr), M_WAITOK);
8531                 break;
8532
8533         case SPRO_PREEMPT:
8534         case SPRO_PRE_ABO: {
8535                 int nretval;
8536
8537                 nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8538                                           residx, ctsio, cdb, param);
8539                 if (nretval != 0)
8540                         return (CTL_RETVAL_COMPLETE);
8541                 break;
8542         }
8543         default:
8544                 panic("%s: Invalid PR type %#x", __func__, cdb->action);
8545         }
8546
8547 done:
8548         free(ctsio->kern_data_ptr, M_CTL);
8549         ctl_set_success(ctsio);
8550         ctl_done((union ctl_io *)ctsio);
8551
8552         return (retval);
8553 }
8554
8555 /*
8556  * This routine is for handling a message from the other SC pertaining to
8557  * persistent reserve out. All the error checking will have been done
8558  * so only perorming the action need be done here to keep the two
8559  * in sync.
8560  */
8561 static void
8562 ctl_hndl_per_res_out_on_other_sc(union ctl_io *io)
8563 {
8564         struct ctl_softc *softc = CTL_SOFTC(io);
8565         union ctl_ha_msg *msg = (union ctl_ha_msg *)&io->presio.pr_msg;
8566         struct ctl_lun *lun;
8567         int i;
8568         uint32_t residx, targ_lun;
8569
8570         targ_lun = msg->hdr.nexus.targ_mapped_lun;
8571         mtx_lock(&softc->ctl_lock);
8572         if (targ_lun >= CTL_MAX_LUNS ||
8573             (lun = softc->ctl_luns[targ_lun]) == NULL) {
8574                 mtx_unlock(&softc->ctl_lock);
8575                 return;
8576         }
8577         mtx_lock(&lun->lun_lock);
8578         mtx_unlock(&softc->ctl_lock);
8579         if (lun->flags & CTL_LUN_DISABLED) {
8580                 mtx_unlock(&lun->lun_lock);
8581                 return;
8582         }
8583         residx = ctl_get_initindex(&msg->hdr.nexus);
8584         switch(msg->pr.pr_info.action) {
8585         case CTL_PR_REG_KEY:
8586                 ctl_alloc_prkey(lun, msg->pr.pr_info.residx);
8587                 if (ctl_get_prkey(lun, msg->pr.pr_info.residx) == 0)
8588                         lun->pr_key_count++;
8589                 ctl_set_prkey(lun, msg->pr.pr_info.residx,
8590                     scsi_8btou64(msg->pr.pr_info.sa_res_key));
8591                 lun->pr_generation++;
8592                 break;
8593
8594         case CTL_PR_UNREG_KEY:
8595                 ctl_clr_prkey(lun, msg->pr.pr_info.residx);
8596                 lun->pr_key_count--;
8597
8598                 /* XXX Need to see if the reservation has been released */
8599                 /* if so do we need to generate UA? */
8600                 if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8601                         lun->flags &= ~CTL_LUN_PR_RESERVED;
8602                         lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8603
8604                         if ((lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8605                              lun->pr_res_type == SPR_TYPE_EX_AC_RO) &&
8606                             lun->pr_key_count) {
8607                                 /*
8608                                  * If the reservation is a registrants
8609                                  * only type we need to generate a UA
8610                                  * for other registered inits.  The
8611                                  * sense code should be RESERVATIONS
8612                                  * RELEASED
8613                                  */
8614
8615                                 for (i = softc->init_min; i < softc->init_max; i++) {
8616                                         if (ctl_get_prkey(lun, i) == 0)
8617                                                 continue;
8618
8619                                         ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8620                                 }
8621                         }
8622                         lun->pr_res_type = 0;
8623                 } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8624                         if (lun->pr_key_count==0) {
8625                                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8626                                 lun->pr_res_type = 0;
8627                                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8628                         }
8629                 }
8630                 lun->pr_generation++;
8631                 break;
8632
8633         case CTL_PR_RESERVE:
8634                 lun->flags |= CTL_LUN_PR_RESERVED;
8635                 lun->pr_res_type = msg->pr.pr_info.res_type;
8636                 lun->pr_res_idx = msg->pr.pr_info.residx;
8637
8638                 break;
8639
8640         case CTL_PR_RELEASE:
8641                 /*
8642                  * If this isn't an exclusive access reservation and NUAR
8643                  * is not set, generate UA for all other registrants.
8644                  */
8645                 if (lun->pr_res_type != SPR_TYPE_EX_AC &&
8646                     lun->pr_res_type != SPR_TYPE_WR_EX &&
8647                     (lun->MODE_CTRL.queue_flags & SCP_NUAR) == 0) {
8648                         for (i = softc->init_min; i < softc->init_max; i++)
8649                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
8650                                         continue;
8651                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8652                 }
8653
8654                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8655                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8656                 lun->pr_res_type = 0;
8657                 break;
8658
8659         case CTL_PR_PREEMPT:
8660                 ctl_pro_preempt_other(lun, msg);
8661                 break;
8662         case CTL_PR_CLEAR:
8663                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8664                 lun->pr_res_type = 0;
8665                 lun->pr_key_count = 0;
8666                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8667
8668                 for (i=0; i < CTL_MAX_INITIATORS; i++) {
8669                         if (ctl_get_prkey(lun, i) == 0)
8670                                 continue;
8671                         ctl_clr_prkey(lun, i);
8672                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8673                 }
8674                 lun->pr_generation++;
8675                 break;
8676         }
8677
8678         mtx_unlock(&lun->lun_lock);
8679 }
8680
8681 int
8682 ctl_read_write(struct ctl_scsiio *ctsio)
8683 {
8684         struct ctl_lun *lun = CTL_LUN(ctsio);
8685         struct ctl_lba_len_flags *lbalen;
8686         uint64_t lba;
8687         uint32_t num_blocks;
8688         int flags, retval;
8689         int isread;
8690
8691         CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
8692
8693         flags = 0;
8694         isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
8695               || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
8696         switch (ctsio->cdb[0]) {
8697         case READ_6:
8698         case WRITE_6: {
8699                 struct scsi_rw_6 *cdb;
8700
8701                 cdb = (struct scsi_rw_6 *)ctsio->cdb;
8702
8703                 lba = scsi_3btoul(cdb->addr);
8704                 /* only 5 bits are valid in the most significant address byte */
8705                 lba &= 0x1fffff;
8706                 num_blocks = cdb->length;
8707                 /*
8708                  * This is correct according to SBC-2.
8709                  */
8710                 if (num_blocks == 0)
8711                         num_blocks = 256;
8712                 break;
8713         }
8714         case READ_10:
8715         case WRITE_10: {
8716                 struct scsi_rw_10 *cdb;
8717
8718                 cdb = (struct scsi_rw_10 *)ctsio->cdb;
8719                 if (cdb->byte2 & SRW10_FUA)
8720                         flags |= CTL_LLF_FUA;
8721                 if (cdb->byte2 & SRW10_DPO)
8722                         flags |= CTL_LLF_DPO;
8723                 lba = scsi_4btoul(cdb->addr);
8724                 num_blocks = scsi_2btoul(cdb->length);
8725                 break;
8726         }
8727         case WRITE_VERIFY_10: {
8728                 struct scsi_write_verify_10 *cdb;
8729
8730                 cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
8731                 flags |= CTL_LLF_FUA;
8732                 if (cdb->byte2 & SWV_DPO)
8733                         flags |= CTL_LLF_DPO;
8734                 lba = scsi_4btoul(cdb->addr);
8735                 num_blocks = scsi_2btoul(cdb->length);
8736                 break;
8737         }
8738         case READ_12:
8739         case WRITE_12: {
8740                 struct scsi_rw_12 *cdb;
8741
8742                 cdb = (struct scsi_rw_12 *)ctsio->cdb;
8743                 if (cdb->byte2 & SRW12_FUA)
8744                         flags |= CTL_LLF_FUA;
8745                 if (cdb->byte2 & SRW12_DPO)
8746                         flags |= CTL_LLF_DPO;
8747                 lba = scsi_4btoul(cdb->addr);
8748                 num_blocks = scsi_4btoul(cdb->length);
8749                 break;
8750         }
8751         case WRITE_VERIFY_12: {
8752                 struct scsi_write_verify_12 *cdb;
8753
8754                 cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
8755                 flags |= CTL_LLF_FUA;
8756                 if (cdb->byte2 & SWV_DPO)
8757                         flags |= CTL_LLF_DPO;
8758                 lba = scsi_4btoul(cdb->addr);
8759                 num_blocks = scsi_4btoul(cdb->length);
8760                 break;
8761         }
8762         case READ_16:
8763         case WRITE_16: {
8764                 struct scsi_rw_16 *cdb;
8765
8766                 cdb = (struct scsi_rw_16 *)ctsio->cdb;
8767                 if (cdb->byte2 & SRW12_FUA)
8768                         flags |= CTL_LLF_FUA;
8769                 if (cdb->byte2 & SRW12_DPO)
8770                         flags |= CTL_LLF_DPO;
8771                 lba = scsi_8btou64(cdb->addr);
8772                 num_blocks = scsi_4btoul(cdb->length);
8773                 break;
8774         }
8775         case WRITE_ATOMIC_16: {
8776                 struct scsi_write_atomic_16 *cdb;
8777
8778                 if (lun->be_lun->atomicblock == 0) {
8779                         ctl_set_invalid_opcode(ctsio);
8780                         ctl_done((union ctl_io *)ctsio);
8781                         return (CTL_RETVAL_COMPLETE);
8782                 }
8783
8784                 cdb = (struct scsi_write_atomic_16 *)ctsio->cdb;
8785                 if (cdb->byte2 & SRW12_FUA)
8786                         flags |= CTL_LLF_FUA;
8787                 if (cdb->byte2 & SRW12_DPO)
8788                         flags |= CTL_LLF_DPO;
8789                 lba = scsi_8btou64(cdb->addr);
8790                 num_blocks = scsi_2btoul(cdb->length);
8791                 if (num_blocks > lun->be_lun->atomicblock) {
8792                         ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
8793                             /*command*/ 1, /*field*/ 12, /*bit_valid*/ 0,
8794                             /*bit*/ 0);
8795                         ctl_done((union ctl_io *)ctsio);
8796                         return (CTL_RETVAL_COMPLETE);
8797                 }
8798                 break;
8799         }
8800         case WRITE_VERIFY_16: {
8801                 struct scsi_write_verify_16 *cdb;
8802
8803                 cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
8804                 flags |= CTL_LLF_FUA;
8805                 if (cdb->byte2 & SWV_DPO)
8806                         flags |= CTL_LLF_DPO;
8807                 lba = scsi_8btou64(cdb->addr);
8808                 num_blocks = scsi_4btoul(cdb->length);
8809                 break;
8810         }
8811         default:
8812                 /*
8813                  * We got a command we don't support.  This shouldn't
8814                  * happen, commands should be filtered out above us.
8815                  */
8816                 ctl_set_invalid_opcode(ctsio);
8817                 ctl_done((union ctl_io *)ctsio);
8818
8819                 return (CTL_RETVAL_COMPLETE);
8820                 break; /* NOTREACHED */
8821         }
8822
8823         /*
8824          * The first check is to make sure we're in bounds, the second
8825          * check is to catch wrap-around problems.  If the lba + num blocks
8826          * is less than the lba, then we've wrapped around and the block
8827          * range is invalid anyway.
8828          */
8829         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8830          || ((lba + num_blocks) < lba)) {
8831                 ctl_set_lba_out_of_range(ctsio,
8832                     MAX(lba, lun->be_lun->maxlba + 1));
8833                 ctl_done((union ctl_io *)ctsio);
8834                 return (CTL_RETVAL_COMPLETE);
8835         }
8836
8837         /*
8838          * According to SBC-3, a transfer length of 0 is not an error.
8839          * Note that this cannot happen with WRITE(6) or READ(6), since 0
8840          * translates to 256 blocks for those commands.
8841          */
8842         if (num_blocks == 0) {
8843                 ctl_set_success(ctsio);
8844                 ctl_done((union ctl_io *)ctsio);
8845                 return (CTL_RETVAL_COMPLETE);
8846         }
8847
8848         /* Set FUA and/or DPO if caches are disabled. */
8849         if (isread) {
8850                 if ((lun->MODE_CACHING.flags1 & SCP_RCD) != 0)
8851                         flags |= CTL_LLF_FUA | CTL_LLF_DPO;
8852         } else {
8853                 if ((lun->MODE_CACHING.flags1 & SCP_WCE) == 0)
8854                         flags |= CTL_LLF_FUA;
8855         }
8856
8857         lbalen = (struct ctl_lba_len_flags *)
8858             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8859         lbalen->lba = lba;
8860         lbalen->len = num_blocks;
8861         lbalen->flags = (isread ? CTL_LLF_READ : CTL_LLF_WRITE) | flags;
8862
8863         ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
8864         ctsio->kern_rel_offset = 0;
8865
8866         CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
8867
8868         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8869         return (retval);
8870 }
8871
8872 static int
8873 ctl_cnw_cont(union ctl_io *io)
8874 {
8875         struct ctl_lun *lun = CTL_LUN(io);
8876         struct ctl_scsiio *ctsio;
8877         struct ctl_lba_len_flags *lbalen;
8878         int retval;
8879
8880         ctsio = &io->scsiio;
8881         ctsio->io_hdr.status = CTL_STATUS_NONE;
8882         ctsio->io_hdr.flags &= ~CTL_FLAG_IO_CONT;
8883         lbalen = (struct ctl_lba_len_flags *)
8884             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8885         lbalen->flags &= ~CTL_LLF_COMPARE;
8886         lbalen->flags |= CTL_LLF_WRITE;
8887
8888         CTL_DEBUG_PRINT(("ctl_cnw_cont: calling data_submit()\n"));
8889         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8890         return (retval);
8891 }
8892
8893 int
8894 ctl_cnw(struct ctl_scsiio *ctsio)
8895 {
8896         struct ctl_lun *lun = CTL_LUN(ctsio);
8897         struct ctl_lba_len_flags *lbalen;
8898         uint64_t lba;
8899         uint32_t num_blocks;
8900         int flags, retval;
8901
8902         CTL_DEBUG_PRINT(("ctl_cnw: command: %#x\n", ctsio->cdb[0]));
8903
8904         flags = 0;
8905         switch (ctsio->cdb[0]) {
8906         case COMPARE_AND_WRITE: {
8907                 struct scsi_compare_and_write *cdb;
8908
8909                 cdb = (struct scsi_compare_and_write *)ctsio->cdb;
8910                 if (cdb->byte2 & SRW10_FUA)
8911                         flags |= CTL_LLF_FUA;
8912                 if (cdb->byte2 & SRW10_DPO)
8913                         flags |= CTL_LLF_DPO;
8914                 lba = scsi_8btou64(cdb->addr);
8915                 num_blocks = cdb->length;
8916                 break;
8917         }
8918         default:
8919                 /*
8920                  * We got a command we don't support.  This shouldn't
8921                  * happen, commands should be filtered out above us.
8922                  */
8923                 ctl_set_invalid_opcode(ctsio);
8924                 ctl_done((union ctl_io *)ctsio);
8925
8926                 return (CTL_RETVAL_COMPLETE);
8927                 break; /* NOTREACHED */
8928         }
8929
8930         /*
8931          * The first check is to make sure we're in bounds, the second
8932          * check is to catch wrap-around problems.  If the lba + num blocks
8933          * is less than the lba, then we've wrapped around and the block
8934          * range is invalid anyway.
8935          */
8936         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8937          || ((lba + num_blocks) < lba)) {
8938                 ctl_set_lba_out_of_range(ctsio,
8939                     MAX(lba, lun->be_lun->maxlba + 1));
8940                 ctl_done((union ctl_io *)ctsio);
8941                 return (CTL_RETVAL_COMPLETE);
8942         }
8943
8944         /*
8945          * According to SBC-3, a transfer length of 0 is not an error.
8946          */
8947         if (num_blocks == 0) {
8948                 ctl_set_success(ctsio);
8949                 ctl_done((union ctl_io *)ctsio);
8950                 return (CTL_RETVAL_COMPLETE);
8951         }
8952
8953         /* Set FUA if write cache is disabled. */
8954         if ((lun->MODE_CACHING.flags1 & SCP_WCE) == 0)
8955                 flags |= CTL_LLF_FUA;
8956
8957         ctsio->kern_total_len = 2 * num_blocks * lun->be_lun->blocksize;
8958         ctsio->kern_rel_offset = 0;
8959
8960         /*
8961          * Set the IO_CONT flag, so that if this I/O gets passed to
8962          * ctl_data_submit_done(), it'll get passed back to
8963          * ctl_ctl_cnw_cont() for further processing.
8964          */
8965         ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
8966         ctsio->io_cont = ctl_cnw_cont;
8967
8968         lbalen = (struct ctl_lba_len_flags *)
8969             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8970         lbalen->lba = lba;
8971         lbalen->len = num_blocks;
8972         lbalen->flags = CTL_LLF_COMPARE | flags;
8973
8974         CTL_DEBUG_PRINT(("ctl_cnw: calling data_submit()\n"));
8975         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8976         return (retval);
8977 }
8978
8979 int
8980 ctl_verify(struct ctl_scsiio *ctsio)
8981 {
8982         struct ctl_lun *lun = CTL_LUN(ctsio);
8983         struct ctl_lba_len_flags *lbalen;
8984         uint64_t lba;
8985         uint32_t num_blocks;
8986         int bytchk, flags;
8987         int retval;
8988
8989         CTL_DEBUG_PRINT(("ctl_verify: command: %#x\n", ctsio->cdb[0]));
8990
8991         bytchk = 0;
8992         flags = CTL_LLF_FUA;
8993         switch (ctsio->cdb[0]) {
8994         case VERIFY_10: {
8995                 struct scsi_verify_10 *cdb;
8996
8997                 cdb = (struct scsi_verify_10 *)ctsio->cdb;
8998                 if (cdb->byte2 & SVFY_BYTCHK)
8999                         bytchk = 1;
9000                 if (cdb->byte2 & SVFY_DPO)
9001                         flags |= CTL_LLF_DPO;
9002                 lba = scsi_4btoul(cdb->addr);
9003                 num_blocks = scsi_2btoul(cdb->length);
9004                 break;
9005         }
9006         case VERIFY_12: {
9007                 struct scsi_verify_12 *cdb;
9008
9009                 cdb = (struct scsi_verify_12 *)ctsio->cdb;
9010                 if (cdb->byte2 & SVFY_BYTCHK)
9011                         bytchk = 1;
9012                 if (cdb->byte2 & SVFY_DPO)
9013                         flags |= CTL_LLF_DPO;
9014                 lba = scsi_4btoul(cdb->addr);
9015                 num_blocks = scsi_4btoul(cdb->length);
9016                 break;
9017         }
9018         case VERIFY_16: {
9019                 struct scsi_rw_16 *cdb;
9020
9021                 cdb = (struct scsi_rw_16 *)ctsio->cdb;
9022                 if (cdb->byte2 & SVFY_BYTCHK)
9023                         bytchk = 1;
9024                 if (cdb->byte2 & SVFY_DPO)
9025                         flags |= CTL_LLF_DPO;
9026                 lba = scsi_8btou64(cdb->addr);
9027                 num_blocks = scsi_4btoul(cdb->length);
9028                 break;
9029         }
9030         default:
9031                 /*
9032                  * We got a command we don't support.  This shouldn't
9033                  * happen, commands should be filtered out above us.
9034                  */
9035                 ctl_set_invalid_opcode(ctsio);
9036                 ctl_done((union ctl_io *)ctsio);
9037                 return (CTL_RETVAL_COMPLETE);
9038         }
9039
9040         /*
9041          * The first check is to make sure we're in bounds, the second
9042          * check is to catch wrap-around problems.  If the lba + num blocks
9043          * is less than the lba, then we've wrapped around and the block
9044          * range is invalid anyway.
9045          */
9046         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9047          || ((lba + num_blocks) < lba)) {
9048                 ctl_set_lba_out_of_range(ctsio,
9049                     MAX(lba, lun->be_lun->maxlba + 1));
9050                 ctl_done((union ctl_io *)ctsio);
9051                 return (CTL_RETVAL_COMPLETE);
9052         }
9053
9054         /*
9055          * According to SBC-3, a transfer length of 0 is not an error.
9056          */
9057         if (num_blocks == 0) {
9058                 ctl_set_success(ctsio);
9059                 ctl_done((union ctl_io *)ctsio);
9060                 return (CTL_RETVAL_COMPLETE);
9061         }
9062
9063         lbalen = (struct ctl_lba_len_flags *)
9064             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9065         lbalen->lba = lba;
9066         lbalen->len = num_blocks;
9067         if (bytchk) {
9068                 lbalen->flags = CTL_LLF_COMPARE | flags;
9069                 ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9070         } else {
9071                 lbalen->flags = CTL_LLF_VERIFY | flags;
9072                 ctsio->kern_total_len = 0;
9073         }
9074         ctsio->kern_rel_offset = 0;
9075
9076         CTL_DEBUG_PRINT(("ctl_verify: calling data_submit()\n"));
9077         retval = lun->backend->data_submit((union ctl_io *)ctsio);
9078         return (retval);
9079 }
9080
9081 int
9082 ctl_report_luns(struct ctl_scsiio *ctsio)
9083 {
9084         struct ctl_softc *softc = CTL_SOFTC(ctsio);
9085         struct ctl_port *port = CTL_PORT(ctsio);
9086         struct ctl_lun *lun, *request_lun = CTL_LUN(ctsio);
9087         struct scsi_report_luns *cdb;
9088         struct scsi_report_luns_data *lun_data;
9089         int num_filled, num_luns, num_port_luns, retval;
9090         uint32_t alloc_len, lun_datalen;
9091         uint32_t initidx, targ_lun_id, lun_id;
9092
9093         retval = CTL_RETVAL_COMPLETE;
9094         cdb = (struct scsi_report_luns *)ctsio->cdb;
9095
9096         CTL_DEBUG_PRINT(("ctl_report_luns\n"));
9097
9098         num_luns = 0;
9099         num_port_luns = port->lun_map ? port->lun_map_size : CTL_MAX_LUNS;
9100         mtx_lock(&softc->ctl_lock);
9101         for (targ_lun_id = 0; targ_lun_id < num_port_luns; targ_lun_id++) {
9102                 if (ctl_lun_map_from_port(port, targ_lun_id) != UINT32_MAX)
9103                         num_luns++;
9104         }
9105         mtx_unlock(&softc->ctl_lock);
9106
9107         switch (cdb->select_report) {
9108         case RPL_REPORT_DEFAULT:
9109         case RPL_REPORT_ALL:
9110         case RPL_REPORT_NONSUBSID:
9111                 break;
9112         case RPL_REPORT_WELLKNOWN:
9113         case RPL_REPORT_ADMIN:
9114         case RPL_REPORT_CONGLOM:
9115                 num_luns = 0;
9116                 break;
9117         default:
9118                 ctl_set_invalid_field(ctsio,
9119                                       /*sks_valid*/ 1,
9120                                       /*command*/ 1,
9121                                       /*field*/ 2,
9122                                       /*bit_valid*/ 0,
9123                                       /*bit*/ 0);
9124                 ctl_done((union ctl_io *)ctsio);
9125                 return (retval);
9126                 break; /* NOTREACHED */
9127         }
9128
9129         alloc_len = scsi_4btoul(cdb->length);
9130         /*
9131          * The initiator has to allocate at least 16 bytes for this request,
9132          * so he can at least get the header and the first LUN.  Otherwise
9133          * we reject the request (per SPC-3 rev 14, section 6.21).
9134          */
9135         if (alloc_len < (sizeof(struct scsi_report_luns_data) +
9136             sizeof(struct scsi_report_luns_lundata))) {
9137                 ctl_set_invalid_field(ctsio,
9138                                       /*sks_valid*/ 1,
9139                                       /*command*/ 1,
9140                                       /*field*/ 6,
9141                                       /*bit_valid*/ 0,
9142                                       /*bit*/ 0);
9143                 ctl_done((union ctl_io *)ctsio);
9144                 return (retval);
9145         }
9146
9147         lun_datalen = sizeof(*lun_data) +
9148                 (num_luns * sizeof(struct scsi_report_luns_lundata));
9149
9150         ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO);
9151         lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
9152         ctsio->kern_sg_entries = 0;
9153
9154         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9155
9156         mtx_lock(&softc->ctl_lock);
9157         for (targ_lun_id = 0, num_filled = 0;
9158             targ_lun_id < num_port_luns && num_filled < num_luns;
9159             targ_lun_id++) {
9160                 lun_id = ctl_lun_map_from_port(port, targ_lun_id);
9161                 if (lun_id == UINT32_MAX)
9162                         continue;
9163                 lun = softc->ctl_luns[lun_id];
9164                 if (lun == NULL)
9165                         continue;
9166
9167                 be64enc(lun_data->luns[num_filled++].lundata,
9168                     ctl_encode_lun(targ_lun_id));
9169
9170                 /*
9171                  * According to SPC-3, rev 14 section 6.21:
9172                  *
9173                  * "The execution of a REPORT LUNS command to any valid and
9174                  * installed logical unit shall clear the REPORTED LUNS DATA
9175                  * HAS CHANGED unit attention condition for all logical
9176                  * units of that target with respect to the requesting
9177                  * initiator. A valid and installed logical unit is one
9178                  * having a PERIPHERAL QUALIFIER of 000b in the standard
9179                  * INQUIRY data (see 6.4.2)."
9180                  *
9181                  * If request_lun is NULL, the LUN this report luns command
9182                  * was issued to is either disabled or doesn't exist. In that
9183                  * case, we shouldn't clear any pending lun change unit
9184                  * attention.
9185                  */
9186                 if (request_lun != NULL) {
9187                         mtx_lock(&lun->lun_lock);
9188                         ctl_clr_ua(lun, initidx, CTL_UA_LUN_CHANGE);
9189                         mtx_unlock(&lun->lun_lock);
9190                 }
9191         }
9192         mtx_unlock(&softc->ctl_lock);
9193
9194         /*
9195          * It's quite possible that we've returned fewer LUNs than we allocated
9196          * space for.  Trim it.
9197          */
9198         lun_datalen = sizeof(*lun_data) +
9199                 (num_filled * sizeof(struct scsi_report_luns_lundata));
9200
9201         if (lun_datalen < alloc_len) {
9202                 ctsio->residual = alloc_len - lun_datalen;
9203                 ctsio->kern_data_len = lun_datalen;
9204                 ctsio->kern_total_len = lun_datalen;
9205         } else {
9206                 ctsio->residual = 0;
9207                 ctsio->kern_data_len = alloc_len;
9208                 ctsio->kern_total_len = alloc_len;
9209         }
9210         ctsio->kern_data_resid = 0;
9211         ctsio->kern_rel_offset = 0;
9212         ctsio->kern_sg_entries = 0;
9213
9214         /*
9215          * We set this to the actual data length, regardless of how much
9216          * space we actually have to return results.  If the user looks at
9217          * this value, he'll know whether or not he allocated enough space
9218          * and reissue the command if necessary.  We don't support well
9219          * known logical units, so if the user asks for that, return none.
9220          */
9221         scsi_ulto4b(lun_datalen - 8, lun_data->length);
9222
9223         /*
9224          * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
9225          * this request.
9226          */
9227         ctl_set_success(ctsio);
9228         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9229         ctsio->be_move_done = ctl_config_move_done;
9230         ctl_datamove((union ctl_io *)ctsio);
9231         return (retval);
9232 }
9233
9234 int
9235 ctl_request_sense(struct ctl_scsiio *ctsio)
9236 {
9237         struct ctl_softc *softc = CTL_SOFTC(ctsio);
9238         struct ctl_lun *lun = CTL_LUN(ctsio);
9239         struct scsi_request_sense *cdb;
9240         struct scsi_sense_data *sense_ptr;
9241         uint32_t initidx;
9242         int have_error;
9243         u_int sense_len = SSD_FULL_SIZE;
9244         scsi_sense_data_type sense_format;
9245         ctl_ua_type ua_type;
9246         uint8_t asc = 0, ascq = 0;
9247
9248         cdb = (struct scsi_request_sense *)ctsio->cdb;
9249
9250         CTL_DEBUG_PRINT(("ctl_request_sense\n"));
9251
9252         /*
9253          * Determine which sense format the user wants.
9254          */
9255         if (cdb->byte2 & SRS_DESC)
9256                 sense_format = SSD_TYPE_DESC;
9257         else
9258                 sense_format = SSD_TYPE_FIXED;
9259
9260         ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
9261         sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
9262         ctsio->kern_sg_entries = 0;
9263
9264         /*
9265          * struct scsi_sense_data, which is currently set to 256 bytes, is
9266          * larger than the largest allowed value for the length field in the
9267          * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
9268          */
9269         ctsio->residual = 0;
9270         ctsio->kern_data_len = cdb->length;
9271         ctsio->kern_total_len = cdb->length;
9272
9273         ctsio->kern_data_resid = 0;
9274         ctsio->kern_rel_offset = 0;
9275         ctsio->kern_sg_entries = 0;
9276
9277         /*
9278          * If we don't have a LUN, we don't have any pending sense.
9279          */
9280         if (lun == NULL ||
9281             ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
9282              softc->ha_link < CTL_HA_LINK_UNKNOWN)) {
9283                 /* "Logical unit not supported" */
9284                 ctl_set_sense_data(sense_ptr, &sense_len, NULL, sense_format,
9285                     /*current_error*/ 1,
9286                     /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
9287                     /*asc*/ 0x25,
9288                     /*ascq*/ 0x00,
9289                     SSD_ELEM_NONE);
9290                 goto send;
9291         }
9292
9293         have_error = 0;
9294         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9295         /*
9296          * Check for pending sense, and then for pending unit attentions.
9297          * Pending sense gets returned first, then pending unit attentions.
9298          */
9299         mtx_lock(&lun->lun_lock);
9300 #ifdef CTL_WITH_CA
9301         if (ctl_is_set(lun->have_ca, initidx)) {
9302                 scsi_sense_data_type stored_format;
9303
9304                 /*
9305                  * Check to see which sense format was used for the stored
9306                  * sense data.
9307                  */
9308                 stored_format = scsi_sense_type(&lun->pending_sense[initidx]);
9309
9310                 /*
9311                  * If the user requested a different sense format than the
9312                  * one we stored, then we need to convert it to the other
9313                  * format.  If we're going from descriptor to fixed format
9314                  * sense data, we may lose things in translation, depending
9315                  * on what options were used.
9316                  *
9317                  * If the stored format is SSD_TYPE_NONE (i.e. invalid),
9318                  * for some reason we'll just copy it out as-is.
9319                  */
9320                 if ((stored_format == SSD_TYPE_FIXED)
9321                  && (sense_format == SSD_TYPE_DESC))
9322                         ctl_sense_to_desc((struct scsi_sense_data_fixed *)
9323                             &lun->pending_sense[initidx],
9324                             (struct scsi_sense_data_desc *)sense_ptr);
9325                 else if ((stored_format == SSD_TYPE_DESC)
9326                       && (sense_format == SSD_TYPE_FIXED))
9327                         ctl_sense_to_fixed((struct scsi_sense_data_desc *)
9328                             &lun->pending_sense[initidx],
9329                             (struct scsi_sense_data_fixed *)sense_ptr);
9330                 else
9331                         memcpy(sense_ptr, &lun->pending_sense[initidx],
9332                                MIN(sizeof(*sense_ptr),
9333                                sizeof(lun->pending_sense[initidx])));
9334
9335                 ctl_clear_mask(lun->have_ca, initidx);
9336                 have_error = 1;
9337         } else
9338 #endif
9339         if (have_error == 0) {
9340                 ua_type = ctl_build_ua(lun, initidx, sense_ptr, &sense_len,
9341                     sense_format);
9342                 if (ua_type != CTL_UA_NONE)
9343                         have_error = 1;
9344         }
9345         if (have_error == 0) {
9346                 /*
9347                  * Report informational exception if have one and allowed.
9348                  */
9349                 if (lun->MODE_IE.mrie != SIEP_MRIE_NO) {
9350                         asc = lun->ie_asc;
9351                         ascq = lun->ie_ascq;
9352                 }
9353                 ctl_set_sense_data(sense_ptr, &sense_len, lun, sense_format,
9354                     /*current_error*/ 1,
9355                     /*sense_key*/ SSD_KEY_NO_SENSE,
9356                     /*asc*/ asc,
9357                     /*ascq*/ ascq,
9358                     SSD_ELEM_NONE);
9359         }
9360         mtx_unlock(&lun->lun_lock);
9361
9362 send:
9363         /*
9364          * We report the SCSI status as OK, since the status of the command
9365          * itself is OK.  We're reporting sense as parameter data.
9366          */
9367         ctl_set_success(ctsio);
9368         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9369         ctsio->be_move_done = ctl_config_move_done;
9370         ctl_datamove((union ctl_io *)ctsio);
9371         return (CTL_RETVAL_COMPLETE);
9372 }
9373
9374 int
9375 ctl_tur(struct ctl_scsiio *ctsio)
9376 {
9377
9378         CTL_DEBUG_PRINT(("ctl_tur\n"));
9379
9380         ctl_set_success(ctsio);
9381         ctl_done((union ctl_io *)ctsio);
9382
9383         return (CTL_RETVAL_COMPLETE);
9384 }
9385
9386 /*
9387  * SCSI VPD page 0x00, the Supported VPD Pages page.
9388  */
9389 static int
9390 ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
9391 {
9392         struct ctl_lun *lun = CTL_LUN(ctsio);
9393         struct scsi_vpd_supported_pages *pages;
9394         int sup_page_size;
9395         int p;
9396
9397         sup_page_size = sizeof(struct scsi_vpd_supported_pages) *
9398             SCSI_EVPD_NUM_SUPPORTED_PAGES;
9399         ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO);
9400         pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
9401         ctsio->kern_sg_entries = 0;
9402
9403         if (sup_page_size < alloc_len) {
9404                 ctsio->residual = alloc_len - sup_page_size;
9405                 ctsio->kern_data_len = sup_page_size;
9406                 ctsio->kern_total_len = sup_page_size;
9407         } else {
9408                 ctsio->residual = 0;
9409                 ctsio->kern_data_len = alloc_len;
9410                 ctsio->kern_total_len = alloc_len;
9411         }
9412         ctsio->kern_data_resid = 0;
9413         ctsio->kern_rel_offset = 0;
9414         ctsio->kern_sg_entries = 0;
9415
9416         /*
9417          * The control device is always connected.  The disk device, on the
9418          * other hand, may not be online all the time.  Need to change this
9419          * to figure out whether the disk device is actually online or not.
9420          */
9421         if (lun != NULL)
9422                 pages->device = (SID_QUAL_LU_CONNECTED << 5) |
9423                                 lun->be_lun->lun_type;
9424         else
9425                 pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9426
9427         p = 0;
9428         /* Supported VPD pages */
9429         pages->page_list[p++] = SVPD_SUPPORTED_PAGES;
9430         /* Serial Number */
9431         pages->page_list[p++] = SVPD_UNIT_SERIAL_NUMBER;
9432         /* Device Identification */
9433         pages->page_list[p++] = SVPD_DEVICE_ID;
9434         /* Extended INQUIRY Data */
9435         pages->page_list[p++] = SVPD_EXTENDED_INQUIRY_DATA;
9436         /* Mode Page Policy */
9437         pages->page_list[p++] = SVPD_MODE_PAGE_POLICY;
9438         /* SCSI Ports */
9439         pages->page_list[p++] = SVPD_SCSI_PORTS;
9440         /* Third-party Copy */
9441         pages->page_list[p++] = SVPD_SCSI_TPC;
9442         if (lun != NULL && lun->be_lun->lun_type == T_DIRECT) {
9443                 /* Block limits */
9444                 pages->page_list[p++] = SVPD_BLOCK_LIMITS;
9445                 /* Block Device Characteristics */
9446                 pages->page_list[p++] = SVPD_BDC;
9447                 /* Logical Block Provisioning */
9448                 pages->page_list[p++] = SVPD_LBP;
9449         }
9450         pages->length = p;
9451
9452         ctl_set_success(ctsio);
9453         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9454         ctsio->be_move_done = ctl_config_move_done;
9455         ctl_datamove((union ctl_io *)ctsio);
9456         return (CTL_RETVAL_COMPLETE);
9457 }
9458
9459 /*
9460  * SCSI VPD page 0x80, the Unit Serial Number page.
9461  */
9462 static int
9463 ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
9464 {
9465         struct ctl_lun *lun = CTL_LUN(ctsio);
9466         struct scsi_vpd_unit_serial_number *sn_ptr;
9467         int data_len;
9468
9469         data_len = 4 + CTL_SN_LEN;
9470         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9471         sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
9472         if (data_len < alloc_len) {
9473                 ctsio->residual = alloc_len - data_len;
9474                 ctsio->kern_data_len = data_len;
9475                 ctsio->kern_total_len = data_len;
9476         } else {
9477                 ctsio->residual = 0;
9478                 ctsio->kern_data_len = alloc_len;
9479                 ctsio->kern_total_len = alloc_len;
9480         }
9481         ctsio->kern_data_resid = 0;
9482         ctsio->kern_rel_offset = 0;
9483         ctsio->kern_sg_entries = 0;
9484
9485         /*
9486          * The control device is always connected.  The disk device, on the
9487          * other hand, may not be online all the time.  Need to change this
9488          * to figure out whether the disk device is actually online or not.
9489          */
9490         if (lun != NULL)
9491                 sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9492                                   lun->be_lun->lun_type;
9493         else
9494                 sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9495
9496         sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9497         sn_ptr->length = CTL_SN_LEN;
9498         /*
9499          * If we don't have a LUN, we just leave the serial number as
9500          * all spaces.
9501          */
9502         if (lun != NULL) {
9503                 strncpy((char *)sn_ptr->serial_num,
9504                         (char *)lun->be_lun->serial_num, CTL_SN_LEN);
9505         } else
9506                 memset(sn_ptr->serial_num, 0x20, CTL_SN_LEN);
9507
9508         ctl_set_success(ctsio);
9509         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9510         ctsio->be_move_done = ctl_config_move_done;
9511         ctl_datamove((union ctl_io *)ctsio);
9512         return (CTL_RETVAL_COMPLETE);
9513 }
9514
9515
9516 /*
9517  * SCSI VPD page 0x86, the Extended INQUIRY Data page.
9518  */
9519 static int
9520 ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len)
9521 {
9522         struct ctl_lun *lun = CTL_LUN(ctsio);
9523         struct scsi_vpd_extended_inquiry_data *eid_ptr;
9524         int data_len;
9525
9526         data_len = sizeof(struct scsi_vpd_extended_inquiry_data);
9527         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9528         eid_ptr = (struct scsi_vpd_extended_inquiry_data *)ctsio->kern_data_ptr;
9529         ctsio->kern_sg_entries = 0;
9530
9531         if (data_len < alloc_len) {
9532                 ctsio->residual = alloc_len - data_len;
9533                 ctsio->kern_data_len = data_len;
9534                 ctsio->kern_total_len = data_len;
9535         } else {
9536                 ctsio->residual = 0;
9537                 ctsio->kern_data_len = alloc_len;
9538                 ctsio->kern_total_len = alloc_len;
9539         }
9540         ctsio->kern_data_resid = 0;
9541         ctsio->kern_rel_offset = 0;
9542         ctsio->kern_sg_entries = 0;
9543
9544         /*
9545          * The control device is always connected.  The disk device, on the
9546          * other hand, may not be online all the time.
9547          */
9548         if (lun != NULL)
9549                 eid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9550                                      lun->be_lun->lun_type;
9551         else
9552                 eid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9553         eid_ptr->page_code = SVPD_EXTENDED_INQUIRY_DATA;
9554         scsi_ulto2b(data_len - 4, eid_ptr->page_length);
9555         /*
9556          * We support head of queue, ordered and simple tags.
9557          */
9558         eid_ptr->flags2 = SVPD_EID_HEADSUP | SVPD_EID_ORDSUP | SVPD_EID_SIMPSUP;
9559         /*
9560          * Volatile cache supported.
9561          */
9562         eid_ptr->flags3 = SVPD_EID_V_SUP;
9563
9564         /*
9565          * This means that we clear the REPORTED LUNS DATA HAS CHANGED unit
9566          * attention for a particular IT nexus on all LUNs once we report
9567          * it to that nexus once.  This bit is required as of SPC-4.
9568          */
9569         eid_ptr->flags4 = SVPD_EID_LUICLR;
9570
9571         /*
9572          * We support revert to defaults (RTD) bit in MODE SELECT.
9573          */
9574         eid_ptr->flags5 = SVPD_EID_RTD_SUP;
9575
9576         /*
9577          * XXX KDM in order to correctly answer this, we would need
9578          * information from the SIM to determine how much sense data it
9579          * can send.  So this would really be a path inquiry field, most
9580          * likely.  This can be set to a maximum of 252 according to SPC-4,
9581          * but the hardware may or may not be able to support that much.
9582          * 0 just means that the maximum sense data length is not reported.
9583          */
9584         eid_ptr->max_sense_length = 0;
9585
9586         ctl_set_success(ctsio);
9587         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9588         ctsio->be_move_done = ctl_config_move_done;
9589         ctl_datamove((union ctl_io *)ctsio);
9590         return (CTL_RETVAL_COMPLETE);
9591 }
9592
9593 static int
9594 ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len)
9595 {
9596         struct ctl_lun *lun = CTL_LUN(ctsio);
9597         struct scsi_vpd_mode_page_policy *mpp_ptr;
9598         int data_len;
9599
9600         data_len = sizeof(struct scsi_vpd_mode_page_policy) +
9601             sizeof(struct scsi_vpd_mode_page_policy_descr);
9602
9603         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9604         mpp_ptr = (struct scsi_vpd_mode_page_policy *)ctsio->kern_data_ptr;
9605         ctsio->kern_sg_entries = 0;
9606
9607         if (data_len < alloc_len) {
9608                 ctsio->residual = alloc_len - data_len;
9609                 ctsio->kern_data_len = data_len;
9610                 ctsio->kern_total_len = data_len;
9611         } else {
9612                 ctsio->residual = 0;
9613                 ctsio->kern_data_len = alloc_len;
9614                 ctsio->kern_total_len = alloc_len;
9615         }
9616         ctsio->kern_data_resid = 0;
9617         ctsio->kern_rel_offset = 0;
9618         ctsio->kern_sg_entries = 0;
9619
9620         /*
9621          * The control device is always connected.  The disk device, on the
9622          * other hand, may not be online all the time.
9623          */
9624         if (lun != NULL)
9625                 mpp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9626                                      lun->be_lun->lun_type;
9627         else
9628                 mpp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9629         mpp_ptr->page_code = SVPD_MODE_PAGE_POLICY;
9630         scsi_ulto2b(data_len - 4, mpp_ptr->page_length);
9631         mpp_ptr->descr[0].page_code = 0x3f;
9632         mpp_ptr->descr[0].subpage_code = 0xff;
9633         mpp_ptr->descr[0].policy = SVPD_MPP_SHARED;
9634
9635         ctl_set_success(ctsio);
9636         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9637         ctsio->be_move_done = ctl_config_move_done;
9638         ctl_datamove((union ctl_io *)ctsio);
9639         return (CTL_RETVAL_COMPLETE);
9640 }
9641
9642 /*
9643  * SCSI VPD page 0x83, the Device Identification page.
9644  */
9645 static int
9646 ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
9647 {
9648         struct ctl_softc *softc = CTL_SOFTC(ctsio);
9649         struct ctl_port *port = CTL_PORT(ctsio);
9650         struct ctl_lun *lun = CTL_LUN(ctsio);
9651         struct scsi_vpd_device_id *devid_ptr;
9652         struct scsi_vpd_id_descriptor *desc;
9653         int data_len, g;
9654         uint8_t proto;
9655
9656         data_len = sizeof(struct scsi_vpd_device_id) +
9657             sizeof(struct scsi_vpd_id_descriptor) +
9658                 sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
9659             sizeof(struct scsi_vpd_id_descriptor) +
9660                 sizeof(struct scsi_vpd_id_trgt_port_grp_id);
9661         if (lun && lun->lun_devid)
9662                 data_len += lun->lun_devid->len;
9663         if (port && port->port_devid)
9664                 data_len += port->port_devid->len;
9665         if (port && port->target_devid)
9666                 data_len += port->target_devid->len;
9667
9668         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9669         devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
9670         ctsio->kern_sg_entries = 0;
9671
9672         if (data_len < alloc_len) {
9673                 ctsio->residual = alloc_len - data_len;
9674                 ctsio->kern_data_len = data_len;
9675                 ctsio->kern_total_len = data_len;
9676         } else {
9677                 ctsio->residual = 0;
9678                 ctsio->kern_data_len = alloc_len;
9679                 ctsio->kern_total_len = alloc_len;
9680         }
9681         ctsio->kern_data_resid = 0;
9682         ctsio->kern_rel_offset = 0;
9683         ctsio->kern_sg_entries = 0;
9684
9685         /*
9686          * The control device is always connected.  The disk device, on the
9687          * other hand, may not be online all the time.
9688          */
9689         if (lun != NULL)
9690                 devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9691                                      lun->be_lun->lun_type;
9692         else
9693                 devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9694         devid_ptr->page_code = SVPD_DEVICE_ID;
9695         scsi_ulto2b(data_len - 4, devid_ptr->length);
9696
9697         if (port && port->port_type == CTL_PORT_FC)
9698                 proto = SCSI_PROTO_FC << 4;
9699         else if (port && port->port_type == CTL_PORT_ISCSI)
9700                 proto = SCSI_PROTO_ISCSI << 4;
9701         else
9702                 proto = SCSI_PROTO_SPI << 4;
9703         desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
9704
9705         /*
9706          * We're using a LUN association here.  i.e., this device ID is a
9707          * per-LUN identifier.
9708          */
9709         if (lun && lun->lun_devid) {
9710                 memcpy(desc, lun->lun_devid->data, lun->lun_devid->len);
9711                 desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9712                     lun->lun_devid->len);
9713         }
9714
9715         /*
9716          * This is for the WWPN which is a port association.
9717          */
9718         if (port && port->port_devid) {
9719                 memcpy(desc, port->port_devid->data, port->port_devid->len);
9720                 desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9721                     port->port_devid->len);
9722         }
9723
9724         /*
9725          * This is for the Relative Target Port(type 4h) identifier
9726          */
9727         desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9728         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9729             SVPD_ID_TYPE_RELTARG;
9730         desc->length = 4;
9731         scsi_ulto2b(ctsio->io_hdr.nexus.targ_port, &desc->identifier[2]);
9732         desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9733             sizeof(struct scsi_vpd_id_rel_trgt_port_id));
9734
9735         /*
9736          * This is for the Target Port Group(type 5h) identifier
9737          */
9738         desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9739         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9740             SVPD_ID_TYPE_TPORTGRP;
9741         desc->length = 4;
9742         if (softc->is_single ||
9743             (port && port->status & CTL_PORT_STATUS_HA_SHARED))
9744                 g = 1;
9745         else
9746                 g = 2 + ctsio->io_hdr.nexus.targ_port / softc->port_cnt;
9747         scsi_ulto2b(g, &desc->identifier[2]);
9748         desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9749             sizeof(struct scsi_vpd_id_trgt_port_grp_id));
9750
9751         /*
9752          * This is for the Target identifier
9753          */
9754         if (port && port->target_devid) {
9755                 memcpy(desc, port->target_devid->data, port->target_devid->len);
9756         }
9757
9758         ctl_set_success(ctsio);
9759         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9760         ctsio->be_move_done = ctl_config_move_done;
9761         ctl_datamove((union ctl_io *)ctsio);
9762         return (CTL_RETVAL_COMPLETE);
9763 }
9764
9765 static int
9766 ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio, int alloc_len)
9767 {
9768         struct ctl_softc *softc = CTL_SOFTC(ctsio);
9769         struct ctl_lun *lun = CTL_LUN(ctsio);
9770         struct scsi_vpd_scsi_ports *sp;
9771         struct scsi_vpd_port_designation *pd;
9772         struct scsi_vpd_port_designation_cont *pdc;
9773         struct ctl_port *port;
9774         int data_len, num_target_ports, iid_len, id_len;
9775
9776         num_target_ports = 0;
9777         iid_len = 0;
9778         id_len = 0;
9779         mtx_lock(&softc->ctl_lock);
9780         STAILQ_FOREACH(port, &softc->port_list, links) {
9781                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
9782                         continue;
9783                 if (lun != NULL &&
9784                     ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
9785                         continue;
9786                 num_target_ports++;
9787                 if (port->init_devid)
9788                         iid_len += port->init_devid->len;
9789                 if (port->port_devid)
9790                         id_len += port->port_devid->len;
9791         }
9792         mtx_unlock(&softc->ctl_lock);
9793
9794         data_len = sizeof(struct scsi_vpd_scsi_ports) +
9795             num_target_ports * (sizeof(struct scsi_vpd_port_designation) +
9796              sizeof(struct scsi_vpd_port_designation_cont)) + iid_len + id_len;
9797         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9798         sp = (struct scsi_vpd_scsi_ports *)ctsio->kern_data_ptr;
9799         ctsio->kern_sg_entries = 0;
9800
9801         if (data_len < alloc_len) {
9802                 ctsio->residual = alloc_len - data_len;
9803                 ctsio->kern_data_len = data_len;
9804                 ctsio->kern_total_len = data_len;
9805         } else {
9806                 ctsio->residual = 0;
9807                 ctsio->kern_data_len = alloc_len;
9808                 ctsio->kern_total_len = alloc_len;
9809         }
9810         ctsio->kern_data_resid = 0;
9811         ctsio->kern_rel_offset = 0;
9812         ctsio->kern_sg_entries = 0;
9813
9814         /*
9815          * The control device is always connected.  The disk device, on the
9816          * other hand, may not be online all the time.  Need to change this
9817          * to figure out whether the disk device is actually online or not.
9818          */
9819         if (lun != NULL)
9820                 sp->device = (SID_QUAL_LU_CONNECTED << 5) |
9821                                   lun->be_lun->lun_type;
9822         else
9823                 sp->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9824
9825         sp->page_code = SVPD_SCSI_PORTS;
9826         scsi_ulto2b(data_len - sizeof(struct scsi_vpd_scsi_ports),
9827             sp->page_length);
9828         pd = &sp->design[0];
9829
9830         mtx_lock(&softc->ctl_lock);
9831         STAILQ_FOREACH(port, &softc->port_list, links) {
9832                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
9833                         continue;
9834                 if (lun != NULL &&
9835                     ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
9836                         continue;
9837                 scsi_ulto2b(port->targ_port, pd->relative_port_id);
9838                 if (port->init_devid) {
9839                         iid_len = port->init_devid->len;
9840                         memcpy(pd->initiator_transportid,
9841                             port->init_devid->data, port->init_devid->len);
9842                 } else
9843                         iid_len = 0;
9844                 scsi_ulto2b(iid_len, pd->initiator_transportid_length);
9845                 pdc = (struct scsi_vpd_port_designation_cont *)
9846                     (&pd->initiator_transportid[iid_len]);
9847                 if (port->port_devid) {
9848                         id_len = port->port_devid->len;
9849                         memcpy(pdc->target_port_descriptors,
9850                             port->port_devid->data, port->port_devid->len);
9851                 } else
9852                         id_len = 0;
9853                 scsi_ulto2b(id_len, pdc->target_port_descriptors_length);
9854                 pd = (struct scsi_vpd_port_designation *)
9855                     ((uint8_t *)pdc->target_port_descriptors + id_len);
9856         }
9857         mtx_unlock(&softc->ctl_lock);
9858
9859         ctl_set_success(ctsio);
9860         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9861         ctsio->be_move_done = ctl_config_move_done;
9862         ctl_datamove((union ctl_io *)ctsio);
9863         return (CTL_RETVAL_COMPLETE);
9864 }
9865
9866 static int
9867 ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio, int alloc_len)
9868 {
9869         struct ctl_lun *lun = CTL_LUN(ctsio);
9870         struct scsi_vpd_block_limits *bl_ptr;
9871         uint64_t ival;
9872
9873         ctsio->kern_data_ptr = malloc(sizeof(*bl_ptr), M_CTL, M_WAITOK | M_ZERO);
9874         bl_ptr = (struct scsi_vpd_block_limits *)ctsio->kern_data_ptr;
9875         ctsio->kern_sg_entries = 0;
9876
9877         if (sizeof(*bl_ptr) < alloc_len) {
9878                 ctsio->residual = alloc_len - sizeof(*bl_ptr);
9879                 ctsio->kern_data_len = sizeof(*bl_ptr);
9880                 ctsio->kern_total_len = sizeof(*bl_ptr);
9881         } else {
9882                 ctsio->residual = 0;
9883                 ctsio->kern_data_len = alloc_len;
9884                 ctsio->kern_total_len = alloc_len;
9885         }
9886         ctsio->kern_data_resid = 0;
9887         ctsio->kern_rel_offset = 0;
9888         ctsio->kern_sg_entries = 0;
9889
9890         /*
9891          * The control device is always connected.  The disk device, on the
9892          * other hand, may not be online all the time.  Need to change this
9893          * to figure out whether the disk device is actually online or not.
9894          */
9895         if (lun != NULL)
9896                 bl_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9897                                   lun->be_lun->lun_type;
9898         else
9899                 bl_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9900
9901         bl_ptr->page_code = SVPD_BLOCK_LIMITS;
9902         scsi_ulto2b(sizeof(*bl_ptr) - 4, bl_ptr->page_length);
9903         bl_ptr->max_cmp_write_len = 0xff;
9904         scsi_ulto4b(0xffffffff, bl_ptr->max_txfer_len);
9905         if (lun != NULL) {
9906                 scsi_ulto4b(lun->be_lun->opttxferlen, bl_ptr->opt_txfer_len);
9907                 if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
9908                         ival = 0xffffffff;
9909                         ctl_get_opt_number(&lun->be_lun->options,
9910                             "unmap_max_lba", &ival);
9911                         scsi_ulto4b(ival, bl_ptr->max_unmap_lba_cnt);
9912                         ival = 0xffffffff;
9913                         ctl_get_opt_number(&lun->be_lun->options,
9914                             "unmap_max_descr", &ival);
9915                         scsi_ulto4b(ival, bl_ptr->max_unmap_blk_cnt);
9916                         if (lun->be_lun->ublockexp != 0) {
9917                                 scsi_ulto4b((1 << lun->be_lun->ublockexp),
9918                                     bl_ptr->opt_unmap_grain);
9919                                 scsi_ulto4b(0x80000000 | lun->be_lun->ublockoff,
9920                                     bl_ptr->unmap_grain_align);
9921                         }
9922                 }
9923                 scsi_ulto4b(lun->be_lun->atomicblock,
9924                     bl_ptr->max_atomic_transfer_length);
9925                 scsi_ulto4b(0, bl_ptr->atomic_alignment);
9926                 scsi_ulto4b(0, bl_ptr->atomic_transfer_length_granularity);
9927                 scsi_ulto4b(0, bl_ptr->max_atomic_transfer_length_with_atomic_boundary);
9928                 scsi_ulto4b(0, bl_ptr->max_atomic_boundary_size);
9929                 ival = UINT64_MAX;
9930                 ctl_get_opt_number(&lun->be_lun->options, "write_same_max_lba", &ival);
9931                 scsi_u64to8b(ival, bl_ptr->max_write_same_length);
9932         }
9933
9934         ctl_set_success(ctsio);
9935         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9936         ctsio->be_move_done = ctl_config_move_done;
9937         ctl_datamove((union ctl_io *)ctsio);
9938         return (CTL_RETVAL_COMPLETE);
9939 }
9940
9941 static int
9942 ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len)
9943 {
9944         struct ctl_lun *lun = CTL_LUN(ctsio);
9945         struct scsi_vpd_block_device_characteristics *bdc_ptr;
9946         const char *value;
9947         u_int i;
9948
9949         ctsio->kern_data_ptr = malloc(sizeof(*bdc_ptr), M_CTL, M_WAITOK | M_ZERO);
9950         bdc_ptr = (struct scsi_vpd_block_device_characteristics *)ctsio->kern_data_ptr;
9951         ctsio->kern_sg_entries = 0;
9952
9953         if (sizeof(*bdc_ptr) < alloc_len) {
9954                 ctsio->residual = alloc_len - sizeof(*bdc_ptr);
9955                 ctsio->kern_data_len = sizeof(*bdc_ptr);
9956                 ctsio->kern_total_len = sizeof(*bdc_ptr);
9957         } else {
9958                 ctsio->residual = 0;
9959                 ctsio->kern_data_len = alloc_len;
9960                 ctsio->kern_total_len = alloc_len;
9961         }
9962         ctsio->kern_data_resid = 0;
9963         ctsio->kern_rel_offset = 0;
9964         ctsio->kern_sg_entries = 0;
9965
9966         /*
9967          * The control device is always connected.  The disk device, on the
9968          * other hand, may not be online all the time.  Need to change this
9969          * to figure out whether the disk device is actually online or not.
9970          */
9971         if (lun != NULL)
9972                 bdc_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9973                                   lun->be_lun->lun_type;
9974         else
9975                 bdc_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9976         bdc_ptr->page_code = SVPD_BDC;
9977         scsi_ulto2b(sizeof(*bdc_ptr) - 4, bdc_ptr->page_length);
9978         if (lun != NULL &&
9979             (value = ctl_get_opt(&lun->be_lun->options, "rpm")) != NULL)
9980                 i = strtol(value, NULL, 0);
9981         else
9982                 i = CTL_DEFAULT_ROTATION_RATE;
9983         scsi_ulto2b(i, bdc_ptr->medium_rotation_rate);
9984         if (lun != NULL &&
9985             (value = ctl_get_opt(&lun->be_lun->options, "formfactor")) != NULL)
9986                 i = strtol(value, NULL, 0);
9987         else
9988                 i = 0;
9989         bdc_ptr->wab_wac_ff = (i & 0x0f);
9990         bdc_ptr->flags = SVPD_FUAB | SVPD_VBULS;
9991
9992         ctl_set_success(ctsio);
9993         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9994         ctsio->be_move_done = ctl_config_move_done;
9995         ctl_datamove((union ctl_io *)ctsio);
9996         return (CTL_RETVAL_COMPLETE);
9997 }
9998
9999 static int
10000 ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len)
10001 {
10002         struct ctl_lun *lun = CTL_LUN(ctsio);
10003         struct scsi_vpd_logical_block_prov *lbp_ptr;
10004         const char *value;
10005
10006         ctsio->kern_data_ptr = malloc(sizeof(*lbp_ptr), M_CTL, M_WAITOK | M_ZERO);
10007         lbp_ptr = (struct scsi_vpd_logical_block_prov *)ctsio->kern_data_ptr;
10008         ctsio->kern_sg_entries = 0;
10009
10010         if (sizeof(*lbp_ptr) < alloc_len) {
10011                 ctsio->residual = alloc_len - sizeof(*lbp_ptr);
10012                 ctsio->kern_data_len = sizeof(*lbp_ptr);
10013                 ctsio->kern_total_len = sizeof(*lbp_ptr);
10014         } else {
10015                 ctsio->residual = 0;
10016                 ctsio->kern_data_len = alloc_len;
10017                 ctsio->kern_total_len = alloc_len;
10018         }
10019         ctsio->kern_data_resid = 0;
10020         ctsio->kern_rel_offset = 0;
10021         ctsio->kern_sg_entries = 0;
10022
10023         /*
10024          * The control device is always connected.  The disk device, on the
10025          * other hand, may not be online all the time.  Need to change this
10026          * to figure out whether the disk device is actually online or not.
10027          */
10028         if (lun != NULL)
10029                 lbp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10030                                   lun->be_lun->lun_type;
10031         else
10032                 lbp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10033
10034         lbp_ptr->page_code = SVPD_LBP;
10035         scsi_ulto2b(sizeof(*lbp_ptr) - 4, lbp_ptr->page_length);
10036         lbp_ptr->threshold_exponent = CTL_LBP_EXPONENT;
10037         if (lun != NULL && lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
10038                 lbp_ptr->flags = SVPD_LBP_UNMAP | SVPD_LBP_WS16 |
10039                     SVPD_LBP_WS10 | SVPD_LBP_RZ | SVPD_LBP_ANC_SUP;
10040                 value = ctl_get_opt(&lun->be_lun->options, "provisioning_type");
10041                 if (value != NULL) {
10042                         if (strcmp(value, "resource") == 0)
10043                                 lbp_ptr->prov_type = SVPD_LBP_RESOURCE;
10044                         else if (strcmp(value, "thin") == 0)
10045                                 lbp_ptr->prov_type = SVPD_LBP_THIN;
10046                 } else
10047                         lbp_ptr->prov_type = SVPD_LBP_THIN;
10048         }
10049
10050         ctl_set_success(ctsio);
10051         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10052         ctsio->be_move_done = ctl_config_move_done;
10053         ctl_datamove((union ctl_io *)ctsio);
10054         return (CTL_RETVAL_COMPLETE);
10055 }
10056
10057 /*
10058  * INQUIRY with the EVPD bit set.
10059  */
10060 static int
10061 ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
10062 {
10063         struct ctl_lun *lun = CTL_LUN(ctsio);
10064         struct scsi_inquiry *cdb;
10065         int alloc_len, retval;
10066
10067         cdb = (struct scsi_inquiry *)ctsio->cdb;
10068         alloc_len = scsi_2btoul(cdb->length);
10069
10070         switch (cdb->page_code) {
10071         case SVPD_SUPPORTED_PAGES:
10072                 retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
10073                 break;
10074         case SVPD_UNIT_SERIAL_NUMBER:
10075                 retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
10076                 break;
10077         case SVPD_DEVICE_ID:
10078                 retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
10079                 break;
10080         case SVPD_EXTENDED_INQUIRY_DATA:
10081                 retval = ctl_inquiry_evpd_eid(ctsio, alloc_len);
10082                 break;
10083         case SVPD_MODE_PAGE_POLICY:
10084                 retval = ctl_inquiry_evpd_mpp(ctsio, alloc_len);
10085                 break;
10086         case SVPD_SCSI_PORTS:
10087                 retval = ctl_inquiry_evpd_scsi_ports(ctsio, alloc_len);
10088                 break;
10089         case SVPD_SCSI_TPC:
10090                 retval = ctl_inquiry_evpd_tpc(ctsio, alloc_len);
10091                 break;
10092         case SVPD_BLOCK_LIMITS:
10093                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
10094                         goto err;
10095                 retval = ctl_inquiry_evpd_block_limits(ctsio, alloc_len);
10096                 break;
10097         case SVPD_BDC:
10098                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
10099                         goto err;
10100                 retval = ctl_inquiry_evpd_bdc(ctsio, alloc_len);
10101                 break;
10102         case SVPD_LBP:
10103                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
10104                         goto err;
10105                 retval = ctl_inquiry_evpd_lbp(ctsio, alloc_len);
10106                 break;
10107         default:
10108 err:
10109                 ctl_set_invalid_field(ctsio,
10110                                       /*sks_valid*/ 1,
10111                                       /*command*/ 1,
10112                                       /*field*/ 2,
10113                                       /*bit_valid*/ 0,
10114                                       /*bit*/ 0);
10115                 ctl_done((union ctl_io *)ctsio);
10116                 retval = CTL_RETVAL_COMPLETE;
10117                 break;
10118         }
10119
10120         return (retval);
10121 }
10122
10123 /*
10124  * Standard INQUIRY data.
10125  */
10126 static int
10127 ctl_inquiry_std(struct ctl_scsiio *ctsio)
10128 {
10129         struct ctl_softc *softc = CTL_SOFTC(ctsio);
10130         struct ctl_port *port = CTL_PORT(ctsio);
10131         struct ctl_lun *lun = CTL_LUN(ctsio);
10132         struct scsi_inquiry_data *inq_ptr;
10133         struct scsi_inquiry *cdb;
10134         char *val;
10135         uint32_t alloc_len, data_len;
10136         ctl_port_type port_type;
10137
10138         port_type = port->port_type;
10139         if (port_type == CTL_PORT_IOCTL || port_type == CTL_PORT_INTERNAL)
10140                 port_type = CTL_PORT_SCSI;
10141
10142         cdb = (struct scsi_inquiry *)ctsio->cdb;
10143         alloc_len = scsi_2btoul(cdb->length);
10144
10145         /*
10146          * We malloc the full inquiry data size here and fill it
10147          * in.  If the user only asks for less, we'll give him
10148          * that much.
10149          */
10150         data_len = offsetof(struct scsi_inquiry_data, vendor_specific1);
10151         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10152         inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
10153         ctsio->kern_sg_entries = 0;
10154         ctsio->kern_data_resid = 0;
10155         ctsio->kern_rel_offset = 0;
10156
10157         if (data_len < alloc_len) {
10158                 ctsio->residual = alloc_len - data_len;
10159                 ctsio->kern_data_len = data_len;
10160                 ctsio->kern_total_len = data_len;
10161         } else {
10162                 ctsio->residual = 0;
10163                 ctsio->kern_data_len = alloc_len;
10164                 ctsio->kern_total_len = alloc_len;
10165         }
10166
10167         if (lun != NULL) {
10168                 if ((lun->flags & CTL_LUN_PRIMARY_SC) ||
10169                     softc->ha_link >= CTL_HA_LINK_UNKNOWN) {
10170                         inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10171                             lun->be_lun->lun_type;
10172                 } else {
10173                         inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) |
10174                             lun->be_lun->lun_type;
10175                 }
10176                 if (lun->flags & CTL_LUN_REMOVABLE)
10177                         inq_ptr->dev_qual2 |= SID_RMB;
10178         } else
10179                 inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE;
10180
10181         /* RMB in byte 2 is 0 */
10182         inq_ptr->version = SCSI_REV_SPC5;
10183
10184         /*
10185          * According to SAM-3, even if a device only supports a single
10186          * level of LUN addressing, it should still set the HISUP bit:
10187          *
10188          * 4.9.1 Logical unit numbers overview
10189          *
10190          * All logical unit number formats described in this standard are
10191          * hierarchical in structure even when only a single level in that
10192          * hierarchy is used. The HISUP bit shall be set to one in the
10193          * standard INQUIRY data (see SPC-2) when any logical unit number
10194          * format described in this standard is used.  Non-hierarchical
10195          * formats are outside the scope of this standard.
10196          *
10197          * Therefore we set the HiSup bit here.
10198          *
10199          * The response format is 2, per SPC-3.
10200          */
10201         inq_ptr->response_format = SID_HiSup | 2;
10202
10203         inq_ptr->additional_length = data_len -
10204             (offsetof(struct scsi_inquiry_data, additional_length) + 1);
10205         CTL_DEBUG_PRINT(("additional_length = %d\n",
10206                          inq_ptr->additional_length));
10207
10208         inq_ptr->spc3_flags = SPC3_SID_3PC | SPC3_SID_TPGS_IMPLICIT;
10209         if (port_type == CTL_PORT_SCSI)
10210                 inq_ptr->spc2_flags = SPC2_SID_ADDR16;
10211         inq_ptr->spc2_flags |= SPC2_SID_MultiP;
10212         inq_ptr->flags = SID_CmdQue;
10213         if (port_type == CTL_PORT_SCSI)
10214                 inq_ptr->flags |= SID_WBus16 | SID_Sync;
10215
10216         /*
10217          * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
10218          * We have 8 bytes for the vendor name, and 16 bytes for the device
10219          * name and 4 bytes for the revision.
10220          */
10221         if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10222             "vendor")) == NULL) {
10223                 strncpy(inq_ptr->vendor, CTL_VENDOR, sizeof(inq_ptr->vendor));
10224         } else {
10225                 memset(inq_ptr->vendor, ' ', sizeof(inq_ptr->vendor));
10226                 strncpy(inq_ptr->vendor, val,
10227                     min(sizeof(inq_ptr->vendor), strlen(val)));
10228         }
10229         if (lun == NULL) {
10230                 strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10231                     sizeof(inq_ptr->product));
10232         } else if ((val = ctl_get_opt(&lun->be_lun->options, "product")) == NULL) {
10233                 switch (lun->be_lun->lun_type) {
10234                 case T_DIRECT:
10235                         strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10236                             sizeof(inq_ptr->product));
10237                         break;
10238                 case T_PROCESSOR:
10239                         strncpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT,
10240                             sizeof(inq_ptr->product));
10241                         break;
10242                 case T_CDROM:
10243                         strncpy(inq_ptr->product, CTL_CDROM_PRODUCT,
10244                             sizeof(inq_ptr->product));
10245                         break;
10246                 default:
10247                         strncpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT,
10248                             sizeof(inq_ptr->product));
10249                         break;
10250                 }
10251         } else {
10252                 memset(inq_ptr->product, ' ', sizeof(inq_ptr->product));
10253                 strncpy(inq_ptr->product, val,
10254                     min(sizeof(inq_ptr->product), strlen(val)));
10255         }
10256
10257         /*
10258          * XXX make this a macro somewhere so it automatically gets
10259          * incremented when we make changes.
10260          */
10261         if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10262             "revision")) == NULL) {
10263                 strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
10264         } else {
10265                 memset(inq_ptr->revision, ' ', sizeof(inq_ptr->revision));
10266                 strncpy(inq_ptr->revision, val,
10267                     min(sizeof(inq_ptr->revision), strlen(val)));
10268         }
10269
10270         /*
10271          * For parallel SCSI, we support double transition and single
10272          * transition clocking.  We also support QAS (Quick Arbitration
10273          * and Selection) and Information Unit transfers on both the
10274          * control and array devices.
10275          */
10276         if (port_type == CTL_PORT_SCSI)
10277                 inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
10278                                     SID_SPI_IUS;
10279
10280         /* SAM-6 (no version claimed) */
10281         scsi_ulto2b(0x00C0, inq_ptr->version1);
10282         /* SPC-5 (no version claimed) */
10283         scsi_ulto2b(0x05C0, inq_ptr->version2);
10284         if (port_type == CTL_PORT_FC) {
10285                 /* FCP-2 ANSI INCITS.350:2003 */
10286                 scsi_ulto2b(0x0917, inq_ptr->version3);
10287         } else if (port_type == CTL_PORT_SCSI) {
10288                 /* SPI-4 ANSI INCITS.362:200x */
10289                 scsi_ulto2b(0x0B56, inq_ptr->version3);
10290         } else if (port_type == CTL_PORT_ISCSI) {
10291                 /* iSCSI (no version claimed) */
10292                 scsi_ulto2b(0x0960, inq_ptr->version3);
10293         } else if (port_type == CTL_PORT_SAS) {
10294                 /* SAS (no version claimed) */
10295                 scsi_ulto2b(0x0BE0, inq_ptr->version3);
10296         }
10297
10298         if (lun == NULL) {
10299                 /* SBC-4 (no version claimed) */
10300                 scsi_ulto2b(0x0600, inq_ptr->version4);
10301         } else {
10302                 switch (lun->be_lun->lun_type) {
10303                 case T_DIRECT:
10304                         /* SBC-4 (no version claimed) */
10305                         scsi_ulto2b(0x0600, inq_ptr->version4);
10306                         break;
10307                 case T_PROCESSOR:
10308                         break;
10309                 case T_CDROM:
10310                         /* MMC-6 (no version claimed) */
10311                         scsi_ulto2b(0x04E0, inq_ptr->version4);
10312                         break;
10313                 default:
10314                         break;
10315                 }
10316         }
10317
10318         ctl_set_success(ctsio);
10319         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10320         ctsio->be_move_done = ctl_config_move_done;
10321         ctl_datamove((union ctl_io *)ctsio);
10322         return (CTL_RETVAL_COMPLETE);
10323 }
10324
10325 int
10326 ctl_inquiry(struct ctl_scsiio *ctsio)
10327 {
10328         struct scsi_inquiry *cdb;
10329         int retval;
10330
10331         CTL_DEBUG_PRINT(("ctl_inquiry\n"));
10332
10333         cdb = (struct scsi_inquiry *)ctsio->cdb;
10334         if (cdb->byte2 & SI_EVPD)
10335                 retval = ctl_inquiry_evpd(ctsio);
10336         else if (cdb->page_code == 0)
10337                 retval = ctl_inquiry_std(ctsio);
10338         else {
10339                 ctl_set_invalid_field(ctsio,
10340                                       /*sks_valid*/ 1,
10341                                       /*command*/ 1,
10342                                       /*field*/ 2,
10343                                       /*bit_valid*/ 0,
10344                                       /*bit*/ 0);
10345                 ctl_done((union ctl_io *)ctsio);
10346                 return (CTL_RETVAL_COMPLETE);
10347         }
10348
10349         return (retval);
10350 }
10351
10352 int
10353 ctl_get_config(struct ctl_scsiio *ctsio)
10354 {
10355         struct ctl_lun *lun = CTL_LUN(ctsio);
10356         struct scsi_get_config_header *hdr;
10357         struct scsi_get_config_feature *feature;
10358         struct scsi_get_config *cdb;
10359         uint32_t alloc_len, data_len;
10360         int rt, starting;
10361
10362         cdb = (struct scsi_get_config *)ctsio->cdb;
10363         rt = (cdb->rt & SGC_RT_MASK);
10364         starting = scsi_2btoul(cdb->starting_feature);
10365         alloc_len = scsi_2btoul(cdb->length);
10366
10367         data_len = sizeof(struct scsi_get_config_header) +
10368             sizeof(struct scsi_get_config_feature) + 8 +
10369             sizeof(struct scsi_get_config_feature) + 8 +
10370             sizeof(struct scsi_get_config_feature) + 4 +
10371             sizeof(struct scsi_get_config_feature) + 4 +
10372             sizeof(struct scsi_get_config_feature) + 8 +
10373             sizeof(struct scsi_get_config_feature) +
10374             sizeof(struct scsi_get_config_feature) + 4 +
10375             sizeof(struct scsi_get_config_feature) + 4 +
10376             sizeof(struct scsi_get_config_feature) + 4 +
10377             sizeof(struct scsi_get_config_feature) + 4 +
10378             sizeof(struct scsi_get_config_feature) + 4 +
10379             sizeof(struct scsi_get_config_feature) + 4;
10380         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10381         ctsio->kern_sg_entries = 0;
10382         ctsio->kern_data_resid = 0;
10383         ctsio->kern_rel_offset = 0;
10384
10385         hdr = (struct scsi_get_config_header *)ctsio->kern_data_ptr;
10386         if (lun->flags & CTL_LUN_NO_MEDIA)
10387                 scsi_ulto2b(0x0000, hdr->current_profile);
10388         else
10389                 scsi_ulto2b(0x0010, hdr->current_profile);
10390         feature = (struct scsi_get_config_feature *)(hdr + 1);
10391
10392         if (starting > 0x003b)
10393                 goto done;
10394         if (starting > 0x003a)
10395                 goto f3b;
10396         if (starting > 0x002b)
10397                 goto f3a;
10398         if (starting > 0x002a)
10399                 goto f2b;
10400         if (starting > 0x001f)
10401                 goto f2a;
10402         if (starting > 0x001e)
10403                 goto f1f;
10404         if (starting > 0x001d)
10405                 goto f1e;
10406         if (starting > 0x0010)
10407                 goto f1d;
10408         if (starting > 0x0003)
10409                 goto f10;
10410         if (starting > 0x0002)
10411                 goto f3;
10412         if (starting > 0x0001)
10413                 goto f2;
10414         if (starting > 0x0000)
10415                 goto f1;
10416
10417         /* Profile List */
10418         scsi_ulto2b(0x0000, feature->feature_code);
10419         feature->flags = SGC_F_PERSISTENT | SGC_F_CURRENT;
10420         feature->add_length = 8;
10421         scsi_ulto2b(0x0008, &feature->feature_data[0]); /* CD-ROM */
10422         feature->feature_data[2] = 0x00;
10423         scsi_ulto2b(0x0010, &feature->feature_data[4]); /* DVD-ROM */
10424         feature->feature_data[6] = 0x01;
10425         feature = (struct scsi_get_config_feature *)
10426             &feature->feature_data[feature->add_length];
10427
10428 f1:     /* Core */
10429         scsi_ulto2b(0x0001, feature->feature_code);
10430         feature->flags = 0x08 | SGC_F_PERSISTENT | SGC_F_CURRENT;
10431         feature->add_length = 8;
10432         scsi_ulto4b(0x00000000, &feature->feature_data[0]);
10433         feature->feature_data[4] = 0x03;
10434         feature = (struct scsi_get_config_feature *)
10435             &feature->feature_data[feature->add_length];
10436
10437 f2:     /* Morphing */
10438         scsi_ulto2b(0x0002, feature->feature_code);
10439         feature->flags = 0x04 | SGC_F_PERSISTENT | SGC_F_CURRENT;
10440         feature->add_length = 4;
10441         feature->feature_data[0] = 0x02;
10442         feature = (struct scsi_get_config_feature *)
10443             &feature->feature_data[feature->add_length];
10444
10445 f3:     /* Removable Medium */
10446         scsi_ulto2b(0x0003, feature->feature_code);
10447         feature->flags = 0x04 | SGC_F_PERSISTENT | SGC_F_CURRENT;
10448         feature->add_length = 4;
10449         feature->feature_data[0] = 0x39;
10450         feature = (struct scsi_get_config_feature *)
10451             &feature->feature_data[feature->add_length];
10452
10453         if (rt == SGC_RT_CURRENT && (lun->flags & CTL_LUN_NO_MEDIA))
10454                 goto done;
10455
10456 f10:    /* Random Read */
10457         scsi_ulto2b(0x0010, feature->feature_code);
10458         feature->flags = 0x00;
10459         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10460                 feature->flags |= SGC_F_CURRENT;
10461         feature->add_length = 8;
10462         scsi_ulto4b(lun->be_lun->blocksize, &feature->feature_data[0]);
10463         scsi_ulto2b(1, &feature->feature_data[4]);
10464         feature->feature_data[6] = 0x00;
10465         feature = (struct scsi_get_config_feature *)
10466             &feature->feature_data[feature->add_length];
10467
10468 f1d:    /* Multi-Read */
10469         scsi_ulto2b(0x001D, feature->feature_code);
10470         feature->flags = 0x00;
10471         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10472                 feature->flags |= SGC_F_CURRENT;
10473         feature->add_length = 0;
10474         feature = (struct scsi_get_config_feature *)
10475             &feature->feature_data[feature->add_length];
10476
10477 f1e:    /* CD Read */
10478         scsi_ulto2b(0x001E, feature->feature_code);
10479         feature->flags = 0x00;
10480         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10481                 feature->flags |= SGC_F_CURRENT;
10482         feature->add_length = 4;
10483         feature->feature_data[0] = 0x00;
10484         feature = (struct scsi_get_config_feature *)
10485             &feature->feature_data[feature->add_length];
10486
10487 f1f:    /* DVD Read */
10488         scsi_ulto2b(0x001F, feature->feature_code);
10489         feature->flags = 0x08;
10490         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10491                 feature->flags |= SGC_F_CURRENT;
10492         feature->add_length = 4;
10493         feature->feature_data[0] = 0x01;
10494         feature->feature_data[2] = 0x03;
10495         feature = (struct scsi_get_config_feature *)
10496             &feature->feature_data[feature->add_length];
10497
10498 f2a:    /* DVD+RW */
10499         scsi_ulto2b(0x002A, feature->feature_code);
10500         feature->flags = 0x04;
10501         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10502                 feature->flags |= SGC_F_CURRENT;
10503         feature->add_length = 4;
10504         feature->feature_data[0] = 0x00;
10505         feature->feature_data[1] = 0x00;
10506         feature = (struct scsi_get_config_feature *)
10507             &feature->feature_data[feature->add_length];
10508
10509 f2b:    /* DVD+R */
10510         scsi_ulto2b(0x002B, feature->feature_code);
10511         feature->flags = 0x00;
10512         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10513                 feature->flags |= SGC_F_CURRENT;
10514         feature->add_length = 4;
10515         feature->feature_data[0] = 0x00;
10516         feature = (struct scsi_get_config_feature *)
10517             &feature->feature_data[feature->add_length];
10518
10519 f3a:    /* DVD+RW Dual Layer */
10520         scsi_ulto2b(0x003A, feature->feature_code);
10521         feature->flags = 0x00;
10522         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10523                 feature->flags |= SGC_F_CURRENT;
10524         feature->add_length = 4;
10525         feature->feature_data[0] = 0x00;
10526         feature->feature_data[1] = 0x00;
10527         feature = (struct scsi_get_config_feature *)
10528             &feature->feature_data[feature->add_length];
10529
10530 f3b:    /* DVD+R Dual Layer */
10531         scsi_ulto2b(0x003B, feature->feature_code);
10532         feature->flags = 0x00;
10533         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10534                 feature->flags |= SGC_F_CURRENT;
10535         feature->add_length = 4;
10536         feature->feature_data[0] = 0x00;
10537         feature = (struct scsi_get_config_feature *)
10538             &feature->feature_data[feature->add_length];
10539
10540 done:
10541         data_len = (uint8_t *)feature - (uint8_t *)hdr;
10542         if (rt == SGC_RT_SPECIFIC && data_len > 4) {
10543                 feature = (struct scsi_get_config_feature *)(hdr + 1);
10544                 if (scsi_2btoul(feature->feature_code) == starting)
10545                         feature = (struct scsi_get_config_feature *)
10546                             &feature->feature_data[feature->add_length];
10547                 data_len = (uint8_t *)feature - (uint8_t *)hdr;
10548         }
10549         scsi_ulto4b(data_len - 4, hdr->data_length);
10550         if (data_len < alloc_len) {
10551                 ctsio->residual = alloc_len - data_len;
10552                 ctsio->kern_data_len = data_len;
10553                 ctsio->kern_total_len = data_len;
10554         } else {
10555                 ctsio->residual = 0;
10556                 ctsio->kern_data_len = alloc_len;
10557                 ctsio->kern_total_len = alloc_len;
10558         }
10559
10560         ctl_set_success(ctsio);
10561         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10562         ctsio->be_move_done = ctl_config_move_done;
10563         ctl_datamove((union ctl_io *)ctsio);
10564         return (CTL_RETVAL_COMPLETE);
10565 }
10566
10567 int
10568 ctl_get_event_status(struct ctl_scsiio *ctsio)
10569 {
10570         struct scsi_get_event_status_header *hdr;
10571         struct scsi_get_event_status *cdb;
10572         uint32_t alloc_len, data_len;
10573         int notif_class;
10574
10575         cdb = (struct scsi_get_event_status *)ctsio->cdb;
10576         if ((cdb->byte2 & SGESN_POLLED) == 0) {
10577                 ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 1,
10578                     /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 0);
10579                 ctl_done((union ctl_io *)ctsio);
10580                 return (CTL_RETVAL_COMPLETE);
10581         }
10582         notif_class = cdb->notif_class;
10583         alloc_len = scsi_2btoul(cdb->length);
10584
10585         data_len = sizeof(struct scsi_get_event_status_header);
10586         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10587         ctsio->kern_sg_entries = 0;
10588         ctsio->kern_data_resid = 0;
10589         ctsio->kern_rel_offset = 0;
10590
10591         if (data_len < alloc_len) {
10592                 ctsio->residual = alloc_len - data_len;
10593                 ctsio->kern_data_len = data_len;
10594                 ctsio->kern_total_len = data_len;
10595         } else {
10596                 ctsio->residual = 0;
10597                 ctsio->kern_data_len = alloc_len;
10598                 ctsio->kern_total_len = alloc_len;
10599         }
10600
10601         hdr = (struct scsi_get_event_status_header *)ctsio->kern_data_ptr;
10602         scsi_ulto2b(0, hdr->descr_length);
10603         hdr->nea_class = SGESN_NEA;
10604         hdr->supported_class = 0;
10605
10606         ctl_set_success(ctsio);
10607         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10608         ctsio->be_move_done = ctl_config_move_done;
10609         ctl_datamove((union ctl_io *)ctsio);
10610         return (CTL_RETVAL_COMPLETE);
10611 }
10612
10613 int
10614 ctl_mechanism_status(struct ctl_scsiio *ctsio)
10615 {
10616         struct scsi_mechanism_status_header *hdr;
10617         struct scsi_mechanism_status *cdb;
10618         uint32_t alloc_len, data_len;
10619
10620         cdb = (struct scsi_mechanism_status *)ctsio->cdb;
10621         alloc_len = scsi_2btoul(cdb->length);
10622
10623         data_len = sizeof(struct scsi_mechanism_status_header);
10624         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10625         ctsio->kern_sg_entries = 0;
10626         ctsio->kern_data_resid = 0;
10627         ctsio->kern_rel_offset = 0;
10628
10629         if (data_len < alloc_len) {
10630                 ctsio->residual = alloc_len - data_len;
10631                 ctsio->kern_data_len = data_len;
10632                 ctsio->kern_total_len = data_len;
10633         } else {
10634                 ctsio->residual = 0;
10635                 ctsio->kern_data_len = alloc_len;
10636                 ctsio->kern_total_len = alloc_len;
10637         }
10638
10639         hdr = (struct scsi_mechanism_status_header *)ctsio->kern_data_ptr;
10640         hdr->state1 = 0x00;
10641         hdr->state2 = 0xe0;
10642         scsi_ulto3b(0, hdr->lba);
10643         hdr->slots_num = 0;
10644         scsi_ulto2b(0, hdr->slots_length);
10645
10646         ctl_set_success(ctsio);
10647         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10648         ctsio->be_move_done = ctl_config_move_done;
10649         ctl_datamove((union ctl_io *)ctsio);
10650         return (CTL_RETVAL_COMPLETE);
10651 }
10652
10653 static void
10654 ctl_ultomsf(uint32_t lba, uint8_t *buf)
10655 {
10656
10657         lba += 150;
10658         buf[0] = 0;
10659         buf[1] = bin2bcd((lba / 75) / 60);
10660         buf[2] = bin2bcd((lba / 75) % 60);
10661         buf[3] = bin2bcd(lba % 75);
10662 }
10663
10664 int
10665 ctl_read_toc(struct ctl_scsiio *ctsio)
10666 {
10667         struct ctl_lun *lun = CTL_LUN(ctsio);
10668         struct scsi_read_toc_hdr *hdr;
10669         struct scsi_read_toc_type01_descr *descr;
10670         struct scsi_read_toc *cdb;
10671         uint32_t alloc_len, data_len;
10672         int format, msf;
10673
10674         cdb = (struct scsi_read_toc *)ctsio->cdb;
10675         msf = (cdb->byte2 & CD_MSF) != 0;
10676         format = cdb->format;
10677         alloc_len = scsi_2btoul(cdb->data_len);
10678
10679         data_len = sizeof(struct scsi_read_toc_hdr);
10680         if (format == 0)
10681                 data_len += 2 * sizeof(struct scsi_read_toc_type01_descr);
10682         else
10683                 data_len += sizeof(struct scsi_read_toc_type01_descr);
10684         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10685         ctsio->kern_sg_entries = 0;
10686         ctsio->kern_data_resid = 0;
10687         ctsio->kern_rel_offset = 0;
10688
10689         if (data_len < alloc_len) {
10690                 ctsio->residual = alloc_len - data_len;
10691                 ctsio->kern_data_len = data_len;
10692                 ctsio->kern_total_len = data_len;
10693         } else {
10694                 ctsio->residual = 0;
10695                 ctsio->kern_data_len = alloc_len;
10696                 ctsio->kern_total_len = alloc_len;
10697         }
10698
10699         hdr = (struct scsi_read_toc_hdr *)ctsio->kern_data_ptr;
10700         if (format == 0) {
10701                 scsi_ulto2b(0x12, hdr->data_length);
10702                 hdr->first = 1;
10703                 hdr->last = 1;
10704                 descr = (struct scsi_read_toc_type01_descr *)(hdr + 1);
10705                 descr->addr_ctl = 0x14;
10706                 descr->track_number = 1;
10707                 if (msf)
10708                         ctl_ultomsf(0, descr->track_start);
10709                 else
10710                         scsi_ulto4b(0, descr->track_start);
10711                 descr++;
10712                 descr->addr_ctl = 0x14;
10713                 descr->track_number = 0xaa;
10714                 if (msf)
10715                         ctl_ultomsf(lun->be_lun->maxlba+1, descr->track_start);
10716                 else
10717                         scsi_ulto4b(lun->be_lun->maxlba+1, descr->track_start);
10718         } else {
10719                 scsi_ulto2b(0x0a, hdr->data_length);
10720                 hdr->first = 1;
10721                 hdr->last = 1;
10722                 descr = (struct scsi_read_toc_type01_descr *)(hdr + 1);
10723                 descr->addr_ctl = 0x14;
10724                 descr->track_number = 1;
10725                 if (msf)
10726                         ctl_ultomsf(0, descr->track_start);
10727                 else
10728                         scsi_ulto4b(0, descr->track_start);
10729         }
10730
10731         ctl_set_success(ctsio);
10732         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10733         ctsio->be_move_done = ctl_config_move_done;
10734         ctl_datamove((union ctl_io *)ctsio);
10735         return (CTL_RETVAL_COMPLETE);
10736 }
10737
10738 /*
10739  * For known CDB types, parse the LBA and length.
10740  */
10741 static int
10742 ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len)
10743 {
10744         if (io->io_hdr.io_type != CTL_IO_SCSI)
10745                 return (1);
10746
10747         switch (io->scsiio.cdb[0]) {
10748         case COMPARE_AND_WRITE: {
10749                 struct scsi_compare_and_write *cdb;
10750
10751                 cdb = (struct scsi_compare_and_write *)io->scsiio.cdb;
10752
10753                 *lba = scsi_8btou64(cdb->addr);
10754                 *len = cdb->length;
10755                 break;
10756         }
10757         case READ_6:
10758         case WRITE_6: {
10759                 struct scsi_rw_6 *cdb;
10760
10761                 cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
10762
10763                 *lba = scsi_3btoul(cdb->addr);
10764                 /* only 5 bits are valid in the most significant address byte */
10765                 *lba &= 0x1fffff;
10766                 *len = cdb->length;
10767                 break;
10768         }
10769         case READ_10:
10770         case WRITE_10: {
10771                 struct scsi_rw_10 *cdb;
10772
10773                 cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
10774
10775                 *lba = scsi_4btoul(cdb->addr);
10776                 *len = scsi_2btoul(cdb->length);
10777                 break;
10778         }
10779         case WRITE_VERIFY_10: {
10780                 struct scsi_write_verify_10 *cdb;
10781
10782                 cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
10783
10784                 *lba = scsi_4btoul(cdb->addr);
10785                 *len = scsi_2btoul(cdb->length);
10786                 break;
10787         }
10788         case READ_12:
10789         case WRITE_12: {
10790                 struct scsi_rw_12 *cdb;
10791
10792                 cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
10793
10794                 *lba = scsi_4btoul(cdb->addr);
10795                 *len = scsi_4btoul(cdb->length);
10796                 break;
10797         }
10798         case WRITE_VERIFY_12: {
10799                 struct scsi_write_verify_12 *cdb;
10800
10801                 cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
10802
10803                 *lba = scsi_4btoul(cdb->addr);
10804                 *len = scsi_4btoul(cdb->length);
10805                 break;
10806         }
10807         case READ_16:
10808         case WRITE_16: {
10809                 struct scsi_rw_16 *cdb;
10810
10811                 cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
10812
10813                 *lba = scsi_8btou64(cdb->addr);
10814                 *len = scsi_4btoul(cdb->length);
10815                 break;
10816         }
10817         case WRITE_ATOMIC_16: {
10818                 struct scsi_write_atomic_16 *cdb;
10819
10820                 cdb = (struct scsi_write_atomic_16 *)io->scsiio.cdb;
10821
10822                 *lba = scsi_8btou64(cdb->addr);
10823                 *len = scsi_2btoul(cdb->length);
10824                 break;
10825         }
10826         case WRITE_VERIFY_16: {
10827                 struct scsi_write_verify_16 *cdb;
10828
10829                 cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
10830
10831                 *lba = scsi_8btou64(cdb->addr);
10832                 *len = scsi_4btoul(cdb->length);
10833                 break;
10834         }
10835         case WRITE_SAME_10: {
10836                 struct scsi_write_same_10 *cdb;
10837
10838                 cdb = (struct scsi_write_same_10 *)io->scsiio.cdb;
10839
10840                 *lba = scsi_4btoul(cdb->addr);
10841                 *len = scsi_2btoul(cdb->length);
10842                 break;
10843         }
10844         case WRITE_SAME_16: {
10845                 struct scsi_write_same_16 *cdb;
10846
10847                 cdb = (struct scsi_write_same_16 *)io->scsiio.cdb;
10848
10849                 *lba = scsi_8btou64(cdb->addr);
10850                 *len = scsi_4btoul(cdb->length);
10851                 break;
10852         }
10853         case VERIFY_10: {
10854                 struct scsi_verify_10 *cdb;
10855
10856                 cdb = (struct scsi_verify_10 *)io->scsiio.cdb;
10857
10858                 *lba = scsi_4btoul(cdb->addr);
10859                 *len = scsi_2btoul(cdb->length);
10860                 break;
10861         }
10862         case VERIFY_12: {
10863                 struct scsi_verify_12 *cdb;
10864
10865                 cdb = (struct scsi_verify_12 *)io->scsiio.cdb;
10866
10867                 *lba = scsi_4btoul(cdb->addr);
10868                 *len = scsi_4btoul(cdb->length);
10869                 break;
10870         }
10871         case VERIFY_16: {
10872                 struct scsi_verify_16 *cdb;
10873
10874                 cdb = (struct scsi_verify_16 *)io->scsiio.cdb;
10875
10876                 *lba = scsi_8btou64(cdb->addr);
10877                 *len = scsi_4btoul(cdb->length);
10878                 break;
10879         }
10880         case UNMAP: {
10881                 *lba = 0;
10882                 *len = UINT64_MAX;
10883                 break;
10884         }
10885         case SERVICE_ACTION_IN: {       /* GET LBA STATUS */
10886                 struct scsi_get_lba_status *cdb;
10887
10888                 cdb = (struct scsi_get_lba_status *)io->scsiio.cdb;
10889                 *lba = scsi_8btou64(cdb->addr);
10890                 *len = UINT32_MAX;
10891                 break;
10892         }
10893         default:
10894                 return (1);
10895                 break; /* NOTREACHED */
10896         }
10897
10898         return (0);
10899 }
10900
10901 static ctl_action
10902 ctl_extent_check_lba(uint64_t lba1, uint64_t len1, uint64_t lba2, uint64_t len2,
10903     bool seq)
10904 {
10905         uint64_t endlba1, endlba2;
10906
10907         endlba1 = lba1 + len1 - (seq ? 0 : 1);
10908         endlba2 = lba2 + len2 - 1;
10909
10910         if ((endlba1 < lba2) || (endlba2 < lba1))
10911                 return (CTL_ACTION_PASS);
10912         else
10913                 return (CTL_ACTION_BLOCK);
10914 }
10915
10916 static int
10917 ctl_extent_check_unmap(union ctl_io *io, uint64_t lba2, uint64_t len2)
10918 {
10919         struct ctl_ptr_len_flags *ptrlen;
10920         struct scsi_unmap_desc *buf, *end, *range;
10921         uint64_t lba;
10922         uint32_t len;
10923
10924         /* If not UNMAP -- go other way. */
10925         if (io->io_hdr.io_type != CTL_IO_SCSI ||
10926             io->scsiio.cdb[0] != UNMAP)
10927                 return (CTL_ACTION_ERROR);
10928
10929         /* If UNMAP without data -- block and wait for data. */
10930         ptrlen = (struct ctl_ptr_len_flags *)
10931             &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
10932         if ((io->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0 ||
10933             ptrlen->ptr == NULL)
10934                 return (CTL_ACTION_BLOCK);
10935
10936         /* UNMAP with data -- check for collision. */
10937         buf = (struct scsi_unmap_desc *)ptrlen->ptr;
10938         end = buf + ptrlen->len / sizeof(*buf);
10939         for (range = buf; range < end; range++) {
10940                 lba = scsi_8btou64(range->lba);
10941                 len = scsi_4btoul(range->length);
10942                 if ((lba < lba2 + len2) && (lba + len > lba2))
10943                         return (CTL_ACTION_BLOCK);
10944         }
10945         return (CTL_ACTION_PASS);
10946 }
10947
10948 static ctl_action
10949 ctl_extent_check(union ctl_io *io1, union ctl_io *io2, bool seq)
10950 {
10951         uint64_t lba1, lba2;
10952         uint64_t len1, len2;
10953         int retval;
10954
10955         if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
10956                 return (CTL_ACTION_ERROR);
10957
10958         retval = ctl_extent_check_unmap(io1, lba2, len2);
10959         if (retval != CTL_ACTION_ERROR)
10960                 return (retval);
10961
10962         if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
10963                 return (CTL_ACTION_ERROR);
10964
10965         if (io1->io_hdr.flags & CTL_FLAG_SERSEQ_DONE)
10966                 seq = FALSE;
10967         return (ctl_extent_check_lba(lba1, len1, lba2, len2, seq));
10968 }
10969
10970 static ctl_action
10971 ctl_extent_check_seq(union ctl_io *io1, union ctl_io *io2)
10972 {
10973         uint64_t lba1, lba2;
10974         uint64_t len1, len2;
10975
10976         if (io1->io_hdr.flags & CTL_FLAG_SERSEQ_DONE)
10977                 return (CTL_ACTION_PASS);
10978         if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
10979                 return (CTL_ACTION_ERROR);
10980         if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
10981                 return (CTL_ACTION_ERROR);
10982
10983         if (lba1 + len1 == lba2)
10984                 return (CTL_ACTION_BLOCK);
10985         return (CTL_ACTION_PASS);
10986 }
10987
10988 static ctl_action
10989 ctl_check_for_blockage(struct ctl_lun *lun, union ctl_io *pending_io,
10990     union ctl_io *ooa_io)
10991 {
10992         const struct ctl_cmd_entry *pending_entry, *ooa_entry;
10993         const ctl_serialize_action *serialize_row;
10994
10995         /*
10996          * The initiator attempted multiple untagged commands at the same
10997          * time.  Can't do that.
10998          */
10999         if ((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
11000          && (ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
11001          && ((pending_io->io_hdr.nexus.targ_port ==
11002               ooa_io->io_hdr.nexus.targ_port)
11003           && (pending_io->io_hdr.nexus.initid ==
11004               ooa_io->io_hdr.nexus.initid))
11005          && ((ooa_io->io_hdr.flags & (CTL_FLAG_ABORT |
11006               CTL_FLAG_STATUS_SENT)) == 0))
11007                 return (CTL_ACTION_OVERLAP);
11008
11009         /*
11010          * The initiator attempted to send multiple tagged commands with
11011          * the same ID.  (It's fine if different initiators have the same
11012          * tag ID.)
11013          *
11014          * Even if all of those conditions are true, we don't kill the I/O
11015          * if the command ahead of us has been aborted.  We won't end up
11016          * sending it to the FETD, and it's perfectly legal to resend a
11017          * command with the same tag number as long as the previous
11018          * instance of this tag number has been aborted somehow.
11019          */
11020         if ((pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
11021          && (ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
11022          && (pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
11023          && ((pending_io->io_hdr.nexus.targ_port ==
11024               ooa_io->io_hdr.nexus.targ_port)
11025           && (pending_io->io_hdr.nexus.initid ==
11026               ooa_io->io_hdr.nexus.initid))
11027          && ((ooa_io->io_hdr.flags & (CTL_FLAG_ABORT |
11028               CTL_FLAG_STATUS_SENT)) == 0))
11029                 return (CTL_ACTION_OVERLAP_TAG);
11030
11031         /*
11032          * If we get a head of queue tag, SAM-3 says that we should
11033          * immediately execute it.
11034          *
11035          * What happens if this command would normally block for some other
11036          * reason?  e.g. a request sense with a head of queue tag
11037          * immediately after a write.  Normally that would block, but this
11038          * will result in its getting executed immediately...
11039          *
11040          * We currently return "pass" instead of "skip", so we'll end up
11041          * going through the rest of the queue to check for overlapped tags.
11042          *
11043          * XXX KDM check for other types of blockage first??
11044          */
11045         if (pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
11046                 return (CTL_ACTION_PASS);
11047
11048         /*
11049          * Ordered tags have to block until all items ahead of them
11050          * have completed.  If we get called with an ordered tag, we always
11051          * block, if something else is ahead of us in the queue.
11052          */
11053         if (pending_io->scsiio.tag_type == CTL_TAG_ORDERED)
11054                 return (CTL_ACTION_BLOCK);
11055
11056         /*
11057          * Simple tags get blocked until all head of queue and ordered tags
11058          * ahead of them have completed.  I'm lumping untagged commands in
11059          * with simple tags here.  XXX KDM is that the right thing to do?
11060          */
11061         if (((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
11062           || (pending_io->scsiio.tag_type == CTL_TAG_SIMPLE))
11063          && ((ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
11064           || (ooa_io->scsiio.tag_type == CTL_TAG_ORDERED)))
11065                 return (CTL_ACTION_BLOCK);
11066
11067         pending_entry = ctl_get_cmd_entry(&pending_io->scsiio, NULL);
11068         KASSERT(pending_entry->seridx < CTL_SERIDX_COUNT,
11069             ("%s: Invalid seridx %d for pending CDB %02x %02x @ %p",
11070              __func__, pending_entry->seridx, pending_io->scsiio.cdb[0],
11071              pending_io->scsiio.cdb[1], pending_io));
11072         ooa_entry = ctl_get_cmd_entry(&ooa_io->scsiio, NULL);
11073         if (ooa_entry->seridx == CTL_SERIDX_INVLD)
11074                 return (CTL_ACTION_PASS); /* Unsupported command in OOA queue */
11075         KASSERT(ooa_entry->seridx < CTL_SERIDX_COUNT,
11076             ("%s: Invalid seridx %d for ooa CDB %02x %02x @ %p",
11077              __func__, ooa_entry->seridx, ooa_io->scsiio.cdb[0],
11078              ooa_io->scsiio.cdb[1], ooa_io));
11079
11080         serialize_row = ctl_serialize_table[ooa_entry->seridx];
11081
11082         switch (serialize_row[pending_entry->seridx]) {
11083         case CTL_SER_BLOCK:
11084                 return (CTL_ACTION_BLOCK);
11085         case CTL_SER_EXTENT:
11086                 return (ctl_extent_check(ooa_io, pending_io,
11087                     (lun->be_lun && lun->be_lun->serseq == CTL_LUN_SERSEQ_ON)));
11088         case CTL_SER_EXTENTOPT:
11089                 if ((lun->MODE_CTRL.queue_flags & SCP_QUEUE_ALG_MASK) !=
11090                     SCP_QUEUE_ALG_UNRESTRICTED)
11091                         return (ctl_extent_check(ooa_io, pending_io,
11092                             (lun->be_lun &&
11093                              lun->be_lun->serseq == CTL_LUN_SERSEQ_ON)));
11094                 return (CTL_ACTION_PASS);
11095         case CTL_SER_EXTENTSEQ:
11096                 if (lun->be_lun && lun->be_lun->serseq != CTL_LUN_SERSEQ_OFF)
11097                         return (ctl_extent_check_seq(ooa_io, pending_io));
11098                 return (CTL_ACTION_PASS);
11099         case CTL_SER_PASS:
11100                 return (CTL_ACTION_PASS);
11101         case CTL_SER_BLOCKOPT:
11102                 if ((lun->MODE_CTRL.queue_flags & SCP_QUEUE_ALG_MASK) !=
11103                     SCP_QUEUE_ALG_UNRESTRICTED)
11104                         return (CTL_ACTION_BLOCK);
11105                 return (CTL_ACTION_PASS);
11106         case CTL_SER_SKIP:
11107                 return (CTL_ACTION_SKIP);
11108         default:
11109                 panic("%s: Invalid serialization value %d for %d => %d",
11110                     __func__, serialize_row[pending_entry->seridx],
11111                     pending_entry->seridx, ooa_entry->seridx);
11112         }
11113
11114         return (CTL_ACTION_ERROR);
11115 }
11116
11117 /*
11118  * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
11119  * Assumptions:
11120  * - pending_io is generally either incoming, or on the blocked queue
11121  * - starting I/O is the I/O we want to start the check with.
11122  */
11123 static ctl_action
11124 ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
11125               union ctl_io *starting_io)
11126 {
11127         union ctl_io *ooa_io;
11128         ctl_action action;
11129
11130         mtx_assert(&lun->lun_lock, MA_OWNED);
11131
11132         /*
11133          * Run back along the OOA queue, starting with the current
11134          * blocked I/O and going through every I/O before it on the
11135          * queue.  If starting_io is NULL, we'll just end up returning
11136          * CTL_ACTION_PASS.
11137          */
11138         for (ooa_io = starting_io; ooa_io != NULL;
11139              ooa_io = (union ctl_io *)TAILQ_PREV(&ooa_io->io_hdr, ctl_ooaq,
11140              ooa_links)){
11141
11142                 /*
11143                  * This routine just checks to see whether
11144                  * cur_blocked is blocked by ooa_io, which is ahead
11145                  * of it in the queue.  It doesn't queue/dequeue
11146                  * cur_blocked.
11147                  */
11148                 action = ctl_check_for_blockage(lun, pending_io, ooa_io);
11149                 switch (action) {
11150                 case CTL_ACTION_BLOCK:
11151                 case CTL_ACTION_OVERLAP:
11152                 case CTL_ACTION_OVERLAP_TAG:
11153                 case CTL_ACTION_SKIP:
11154                 case CTL_ACTION_ERROR:
11155                         return (action);
11156                         break; /* NOTREACHED */
11157                 case CTL_ACTION_PASS:
11158                         break;
11159                 default:
11160                         panic("%s: Invalid action %d\n", __func__, action);
11161                 }
11162         }
11163
11164         return (CTL_ACTION_PASS);
11165 }
11166
11167 /*
11168  * Assumptions:
11169  * - An I/O has just completed, and has been removed from the per-LUN OOA
11170  *   queue, so some items on the blocked queue may now be unblocked.
11171  */
11172 static int
11173 ctl_check_blocked(struct ctl_lun *lun)
11174 {
11175         struct ctl_softc *softc = lun->ctl_softc;
11176         union ctl_io *cur_blocked, *next_blocked;
11177
11178         mtx_assert(&lun->lun_lock, MA_OWNED);
11179
11180         /*
11181          * Run forward from the head of the blocked queue, checking each
11182          * entry against the I/Os prior to it on the OOA queue to see if
11183          * there is still any blockage.
11184          *
11185          * We cannot use the TAILQ_FOREACH() macro, because it can't deal
11186          * with our removing a variable on it while it is traversing the
11187          * list.
11188          */
11189         for (cur_blocked = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue);
11190              cur_blocked != NULL; cur_blocked = next_blocked) {
11191                 union ctl_io *prev_ooa;
11192                 ctl_action action;
11193
11194                 next_blocked = (union ctl_io *)TAILQ_NEXT(&cur_blocked->io_hdr,
11195                                                           blocked_links);
11196
11197                 prev_ooa = (union ctl_io *)TAILQ_PREV(&cur_blocked->io_hdr,
11198                                                       ctl_ooaq, ooa_links);
11199
11200                 /*
11201                  * If cur_blocked happens to be the first item in the OOA
11202                  * queue now, prev_ooa will be NULL, and the action
11203                  * returned will just be CTL_ACTION_PASS.
11204                  */
11205                 action = ctl_check_ooa(lun, cur_blocked, prev_ooa);
11206
11207                 switch (action) {
11208                 case CTL_ACTION_BLOCK:
11209                         /* Nothing to do here, still blocked */
11210                         break;
11211                 case CTL_ACTION_OVERLAP:
11212                 case CTL_ACTION_OVERLAP_TAG:
11213                         /*
11214                          * This shouldn't happen!  In theory we've already
11215                          * checked this command for overlap...
11216                          */
11217                         break;
11218                 case CTL_ACTION_PASS:
11219                 case CTL_ACTION_SKIP: {
11220                         const struct ctl_cmd_entry *entry;
11221
11222                         /*
11223                          * The skip case shouldn't happen, this transaction
11224                          * should have never made it onto the blocked queue.
11225                          */
11226                         /*
11227                          * This I/O is no longer blocked, we can remove it
11228                          * from the blocked queue.  Since this is a TAILQ
11229                          * (doubly linked list), we can do O(1) removals
11230                          * from any place on the list.
11231                          */
11232                         TAILQ_REMOVE(&lun->blocked_queue, &cur_blocked->io_hdr,
11233                                      blocked_links);
11234                         cur_blocked->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
11235
11236                         if ((softc->ha_mode != CTL_HA_MODE_XFER) &&
11237                             (cur_blocked->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)){
11238                                 /*
11239                                  * Need to send IO back to original side to
11240                                  * run
11241                                  */
11242                                 union ctl_ha_msg msg_info;
11243
11244                                 cur_blocked->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11245                                 msg_info.hdr.original_sc =
11246                                         cur_blocked->io_hdr.original_sc;
11247                                 msg_info.hdr.serializing_sc = cur_blocked;
11248                                 msg_info.hdr.msg_type = CTL_MSG_R2R;
11249                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11250                                     sizeof(msg_info.hdr), M_NOWAIT);
11251                                 break;
11252                         }
11253                         entry = ctl_get_cmd_entry(&cur_blocked->scsiio, NULL);
11254
11255                         /*
11256                          * Check this I/O for LUN state changes that may
11257                          * have happened while this command was blocked.
11258                          * The LUN state may have been changed by a command
11259                          * ahead of us in the queue, so we need to re-check
11260                          * for any states that can be caused by SCSI
11261                          * commands.
11262                          */
11263                         if (ctl_scsiio_lun_check(lun, entry,
11264                                                  &cur_blocked->scsiio) == 0) {
11265                                 cur_blocked->io_hdr.flags |=
11266                                                       CTL_FLAG_IS_WAS_ON_RTR;
11267                                 ctl_enqueue_rtr(cur_blocked);
11268                         } else
11269                                 ctl_done(cur_blocked);
11270                         break;
11271                 }
11272                 default:
11273                         /*
11274                          * This probably shouldn't happen -- we shouldn't
11275                          * get CTL_ACTION_ERROR, or anything else.
11276                          */
11277                         break;
11278                 }
11279         }
11280
11281         return (CTL_RETVAL_COMPLETE);
11282 }
11283
11284 /*
11285  * This routine (with one exception) checks LUN flags that can be set by
11286  * commands ahead of us in the OOA queue.  These flags have to be checked
11287  * when a command initially comes in, and when we pull a command off the
11288  * blocked queue and are preparing to execute it.  The reason we have to
11289  * check these flags for commands on the blocked queue is that the LUN
11290  * state may have been changed by a command ahead of us while we're on the
11291  * blocked queue.
11292  *
11293  * Ordering is somewhat important with these checks, so please pay
11294  * careful attention to the placement of any new checks.
11295  */
11296 static int
11297 ctl_scsiio_lun_check(struct ctl_lun *lun,
11298     const struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
11299 {
11300         struct ctl_softc *softc = lun->ctl_softc;
11301         int retval;
11302         uint32_t residx;
11303
11304         retval = 0;
11305
11306         mtx_assert(&lun->lun_lock, MA_OWNED);
11307
11308         /*
11309          * If this shelf is a secondary shelf controller, we may have to
11310          * reject some commands disallowed by HA mode and link state.
11311          */
11312         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
11313                 if (softc->ha_link == CTL_HA_LINK_OFFLINE &&
11314                     (entry->flags & CTL_CMD_FLAG_OK_ON_UNAVAIL) == 0) {
11315                         ctl_set_lun_unavail(ctsio);
11316                         retval = 1;
11317                         goto bailout;
11318                 }
11319                 if ((lun->flags & CTL_LUN_PEER_SC_PRIMARY) == 0 &&
11320                     (entry->flags & CTL_CMD_FLAG_OK_ON_UNAVAIL) == 0) {
11321                         ctl_set_lun_transit(ctsio);
11322                         retval = 1;
11323                         goto bailout;
11324                 }
11325                 if (softc->ha_mode == CTL_HA_MODE_ACT_STBY &&
11326                     (entry->flags & CTL_CMD_FLAG_OK_ON_STANDBY) == 0) {
11327                         ctl_set_lun_standby(ctsio);
11328                         retval = 1;
11329                         goto bailout;
11330                 }
11331
11332                 /* The rest of checks are only done on executing side */
11333                 if (softc->ha_mode == CTL_HA_MODE_XFER)
11334                         goto bailout;
11335         }
11336
11337         if (entry->pattern & CTL_LUN_PAT_WRITE) {
11338                 if (lun->be_lun &&
11339                     lun->be_lun->flags & CTL_LUN_FLAG_READONLY) {
11340                         ctl_set_hw_write_protected(ctsio);
11341                         retval = 1;
11342                         goto bailout;
11343                 }
11344                 if ((lun->MODE_CTRL.eca_and_aen & SCP_SWP) != 0) {
11345                         ctl_set_sense(ctsio, /*current_error*/ 1,
11346                             /*sense_key*/ SSD_KEY_DATA_PROTECT,
11347                             /*asc*/ 0x27, /*ascq*/ 0x02, SSD_ELEM_NONE);
11348                         retval = 1;
11349                         goto bailout;
11350                 }
11351         }
11352
11353         /*
11354          * Check for a reservation conflict.  If this command isn't allowed
11355          * even on reserved LUNs, and if this initiator isn't the one who
11356          * reserved us, reject the command with a reservation conflict.
11357          */
11358         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11359         if ((lun->flags & CTL_LUN_RESERVED)
11360          && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
11361                 if (lun->res_idx != residx) {
11362                         ctl_set_reservation_conflict(ctsio);
11363                         retval = 1;
11364                         goto bailout;
11365                 }
11366         }
11367
11368         if ((lun->flags & CTL_LUN_PR_RESERVED) == 0 ||
11369             (entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV)) {
11370                 /* No reservation or command is allowed. */;
11371         } else if ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_WRESV) &&
11372             (lun->pr_res_type == SPR_TYPE_WR_EX ||
11373              lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
11374              lun->pr_res_type == SPR_TYPE_WR_EX_AR)) {
11375                 /* The command is allowed for Write Exclusive resv. */;
11376         } else {
11377                 /*
11378                  * if we aren't registered or it's a res holder type
11379                  * reservation and this isn't the res holder then set a
11380                  * conflict.
11381                  */
11382                 if (ctl_get_prkey(lun, residx) == 0 ||
11383                     (residx != lun->pr_res_idx && lun->pr_res_type < 4)) {
11384                         ctl_set_reservation_conflict(ctsio);
11385                         retval = 1;
11386                         goto bailout;
11387                 }
11388         }
11389
11390         if ((entry->flags & CTL_CMD_FLAG_OK_ON_NO_MEDIA) == 0) {
11391                 if (lun->flags & CTL_LUN_EJECTED)
11392                         ctl_set_lun_ejected(ctsio);
11393                 else if (lun->flags & CTL_LUN_NO_MEDIA) {
11394                         if (lun->flags & CTL_LUN_REMOVABLE)
11395                                 ctl_set_lun_no_media(ctsio);
11396                         else
11397                                 ctl_set_lun_int_reqd(ctsio);
11398                 } else if (lun->flags & CTL_LUN_STOPPED)
11399                         ctl_set_lun_stopped(ctsio);
11400                 else
11401                         goto bailout;
11402                 retval = 1;
11403                 goto bailout;
11404         }
11405
11406 bailout:
11407         return (retval);
11408 }
11409
11410 static void
11411 ctl_failover_io(union ctl_io *io, int have_lock)
11412 {
11413         ctl_set_busy(&io->scsiio);
11414         ctl_done(io);
11415 }
11416
11417 static void
11418 ctl_failover_lun(union ctl_io *rio)
11419 {
11420         struct ctl_softc *softc = CTL_SOFTC(rio);
11421         struct ctl_lun *lun;
11422         struct ctl_io_hdr *io, *next_io;
11423         uint32_t targ_lun;
11424
11425         targ_lun = rio->io_hdr.nexus.targ_mapped_lun;
11426         CTL_DEBUG_PRINT(("FAILOVER for lun %ju\n", targ_lun));
11427
11428         /* Find and lock the LUN. */
11429         mtx_lock(&softc->ctl_lock);
11430         if (targ_lun > CTL_MAX_LUNS ||
11431             (lun = softc->ctl_luns[targ_lun]) == NULL) {
11432                 mtx_unlock(&softc->ctl_lock);
11433                 return;
11434         }
11435         mtx_lock(&lun->lun_lock);
11436         mtx_unlock(&softc->ctl_lock);
11437         if (lun->flags & CTL_LUN_DISABLED) {
11438                 mtx_unlock(&lun->lun_lock);
11439                 return;
11440         }
11441
11442         if (softc->ha_mode == CTL_HA_MODE_XFER) {
11443                 TAILQ_FOREACH_SAFE(io, &lun->ooa_queue, ooa_links, next_io) {
11444                         /* We are master */
11445                         if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
11446                                 if (io->flags & CTL_FLAG_IO_ACTIVE) {
11447                                         io->flags |= CTL_FLAG_ABORT;
11448                                         io->flags |= CTL_FLAG_FAILOVER;
11449                                 } else { /* This can be only due to DATAMOVE */
11450                                         io->msg_type = CTL_MSG_DATAMOVE_DONE;
11451                                         io->flags &= ~CTL_FLAG_DMA_INPROG;
11452                                         io->flags |= CTL_FLAG_IO_ACTIVE;
11453                                         io->port_status = 31340;
11454                                         ctl_enqueue_isc((union ctl_io *)io);
11455                                 }
11456                         }
11457                         /* We are slave */
11458                         if (io->flags & CTL_FLAG_SENT_2OTHER_SC) {
11459                                 io->flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11460                                 if (io->flags & CTL_FLAG_IO_ACTIVE) {
11461                                         io->flags |= CTL_FLAG_FAILOVER;
11462                                 } else {
11463                                         ctl_set_busy(&((union ctl_io *)io)->
11464                                             scsiio);
11465                                         ctl_done((union ctl_io *)io);
11466                                 }
11467                         }
11468                 }
11469         } else { /* SERIALIZE modes */
11470                 TAILQ_FOREACH_SAFE(io, &lun->blocked_queue, blocked_links,
11471                     next_io) {
11472                         /* We are master */
11473                         if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
11474                                 TAILQ_REMOVE(&lun->blocked_queue, io,
11475                                     blocked_links);
11476                                 io->flags &= ~CTL_FLAG_BLOCKED;
11477                                 TAILQ_REMOVE(&lun->ooa_queue, io, ooa_links);
11478                                 ctl_free_io((union ctl_io *)io);
11479                         }
11480                 }
11481                 TAILQ_FOREACH_SAFE(io, &lun->ooa_queue, ooa_links, next_io) {
11482                         /* We are master */
11483                         if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
11484                                 TAILQ_REMOVE(&lun->ooa_queue, io, ooa_links);
11485                                 ctl_free_io((union ctl_io *)io);
11486                         }
11487                         /* We are slave */
11488                         if (io->flags & CTL_FLAG_SENT_2OTHER_SC) {
11489                                 io->flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11490                                 if (!(io->flags & CTL_FLAG_IO_ACTIVE)) {
11491                                         ctl_set_busy(&((union ctl_io *)io)->
11492                                             scsiio);
11493                                         ctl_done((union ctl_io *)io);
11494                                 }
11495                         }
11496                 }
11497                 ctl_check_blocked(lun);
11498         }
11499         mtx_unlock(&lun->lun_lock);
11500 }
11501
11502 static int
11503 ctl_scsiio_precheck(struct ctl_softc *softc, struct ctl_scsiio *ctsio)
11504 {
11505         struct ctl_lun *lun;
11506         const struct ctl_cmd_entry *entry;
11507         uint32_t initidx, targ_lun;
11508         int retval = 0;
11509
11510         lun = NULL;
11511         targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
11512         if (targ_lun < CTL_MAX_LUNS)
11513                 lun = softc->ctl_luns[targ_lun];
11514         if (lun) {
11515                 /*
11516                  * If the LUN is invalid, pretend that it doesn't exist.
11517                  * It will go away as soon as all pending I/O has been
11518                  * completed.
11519                  */
11520                 mtx_lock(&lun->lun_lock);
11521                 if (lun->flags & CTL_LUN_DISABLED) {
11522                         mtx_unlock(&lun->lun_lock);
11523                         lun = NULL;
11524                 }
11525         }
11526         CTL_LUN(ctsio) = lun;
11527         if (lun) {
11528                 CTL_BACKEND_LUN(ctsio) = lun->be_lun;
11529
11530                 /*
11531                  * Every I/O goes into the OOA queue for a particular LUN,
11532                  * and stays there until completion.
11533                  */
11534 #ifdef CTL_TIME_IO
11535                 if (TAILQ_EMPTY(&lun->ooa_queue))
11536                         lun->idle_time += getsbinuptime() - lun->last_busy;
11537 #endif
11538                 TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
11539         }
11540
11541         /* Get command entry and return error if it is unsuppotyed. */
11542         entry = ctl_validate_command(ctsio);
11543         if (entry == NULL) {
11544                 if (lun)
11545                         mtx_unlock(&lun->lun_lock);
11546                 return (retval);
11547         }
11548
11549         ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
11550         ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
11551
11552         /*
11553          * Check to see whether we can send this command to LUNs that don't
11554          * exist.  This should pretty much only be the case for inquiry
11555          * and request sense.  Further checks, below, really require having
11556          * a LUN, so we can't really check the command anymore.  Just put
11557          * it on the rtr queue.
11558          */
11559         if (lun == NULL) {
11560                 if (entry->flags & CTL_CMD_FLAG_OK_ON_NO_LUN) {
11561                         ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11562                         ctl_enqueue_rtr((union ctl_io *)ctsio);
11563                         return (retval);
11564                 }
11565
11566                 ctl_set_unsupported_lun(ctsio);
11567                 ctl_done((union ctl_io *)ctsio);
11568                 CTL_DEBUG_PRINT(("ctl_scsiio_precheck: bailing out due to invalid LUN\n"));
11569                 return (retval);
11570         } else {
11571                 /*
11572                  * Make sure we support this particular command on this LUN.
11573                  * e.g., we don't support writes to the control LUN.
11574                  */
11575                 if (!ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
11576                         mtx_unlock(&lun->lun_lock);
11577                         ctl_set_invalid_opcode(ctsio);
11578                         ctl_done((union ctl_io *)ctsio);
11579                         return (retval);
11580                 }
11581         }
11582
11583         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11584
11585 #ifdef CTL_WITH_CA
11586         /*
11587          * If we've got a request sense, it'll clear the contingent
11588          * allegiance condition.  Otherwise, if we have a CA condition for
11589          * this initiator, clear it, because it sent down a command other
11590          * than request sense.
11591          */
11592         if ((ctsio->cdb[0] != REQUEST_SENSE)
11593          && (ctl_is_set(lun->have_ca, initidx)))
11594                 ctl_clear_mask(lun->have_ca, initidx);
11595 #endif
11596
11597         /*
11598          * If the command has this flag set, it handles its own unit
11599          * attention reporting, we shouldn't do anything.  Otherwise we
11600          * check for any pending unit attentions, and send them back to the
11601          * initiator.  We only do this when a command initially comes in,
11602          * not when we pull it off the blocked queue.
11603          *
11604          * According to SAM-3, section 5.3.2, the order that things get
11605          * presented back to the host is basically unit attentions caused
11606          * by some sort of reset event, busy status, reservation conflicts
11607          * or task set full, and finally any other status.
11608          *
11609          * One issue here is that some of the unit attentions we report
11610          * don't fall into the "reset" category (e.g. "reported luns data
11611          * has changed").  So reporting it here, before the reservation
11612          * check, may be technically wrong.  I guess the only thing to do
11613          * would be to check for and report the reset events here, and then
11614          * check for the other unit attention types after we check for a
11615          * reservation conflict.
11616          *
11617          * XXX KDM need to fix this
11618          */
11619         if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
11620                 ctl_ua_type ua_type;
11621                 u_int sense_len = 0;
11622
11623                 ua_type = ctl_build_ua(lun, initidx, &ctsio->sense_data,
11624                     &sense_len, SSD_TYPE_NONE);
11625                 if (ua_type != CTL_UA_NONE) {
11626                         mtx_unlock(&lun->lun_lock);
11627                         ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
11628                         ctsio->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
11629                         ctsio->sense_len = sense_len;
11630                         ctl_done((union ctl_io *)ctsio);
11631                         return (retval);
11632                 }
11633         }
11634
11635
11636         if (ctl_scsiio_lun_check(lun, entry, ctsio) != 0) {
11637                 mtx_unlock(&lun->lun_lock);
11638                 ctl_done((union ctl_io *)ctsio);
11639                 return (retval);
11640         }
11641
11642         /*
11643          * XXX CHD this is where we want to send IO to other side if
11644          * this LUN is secondary on this SC. We will need to make a copy
11645          * of the IO and flag the IO on this side as SENT_2OTHER and the flag
11646          * the copy we send as FROM_OTHER.
11647          * We also need to stuff the address of the original IO so we can
11648          * find it easily. Something similar will need be done on the other
11649          * side so when we are done we can find the copy.
11650          */
11651         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
11652             (lun->flags & CTL_LUN_PEER_SC_PRIMARY) != 0 &&
11653             (entry->flags & CTL_CMD_FLAG_RUN_HERE) == 0) {
11654                 union ctl_ha_msg msg_info;
11655                 int isc_retval;
11656
11657                 ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11658                 ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11659                 mtx_unlock(&lun->lun_lock);
11660
11661                 msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
11662                 msg_info.hdr.original_sc = (union ctl_io *)ctsio;
11663                 msg_info.hdr.serializing_sc = NULL;
11664                 msg_info.hdr.nexus = ctsio->io_hdr.nexus;
11665                 msg_info.scsi.tag_num = ctsio->tag_num;
11666                 msg_info.scsi.tag_type = ctsio->tag_type;
11667                 msg_info.scsi.cdb_len = ctsio->cdb_len;
11668                 memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
11669
11670                 if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11671                     sizeof(msg_info.scsi) - sizeof(msg_info.scsi.sense_data),
11672                     M_WAITOK)) > CTL_HA_STATUS_SUCCESS) {
11673                         ctl_set_busy(ctsio);
11674                         ctl_done((union ctl_io *)ctsio);
11675                         return (retval);
11676                 }
11677                 return (retval);
11678         }
11679
11680         switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
11681                               (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr,
11682                               ctl_ooaq, ooa_links))) {
11683         case CTL_ACTION_BLOCK:
11684                 ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
11685                 TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
11686                                   blocked_links);
11687                 mtx_unlock(&lun->lun_lock);
11688                 return (retval);
11689         case CTL_ACTION_PASS:
11690         case CTL_ACTION_SKIP:
11691                 ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11692                 mtx_unlock(&lun->lun_lock);
11693                 ctl_enqueue_rtr((union ctl_io *)ctsio);
11694                 break;
11695         case CTL_ACTION_OVERLAP:
11696                 mtx_unlock(&lun->lun_lock);
11697                 ctl_set_overlapped_cmd(ctsio);
11698                 ctl_done((union ctl_io *)ctsio);
11699                 break;
11700         case CTL_ACTION_OVERLAP_TAG:
11701                 mtx_unlock(&lun->lun_lock);
11702                 ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
11703                 ctl_done((union ctl_io *)ctsio);
11704                 break;
11705         case CTL_ACTION_ERROR:
11706         default:
11707                 mtx_unlock(&lun->lun_lock);
11708                 ctl_set_internal_failure(ctsio,
11709                                          /*sks_valid*/ 0,
11710                                          /*retry_count*/ 0);
11711                 ctl_done((union ctl_io *)ctsio);
11712                 break;
11713         }
11714         return (retval);
11715 }
11716
11717 const struct ctl_cmd_entry *
11718 ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa)
11719 {
11720         const struct ctl_cmd_entry *entry;
11721         int service_action;
11722
11723         entry = &ctl_cmd_table[ctsio->cdb[0]];
11724         if (sa)
11725                 *sa = ((entry->flags & CTL_CMD_FLAG_SA5) != 0);
11726         if (entry->flags & CTL_CMD_FLAG_SA5) {
11727                 service_action = ctsio->cdb[1] & SERVICE_ACTION_MASK;
11728                 entry = &((const struct ctl_cmd_entry *)
11729                     entry->execute)[service_action];
11730         }
11731         return (entry);
11732 }
11733
11734 const struct ctl_cmd_entry *
11735 ctl_validate_command(struct ctl_scsiio *ctsio)
11736 {
11737         const struct ctl_cmd_entry *entry;
11738         int i, sa;
11739         uint8_t diff;
11740
11741         entry = ctl_get_cmd_entry(ctsio, &sa);
11742         if (entry->execute == NULL) {
11743                 if (sa)
11744                         ctl_set_invalid_field(ctsio,
11745                                               /*sks_valid*/ 1,
11746                                               /*command*/ 1,
11747                                               /*field*/ 1,
11748                                               /*bit_valid*/ 1,
11749                                               /*bit*/ 4);
11750                 else
11751                         ctl_set_invalid_opcode(ctsio);
11752                 ctl_done((union ctl_io *)ctsio);
11753                 return (NULL);
11754         }
11755         KASSERT(entry->length > 0,
11756             ("Not defined length for command 0x%02x/0x%02x",
11757              ctsio->cdb[0], ctsio->cdb[1]));
11758         for (i = 1; i < entry->length; i++) {
11759                 diff = ctsio->cdb[i] & ~entry->usage[i - 1];
11760                 if (diff == 0)
11761                         continue;
11762                 ctl_set_invalid_field(ctsio,
11763                                       /*sks_valid*/ 1,
11764                                       /*command*/ 1,
11765                                       /*field*/ i,
11766                                       /*bit_valid*/ 1,
11767                                       /*bit*/ fls(diff) - 1);
11768                 ctl_done((union ctl_io *)ctsio);
11769                 return (NULL);
11770         }
11771         return (entry);
11772 }
11773
11774 static int
11775 ctl_cmd_applicable(uint8_t lun_type, const struct ctl_cmd_entry *entry)
11776 {
11777
11778         switch (lun_type) {
11779         case T_DIRECT:
11780                 if ((entry->flags & CTL_CMD_FLAG_OK_ON_DIRECT) == 0)
11781                         return (0);
11782                 break;
11783         case T_PROCESSOR:
11784                 if ((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0)
11785                         return (0);
11786                 break;
11787         case T_CDROM:
11788                 if ((entry->flags & CTL_CMD_FLAG_OK_ON_CDROM) == 0)
11789                         return (0);
11790                 break;
11791         default:
11792                 return (0);
11793         }
11794         return (1);
11795 }
11796
11797 static int
11798 ctl_scsiio(struct ctl_scsiio *ctsio)
11799 {
11800         int retval;
11801         const struct ctl_cmd_entry *entry;
11802
11803         retval = CTL_RETVAL_COMPLETE;
11804
11805         CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
11806
11807         entry = ctl_get_cmd_entry(ctsio, NULL);
11808
11809         /*
11810          * If this I/O has been aborted, just send it straight to
11811          * ctl_done() without executing it.
11812          */
11813         if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
11814                 ctl_done((union ctl_io *)ctsio);
11815                 goto bailout;
11816         }
11817
11818         /*
11819          * All the checks should have been handled by ctl_scsiio_precheck().
11820          * We should be clear now to just execute the I/O.
11821          */
11822         retval = entry->execute(ctsio);
11823
11824 bailout:
11825         return (retval);
11826 }
11827
11828 /*
11829  * Since we only implement one target right now, a bus reset simply resets
11830  * our single target.
11831  */
11832 static int
11833 ctl_bus_reset(struct ctl_softc *softc, union ctl_io *io)
11834 {
11835         return(ctl_target_reset(softc, io, CTL_UA_BUS_RESET));
11836 }
11837
11838 static int
11839 ctl_target_reset(struct ctl_softc *softc, union ctl_io *io,
11840                  ctl_ua_type ua_type)
11841 {
11842         struct ctl_port *port = CTL_PORT(io);
11843         struct ctl_lun *lun;
11844         int retval;
11845
11846         if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11847                 union ctl_ha_msg msg_info;
11848
11849                 msg_info.hdr.nexus = io->io_hdr.nexus;
11850                 if (ua_type==CTL_UA_TARG_RESET)
11851                         msg_info.task.task_action = CTL_TASK_TARGET_RESET;
11852                 else
11853                         msg_info.task.task_action = CTL_TASK_BUS_RESET;
11854                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11855                 msg_info.hdr.original_sc = NULL;
11856                 msg_info.hdr.serializing_sc = NULL;
11857                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11858                     sizeof(msg_info.task), M_WAITOK);
11859         }
11860         retval = 0;
11861
11862         mtx_lock(&softc->ctl_lock);
11863         STAILQ_FOREACH(lun, &softc->lun_list, links) {
11864                 if (port != NULL &&
11865                     ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
11866                         continue;
11867                 retval += ctl_do_lun_reset(lun, io, ua_type);
11868         }
11869         mtx_unlock(&softc->ctl_lock);
11870         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11871         return (retval);
11872 }
11873
11874 /*
11875  * The LUN should always be set.  The I/O is optional, and is used to
11876  * distinguish between I/Os sent by this initiator, and by other
11877  * initiators.  We set unit attention for initiators other than this one.
11878  * SAM-3 is vague on this point.  It does say that a unit attention should
11879  * be established for other initiators when a LUN is reset (see section
11880  * 5.7.3), but it doesn't specifically say that the unit attention should
11881  * be established for this particular initiator when a LUN is reset.  Here
11882  * is the relevant text, from SAM-3 rev 8:
11883  *
11884  * 5.7.2 When a SCSI initiator port aborts its own tasks
11885  *
11886  * When a SCSI initiator port causes its own task(s) to be aborted, no
11887  * notification that the task(s) have been aborted shall be returned to
11888  * the SCSI initiator port other than the completion response for the
11889  * command or task management function action that caused the task(s) to
11890  * be aborted and notification(s) associated with related effects of the
11891  * action (e.g., a reset unit attention condition).
11892  *
11893  * XXX KDM for now, we're setting unit attention for all initiators.
11894  */
11895 static int
11896 ctl_do_lun_reset(struct ctl_lun *lun, union ctl_io *io, ctl_ua_type ua_type)
11897 {
11898         union ctl_io *xio;
11899 #if 0
11900         uint32_t initidx;
11901 #endif
11902         int i;
11903
11904         mtx_lock(&lun->lun_lock);
11905         /*
11906          * Run through the OOA queue and abort each I/O.
11907          */
11908         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11909              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11910                 xio->io_hdr.flags |= CTL_FLAG_ABORT | CTL_FLAG_ABORT_STATUS;
11911         }
11912
11913         /*
11914          * This version sets unit attention for every
11915          */
11916 #if 0
11917         initidx = ctl_get_initindex(&io->io_hdr.nexus);
11918         ctl_est_ua_all(lun, initidx, ua_type);
11919 #else
11920         ctl_est_ua_all(lun, -1, ua_type);
11921 #endif
11922
11923         /*
11924          * A reset (any kind, really) clears reservations established with
11925          * RESERVE/RELEASE.  It does not clear reservations established
11926          * with PERSISTENT RESERVE OUT, but we don't support that at the
11927          * moment anyway.  See SPC-2, section 5.6.  SPC-3 doesn't address
11928          * reservations made with the RESERVE/RELEASE commands, because
11929          * those commands are obsolete in SPC-3.
11930          */
11931         lun->flags &= ~CTL_LUN_RESERVED;
11932
11933 #ifdef CTL_WITH_CA
11934         for (i = 0; i < CTL_MAX_INITIATORS; i++)
11935                 ctl_clear_mask(lun->have_ca, i);
11936 #endif
11937         lun->prevent_count = 0;
11938         if (lun->prevent) {
11939                 for (i = 0; i < CTL_MAX_INITIATORS; i++)
11940                         ctl_clear_mask(lun->prevent, i);
11941         }
11942         mtx_unlock(&lun->lun_lock);
11943
11944         return (0);
11945 }
11946
11947 static int
11948 ctl_lun_reset(struct ctl_softc *softc, union ctl_io *io)
11949 {
11950         struct ctl_lun *lun;
11951         uint32_t targ_lun;
11952         int retval;
11953
11954         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11955         mtx_lock(&softc->ctl_lock);
11956         if (targ_lun >= CTL_MAX_LUNS ||
11957             (lun = softc->ctl_luns[targ_lun]) == NULL) {
11958                 mtx_unlock(&softc->ctl_lock);
11959                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
11960                 return (1);
11961         }
11962         retval = ctl_do_lun_reset(lun, io, CTL_UA_LUN_RESET);
11963         mtx_unlock(&softc->ctl_lock);
11964         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11965
11966         if ((io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) == 0) {
11967                 union ctl_ha_msg msg_info;
11968
11969                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11970                 msg_info.hdr.nexus = io->io_hdr.nexus;
11971                 msg_info.task.task_action = CTL_TASK_LUN_RESET;
11972                 msg_info.hdr.original_sc = NULL;
11973                 msg_info.hdr.serializing_sc = NULL;
11974                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11975                     sizeof(msg_info.task), M_WAITOK);
11976         }
11977         return (retval);
11978 }
11979
11980 static void
11981 ctl_abort_tasks_lun(struct ctl_lun *lun, uint32_t targ_port, uint32_t init_id,
11982     int other_sc)
11983 {
11984         union ctl_io *xio;
11985
11986         mtx_assert(&lun->lun_lock, MA_OWNED);
11987
11988         /*
11989          * Run through the OOA queue and attempt to find the given I/O.
11990          * The target port, initiator ID, tag type and tag number have to
11991          * match the values that we got from the initiator.  If we have an
11992          * untagged command to abort, simply abort the first untagged command
11993          * we come to.  We only allow one untagged command at a time of course.
11994          */
11995         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11996              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11997
11998                 if ((targ_port == UINT32_MAX ||
11999                      targ_port == xio->io_hdr.nexus.targ_port) &&
12000                     (init_id == UINT32_MAX ||
12001                      init_id == xio->io_hdr.nexus.initid)) {
12002                         if (targ_port != xio->io_hdr.nexus.targ_port ||
12003                             init_id != xio->io_hdr.nexus.initid)
12004                                 xio->io_hdr.flags |= CTL_FLAG_ABORT_STATUS;
12005                         xio->io_hdr.flags |= CTL_FLAG_ABORT;
12006                         if (!other_sc && !(lun->flags & CTL_LUN_PRIMARY_SC)) {
12007                                 union ctl_ha_msg msg_info;
12008
12009                                 msg_info.hdr.nexus = xio->io_hdr.nexus;
12010                                 msg_info.task.task_action = CTL_TASK_ABORT_TASK;
12011                                 msg_info.task.tag_num = xio->scsiio.tag_num;
12012                                 msg_info.task.tag_type = xio->scsiio.tag_type;
12013                                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
12014                                 msg_info.hdr.original_sc = NULL;
12015                                 msg_info.hdr.serializing_sc = NULL;
12016                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
12017                                     sizeof(msg_info.task), M_NOWAIT);
12018                         }
12019                 }
12020         }
12021 }
12022
12023 static int
12024 ctl_abort_task_set(union ctl_io *io)
12025 {
12026         struct ctl_softc *softc = CTL_SOFTC(io);
12027         struct ctl_lun *lun;
12028         uint32_t targ_lun;
12029
12030         /*
12031          * Look up the LUN.
12032          */
12033         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12034         mtx_lock(&softc->ctl_lock);
12035         if (targ_lun >= CTL_MAX_LUNS ||
12036             (lun = softc->ctl_luns[targ_lun]) == NULL) {
12037                 mtx_unlock(&softc->ctl_lock);
12038                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
12039                 return (1);
12040         }
12041
12042         mtx_lock(&lun->lun_lock);
12043         mtx_unlock(&softc->ctl_lock);
12044         if (io->taskio.task_action == CTL_TASK_ABORT_TASK_SET) {
12045                 ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
12046                     io->io_hdr.nexus.initid,
12047                     (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12048         } else { /* CTL_TASK_CLEAR_TASK_SET */
12049                 ctl_abort_tasks_lun(lun, UINT32_MAX, UINT32_MAX,
12050                     (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12051         }
12052         mtx_unlock(&lun->lun_lock);
12053         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12054         return (0);
12055 }
12056
12057 static int
12058 ctl_i_t_nexus_reset(union ctl_io *io)
12059 {
12060         struct ctl_softc *softc = CTL_SOFTC(io);
12061         struct ctl_lun *lun;
12062         uint32_t initidx;
12063
12064         if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
12065                 union ctl_ha_msg msg_info;
12066
12067                 msg_info.hdr.nexus = io->io_hdr.nexus;
12068                 msg_info.task.task_action = CTL_TASK_I_T_NEXUS_RESET;
12069                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
12070                 msg_info.hdr.original_sc = NULL;
12071                 msg_info.hdr.serializing_sc = NULL;
12072                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
12073                     sizeof(msg_info.task), M_WAITOK);
12074         }
12075
12076         initidx = ctl_get_initindex(&io->io_hdr.nexus);
12077         mtx_lock(&softc->ctl_lock);
12078         STAILQ_FOREACH(lun, &softc->lun_list, links) {
12079                 mtx_lock(&lun->lun_lock);
12080                 ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
12081                     io->io_hdr.nexus.initid, 1);
12082 #ifdef CTL_WITH_CA
12083                 ctl_clear_mask(lun->have_ca, initidx);
12084 #endif
12085                 if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == initidx))
12086                         lun->flags &= ~CTL_LUN_RESERVED;
12087                 if (lun->prevent && ctl_is_set(lun->prevent, initidx)) {
12088                         ctl_clear_mask(lun->prevent, initidx);
12089                         lun->prevent_count--;
12090                 }
12091                 ctl_est_ua(lun, initidx, CTL_UA_I_T_NEXUS_LOSS);
12092                 mtx_unlock(&lun->lun_lock);
12093         }
12094         mtx_unlock(&softc->ctl_lock);
12095         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12096         return (0);
12097 }
12098
12099 static int
12100 ctl_abort_task(union ctl_io *io)
12101 {
12102         struct ctl_softc *softc = CTL_SOFTC(io);
12103         union ctl_io *xio;
12104         struct ctl_lun *lun;
12105 #if 0
12106         struct sbuf sb;
12107         char printbuf[128];
12108 #endif
12109         int found;
12110         uint32_t targ_lun;
12111
12112         found = 0;
12113
12114         /*
12115          * Look up the LUN.
12116          */
12117         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12118         mtx_lock(&softc->ctl_lock);
12119         if (targ_lun >= CTL_MAX_LUNS ||
12120             (lun = softc->ctl_luns[targ_lun]) == NULL) {
12121                 mtx_unlock(&softc->ctl_lock);
12122                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
12123                 return (1);
12124         }
12125
12126 #if 0
12127         printf("ctl_abort_task: called for lun %lld, tag %d type %d\n",
12128                lun->lun, io->taskio.tag_num, io->taskio.tag_type);
12129 #endif
12130
12131         mtx_lock(&lun->lun_lock);
12132         mtx_unlock(&softc->ctl_lock);
12133         /*
12134          * Run through the OOA queue and attempt to find the given I/O.
12135          * The target port, initiator ID, tag type and tag number have to
12136          * match the values that we got from the initiator.  If we have an
12137          * untagged command to abort, simply abort the first untagged command
12138          * we come to.  We only allow one untagged command at a time of course.
12139          */
12140         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12141              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12142 #if 0
12143                 sbuf_new(&sb, printbuf, sizeof(printbuf), SBUF_FIXEDLEN);
12144
12145                 sbuf_printf(&sb, "LUN %lld tag %d type %d%s%s%s%s: ",
12146                             lun->lun, xio->scsiio.tag_num,
12147                             xio->scsiio.tag_type,
12148                             (xio->io_hdr.blocked_links.tqe_prev
12149                             == NULL) ? "" : " BLOCKED",
12150                             (xio->io_hdr.flags &
12151                             CTL_FLAG_DMA_INPROG) ? " DMA" : "",
12152                             (xio->io_hdr.flags &
12153                             CTL_FLAG_ABORT) ? " ABORT" : "",
12154                             (xio->io_hdr.flags &
12155                             CTL_FLAG_IS_WAS_ON_RTR ? " RTR" : ""));
12156                 ctl_scsi_command_string(&xio->scsiio, NULL, &sb);
12157                 sbuf_finish(&sb);
12158                 printf("%s\n", sbuf_data(&sb));
12159 #endif
12160
12161                 if ((xio->io_hdr.nexus.targ_port != io->io_hdr.nexus.targ_port)
12162                  || (xio->io_hdr.nexus.initid != io->io_hdr.nexus.initid)
12163                  || (xio->io_hdr.flags & CTL_FLAG_ABORT))
12164                         continue;
12165
12166                 /*
12167                  * If the abort says that the task is untagged, the
12168                  * task in the queue must be untagged.  Otherwise,
12169                  * we just check to see whether the tag numbers
12170                  * match.  This is because the QLogic firmware
12171                  * doesn't pass back the tag type in an abort
12172                  * request.
12173                  */
12174 #if 0
12175                 if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
12176                   && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
12177                  || (xio->scsiio.tag_num == io->taskio.tag_num))
12178 #endif
12179                 /*
12180                  * XXX KDM we've got problems with FC, because it
12181                  * doesn't send down a tag type with aborts.  So we
12182                  * can only really go by the tag number...
12183                  * This may cause problems with parallel SCSI.
12184                  * Need to figure that out!!
12185                  */
12186                 if (xio->scsiio.tag_num == io->taskio.tag_num) {
12187                         xio->io_hdr.flags |= CTL_FLAG_ABORT;
12188                         found = 1;
12189                         if ((io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) == 0 &&
12190                             !(lun->flags & CTL_LUN_PRIMARY_SC)) {
12191                                 union ctl_ha_msg msg_info;
12192
12193                                 msg_info.hdr.nexus = io->io_hdr.nexus;
12194                                 msg_info.task.task_action = CTL_TASK_ABORT_TASK;
12195                                 msg_info.task.tag_num = io->taskio.tag_num;
12196                                 msg_info.task.tag_type = io->taskio.tag_type;
12197                                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
12198                                 msg_info.hdr.original_sc = NULL;
12199                                 msg_info.hdr.serializing_sc = NULL;
12200 #if 0
12201                                 printf("Sent Abort to other side\n");
12202 #endif
12203                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
12204                                     sizeof(msg_info.task), M_NOWAIT);
12205                         }
12206 #if 0
12207                         printf("ctl_abort_task: found I/O to abort\n");
12208 #endif
12209                 }
12210         }
12211         mtx_unlock(&lun->lun_lock);
12212
12213         if (found == 0) {
12214                 /*
12215                  * This isn't really an error.  It's entirely possible for
12216                  * the abort and command completion to cross on the wire.
12217                  * This is more of an informative/diagnostic error.
12218                  */
12219 #if 0
12220                 printf("ctl_abort_task: ABORT sent for nonexistent I/O: "
12221                        "%u:%u:%u tag %d type %d\n",
12222                        io->io_hdr.nexus.initid,
12223                        io->io_hdr.nexus.targ_port,
12224                        io->io_hdr.nexus.targ_lun, io->taskio.tag_num,
12225                        io->taskio.tag_type);
12226 #endif
12227         }
12228         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12229         return (0);
12230 }
12231
12232 static int
12233 ctl_query_task(union ctl_io *io, int task_set)
12234 {
12235         struct ctl_softc *softc = CTL_SOFTC(io);
12236         union ctl_io *xio;
12237         struct ctl_lun *lun;
12238         int found = 0;
12239         uint32_t targ_lun;
12240
12241         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12242         mtx_lock(&softc->ctl_lock);
12243         if (targ_lun >= CTL_MAX_LUNS ||
12244             (lun = softc->ctl_luns[targ_lun]) == NULL) {
12245                 mtx_unlock(&softc->ctl_lock);
12246                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
12247                 return (1);
12248         }
12249         mtx_lock(&lun->lun_lock);
12250         mtx_unlock(&softc->ctl_lock);
12251         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12252              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12253
12254                 if ((xio->io_hdr.nexus.targ_port != io->io_hdr.nexus.targ_port)
12255                  || (xio->io_hdr.nexus.initid != io->io_hdr.nexus.initid)
12256                  || (xio->io_hdr.flags & CTL_FLAG_ABORT))
12257                         continue;
12258
12259                 if (task_set || xio->scsiio.tag_num == io->taskio.tag_num) {
12260                         found = 1;
12261                         break;
12262                 }
12263         }
12264         mtx_unlock(&lun->lun_lock);
12265         if (found)
12266                 io->taskio.task_status = CTL_TASK_FUNCTION_SUCCEEDED;
12267         else
12268                 io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12269         return (0);
12270 }
12271
12272 static int
12273 ctl_query_async_event(union ctl_io *io)
12274 {
12275         struct ctl_softc *softc = CTL_SOFTC(io);
12276         struct ctl_lun *lun;
12277         ctl_ua_type ua;
12278         uint32_t targ_lun, initidx;
12279
12280         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12281         mtx_lock(&softc->ctl_lock);
12282         if (targ_lun >= CTL_MAX_LUNS ||
12283             (lun = softc->ctl_luns[targ_lun]) == NULL) {
12284                 mtx_unlock(&softc->ctl_lock);
12285                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
12286                 return (1);
12287         }
12288         mtx_lock(&lun->lun_lock);
12289         mtx_unlock(&softc->ctl_lock);
12290         initidx = ctl_get_initindex(&io->io_hdr.nexus);
12291         ua = ctl_build_qae(lun, initidx, io->taskio.task_resp);
12292         mtx_unlock(&lun->lun_lock);
12293         if (ua != CTL_UA_NONE)
12294                 io->taskio.task_status = CTL_TASK_FUNCTION_SUCCEEDED;
12295         else
12296                 io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12297         return (0);
12298 }
12299
12300 static void
12301 ctl_run_task(union ctl_io *io)
12302 {
12303         struct ctl_softc *softc = CTL_SOFTC(io);
12304         int retval = 1;
12305
12306         CTL_DEBUG_PRINT(("ctl_run_task\n"));
12307         KASSERT(io->io_hdr.io_type == CTL_IO_TASK,
12308             ("ctl_run_task: Unextected io_type %d\n", io->io_hdr.io_type));
12309         io->taskio.task_status = CTL_TASK_FUNCTION_NOT_SUPPORTED;
12310         bzero(io->taskio.task_resp, sizeof(io->taskio.task_resp));
12311         switch (io->taskio.task_action) {
12312         case CTL_TASK_ABORT_TASK:
12313                 retval = ctl_abort_task(io);
12314                 break;
12315         case CTL_TASK_ABORT_TASK_SET:
12316         case CTL_TASK_CLEAR_TASK_SET:
12317                 retval = ctl_abort_task_set(io);
12318                 break;
12319         case CTL_TASK_CLEAR_ACA:
12320                 break;
12321         case CTL_TASK_I_T_NEXUS_RESET:
12322                 retval = ctl_i_t_nexus_reset(io);
12323                 break;
12324         case CTL_TASK_LUN_RESET:
12325                 retval = ctl_lun_reset(softc, io);
12326                 break;
12327         case CTL_TASK_TARGET_RESET:
12328                 retval = ctl_target_reset(softc, io, CTL_UA_TARG_RESET);
12329                 break;
12330         case CTL_TASK_BUS_RESET:
12331                 retval = ctl_bus_reset(softc, io);
12332                 break;
12333         case CTL_TASK_PORT_LOGIN:
12334                 break;
12335         case CTL_TASK_PORT_LOGOUT:
12336                 break;
12337         case CTL_TASK_QUERY_TASK:
12338                 retval = ctl_query_task(io, 0);
12339                 break;
12340         case CTL_TASK_QUERY_TASK_SET:
12341                 retval = ctl_query_task(io, 1);
12342                 break;
12343         case CTL_TASK_QUERY_ASYNC_EVENT:
12344                 retval = ctl_query_async_event(io);
12345                 break;
12346         default:
12347                 printf("%s: got unknown task management event %d\n",
12348                        __func__, io->taskio.task_action);
12349                 break;
12350         }
12351         if (retval == 0)
12352                 io->io_hdr.status = CTL_SUCCESS;
12353         else
12354                 io->io_hdr.status = CTL_ERROR;
12355         ctl_done(io);
12356 }
12357
12358 /*
12359  * For HA operation.  Handle commands that come in from the other
12360  * controller.
12361  */
12362 static void
12363 ctl_handle_isc(union ctl_io *io)
12364 {
12365         struct ctl_softc *softc = CTL_SOFTC(io);
12366         struct ctl_lun *lun;
12367         const struct ctl_cmd_entry *entry;
12368         uint32_t targ_lun;
12369
12370         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12371         switch (io->io_hdr.msg_type) {
12372         case CTL_MSG_SERIALIZE:
12373                 ctl_serialize_other_sc_cmd(&io->scsiio);
12374                 break;
12375         case CTL_MSG_R2R:               /* Only used in SER_ONLY mode. */
12376                 entry = ctl_get_cmd_entry(&io->scsiio, NULL);
12377                 if (targ_lun >= CTL_MAX_LUNS ||
12378                     (lun = softc->ctl_luns[targ_lun]) == NULL) {
12379                         ctl_done(io);
12380                         break;
12381                 }
12382                 mtx_lock(&lun->lun_lock);
12383                 if (ctl_scsiio_lun_check(lun, entry, &io->scsiio) != 0) {
12384                         mtx_unlock(&lun->lun_lock);
12385                         ctl_done(io);
12386                         break;
12387                 }
12388                 io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
12389                 mtx_unlock(&lun->lun_lock);
12390                 ctl_enqueue_rtr(io);
12391                 break;
12392         case CTL_MSG_FINISH_IO:
12393                 if (softc->ha_mode == CTL_HA_MODE_XFER) {
12394                         ctl_done(io);
12395                         break;
12396                 }
12397                 if (targ_lun >= CTL_MAX_LUNS ||
12398                     (lun = softc->ctl_luns[targ_lun]) == NULL) {
12399                         ctl_free_io(io);
12400                         break;
12401                 }
12402                 mtx_lock(&lun->lun_lock);
12403                 TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
12404                 ctl_check_blocked(lun);
12405                 mtx_unlock(&lun->lun_lock);
12406                 ctl_free_io(io);
12407                 break;
12408         case CTL_MSG_PERS_ACTION:
12409                 ctl_hndl_per_res_out_on_other_sc(io);
12410                 ctl_free_io(io);
12411                 break;
12412         case CTL_MSG_BAD_JUJU:
12413                 ctl_done(io);
12414                 break;
12415         case CTL_MSG_DATAMOVE:          /* Only used in XFER mode */
12416                 ctl_datamove_remote(io);
12417                 break;
12418         case CTL_MSG_DATAMOVE_DONE:     /* Only used in XFER mode */
12419                 io->scsiio.be_move_done(io);
12420                 break;
12421         case CTL_MSG_FAILOVER:
12422                 ctl_failover_lun(io);
12423                 ctl_free_io(io);
12424                 break;
12425         default:
12426                 printf("%s: Invalid message type %d\n",
12427                        __func__, io->io_hdr.msg_type);
12428                 ctl_free_io(io);
12429                 break;
12430         }
12431
12432 }
12433
12434
12435 /*
12436  * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
12437  * there is no match.
12438  */
12439 static ctl_lun_error_pattern
12440 ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
12441 {
12442         const struct ctl_cmd_entry *entry;
12443         ctl_lun_error_pattern filtered_pattern, pattern;
12444
12445         pattern = desc->error_pattern;
12446
12447         /*
12448          * XXX KDM we need more data passed into this function to match a
12449          * custom pattern, and we actually need to implement custom pattern
12450          * matching.
12451          */
12452         if (pattern & CTL_LUN_PAT_CMD)
12453                 return (CTL_LUN_PAT_CMD);
12454
12455         if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
12456                 return (CTL_LUN_PAT_ANY);
12457
12458         entry = ctl_get_cmd_entry(ctsio, NULL);
12459
12460         filtered_pattern = entry->pattern & pattern;
12461
12462         /*
12463          * If the user requested specific flags in the pattern (e.g.
12464          * CTL_LUN_PAT_RANGE), make sure the command supports all of those
12465          * flags.
12466          *
12467          * If the user did not specify any flags, it doesn't matter whether
12468          * or not the command supports the flags.
12469          */
12470         if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
12471              (pattern & ~CTL_LUN_PAT_MASK))
12472                 return (CTL_LUN_PAT_NONE);
12473
12474         /*
12475          * If the user asked for a range check, see if the requested LBA
12476          * range overlaps with this command's LBA range.
12477          */
12478         if (filtered_pattern & CTL_LUN_PAT_RANGE) {
12479                 uint64_t lba1;
12480                 uint64_t len1;
12481                 ctl_action action;
12482                 int retval;
12483
12484                 retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
12485                 if (retval != 0)
12486                         return (CTL_LUN_PAT_NONE);
12487
12488                 action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
12489                                               desc->lba_range.len, FALSE);
12490                 /*
12491                  * A "pass" means that the LBA ranges don't overlap, so
12492                  * this doesn't match the user's range criteria.
12493                  */
12494                 if (action == CTL_ACTION_PASS)
12495                         return (CTL_LUN_PAT_NONE);
12496         }
12497
12498         return (filtered_pattern);
12499 }
12500
12501 static void
12502 ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
12503 {
12504         struct ctl_error_desc *desc, *desc2;
12505
12506         mtx_assert(&lun->lun_lock, MA_OWNED);
12507
12508         STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
12509                 ctl_lun_error_pattern pattern;
12510                 /*
12511                  * Check to see whether this particular command matches
12512                  * the pattern in the descriptor.
12513                  */
12514                 pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
12515                 if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
12516                         continue;
12517
12518                 switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
12519                 case CTL_LUN_INJ_ABORTED:
12520                         ctl_set_aborted(&io->scsiio);
12521                         break;
12522                 case CTL_LUN_INJ_MEDIUM_ERR:
12523                         ctl_set_medium_error(&io->scsiio,
12524                             (io->io_hdr.flags & CTL_FLAG_DATA_MASK) !=
12525                              CTL_FLAG_DATA_OUT);
12526                         break;
12527                 case CTL_LUN_INJ_UA:
12528                         /* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
12529                          * OCCURRED */
12530                         ctl_set_ua(&io->scsiio, 0x29, 0x00);
12531                         break;
12532                 case CTL_LUN_INJ_CUSTOM:
12533                         /*
12534                          * We're assuming the user knows what he is doing.
12535                          * Just copy the sense information without doing
12536                          * checks.
12537                          */
12538                         bcopy(&desc->custom_sense, &io->scsiio.sense_data,
12539                               MIN(sizeof(desc->custom_sense),
12540                                   sizeof(io->scsiio.sense_data)));
12541                         io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
12542                         io->scsiio.sense_len = SSD_FULL_SIZE;
12543                         io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
12544                         break;
12545                 case CTL_LUN_INJ_NONE:
12546                 default:
12547                         /*
12548                          * If this is an error injection type we don't know
12549                          * about, clear the continuous flag (if it is set)
12550                          * so it will get deleted below.
12551                          */
12552                         desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
12553                         break;
12554                 }
12555                 /*
12556                  * By default, each error injection action is a one-shot
12557                  */
12558                 if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
12559                         continue;
12560
12561                 STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
12562
12563                 free(desc, M_CTL);
12564         }
12565 }
12566
12567 #ifdef CTL_IO_DELAY
12568 static void
12569 ctl_datamove_timer_wakeup(void *arg)
12570 {
12571         union ctl_io *io;
12572
12573         io = (union ctl_io *)arg;
12574
12575         ctl_datamove(io);
12576 }
12577 #endif /* CTL_IO_DELAY */
12578
12579 void
12580 ctl_datamove(union ctl_io *io)
12581 {
12582         void (*fe_datamove)(union ctl_io *io);
12583
12584         mtx_assert(&((struct ctl_softc *)CTL_SOFTC(io))->ctl_lock, MA_NOTOWNED);
12585
12586         CTL_DEBUG_PRINT(("ctl_datamove\n"));
12587
12588 #ifdef CTL_TIME_IO
12589         if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12590                 char str[256];
12591                 char path_str[64];
12592                 struct sbuf sb;
12593
12594                 ctl_scsi_path_string(io, path_str, sizeof(path_str));
12595                 sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12596
12597                 sbuf_cat(&sb, path_str);
12598                 switch (io->io_hdr.io_type) {
12599                 case CTL_IO_SCSI:
12600                         ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12601                         sbuf_printf(&sb, "\n");
12602                         sbuf_cat(&sb, path_str);
12603                         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12604                                     io->scsiio.tag_num, io->scsiio.tag_type);
12605                         break;
12606                 case CTL_IO_TASK:
12607                         sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12608                                     "Tag Type: %d\n", io->taskio.task_action,
12609                                     io->taskio.tag_num, io->taskio.tag_type);
12610                         break;
12611                 default:
12612                         panic("%s: Invalid CTL I/O type %d\n",
12613                             __func__, io->io_hdr.io_type);
12614                 }
12615                 sbuf_cat(&sb, path_str);
12616                 sbuf_printf(&sb, "ctl_datamove: %jd seconds\n",
12617                             (intmax_t)time_uptime - io->io_hdr.start_time);
12618                 sbuf_finish(&sb);
12619                 printf("%s", sbuf_data(&sb));
12620         }
12621 #endif /* CTL_TIME_IO */
12622
12623 #ifdef CTL_IO_DELAY
12624         if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
12625                 io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
12626         } else {
12627                 if ((lun != NULL)
12628                  && (lun->delay_info.datamove_delay > 0)) {
12629
12630                         callout_init(&io->io_hdr.delay_callout, /*mpsafe*/ 1);
12631                         io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
12632                         callout_reset(&io->io_hdr.delay_callout,
12633                                       lun->delay_info.datamove_delay * hz,
12634                                       ctl_datamove_timer_wakeup, io);
12635                         if (lun->delay_info.datamove_type ==
12636                             CTL_DELAY_TYPE_ONESHOT)
12637                                 lun->delay_info.datamove_delay = 0;
12638                         return;
12639                 }
12640         }
12641 #endif
12642
12643         /*
12644          * This command has been aborted.  Set the port status, so we fail
12645          * the data move.
12646          */
12647         if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12648                 printf("ctl_datamove: tag 0x%04x on (%u:%u:%u) aborted\n",
12649                        io->scsiio.tag_num, io->io_hdr.nexus.initid,
12650                        io->io_hdr.nexus.targ_port,
12651                        io->io_hdr.nexus.targ_lun);
12652                 io->io_hdr.port_status = 31337;
12653                 /*
12654                  * Note that the backend, in this case, will get the
12655                  * callback in its context.  In other cases it may get
12656                  * called in the frontend's interrupt thread context.
12657                  */
12658                 io->scsiio.be_move_done(io);
12659                 return;
12660         }
12661
12662         /* Don't confuse frontend with zero length data move. */
12663         if (io->scsiio.kern_data_len == 0) {
12664                 io->scsiio.be_move_done(io);
12665                 return;
12666         }
12667
12668         fe_datamove = CTL_PORT(io)->fe_datamove;
12669         fe_datamove(io);
12670 }
12671
12672 static void
12673 ctl_send_datamove_done(union ctl_io *io, int have_lock)
12674 {
12675         union ctl_ha_msg msg;
12676 #ifdef CTL_TIME_IO
12677         struct bintime cur_bt;
12678 #endif
12679
12680         memset(&msg, 0, sizeof(msg));
12681         msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
12682         msg.hdr.original_sc = io;
12683         msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
12684         msg.hdr.nexus = io->io_hdr.nexus;
12685         msg.hdr.status = io->io_hdr.status;
12686         msg.scsi.tag_num = io->scsiio.tag_num;
12687         msg.scsi.tag_type = io->scsiio.tag_type;
12688         msg.scsi.scsi_status = io->scsiio.scsi_status;
12689         memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
12690                io->scsiio.sense_len);
12691         msg.scsi.sense_len = io->scsiio.sense_len;
12692         msg.scsi.sense_residual = io->scsiio.sense_residual;
12693         msg.scsi.fetd_status = io->io_hdr.port_status;
12694         msg.scsi.residual = io->scsiio.residual;
12695         io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12696         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12697                 ctl_failover_io(io, /*have_lock*/ have_lock);
12698                 return;
12699         }
12700         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12701             sizeof(msg.scsi) - sizeof(msg.scsi.sense_data) +
12702             msg.scsi.sense_len, M_WAITOK);
12703
12704 #ifdef CTL_TIME_IO
12705         getbinuptime(&cur_bt);
12706         bintime_sub(&cur_bt, &io->io_hdr.dma_start_bt);
12707         bintime_add(&io->io_hdr.dma_bt, &cur_bt);
12708 #endif
12709         io->io_hdr.num_dmas++;
12710 }
12711
12712 /*
12713  * The DMA to the remote side is done, now we need to tell the other side
12714  * we're done so it can continue with its data movement.
12715  */
12716 static void
12717 ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
12718 {
12719         union ctl_io *io;
12720         uint32_t i;
12721
12722         io = rq->context;
12723
12724         if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12725                 printf("%s: ISC DMA write failed with error %d", __func__,
12726                        rq->ret);
12727                 ctl_set_internal_failure(&io->scsiio,
12728                                          /*sks_valid*/ 1,
12729                                          /*retry_count*/ rq->ret);
12730         }
12731
12732         ctl_dt_req_free(rq);
12733
12734         for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12735                 free(io->io_hdr.local_sglist[i].addr, M_CTL);
12736         free(io->io_hdr.remote_sglist, M_CTL);
12737         io->io_hdr.remote_sglist = NULL;
12738         io->io_hdr.local_sglist = NULL;
12739
12740         /*
12741          * The data is in local and remote memory, so now we need to send
12742          * status (good or back) back to the other side.
12743          */
12744         ctl_send_datamove_done(io, /*have_lock*/ 0);
12745 }
12746
12747 /*
12748  * We've moved the data from the host/controller into local memory.  Now we
12749  * need to push it over to the remote controller's memory.
12750  */
12751 static int
12752 ctl_datamove_remote_dm_write_cb(union ctl_io *io)
12753 {
12754         int retval;
12755
12756         retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
12757                                           ctl_datamove_remote_write_cb);
12758         return (retval);
12759 }
12760
12761 static void
12762 ctl_datamove_remote_write(union ctl_io *io)
12763 {
12764         int retval;
12765         void (*fe_datamove)(union ctl_io *io);
12766
12767         /*
12768          * - Get the data from the host/HBA into local memory.
12769          * - DMA memory from the local controller to the remote controller.
12770          * - Send status back to the remote controller.
12771          */
12772
12773         retval = ctl_datamove_remote_sgl_setup(io);
12774         if (retval != 0)
12775                 return;
12776
12777         /* Switch the pointer over so the FETD knows what to do */
12778         io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12779
12780         /*
12781          * Use a custom move done callback, since we need to send completion
12782          * back to the other controller, not to the backend on this side.
12783          */
12784         io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
12785
12786         fe_datamove = CTL_PORT(io)->fe_datamove;
12787         fe_datamove(io);
12788 }
12789
12790 static int
12791 ctl_datamove_remote_dm_read_cb(union ctl_io *io)
12792 {
12793 #if 0
12794         char str[256];
12795         char path_str[64];
12796         struct sbuf sb;
12797 #endif
12798         uint32_t i;
12799
12800         for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12801                 free(io->io_hdr.local_sglist[i].addr, M_CTL);
12802         free(io->io_hdr.remote_sglist, M_CTL);
12803         io->io_hdr.remote_sglist = NULL;
12804         io->io_hdr.local_sglist = NULL;
12805
12806 #if 0
12807         scsi_path_string(io, path_str, sizeof(path_str));
12808         sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12809         sbuf_cat(&sb, path_str);
12810         scsi_command_string(&io->scsiio, NULL, &sb);
12811         sbuf_printf(&sb, "\n");
12812         sbuf_cat(&sb, path_str);
12813         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12814                     io->scsiio.tag_num, io->scsiio.tag_type);
12815         sbuf_cat(&sb, path_str);
12816         sbuf_printf(&sb, "%s: flags %#x, status %#x\n", __func__,
12817                     io->io_hdr.flags, io->io_hdr.status);
12818         sbuf_finish(&sb);
12819         printk("%s", sbuf_data(&sb));
12820 #endif
12821
12822
12823         /*
12824          * The read is done, now we need to send status (good or bad) back
12825          * to the other side.
12826          */
12827         ctl_send_datamove_done(io, /*have_lock*/ 0);
12828
12829         return (0);
12830 }
12831
12832 static void
12833 ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
12834 {
12835         union ctl_io *io;
12836         void (*fe_datamove)(union ctl_io *io);
12837
12838         io = rq->context;
12839
12840         if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12841                 printf("%s: ISC DMA read failed with error %d\n", __func__,
12842                        rq->ret);
12843                 ctl_set_internal_failure(&io->scsiio,
12844                                          /*sks_valid*/ 1,
12845                                          /*retry_count*/ rq->ret);
12846         }
12847
12848         ctl_dt_req_free(rq);
12849
12850         /* Switch the pointer over so the FETD knows what to do */
12851         io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12852
12853         /*
12854          * Use a custom move done callback, since we need to send completion
12855          * back to the other controller, not to the backend on this side.
12856          */
12857         io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
12858
12859         /* XXX KDM add checks like the ones in ctl_datamove? */
12860
12861         fe_datamove = CTL_PORT(io)->fe_datamove;
12862         fe_datamove(io);
12863 }
12864
12865 static int
12866 ctl_datamove_remote_sgl_setup(union ctl_io *io)
12867 {
12868         struct ctl_sg_entry *local_sglist;
12869         uint32_t len_to_go;
12870         int retval;
12871         int i;
12872
12873         retval = 0;
12874         local_sglist = io->io_hdr.local_sglist;
12875         len_to_go = io->scsiio.kern_data_len;
12876
12877         /*
12878          * The difficult thing here is that the size of the various
12879          * S/G segments may be different than the size from the
12880          * remote controller.  That'll make it harder when DMAing
12881          * the data back to the other side.
12882          */
12883         for (i = 0; len_to_go > 0; i++) {
12884                 local_sglist[i].len = MIN(len_to_go, CTL_HA_DATAMOVE_SEGMENT);
12885                 local_sglist[i].addr =
12886                     malloc(local_sglist[i].len, M_CTL, M_WAITOK);
12887
12888                 len_to_go -= local_sglist[i].len;
12889         }
12890         /*
12891          * Reset the number of S/G entries accordingly.  The original
12892          * number of S/G entries is available in rem_sg_entries.
12893          */
12894         io->scsiio.kern_sg_entries = i;
12895
12896 #if 0
12897         printf("%s: kern_sg_entries = %d\n", __func__,
12898                io->scsiio.kern_sg_entries);
12899         for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12900                 printf("%s: sg[%d] = %p, %lu\n", __func__, i,
12901                        local_sglist[i].addr, local_sglist[i].len);
12902 #endif
12903
12904         return (retval);
12905 }
12906
12907 static int
12908 ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
12909                          ctl_ha_dt_cb callback)
12910 {
12911         struct ctl_ha_dt_req *rq;
12912         struct ctl_sg_entry *remote_sglist, *local_sglist;
12913         uint32_t local_used, remote_used, total_used;
12914         int i, j, isc_ret;
12915
12916         rq = ctl_dt_req_alloc();
12917
12918         /*
12919          * If we failed to allocate the request, and if the DMA didn't fail
12920          * anyway, set busy status.  This is just a resource allocation
12921          * failure.
12922          */
12923         if ((rq == NULL)
12924          && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
12925              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS))
12926                 ctl_set_busy(&io->scsiio);
12927
12928         if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
12929             (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS) {
12930
12931                 if (rq != NULL)
12932                         ctl_dt_req_free(rq);
12933
12934                 /*
12935                  * The data move failed.  We need to return status back
12936                  * to the other controller.  No point in trying to DMA
12937                  * data to the remote controller.
12938                  */
12939
12940                 ctl_send_datamove_done(io, /*have_lock*/ 0);
12941
12942                 return (1);
12943         }
12944
12945         local_sglist = io->io_hdr.local_sglist;
12946         remote_sglist = io->io_hdr.remote_sglist;
12947         local_used = 0;
12948         remote_used = 0;
12949         total_used = 0;
12950
12951         /*
12952          * Pull/push the data over the wire from/to the other controller.
12953          * This takes into account the possibility that the local and
12954          * remote sglists may not be identical in terms of the size of
12955          * the elements and the number of elements.
12956          *
12957          * One fundamental assumption here is that the length allocated for
12958          * both the local and remote sglists is identical.  Otherwise, we've
12959          * essentially got a coding error of some sort.
12960          */
12961         isc_ret = CTL_HA_STATUS_SUCCESS;
12962         for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
12963                 uint32_t cur_len;
12964                 uint8_t *tmp_ptr;
12965
12966                 rq->command = command;
12967                 rq->context = io;
12968
12969                 /*
12970                  * Both pointers should be aligned.  But it is possible
12971                  * that the allocation length is not.  They should both
12972                  * also have enough slack left over at the end, though,
12973                  * to round up to the next 8 byte boundary.
12974                  */
12975                 cur_len = MIN(local_sglist[i].len - local_used,
12976                               remote_sglist[j].len - remote_used);
12977                 rq->size = cur_len;
12978
12979                 tmp_ptr = (uint8_t *)local_sglist[i].addr;
12980                 tmp_ptr += local_used;
12981
12982 #if 0
12983                 /* Use physical addresses when talking to ISC hardware */
12984                 if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
12985                         /* XXX KDM use busdma */
12986                         rq->local = vtophys(tmp_ptr);
12987                 } else
12988                         rq->local = tmp_ptr;
12989 #else
12990                 KASSERT((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0,
12991                     ("HA does not support BUS_ADDR"));
12992                 rq->local = tmp_ptr;
12993 #endif
12994
12995                 tmp_ptr = (uint8_t *)remote_sglist[j].addr;
12996                 tmp_ptr += remote_used;
12997                 rq->remote = tmp_ptr;
12998
12999                 rq->callback = NULL;
13000
13001                 local_used += cur_len;
13002                 if (local_used >= local_sglist[i].len) {
13003                         i++;
13004                         local_used = 0;
13005                 }
13006
13007                 remote_used += cur_len;
13008                 if (remote_used >= remote_sglist[j].len) {
13009                         j++;
13010                         remote_used = 0;
13011                 }
13012                 total_used += cur_len;
13013
13014                 if (total_used >= io->scsiio.kern_data_len)
13015                         rq->callback = callback;
13016
13017 #if 0
13018                 printf("%s: %s: local %p remote %p size %d\n", __func__,
13019                        (command == CTL_HA_DT_CMD_WRITE) ? "WRITE" : "READ",
13020                        rq->local, rq->remote, rq->size);
13021 #endif
13022
13023                 isc_ret = ctl_dt_single(rq);
13024                 if (isc_ret > CTL_HA_STATUS_SUCCESS)
13025                         break;
13026         }
13027         if (isc_ret != CTL_HA_STATUS_WAIT) {
13028                 rq->ret = isc_ret;
13029                 callback(rq);
13030         }
13031
13032         return (0);
13033 }
13034
13035 static void
13036 ctl_datamove_remote_read(union ctl_io *io)
13037 {
13038         int retval;
13039         uint32_t i;
13040
13041         /*
13042          * This will send an error to the other controller in the case of a
13043          * failure.
13044          */
13045         retval = ctl_datamove_remote_sgl_setup(io);
13046         if (retval != 0)
13047                 return;
13048
13049         retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
13050                                           ctl_datamove_remote_read_cb);
13051         if (retval != 0) {
13052                 /*
13053                  * Make sure we free memory if there was an error..  The
13054                  * ctl_datamove_remote_xfer() function will send the
13055                  * datamove done message, or call the callback with an
13056                  * error if there is a problem.
13057                  */
13058                 for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13059                         free(io->io_hdr.local_sglist[i].addr, M_CTL);
13060                 free(io->io_hdr.remote_sglist, M_CTL);
13061                 io->io_hdr.remote_sglist = NULL;
13062                 io->io_hdr.local_sglist = NULL;
13063         }
13064 }
13065
13066 /*
13067  * Process a datamove request from the other controller.  This is used for
13068  * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
13069  * first.  Once that is complete, the data gets DMAed into the remote
13070  * controller's memory.  For reads, we DMA from the remote controller's
13071  * memory into our memory first, and then move it out to the FETD.
13072  */
13073 static void
13074 ctl_datamove_remote(union ctl_io *io)
13075 {
13076
13077         mtx_assert(&((struct ctl_softc *)CTL_SOFTC(io))->ctl_lock, MA_NOTOWNED);
13078
13079         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
13080                 ctl_failover_io(io, /*have_lock*/ 0);
13081                 return;
13082         }
13083
13084         /*
13085          * Note that we look for an aborted I/O here, but don't do some of
13086          * the other checks that ctl_datamove() normally does.
13087          * We don't need to run the datamove delay code, since that should
13088          * have been done if need be on the other controller.
13089          */
13090         if (io->io_hdr.flags & CTL_FLAG_ABORT) {
13091                 printf("%s: tag 0x%04x on (%u:%u:%u) aborted\n", __func__,
13092                        io->scsiio.tag_num, io->io_hdr.nexus.initid,
13093                        io->io_hdr.nexus.targ_port,
13094                        io->io_hdr.nexus.targ_lun);
13095                 io->io_hdr.port_status = 31338;
13096                 ctl_send_datamove_done(io, /*have_lock*/ 0);
13097                 return;
13098         }
13099
13100         if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT)
13101                 ctl_datamove_remote_write(io);
13102         else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN)
13103                 ctl_datamove_remote_read(io);
13104         else {
13105                 io->io_hdr.port_status = 31339;
13106                 ctl_send_datamove_done(io, /*have_lock*/ 0);
13107         }
13108 }
13109
13110 static void
13111 ctl_process_done(union ctl_io *io)
13112 {
13113         struct ctl_softc *softc = CTL_SOFTC(io);
13114         struct ctl_port *port = CTL_PORT(io);
13115         struct ctl_lun *lun = CTL_LUN(io);
13116         void (*fe_done)(union ctl_io *io);
13117         union ctl_ha_msg msg;
13118
13119         CTL_DEBUG_PRINT(("ctl_process_done\n"));
13120         fe_done = port->fe_done;
13121
13122 #ifdef CTL_TIME_IO
13123         if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
13124                 char str[256];
13125                 char path_str[64];
13126                 struct sbuf sb;
13127
13128                 ctl_scsi_path_string(io, path_str, sizeof(path_str));
13129                 sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
13130
13131                 sbuf_cat(&sb, path_str);
13132                 switch (io->io_hdr.io_type) {
13133                 case CTL_IO_SCSI:
13134                         ctl_scsi_command_string(&io->scsiio, NULL, &sb);
13135                         sbuf_printf(&sb, "\n");
13136                         sbuf_cat(&sb, path_str);
13137                         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
13138                                     io->scsiio.tag_num, io->scsiio.tag_type);
13139                         break;
13140                 case CTL_IO_TASK:
13141                         sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
13142                                     "Tag Type: %d\n", io->taskio.task_action,
13143                                     io->taskio.tag_num, io->taskio.tag_type);
13144                         break;
13145                 default:
13146                         panic("%s: Invalid CTL I/O type %d\n",
13147                             __func__, io->io_hdr.io_type);
13148                 }
13149                 sbuf_cat(&sb, path_str);
13150                 sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
13151                             (intmax_t)time_uptime - io->io_hdr.start_time);
13152                 sbuf_finish(&sb);
13153                 printf("%s", sbuf_data(&sb));
13154         }
13155 #endif /* CTL_TIME_IO */
13156
13157         switch (io->io_hdr.io_type) {
13158         case CTL_IO_SCSI:
13159                 break;
13160         case CTL_IO_TASK:
13161                 if (ctl_debug & CTL_DEBUG_INFO)
13162                         ctl_io_error_print(io, NULL);
13163                 fe_done(io);
13164                 return;
13165         default:
13166                 panic("%s: Invalid CTL I/O type %d\n",
13167                     __func__, io->io_hdr.io_type);
13168         }
13169
13170         if (lun == NULL) {
13171                 CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
13172                                  io->io_hdr.nexus.targ_mapped_lun));
13173                 goto bailout;
13174         }
13175
13176         mtx_lock(&lun->lun_lock);
13177
13178         /*
13179          * Check to see if we have any informational exception and status
13180          * of this command can be modified to report it in form of either
13181          * RECOVERED ERROR or NO SENSE, depending on MRIE mode page field.
13182          */
13183         if (lun->ie_reported == 0 && lun->ie_asc != 0 &&
13184             io->io_hdr.status == CTL_SUCCESS &&
13185             (io->io_hdr.flags & CTL_FLAG_STATUS_SENT) == 0) {
13186                 uint8_t mrie = lun->MODE_IE.mrie;
13187                 uint8_t per = ((lun->MODE_RWER.byte3 & SMS_RWER_PER) ||
13188                     (lun->MODE_VER.byte3 & SMS_VER_PER));
13189                 if (((mrie == SIEP_MRIE_REC_COND && per) ||
13190                      mrie == SIEP_MRIE_REC_UNCOND ||
13191                      mrie == SIEP_MRIE_NO_SENSE) &&
13192                     (ctl_get_cmd_entry(&io->scsiio, NULL)->flags &
13193                      CTL_CMD_FLAG_NO_SENSE) == 0) {
13194                         ctl_set_sense(&io->scsiio,
13195                               /*current_error*/ 1,
13196                               /*sense_key*/ (mrie == SIEP_MRIE_NO_SENSE) ?
13197                                 SSD_KEY_NO_SENSE : SSD_KEY_RECOVERED_ERROR,
13198                               /*asc*/ lun->ie_asc,
13199                               /*ascq*/ lun->ie_ascq,
13200                               SSD_ELEM_NONE);
13201                         lun->ie_reported = 1;
13202                 }
13203         } else if (lun->ie_reported < 0)
13204                 lun->ie_reported = 0;
13205
13206         /*
13207          * Check to see if we have any errors to inject here.  We only
13208          * inject errors for commands that don't already have errors set.
13209          */
13210         if (!STAILQ_EMPTY(&lun->error_list) &&
13211             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS) &&
13212             ((io->io_hdr.flags & CTL_FLAG_STATUS_SENT) == 0))
13213                 ctl_inject_error(lun, io);
13214
13215         /*
13216          * XXX KDM how do we treat commands that aren't completed
13217          * successfully?
13218          *
13219          * XXX KDM should we also track I/O latency?
13220          */
13221         if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS &&
13222             io->io_hdr.io_type == CTL_IO_SCSI) {
13223                 int type;
13224 #ifdef CTL_TIME_IO
13225                 struct bintime bt;
13226
13227                 getbinuptime(&bt);
13228                 bintime_sub(&bt, &io->io_hdr.start_bt);
13229 #endif
13230                 if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13231                     CTL_FLAG_DATA_IN)
13232                         type = CTL_STATS_READ;
13233                 else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13234                     CTL_FLAG_DATA_OUT)
13235                         type = CTL_STATS_WRITE;
13236                 else
13237                         type = CTL_STATS_NO_IO;
13238
13239 #ifdef CTL_LEGACY_STATS
13240                 uint32_t targ_port = port->targ_port;
13241                 lun->legacy_stats.ports[targ_port].bytes[type] +=
13242                     io->scsiio.kern_total_len;
13243                 lun->legacy_stats.ports[targ_port].operations[type] ++;
13244                 lun->legacy_stats.ports[targ_port].num_dmas[type] +=
13245                     io->io_hdr.num_dmas;
13246 #ifdef CTL_TIME_IO
13247                 bintime_add(&lun->legacy_stats.ports[targ_port].dma_time[type],
13248                    &io->io_hdr.dma_bt);
13249                 bintime_add(&lun->legacy_stats.ports[targ_port].time[type],
13250                     &bt);
13251 #endif
13252 #endif /* CTL_LEGACY_STATS */
13253
13254                 lun->stats.bytes[type] += io->scsiio.kern_total_len;
13255                 lun->stats.operations[type] ++;
13256                 lun->stats.dmas[type] += io->io_hdr.num_dmas;
13257 #ifdef CTL_TIME_IO
13258                 bintime_add(&lun->stats.dma_time[type], &io->io_hdr.dma_bt);
13259                 bintime_add(&lun->stats.time[type], &bt);
13260 #endif
13261
13262                 mtx_lock(&port->port_lock);
13263                 port->stats.bytes[type] += io->scsiio.kern_total_len;
13264                 port->stats.operations[type] ++;
13265                 port->stats.dmas[type] += io->io_hdr.num_dmas;
13266 #ifdef CTL_TIME_IO
13267                 bintime_add(&port->stats.dma_time[type], &io->io_hdr.dma_bt);
13268                 bintime_add(&port->stats.time[type], &bt);
13269 #endif
13270                 mtx_unlock(&port->port_lock);
13271         }
13272
13273         /*
13274          * Remove this from the OOA queue.
13275          */
13276         TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
13277 #ifdef CTL_TIME_IO
13278         if (TAILQ_EMPTY(&lun->ooa_queue))
13279                 lun->last_busy = getsbinuptime();
13280 #endif
13281
13282         /*
13283          * Run through the blocked queue on this LUN and see if anything
13284          * has become unblocked, now that this transaction is done.
13285          */
13286         ctl_check_blocked(lun);
13287
13288         /*
13289          * If the LUN has been invalidated, free it if there is nothing
13290          * left on its OOA queue.
13291          */
13292         if ((lun->flags & CTL_LUN_INVALID)
13293          && TAILQ_EMPTY(&lun->ooa_queue)) {
13294                 mtx_unlock(&lun->lun_lock);
13295                 mtx_lock(&softc->ctl_lock);
13296                 ctl_free_lun(lun);
13297                 mtx_unlock(&softc->ctl_lock);
13298         } else
13299                 mtx_unlock(&lun->lun_lock);
13300
13301 bailout:
13302
13303         /*
13304          * If this command has been aborted, make sure we set the status
13305          * properly.  The FETD is responsible for freeing the I/O and doing
13306          * whatever it needs to do to clean up its state.
13307          */
13308         if (io->io_hdr.flags & CTL_FLAG_ABORT)
13309                 ctl_set_task_aborted(&io->scsiio);
13310
13311         /*
13312          * If enabled, print command error status.
13313          */
13314         if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS &&
13315             (ctl_debug & CTL_DEBUG_INFO) != 0)
13316                 ctl_io_error_print(io, NULL);
13317
13318         /*
13319          * Tell the FETD or the other shelf controller we're done with this
13320          * command.  Note that only SCSI commands get to this point.  Task
13321          * management commands are completed above.
13322          */
13323         if ((softc->ha_mode != CTL_HA_MODE_XFER) &&
13324             (io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)) {
13325                 memset(&msg, 0, sizeof(msg));
13326                 msg.hdr.msg_type = CTL_MSG_FINISH_IO;
13327                 msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
13328                 msg.hdr.nexus = io->io_hdr.nexus;
13329                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13330                     sizeof(msg.scsi) - sizeof(msg.scsi.sense_data),
13331                     M_WAITOK);
13332         }
13333
13334         fe_done(io);
13335 }
13336
13337 #ifdef CTL_WITH_CA
13338 /*
13339  * Front end should call this if it doesn't do autosense.  When the request
13340  * sense comes back in from the initiator, we'll dequeue this and send it.
13341  */
13342 int
13343 ctl_queue_sense(union ctl_io *io)
13344 {
13345         struct ctl_softc *softc = CTL_SOFTC(io);
13346         struct ctl_port *port = CTL_PORT(io);
13347         struct ctl_lun *lun;
13348         uint32_t initidx, targ_lun;
13349
13350         CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
13351
13352         targ_lun = ctl_lun_map_from_port(port, io->io_hdr.nexus.targ_lun);
13353
13354         /*
13355          * LUN lookup will likely move to the ctl_work_thread() once we
13356          * have our new queueing infrastructure (that doesn't put things on
13357          * a per-LUN queue initially).  That is so that we can handle
13358          * things like an INQUIRY to a LUN that we don't have enabled.  We
13359          * can't deal with that right now.
13360          * If we don't have a LUN for this, just toss the sense information.
13361          */
13362         mtx_lock(&softc->ctl_lock);
13363         if (targ_lun >= CTL_MAX_LUNS ||
13364             (lun = softc->ctl_luns[targ_lun]) == NULL) {
13365                 mtx_unlock(&softc->ctl_lock);
13366                 goto bailout;
13367         }
13368         mtx_lock(&lun->lun_lock);
13369         mtx_unlock(&softc->ctl_lock);
13370
13371         /*
13372          * Already have CA set for this LUN...toss the sense information.
13373          */
13374         initidx = ctl_get_initindex(&io->io_hdr.nexus);
13375         if (ctl_is_set(lun->have_ca, initidx)) {
13376                 mtx_unlock(&lun->lun_lock);
13377                 goto bailout;
13378         }
13379
13380         memcpy(&lun->pending_sense[initidx], &io->scsiio.sense_data,
13381                MIN(sizeof(lun->pending_sense[initidx]),
13382                sizeof(io->scsiio.sense_data)));
13383         ctl_set_mask(lun->have_ca, initidx);
13384         mtx_unlock(&lun->lun_lock);
13385
13386 bailout:
13387         ctl_free_io(io);
13388         return (CTL_RETVAL_COMPLETE);
13389 }
13390 #endif
13391
13392 /*
13393  * Primary command inlet from frontend ports.  All SCSI and task I/O
13394  * requests must go through this function.
13395  */
13396 int
13397 ctl_queue(union ctl_io *io)
13398 {
13399         struct ctl_port *port = CTL_PORT(io);
13400
13401         CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
13402
13403 #ifdef CTL_TIME_IO
13404         io->io_hdr.start_time = time_uptime;
13405         getbinuptime(&io->io_hdr.start_bt);
13406 #endif /* CTL_TIME_IO */
13407
13408         /* Map FE-specific LUN ID into global one. */
13409         io->io_hdr.nexus.targ_mapped_lun =
13410             ctl_lun_map_from_port(port, io->io_hdr.nexus.targ_lun);
13411
13412         switch (io->io_hdr.io_type) {
13413         case CTL_IO_SCSI:
13414         case CTL_IO_TASK:
13415                 if (ctl_debug & CTL_DEBUG_CDB)
13416                         ctl_io_print(io);
13417                 ctl_enqueue_incoming(io);
13418                 break;
13419         default:
13420                 printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
13421                 return (EINVAL);
13422         }
13423
13424         return (CTL_RETVAL_COMPLETE);
13425 }
13426
13427 #ifdef CTL_IO_DELAY
13428 static void
13429 ctl_done_timer_wakeup(void *arg)
13430 {
13431         union ctl_io *io;
13432
13433         io = (union ctl_io *)arg;
13434         ctl_done(io);
13435 }
13436 #endif /* CTL_IO_DELAY */
13437
13438 void
13439 ctl_serseq_done(union ctl_io *io)
13440 {
13441         struct ctl_lun *lun = CTL_LUN(io);;
13442
13443         if (lun->be_lun == NULL ||
13444             lun->be_lun->serseq == CTL_LUN_SERSEQ_OFF)
13445                 return;
13446         mtx_lock(&lun->lun_lock);
13447         io->io_hdr.flags |= CTL_FLAG_SERSEQ_DONE;
13448         ctl_check_blocked(lun);
13449         mtx_unlock(&lun->lun_lock);
13450 }
13451
13452 void
13453 ctl_done(union ctl_io *io)
13454 {
13455
13456         /*
13457          * Enable this to catch duplicate completion issues.
13458          */
13459 #if 0
13460         if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
13461                 printf("%s: type %d msg %d cdb %x iptl: "
13462                        "%u:%u:%u tag 0x%04x "
13463                        "flag %#x status %x\n",
13464                         __func__,
13465                         io->io_hdr.io_type,
13466                         io->io_hdr.msg_type,
13467                         io->scsiio.cdb[0],
13468                         io->io_hdr.nexus.initid,
13469                         io->io_hdr.nexus.targ_port,
13470                         io->io_hdr.nexus.targ_lun,
13471                         (io->io_hdr.io_type ==
13472                         CTL_IO_TASK) ?
13473                         io->taskio.tag_num :
13474                         io->scsiio.tag_num,
13475                         io->io_hdr.flags,
13476                         io->io_hdr.status);
13477         } else
13478                 io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
13479 #endif
13480
13481         /*
13482          * This is an internal copy of an I/O, and should not go through
13483          * the normal done processing logic.
13484          */
13485         if (io->io_hdr.flags & CTL_FLAG_INT_COPY)
13486                 return;
13487
13488 #ifdef CTL_IO_DELAY
13489         if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
13490                 io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
13491         } else {
13492                 struct ctl_lun *lun = CTL_LUN(io);
13493
13494                 if ((lun != NULL)
13495                  && (lun->delay_info.done_delay > 0)) {
13496
13497                         callout_init(&io->io_hdr.delay_callout, /*mpsafe*/ 1);
13498                         io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
13499                         callout_reset(&io->io_hdr.delay_callout,
13500                                       lun->delay_info.done_delay * hz,
13501                                       ctl_done_timer_wakeup, io);
13502                         if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
13503                                 lun->delay_info.done_delay = 0;
13504                         return;
13505                 }
13506         }
13507 #endif /* CTL_IO_DELAY */
13508
13509         ctl_enqueue_done(io);
13510 }
13511
13512 static void
13513 ctl_work_thread(void *arg)
13514 {
13515         struct ctl_thread *thr = (struct ctl_thread *)arg;
13516         struct ctl_softc *softc = thr->ctl_softc;
13517         union ctl_io *io;
13518         int retval;
13519
13520         CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
13521
13522         for (;;) {
13523                 /*
13524                  * We handle the queues in this order:
13525                  * - ISC
13526                  * - done queue (to free up resources, unblock other commands)
13527                  * - RtR queue
13528                  * - incoming queue
13529                  *
13530                  * If those queues are empty, we break out of the loop and
13531                  * go to sleep.
13532                  */
13533                 mtx_lock(&thr->queue_lock);
13534                 io = (union ctl_io *)STAILQ_FIRST(&thr->isc_queue);
13535                 if (io != NULL) {
13536                         STAILQ_REMOVE_HEAD(&thr->isc_queue, links);
13537                         mtx_unlock(&thr->queue_lock);
13538                         ctl_handle_isc(io);
13539                         continue;
13540                 }
13541                 io = (union ctl_io *)STAILQ_FIRST(&thr->done_queue);
13542                 if (io != NULL) {
13543                         STAILQ_REMOVE_HEAD(&thr->done_queue, links);
13544                         /* clear any blocked commands, call fe_done */
13545                         mtx_unlock(&thr->queue_lock);
13546                         ctl_process_done(io);
13547                         continue;
13548                 }
13549                 io = (union ctl_io *)STAILQ_FIRST(&thr->incoming_queue);
13550                 if (io != NULL) {
13551                         STAILQ_REMOVE_HEAD(&thr->incoming_queue, links);
13552                         mtx_unlock(&thr->queue_lock);
13553                         if (io->io_hdr.io_type == CTL_IO_TASK)
13554                                 ctl_run_task(io);
13555                         else
13556                                 ctl_scsiio_precheck(softc, &io->scsiio);
13557                         continue;
13558                 }
13559                 io = (union ctl_io *)STAILQ_FIRST(&thr->rtr_queue);
13560                 if (io != NULL) {
13561                         STAILQ_REMOVE_HEAD(&thr->rtr_queue, links);
13562                         mtx_unlock(&thr->queue_lock);
13563                         retval = ctl_scsiio(&io->scsiio);
13564                         if (retval != CTL_RETVAL_COMPLETE)
13565                                 CTL_DEBUG_PRINT(("ctl_scsiio failed\n"));
13566                         continue;
13567                 }
13568
13569                 /* Sleep until we have something to do. */
13570                 mtx_sleep(thr, &thr->queue_lock, PDROP | PRIBIO, "-", 0);
13571         }
13572 }
13573
13574 static void
13575 ctl_lun_thread(void *arg)
13576 {
13577         struct ctl_softc *softc = (struct ctl_softc *)arg;
13578         struct ctl_be_lun *be_lun;
13579
13580         CTL_DEBUG_PRINT(("ctl_lun_thread starting\n"));
13581
13582         for (;;) {
13583                 mtx_lock(&softc->ctl_lock);
13584                 be_lun = STAILQ_FIRST(&softc->pending_lun_queue);
13585                 if (be_lun != NULL) {
13586                         STAILQ_REMOVE_HEAD(&softc->pending_lun_queue, links);
13587                         mtx_unlock(&softc->ctl_lock);
13588                         ctl_create_lun(be_lun);
13589                         continue;
13590                 }
13591
13592                 /* Sleep until we have something to do. */
13593                 mtx_sleep(&softc->pending_lun_queue, &softc->ctl_lock,
13594                     PDROP | PRIBIO, "-", 0);
13595         }
13596 }
13597
13598 static void
13599 ctl_thresh_thread(void *arg)
13600 {
13601         struct ctl_softc *softc = (struct ctl_softc *)arg;
13602         struct ctl_lun *lun;
13603         struct ctl_logical_block_provisioning_page *page;
13604         const char *attr;
13605         union ctl_ha_msg msg;
13606         uint64_t thres, val;
13607         int i, e, set;
13608
13609         CTL_DEBUG_PRINT(("ctl_thresh_thread starting\n"));
13610
13611         for (;;) {
13612                 mtx_lock(&softc->ctl_lock);
13613                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
13614                         if ((lun->flags & CTL_LUN_DISABLED) ||
13615                             (lun->flags & CTL_LUN_NO_MEDIA) ||
13616                             lun->backend->lun_attr == NULL)
13617                                 continue;
13618                         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
13619                             softc->ha_mode == CTL_HA_MODE_XFER)
13620                                 continue;
13621                         if ((lun->MODE_RWER.byte8 & SMS_RWER_LBPERE) == 0)
13622                                 continue;
13623                         e = 0;
13624                         page = &lun->MODE_LBP;
13625                         for (i = 0; i < CTL_NUM_LBP_THRESH; i++) {
13626                                 if ((page->descr[i].flags & SLBPPD_ENABLED) == 0)
13627                                         continue;
13628                                 thres = scsi_4btoul(page->descr[i].count);
13629                                 thres <<= CTL_LBP_EXPONENT;
13630                                 switch (page->descr[i].resource) {
13631                                 case 0x01:
13632                                         attr = "blocksavail";
13633                                         break;
13634                                 case 0x02:
13635                                         attr = "blocksused";
13636                                         break;
13637                                 case 0xf1:
13638                                         attr = "poolblocksavail";
13639                                         break;
13640                                 case 0xf2:
13641                                         attr = "poolblocksused";
13642                                         break;
13643                                 default:
13644                                         continue;
13645                                 }
13646                                 mtx_unlock(&softc->ctl_lock); // XXX
13647                                 val = lun->backend->lun_attr(
13648                                     lun->be_lun->be_lun, attr);
13649                                 mtx_lock(&softc->ctl_lock);
13650                                 if (val == UINT64_MAX)
13651                                         continue;
13652                                 if ((page->descr[i].flags & SLBPPD_ARMING_MASK)
13653                                     == SLBPPD_ARMING_INC)
13654                                         e = (val >= thres);
13655                                 else
13656                                         e = (val <= thres);
13657                                 if (e)
13658                                         break;
13659                         }
13660                         mtx_lock(&lun->lun_lock);
13661                         if (e) {
13662                                 scsi_u64to8b((uint8_t *)&page->descr[i] -
13663                                     (uint8_t *)page, lun->ua_tpt_info);
13664                                 if (lun->lasttpt == 0 ||
13665                                     time_uptime - lun->lasttpt >= CTL_LBP_UA_PERIOD) {
13666                                         lun->lasttpt = time_uptime;
13667                                         ctl_est_ua_all(lun, -1, CTL_UA_THIN_PROV_THRES);
13668                                         set = 1;
13669                                 } else
13670                                         set = 0;
13671                         } else {
13672                                 lun->lasttpt = 0;
13673                                 ctl_clr_ua_all(lun, -1, CTL_UA_THIN_PROV_THRES);
13674                                 set = -1;
13675                         }
13676                         mtx_unlock(&lun->lun_lock);
13677                         if (set != 0 &&
13678                             lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
13679                                 /* Send msg to other side. */
13680                                 bzero(&msg.ua, sizeof(msg.ua));
13681                                 msg.hdr.msg_type = CTL_MSG_UA;
13682                                 msg.hdr.nexus.initid = -1;
13683                                 msg.hdr.nexus.targ_port = -1;
13684                                 msg.hdr.nexus.targ_lun = lun->lun;
13685                                 msg.hdr.nexus.targ_mapped_lun = lun->lun;
13686                                 msg.ua.ua_all = 1;
13687                                 msg.ua.ua_set = (set > 0);
13688                                 msg.ua.ua_type = CTL_UA_THIN_PROV_THRES;
13689                                 memcpy(msg.ua.ua_info, lun->ua_tpt_info, 8);
13690                                 mtx_unlock(&softc->ctl_lock); // XXX
13691                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13692                                     sizeof(msg.ua), M_WAITOK);
13693                                 mtx_lock(&softc->ctl_lock);
13694                         }
13695                 }
13696                 mtx_unlock(&softc->ctl_lock);
13697                 pause("-", CTL_LBP_PERIOD * hz);
13698         }
13699 }
13700
13701 static void
13702 ctl_enqueue_incoming(union ctl_io *io)
13703 {
13704         struct ctl_softc *softc = CTL_SOFTC(io);
13705         struct ctl_thread *thr;
13706         u_int idx;
13707
13708         idx = (io->io_hdr.nexus.targ_port * 127 +
13709                io->io_hdr.nexus.initid) % worker_threads;
13710         thr = &softc->threads[idx];
13711         mtx_lock(&thr->queue_lock);
13712         STAILQ_INSERT_TAIL(&thr->incoming_queue, &io->io_hdr, links);
13713         mtx_unlock(&thr->queue_lock);
13714         wakeup(thr);
13715 }
13716
13717 static void
13718 ctl_enqueue_rtr(union ctl_io *io)
13719 {
13720         struct ctl_softc *softc = CTL_SOFTC(io);
13721         struct ctl_thread *thr;
13722
13723         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13724         mtx_lock(&thr->queue_lock);
13725         STAILQ_INSERT_TAIL(&thr->rtr_queue, &io->io_hdr, links);
13726         mtx_unlock(&thr->queue_lock);
13727         wakeup(thr);
13728 }
13729
13730 static void
13731 ctl_enqueue_done(union ctl_io *io)
13732 {
13733         struct ctl_softc *softc = CTL_SOFTC(io);
13734         struct ctl_thread *thr;
13735
13736         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13737         mtx_lock(&thr->queue_lock);
13738         STAILQ_INSERT_TAIL(&thr->done_queue, &io->io_hdr, links);
13739         mtx_unlock(&thr->queue_lock);
13740         wakeup(thr);
13741 }
13742
13743 static void
13744 ctl_enqueue_isc(union ctl_io *io)
13745 {
13746         struct ctl_softc *softc = CTL_SOFTC(io);
13747         struct ctl_thread *thr;
13748
13749         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13750         mtx_lock(&thr->queue_lock);
13751         STAILQ_INSERT_TAIL(&thr->isc_queue, &io->io_hdr, links);
13752         mtx_unlock(&thr->queue_lock);
13753         wakeup(thr);
13754 }
13755
13756 /*
13757  *  vim: ts=8
13758  */