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) 2015 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*/0
 282 };
 283
 284 const static struct scsi_info_exceptions_page ie_page_default = {
 285         /*page_code*/SMS_INFO_EXCEPTIONS_PAGE,
 286         /*page_length*/sizeof(struct scsi_info_exceptions_page) - 2,
 287         /*info_flags*/SIEP_FLAGS_EWASC,
 288         /*mrie*/SIEP_MRIE_NO,
 289         /*interval_timer*/{0, 0, 0, 0},
 290         /*report_count*/{0, 0, 0, 1}
 291 };
 292
 293 const static struct scsi_info_exceptions_page ie_page_changeable = {
 294         /*page_code*/SMS_INFO_EXCEPTIONS_PAGE,
 295         /*page_length*/sizeof(struct scsi_info_exceptions_page) - 2,
 296         /*info_flags*/SIEP_FLAGS_EWASC | SIEP_FLAGS_DEXCPT | SIEP_FLAGS_TEST |
 297             SIEP_FLAGS_LOGERR,
 298         /*mrie*/0x0f,
 299         /*interval_timer*/{0xff, 0xff, 0xff, 0xff},
 300         /*report_count*/{0xff, 0xff, 0xff, 0xff}
 301 };
 302
 303 #define CTL_LBPM_LEN    (sizeof(struct ctl_logical_block_provisioning_page) - 4)
 304
 305 const static struct ctl_logical_block_provisioning_page lbp_page_default = {{
 306         /*page_code*/SMS_INFO_EXCEPTIONS_PAGE | SMPH_SPF,
 307         /*subpage_code*/0x02,
 308         /*page_length*/{CTL_LBPM_LEN >> 8, CTL_LBPM_LEN},
 309         /*flags*/0,
 310         /*reserved*/{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
 311         /*descr*/{}},
 312         {{/*flags*/0,
 313           /*resource*/0x01,
 314           /*reserved*/{0, 0},
 315           /*count*/{0, 0, 0, 0}},
 316          {/*flags*/0,
 317           /*resource*/0x02,
 318           /*reserved*/{0, 0},
 319           /*count*/{0, 0, 0, 0}},
 320          {/*flags*/0,
 321           /*resource*/0xf1,
 322           /*reserved*/{0, 0},
 323           /*count*/{0, 0, 0, 0}},
 324          {/*flags*/0,
 325           /*resource*/0xf2,
 326           /*reserved*/{0, 0},
 327           /*count*/{0, 0, 0, 0}}
 328         }
 329 };
 330
 331 const static struct ctl_logical_block_provisioning_page lbp_page_changeable = {{
 332         /*page_code*/SMS_INFO_EXCEPTIONS_PAGE | SMPH_SPF,
 333         /*subpage_code*/0x02,
 334         /*page_length*/{CTL_LBPM_LEN >> 8, CTL_LBPM_LEN},
 335         /*flags*/SLBPP_SITUA,
 336         /*reserved*/{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
 337         /*descr*/{}},
 338         {{/*flags*/0,
 339           /*resource*/0,
 340           /*reserved*/{0, 0},
 341           /*count*/{0, 0, 0, 0}},
 342          {/*flags*/0,
 343           /*resource*/0,
 344           /*reserved*/{0, 0},
 345           /*count*/{0, 0, 0, 0}},
 346          {/*flags*/0,
 347           /*resource*/0,
 348           /*reserved*/{0, 0},
 349           /*count*/{0, 0, 0, 0}},
 350          {/*flags*/0,
 351           /*resource*/0,
 352           /*reserved*/{0, 0},
 353           /*count*/{0, 0, 0, 0}}
 354         }
 355 };
 356
 357 const static struct scsi_cddvd_capabilities_page cddvd_page_default = {
 358         /*page_code*/SMS_CDDVD_CAPS_PAGE,
 359         /*page_length*/sizeof(struct scsi_cddvd_capabilities_page) - 2,
 360         /*caps1*/0x3f,
 361         /*caps2*/0x00,
 362         /*caps3*/0xf0,
 363         /*caps4*/0x00,
 364         /*caps5*/0x29,
 365         /*caps6*/0x00,
 366         /*obsolete*/{0, 0},
 367         /*nvol_levels*/{0, 0},
 368         /*buffer_size*/{8, 0},
 369         /*obsolete2*/{0, 0},
 370         /*reserved*/0,
 371         /*digital*/0,
 372         /*obsolete3*/0,
 373         /*copy_management*/0,
 374         /*reserved2*/0,
 375         /*rotation_control*/0,
 376         /*cur_write_speed*/0,
 377         /*num_speed_descr*/0,
 378 };
 379
 380 const static struct scsi_cddvd_capabilities_page cddvd_page_changeable = {
 381         /*page_code*/SMS_CDDVD_CAPS_PAGE,
 382         /*page_length*/sizeof(struct scsi_cddvd_capabilities_page) - 2,
 383         /*caps1*/0,
 384         /*caps2*/0,
 385         /*caps3*/0,
 386         /*caps4*/0,
 387         /*caps5*/0,
 388         /*caps6*/0,
 389         /*obsolete*/{0, 0},
 390         /*nvol_levels*/{0, 0},
 391         /*buffer_size*/{0, 0},
 392         /*obsolete2*/{0, 0},
 393         /*reserved*/0,
 394         /*digital*/0,
 395         /*obsolete3*/0,
 396         /*copy_management*/0,
 397         /*reserved2*/0,
 398         /*rotation_control*/0,
 399         /*cur_write_speed*/0,
 400         /*num_speed_descr*/0,
 401 };
 402
 403 SYSCTL_NODE(_kern_cam, OID_AUTO, ctl, CTLFLAG_RD, 0, "CAM Target Layer");
 404 static int worker_threads = -1;
 405 TUNABLE_INT("kern.cam.ctl.worker_threads", &worker_threads);
 406 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, worker_threads, CTLFLAG_RDTUN,
 407     &worker_threads, 1, "Number of worker threads");
 408 static int ctl_debug = CTL_DEBUG_NONE;
 409 TUNABLE_INT("kern.cam.ctl.debug", &ctl_debug);
 410 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, debug, CTLFLAG_RWTUN,
 411     &ctl_debug, 0, "Enabled debug flags");
 412
 413 /*
 414  * Supported pages (0x00), Serial number (0x80), Device ID (0x83),
 415  * Extended INQUIRY Data (0x86), Mode Page Policy (0x87),
 416  * SCSI Ports (0x88), Third-party Copy (0x8F), Block limits (0xB0),
 417  * Block Device Characteristics (0xB1) and Logical Block Provisioning (0xB2)
 418  */
 419 #define SCSI_EVPD_NUM_SUPPORTED_PAGES   10
 420
 421 static void ctl_isc_event_handler(ctl_ha_channel chanel, ctl_ha_event event,
 422                                   int param);
 423 static void ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest);
 424 static void ctl_copy_sense_data_back(union ctl_io *src, union ctl_ha_msg *dest);
 425 static int ctl_init(void);
 426 void ctl_shutdown(void);
 427 static int ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td);
 428 static int ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td);
 429 static int ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio);
 430 static void ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
 431                               struct ctl_ooa *ooa_hdr,
 432                               struct ctl_ooa_entry *kern_entries);
 433 static int ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
 434                      struct thread *td);
 435 static int ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
 436                          struct ctl_be_lun *be_lun);
 437 static int ctl_free_lun(struct ctl_lun *lun);
 438 static void ctl_create_lun(struct ctl_be_lun *be_lun);
 439 static struct ctl_port * ctl_io_port(struct ctl_io_hdr *io_hdr);
 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_ha_msg *msg);
 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;
 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         lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
 576         memset(&msg.dt, 0, sizeof(msg.dt));
 577         msg.hdr.msg_type = CTL_MSG_DATAMOVE;
 578         msg.hdr.original_sc = io->io_hdr.original_sc;
 579         msg.hdr.serializing_sc = io;
 580         msg.hdr.nexus = io->io_hdr.nexus;
 581         msg.hdr.status = io->io_hdr.status;
 582         msg.dt.flags = io->io_hdr.flags;
 583
 584         /*
 585          * We convert everything into a S/G list here.  We can't
 586          * pass by reference, only by value between controllers.
 587          * So we can't pass a pointer to the S/G list, only as many
 588          * S/G entries as we can fit in here.  If it's possible for
 589          * us to get more than CTL_HA_MAX_SG_ENTRIES S/G entries,
 590          * then we need to break this up into multiple transfers.
 591          */
 592         if (io->scsiio.kern_sg_entries == 0) {
 593                 msg.dt.kern_sg_entries = 1;
 594 #if 0
 595                 if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
 596                         msg.dt.sg_list[0].addr = io->scsiio.kern_data_ptr;
 597                 } else {
 598                         /* XXX KDM use busdma here! */
 599                         msg.dt.sg_list[0].addr =
 600                             (void *)vtophys(io->scsiio.kern_data_ptr);
 601                 }
 602 #else
 603                 KASSERT((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0,
 604                     ("HA does not support BUS_ADDR"));
 605                 msg.dt.sg_list[0].addr = io->scsiio.kern_data_ptr;
 606 #endif
 607                 msg.dt.sg_list[0].len = io->scsiio.kern_data_len;
 608                 do_sg_copy = 0;
 609         } else {
 610                 msg.dt.kern_sg_entries = io->scsiio.kern_sg_entries;
 611                 do_sg_copy = 1;
 612         }
 613
 614         msg.dt.kern_data_len = io->scsiio.kern_data_len;
 615         msg.dt.kern_total_len = io->scsiio.kern_total_len;
 616         msg.dt.kern_data_resid = io->scsiio.kern_data_resid;
 617         msg.dt.kern_rel_offset = io->scsiio.kern_rel_offset;
 618         msg.dt.sg_sequence = 0;
 619
 620         /*
 621          * Loop until we've sent all of the S/G entries.  On the
 622          * other end, we'll recompose these S/G entries into one
 623          * contiguous list before processing.
 624          */
 625         for (sg_entries_sent = 0; sg_entries_sent < msg.dt.kern_sg_entries;
 626             msg.dt.sg_sequence++) {
 627                 msg.dt.cur_sg_entries = MIN((sizeof(msg.dt.sg_list) /
 628                     sizeof(msg.dt.sg_list[0])),
 629                     msg.dt.kern_sg_entries - sg_entries_sent);
 630                 if (do_sg_copy != 0) {
 631                         sgl = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr;
 632                         for (i = sg_entries_sent, j = 0;
 633                              i < msg.dt.cur_sg_entries; i++, j++) {
 634 #if 0
 635                                 if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
 636                                         msg.dt.sg_list[j].addr = sgl[i].addr;
 637                                 } else {
 638                                         /* XXX KDM use busdma here! */
 639                                         msg.dt.sg_list[j].addr =
 640                                             (void *)vtophys(sgl[i].addr);
 641                                 }
 642 #else
 643                                 KASSERT((io->io_hdr.flags &
 644                                     CTL_FLAG_BUS_ADDR) == 0,
 645                                     ("HA does not support BUS_ADDR"));
 646                                 msg.dt.sg_list[j].addr = sgl[i].addr;
 647 #endif
 648                                 msg.dt.sg_list[j].len = sgl[i].len;
 649                         }
 650                 }
 651
 652                 sg_entries_sent += msg.dt.cur_sg_entries;
 653                 msg.dt.sg_last = (sg_entries_sent >= msg.dt.kern_sg_entries);
 654                 if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
 655                     sizeof(msg.dt) - sizeof(msg.dt.sg_list) +
 656                     sizeof(struct ctl_sg_entry) * msg.dt.cur_sg_entries,
 657                     M_WAITOK) > CTL_HA_STATUS_SUCCESS) {
 658                         io->io_hdr.port_status = 31341;
 659                         io->scsiio.be_move_done(io);
 660                         return;
 661                 }
 662                 msg.dt.sent_sg_entries = sg_entries_sent;
 663         }
 664
 665         /*
 666          * Officially handover the request from us to peer.
 667          * If failover has just happened, then we must return error.
 668          * If failover happen just after, then it is not our problem.
 669          */
 670         if (lun)
 671                 mtx_lock(&lun->lun_lock);
 672         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
 673                 if (lun)
 674                         mtx_unlock(&lun->lun_lock);
 675                 io->io_hdr.port_status = 31342;
 676                 io->scsiio.be_move_done(io);
 677                 return;
 678         }
 679         io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
 680         io->io_hdr.flags |= CTL_FLAG_DMA_INPROG;
 681         if (lun)
 682                 mtx_unlock(&lun->lun_lock);
 683 }
 684
 685 static void
 686 ctl_ha_done(union ctl_io *io)
 687 {
 688         union ctl_ha_msg msg;
 689
 690         if (io->io_hdr.io_type == CTL_IO_SCSI) {
 691                 memset(&msg, 0, sizeof(msg));
 692                 msg.hdr.msg_type = CTL_MSG_FINISH_IO;
 693                 msg.hdr.original_sc = io->io_hdr.original_sc;
 694                 msg.hdr.nexus = io->io_hdr.nexus;
 695                 msg.hdr.status = io->io_hdr.status;
 696                 msg.scsi.scsi_status = io->scsiio.scsi_status;
 697                 msg.scsi.tag_num = io->scsiio.tag_num;
 698                 msg.scsi.tag_type = io->scsiio.tag_type;
 699                 msg.scsi.sense_len = io->scsiio.sense_len;
 700                 msg.scsi.sense_residual = io->scsiio.sense_residual;
 701                 msg.scsi.residual = io->scsiio.residual;
 702                 memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
 703                     io->scsiio.sense_len);
 704                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
 705                     sizeof(msg.scsi) - sizeof(msg.scsi.sense_data) +
 706                     msg.scsi.sense_len, M_WAITOK);
 707         }
 708         ctl_free_io(io);
 709 }
 710
 711 static void
 712 ctl_isc_handler_finish_xfer(struct ctl_softc *ctl_softc,
 713                             union ctl_ha_msg *msg_info)
 714 {
 715         struct ctl_scsiio *ctsio;
 716
 717         if (msg_info->hdr.original_sc == NULL) {
 718                 printf("%s: original_sc == NULL!\n", __func__);
 719                 /* XXX KDM now what? */
 720                 return;
 721         }
 722
 723         ctsio = &msg_info->hdr.original_sc->scsiio;
 724         ctsio->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
 725         ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
 726         ctsio->io_hdr.status = msg_info->hdr.status;
 727         ctsio->scsi_status = msg_info->scsi.scsi_status;
 728         ctsio->sense_len = msg_info->scsi.sense_len;
 729         ctsio->sense_residual = msg_info->scsi.sense_residual;
 730         ctsio->residual = msg_info->scsi.residual;
 731         memcpy(&ctsio->sense_data, &msg_info->scsi.sense_data,
 732                msg_info->scsi.sense_len);
 733         ctl_enqueue_isc((union ctl_io *)ctsio);
 734 }
 735
 736 static void
 737 ctl_isc_handler_finish_ser_only(struct ctl_softc *ctl_softc,
 738                                 union ctl_ha_msg *msg_info)
 739 {
 740         struct ctl_scsiio *ctsio;
 741
 742         if (msg_info->hdr.serializing_sc == NULL) {
 743                 printf("%s: serializing_sc == NULL!\n", __func__);
 744                 /* XXX KDM now what? */
 745                 return;
 746         }
 747
 748         ctsio = &msg_info->hdr.serializing_sc->scsiio;
 749         ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
 750         ctl_enqueue_isc((union ctl_io *)ctsio);
 751 }
 752
 753 void
 754 ctl_isc_announce_lun(struct ctl_lun *lun)
 755 {
 756         struct ctl_softc *softc = lun->ctl_softc;
 757         union ctl_ha_msg *msg;
 758         struct ctl_ha_msg_lun_pr_key pr_key;
 759         int i, k;
 760
 761         if (softc->ha_link != CTL_HA_LINK_ONLINE)
 762                 return;
 763         mtx_lock(&lun->lun_lock);
 764         i = sizeof(msg->lun);
 765         if (lun->lun_devid)
 766                 i += lun->lun_devid->len;
 767         i += sizeof(pr_key) * lun->pr_key_count;
 768 alloc:
 769         mtx_unlock(&lun->lun_lock);
 770         msg = malloc(i, M_CTL, M_WAITOK);
 771         mtx_lock(&lun->lun_lock);
 772         k = sizeof(msg->lun);
 773         if (lun->lun_devid)
 774                 k += lun->lun_devid->len;
 775         k += sizeof(pr_key) * lun->pr_key_count;
 776         if (i < k) {
 777                 free(msg, M_CTL);
 778                 i = k;
 779                 goto alloc;
 780         }
 781         bzero(&msg->lun, sizeof(msg->lun));
 782         msg->hdr.msg_type = CTL_MSG_LUN_SYNC;
 783         msg->hdr.nexus.targ_lun = lun->lun;
 784         msg->hdr.nexus.targ_mapped_lun = lun->lun;
 785         msg->lun.flags = lun->flags;
 786         msg->lun.pr_generation = lun->pr_generation;
 787         msg->lun.pr_res_idx = lun->pr_res_idx;
 788         msg->lun.pr_res_type = lun->pr_res_type;
 789         msg->lun.pr_key_count = lun->pr_key_count;
 790         i = 0;
 791         if (lun->lun_devid) {
 792                 msg->lun.lun_devid_len = lun->lun_devid->len;
 793                 memcpy(&msg->lun.data[i], lun->lun_devid->data,
 794                     msg->lun.lun_devid_len);
 795                 i += msg->lun.lun_devid_len;
 796         }
 797         for (k = 0; k < CTL_MAX_INITIATORS; k++) {
 798                 if ((pr_key.pr_key = ctl_get_prkey(lun, k)) == 0)
 799                         continue;
 800                 pr_key.pr_iid = k;
 801                 memcpy(&msg->lun.data[i], &pr_key, sizeof(pr_key));
 802                 i += sizeof(pr_key);
 803         }
 804         mtx_unlock(&lun->lun_lock);
 805         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->port, sizeof(msg->port) + i,
 806             M_WAITOK);
 807         free(msg, M_CTL);
 808
 809         if (lun->flags & CTL_LUN_PRIMARY_SC) {
 810                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
 811                         ctl_isc_announce_mode(lun, -1,
 812                             lun->mode_pages.index[i].page_code & SMPH_PC_MASK,
 813                             lun->mode_pages.index[i].subpage);
 814                 }
 815         }
 816 }
 817
 818 void
 819 ctl_isc_announce_port(struct ctl_port *port)
 820 {
 821         struct ctl_softc *softc = port->ctl_softc;
 822         union ctl_ha_msg *msg;
 823         int i;
 824
 825         if (port->targ_port < softc->port_min ||
 826             port->targ_port >= softc->port_max ||
 827             softc->ha_link != CTL_HA_LINK_ONLINE)
 828                 return;
 829         i = sizeof(msg->port) + strlen(port->port_name) + 1;
 830         if (port->lun_map)
 831                 i += sizeof(uint32_t) * CTL_MAX_LUNS;
 832         if (port->port_devid)
 833                 i += port->port_devid->len;
 834         if (port->target_devid)
 835                 i += port->target_devid->len;
 836         if (port->init_devid)
 837                 i += port->init_devid->len;
 838         msg = malloc(i, M_CTL, M_WAITOK);
 839         bzero(&msg->port, sizeof(msg->port));
 840         msg->hdr.msg_type = CTL_MSG_PORT_SYNC;
 841         msg->hdr.nexus.targ_port = port->targ_port;
 842         msg->port.port_type = port->port_type;
 843         msg->port.physical_port = port->physical_port;
 844         msg->port.virtual_port = port->virtual_port;
 845         msg->port.status = port->status;
 846         i = 0;
 847         msg->port.name_len = sprintf(&msg->port.data[i],
 848             "%d:%s", softc->ha_id, port->port_name) + 1;
 849         i += msg->port.name_len;
 850         if (port->lun_map) {
 851                 msg->port.lun_map_len = sizeof(uint32_t) * CTL_MAX_LUNS;
 852                 memcpy(&msg->port.data[i], port->lun_map,
 853                     msg->port.lun_map_len);
 854                 i += msg->port.lun_map_len;
 855         }
 856         if (port->port_devid) {
 857                 msg->port.port_devid_len = port->port_devid->len;
 858                 memcpy(&msg->port.data[i], port->port_devid->data,
 859                     msg->port.port_devid_len);
 860                 i += msg->port.port_devid_len;
 861         }
 862         if (port->target_devid) {
 863                 msg->port.target_devid_len = port->target_devid->len;
 864                 memcpy(&msg->port.data[i], port->target_devid->data,
 865                     msg->port.target_devid_len);
 866                 i += msg->port.target_devid_len;
 867         }
 868         if (port->init_devid) {
 869                 msg->port.init_devid_len = port->init_devid->len;
 870                 memcpy(&msg->port.data[i], port->init_devid->data,
 871                     msg->port.init_devid_len);
 872                 i += msg->port.init_devid_len;
 873         }
 874         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->port, sizeof(msg->port) + i,
 875             M_WAITOK);
 876         free(msg, M_CTL);
 877 }
 878
 879 void
 880 ctl_isc_announce_iid(struct ctl_port *port, int iid)
 881 {
 882         struct ctl_softc *softc = port->ctl_softc;
 883         union ctl_ha_msg *msg;
 884         int i, l;
 885
 886         if (port->targ_port < softc->port_min ||
 887             port->targ_port >= softc->port_max ||
 888             softc->ha_link != CTL_HA_LINK_ONLINE)
 889                 return;
 890         mtx_lock(&softc->ctl_lock);
 891         i = sizeof(msg->iid);
 892         l = 0;
 893         if (port->wwpn_iid[iid].name)
 894                 l = strlen(port->wwpn_iid[iid].name) + 1;
 895         i += l;
 896         msg = malloc(i, M_CTL, M_NOWAIT);
 897         if (msg == NULL) {
 898                 mtx_unlock(&softc->ctl_lock);
 899                 return;
 900         }
 901         bzero(&msg->iid, sizeof(msg->iid));
 902         msg->hdr.msg_type = CTL_MSG_IID_SYNC;
 903         msg->hdr.nexus.targ_port = port->targ_port;
 904         msg->hdr.nexus.initid = iid;
 905         msg->iid.in_use = port->wwpn_iid[iid].in_use;
 906         msg->iid.name_len = l;
 907         msg->iid.wwpn = port->wwpn_iid[iid].wwpn;
 908         if (port->wwpn_iid[iid].name)
 909                 strlcpy(msg->iid.data, port->wwpn_iid[iid].name, l);
 910         mtx_unlock(&softc->ctl_lock);
 911         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->iid, i, M_NOWAIT);
 912         free(msg, M_CTL);
 913 }
 914
 915 void
 916 ctl_isc_announce_mode(struct ctl_lun *lun, uint32_t initidx,
 917     uint8_t page, uint8_t subpage)
 918 {
 919         struct ctl_softc *softc = lun->ctl_softc;
 920         union ctl_ha_msg msg;
 921         u_int i;
 922
 923         if (softc->ha_link != CTL_HA_LINK_ONLINE)
 924                 return;
 925         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
 926                 if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) ==
 927                     page && lun->mode_pages.index[i].subpage == subpage)
 928                         break;
 929         }
 930         if (i == CTL_NUM_MODE_PAGES)
 931                 return;
 932
 933         /* Don't try to replicate pages not present on this device. */
 934         if (lun->mode_pages.index[i].page_data == NULL)
 935                 return;
 936
 937         bzero(&msg.mode, sizeof(msg.mode));
 938         msg.hdr.msg_type = CTL_MSG_MODE_SYNC;
 939         msg.hdr.nexus.targ_port = initidx / CTL_MAX_INIT_PER_PORT;
 940         msg.hdr.nexus.initid = initidx % CTL_MAX_INIT_PER_PORT;
 941         msg.hdr.nexus.targ_lun = lun->lun;
 942         msg.hdr.nexus.targ_mapped_lun = lun->lun;
 943         msg.mode.page_code = page;
 944         msg.mode.subpage = subpage;
 945         msg.mode.page_len = lun->mode_pages.index[i].page_len;
 946         memcpy(msg.mode.data, lun->mode_pages.index[i].page_data,
 947             msg.mode.page_len);
 948         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg.mode, sizeof(msg.mode),
 949             M_WAITOK);
 950 }
 951
 952 static void
 953 ctl_isc_ha_link_up(struct ctl_softc *softc)
 954 {
 955         struct ctl_port *port;
 956         struct ctl_lun *lun;
 957         union ctl_ha_msg msg;
 958         int i;
 959
 960         /* Announce this node parameters to peer for validation. */
 961         msg.login.msg_type = CTL_MSG_LOGIN;
 962         msg.login.version = CTL_HA_VERSION;
 963         msg.login.ha_mode = softc->ha_mode;
 964         msg.login.ha_id = softc->ha_id;
 965         msg.login.max_luns = CTL_MAX_LUNS;
 966         msg.login.max_ports = CTL_MAX_PORTS;
 967         msg.login.max_init_per_port = CTL_MAX_INIT_PER_PORT;
 968         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg.login, sizeof(msg.login),
 969             M_WAITOK);
 970
 971         STAILQ_FOREACH(port, &softc->port_list, links) {
 972                 ctl_isc_announce_port(port);
 973                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
 974                         if (port->wwpn_iid[i].in_use)
 975                                 ctl_isc_announce_iid(port, i);
 976                 }
 977         }
 978         STAILQ_FOREACH(lun, &softc->lun_list, links)
 979                 ctl_isc_announce_lun(lun);
 980 }
 981
 982 static void
 983 ctl_isc_ha_link_down(struct ctl_softc *softc)
 984 {
 985         struct ctl_port *port;
 986         struct ctl_lun *lun;
 987         union ctl_io *io;
 988         int i;
 989
 990         mtx_lock(&softc->ctl_lock);
 991         STAILQ_FOREACH(lun, &softc->lun_list, links) {
 992                 mtx_lock(&lun->lun_lock);
 993                 if (lun->flags & CTL_LUN_PEER_SC_PRIMARY) {
 994                         lun->flags &= ~CTL_LUN_PEER_SC_PRIMARY;
 995                         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
 996                 }
 997                 mtx_unlock(&lun->lun_lock);
 998
 999                 mtx_unlock(&softc->ctl_lock);
1000                 io = ctl_alloc_io(softc->othersc_pool);
1001                 mtx_lock(&softc->ctl_lock);
1002                 ctl_zero_io(io);
1003                 io->io_hdr.msg_type = CTL_MSG_FAILOVER;
1004                 io->io_hdr.nexus.targ_mapped_lun = lun->lun;
1005                 ctl_enqueue_isc(io);
1006         }
1007
1008         STAILQ_FOREACH(port, &softc->port_list, links) {
1009                 if (port->targ_port >= softc->port_min &&
1010                     port->targ_port < softc->port_max)
1011                         continue;
1012                 port->status &= ~CTL_PORT_STATUS_ONLINE;
1013                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1014                         port->wwpn_iid[i].in_use = 0;
1015                         free(port->wwpn_iid[i].name, M_CTL);
1016                         port->wwpn_iid[i].name = NULL;
1017                 }
1018         }
1019         mtx_unlock(&softc->ctl_lock);
1020 }
1021
1022 static void
1023 ctl_isc_ua(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1024 {
1025         struct ctl_lun *lun;
1026         uint32_t iid = ctl_get_initindex(&msg->hdr.nexus);
1027
1028         mtx_lock(&softc->ctl_lock);
1029         if (msg->hdr.nexus.targ_lun < CTL_MAX_LUNS &&
1030             (lun = softc->ctl_luns[msg->hdr.nexus.targ_mapped_lun]) != NULL) {
1031                 mtx_lock(&lun->lun_lock);
1032                 mtx_unlock(&softc->ctl_lock);
1033                 if (msg->ua.ua_type == CTL_UA_THIN_PROV_THRES &&
1034                     msg->ua.ua_set)
1035                         memcpy(lun->ua_tpt_info, msg->ua.ua_info, 8);
1036                 if (msg->ua.ua_all) {
1037                         if (msg->ua.ua_set)
1038                                 ctl_est_ua_all(lun, iid, msg->ua.ua_type);
1039                         else
1040                                 ctl_clr_ua_all(lun, iid, msg->ua.ua_type);
1041                 } else {
1042                         if (msg->ua.ua_set)
1043                                 ctl_est_ua(lun, iid, msg->ua.ua_type);
1044                         else
1045                                 ctl_clr_ua(lun, iid, msg->ua.ua_type);
1046                 }
1047                 mtx_unlock(&lun->lun_lock);
1048         } else
1049                 mtx_unlock(&softc->ctl_lock);
1050 }
1051
1052 static void
1053 ctl_isc_lun_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1054 {
1055         struct ctl_lun *lun;
1056         struct ctl_ha_msg_lun_pr_key pr_key;
1057         int i, k;
1058         ctl_lun_flags oflags;
1059         uint32_t targ_lun;
1060
1061         targ_lun = msg->hdr.nexus.targ_mapped_lun;
1062         mtx_lock(&softc->ctl_lock);
1063         if ((targ_lun >= CTL_MAX_LUNS) ||
1064             ((lun = softc->ctl_luns[targ_lun]) == NULL)) {
1065                 mtx_unlock(&softc->ctl_lock);
1066                 return;
1067         }
1068         mtx_lock(&lun->lun_lock);
1069         mtx_unlock(&softc->ctl_lock);
1070         if (lun->flags & CTL_LUN_DISABLED) {
1071                 mtx_unlock(&lun->lun_lock);
1072                 return;
1073         }
1074         i = (lun->lun_devid != NULL) ? lun->lun_devid->len : 0;
1075         if (msg->lun.lun_devid_len != i || (i > 0 &&
1076             memcmp(&msg->lun.data[0], lun->lun_devid->data, i) != 0)) {
1077                 mtx_unlock(&lun->lun_lock);
1078                 printf("%s: Received conflicting HA LUN %d\n",
1079                     __func__, msg->hdr.nexus.targ_lun);
1080                 return;
1081         } else {
1082                 /* Record whether peer is primary. */
1083                 oflags = lun->flags;
1084                 if ((msg->lun.flags & CTL_LUN_PRIMARY_SC) &&
1085                     (msg->lun.flags & CTL_LUN_DISABLED) == 0)
1086                         lun->flags |= CTL_LUN_PEER_SC_PRIMARY;
1087                 else
1088                         lun->flags &= ~CTL_LUN_PEER_SC_PRIMARY;
1089                 if (oflags != lun->flags)
1090                         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
1091
1092                 /* If peer is primary and we are not -- use data */
1093                 if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
1094                     (lun->flags & CTL_LUN_PEER_SC_PRIMARY)) {
1095                         lun->pr_generation = msg->lun.pr_generation;
1096                         lun->pr_res_idx = msg->lun.pr_res_idx;
1097                         lun->pr_res_type = msg->lun.pr_res_type;
1098                         lun->pr_key_count = msg->lun.pr_key_count;
1099                         for (k = 0; k < CTL_MAX_INITIATORS; k++)
1100                                 ctl_clr_prkey(lun, k);
1101                         for (k = 0; k < msg->lun.pr_key_count; k++) {
1102                                 memcpy(&pr_key, &msg->lun.data[i],
1103                                     sizeof(pr_key));
1104                                 ctl_alloc_prkey(lun, pr_key.pr_iid);
1105                                 ctl_set_prkey(lun, pr_key.pr_iid,
1106                                     pr_key.pr_key);
1107                                 i += sizeof(pr_key);
1108                         }
1109                 }
1110
1111                 mtx_unlock(&lun->lun_lock);
1112                 CTL_DEBUG_PRINT(("%s: Known LUN %d, peer is %s\n",
1113                     __func__, msg->hdr.nexus.targ_lun,
1114                     (msg->lun.flags & CTL_LUN_PRIMARY_SC) ?
1115                     "primary" : "secondary"));
1116
1117                 /* If we are primary but peer doesn't know -- notify */
1118                 if ((lun->flags & CTL_LUN_PRIMARY_SC) &&
1119                     (msg->lun.flags & CTL_LUN_PEER_SC_PRIMARY) == 0)
1120                         ctl_isc_announce_lun(lun);
1121         }
1122 }
1123
1124 static void
1125 ctl_isc_port_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1126 {
1127         struct ctl_port *port;
1128         struct ctl_lun *lun;
1129         int i, new;
1130
1131         port = softc->ctl_ports[msg->hdr.nexus.targ_port];
1132         if (port == NULL) {
1133                 CTL_DEBUG_PRINT(("%s: New port %d\n", __func__,
1134                     msg->hdr.nexus.targ_port));
1135                 new = 1;
1136                 port = malloc(sizeof(*port), M_CTL, M_WAITOK | M_ZERO);
1137                 port->frontend = &ha_frontend;
1138                 port->targ_port = msg->hdr.nexus.targ_port;
1139                 port->fe_datamove = ctl_ha_datamove;
1140                 port->fe_done = ctl_ha_done;
1141         } else if (port->frontend == &ha_frontend) {
1142                 CTL_DEBUG_PRINT(("%s: Updated port %d\n", __func__,
1143                     msg->hdr.nexus.targ_port));
1144                 new = 0;
1145         } else {
1146                 printf("%s: Received conflicting HA port %d\n",
1147                     __func__, msg->hdr.nexus.targ_port);
1148                 return;
1149         }
1150         port->port_type = msg->port.port_type;
1151         port->physical_port = msg->port.physical_port;
1152         port->virtual_port = msg->port.virtual_port;
1153         port->status = msg->port.status;
1154         i = 0;
1155         free(port->port_name, M_CTL);
1156         port->port_name = strndup(&msg->port.data[i], msg->port.name_len,
1157             M_CTL);
1158         i += msg->port.name_len;
1159         if (msg->port.lun_map_len != 0) {
1160                 if (port->lun_map == NULL)
1161                         port->lun_map = malloc(sizeof(uint32_t) * CTL_MAX_LUNS,
1162                             M_CTL, M_WAITOK);
1163                 memcpy(port->lun_map, &msg->port.data[i],
1164                     sizeof(uint32_t) * CTL_MAX_LUNS);
1165                 i += msg->port.lun_map_len;
1166         } else {
1167                 free(port->lun_map, M_CTL);
1168                 port->lun_map = NULL;
1169         }
1170         if (msg->port.port_devid_len != 0) {
1171                 if (port->port_devid == NULL ||
1172                     port->port_devid->len != msg->port.port_devid_len) {
1173                         free(port->port_devid, M_CTL);
1174                         port->port_devid = malloc(sizeof(struct ctl_devid) +
1175                             msg->port.port_devid_len, M_CTL, M_WAITOK);
1176                 }
1177                 memcpy(port->port_devid->data, &msg->port.data[i],
1178                     msg->port.port_devid_len);
1179                 port->port_devid->len = msg->port.port_devid_len;
1180                 i += msg->port.port_devid_len;
1181         } else {
1182                 free(port->port_devid, M_CTL);
1183                 port->port_devid = NULL;
1184         }
1185         if (msg->port.target_devid_len != 0) {
1186                 if (port->target_devid == NULL ||
1187                     port->target_devid->len != msg->port.target_devid_len) {
1188                         free(port->target_devid, M_CTL);
1189                         port->target_devid = malloc(sizeof(struct ctl_devid) +
1190                             msg->port.target_devid_len, M_CTL, M_WAITOK);
1191                 }
1192                 memcpy(port->target_devid->data, &msg->port.data[i],
1193                     msg->port.target_devid_len);
1194                 port->target_devid->len = msg->port.target_devid_len;
1195                 i += msg->port.target_devid_len;
1196         } else {
1197                 free(port->target_devid, M_CTL);
1198                 port->target_devid = NULL;
1199         }
1200         if (msg->port.init_devid_len != 0) {
1201                 if (port->init_devid == NULL ||
1202                     port->init_devid->len != msg->port.init_devid_len) {
1203                         free(port->init_devid, M_CTL);
1204                         port->init_devid = malloc(sizeof(struct ctl_devid) +
1205                             msg->port.init_devid_len, M_CTL, M_WAITOK);
1206                 }
1207                 memcpy(port->init_devid->data, &msg->port.data[i],
1208                     msg->port.init_devid_len);
1209                 port->init_devid->len = msg->port.init_devid_len;
1210                 i += msg->port.init_devid_len;
1211         } else {
1212                 free(port->init_devid, M_CTL);
1213                 port->init_devid = NULL;
1214         }
1215         if (new) {
1216                 if (ctl_port_register(port) != 0) {
1217                         printf("%s: ctl_port_register() failed with error\n",
1218                             __func__);
1219                 }
1220         }
1221         mtx_lock(&softc->ctl_lock);
1222         STAILQ_FOREACH(lun, &softc->lun_list, links) {
1223                 if (ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
1224                         continue;
1225                 mtx_lock(&lun->lun_lock);
1226                 ctl_est_ua_all(lun, -1, CTL_UA_INQ_CHANGE);
1227                 mtx_unlock(&lun->lun_lock);
1228         }
1229         mtx_unlock(&softc->ctl_lock);
1230 }
1231
1232 static void
1233 ctl_isc_iid_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1234 {
1235         struct ctl_port *port;
1236         int iid;
1237
1238         port = softc->ctl_ports[msg->hdr.nexus.targ_port];
1239         if (port == NULL) {
1240                 printf("%s: Received IID for unknown port %d\n",
1241                     __func__, msg->hdr.nexus.targ_port);
1242                 return;
1243         }
1244         iid = msg->hdr.nexus.initid;
1245         port->wwpn_iid[iid].in_use = msg->iid.in_use;
1246         port->wwpn_iid[iid].wwpn = msg->iid.wwpn;
1247         free(port->wwpn_iid[iid].name, M_CTL);
1248         if (msg->iid.name_len) {
1249                 port->wwpn_iid[iid].name = strndup(&msg->iid.data[0],
1250                     msg->iid.name_len, M_CTL);
1251         } else
1252                 port->wwpn_iid[iid].name = NULL;
1253 }
1254
1255 static void
1256 ctl_isc_login(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1257 {
1258
1259         if (msg->login.version != CTL_HA_VERSION) {
1260                 printf("CTL HA peers have different versions %d != %d\n",
1261                     msg->login.version, CTL_HA_VERSION);
1262                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1263                 return;
1264         }
1265         if (msg->login.ha_mode != softc->ha_mode) {
1266                 printf("CTL HA peers have different ha_mode %d != %d\n",
1267                     msg->login.ha_mode, softc->ha_mode);
1268                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1269                 return;
1270         }
1271         if (msg->login.ha_id == softc->ha_id) {
1272                 printf("CTL HA peers have same ha_id %d\n", msg->login.ha_id);
1273                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1274                 return;
1275         }
1276         if (msg->login.max_luns != CTL_MAX_LUNS ||
1277             msg->login.max_ports != CTL_MAX_PORTS ||
1278             msg->login.max_init_per_port != CTL_MAX_INIT_PER_PORT) {
1279                 printf("CTL HA peers have different limits\n");
1280                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1281                 return;
1282         }
1283 }
1284
1285 static void
1286 ctl_isc_mode_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1287 {
1288         struct ctl_lun *lun;
1289         u_int i;
1290         uint32_t initidx, targ_lun;
1291
1292         targ_lun = msg->hdr.nexus.targ_mapped_lun;
1293         mtx_lock(&softc->ctl_lock);
1294         if ((targ_lun >= CTL_MAX_LUNS) ||
1295             ((lun = softc->ctl_luns[targ_lun]) == NULL)) {
1296                 mtx_unlock(&softc->ctl_lock);
1297                 return;
1298         }
1299         mtx_lock(&lun->lun_lock);
1300         mtx_unlock(&softc->ctl_lock);
1301         if (lun->flags & CTL_LUN_DISABLED) {
1302                 mtx_unlock(&lun->lun_lock);
1303                 return;
1304         }
1305         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
1306                 if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) ==
1307                     msg->mode.page_code &&
1308                     lun->mode_pages.index[i].subpage == msg->mode.subpage)
1309                         break;
1310         }
1311         if (i == CTL_NUM_MODE_PAGES) {
1312                 mtx_unlock(&lun->lun_lock);
1313                 return;
1314         }
1315         memcpy(lun->mode_pages.index[i].page_data, msg->mode.data,
1316             lun->mode_pages.index[i].page_len);
1317         initidx = ctl_get_initindex(&msg->hdr.nexus);
1318         if (initidx != -1)
1319                 ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
1320         mtx_unlock(&lun->lun_lock);
1321 }
1322
1323 /*
1324  * ISC (Inter Shelf Communication) event handler.  Events from the HA
1325  * subsystem come in here.
1326  */
1327 static void
1328 ctl_isc_event_handler(ctl_ha_channel channel, ctl_ha_event event, int param)
1329 {
1330         struct ctl_softc *softc = control_softc;
1331         union ctl_io *io;
1332         struct ctl_prio *presio;
1333         ctl_ha_status isc_status;
1334
1335         CTL_DEBUG_PRINT(("CTL: Isc Msg event %d\n", event));
1336         if (event == CTL_HA_EVT_MSG_RECV) {
1337                 union ctl_ha_msg *msg, msgbuf;
1338
1339                 if (param > sizeof(msgbuf))
1340                         msg = malloc(param, M_CTL, M_WAITOK);
1341                 else
1342                         msg = &msgbuf;
1343                 isc_status = ctl_ha_msg_recv(CTL_HA_CHAN_CTL, msg, param,
1344                     M_WAITOK);
1345                 if (isc_status != CTL_HA_STATUS_SUCCESS) {
1346                         printf("%s: Error receiving message: %d\n",
1347                             __func__, isc_status);
1348                         if (msg != &msgbuf)
1349                                 free(msg, M_CTL);
1350                         return;
1351                 }
1352
1353                 CTL_DEBUG_PRINT(("CTL: msg_type %d\n", msg->msg_type));
1354                 switch (msg->hdr.msg_type) {
1355                 case CTL_MSG_SERIALIZE:
1356                         io = ctl_alloc_io(softc->othersc_pool);
1357                         ctl_zero_io(io);
1358                         // populate ctsio from msg
1359                         io->io_hdr.io_type = CTL_IO_SCSI;
1360                         io->io_hdr.msg_type = CTL_MSG_SERIALIZE;
1361                         io->io_hdr.original_sc = msg->hdr.original_sc;
1362                         io->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC |
1363                                             CTL_FLAG_IO_ACTIVE;
1364                         /*
1365                          * If we're in serialization-only mode, we don't
1366                          * want to go through full done processing.  Thus
1367                          * the COPY flag.
1368                          *
1369                          * XXX KDM add another flag that is more specific.
1370                          */
1371                         if (softc->ha_mode != CTL_HA_MODE_XFER)
1372                                 io->io_hdr.flags |= CTL_FLAG_INT_COPY;
1373                         io->io_hdr.nexus = msg->hdr.nexus;
1374 #if 0
1375                         printf("port %u, iid %u, lun %u\n",
1376                                io->io_hdr.nexus.targ_port,
1377                                io->io_hdr.nexus.initid,
1378                                io->io_hdr.nexus.targ_lun);
1379 #endif
1380                         io->scsiio.tag_num = msg->scsi.tag_num;
1381                         io->scsiio.tag_type = msg->scsi.tag_type;
1382 #ifdef CTL_TIME_IO
1383                         io->io_hdr.start_time = time_uptime;
1384                         getbinuptime(&io->io_hdr.start_bt);
1385 #endif /* CTL_TIME_IO */
1386                         io->scsiio.cdb_len = msg->scsi.cdb_len;
1387                         memcpy(io->scsiio.cdb, msg->scsi.cdb,
1388                                CTL_MAX_CDBLEN);
1389                         if (softc->ha_mode == CTL_HA_MODE_XFER) {
1390                                 const struct ctl_cmd_entry *entry;
1391
1392                                 entry = ctl_get_cmd_entry(&io->scsiio, NULL);
1393                                 io->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
1394                                 io->io_hdr.flags |=
1395                                         entry->flags & CTL_FLAG_DATA_MASK;
1396                         }
1397                         ctl_enqueue_isc(io);
1398                         break;
1399
1400                 /* Performed on the Originating SC, XFER mode only */
1401                 case CTL_MSG_DATAMOVE: {
1402                         struct ctl_sg_entry *sgl;
1403                         int i, j;
1404
1405                         io = msg->hdr.original_sc;
1406                         if (io == NULL) {
1407                                 printf("%s: original_sc == NULL!\n", __func__);
1408                                 /* XXX KDM do something here */
1409                                 break;
1410                         }
1411                         io->io_hdr.msg_type = CTL_MSG_DATAMOVE;
1412                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1413                         /*
1414                          * Keep track of this, we need to send it back over
1415                          * when the datamove is complete.
1416                          */
1417                         io->io_hdr.serializing_sc = msg->hdr.serializing_sc;
1418                         if (msg->hdr.status == CTL_SUCCESS)
1419                                 io->io_hdr.status = msg->hdr.status;
1420
1421                         if (msg->dt.sg_sequence == 0) {
1422 #ifdef CTL_TIME_IO
1423                                 getbinuptime(&io->io_hdr.dma_start_bt);
1424 #endif
1425                                 i = msg->dt.kern_sg_entries +
1426                                     msg->dt.kern_data_len /
1427                                     CTL_HA_DATAMOVE_SEGMENT + 1;
1428                                 sgl = malloc(sizeof(*sgl) * i, M_CTL,
1429                                     M_WAITOK | M_ZERO);
1430                                 io->io_hdr.remote_sglist = sgl;
1431                                 io->io_hdr.local_sglist =
1432                                     &sgl[msg->dt.kern_sg_entries];
1433
1434                                 io->scsiio.kern_data_ptr = (uint8_t *)sgl;
1435
1436                                 io->scsiio.kern_sg_entries =
1437                                         msg->dt.kern_sg_entries;
1438                                 io->scsiio.rem_sg_entries =
1439                                         msg->dt.kern_sg_entries;
1440                                 io->scsiio.kern_data_len =
1441                                         msg->dt.kern_data_len;
1442                                 io->scsiio.kern_total_len =
1443                                         msg->dt.kern_total_len;
1444                                 io->scsiio.kern_data_resid =
1445                                         msg->dt.kern_data_resid;
1446                                 io->scsiio.kern_rel_offset =
1447                                         msg->dt.kern_rel_offset;
1448                                 io->io_hdr.flags &= ~CTL_FLAG_BUS_ADDR;
1449                                 io->io_hdr.flags |= msg->dt.flags &
1450                                     CTL_FLAG_BUS_ADDR;
1451                         } else
1452                                 sgl = (struct ctl_sg_entry *)
1453                                         io->scsiio.kern_data_ptr;
1454
1455                         for (i = msg->dt.sent_sg_entries, j = 0;
1456                              i < (msg->dt.sent_sg_entries +
1457                              msg->dt.cur_sg_entries); i++, j++) {
1458                                 sgl[i].addr = msg->dt.sg_list[j].addr;
1459                                 sgl[i].len = msg->dt.sg_list[j].len;
1460
1461 #if 0
1462                                 printf("%s: DATAMOVE: %p,%lu j=%d, i=%d\n",
1463                                     __func__, sgl[i].addr, sgl[i].len, j, i);
1464 #endif
1465                         }
1466
1467                         /*
1468                          * If this is the last piece of the I/O, we've got
1469                          * the full S/G list.  Queue processing in the thread.
1470                          * Otherwise wait for the next piece.
1471                          */
1472                         if (msg->dt.sg_last != 0)
1473                                 ctl_enqueue_isc(io);
1474                         break;
1475                 }
1476                 /* Performed on the Serializing (primary) SC, XFER mode only */
1477                 case CTL_MSG_DATAMOVE_DONE: {
1478                         if (msg->hdr.serializing_sc == NULL) {
1479                                 printf("%s: serializing_sc == NULL!\n",
1480                                        __func__);
1481                                 /* XXX KDM now what? */
1482                                 break;
1483                         }
1484                         /*
1485                          * We grab the sense information here in case
1486                          * there was a failure, so we can return status
1487                          * back to the initiator.
1488                          */
1489                         io = msg->hdr.serializing_sc;
1490                         io->io_hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
1491                         io->io_hdr.flags &= ~CTL_FLAG_DMA_INPROG;
1492                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1493                         io->io_hdr.port_status = msg->scsi.fetd_status;
1494                         io->scsiio.residual = msg->scsi.residual;
1495                         if (msg->hdr.status != CTL_STATUS_NONE) {
1496                                 io->io_hdr.status = msg->hdr.status;
1497                                 io->scsiio.scsi_status = msg->scsi.scsi_status;
1498                                 io->scsiio.sense_len = msg->scsi.sense_len;
1499                                 io->scsiio.sense_residual =msg->scsi.sense_residual;
1500                                 memcpy(&io->scsiio.sense_data,
1501                                     &msg->scsi.sense_data,
1502                                     msg->scsi.sense_len);
1503                                 if (msg->hdr.status == CTL_SUCCESS)
1504                                         io->io_hdr.flags |= CTL_FLAG_STATUS_SENT;
1505                         }
1506                         ctl_enqueue_isc(io);
1507                         break;
1508                 }
1509
1510                 /* Preformed on Originating SC, SER_ONLY mode */
1511                 case CTL_MSG_R2R:
1512                         io = msg->hdr.original_sc;
1513                         if (io == NULL) {
1514                                 printf("%s: original_sc == NULL!\n",
1515                                     __func__);
1516                                 break;
1517                         }
1518                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1519                         io->io_hdr.msg_type = CTL_MSG_R2R;
1520                         io->io_hdr.serializing_sc = msg->hdr.serializing_sc;
1521                         ctl_enqueue_isc(io);
1522                         break;
1523
1524                 /*
1525                  * Performed on Serializing(i.e. primary SC) SC in SER_ONLY
1526                  * mode.
1527                  * Performed on the Originating (i.e. secondary) SC in XFER
1528                  * mode
1529                  */
1530                 case CTL_MSG_FINISH_IO:
1531                         if (softc->ha_mode == CTL_HA_MODE_XFER)
1532                                 ctl_isc_handler_finish_xfer(softc, msg);
1533                         else
1534                                 ctl_isc_handler_finish_ser_only(softc, msg);
1535                         break;
1536
1537                 /* Preformed on Originating SC */
1538                 case CTL_MSG_BAD_JUJU:
1539                         io = msg->hdr.original_sc;
1540                         if (io == NULL) {
1541                                 printf("%s: Bad JUJU!, original_sc is NULL!\n",
1542                                        __func__);
1543                                 break;
1544                         }
1545                         ctl_copy_sense_data(msg, io);
1546                         /*
1547                          * IO should have already been cleaned up on other
1548                          * SC so clear this flag so we won't send a message
1549                          * back to finish the IO there.
1550                          */
1551                         io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
1552                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1553
1554                         /* io = msg->hdr.serializing_sc; */
1555                         io->io_hdr.msg_type = CTL_MSG_BAD_JUJU;
1556                         ctl_enqueue_isc(io);
1557                         break;
1558
1559                 /* Handle resets sent from the other side */
1560                 case CTL_MSG_MANAGE_TASKS: {
1561                         struct ctl_taskio *taskio;
1562                         taskio = (struct ctl_taskio *)ctl_alloc_io(
1563                             softc->othersc_pool);
1564                         ctl_zero_io((union ctl_io *)taskio);
1565                         taskio->io_hdr.io_type = CTL_IO_TASK;
1566                         taskio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
1567                         taskio->io_hdr.nexus = msg->hdr.nexus;
1568                         taskio->task_action = msg->task.task_action;
1569                         taskio->tag_num = msg->task.tag_num;
1570                         taskio->tag_type = msg->task.tag_type;
1571 #ifdef CTL_TIME_IO
1572                         taskio->io_hdr.start_time = time_uptime;
1573                         getbinuptime(&taskio->io_hdr.start_bt);
1574 #endif /* CTL_TIME_IO */
1575                         ctl_run_task((union ctl_io *)taskio);
1576                         break;
1577                 }
1578                 /* Persistent Reserve action which needs attention */
1579                 case CTL_MSG_PERS_ACTION:
1580                         presio = (struct ctl_prio *)ctl_alloc_io(
1581                             softc->othersc_pool);
1582                         ctl_zero_io((union ctl_io *)presio);
1583                         presio->io_hdr.msg_type = CTL_MSG_PERS_ACTION;
1584                         presio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
1585                         presio->io_hdr.nexus = msg->hdr.nexus;
1586                         presio->pr_msg = msg->pr;
1587                         ctl_enqueue_isc((union ctl_io *)presio);
1588                         break;
1589                 case CTL_MSG_UA:
1590                         ctl_isc_ua(softc, msg, param);
1591                         break;
1592                 case CTL_MSG_PORT_SYNC:
1593                         ctl_isc_port_sync(softc, msg, param);
1594                         break;
1595                 case CTL_MSG_LUN_SYNC:
1596                         ctl_isc_lun_sync(softc, msg, param);
1597                         break;
1598                 case CTL_MSG_IID_SYNC:
1599                         ctl_isc_iid_sync(softc, msg, param);
1600                         break;
1601                 case CTL_MSG_LOGIN:
1602                         ctl_isc_login(softc, msg, param);
1603                         break;
1604                 case CTL_MSG_MODE_SYNC:
1605                         ctl_isc_mode_sync(softc, msg, param);
1606                         break;
1607                 default:
1608                         printf("Received HA message of unknown type %d\n",
1609                             msg->hdr.msg_type);
1610                         ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1611                         break;
1612                 }
1613                 if (msg != &msgbuf)
1614                         free(msg, M_CTL);
1615         } else if (event == CTL_HA_EVT_LINK_CHANGE) {
1616                 printf("CTL: HA link status changed from %d to %d\n",
1617                     softc->ha_link, param);
1618                 if (param == softc->ha_link)
1619                         return;
1620                 if (softc->ha_link == CTL_HA_LINK_ONLINE) {
1621                         softc->ha_link = param;
1622                         ctl_isc_ha_link_down(softc);
1623                 } else {
1624                         softc->ha_link = param;
1625                         if (softc->ha_link == CTL_HA_LINK_ONLINE)
1626                                 ctl_isc_ha_link_up(softc);
1627                 }
1628                 return;
1629         } else {
1630                 printf("ctl_isc_event_handler: Unknown event %d\n", event);
1631                 return;
1632         }
1633 }
1634
1635 static void
1636 ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest)
1637 {
1638
1639         memcpy(&dest->scsiio.sense_data, &src->scsi.sense_data,
1640             src->scsi.sense_len);
1641         dest->scsiio.scsi_status = src->scsi.scsi_status;
1642         dest->scsiio.sense_len = src->scsi.sense_len;
1643         dest->io_hdr.status = src->hdr.status;
1644 }
1645
1646 static void
1647 ctl_copy_sense_data_back(union ctl_io *src, union ctl_ha_msg *dest)
1648 {
1649
1650         memcpy(&dest->scsi.sense_data, &src->scsiio.sense_data,
1651             src->scsiio.sense_len);
1652         dest->scsi.scsi_status = src->scsiio.scsi_status;
1653         dest->scsi.sense_len = src->scsiio.sense_len;
1654         dest->hdr.status = src->io_hdr.status;
1655 }
1656
1657 void
1658 ctl_est_ua(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua)
1659 {
1660         struct ctl_softc *softc = lun->ctl_softc;
1661         ctl_ua_type *pu;
1662
1663         if (initidx < softc->init_min || initidx >= softc->init_max)
1664                 return;
1665         mtx_assert(&lun->lun_lock, MA_OWNED);
1666         pu = lun->pending_ua[initidx / CTL_MAX_INIT_PER_PORT];
1667         if (pu == NULL)
1668                 return;
1669         pu[initidx % CTL_MAX_INIT_PER_PORT] |= ua;
1670 }
1671
1672 void
1673 ctl_est_ua_port(struct ctl_lun *lun, int port, uint32_t except, ctl_ua_type ua)
1674 {
1675         int i;
1676
1677         mtx_assert(&lun->lun_lock, MA_OWNED);
1678         if (lun->pending_ua[port] == NULL)
1679                 return;
1680         for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1681                 if (port * CTL_MAX_INIT_PER_PORT + i == except)
1682                         continue;
1683                 lun->pending_ua[port][i] |= ua;
1684         }
1685 }
1686
1687 void
1688 ctl_est_ua_all(struct ctl_lun *lun, uint32_t except, ctl_ua_type ua)
1689 {
1690         struct ctl_softc *softc = lun->ctl_softc;
1691         int i;
1692
1693         mtx_assert(&lun->lun_lock, MA_OWNED);
1694         for (i = softc->port_min; i < softc->port_max; i++)
1695                 ctl_est_ua_port(lun, i, except, ua);
1696 }
1697
1698 void
1699 ctl_clr_ua(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua)
1700 {
1701         struct ctl_softc *softc = lun->ctl_softc;
1702         ctl_ua_type *pu;
1703
1704         if (initidx < softc->init_min || initidx >= softc->init_max)
1705                 return;
1706         mtx_assert(&lun->lun_lock, MA_OWNED);
1707         pu = lun->pending_ua[initidx / CTL_MAX_INIT_PER_PORT];
1708         if (pu == NULL)
1709                 return;
1710         pu[initidx % CTL_MAX_INIT_PER_PORT] &= ~ua;
1711 }
1712
1713 void
1714 ctl_clr_ua_all(struct ctl_lun *lun, uint32_t except, ctl_ua_type ua)
1715 {
1716         struct ctl_softc *softc = lun->ctl_softc;
1717         int i, j;
1718
1719         mtx_assert(&lun->lun_lock, MA_OWNED);
1720         for (i = softc->port_min; i < softc->port_max; i++) {
1721                 if (lun->pending_ua[i] == NULL)
1722                         continue;
1723                 for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
1724                         if (i * CTL_MAX_INIT_PER_PORT + j == except)
1725                                 continue;
1726                         lun->pending_ua[i][j] &= ~ua;
1727                 }
1728         }
1729 }
1730
1731 void
1732 ctl_clr_ua_allluns(struct ctl_softc *ctl_softc, uint32_t initidx,
1733     ctl_ua_type ua_type)
1734 {
1735         struct ctl_lun *lun;
1736
1737         mtx_assert(&ctl_softc->ctl_lock, MA_OWNED);
1738         STAILQ_FOREACH(lun, &ctl_softc->lun_list, links) {
1739                 mtx_lock(&lun->lun_lock);
1740                 ctl_clr_ua(lun, initidx, ua_type);
1741                 mtx_unlock(&lun->lun_lock);
1742         }
1743 }
1744
1745 static int
1746 ctl_ha_role_sysctl(SYSCTL_HANDLER_ARGS)
1747 {
1748         struct ctl_softc *softc = (struct ctl_softc *)arg1;
1749         struct ctl_lun *lun;
1750         struct ctl_lun_req ireq;
1751         int error, value;
1752
1753         value = (softc->flags & CTL_FLAG_ACTIVE_SHELF) ? 0 : 1;
1754         error = sysctl_handle_int(oidp, &value, 0, req);
1755         if ((error != 0) || (req->newptr == NULL))
1756                 return (error);
1757
1758         mtx_lock(&softc->ctl_lock);
1759         if (value == 0)
1760                 softc->flags |= CTL_FLAG_ACTIVE_SHELF;
1761         else
1762                 softc->flags &= ~CTL_FLAG_ACTIVE_SHELF;
1763         STAILQ_FOREACH(lun, &softc->lun_list, links) {
1764                 mtx_unlock(&softc->ctl_lock);
1765                 bzero(&ireq, sizeof(ireq));
1766                 ireq.reqtype = CTL_LUNREQ_MODIFY;
1767                 ireq.reqdata.modify.lun_id = lun->lun;
1768                 lun->backend->ioctl(NULL, CTL_LUN_REQ, (caddr_t)&ireq, 0,
1769                     curthread);
1770                 if (ireq.status != CTL_LUN_OK) {
1771                         printf("%s: CTL_LUNREQ_MODIFY returned %d '%s'\n",
1772                             __func__, ireq.status, ireq.error_str);
1773                 }
1774                 mtx_lock(&softc->ctl_lock);
1775         }
1776         mtx_unlock(&softc->ctl_lock);
1777         return (0);
1778 }
1779
1780 static int
1781 ctl_init(void)
1782 {
1783         struct make_dev_args args;
1784         struct ctl_softc *softc;
1785         void *other_pool;
1786         int i, error;
1787
1788         softc = control_softc = malloc(sizeof(*control_softc), M_DEVBUF,
1789                                M_WAITOK | M_ZERO);
1790
1791         make_dev_args_init(&args);
1792         args.mda_devsw = &ctl_cdevsw;
1793         args.mda_uid = UID_ROOT;
1794         args.mda_gid = GID_OPERATOR;
1795         args.mda_mode = 0600;
1796         args.mda_si_drv1 = softc;
1797         error = make_dev_s(&args, &softc->dev, "cam/ctl");
1798         if (error != 0) {
1799                 free(control_softc, M_DEVBUF);
1800                 return (error);
1801         }
1802
1803         sysctl_ctx_init(&softc->sysctl_ctx);
1804         softc->sysctl_tree = SYSCTL_ADD_NODE(&softc->sysctl_ctx,
1805                 SYSCTL_STATIC_CHILDREN(_kern_cam), OID_AUTO, "ctl",
1806                 CTLFLAG_RD, 0, "CAM Target Layer");
1807
1808         if (softc->sysctl_tree == NULL) {
1809                 printf("%s: unable to allocate sysctl tree\n", __func__);
1810                 destroy_dev(softc->dev);
1811                 free(control_softc, M_DEVBUF);
1812                 control_softc = NULL;
1813                 return (ENOMEM);
1814         }
1815
1816         mtx_init(&softc->ctl_lock, "CTL mutex", NULL, MTX_DEF);
1817         softc->io_zone = uma_zcreate("CTL IO", sizeof(union ctl_io),
1818             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1819         softc->flags = 0;
1820
1821         TUNABLE_INT_FETCH("kern.cam.ctl.ha_mode", (int *)&softc->ha_mode);
1822         SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1823             OID_AUTO, "ha_mode", CTLFLAG_RDTUN, (int *)&softc->ha_mode, 0,
1824             "HA mode (0 - act/stby, 1 - serialize only, 2 - xfer)");
1825
1826         /*
1827          * In Copan's HA scheme, the "master" and "slave" roles are
1828          * figured out through the slot the controller is in.  Although it
1829          * is an active/active system, someone has to be in charge.
1830          */
1831         TUNABLE_INT_FETCH("kern.cam.ctl.ha_id", &softc->ha_id);
1832         SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1833             OID_AUTO, "ha_id", CTLFLAG_RDTUN, &softc->ha_id, 0,
1834             "HA head ID (0 - no HA)");
1835         if (softc->ha_id == 0 || softc->ha_id > NUM_HA_SHELVES) {
1836                 softc->flags |= CTL_FLAG_ACTIVE_SHELF;
1837                 softc->is_single = 1;
1838                 softc->port_cnt = CTL_MAX_PORTS;
1839                 softc->port_min = 0;
1840         } else {
1841                 softc->port_cnt = CTL_MAX_PORTS / NUM_HA_SHELVES;
1842                 softc->port_min = (softc->ha_id - 1) * softc->port_cnt;
1843         }
1844         softc->port_max = softc->port_min + softc->port_cnt;
1845         softc->init_min = softc->port_min * CTL_MAX_INIT_PER_PORT;
1846         softc->init_max = softc->port_max * CTL_MAX_INIT_PER_PORT;
1847
1848         SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1849             OID_AUTO, "ha_link", CTLFLAG_RD, (int *)&softc->ha_link, 0,
1850             "HA link state (0 - offline, 1 - unknown, 2 - online)");
1851
1852         STAILQ_INIT(&softc->lun_list);
1853         STAILQ_INIT(&softc->pending_lun_queue);
1854         STAILQ_INIT(&softc->fe_list);
1855         STAILQ_INIT(&softc->port_list);
1856         STAILQ_INIT(&softc->be_list);
1857         ctl_tpc_init(softc);
1858
1859         if (ctl_pool_create(softc, "othersc", CTL_POOL_ENTRIES_OTHER_SC,
1860                             &other_pool) != 0)
1861         {
1862                 printf("ctl: can't allocate %d entry other SC pool, "
1863                        "exiting\n", CTL_POOL_ENTRIES_OTHER_SC);
1864                 return (ENOMEM);
1865         }
1866         softc->othersc_pool = other_pool;
1867
1868         if (worker_threads <= 0)
1869                 worker_threads = max(1, mp_ncpus / 4);
1870         if (worker_threads > CTL_MAX_THREADS)
1871                 worker_threads = CTL_MAX_THREADS;
1872
1873         for (i = 0; i < worker_threads; i++) {
1874                 struct ctl_thread *thr = &softc->threads[i];
1875
1876                 mtx_init(&thr->queue_lock, "CTL queue mutex", NULL, MTX_DEF);
1877                 thr->ctl_softc = softc;
1878                 STAILQ_INIT(&thr->incoming_queue);
1879                 STAILQ_INIT(&thr->rtr_queue);
1880                 STAILQ_INIT(&thr->done_queue);
1881                 STAILQ_INIT(&thr->isc_queue);
1882
1883                 error = kproc_kthread_add(ctl_work_thread, thr,
1884                     &softc->ctl_proc, &thr->thread, 0, 0, "ctl", "work%d", i);
1885                 if (error != 0) {
1886                         printf("error creating CTL work thread!\n");
1887                         ctl_pool_free(other_pool);
1888                         return (error);
1889                 }
1890         }
1891         error = kproc_kthread_add(ctl_lun_thread, softc,
1892             &softc->ctl_proc, NULL, 0, 0, "ctl", "lun");
1893         if (error != 0) {
1894                 printf("error creating CTL lun thread!\n");
1895                 ctl_pool_free(other_pool);
1896                 return (error);
1897         }
1898         error = kproc_kthread_add(ctl_thresh_thread, softc,
1899             &softc->ctl_proc, NULL, 0, 0, "ctl", "thresh");
1900         if (error != 0) {
1901                 printf("error creating CTL threshold thread!\n");
1902                 ctl_pool_free(other_pool);
1903                 return (error);
1904         }
1905
1906         SYSCTL_ADD_PROC(&softc->sysctl_ctx,SYSCTL_CHILDREN(softc->sysctl_tree),
1907             OID_AUTO, "ha_role", CTLTYPE_INT | CTLFLAG_RWTUN,
1908             softc, 0, ctl_ha_role_sysctl, "I", "HA role for this head");
1909
1910         if (softc->is_single == 0) {
1911                 ctl_frontend_register(&ha_frontend);
1912                 if (ctl_ha_msg_init(softc) != CTL_HA_STATUS_SUCCESS) {
1913                         printf("ctl_init: ctl_ha_msg_init failed.\n");
1914                         softc->is_single = 1;
1915                 } else
1916                 if (ctl_ha_msg_register(CTL_HA_CHAN_CTL, ctl_isc_event_handler)
1917                     != CTL_HA_STATUS_SUCCESS) {
1918                         printf("ctl_init: ctl_ha_msg_register failed.\n");
1919                         softc->is_single = 1;
1920                 }
1921         }
1922         return (0);
1923 }
1924
1925 void
1926 ctl_shutdown(void)
1927 {
1928         struct ctl_softc *softc = control_softc;
1929         struct ctl_lun *lun, *next_lun;
1930
1931         if (softc->is_single == 0) {
1932                 ctl_ha_msg_shutdown(softc);
1933                 if (ctl_ha_msg_deregister(CTL_HA_CHAN_CTL)
1934                     != CTL_HA_STATUS_SUCCESS)
1935                         printf("%s: ctl_ha_msg_deregister failed.\n", __func__);
1936                 if (ctl_ha_msg_destroy(softc) != CTL_HA_STATUS_SUCCESS)
1937                         printf("%s: ctl_ha_msg_destroy failed.\n", __func__);
1938                 ctl_frontend_deregister(&ha_frontend);
1939         }
1940
1941         mtx_lock(&softc->ctl_lock);
1942
1943         STAILQ_FOREACH_SAFE(lun, &softc->lun_list, links, next_lun)
1944                 ctl_free_lun(lun);
1945
1946         mtx_unlock(&softc->ctl_lock);
1947
1948 #if 0
1949         ctl_shutdown_thread(softc->work_thread);
1950         mtx_destroy(&softc->queue_lock);
1951 #endif
1952
1953         ctl_tpc_shutdown(softc);
1954         uma_zdestroy(softc->io_zone);
1955         mtx_destroy(&softc->ctl_lock);
1956
1957         destroy_dev(softc->dev);
1958
1959         sysctl_ctx_free(&softc->sysctl_ctx);
1960
1961         free(control_softc, M_DEVBUF);
1962         control_softc = NULL;
1963 }
1964
1965 static int
1966 ctl_module_event_handler(module_t mod, int what, void *arg)
1967 {
1968
1969         switch (what) {
1970         case MOD_LOAD:
1971                 return (ctl_init());
1972         case MOD_UNLOAD:
1973                 return (EBUSY);
1974         default:
1975                 return (EOPNOTSUPP);
1976         }
1977 }
1978
1979 /*
1980  * XXX KDM should we do some access checks here?  Bump a reference count to
1981  * prevent a CTL module from being unloaded while someone has it open?
1982  */
1983 static int
1984 ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td)
1985 {
1986         return (0);
1987 }
1988
1989 static int
1990 ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td)
1991 {
1992         return (0);
1993 }
1994
1995 /*
1996  * Remove an initiator by port number and initiator ID.
1997  * Returns 0 for success, -1 for failure.
1998  */
1999 int
2000 ctl_remove_initiator(struct ctl_port *port, int iid)
2001 {
2002         struct ctl_softc *softc = port->ctl_softc;
2003
2004         mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
2005
2006         if (iid > CTL_MAX_INIT_PER_PORT) {
2007                 printf("%s: initiator ID %u > maximun %u!\n",
2008                        __func__, iid, CTL_MAX_INIT_PER_PORT);
2009                 return (-1);
2010         }
2011
2012         mtx_lock(&softc->ctl_lock);
2013         port->wwpn_iid[iid].in_use--;
2014         port->wwpn_iid[iid].last_use = time_uptime;
2015         mtx_unlock(&softc->ctl_lock);
2016         ctl_isc_announce_iid(port, iid);
2017
2018         return (0);
2019 }
2020
2021 /*
2022  * Add an initiator to the initiator map.
2023  * Returns iid for success, < 0 for failure.
2024  */
2025 int
2026 ctl_add_initiator(struct ctl_port *port, int iid, uint64_t wwpn, char *name)
2027 {
2028         struct ctl_softc *softc = port->ctl_softc;
2029         time_t best_time;
2030         int i, best;
2031
2032         mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
2033
2034         if (iid >= CTL_MAX_INIT_PER_PORT) {
2035                 printf("%s: WWPN %#jx initiator ID %u > maximum %u!\n",
2036                        __func__, wwpn, iid, CTL_MAX_INIT_PER_PORT);
2037                 free(name, M_CTL);
2038                 return (-1);
2039         }
2040
2041         mtx_lock(&softc->ctl_lock);
2042
2043         if (iid < 0 && (wwpn != 0 || name != NULL)) {
2044                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
2045                         if (wwpn != 0 && wwpn == port->wwpn_iid[i].wwpn) {
2046                                 iid = i;
2047                                 break;
2048                         }
2049                         if (name != NULL && port->wwpn_iid[i].name != NULL &&
2050                             strcmp(name, port->wwpn_iid[i].name) == 0) {
2051                                 iid = i;
2052                                 break;
2053                         }
2054                 }
2055         }
2056
2057         if (iid < 0) {
2058                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
2059                         if (port->wwpn_iid[i].in_use == 0 &&
2060                             port->wwpn_iid[i].wwpn == 0 &&
2061                             port->wwpn_iid[i].name == NULL) {
2062                                 iid = i;
2063                                 break;
2064                         }
2065                 }
2066         }
2067
2068         if (iid < 0) {
2069                 best = -1;
2070                 best_time = INT32_MAX;
2071                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
2072                         if (port->wwpn_iid[i].in_use == 0) {
2073                                 if (port->wwpn_iid[i].last_use < best_time) {
2074                                         best = i;
2075                                         best_time = port->wwpn_iid[i].last_use;
2076                                 }
2077                         }
2078                 }
2079                 iid = best;
2080         }
2081
2082         if (iid < 0) {
2083                 mtx_unlock(&softc->ctl_lock);
2084                 free(name, M_CTL);
2085                 return (-2);
2086         }
2087
2088         if (port->wwpn_iid[iid].in_use > 0 && (wwpn != 0 || name != NULL)) {
2089                 /*
2090                  * This is not an error yet.
2091                  */
2092                 if (wwpn != 0 && wwpn == port->wwpn_iid[iid].wwpn) {
2093 #if 0
2094                         printf("%s: port %d iid %u WWPN %#jx arrived"
2095                             " again\n", __func__, port->targ_port,
2096                             iid, (uintmax_t)wwpn);
2097 #endif
2098                         goto take;
2099                 }
2100                 if (name != NULL && port->wwpn_iid[iid].name != NULL &&
2101                     strcmp(name, port->wwpn_iid[iid].name) == 0) {
2102 #if 0
2103                         printf("%s: port %d iid %u name '%s' arrived"
2104                             " again\n", __func__, port->targ_port,
2105                             iid, name);
2106 #endif
2107                         goto take;
2108                 }
2109
2110                 /*
2111                  * This is an error, but what do we do about it?  The
2112                  * driver is telling us we have a new WWPN for this
2113                  * initiator ID, so we pretty much need to use it.
2114                  */
2115                 printf("%s: port %d iid %u WWPN %#jx '%s' arrived,"
2116                     " but WWPN %#jx '%s' is still at that address\n",
2117                     __func__, port->targ_port, iid, wwpn, name,
2118                     (uintmax_t)port->wwpn_iid[iid].wwpn,
2119                     port->wwpn_iid[iid].name);
2120
2121                 /*
2122                  * XXX KDM clear have_ca and ua_pending on each LUN for
2123                  * this initiator.
2124                  */
2125         }
2126 take:
2127         free(port->wwpn_iid[iid].name, M_CTL);
2128         port->wwpn_iid[iid].name = name;
2129         port->wwpn_iid[iid].wwpn = wwpn;
2130         port->wwpn_iid[iid].in_use++;
2131         mtx_unlock(&softc->ctl_lock);
2132         ctl_isc_announce_iid(port, iid);
2133
2134         return (iid);
2135 }
2136
2137 static int
2138 ctl_create_iid(struct ctl_port *port, int iid, uint8_t *buf)
2139 {
2140         int len;
2141
2142         switch (port->port_type) {
2143         case CTL_PORT_FC:
2144         {
2145                 struct scsi_transportid_fcp *id =
2146                     (struct scsi_transportid_fcp *)buf;
2147                 if (port->wwpn_iid[iid].wwpn == 0)
2148                         return (0);
2149                 memset(id, 0, sizeof(*id));
2150                 id->format_protocol = SCSI_PROTO_FC;
2151                 scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->n_port_name);
2152                 return (sizeof(*id));
2153         }
2154         case CTL_PORT_ISCSI:
2155         {
2156                 struct scsi_transportid_iscsi_port *id =
2157                     (struct scsi_transportid_iscsi_port *)buf;
2158                 if (port->wwpn_iid[iid].name == NULL)
2159                         return (0);
2160                 memset(id, 0, 256);
2161                 id->format_protocol = SCSI_TRN_ISCSI_FORMAT_PORT |
2162                     SCSI_PROTO_ISCSI;
2163                 len = strlcpy(id->iscsi_name, port->wwpn_iid[iid].name, 252) + 1;
2164                 len = roundup2(min(len, 252), 4);
2165                 scsi_ulto2b(len, id->additional_length);
2166                 return (sizeof(*id) + len);
2167         }
2168         case CTL_PORT_SAS:
2169         {
2170                 struct scsi_transportid_sas *id =
2171                     (struct scsi_transportid_sas *)buf;
2172                 if (port->wwpn_iid[iid].wwpn == 0)
2173                         return (0);
2174                 memset(id, 0, sizeof(*id));
2175                 id->format_protocol = SCSI_PROTO_SAS;
2176                 scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->sas_address);
2177                 return (sizeof(*id));
2178         }
2179         default:
2180         {
2181                 struct scsi_transportid_spi *id =
2182                     (struct scsi_transportid_spi *)buf;
2183                 memset(id, 0, sizeof(*id));
2184                 id->format_protocol = SCSI_PROTO_SPI;
2185                 scsi_ulto2b(iid, id->scsi_addr);
2186                 scsi_ulto2b(port->targ_port, id->rel_trgt_port_id);
2187                 return (sizeof(*id));
2188         }
2189         }
2190 }
2191
2192 /*
2193  * Serialize a command that went down the "wrong" side, and so was sent to
2194  * this controller for execution.  The logic is a little different than the
2195  * standard case in ctl_scsiio_precheck().  Errors in this case need to get
2196  * sent back to the other side, but in the success case, we execute the
2197  * command on this side (XFER mode) or tell the other side to execute it
2198  * (SER_ONLY mode).
2199  */
2200 static int
2201 ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio)
2202 {
2203         struct ctl_softc *softc = control_softc;
2204         union ctl_ha_msg msg_info;
2205         struct ctl_port *port;
2206         struct ctl_lun *lun;
2207         const struct ctl_cmd_entry *entry;
2208         int retval = 0;
2209         uint32_t targ_lun;
2210
2211         targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
2212         mtx_lock(&softc->ctl_lock);
2213
2214         /* Make sure that we know about this port. */
2215         port = ctl_io_port(&ctsio->io_hdr);
2216         if (port == NULL || (port->status & CTL_PORT_STATUS_ONLINE) == 0) {
2217                 ctl_set_internal_failure(ctsio, /*sks_valid*/ 0,
2218                                          /*retry_count*/ 1);
2219                 goto badjuju;
2220         }
2221
2222         /* Make sure that we know about this LUN. */
2223         if ((targ_lun < CTL_MAX_LUNS) &&
2224             ((lun = softc->ctl_luns[targ_lun]) != NULL)) {
2225                 mtx_lock(&lun->lun_lock);
2226                 mtx_unlock(&softc->ctl_lock);
2227                 /*
2228                  * If the LUN is invalid, pretend that it doesn't exist.
2229                  * It will go away as soon as all pending I/O has been
2230                  * completed.
2231                  */
2232                 if (lun->flags & CTL_LUN_DISABLED) {
2233                         mtx_unlock(&lun->lun_lock);
2234                         lun = NULL;
2235                 }
2236         } else {
2237                 mtx_unlock(&softc->ctl_lock);
2238                 lun = NULL;
2239         }
2240         if (lun == NULL) {
2241                 /*
2242                  * The other node would not send this request to us unless
2243                  * received announce that we are primary node for this LUN.
2244                  * If this LUN does not exist now, it is probably result of
2245                  * a race, so respond to initiator in the most opaque way.
2246                  */
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         ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = lun;
2258         ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = 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                 retval = 1;
2322                 break;
2323         }
2324         return (retval);
2325 }
2326
2327 /*
2328  * Returns 0 for success, errno for failure.
2329  */
2330 static void
2331 ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
2332                    struct ctl_ooa *ooa_hdr, struct ctl_ooa_entry *kern_entries)
2333 {
2334         union ctl_io *io;
2335
2336         mtx_lock(&lun->lun_lock);
2337         for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); (io != NULL);
2338              (*cur_fill_num)++, io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2339              ooa_links)) {
2340                 struct ctl_ooa_entry *entry;
2341
2342                 /*
2343                  * If we've got more than we can fit, just count the
2344                  * remaining entries.
2345                  */
2346                 if (*cur_fill_num >= ooa_hdr->alloc_num)
2347                         continue;
2348
2349                 entry = &kern_entries[*cur_fill_num];
2350
2351                 entry->tag_num = io->scsiio.tag_num;
2352                 entry->lun_num = lun->lun;
2353 #ifdef CTL_TIME_IO
2354                 entry->start_bt = io->io_hdr.start_bt;
2355 #endif
2356                 bcopy(io->scsiio.cdb, entry->cdb, io->scsiio.cdb_len);
2357                 entry->cdb_len = io->scsiio.cdb_len;
2358                 if (io->io_hdr.flags & CTL_FLAG_BLOCKED)
2359                         entry->cmd_flags |= CTL_OOACMD_FLAG_BLOCKED;
2360
2361                 if (io->io_hdr.flags & CTL_FLAG_DMA_INPROG)
2362                         entry->cmd_flags |= CTL_OOACMD_FLAG_DMA;
2363
2364                 if (io->io_hdr.flags & CTL_FLAG_ABORT)
2365                         entry->cmd_flags |= CTL_OOACMD_FLAG_ABORT;
2366
2367                 if (io->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR)
2368                         entry->cmd_flags |= CTL_OOACMD_FLAG_RTR;
2369
2370                 if (io->io_hdr.flags & CTL_FLAG_DMA_QUEUED)
2371                         entry->cmd_flags |= CTL_OOACMD_FLAG_DMA_QUEUED;
2372         }
2373         mtx_unlock(&lun->lun_lock);
2374 }
2375
2376 static void *
2377 ctl_copyin_alloc(void *user_addr, int len, char *error_str,
2378                  size_t error_str_len)
2379 {
2380         void *kptr;
2381
2382         kptr = malloc(len, M_CTL, M_WAITOK | M_ZERO);
2383
2384         if (copyin(user_addr, kptr, len) != 0) {
2385                 snprintf(error_str, error_str_len, "Error copying %d bytes "
2386                          "from user address %p to kernel address %p", len,
2387                          user_addr, kptr);
2388                 free(kptr, M_CTL);
2389                 return (NULL);
2390         }
2391
2392         return (kptr);
2393 }
2394
2395 static void
2396 ctl_free_args(int num_args, struct ctl_be_arg *args)
2397 {
2398         int i;
2399
2400         if (args == NULL)
2401                 return;
2402
2403         for (i = 0; i < num_args; i++) {
2404                 free(args[i].kname, M_CTL);
2405                 free(args[i].kvalue, M_CTL);
2406         }
2407
2408         free(args, M_CTL);
2409 }
2410
2411 static struct ctl_be_arg *
2412 ctl_copyin_args(int num_args, struct ctl_be_arg *uargs,
2413                 char *error_str, size_t error_str_len)
2414 {
2415         struct ctl_be_arg *args;
2416         int i;
2417
2418         args = ctl_copyin_alloc(uargs, num_args * sizeof(*args),
2419                                 error_str, error_str_len);
2420
2421         if (args == NULL)
2422                 goto bailout;
2423
2424         for (i = 0; i < num_args; i++) {
2425                 args[i].kname = NULL;
2426                 args[i].kvalue = NULL;
2427         }
2428
2429         for (i = 0; i < num_args; i++) {
2430                 uint8_t *tmpptr;
2431
2432                 args[i].kname = ctl_copyin_alloc(args[i].name,
2433                         args[i].namelen, error_str, error_str_len);
2434                 if (args[i].kname == NULL)
2435                         goto bailout;
2436
2437                 if (args[i].kname[args[i].namelen - 1] != '\0') {
2438                         snprintf(error_str, error_str_len, "Argument %d "
2439                                  "name is not NUL-terminated", i);
2440                         goto bailout;
2441                 }
2442
2443                 if (args[i].flags & CTL_BEARG_RD) {
2444                         tmpptr = ctl_copyin_alloc(args[i].value,
2445                                 args[i].vallen, error_str, error_str_len);
2446                         if (tmpptr == NULL)
2447                                 goto bailout;
2448                         if ((args[i].flags & CTL_BEARG_ASCII)
2449                          && (tmpptr[args[i].vallen - 1] != '\0')) {
2450                                 snprintf(error_str, error_str_len, "Argument "
2451                                     "%d value is not NUL-terminated", i);
2452                                 free(tmpptr, M_CTL);
2453                                 goto bailout;
2454                         }
2455                         args[i].kvalue = tmpptr;
2456                 } else {
2457                         args[i].kvalue = malloc(args[i].vallen,
2458                             M_CTL, M_WAITOK | M_ZERO);
2459                 }
2460         }
2461
2462         return (args);
2463 bailout:
2464
2465         ctl_free_args(num_args, args);
2466
2467         return (NULL);
2468 }
2469
2470 static void
2471 ctl_copyout_args(int num_args, struct ctl_be_arg *args)
2472 {
2473         int i;
2474
2475         for (i = 0; i < num_args; i++) {
2476                 if (args[i].flags & CTL_BEARG_WR)
2477                         copyout(args[i].kvalue, args[i].value, args[i].vallen);
2478         }
2479 }
2480
2481 /*
2482  * Escape characters that are illegal or not recommended in XML.
2483  */
2484 int
2485 ctl_sbuf_printf_esc(struct sbuf *sb, char *str, int size)
2486 {
2487         char *end = str + size;
2488         int retval;
2489
2490         retval = 0;
2491
2492         for (; *str && str < end; str++) {
2493                 switch (*str) {
2494                 case '&':
2495                         retval = sbuf_printf(sb, "&amp;");
2496                         break;
2497                 case '>':
2498                         retval = sbuf_printf(sb, "&gt;");
2499                         break;
2500                 case '<':
2501                         retval = sbuf_printf(sb, "&lt;");
2502                         break;
2503                 default:
2504                         retval = sbuf_putc(sb, *str);
2505                         break;
2506                 }
2507
2508                 if (retval != 0)
2509                         break;
2510
2511         }
2512
2513         return (retval);
2514 }
2515
2516 static void
2517 ctl_id_sbuf(struct ctl_devid *id, struct sbuf *sb)
2518 {
2519         struct scsi_vpd_id_descriptor *desc;
2520         int i;
2521
2522         if (id == NULL || id->len < 4)
2523                 return;
2524         desc = (struct scsi_vpd_id_descriptor *)id->data;
2525         switch (desc->id_type & SVPD_ID_TYPE_MASK) {
2526         case SVPD_ID_TYPE_T10:
2527                 sbuf_printf(sb, "t10.");
2528                 break;
2529         case SVPD_ID_TYPE_EUI64:
2530                 sbuf_printf(sb, "eui.");
2531                 break;
2532         case SVPD_ID_TYPE_NAA:
2533                 sbuf_printf(sb, "naa.");
2534                 break;
2535         case SVPD_ID_TYPE_SCSI_NAME:
2536                 break;
2537         }
2538         switch (desc->proto_codeset & SVPD_ID_CODESET_MASK) {
2539         case SVPD_ID_CODESET_BINARY:
2540                 for (i = 0; i < desc->length; i++)
2541                         sbuf_printf(sb, "%02x", desc->identifier[i]);
2542                 break;
2543         case SVPD_ID_CODESET_ASCII:
2544                 sbuf_printf(sb, "%.*s", (int)desc->length,
2545                     (char *)desc->identifier);
2546                 break;
2547         case SVPD_ID_CODESET_UTF8:
2548                 sbuf_printf(sb, "%s", (char *)desc->identifier);
2549                 break;
2550         }
2551 }
2552
2553 static int
2554 ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
2555           struct thread *td)
2556 {
2557         struct ctl_softc *softc = dev->si_drv1;
2558         struct ctl_lun *lun;
2559         int retval;
2560
2561         retval = 0;
2562
2563         switch (cmd) {
2564         case CTL_IO:
2565                 retval = ctl_ioctl_io(dev, cmd, addr, flag, td);
2566                 break;
2567         case CTL_ENABLE_PORT:
2568         case CTL_DISABLE_PORT:
2569         case CTL_SET_PORT_WWNS: {
2570                 struct ctl_port *port;
2571                 struct ctl_port_entry *entry;
2572
2573                 entry = (struct ctl_port_entry *)addr;
2574
2575                 mtx_lock(&softc->ctl_lock);
2576                 STAILQ_FOREACH(port, &softc->port_list, links) {
2577                         int action, done;
2578
2579                         if (port->targ_port < softc->port_min ||
2580                             port->targ_port >= softc->port_max)
2581                                 continue;
2582
2583                         action = 0;
2584                         done = 0;
2585                         if ((entry->port_type == CTL_PORT_NONE)
2586                          && (entry->targ_port == port->targ_port)) {
2587                                 /*
2588                                  * If the user only wants to enable or
2589                                  * disable or set WWNs on a specific port,
2590                                  * do the operation and we're done.
2591                                  */
2592                                 action = 1;
2593                                 done = 1;
2594                         } else if (entry->port_type & port->port_type) {
2595                                 /*
2596                                  * Compare the user's type mask with the
2597                                  * particular frontend type to see if we
2598                                  * have a match.
2599                                  */
2600                                 action = 1;
2601                                 done = 0;
2602
2603                                 /*
2604                                  * Make sure the user isn't trying to set
2605                                  * WWNs on multiple ports at the same time.
2606                                  */
2607                                 if (cmd == CTL_SET_PORT_WWNS) {
2608                                         printf("%s: Can't set WWNs on "
2609                                                "multiple ports\n", __func__);
2610                                         retval = EINVAL;
2611                                         break;
2612                                 }
2613                         }
2614                         if (action == 0)
2615                                 continue;
2616
2617                         /*
2618                          * XXX KDM we have to drop the lock here, because
2619                          * the online/offline operations can potentially
2620                          * block.  We need to reference count the frontends
2621                          * so they can't go away,
2622                          */
2623                         if (cmd == CTL_ENABLE_PORT) {
2624                                 mtx_unlock(&softc->ctl_lock);
2625                                 ctl_port_online(port);
2626                                 mtx_lock(&softc->ctl_lock);
2627                         } else if (cmd == CTL_DISABLE_PORT) {
2628                                 mtx_unlock(&softc->ctl_lock);
2629                                 ctl_port_offline(port);
2630                                 mtx_lock(&softc->ctl_lock);
2631                         } else if (cmd == CTL_SET_PORT_WWNS) {
2632                                 ctl_port_set_wwns(port,
2633                                     (entry->flags & CTL_PORT_WWNN_VALID) ?
2634                                     1 : 0, entry->wwnn,
2635                                     (entry->flags & CTL_PORT_WWPN_VALID) ?
2636                                     1 : 0, entry->wwpn);
2637                         }
2638                         if (done != 0)
2639                                 break;
2640                 }
2641                 mtx_unlock(&softc->ctl_lock);
2642                 break;
2643         }
2644         case CTL_GET_OOA: {
2645                 struct ctl_ooa *ooa_hdr;
2646                 struct ctl_ooa_entry *entries;
2647                 uint32_t cur_fill_num;
2648
2649                 ooa_hdr = (struct ctl_ooa *)addr;
2650
2651                 if ((ooa_hdr->alloc_len == 0)
2652                  || (ooa_hdr->alloc_num == 0)) {
2653                         printf("%s: CTL_GET_OOA: alloc len %u and alloc num %u "
2654                                "must be non-zero\n", __func__,
2655                                ooa_hdr->alloc_len, ooa_hdr->alloc_num);
2656                         retval = EINVAL;
2657                         break;
2658                 }
2659
2660                 if (ooa_hdr->alloc_len != (ooa_hdr->alloc_num *
2661                     sizeof(struct ctl_ooa_entry))) {
2662                         printf("%s: CTL_GET_OOA: alloc len %u must be alloc "
2663                                "num %d * sizeof(struct ctl_ooa_entry) %zd\n",
2664                                __func__, ooa_hdr->alloc_len,
2665                                ooa_hdr->alloc_num,sizeof(struct ctl_ooa_entry));
2666                         retval = EINVAL;
2667                         break;
2668                 }
2669
2670                 entries = malloc(ooa_hdr->alloc_len, M_CTL, M_WAITOK | M_ZERO);
2671                 if (entries == NULL) {
2672                         printf("%s: could not allocate %d bytes for OOA "
2673                                "dump\n", __func__, ooa_hdr->alloc_len);
2674                         retval = ENOMEM;
2675                         break;
2676                 }
2677
2678                 mtx_lock(&softc->ctl_lock);
2679                 if (((ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) == 0)
2680                  && ((ooa_hdr->lun_num >= CTL_MAX_LUNS)
2681                   || (softc->ctl_luns[ooa_hdr->lun_num] == NULL))) {
2682                         mtx_unlock(&softc->ctl_lock);
2683                         free(entries, M_CTL);
2684                         printf("%s: CTL_GET_OOA: invalid LUN %ju\n",
2685                                __func__, (uintmax_t)ooa_hdr->lun_num);
2686                         retval = EINVAL;
2687                         break;
2688                 }
2689
2690                 cur_fill_num = 0;
2691
2692                 if (ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) {
2693                         STAILQ_FOREACH(lun, &softc->lun_list, links) {
2694                                 ctl_ioctl_fill_ooa(lun, &cur_fill_num,
2695                                     ooa_hdr, entries);
2696                         }
2697                 } else {
2698                         lun = softc->ctl_luns[ooa_hdr->lun_num];
2699                         ctl_ioctl_fill_ooa(lun, &cur_fill_num, ooa_hdr,
2700                             entries);
2701                 }
2702                 mtx_unlock(&softc->ctl_lock);
2703
2704                 ooa_hdr->fill_num = min(cur_fill_num, ooa_hdr->alloc_num);
2705                 ooa_hdr->fill_len = ooa_hdr->fill_num *
2706                         sizeof(struct ctl_ooa_entry);
2707                 retval = copyout(entries, ooa_hdr->entries, ooa_hdr->fill_len);
2708                 if (retval != 0) {
2709                         printf("%s: error copying out %d bytes for OOA dump\n",
2710                                __func__, ooa_hdr->fill_len);
2711                 }
2712
2713                 getbinuptime(&ooa_hdr->cur_bt);
2714
2715                 if (cur_fill_num > ooa_hdr->alloc_num) {
2716                         ooa_hdr->dropped_num = cur_fill_num -ooa_hdr->alloc_num;
2717                         ooa_hdr->status = CTL_OOA_NEED_MORE_SPACE;
2718                 } else {
2719                         ooa_hdr->dropped_num = 0;
2720                         ooa_hdr->status = CTL_OOA_OK;
2721                 }
2722
2723                 free(entries, M_CTL);
2724                 break;
2725         }
2726         case CTL_DELAY_IO: {
2727                 struct ctl_io_delay_info *delay_info;
2728
2729                 delay_info = (struct ctl_io_delay_info *)addr;
2730
2731 #ifdef CTL_IO_DELAY
2732                 mtx_lock(&softc->ctl_lock);
2733
2734                 if ((delay_info->lun_id >= CTL_MAX_LUNS)
2735                  || (softc->ctl_luns[delay_info->lun_id] == NULL)) {
2736                         delay_info->status = CTL_DELAY_STATUS_INVALID_LUN;
2737                 } else {
2738                         lun = softc->ctl_luns[delay_info->lun_id];
2739                         mtx_lock(&lun->lun_lock);
2740
2741                         delay_info->status = CTL_DELAY_STATUS_OK;
2742
2743                         switch (delay_info->delay_type) {
2744                         case CTL_DELAY_TYPE_CONT:
2745                                 break;
2746                         case CTL_DELAY_TYPE_ONESHOT:
2747                                 break;
2748                         default:
2749                                 delay_info->status =
2750                                         CTL_DELAY_STATUS_INVALID_TYPE;
2751                                 break;
2752                         }
2753
2754                         switch (delay_info->delay_loc) {
2755                         case CTL_DELAY_LOC_DATAMOVE:
2756                                 lun->delay_info.datamove_type =
2757                                         delay_info->delay_type;
2758                                 lun->delay_info.datamove_delay =
2759                                         delay_info->delay_secs;
2760                                 break;
2761                         case CTL_DELAY_LOC_DONE:
2762                                 lun->delay_info.done_type =
2763                                         delay_info->delay_type;
2764                                 lun->delay_info.done_delay =
2765                                         delay_info->delay_secs;
2766                                 break;
2767                         default:
2768                                 delay_info->status =
2769                                         CTL_DELAY_STATUS_INVALID_LOC;
2770                                 break;
2771                         }
2772                         mtx_unlock(&lun->lun_lock);
2773                 }
2774
2775                 mtx_unlock(&softc->ctl_lock);
2776 #else
2777                 delay_info->status = CTL_DELAY_STATUS_NOT_IMPLEMENTED;
2778 #endif /* CTL_IO_DELAY */
2779                 break;
2780         }
2781         case CTL_GETSTATS: {
2782                 struct ctl_stats *stats;
2783                 int i;
2784
2785                 stats = (struct ctl_stats *)addr;
2786
2787                 if ((sizeof(struct ctl_lun_io_stats) * softc->num_luns) >
2788                      stats->alloc_len) {
2789                         stats->status = CTL_SS_NEED_MORE_SPACE;
2790                         stats->num_luns = softc->num_luns;
2791                         break;
2792                 }
2793                 /*
2794                  * XXX KDM no locking here.  If the LUN list changes,
2795                  * things can blow up.
2796                  */
2797                 i = 0;
2798                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
2799                         retval = copyout(&lun->stats, &stats->lun_stats[i++],
2800                                          sizeof(lun->stats));
2801                         if (retval != 0)
2802                                 break;
2803                 }
2804                 stats->num_luns = softc->num_luns;
2805                 stats->fill_len = sizeof(struct ctl_lun_io_stats) *
2806                                  softc->num_luns;
2807                 stats->status = CTL_SS_OK;
2808 #ifdef CTL_TIME_IO
2809                 stats->flags = CTL_STATS_FLAG_TIME_VALID;
2810 #else
2811                 stats->flags = CTL_STATS_FLAG_NONE;
2812 #endif
2813                 getnanouptime(&stats->timestamp);
2814                 break;
2815         }
2816         case CTL_ERROR_INJECT: {
2817                 struct ctl_error_desc *err_desc, *new_err_desc;
2818
2819                 err_desc = (struct ctl_error_desc *)addr;
2820
2821                 new_err_desc = malloc(sizeof(*new_err_desc), M_CTL,
2822                                       M_WAITOK | M_ZERO);
2823                 bcopy(err_desc, new_err_desc, sizeof(*new_err_desc));
2824
2825                 mtx_lock(&softc->ctl_lock);
2826                 lun = softc->ctl_luns[err_desc->lun_id];
2827                 if (lun == NULL) {
2828                         mtx_unlock(&softc->ctl_lock);
2829                         free(new_err_desc, M_CTL);
2830                         printf("%s: CTL_ERROR_INJECT: invalid LUN %ju\n",
2831                                __func__, (uintmax_t)err_desc->lun_id);
2832                         retval = EINVAL;
2833                         break;
2834                 }
2835                 mtx_lock(&lun->lun_lock);
2836                 mtx_unlock(&softc->ctl_lock);
2837
2838                 /*
2839                  * We could do some checking here to verify the validity
2840                  * of the request, but given the complexity of error
2841                  * injection requests, the checking logic would be fairly
2842                  * complex.
2843                  *
2844                  * For now, if the request is invalid, it just won't get
2845                  * executed and might get deleted.
2846                  */
2847                 STAILQ_INSERT_TAIL(&lun->error_list, new_err_desc, links);
2848
2849                 /*
2850                  * XXX KDM check to make sure the serial number is unique,
2851                  * in case we somehow manage to wrap.  That shouldn't
2852                  * happen for a very long time, but it's the right thing to
2853                  * do.
2854                  */
2855                 new_err_desc->serial = lun->error_serial;
2856                 err_desc->serial = lun->error_serial;
2857                 lun->error_serial++;
2858
2859                 mtx_unlock(&lun->lun_lock);
2860                 break;
2861         }
2862         case CTL_ERROR_INJECT_DELETE: {
2863                 struct ctl_error_desc *delete_desc, *desc, *desc2;
2864                 int delete_done;
2865
2866                 delete_desc = (struct ctl_error_desc *)addr;
2867                 delete_done = 0;
2868
2869                 mtx_lock(&softc->ctl_lock);
2870                 lun = softc->ctl_luns[delete_desc->lun_id];
2871                 if (lun == NULL) {
2872                         mtx_unlock(&softc->ctl_lock);
2873                         printf("%s: CTL_ERROR_INJECT_DELETE: invalid LUN %ju\n",
2874                                __func__, (uintmax_t)delete_desc->lun_id);
2875                         retval = EINVAL;
2876                         break;
2877                 }
2878                 mtx_lock(&lun->lun_lock);
2879                 mtx_unlock(&softc->ctl_lock);
2880                 STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
2881                         if (desc->serial != delete_desc->serial)
2882                                 continue;
2883
2884                         STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc,
2885                                       links);
2886                         free(desc, M_CTL);
2887                         delete_done = 1;
2888                 }
2889                 mtx_unlock(&lun->lun_lock);
2890                 if (delete_done == 0) {
2891                         printf("%s: CTL_ERROR_INJECT_DELETE: can't find "
2892                                "error serial %ju on LUN %u\n", __func__,
2893                                delete_desc->serial, delete_desc->lun_id);
2894                         retval = EINVAL;
2895                         break;
2896                 }
2897                 break;
2898         }
2899         case CTL_DUMP_STRUCTS: {
2900                 int i, j, k;
2901                 struct ctl_port *port;
2902                 struct ctl_frontend *fe;
2903
2904                 mtx_lock(&softc->ctl_lock);
2905                 printf("CTL Persistent Reservation information start:\n");
2906                 for (i = 0; i < CTL_MAX_LUNS; i++) {
2907                         lun = softc->ctl_luns[i];
2908
2909                         if ((lun == NULL)
2910                          || ((lun->flags & CTL_LUN_DISABLED) != 0))
2911                                 continue;
2912
2913                         for (j = 0; j < CTL_MAX_PORTS; j++) {
2914                                 if (lun->pr_keys[j] == NULL)
2915                                         continue;
2916                                 for (k = 0; k < CTL_MAX_INIT_PER_PORT; k++){
2917                                         if (lun->pr_keys[j][k] == 0)
2918                                                 continue;
2919                                         printf("  LUN %d port %d iid %d key "
2920                                                "%#jx\n", i, j, k,
2921                                                (uintmax_t)lun->pr_keys[j][k]);
2922                                 }
2923                         }
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 < CTL_MAX_LUNS; j++) {
3305                                         plun = ctl_lun_map_from_port(port, j);
3306                                         if (plun >= CTL_MAX_LUNS)
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                                     CTL_MAX_LUNS)
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 < CTL_MAX_LUNS) {
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 if (lm->plun == UINT32_MAX) {
3393                         if (lm->lun == UINT32_MAX)
3394                                 retval = ctl_lun_map_deinit(port);
3395                         else
3396                                 retval = ctl_lun_map_init(port);
3397                 } else
3398                         return (ENXIO);
3399                 if (port->status & CTL_PORT_STATUS_ONLINE)
3400                         ctl_isc_announce_port(port);
3401                 break;
3402         }
3403         default: {
3404                 /* XXX KDM should we fix this? */
3405 #if 0
3406                 struct ctl_backend_driver *backend;
3407                 unsigned int type;
3408                 int found;
3409
3410                 found = 0;
3411
3412                 /*
3413                  * We encode the backend type as the ioctl type for backend
3414                  * ioctls.  So parse it out here, and then search for a
3415                  * backend of this type.
3416                  */
3417                 type = _IOC_TYPE(cmd);
3418
3419                 STAILQ_FOREACH(backend, &softc->be_list, links) {
3420                         if (backend->type == type) {
3421                                 found = 1;
3422                                 break;
3423                         }
3424                 }
3425                 if (found == 0) {
3426                         printf("ctl: unknown ioctl command %#lx or backend "
3427                                "%d\n", cmd, type);
3428                         retval = EINVAL;
3429                         break;
3430                 }
3431                 retval = backend->ioctl(dev, cmd, addr, flag, td);
3432 #endif
3433                 retval = ENOTTY;
3434                 break;
3435         }
3436         }
3437         return (retval);
3438 }
3439
3440 uint32_t
3441 ctl_get_initindex(struct ctl_nexus *nexus)
3442 {
3443         return (nexus->initid + (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3444 }
3445
3446 int
3447 ctl_lun_map_init(struct ctl_port *port)
3448 {
3449         struct ctl_softc *softc = port->ctl_softc;
3450         struct ctl_lun *lun;
3451         uint32_t i;
3452
3453         if (port->lun_map == NULL)
3454                 port->lun_map = malloc(sizeof(uint32_t) * CTL_MAX_LUNS,
3455                     M_CTL, M_NOWAIT);
3456         if (port->lun_map == NULL)
3457                 return (ENOMEM);
3458         for (i = 0; i < CTL_MAX_LUNS; i++)
3459                 port->lun_map[i] = UINT32_MAX;
3460         if (port->status & CTL_PORT_STATUS_ONLINE) {
3461                 if (port->lun_disable != NULL) {
3462                         STAILQ_FOREACH(lun, &softc->lun_list, links)
3463                                 port->lun_disable(port->targ_lun_arg, lun->lun);
3464                 }
3465                 ctl_isc_announce_port(port);
3466         }
3467         return (0);
3468 }
3469
3470 int
3471 ctl_lun_map_deinit(struct ctl_port *port)
3472 {
3473         struct ctl_softc *softc = port->ctl_softc;
3474         struct ctl_lun *lun;
3475
3476         if (port->lun_map == NULL)
3477                 return (0);
3478         free(port->lun_map, M_CTL);
3479         port->lun_map = NULL;
3480         if (port->status & CTL_PORT_STATUS_ONLINE) {
3481                 if (port->lun_enable != NULL) {
3482                         STAILQ_FOREACH(lun, &softc->lun_list, links)
3483                                 port->lun_enable(port->targ_lun_arg, lun->lun);
3484                 }
3485                 ctl_isc_announce_port(port);
3486         }
3487         return (0);
3488 }
3489
3490 int
3491 ctl_lun_map_set(struct ctl_port *port, uint32_t plun, uint32_t glun)
3492 {
3493         int status;
3494         uint32_t old;
3495
3496         if (port->lun_map == NULL) {
3497                 status = ctl_lun_map_init(port);
3498                 if (status != 0)
3499                         return (status);
3500         }
3501         old = port->lun_map[plun];
3502         port->lun_map[plun] = glun;
3503         if ((port->status & CTL_PORT_STATUS_ONLINE) && old >= CTL_MAX_LUNS) {
3504                 if (port->lun_enable != NULL)
3505                         port->lun_enable(port->targ_lun_arg, plun);
3506                 ctl_isc_announce_port(port);
3507         }
3508         return (0);
3509 }
3510
3511 int
3512 ctl_lun_map_unset(struct ctl_port *port, uint32_t plun)
3513 {
3514         uint32_t old;
3515
3516         if (port->lun_map == NULL)
3517                 return (0);
3518         old = port->lun_map[plun];
3519         port->lun_map[plun] = UINT32_MAX;
3520         if ((port->status & CTL_PORT_STATUS_ONLINE) && old < CTL_MAX_LUNS) {
3521                 if (port->lun_disable != NULL)
3522                         port->lun_disable(port->targ_lun_arg, plun);
3523                 ctl_isc_announce_port(port);
3524         }
3525         return (0);
3526 }
3527
3528 uint32_t
3529 ctl_lun_map_from_port(struct ctl_port *port, uint32_t lun_id)
3530 {
3531
3532         if (port == NULL)
3533                 return (UINT32_MAX);
3534         if (port->lun_map == NULL || lun_id >= CTL_MAX_LUNS)
3535                 return (lun_id);
3536         return (port->lun_map[lun_id]);
3537 }
3538
3539 uint32_t
3540 ctl_lun_map_to_port(struct ctl_port *port, uint32_t lun_id)
3541 {
3542         uint32_t i;
3543
3544         if (port == NULL)
3545                 return (UINT32_MAX);
3546         if (port->lun_map == NULL)
3547                 return (lun_id);
3548         for (i = 0; i < CTL_MAX_LUNS; i++) {
3549                 if (port->lun_map[i] == lun_id)
3550                         return (i);
3551         }
3552         return (UINT32_MAX);
3553 }
3554
3555 uint32_t
3556 ctl_decode_lun(uint64_t encoded)
3557 {
3558         uint8_t lun[8];
3559         uint32_t result = 0xffffffff;
3560
3561         be64enc(lun, encoded);
3562         switch (lun[0] & RPL_LUNDATA_ATYP_MASK) {
3563         case RPL_LUNDATA_ATYP_PERIPH:
3564                 if ((lun[0] & 0x3f) == 0 && lun[2] == 0 && lun[3] == 0 &&
3565                     lun[4] == 0 && lun[5] == 0 && lun[6] == 0 && lun[7] == 0)
3566                         result = lun[1];
3567                 break;
3568         case RPL_LUNDATA_ATYP_FLAT:
3569                 if (lun[2] == 0 && lun[3] == 0 && lun[4] == 0 && lun[5] == 0 &&
3570                     lun[6] == 0 && lun[7] == 0)
3571                         result = ((lun[0] & 0x3f) << 8) + lun[1];
3572                 break;
3573         case RPL_LUNDATA_ATYP_EXTLUN:
3574                 switch (lun[0] & RPL_LUNDATA_EXT_EAM_MASK) {
3575                 case 0x02:
3576                         switch (lun[0] & RPL_LUNDATA_EXT_LEN_MASK) {
3577                         case 0x00:
3578                                 result = lun[1];
3579                                 break;
3580                         case 0x10:
3581                                 result = (lun[1] << 16) + (lun[2] << 8) +
3582                                     lun[3];
3583                                 break;
3584                         case 0x20:
3585                                 if (lun[1] == 0 && lun[6] == 0 && lun[7] == 0)
3586                                         result = (lun[2] << 24) +
3587                                             (lun[3] << 16) + (lun[4] << 8) +
3588                                             lun[5];
3589                                 break;
3590                         }
3591                         break;
3592                 case RPL_LUNDATA_EXT_EAM_NOT_SPEC:
3593                         result = 0xffffffff;
3594                         break;
3595                 }
3596                 break;
3597         }
3598         return (result);
3599 }
3600
3601 uint64_t
3602 ctl_encode_lun(uint32_t decoded)
3603 {
3604         uint64_t l = decoded;
3605
3606         if (l <= 0xff)
3607                 return (((uint64_t)RPL_LUNDATA_ATYP_PERIPH << 56) | (l << 48));
3608         if (l <= 0x3fff)
3609                 return (((uint64_t)RPL_LUNDATA_ATYP_FLAT << 56) | (l << 48));
3610         if (l <= 0xffffff)
3611                 return (((uint64_t)(RPL_LUNDATA_ATYP_EXTLUN | 0x12) << 56) |
3612                     (l << 32));
3613         return ((((uint64_t)RPL_LUNDATA_ATYP_EXTLUN | 0x22) << 56) | (l << 16));
3614 }
3615
3616 static struct ctl_port *
3617 ctl_io_port(struct ctl_io_hdr *io_hdr)
3618 {
3619
3620         return (control_softc->ctl_ports[io_hdr->nexus.targ_port]);
3621 }
3622
3623 int
3624 ctl_ffz(uint32_t *mask, uint32_t first, uint32_t last)
3625 {
3626         int i;
3627
3628         for (i = first; i < last; i++) {
3629                 if ((mask[i / 32] & (1 << (i % 32))) == 0)
3630                         return (i);
3631         }
3632         return (-1);
3633 }
3634
3635 int
3636 ctl_set_mask(uint32_t *mask, uint32_t bit)
3637 {
3638         uint32_t chunk, piece;
3639
3640         chunk = bit >> 5;
3641         piece = bit % (sizeof(uint32_t) * 8);
3642
3643         if ((mask[chunk] & (1 << piece)) != 0)
3644                 return (-1);
3645         else
3646                 mask[chunk] |= (1 << piece);
3647
3648         return (0);
3649 }
3650
3651 int
3652 ctl_clear_mask(uint32_t *mask, uint32_t bit)
3653 {
3654         uint32_t chunk, piece;
3655
3656         chunk = bit >> 5;
3657         piece = bit % (sizeof(uint32_t) * 8);
3658
3659         if ((mask[chunk] & (1 << piece)) == 0)
3660                 return (-1);
3661         else
3662                 mask[chunk] &= ~(1 << piece);
3663
3664         return (0);
3665 }
3666
3667 int
3668 ctl_is_set(uint32_t *mask, uint32_t bit)
3669 {
3670         uint32_t chunk, piece;
3671
3672         chunk = bit >> 5;
3673         piece = bit % (sizeof(uint32_t) * 8);
3674
3675         if ((mask[chunk] & (1 << piece)) == 0)
3676                 return (0);
3677         else
3678                 return (1);
3679 }
3680
3681 static uint64_t
3682 ctl_get_prkey(struct ctl_lun *lun, uint32_t residx)
3683 {
3684         uint64_t *t;
3685
3686         t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3687         if (t == NULL)
3688                 return (0);
3689         return (t[residx % CTL_MAX_INIT_PER_PORT]);
3690 }
3691
3692 static void
3693 ctl_clr_prkey(struct ctl_lun *lun, uint32_t residx)
3694 {
3695         uint64_t *t;
3696
3697         t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3698         if (t == NULL)
3699                 return;
3700         t[residx % CTL_MAX_INIT_PER_PORT] = 0;
3701 }
3702
3703 static void
3704 ctl_alloc_prkey(struct ctl_lun *lun, uint32_t residx)
3705 {
3706         uint64_t *p;
3707         u_int i;
3708
3709         i = residx/CTL_MAX_INIT_PER_PORT;
3710         if (lun->pr_keys[i] != NULL)
3711                 return;
3712         mtx_unlock(&lun->lun_lock);
3713         p = malloc(sizeof(uint64_t) * CTL_MAX_INIT_PER_PORT, M_CTL,
3714             M_WAITOK | M_ZERO);
3715         mtx_lock(&lun->lun_lock);
3716         if (lun->pr_keys[i] == NULL)
3717                 lun->pr_keys[i] = p;
3718         else
3719                 free(p, M_CTL);
3720 }
3721
3722 static void
3723 ctl_set_prkey(struct ctl_lun *lun, uint32_t residx, uint64_t key)
3724 {
3725         uint64_t *t;
3726
3727         t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3728         KASSERT(t != NULL, ("prkey %d is not allocated", residx));
3729         t[residx % CTL_MAX_INIT_PER_PORT] = key;
3730 }
3731
3732 /*
3733  * ctl_softc, pool_name, total_ctl_io are passed in.
3734  * npool is passed out.
3735  */
3736 int
3737 ctl_pool_create(struct ctl_softc *ctl_softc, const char *pool_name,
3738                 uint32_t total_ctl_io, void **npool)
3739 {
3740 #ifdef IO_POOLS
3741         struct ctl_io_pool *pool;
3742
3743         pool = (struct ctl_io_pool *)malloc(sizeof(*pool), M_CTL,
3744                                             M_NOWAIT | M_ZERO);
3745         if (pool == NULL)
3746                 return (ENOMEM);
3747
3748         snprintf(pool->name, sizeof(pool->name), "CTL IO %s", pool_name);
3749         pool->ctl_softc = ctl_softc;
3750         pool->zone = uma_zsecond_create(pool->name, NULL,
3751             NULL, NULL, NULL, ctl_softc->io_zone);
3752         /* uma_prealloc(pool->zone, total_ctl_io); */
3753
3754         *npool = pool;
3755 #else
3756         *npool = ctl_softc->io_zone;
3757 #endif
3758         return (0);
3759 }
3760
3761 void
3762 ctl_pool_free(struct ctl_io_pool *pool)
3763 {
3764
3765         if (pool == NULL)
3766                 return;
3767
3768 #ifdef IO_POOLS
3769         uma_zdestroy(pool->zone);
3770         free(pool, M_CTL);
3771 #endif
3772 }
3773
3774 union ctl_io *
3775 ctl_alloc_io(void *pool_ref)
3776 {
3777         union ctl_io *io;
3778 #ifdef IO_POOLS
3779         struct ctl_io_pool *pool = (struct ctl_io_pool *)pool_ref;
3780
3781         io = uma_zalloc(pool->zone, M_WAITOK);
3782 #else
3783         io = uma_zalloc((uma_zone_t)pool_ref, M_WAITOK);
3784 #endif
3785         if (io != NULL)
3786                 io->io_hdr.pool = pool_ref;
3787         return (io);
3788 }
3789
3790 union ctl_io *
3791 ctl_alloc_io_nowait(void *pool_ref)
3792 {
3793         union ctl_io *io;
3794 #ifdef IO_POOLS
3795         struct ctl_io_pool *pool = (struct ctl_io_pool *)pool_ref;
3796
3797         io = uma_zalloc(pool->zone, M_NOWAIT);
3798 #else
3799         io = uma_zalloc((uma_zone_t)pool_ref, M_NOWAIT);
3800 #endif
3801         if (io != NULL)
3802                 io->io_hdr.pool = pool_ref;
3803         return (io);
3804 }
3805
3806 void
3807 ctl_free_io(union ctl_io *io)
3808 {
3809 #ifdef IO_POOLS
3810         struct ctl_io_pool *pool;
3811 #endif
3812
3813         if (io == NULL)
3814                 return;
3815
3816 #ifdef IO_POOLS
3817         pool = (struct ctl_io_pool *)io->io_hdr.pool;
3818         uma_zfree(pool->zone, io);
3819 #else
3820         uma_zfree((uma_zone_t)io->io_hdr.pool, io);
3821 #endif
3822 }
3823
3824 void
3825 ctl_zero_io(union ctl_io *io)
3826 {
3827         void *pool_ref;
3828
3829         if (io == NULL)
3830                 return;
3831
3832         /*
3833          * May need to preserve linked list pointers at some point too.
3834          */
3835         pool_ref = io->io_hdr.pool;
3836         memset(io, 0, sizeof(*io));
3837         io->io_hdr.pool = pool_ref;
3838 }
3839
3840 int
3841 ctl_expand_number(const char *buf, uint64_t *num)
3842 {
3843         char *endptr;
3844         uint64_t number;
3845         unsigned shift;
3846
3847         number = strtoq(buf, &endptr, 0);
3848
3849         switch (tolower((unsigned char)*endptr)) {
3850         case 'e':
3851                 shift = 60;
3852                 break;
3853         case 'p':
3854                 shift = 50;
3855                 break;
3856         case 't':
3857                 shift = 40;
3858                 break;
3859         case 'g':
3860                 shift = 30;
3861                 break;
3862         case 'm':
3863                 shift = 20;
3864                 break;
3865         case 'k':
3866                 shift = 10;
3867                 break;
3868         case 'b':
3869         case '\0': /* No unit. */
3870                 *num = number;
3871                 return (0);
3872         default:
3873                 /* Unrecognized unit. */
3874                 return (-1);
3875         }
3876
3877         if ((number << shift) >> shift != number) {
3878                 /* Overflow */
3879                 return (-1);
3880         }
3881         *num = number << shift;
3882         return (0);
3883 }
3884
3885
3886 /*
3887  * This routine could be used in the future to load default and/or saved
3888  * mode page parameters for a particuar lun.
3889  */
3890 static int
3891 ctl_init_page_index(struct ctl_lun *lun)
3892 {
3893         int i, page_code;
3894         struct ctl_page_index *page_index;
3895         const char *value;
3896         uint64_t ival;
3897
3898         memcpy(&lun->mode_pages.index, page_index_template,
3899                sizeof(page_index_template));
3900
3901         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
3902
3903                 page_index = &lun->mode_pages.index[i];
3904                 if (lun->be_lun->lun_type == T_DIRECT &&
3905                     (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
3906                         continue;
3907                 if (lun->be_lun->lun_type == T_PROCESSOR &&
3908                     (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
3909                         continue;
3910                 if (lun->be_lun->lun_type == T_CDROM &&
3911                     (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
3912                         continue;
3913
3914                 page_code = page_index->page_code & SMPH_PC_MASK;
3915                 switch (page_code) {
3916                 case SMS_RW_ERROR_RECOVERY_PAGE: {
3917                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
3918                             ("subpage %#x for page %#x is incorrect!",
3919                             page_index->subpage, page_code));
3920                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_CURRENT],
3921                                &rw_er_page_default,
3922                                sizeof(rw_er_page_default));
3923                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_CHANGEABLE],
3924                                &rw_er_page_changeable,
3925                                sizeof(rw_er_page_changeable));
3926                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_DEFAULT],
3927                                &rw_er_page_default,
3928                                sizeof(rw_er_page_default));
3929                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_SAVED],
3930                                &rw_er_page_default,
3931                                sizeof(rw_er_page_default));
3932                         page_index->page_data =
3933                                 (uint8_t *)lun->mode_pages.rw_er_page;
3934                         break;
3935                 }
3936                 case SMS_FORMAT_DEVICE_PAGE: {
3937                         struct scsi_format_page *format_page;
3938
3939                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
3940                             ("subpage %#x for page %#x is incorrect!",
3941                             page_index->subpage, page_code));
3942
3943                         /*
3944                          * Sectors per track are set above.  Bytes per
3945                          * sector need to be set here on a per-LUN basis.
3946                          */
3947                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_CURRENT],
3948                                &format_page_default,
3949                                sizeof(format_page_default));
3950                         memcpy(&lun->mode_pages.format_page[
3951                                CTL_PAGE_CHANGEABLE], &format_page_changeable,
3952                                sizeof(format_page_changeable));
3953                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_DEFAULT],
3954                                &format_page_default,
3955                                sizeof(format_page_default));
3956                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_SAVED],
3957                                &format_page_default,
3958                                sizeof(format_page_default));
3959
3960                         format_page = &lun->mode_pages.format_page[
3961                                 CTL_PAGE_CURRENT];
3962                         scsi_ulto2b(lun->be_lun->blocksize,
3963                                     format_page->bytes_per_sector);
3964
3965                         format_page = &lun->mode_pages.format_page[
3966                                 CTL_PAGE_DEFAULT];
3967                         scsi_ulto2b(lun->be_lun->blocksize,
3968                                     format_page->bytes_per_sector);
3969
3970                         format_page = &lun->mode_pages.format_page[
3971                                 CTL_PAGE_SAVED];
3972                         scsi_ulto2b(lun->be_lun->blocksize,
3973                                     format_page->bytes_per_sector);
3974
3975                         page_index->page_data =
3976                                 (uint8_t *)lun->mode_pages.format_page;
3977                         break;
3978                 }
3979                 case SMS_RIGID_DISK_PAGE: {
3980                         struct scsi_rigid_disk_page *rigid_disk_page;
3981                         uint32_t sectors_per_cylinder;
3982                         uint64_t cylinders;
3983 #ifndef __XSCALE__
3984                         int shift;
3985 #endif /* !__XSCALE__ */
3986
3987                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
3988                             ("subpage %#x for page %#x is incorrect!",
3989                             page_index->subpage, page_code));
3990
3991                         /*
3992                          * Rotation rate and sectors per track are set
3993                          * above.  We calculate the cylinders here based on
3994                          * capacity.  Due to the number of heads and
3995                          * sectors per track we're using, smaller arrays
3996                          * may turn out to have 0 cylinders.  Linux and
3997                          * FreeBSD don't pay attention to these mode pages
3998                          * to figure out capacity, but Solaris does.  It
3999                          * seems to deal with 0 cylinders just fine, and
4000                          * works out a fake geometry based on the capacity.
4001                          */
4002                         memcpy(&lun->mode_pages.rigid_disk_page[
4003                                CTL_PAGE_DEFAULT], &rigid_disk_page_default,
4004                                sizeof(rigid_disk_page_default));
4005                         memcpy(&lun->mode_pages.rigid_disk_page[
4006                                CTL_PAGE_CHANGEABLE],&rigid_disk_page_changeable,
4007                                sizeof(rigid_disk_page_changeable));
4008
4009                         sectors_per_cylinder = CTL_DEFAULT_SECTORS_PER_TRACK *
4010                                 CTL_DEFAULT_HEADS;
4011
4012                         /*
4013                          * The divide method here will be more accurate,
4014                          * probably, but results in floating point being
4015                          * used in the kernel on i386 (__udivdi3()).  On the
4016                          * XScale, though, __udivdi3() is implemented in
4017                          * software.
4018                          *
4019                          * The shift method for cylinder calculation is
4020                          * accurate if sectors_per_cylinder is a power of
4021                          * 2.  Otherwise it might be slightly off -- you
4022                          * might have a bit of a truncation problem.
4023                          */
4024 #ifdef  __XSCALE__
4025                         cylinders = (lun->be_lun->maxlba + 1) /
4026                                 sectors_per_cylinder;
4027 #else
4028                         for (shift = 31; shift > 0; shift--) {
4029                                 if (sectors_per_cylinder & (1 << shift))
4030                                         break;
4031                         }
4032                         cylinders = (lun->be_lun->maxlba + 1) >> shift;
4033 #endif
4034
4035                         /*
4036                          * We've basically got 3 bytes, or 24 bits for the
4037                          * cylinder size in the mode page.  If we're over,
4038                          * just round down to 2^24.
4039                          */
4040                         if (cylinders > 0xffffff)
4041                                 cylinders = 0xffffff;
4042
4043                         rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4044                                 CTL_PAGE_DEFAULT];
4045                         scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4046
4047                         if ((value = ctl_get_opt(&lun->be_lun->options,
4048                             "rpm")) != NULL) {
4049                                 scsi_ulto2b(strtol(value, NULL, 0),
4050                                      rigid_disk_page->rotation_rate);
4051                         }
4052
4053                         memcpy(&lun->mode_pages.rigid_disk_page[CTL_PAGE_CURRENT],
4054                                &lun->mode_pages.rigid_disk_page[CTL_PAGE_DEFAULT],
4055                                sizeof(rigid_disk_page_default));
4056                         memcpy(&lun->mode_pages.rigid_disk_page[CTL_PAGE_SAVED],
4057                                &lun->mode_pages.rigid_disk_page[CTL_PAGE_DEFAULT],
4058                                sizeof(rigid_disk_page_default));
4059
4060                         page_index->page_data =
4061                                 (uint8_t *)lun->mode_pages.rigid_disk_page;
4062                         break;
4063                 }
4064                 case SMS_VERIFY_ERROR_RECOVERY_PAGE: {
4065                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4066                             ("subpage %#x for page %#x is incorrect!",
4067                             page_index->subpage, page_code));
4068                         memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_CURRENT],
4069                                &verify_er_page_default,
4070                                sizeof(verify_er_page_default));
4071                         memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_CHANGEABLE],
4072                                &verify_er_page_changeable,
4073                                sizeof(verify_er_page_changeable));
4074                         memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_DEFAULT],
4075                                &verify_er_page_default,
4076                                sizeof(verify_er_page_default));
4077                         memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_SAVED],
4078                                &verify_er_page_default,
4079                                sizeof(verify_er_page_default));
4080                         page_index->page_data =
4081                                 (uint8_t *)lun->mode_pages.verify_er_page;
4082                         break;
4083                 }
4084                 case SMS_CACHING_PAGE: {
4085                         struct scsi_caching_page *caching_page;
4086
4087                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4088                             ("subpage %#x for page %#x is incorrect!",
4089                             page_index->subpage, page_code));
4090                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_DEFAULT],
4091                                &caching_page_default,
4092                                sizeof(caching_page_default));
4093                         memcpy(&lun->mode_pages.caching_page[
4094                                CTL_PAGE_CHANGEABLE], &caching_page_changeable,
4095                                sizeof(caching_page_changeable));
4096                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4097                                &caching_page_default,
4098                                sizeof(caching_page_default));
4099                         caching_page = &lun->mode_pages.caching_page[
4100                             CTL_PAGE_SAVED];
4101                         value = ctl_get_opt(&lun->be_lun->options, "writecache");
4102                         if (value != NULL && strcmp(value, "off") == 0)
4103                                 caching_page->flags1 &= ~SCP_WCE;
4104                         value = ctl_get_opt(&lun->be_lun->options, "readcache");
4105                         if (value != NULL && strcmp(value, "off") == 0)
4106                                 caching_page->flags1 |= SCP_RCD;
4107                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_CURRENT],
4108                                &lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4109                                sizeof(caching_page_default));
4110                         page_index->page_data =
4111                                 (uint8_t *)lun->mode_pages.caching_page;
4112                         break;
4113                 }
4114                 case SMS_CONTROL_MODE_PAGE: {
4115                         switch (page_index->subpage) {
4116                         case SMS_SUBPAGE_PAGE_0: {
4117                                 struct scsi_control_page *control_page;
4118
4119                                 memcpy(&lun->mode_pages.control_page[
4120                                     CTL_PAGE_DEFAULT],
4121                                        &control_page_default,
4122                                        sizeof(control_page_default));
4123                                 memcpy(&lun->mode_pages.control_page[
4124                                     CTL_PAGE_CHANGEABLE],
4125                                        &control_page_changeable,
4126                                        sizeof(control_page_changeable));
4127                                 memcpy(&lun->mode_pages.control_page[
4128                                     CTL_PAGE_SAVED],
4129                                        &control_page_default,
4130                                        sizeof(control_page_default));
4131                                 control_page = &lun->mode_pages.control_page[
4132                                     CTL_PAGE_SAVED];
4133                                 value = ctl_get_opt(&lun->be_lun->options,
4134                                     "reordering");
4135                                 if (value != NULL &&
4136                                     strcmp(value, "unrestricted") == 0) {
4137                                         control_page->queue_flags &=
4138                                             ~SCP_QUEUE_ALG_MASK;
4139                                         control_page->queue_flags |=
4140                                             SCP_QUEUE_ALG_UNRESTRICTED;
4141                                 }
4142                                 memcpy(&lun->mode_pages.control_page[
4143                                     CTL_PAGE_CURRENT],
4144                                        &lun->mode_pages.control_page[
4145                                     CTL_PAGE_SAVED],
4146                                        sizeof(control_page_default));
4147                                 page_index->page_data =
4148                                     (uint8_t *)lun->mode_pages.control_page;
4149                                 break;
4150                         }
4151                         case 0x01:
4152                                 memcpy(&lun->mode_pages.control_ext_page[
4153                                     CTL_PAGE_DEFAULT],
4154                                        &control_ext_page_default,
4155                                        sizeof(control_ext_page_default));
4156                                 memcpy(&lun->mode_pages.control_ext_page[
4157                                     CTL_PAGE_CHANGEABLE],
4158                                        &control_ext_page_changeable,
4159                                        sizeof(control_ext_page_changeable));
4160                                 memcpy(&lun->mode_pages.control_ext_page[
4161                                     CTL_PAGE_SAVED],
4162                                        &control_ext_page_default,
4163                                        sizeof(control_ext_page_default));
4164                                 memcpy(&lun->mode_pages.control_ext_page[
4165                                     CTL_PAGE_CURRENT],
4166                                        &lun->mode_pages.control_ext_page[
4167                                     CTL_PAGE_SAVED],
4168                                        sizeof(control_ext_page_default));
4169                                 page_index->page_data =
4170                                     (uint8_t *)lun->mode_pages.control_ext_page;
4171                                 break;
4172                         default:
4173                                 panic("subpage %#x for page %#x is incorrect!",
4174                                       page_index->subpage, page_code);
4175                         }
4176                         break;
4177                 }
4178                 case SMS_INFO_EXCEPTIONS_PAGE: {
4179                         switch (page_index->subpage) {
4180                         case SMS_SUBPAGE_PAGE_0:
4181                                 memcpy(&lun->mode_pages.ie_page[CTL_PAGE_CURRENT],
4182                                        &ie_page_default,
4183                                        sizeof(ie_page_default));
4184                                 memcpy(&lun->mode_pages.ie_page[
4185                                        CTL_PAGE_CHANGEABLE], &ie_page_changeable,
4186                                        sizeof(ie_page_changeable));
4187                                 memcpy(&lun->mode_pages.ie_page[CTL_PAGE_DEFAULT],
4188                                        &ie_page_default,
4189                                        sizeof(ie_page_default));
4190                                 memcpy(&lun->mode_pages.ie_page[CTL_PAGE_SAVED],
4191                                        &ie_page_default,
4192                                        sizeof(ie_page_default));
4193                                 page_index->page_data =
4194                                         (uint8_t *)lun->mode_pages.ie_page;
4195                                 break;
4196                         case 0x02: {
4197                                 struct ctl_logical_block_provisioning_page *page;
4198
4199                                 memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_DEFAULT],
4200                                        &lbp_page_default,
4201                                        sizeof(lbp_page_default));
4202                                 memcpy(&lun->mode_pages.lbp_page[
4203                                        CTL_PAGE_CHANGEABLE], &lbp_page_changeable,
4204                                        sizeof(lbp_page_changeable));
4205                                 memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_SAVED],
4206                                        &lbp_page_default,
4207                                        sizeof(lbp_page_default));
4208                                 page = &lun->mode_pages.lbp_page[CTL_PAGE_SAVED];
4209                                 value = ctl_get_opt(&lun->be_lun->options,
4210                                     "avail-threshold");
4211                                 if (value != NULL &&
4212                                     ctl_expand_number(value, &ival) == 0) {
4213                                         page->descr[0].flags |= SLBPPD_ENABLED |
4214                                             SLBPPD_ARMING_DEC;
4215                                         if (lun->be_lun->blocksize)
4216                                                 ival /= lun->be_lun->blocksize;
4217                                         else
4218                                                 ival /= 512;
4219                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4220                                             page->descr[0].count);
4221                                 }
4222                                 value = ctl_get_opt(&lun->be_lun->options,
4223                                     "used-threshold");
4224                                 if (value != NULL &&
4225                                     ctl_expand_number(value, &ival) == 0) {
4226                                         page->descr[1].flags |= SLBPPD_ENABLED |
4227                                             SLBPPD_ARMING_INC;
4228                                         if (lun->be_lun->blocksize)
4229                                                 ival /= lun->be_lun->blocksize;
4230                                         else
4231                                                 ival /= 512;
4232                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4233                                             page->descr[1].count);
4234                                 }
4235                                 value = ctl_get_opt(&lun->be_lun->options,
4236                                     "pool-avail-threshold");
4237                                 if (value != NULL &&
4238                                     ctl_expand_number(value, &ival) == 0) {
4239                                         page->descr[2].flags |= SLBPPD_ENABLED |
4240                                             SLBPPD_ARMING_DEC;
4241                                         if (lun->be_lun->blocksize)
4242                                                 ival /= lun->be_lun->blocksize;
4243                                         else
4244                                                 ival /= 512;
4245                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4246                                             page->descr[2].count);
4247                                 }
4248                                 value = ctl_get_opt(&lun->be_lun->options,
4249                                     "pool-used-threshold");
4250                                 if (value != NULL &&
4251                                     ctl_expand_number(value, &ival) == 0) {
4252                                         page->descr[3].flags |= SLBPPD_ENABLED |
4253                                             SLBPPD_ARMING_INC;
4254                                         if (lun->be_lun->blocksize)
4255                                                 ival /= lun->be_lun->blocksize;
4256                                         else
4257                                                 ival /= 512;
4258                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4259                                             page->descr[3].count);
4260                                 }
4261                                 memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_CURRENT],
4262                                        &lun->mode_pages.lbp_page[CTL_PAGE_SAVED],
4263                                        sizeof(lbp_page_default));
4264                                 page_index->page_data =
4265                                         (uint8_t *)lun->mode_pages.lbp_page;
4266                                 break;
4267                         }
4268                         default:
4269                                 panic("subpage %#x for page %#x is incorrect!",
4270                                       page_index->subpage, page_code);
4271                         }
4272                         break;
4273                 }
4274                 case SMS_CDDVD_CAPS_PAGE:{
4275                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4276                             ("subpage %#x for page %#x is incorrect!",
4277                             page_index->subpage, page_code));
4278                         memcpy(&lun->mode_pages.cddvd_page[CTL_PAGE_DEFAULT],
4279                                &cddvd_page_default,
4280                                sizeof(cddvd_page_default));
4281                         memcpy(&lun->mode_pages.cddvd_page[
4282                                CTL_PAGE_CHANGEABLE], &cddvd_page_changeable,
4283                                sizeof(cddvd_page_changeable));
4284                         memcpy(&lun->mode_pages.cddvd_page[CTL_PAGE_SAVED],
4285                                &cddvd_page_default,
4286                                sizeof(cddvd_page_default));
4287                         memcpy(&lun->mode_pages.cddvd_page[CTL_PAGE_CURRENT],
4288                                &lun->mode_pages.cddvd_page[CTL_PAGE_SAVED],
4289                                sizeof(cddvd_page_default));
4290                         page_index->page_data =
4291                                 (uint8_t *)lun->mode_pages.cddvd_page;
4292                         break;
4293                 }
4294                 default:
4295                         panic("invalid page code value %#x", page_code);
4296                 }
4297         }
4298
4299         return (CTL_RETVAL_COMPLETE);
4300 }
4301
4302 static int
4303 ctl_init_log_page_index(struct ctl_lun *lun)
4304 {
4305         struct ctl_page_index *page_index;
4306         int i, j, k, prev;
4307
4308         memcpy(&lun->log_pages.index, log_page_index_template,
4309                sizeof(log_page_index_template));
4310
4311         prev = -1;
4312         for (i = 0, j = 0, k = 0; i < CTL_NUM_LOG_PAGES; i++) {
4313
4314                 page_index = &lun->log_pages.index[i];
4315                 if (lun->be_lun->lun_type == T_DIRECT &&
4316                     (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
4317                         continue;
4318                 if (lun->be_lun->lun_type == T_PROCESSOR &&
4319                     (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
4320                         continue;
4321                 if (lun->be_lun->lun_type == T_CDROM &&
4322                     (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
4323                         continue;
4324
4325                 if (page_index->page_code == SLS_LOGICAL_BLOCK_PROVISIONING &&
4326                     lun->backend->lun_attr == NULL)
4327                         continue;
4328
4329                 if (page_index->page_code != prev) {
4330                         lun->log_pages.pages_page[j] = page_index->page_code;
4331                         prev = page_index->page_code;
4332                         j++;
4333                 }
4334                 lun->log_pages.subpages_page[k*2] = page_index->page_code;
4335                 lun->log_pages.subpages_page[k*2+1] = page_index->subpage;
4336                 k++;
4337         }
4338         lun->log_pages.index[0].page_data = &lun->log_pages.pages_page[0];
4339         lun->log_pages.index[0].page_len = j;
4340         lun->log_pages.index[1].page_data = &lun->log_pages.subpages_page[0];
4341         lun->log_pages.index[1].page_len = k * 2;
4342         lun->log_pages.index[2].page_data = &lun->log_pages.lbp_page[0];
4343         lun->log_pages.index[2].page_len = 12*CTL_NUM_LBP_PARAMS;
4344         lun->log_pages.index[3].page_data = (uint8_t *)&lun->log_pages.stat_page;
4345         lun->log_pages.index[3].page_len = sizeof(lun->log_pages.stat_page);
4346         lun->log_pages.index[4].page_data = (uint8_t *)&lun->log_pages.ie_page;
4347         lun->log_pages.index[4].page_len = sizeof(lun->log_pages.ie_page);
4348
4349         return (CTL_RETVAL_COMPLETE);
4350 }
4351
4352 static int
4353 hex2bin(const char *str, uint8_t *buf, int buf_size)
4354 {
4355         int i;
4356         u_char c;
4357
4358         memset(buf, 0, buf_size);
4359         while (isspace(str[0]))
4360                 str++;
4361         if (str[0] == '0' && (str[1] == 'x' || str[1] == 'X'))
4362                 str += 2;
4363         buf_size *= 2;
4364         for (i = 0; str[i] != 0 && i < buf_size; i++) {
4365                 while (str[i] == '-')   /* Skip dashes in UUIDs. */
4366                         str++;
4367                 c = str[i];
4368                 if (isdigit(c))
4369                         c -= '0';
4370                 else if (isalpha(c))
4371                         c -= isupper(c) ? 'A' - 10 : 'a' - 10;
4372                 else
4373                         break;
4374                 if (c >= 16)
4375                         break;
4376                 if ((i & 1) == 0)
4377                         buf[i / 2] |= (c << 4);
4378                 else
4379                         buf[i / 2] |= c;
4380         }
4381         return ((i + 1) / 2);
4382 }
4383
4384 /*
4385  * LUN allocation.
4386  *
4387  * Requirements:
4388  * - caller allocates and zeros LUN storage, or passes in a NULL LUN if he
4389  *   wants us to allocate the LUN and he can block.
4390  * - ctl_softc is always set
4391  * - be_lun is set if the LUN has a backend (needed for disk LUNs)
4392  *
4393  * Returns 0 for success, non-zero (errno) for failure.
4394  */
4395 static int
4396 ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *ctl_lun,
4397               struct ctl_be_lun *const be_lun)
4398 {
4399         struct ctl_lun *nlun, *lun;
4400         struct scsi_vpd_id_descriptor *desc;
4401         struct scsi_vpd_id_t10 *t10id;
4402         const char *eui, *naa, *scsiname, *uuid, *vendor, *value;
4403         int lun_number, i, lun_malloced;
4404         int devidlen, idlen1, idlen2 = 0, len;
4405
4406         if (be_lun == NULL)
4407                 return (EINVAL);
4408
4409         /*
4410          * We currently only support Direct Access or Processor LUN types.
4411          */
4412         switch (be_lun->lun_type) {
4413         case T_DIRECT:
4414         case T_PROCESSOR:
4415         case T_CDROM:
4416                 break;
4417         case T_SEQUENTIAL:
4418         case T_CHANGER:
4419         default:
4420                 be_lun->lun_config_status(be_lun->be_lun,
4421                                           CTL_LUN_CONFIG_FAILURE);
4422                 break;
4423         }
4424         if (ctl_lun == NULL) {
4425                 lun = malloc(sizeof(*lun), M_CTL, M_WAITOK);
4426                 lun_malloced = 1;
4427         } else {
4428                 lun_malloced = 0;
4429                 lun = ctl_lun;
4430         }
4431
4432         memset(lun, 0, sizeof(*lun));
4433         if (lun_malloced)
4434                 lun->flags = CTL_LUN_MALLOCED;
4435
4436         /* Generate LUN ID. */
4437         devidlen = max(CTL_DEVID_MIN_LEN,
4438             strnlen(be_lun->device_id, CTL_DEVID_LEN));
4439         idlen1 = sizeof(*t10id) + devidlen;
4440         len = sizeof(struct scsi_vpd_id_descriptor) + idlen1;
4441         scsiname = ctl_get_opt(&be_lun->options, "scsiname");
4442         if (scsiname != NULL) {
4443                 idlen2 = roundup2(strlen(scsiname) + 1, 4);
4444                 len += sizeof(struct scsi_vpd_id_descriptor) + idlen2;
4445         }
4446         eui = ctl_get_opt(&be_lun->options, "eui");
4447         if (eui != NULL) {
4448                 len += sizeof(struct scsi_vpd_id_descriptor) + 16;
4449         }
4450         naa = ctl_get_opt(&be_lun->options, "naa");
4451         if (naa != NULL) {
4452                 len += sizeof(struct scsi_vpd_id_descriptor) + 16;
4453         }
4454         uuid = ctl_get_opt(&be_lun->options, "uuid");
4455         if (uuid != NULL) {
4456                 len += sizeof(struct scsi_vpd_id_descriptor) + 18;
4457         }
4458         lun->lun_devid = malloc(sizeof(struct ctl_devid) + len,
4459             M_CTL, M_WAITOK | M_ZERO);
4460         desc = (struct scsi_vpd_id_descriptor *)lun->lun_devid->data;
4461         desc->proto_codeset = SVPD_ID_CODESET_ASCII;
4462         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_T10;
4463         desc->length = idlen1;
4464         t10id = (struct scsi_vpd_id_t10 *)&desc->identifier[0];
4465         memset(t10id->vendor, ' ', sizeof(t10id->vendor));
4466         if ((vendor = ctl_get_opt(&be_lun->options, "vendor")) == NULL) {
4467                 strncpy((char *)t10id->vendor, CTL_VENDOR, sizeof(t10id->vendor));
4468         } else {
4469                 strncpy(t10id->vendor, vendor,
4470                     min(sizeof(t10id->vendor), strlen(vendor)));
4471         }
4472         strncpy((char *)t10id->vendor_spec_id,
4473             (char *)be_lun->device_id, devidlen);
4474         if (scsiname != NULL) {
4475                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4476                     desc->length);
4477                 desc->proto_codeset = SVPD_ID_CODESET_UTF8;
4478                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4479                     SVPD_ID_TYPE_SCSI_NAME;
4480                 desc->length = idlen2;
4481                 strlcpy(desc->identifier, scsiname, idlen2);
4482         }
4483         if (eui != NULL) {
4484                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4485                     desc->length);
4486                 desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4487                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4488                     SVPD_ID_TYPE_EUI64;
4489                 desc->length = hex2bin(eui, desc->identifier, 16);
4490                 desc->length = desc->length > 12 ? 16 :
4491                     (desc->length > 8 ? 12 : 8);
4492                 len -= 16 - desc->length;
4493         }
4494         if (naa != NULL) {
4495                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4496                     desc->length);
4497                 desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4498                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4499                     SVPD_ID_TYPE_NAA;
4500                 desc->length = hex2bin(naa, desc->identifier, 16);
4501                 desc->length = desc->length > 8 ? 16 : 8;
4502                 len -= 16 - desc->length;
4503         }
4504         if (uuid != NULL) {
4505                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4506                     desc->length);
4507                 desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4508                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4509                     SVPD_ID_TYPE_UUID;
4510                 desc->identifier[0] = 0x10;
4511                 hex2bin(uuid, &desc->identifier[2], 16);
4512                 desc->length = 18;
4513         }
4514         lun->lun_devid->len = len;
4515
4516         mtx_lock(&ctl_softc->ctl_lock);
4517         /*
4518          * See if the caller requested a particular LUN number.  If so, see
4519          * if it is available.  Otherwise, allocate the first available LUN.
4520          */
4521         if (be_lun->flags & CTL_LUN_FLAG_ID_REQ) {
4522                 if ((be_lun->req_lun_id > (CTL_MAX_LUNS - 1))
4523                  || (ctl_is_set(ctl_softc->ctl_lun_mask, be_lun->req_lun_id))) {
4524                         mtx_unlock(&ctl_softc->ctl_lock);
4525                         if (be_lun->req_lun_id > (CTL_MAX_LUNS - 1)) {
4526                                 printf("ctl: requested LUN ID %d is higher "
4527                                        "than CTL_MAX_LUNS - 1 (%d)\n",
4528                                        be_lun->req_lun_id, CTL_MAX_LUNS - 1);
4529                         } else {
4530                                 /*
4531                                  * XXX KDM return an error, or just assign
4532                                  * another LUN ID in this case??
4533                                  */
4534                                 printf("ctl: requested LUN ID %d is already "
4535                                        "in use\n", be_lun->req_lun_id);
4536                         }
4537                         if (lun->flags & CTL_LUN_MALLOCED)
4538                                 free(lun, M_CTL);
4539                         be_lun->lun_config_status(be_lun->be_lun,
4540                                                   CTL_LUN_CONFIG_FAILURE);
4541                         return (ENOSPC);
4542                 }
4543                 lun_number = be_lun->req_lun_id;
4544         } else {
4545                 lun_number = ctl_ffz(ctl_softc->ctl_lun_mask, 0, CTL_MAX_LUNS);
4546                 if (lun_number == -1) {
4547                         mtx_unlock(&ctl_softc->ctl_lock);
4548                         printf("ctl: can't allocate LUN, out of LUNs\n");
4549                         if (lun->flags & CTL_LUN_MALLOCED)
4550                                 free(lun, M_CTL);
4551                         be_lun->lun_config_status(be_lun->be_lun,
4552                                                   CTL_LUN_CONFIG_FAILURE);
4553                         return (ENOSPC);
4554                 }
4555         }
4556         ctl_set_mask(ctl_softc->ctl_lun_mask, lun_number);
4557
4558         mtx_init(&lun->lun_lock, "CTL LUN", NULL, MTX_DEF);
4559         lun->lun = lun_number;
4560         lun->be_lun = be_lun;
4561         /*
4562          * The processor LUN is always enabled.  Disk LUNs come on line
4563          * disabled, and must be enabled by the backend.
4564          */
4565         lun->flags |= CTL_LUN_DISABLED;
4566         lun->backend = be_lun->be;
4567         be_lun->ctl_lun = lun;
4568         be_lun->lun_id = lun_number;
4569         atomic_add_int(&be_lun->be->num_luns, 1);
4570         if (be_lun->flags & CTL_LUN_FLAG_EJECTED)
4571                 lun->flags |= CTL_LUN_EJECTED;
4572         if (be_lun->flags & CTL_LUN_FLAG_NO_MEDIA)
4573                 lun->flags |= CTL_LUN_NO_MEDIA;
4574         if (be_lun->flags & CTL_LUN_FLAG_STOPPED)
4575                 lun->flags |= CTL_LUN_STOPPED;
4576
4577         if (be_lun->flags & CTL_LUN_FLAG_PRIMARY)
4578                 lun->flags |= CTL_LUN_PRIMARY_SC;
4579
4580         value = ctl_get_opt(&be_lun->options, "removable");
4581         if (value != NULL) {
4582                 if (strcmp(value, "on") == 0)
4583                         lun->flags |= CTL_LUN_REMOVABLE;
4584         } else if (be_lun->lun_type == T_CDROM)
4585                 lun->flags |= CTL_LUN_REMOVABLE;
4586
4587         lun->ctl_softc = ctl_softc;
4588 #ifdef CTL_TIME_IO
4589         lun->last_busy = getsbinuptime();
4590 #endif
4591         TAILQ_INIT(&lun->ooa_queue);
4592         TAILQ_INIT(&lun->blocked_queue);
4593         STAILQ_INIT(&lun->error_list);
4594         lun->ie_reported = 1;
4595         callout_init_mtx(&lun->ie_callout, &lun->lun_lock, 0);
4596         ctl_tpc_lun_init(lun);
4597
4598         /*
4599          * Initialize the mode and log page index.
4600          */
4601         ctl_init_page_index(lun);
4602         ctl_init_log_page_index(lun);
4603
4604         /*
4605          * Now, before we insert this lun on the lun list, set the lun
4606          * inventory changed UA for all other luns.
4607          */
4608         STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) {
4609                 mtx_lock(&nlun->lun_lock);
4610                 ctl_est_ua_all(nlun, -1, CTL_UA_LUN_CHANGE);
4611                 mtx_unlock(&nlun->lun_lock);
4612         }
4613
4614         STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links);
4615
4616         ctl_softc->ctl_luns[lun_number] = lun;
4617
4618         ctl_softc->num_luns++;
4619
4620         /* Setup statistics gathering */
4621         lun->stats.device_type = be_lun->lun_type;
4622         lun->stats.lun_number = lun_number;
4623         lun->stats.blocksize = be_lun->blocksize;
4624         if (be_lun->blocksize == 0)
4625                 lun->stats.flags = CTL_LUN_STATS_NO_BLOCKSIZE;
4626         for (i = 0;i < CTL_MAX_PORTS;i++)
4627                 lun->stats.ports[i].targ_port = i;
4628
4629         mtx_unlock(&ctl_softc->ctl_lock);
4630
4631         lun->be_lun->lun_config_status(lun->be_lun->be_lun, CTL_LUN_CONFIG_OK);
4632         return (0);
4633 }
4634
4635 /*
4636  * Delete a LUN.
4637  * Assumptions:
4638  * - LUN has already been marked invalid and any pending I/O has been taken
4639  *   care of.
4640  */
4641 static int
4642 ctl_free_lun(struct ctl_lun *lun)
4643 {
4644         struct ctl_softc *softc;
4645         struct ctl_lun *nlun;
4646         int i;
4647
4648         softc = lun->ctl_softc;
4649
4650         mtx_assert(&softc->ctl_lock, MA_OWNED);
4651
4652         STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links);
4653
4654         ctl_clear_mask(softc->ctl_lun_mask, lun->lun);
4655
4656         softc->ctl_luns[lun->lun] = NULL;
4657
4658         if (!TAILQ_EMPTY(&lun->ooa_queue))
4659                 panic("Freeing a LUN %p with outstanding I/O!!\n", lun);
4660
4661         softc->num_luns--;
4662
4663         /*
4664          * Tell the backend to free resources, if this LUN has a backend.
4665          */
4666         atomic_subtract_int(&lun->be_lun->be->num_luns, 1);
4667         lun->be_lun->lun_shutdown(lun->be_lun->be_lun);
4668
4669         lun->ie_reportcnt = UINT32_MAX;
4670         callout_drain(&lun->ie_callout);
4671
4672         ctl_tpc_lun_shutdown(lun);
4673         mtx_destroy(&lun->lun_lock);
4674         free(lun->lun_devid, M_CTL);
4675         for (i = 0; i < CTL_MAX_PORTS; i++)
4676                 free(lun->pending_ua[i], M_CTL);
4677         for (i = 0; i < CTL_MAX_PORTS; i++)
4678                 free(lun->pr_keys[i], M_CTL);
4679         free(lun->write_buffer, M_CTL);
4680         if (lun->flags & CTL_LUN_MALLOCED)
4681                 free(lun, M_CTL);
4682
4683         STAILQ_FOREACH(nlun, &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
4689         return (0);
4690 }
4691
4692 static void
4693 ctl_create_lun(struct ctl_be_lun *be_lun)
4694 {
4695
4696         /*
4697          * ctl_alloc_lun() should handle all potential failure cases.
4698          */
4699         ctl_alloc_lun(control_softc, NULL, be_lun);
4700 }
4701
4702 int
4703 ctl_add_lun(struct ctl_be_lun *be_lun)
4704 {
4705         struct ctl_softc *softc = control_softc;
4706
4707         mtx_lock(&softc->ctl_lock);
4708         STAILQ_INSERT_TAIL(&softc->pending_lun_queue, be_lun, links);
4709         mtx_unlock(&softc->ctl_lock);
4710         wakeup(&softc->pending_lun_queue);
4711
4712         return (0);
4713 }
4714
4715 int
4716 ctl_enable_lun(struct ctl_be_lun *be_lun)
4717 {
4718         struct ctl_softc *softc;
4719         struct ctl_port *port, *nport;
4720         struct ctl_lun *lun;
4721         int retval;
4722
4723         lun = (struct ctl_lun *)be_lun->ctl_lun;
4724         softc = lun->ctl_softc;
4725
4726         mtx_lock(&softc->ctl_lock);
4727         mtx_lock(&lun->lun_lock);
4728         if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4729                 /*
4730                  * eh?  Why did we get called if the LUN is already
4731                  * enabled?
4732                  */
4733                 mtx_unlock(&lun->lun_lock);
4734                 mtx_unlock(&softc->ctl_lock);
4735                 return (0);
4736         }
4737         lun->flags &= ~CTL_LUN_DISABLED;
4738         mtx_unlock(&lun->lun_lock);
4739
4740         STAILQ_FOREACH_SAFE(port, &softc->port_list, links, nport) {
4741                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0 ||
4742                     port->lun_map != NULL || port->lun_enable == NULL)
4743                         continue;
4744
4745                 /*
4746                  * Drop the lock while we call the FETD's enable routine.
4747                  * This can lead to a callback into CTL (at least in the
4748                  * case of the internal initiator frontend.
4749                  */
4750                 mtx_unlock(&softc->ctl_lock);
4751                 retval = port->lun_enable(port->targ_lun_arg, lun->lun);
4752                 mtx_lock(&softc->ctl_lock);
4753                 if (retval != 0) {
4754                         printf("%s: FETD %s port %d returned error "
4755                                "%d for lun_enable on lun %jd\n",
4756                                __func__, port->port_name, port->targ_port,
4757                                retval, (intmax_t)lun->lun);
4758                 }
4759         }
4760
4761         mtx_unlock(&softc->ctl_lock);
4762         ctl_isc_announce_lun(lun);
4763
4764         return (0);
4765 }
4766
4767 int
4768 ctl_disable_lun(struct ctl_be_lun *be_lun)
4769 {
4770         struct ctl_softc *softc;
4771         struct ctl_port *port;
4772         struct ctl_lun *lun;
4773         int retval;
4774
4775         lun = (struct ctl_lun *)be_lun->ctl_lun;
4776         softc = lun->ctl_softc;
4777
4778         mtx_lock(&softc->ctl_lock);
4779         mtx_lock(&lun->lun_lock);
4780         if (lun->flags & CTL_LUN_DISABLED) {
4781                 mtx_unlock(&lun->lun_lock);
4782                 mtx_unlock(&softc->ctl_lock);
4783                 return (0);
4784         }
4785         lun->flags |= CTL_LUN_DISABLED;
4786         mtx_unlock(&lun->lun_lock);
4787
4788         STAILQ_FOREACH(port, &softc->port_list, links) {
4789                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0 ||
4790                     port->lun_map != NULL || port->lun_disable == NULL)
4791                         continue;
4792
4793                 /*
4794                  * Drop the lock before we call the frontend's disable
4795                  * routine, to avoid lock order reversals.
4796                  *
4797                  * XXX KDM what happens if the frontend list changes while
4798                  * we're traversing it?  It's unlikely, but should be handled.
4799                  */
4800                 mtx_unlock(&softc->ctl_lock);
4801                 retval = port->lun_disable(port->targ_lun_arg, lun->lun);
4802                 mtx_lock(&softc->ctl_lock);
4803                 if (retval != 0) {
4804                         printf("%s: FETD %s port %d returned error "
4805                                "%d for lun_disable on lun %jd\n",
4806                                __func__, port->port_name, port->targ_port,
4807                                retval, (intmax_t)lun->lun);
4808                 }
4809         }
4810
4811         mtx_unlock(&softc->ctl_lock);
4812         ctl_isc_announce_lun(lun);
4813
4814         return (0);
4815 }
4816
4817 int
4818 ctl_start_lun(struct ctl_be_lun *be_lun)
4819 {
4820         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4821
4822         mtx_lock(&lun->lun_lock);
4823         lun->flags &= ~CTL_LUN_STOPPED;
4824         mtx_unlock(&lun->lun_lock);
4825         return (0);
4826 }
4827
4828 int
4829 ctl_stop_lun(struct ctl_be_lun *be_lun)
4830 {
4831         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4832
4833         mtx_lock(&lun->lun_lock);
4834         lun->flags |= CTL_LUN_STOPPED;
4835         mtx_unlock(&lun->lun_lock);
4836         return (0);
4837 }
4838
4839 int
4840 ctl_lun_no_media(struct ctl_be_lun *be_lun)
4841 {
4842         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4843
4844         mtx_lock(&lun->lun_lock);
4845         lun->flags |= CTL_LUN_NO_MEDIA;
4846         mtx_unlock(&lun->lun_lock);
4847         return (0);
4848 }
4849
4850 int
4851 ctl_lun_has_media(struct ctl_be_lun *be_lun)
4852 {
4853         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4854         union ctl_ha_msg msg;
4855
4856         mtx_lock(&lun->lun_lock);
4857         lun->flags &= ~(CTL_LUN_NO_MEDIA | CTL_LUN_EJECTED);
4858         if (lun->flags & CTL_LUN_REMOVABLE)
4859                 ctl_est_ua_all(lun, -1, CTL_UA_MEDIUM_CHANGE);
4860         mtx_unlock(&lun->lun_lock);
4861         if ((lun->flags & CTL_LUN_REMOVABLE) &&
4862             lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
4863                 bzero(&msg.ua, sizeof(msg.ua));
4864                 msg.hdr.msg_type = CTL_MSG_UA;
4865                 msg.hdr.nexus.initid = -1;
4866                 msg.hdr.nexus.targ_port = -1;
4867                 msg.hdr.nexus.targ_lun = lun->lun;
4868                 msg.hdr.nexus.targ_mapped_lun = lun->lun;
4869                 msg.ua.ua_all = 1;
4870                 msg.ua.ua_set = 1;
4871                 msg.ua.ua_type = CTL_UA_MEDIUM_CHANGE;
4872                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg.ua),
4873                     M_WAITOK);
4874         }
4875         return (0);
4876 }
4877
4878 int
4879 ctl_lun_ejected(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_EJECTED;
4885         mtx_unlock(&lun->lun_lock);
4886         return (0);
4887 }
4888
4889 int
4890 ctl_lun_primary(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_PRIMARY_SC;
4896         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
4897         mtx_unlock(&lun->lun_lock);
4898         ctl_isc_announce_lun(lun);
4899         return (0);
4900 }
4901
4902 int
4903 ctl_lun_secondary(struct ctl_be_lun *be_lun)
4904 {
4905         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4906
4907         mtx_lock(&lun->lun_lock);
4908         lun->flags &= ~CTL_LUN_PRIMARY_SC;
4909         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
4910         mtx_unlock(&lun->lun_lock);
4911         ctl_isc_announce_lun(lun);
4912         return (0);
4913 }
4914
4915 int
4916 ctl_invalidate_lun(struct ctl_be_lun *be_lun)
4917 {
4918         struct ctl_softc *softc;
4919         struct ctl_lun *lun;
4920
4921         lun = (struct ctl_lun *)be_lun->ctl_lun;
4922         softc = lun->ctl_softc;
4923
4924         mtx_lock(&lun->lun_lock);
4925
4926         /*
4927          * The LUN needs to be disabled before it can be marked invalid.
4928          */
4929         if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4930                 mtx_unlock(&lun->lun_lock);
4931                 return (-1);
4932         }
4933         /*
4934          * Mark the LUN invalid.
4935          */
4936         lun->flags |= CTL_LUN_INVALID;
4937
4938         /*
4939          * If there is nothing in the OOA queue, go ahead and free the LUN.
4940          * If we have something in the OOA queue, we'll free it when the
4941          * last I/O completes.
4942          */
4943         if (TAILQ_EMPTY(&lun->ooa_queue)) {
4944                 mtx_unlock(&lun->lun_lock);
4945                 mtx_lock(&softc->ctl_lock);
4946                 ctl_free_lun(lun);
4947                 mtx_unlock(&softc->ctl_lock);
4948         } else
4949                 mtx_unlock(&lun->lun_lock);
4950
4951         return (0);
4952 }
4953
4954 void
4955 ctl_lun_capacity_changed(struct ctl_be_lun *be_lun)
4956 {
4957         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4958         union ctl_ha_msg msg;
4959
4960         mtx_lock(&lun->lun_lock);
4961         ctl_est_ua_all(lun, -1, CTL_UA_CAPACITY_CHANGE);
4962         mtx_unlock(&lun->lun_lock);
4963         if (lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
4964                 /* Send msg to other side. */
4965                 bzero(&msg.ua, sizeof(msg.ua));
4966                 msg.hdr.msg_type = CTL_MSG_UA;
4967                 msg.hdr.nexus.initid = -1;
4968                 msg.hdr.nexus.targ_port = -1;
4969                 msg.hdr.nexus.targ_lun = lun->lun;
4970                 msg.hdr.nexus.targ_mapped_lun = lun->lun;
4971                 msg.ua.ua_all = 1;
4972                 msg.ua.ua_set = 1;
4973                 msg.ua.ua_type = CTL_UA_CAPACITY_CHANGE;
4974                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg.ua),
4975                     M_WAITOK);
4976         }
4977 }
4978
4979 /*
4980  * Backend "memory move is complete" callback for requests that never
4981  * make it down to say RAIDCore's configuration code.
4982  */
4983 int
4984 ctl_config_move_done(union ctl_io *io)
4985 {
4986         int retval;
4987
4988         CTL_DEBUG_PRINT(("ctl_config_move_done\n"));
4989         KASSERT(io->io_hdr.io_type == CTL_IO_SCSI,
4990             ("Config I/O type isn't CTL_IO_SCSI (%d)!", io->io_hdr.io_type));
4991
4992         if ((io->io_hdr.port_status != 0) &&
4993             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
4994              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
4995                 /*
4996                  * For hardware error sense keys, the sense key
4997                  * specific value is defined to be a retry count,
4998                  * but we use it to pass back an internal FETD
4999                  * error code.  XXX KDM  Hopefully the FETD is only
5000                  * using 16 bits for an error code, since that's
5001                  * all the space we have in the sks field.
5002                  */
5003                 ctl_set_internal_failure(&io->scsiio,
5004                                          /*sks_valid*/ 1,
5005                                          /*retry_count*/
5006                                          io->io_hdr.port_status);
5007         }
5008
5009         if (ctl_debug & CTL_DEBUG_CDB_DATA)
5010                 ctl_data_print(io);
5011         if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN) ||
5012             ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
5013              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS) ||
5014             ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) {
5015                 /*
5016                  * XXX KDM just assuming a single pointer here, and not a
5017                  * S/G list.  If we start using S/G lists for config data,
5018                  * we'll need to know how to clean them up here as well.
5019                  */
5020                 if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5021                         free(io->scsiio.kern_data_ptr, M_CTL);
5022                 ctl_done(io);
5023                 retval = CTL_RETVAL_COMPLETE;
5024         } else {
5025                 /*
5026                  * XXX KDM now we need to continue data movement.  Some
5027                  * options:
5028                  * - call ctl_scsiio() again?  We don't do this for data
5029                  *   writes, because for those at least we know ahead of
5030                  *   time where the write will go and how long it is.  For
5031                  *   config writes, though, that information is largely
5032                  *   contained within the write itself, thus we need to
5033                  *   parse out the data again.
5034                  *
5035                  * - Call some other function once the data is in?
5036                  */
5037
5038                 /*
5039                  * XXX KDM call ctl_scsiio() again for now, and check flag
5040                  * bits to see whether we're allocated or not.
5041                  */
5042                 retval = ctl_scsiio(&io->scsiio);
5043         }
5044         return (retval);
5045 }
5046
5047 /*
5048  * This gets called by a backend driver when it is done with a
5049  * data_submit method.
5050  */
5051 void
5052 ctl_data_submit_done(union ctl_io *io)
5053 {
5054         /*
5055          * If the IO_CONT flag is set, we need to call the supplied
5056          * function to continue processing the I/O, instead of completing
5057          * the I/O just yet.
5058          *
5059          * If there is an error, though, we don't want to keep processing.
5060          * Instead, just send status back to the initiator.
5061          */
5062         if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5063             (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5064             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5065              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5066                 io->scsiio.io_cont(io);
5067                 return;
5068         }
5069         ctl_done(io);
5070 }
5071
5072 /*
5073  * This gets called by a backend driver when it is done with a
5074  * configuration write.
5075  */
5076 void
5077 ctl_config_write_done(union ctl_io *io)
5078 {
5079         uint8_t *buf;
5080
5081         /*
5082          * If the IO_CONT flag is set, we need to call the supplied
5083          * function to continue processing the I/O, instead of completing
5084          * the I/O just yet.
5085          *
5086          * If there is an error, though, we don't want to keep processing.
5087          * Instead, just send status back to the initiator.
5088          */
5089         if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5090             (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5091             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5092              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5093                 io->scsiio.io_cont(io);
5094                 return;
5095         }
5096         /*
5097          * Since a configuration write can be done for commands that actually
5098          * have data allocated, like write buffer, and commands that have
5099          * no data, like start/stop unit, we need to check here.
5100          */
5101         if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5102                 buf = io->scsiio.kern_data_ptr;
5103         else
5104                 buf = NULL;
5105         ctl_done(io);
5106         if (buf)
5107                 free(buf, M_CTL);
5108 }
5109
5110 void
5111 ctl_config_read_done(union ctl_io *io)
5112 {
5113         uint8_t *buf;
5114
5115         /*
5116          * If there is some error -- we are done, skip data transfer.
5117          */
5118         if ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0 ||
5119             ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
5120              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) {
5121                 if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5122                         buf = io->scsiio.kern_data_ptr;
5123                 else
5124                         buf = NULL;
5125                 ctl_done(io);
5126                 if (buf)
5127                         free(buf, M_CTL);
5128                 return;
5129         }
5130
5131         /*
5132          * If the IO_CONT flag is set, we need to call the supplied
5133          * function to continue processing the I/O, instead of completing
5134          * the I/O just yet.
5135          */
5136         if (io->io_hdr.flags & CTL_FLAG_IO_CONT) {
5137                 io->scsiio.io_cont(io);
5138                 return;
5139         }
5140
5141         ctl_datamove(io);
5142 }
5143
5144 /*
5145  * SCSI release command.
5146  */
5147 int
5148 ctl_scsi_release(struct ctl_scsiio *ctsio)
5149 {
5150         struct ctl_lun *lun;
5151         uint32_t residx;
5152
5153         CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
5154
5155         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5156         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5157
5158         /*
5159          * XXX KDM right now, we only support LUN reservation.  We don't
5160          * support 3rd party reservations, or extent reservations, which
5161          * might actually need the parameter list.  If we've gotten this
5162          * far, we've got a LUN reservation.  Anything else got kicked out
5163          * above.  So, according to SPC, ignore the length.
5164          */
5165
5166         mtx_lock(&lun->lun_lock);
5167
5168         /*
5169          * According to SPC, it is not an error for an intiator to attempt
5170          * to release a reservation on a LUN that isn't reserved, or that
5171          * is reserved by another initiator.  The reservation can only be
5172          * released, though, by the initiator who made it or by one of
5173          * several reset type events.
5174          */
5175         if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == residx))
5176                         lun->flags &= ~CTL_LUN_RESERVED;
5177
5178         mtx_unlock(&lun->lun_lock);
5179
5180         ctl_set_success(ctsio);
5181         ctl_done((union ctl_io *)ctsio);
5182         return (CTL_RETVAL_COMPLETE);
5183 }
5184
5185 int
5186 ctl_scsi_reserve(struct ctl_scsiio *ctsio)
5187 {
5188         struct ctl_lun *lun;
5189         uint32_t residx;
5190
5191         CTL_DEBUG_PRINT(("ctl_reserve\n"));
5192
5193         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5194         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5195
5196         /*
5197          * XXX KDM right now, we only support LUN reservation.  We don't
5198          * support 3rd party reservations, or extent reservations, which
5199          * might actually need the parameter list.  If we've gotten this
5200          * far, we've got a LUN reservation.  Anything else got kicked out
5201          * above.  So, according to SPC, ignore the length.
5202          */
5203
5204         mtx_lock(&lun->lun_lock);
5205         if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx != residx)) {
5206                 ctl_set_reservation_conflict(ctsio);
5207                 goto bailout;
5208         }
5209
5210         /* SPC-3 exceptions to SPC-2 RESERVE and RELEASE behavior. */
5211         if (lun->flags & CTL_LUN_PR_RESERVED) {
5212                 ctl_set_success(ctsio);
5213                 goto bailout;
5214         }
5215
5216         lun->flags |= CTL_LUN_RESERVED;
5217         lun->res_idx = residx;
5218         ctl_set_success(ctsio);
5219
5220 bailout:
5221         mtx_unlock(&lun->lun_lock);
5222         ctl_done((union ctl_io *)ctsio);
5223         return (CTL_RETVAL_COMPLETE);
5224 }
5225
5226 int
5227 ctl_start_stop(struct ctl_scsiio *ctsio)
5228 {
5229         struct scsi_start_stop_unit *cdb;
5230         struct ctl_lun *lun;
5231         int retval;
5232
5233         CTL_DEBUG_PRINT(("ctl_start_stop\n"));
5234
5235         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5236         cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
5237
5238         if ((cdb->how & SSS_PC_MASK) == 0) {
5239                 if ((lun->flags & CTL_LUN_PR_RESERVED) &&
5240                     (cdb->how & SSS_START) == 0) {
5241                         uint32_t residx;
5242
5243                         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5244                         if (ctl_get_prkey(lun, residx) == 0 ||
5245                             (lun->pr_res_idx != residx && lun->pr_res_type < 4)) {
5246
5247                                 ctl_set_reservation_conflict(ctsio);
5248                                 ctl_done((union ctl_io *)ctsio);
5249                                 return (CTL_RETVAL_COMPLETE);
5250                         }
5251                 }
5252
5253                 if ((cdb->how & SSS_LOEJ) &&
5254                     (lun->flags & CTL_LUN_REMOVABLE) == 0) {
5255                         ctl_set_invalid_field(ctsio,
5256                                               /*sks_valid*/ 1,
5257                                               /*command*/ 1,
5258                                               /*field*/ 4,
5259                                               /*bit_valid*/ 1,
5260                                               /*bit*/ 1);
5261                         ctl_done((union ctl_io *)ctsio);
5262                         return (CTL_RETVAL_COMPLETE);
5263                 }
5264
5265                 if ((cdb->how & SSS_START) == 0 && (cdb->how & SSS_LOEJ) &&
5266                     lun->prevent_count > 0) {
5267                         /* "Medium removal prevented" */
5268                         ctl_set_sense(ctsio, /*current_error*/ 1,
5269                             /*sense_key*/(lun->flags & CTL_LUN_NO_MEDIA) ?
5270                              SSD_KEY_NOT_READY : SSD_KEY_ILLEGAL_REQUEST,
5271                             /*asc*/ 0x53, /*ascq*/ 0x02, SSD_ELEM_NONE);
5272                         ctl_done((union ctl_io *)ctsio);
5273                         return (CTL_RETVAL_COMPLETE);
5274                 }
5275         }
5276
5277         retval = lun->backend->config_write((union ctl_io *)ctsio);
5278         return (retval);
5279 }
5280
5281 int
5282 ctl_prevent_allow(struct ctl_scsiio *ctsio)
5283 {
5284         struct ctl_lun *lun;
5285         struct scsi_prevent *cdb;
5286         int retval;
5287         uint32_t initidx;
5288
5289         CTL_DEBUG_PRINT(("ctl_prevent_allow\n"));
5290
5291         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5292         cdb = (struct scsi_prevent *)ctsio->cdb;
5293
5294         if ((lun->flags & CTL_LUN_REMOVABLE) == 0) {
5295                 ctl_set_invalid_opcode(ctsio);
5296                 ctl_done((union ctl_io *)ctsio);
5297                 return (CTL_RETVAL_COMPLETE);
5298         }
5299
5300         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5301         mtx_lock(&lun->lun_lock);
5302         if ((cdb->how & PR_PREVENT) &&
5303             ctl_is_set(lun->prevent, initidx) == 0) {
5304                 ctl_set_mask(lun->prevent, initidx);
5305                 lun->prevent_count++;
5306         } else if ((cdb->how & PR_PREVENT) == 0 &&
5307             ctl_is_set(lun->prevent, initidx)) {
5308                 ctl_clear_mask(lun->prevent, initidx);
5309                 lun->prevent_count--;
5310         }
5311         mtx_unlock(&lun->lun_lock);
5312         retval = lun->backend->config_write((union ctl_io *)ctsio);
5313         return (retval);
5314 }
5315
5316 /*
5317  * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
5318  * we don't really do anything with the LBA and length fields if the user
5319  * passes them in.  Instead we'll just flush out the cache for the entire
5320  * LUN.
5321  */
5322 int
5323 ctl_sync_cache(struct ctl_scsiio *ctsio)
5324 {
5325         struct ctl_lun *lun;
5326         struct ctl_softc *softc;
5327         struct ctl_lba_len_flags *lbalen;
5328         uint64_t starting_lba;
5329         uint32_t block_count;
5330         int retval;
5331         uint8_t byte2;
5332
5333         CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
5334
5335         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5336         softc = lun->ctl_softc;
5337         retval = 0;
5338
5339         switch (ctsio->cdb[0]) {
5340         case SYNCHRONIZE_CACHE: {
5341                 struct scsi_sync_cache *cdb;
5342                 cdb = (struct scsi_sync_cache *)ctsio->cdb;
5343
5344                 starting_lba = scsi_4btoul(cdb->begin_lba);
5345                 block_count = scsi_2btoul(cdb->lb_count);
5346                 byte2 = cdb->byte2;
5347                 break;
5348         }
5349         case SYNCHRONIZE_CACHE_16: {
5350                 struct scsi_sync_cache_16 *cdb;
5351                 cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5352
5353                 starting_lba = scsi_8btou64(cdb->begin_lba);
5354                 block_count = scsi_4btoul(cdb->lb_count);
5355                 byte2 = cdb->byte2;
5356                 break;
5357         }
5358         default:
5359                 ctl_set_invalid_opcode(ctsio);
5360                 ctl_done((union ctl_io *)ctsio);
5361                 goto bailout;
5362                 break; /* NOTREACHED */
5363         }
5364
5365         /*
5366          * We check the LBA and length, but don't do anything with them.
5367          * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5368          * get flushed.  This check will just help satisfy anyone who wants
5369          * to see an error for an out of range LBA.
5370          */
5371         if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5372                 ctl_set_lba_out_of_range(ctsio,
5373                     MAX(starting_lba, lun->be_lun->maxlba + 1));
5374                 ctl_done((union ctl_io *)ctsio);
5375                 goto bailout;
5376         }
5377
5378         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5379         lbalen->lba = starting_lba;
5380         lbalen->len = block_count;
5381         lbalen->flags = byte2;
5382         retval = lun->backend->config_write((union ctl_io *)ctsio);
5383
5384 bailout:
5385         return (retval);
5386 }
5387
5388 int
5389 ctl_format(struct ctl_scsiio *ctsio)
5390 {
5391         struct scsi_format *cdb;
5392         struct ctl_lun *lun;
5393         int length, defect_list_len;
5394
5395         CTL_DEBUG_PRINT(("ctl_format\n"));
5396
5397         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5398
5399         cdb = (struct scsi_format *)ctsio->cdb;
5400
5401         length = 0;
5402         if (cdb->byte2 & SF_FMTDATA) {
5403                 if (cdb->byte2 & SF_LONGLIST)
5404                         length = sizeof(struct scsi_format_header_long);
5405                 else
5406                         length = sizeof(struct scsi_format_header_short);
5407         }
5408
5409         if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5410          && (length > 0)) {
5411                 ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5412                 ctsio->kern_data_len = length;
5413                 ctsio->kern_total_len = length;
5414                 ctsio->kern_data_resid = 0;
5415                 ctsio->kern_rel_offset = 0;
5416                 ctsio->kern_sg_entries = 0;
5417                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5418                 ctsio->be_move_done = ctl_config_move_done;
5419                 ctl_datamove((union ctl_io *)ctsio);
5420
5421                 return (CTL_RETVAL_COMPLETE);
5422         }
5423
5424         defect_list_len = 0;
5425
5426         if (cdb->byte2 & SF_FMTDATA) {
5427                 if (cdb->byte2 & SF_LONGLIST) {
5428                         struct scsi_format_header_long *header;
5429
5430                         header = (struct scsi_format_header_long *)
5431                                 ctsio->kern_data_ptr;
5432
5433                         defect_list_len = scsi_4btoul(header->defect_list_len);
5434                         if (defect_list_len != 0) {
5435                                 ctl_set_invalid_field(ctsio,
5436                                                       /*sks_valid*/ 1,
5437                                                       /*command*/ 0,
5438                                                       /*field*/ 2,
5439                                                       /*bit_valid*/ 0,
5440                                                       /*bit*/ 0);
5441                                 goto bailout;
5442                         }
5443                 } else {
5444                         struct scsi_format_header_short *header;
5445
5446                         header = (struct scsi_format_header_short *)
5447                                 ctsio->kern_data_ptr;
5448
5449                         defect_list_len = scsi_2btoul(header->defect_list_len);
5450                         if (defect_list_len != 0) {
5451                                 ctl_set_invalid_field(ctsio,
5452                                                       /*sks_valid*/ 1,
5453                                                       /*command*/ 0,
5454                                                       /*field*/ 2,
5455                                                       /*bit_valid*/ 0,
5456                                                       /*bit*/ 0);
5457                                 goto bailout;
5458                         }
5459                 }
5460         }
5461
5462         ctl_set_success(ctsio);
5463 bailout:
5464
5465         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5466                 free(ctsio->kern_data_ptr, M_CTL);
5467                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5468         }
5469
5470         ctl_done((union ctl_io *)ctsio);
5471         return (CTL_RETVAL_COMPLETE);
5472 }
5473
5474 int
5475 ctl_read_buffer(struct ctl_scsiio *ctsio)
5476 {
5477         struct ctl_lun *lun;
5478         uint64_t buffer_offset;
5479         uint32_t len;
5480         uint8_t byte2;
5481         static uint8_t descr[4];
5482         static uint8_t echo_descr[4] = { 0 };
5483
5484         CTL_DEBUG_PRINT(("ctl_read_buffer\n"));
5485         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5486         switch (ctsio->cdb[0]) {
5487         case READ_BUFFER: {
5488                 struct scsi_read_buffer *cdb;
5489
5490                 cdb = (struct scsi_read_buffer *)ctsio->cdb;
5491                 buffer_offset = scsi_3btoul(cdb->offset);
5492                 len = scsi_3btoul(cdb->length);
5493                 byte2 = cdb->byte2;
5494                 break;
5495         }
5496         case READ_BUFFER_16: {
5497                 struct scsi_read_buffer_16 *cdb;
5498
5499                 cdb = (struct scsi_read_buffer_16 *)ctsio->cdb;
5500                 buffer_offset = scsi_8btou64(cdb->offset);
5501                 len = scsi_4btoul(cdb->length);
5502                 byte2 = cdb->byte2;
5503                 break;
5504         }
5505         default: /* This shouldn't happen. */
5506                 ctl_set_invalid_opcode(ctsio);
5507                 ctl_done((union ctl_io *)ctsio);
5508                 return (CTL_RETVAL_COMPLETE);
5509         }
5510
5511         if ((byte2 & RWB_MODE) != RWB_MODE_DATA &&
5512             (byte2 & RWB_MODE) != RWB_MODE_ECHO_DESCR &&
5513             (byte2 & RWB_MODE) != RWB_MODE_DESCR) {
5514                 ctl_set_invalid_field(ctsio,
5515                                       /*sks_valid*/ 1,
5516                                       /*command*/ 1,
5517                                       /*field*/ 1,
5518                                       /*bit_valid*/ 1,
5519                                       /*bit*/ 4);
5520                 ctl_done((union ctl_io *)ctsio);
5521                 return (CTL_RETVAL_COMPLETE);
5522         }
5523
5524         if (buffer_offset > CTL_WRITE_BUFFER_SIZE ||
5525             buffer_offset + len > CTL_WRITE_BUFFER_SIZE) {
5526                 ctl_set_invalid_field(ctsio,
5527                                       /*sks_valid*/ 1,
5528                                       /*command*/ 1,
5529                                       /*field*/ 6,
5530                                       /*bit_valid*/ 0,
5531                                       /*bit*/ 0);
5532                 ctl_done((union ctl_io *)ctsio);
5533                 return (CTL_RETVAL_COMPLETE);
5534         }
5535
5536         if ((byte2 & RWB_MODE) == RWB_MODE_DESCR) {
5537                 descr[0] = 0;
5538                 scsi_ulto3b(CTL_WRITE_BUFFER_SIZE, &descr[1]);
5539                 ctsio->kern_data_ptr = descr;
5540                 len = min(len, sizeof(descr));
5541         } else if ((byte2 & RWB_MODE) == RWB_MODE_ECHO_DESCR) {
5542                 ctsio->kern_data_ptr = echo_descr;
5543                 len = min(len, sizeof(echo_descr));
5544         } else {
5545                 if (lun->write_buffer == NULL) {
5546                         lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE,
5547                             M_CTL, M_WAITOK);
5548                 }
5549                 ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5550         }
5551         ctsio->kern_data_len = len;
5552         ctsio->kern_total_len = len;
5553         ctsio->kern_data_resid = 0;
5554         ctsio->kern_rel_offset = 0;
5555         ctsio->kern_sg_entries = 0;
5556         ctl_set_success(ctsio);
5557         ctsio->be_move_done = ctl_config_move_done;
5558         ctl_datamove((union ctl_io *)ctsio);
5559         return (CTL_RETVAL_COMPLETE);
5560 }
5561
5562 int
5563 ctl_write_buffer(struct ctl_scsiio *ctsio)
5564 {
5565         struct scsi_write_buffer *cdb;
5566         struct ctl_lun *lun;
5567         int buffer_offset, len;
5568
5569         CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5570
5571         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5572         cdb = (struct scsi_write_buffer *)ctsio->cdb;
5573
5574         if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA) {
5575                 ctl_set_invalid_field(ctsio,
5576                                       /*sks_valid*/ 1,
5577                                       /*command*/ 1,
5578                                       /*field*/ 1,
5579                                       /*bit_valid*/ 1,
5580                                       /*bit*/ 4);
5581                 ctl_done((union ctl_io *)ctsio);
5582                 return (CTL_RETVAL_COMPLETE);
5583         }
5584
5585         len = scsi_3btoul(cdb->length);
5586         buffer_offset = scsi_3btoul(cdb->offset);
5587
5588         if (buffer_offset + len > CTL_WRITE_BUFFER_SIZE) {
5589                 ctl_set_invalid_field(ctsio,
5590                                       /*sks_valid*/ 1,
5591                                       /*command*/ 1,
5592                                       /*field*/ 6,
5593                                       /*bit_valid*/ 0,
5594                                       /*bit*/ 0);
5595                 ctl_done((union ctl_io *)ctsio);
5596                 return (CTL_RETVAL_COMPLETE);
5597         }
5598
5599         /*
5600          * If we've got a kernel request that hasn't been malloced yet,
5601          * malloc it and tell the caller the data buffer is here.
5602          */
5603         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5604                 if (lun->write_buffer == NULL) {
5605                         lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE,
5606                             M_CTL, M_WAITOK);
5607                 }
5608                 ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5609                 ctsio->kern_data_len = len;
5610                 ctsio->kern_total_len = len;
5611                 ctsio->kern_data_resid = 0;
5612                 ctsio->kern_rel_offset = 0;
5613                 ctsio->kern_sg_entries = 0;
5614                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5615                 ctsio->be_move_done = ctl_config_move_done;
5616                 ctl_datamove((union ctl_io *)ctsio);
5617
5618                 return (CTL_RETVAL_COMPLETE);
5619         }
5620
5621         ctl_set_success(ctsio);
5622         ctl_done((union ctl_io *)ctsio);
5623         return (CTL_RETVAL_COMPLETE);
5624 }
5625
5626 int
5627 ctl_write_same(struct ctl_scsiio *ctsio)
5628 {
5629         struct ctl_lun *lun;
5630         struct ctl_lba_len_flags *lbalen;
5631         uint64_t lba;
5632         uint32_t num_blocks;
5633         int len, retval;
5634         uint8_t byte2;
5635
5636         CTL_DEBUG_PRINT(("ctl_write_same\n"));
5637
5638         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5639
5640         switch (ctsio->cdb[0]) {
5641         case WRITE_SAME_10: {
5642                 struct scsi_write_same_10 *cdb;
5643
5644                 cdb = (struct scsi_write_same_10 *)ctsio->cdb;
5645
5646                 lba = scsi_4btoul(cdb->addr);
5647                 num_blocks = scsi_2btoul(cdb->length);
5648                 byte2 = cdb->byte2;
5649                 break;
5650         }
5651         case WRITE_SAME_16: {
5652                 struct scsi_write_same_16 *cdb;
5653
5654                 cdb = (struct scsi_write_same_16 *)ctsio->cdb;
5655
5656                 lba = scsi_8btou64(cdb->addr);
5657                 num_blocks = scsi_4btoul(cdb->length);
5658                 byte2 = cdb->byte2;
5659                 break;
5660         }
5661         default:
5662                 /*
5663                  * We got a command we don't support.  This shouldn't
5664                  * happen, commands should be filtered out above us.
5665                  */
5666                 ctl_set_invalid_opcode(ctsio);
5667                 ctl_done((union ctl_io *)ctsio);
5668
5669                 return (CTL_RETVAL_COMPLETE);
5670                 break; /* NOTREACHED */
5671         }
5672
5673         /* ANCHOR flag can be used only together with UNMAP */
5674         if ((byte2 & SWS_UNMAP) == 0 && (byte2 & SWS_ANCHOR) != 0) {
5675                 ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
5676                     /*command*/ 1, /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 0);
5677                 ctl_done((union ctl_io *)ctsio);
5678                 return (CTL_RETVAL_COMPLETE);
5679         }
5680
5681         /*
5682          * The first check is to make sure we're in bounds, the second
5683          * check is to catch wrap-around problems.  If the lba + num blocks
5684          * is less than the lba, then we've wrapped around and the block
5685          * range is invalid anyway.
5686          */
5687         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5688          || ((lba + num_blocks) < lba)) {
5689                 ctl_set_lba_out_of_range(ctsio,
5690                     MAX(lba, lun->be_lun->maxlba + 1));
5691                 ctl_done((union ctl_io *)ctsio);
5692                 return (CTL_RETVAL_COMPLETE);
5693         }
5694
5695         /* Zero number of blocks means "to the last logical block" */
5696         if (num_blocks == 0) {
5697                 if ((lun->be_lun->maxlba + 1) - lba > UINT32_MAX) {
5698                         ctl_set_invalid_field(ctsio,
5699                                               /*sks_valid*/ 0,
5700                                               /*command*/ 1,
5701                                               /*field*/ 0,
5702                                               /*bit_valid*/ 0,
5703                                               /*bit*/ 0);
5704                         ctl_done((union ctl_io *)ctsio);
5705                         return (CTL_RETVAL_COMPLETE);
5706                 }
5707                 num_blocks = (lun->be_lun->maxlba + 1) - lba;
5708         }
5709
5710         len = lun->be_lun->blocksize;
5711
5712         /*
5713          * If we've got a kernel request that hasn't been malloced yet,
5714          * malloc it and tell the caller the data buffer is here.
5715          */
5716         if ((byte2 & SWS_NDOB) == 0 &&
5717             (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5718                 ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);
5719                 ctsio->kern_data_len = len;
5720                 ctsio->kern_total_len = len;
5721                 ctsio->kern_data_resid = 0;
5722                 ctsio->kern_rel_offset = 0;
5723                 ctsio->kern_sg_entries = 0;
5724                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5725                 ctsio->be_move_done = ctl_config_move_done;
5726                 ctl_datamove((union ctl_io *)ctsio);
5727
5728                 return (CTL_RETVAL_COMPLETE);
5729         }
5730
5731         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5732         lbalen->lba = lba;
5733         lbalen->len = num_blocks;
5734         lbalen->flags = byte2;
5735         retval = lun->backend->config_write((union ctl_io *)ctsio);
5736
5737         return (retval);
5738 }
5739
5740 int
5741 ctl_unmap(struct ctl_scsiio *ctsio)
5742 {
5743         struct ctl_lun *lun;
5744         struct scsi_unmap *cdb;
5745         struct ctl_ptr_len_flags *ptrlen;
5746         struct scsi_unmap_header *hdr;
5747         struct scsi_unmap_desc *buf, *end, *endnz, *range;
5748         uint64_t lba;
5749         uint32_t num_blocks;
5750         int len, retval;
5751         uint8_t byte2;
5752
5753         CTL_DEBUG_PRINT(("ctl_unmap\n"));
5754
5755         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5756         cdb = (struct scsi_unmap *)ctsio->cdb;
5757
5758         len = scsi_2btoul(cdb->length);
5759         byte2 = cdb->byte2;
5760
5761         /*
5762          * If we've got a kernel request that hasn't been malloced yet,
5763          * malloc it and tell the caller the data buffer is here.
5764          */
5765         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5766                 ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);
5767                 ctsio->kern_data_len = len;
5768                 ctsio->kern_total_len = len;
5769                 ctsio->kern_data_resid = 0;
5770                 ctsio->kern_rel_offset = 0;
5771                 ctsio->kern_sg_entries = 0;
5772                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5773                 ctsio->be_move_done = ctl_config_move_done;
5774                 ctl_datamove((union ctl_io *)ctsio);
5775
5776                 return (CTL_RETVAL_COMPLETE);
5777         }
5778
5779         len = ctsio->kern_total_len - ctsio->kern_data_resid;
5780         hdr = (struct scsi_unmap_header *)ctsio->kern_data_ptr;
5781         if (len < sizeof (*hdr) ||
5782             len < (scsi_2btoul(hdr->length) + sizeof(hdr->length)) ||
5783             len < (scsi_2btoul(hdr->desc_length) + sizeof (*hdr)) ||
5784             scsi_2btoul(hdr->desc_length) % sizeof(*buf) != 0) {
5785                 ctl_set_invalid_field(ctsio,
5786                                       /*sks_valid*/ 0,
5787                                       /*command*/ 0,
5788                                       /*field*/ 0,
5789                                       /*bit_valid*/ 0,
5790                                       /*bit*/ 0);
5791                 goto done;
5792         }
5793         len = scsi_2btoul(hdr->desc_length);
5794         buf = (struct scsi_unmap_desc *)(hdr + 1);
5795         end = buf + len / sizeof(*buf);
5796
5797         endnz = buf;
5798         for (range = buf; range < end; range++) {
5799                 lba = scsi_8btou64(range->lba);
5800                 num_blocks = scsi_4btoul(range->length);
5801                 if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5802                  || ((lba + num_blocks) < lba)) {
5803                         ctl_set_lba_out_of_range(ctsio,
5804                             MAX(lba, lun->be_lun->maxlba + 1));
5805                         ctl_done((union ctl_io *)ctsio);
5806                         return (CTL_RETVAL_COMPLETE);
5807                 }
5808                 if (num_blocks != 0)
5809                         endnz = range + 1;
5810         }
5811
5812         /*
5813          * Block backend can not handle zero last range.
5814          * Filter it out and return if there is nothing left.
5815          */
5816         len = (uint8_t *)endnz - (uint8_t *)buf;
5817         if (len == 0) {
5818                 ctl_set_success(ctsio);
5819                 goto done;
5820         }
5821
5822         mtx_lock(&lun->lun_lock);
5823         ptrlen = (struct ctl_ptr_len_flags *)
5824             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5825         ptrlen->ptr = (void *)buf;
5826         ptrlen->len = len;
5827         ptrlen->flags = byte2;
5828         ctl_check_blocked(lun);
5829         mtx_unlock(&lun->lun_lock);
5830
5831         retval = lun->backend->config_write((union ctl_io *)ctsio);
5832         return (retval);
5833
5834 done:
5835         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5836                 free(ctsio->kern_data_ptr, M_CTL);
5837                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5838         }
5839         ctl_done((union ctl_io *)ctsio);
5840         return (CTL_RETVAL_COMPLETE);
5841 }
5842
5843 int
5844 ctl_default_page_handler(struct ctl_scsiio *ctsio,
5845                          struct ctl_page_index *page_index, uint8_t *page_ptr)
5846 {
5847         struct ctl_lun *lun;
5848         uint8_t *current_cp, *saved_cp;
5849         int set_ua;
5850         uint32_t initidx;
5851
5852         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5853         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5854         set_ua = 0;
5855
5856         current_cp = (page_index->page_data + (page_index->page_len *
5857             CTL_PAGE_CURRENT));
5858         saved_cp = (page_index->page_data + (page_index->page_len *
5859             CTL_PAGE_SAVED));
5860
5861         mtx_lock(&lun->lun_lock);
5862         if (memcmp(current_cp, page_ptr, page_index->page_len)) {
5863                 memcpy(current_cp, page_ptr, page_index->page_len);
5864                 memcpy(saved_cp, page_ptr, page_index->page_len);
5865                 set_ua = 1;
5866         }
5867         if (set_ua != 0)
5868                 ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
5869         mtx_unlock(&lun->lun_lock);
5870         if (set_ua) {
5871                 ctl_isc_announce_mode(lun,
5872                     ctl_get_initindex(&ctsio->io_hdr.nexus),
5873                     page_index->page_code, page_index->subpage);
5874         }
5875         return (CTL_RETVAL_COMPLETE);
5876 }
5877
5878 static void
5879 ctl_ie_timer(void *arg)
5880 {
5881         struct ctl_lun *lun = arg;
5882         uint64_t t;
5883
5884         if (lun->ie_asc == 0)
5885                 return;
5886
5887         if (lun->MODE_IE.mrie == SIEP_MRIE_UA)
5888                 ctl_est_ua_all(lun, -1, CTL_UA_IE);
5889         else
5890                 lun->ie_reported = 0;
5891
5892         if (lun->ie_reportcnt < scsi_4btoul(lun->MODE_IE.report_count)) {
5893                 lun->ie_reportcnt++;
5894                 t = scsi_4btoul(lun->MODE_IE.interval_timer);
5895                 if (t == 0 || t == UINT32_MAX)
5896                         t = 3000;  /* 5 min */
5897                 callout_schedule(&lun->ie_callout, t * hz / 10);
5898         }
5899 }
5900
5901 int
5902 ctl_ie_page_handler(struct ctl_scsiio *ctsio,
5903                          struct ctl_page_index *page_index, uint8_t *page_ptr)
5904 {
5905         struct scsi_info_exceptions_page *pg;
5906         struct ctl_lun *lun;
5907         uint64_t t;
5908
5909         (void)ctl_default_page_handler(ctsio, page_index, page_ptr);
5910
5911         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5912         pg = (struct scsi_info_exceptions_page *)page_ptr;
5913         mtx_lock(&lun->lun_lock);
5914         if (pg->info_flags & SIEP_FLAGS_TEST) {
5915                 lun->ie_asc = 0x5d;
5916                 lun->ie_ascq = 0xff;
5917                 if (pg->mrie == SIEP_MRIE_UA) {
5918                         ctl_est_ua_all(lun, -1, CTL_UA_IE);
5919                         lun->ie_reported = 1;
5920                 } else {
5921                         ctl_clr_ua_all(lun, -1, CTL_UA_IE);
5922                         lun->ie_reported = -1;
5923                 }
5924                 lun->ie_reportcnt = 1;
5925                 if (lun->ie_reportcnt < scsi_4btoul(pg->report_count)) {
5926                         lun->ie_reportcnt++;
5927                         t = scsi_4btoul(pg->interval_timer);
5928                         if (t == 0 || t == UINT32_MAX)
5929                                 t = 3000;  /* 5 min */
5930                         callout_reset(&lun->ie_callout, t * hz / 10,
5931                             ctl_ie_timer, lun);
5932                 }
5933         } else {
5934                 lun->ie_asc = 0;
5935                 lun->ie_ascq = 0;
5936                 lun->ie_reported = 1;
5937                 ctl_clr_ua_all(lun, -1, CTL_UA_IE);
5938                 lun->ie_reportcnt = UINT32_MAX;
5939                 callout_stop(&lun->ie_callout);
5940         }
5941         mtx_unlock(&lun->lun_lock);
5942         return (CTL_RETVAL_COMPLETE);
5943 }
5944
5945 static int
5946 ctl_do_mode_select(union ctl_io *io)
5947 {
5948         struct scsi_mode_page_header *page_header;
5949         struct ctl_page_index *page_index;
5950         struct ctl_scsiio *ctsio;
5951         int page_len, page_len_offset, page_len_size;
5952         union ctl_modepage_info *modepage_info;
5953         struct ctl_lun *lun;
5954         int *len_left, *len_used;
5955         int retval, i;
5956
5957         ctsio = &io->scsiio;
5958         page_index = NULL;
5959         page_len = 0;
5960         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5961
5962         modepage_info = (union ctl_modepage_info *)
5963                 ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
5964         len_left = &modepage_info->header.len_left;
5965         len_used = &modepage_info->header.len_used;
5966
5967 do_next_page:
5968
5969         page_header = (struct scsi_mode_page_header *)
5970                 (ctsio->kern_data_ptr + *len_used);
5971
5972         if (*len_left == 0) {
5973                 free(ctsio->kern_data_ptr, M_CTL);
5974                 ctl_set_success(ctsio);
5975                 ctl_done((union ctl_io *)ctsio);
5976                 return (CTL_RETVAL_COMPLETE);
5977         } else if (*len_left < sizeof(struct scsi_mode_page_header)) {
5978
5979                 free(ctsio->kern_data_ptr, M_CTL);
5980                 ctl_set_param_len_error(ctsio);
5981                 ctl_done((union ctl_io *)ctsio);
5982                 return (CTL_RETVAL_COMPLETE);
5983
5984         } else if ((page_header->page_code & SMPH_SPF)
5985                 && (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
5986
5987                 free(ctsio->kern_data_ptr, M_CTL);
5988                 ctl_set_param_len_error(ctsio);
5989                 ctl_done((union ctl_io *)ctsio);
5990                 return (CTL_RETVAL_COMPLETE);
5991         }
5992
5993
5994         /*
5995          * XXX KDM should we do something with the block descriptor?
5996          */
5997         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
5998                 page_index = &lun->mode_pages.index[i];
5999                 if (lun->be_lun->lun_type == T_DIRECT &&
6000                     (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6001                         continue;
6002                 if (lun->be_lun->lun_type == T_PROCESSOR &&
6003                     (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6004                         continue;
6005                 if (lun->be_lun->lun_type == T_CDROM &&
6006                     (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6007                         continue;
6008
6009                 if ((page_index->page_code & SMPH_PC_MASK) !=
6010                     (page_header->page_code & SMPH_PC_MASK))
6011                         continue;
6012
6013                 /*
6014                  * If neither page has a subpage code, then we've got a
6015                  * match.
6016                  */
6017                 if (((page_index->page_code & SMPH_SPF) == 0)
6018                  && ((page_header->page_code & SMPH_SPF) == 0)) {
6019                         page_len = page_header->page_length;
6020                         break;
6021                 }
6022
6023                 /*
6024                  * If both pages have subpages, then the subpage numbers
6025                  * have to match.
6026                  */
6027                 if ((page_index->page_code & SMPH_SPF)
6028                   && (page_header->page_code & SMPH_SPF)) {
6029                         struct scsi_mode_page_header_sp *sph;
6030
6031                         sph = (struct scsi_mode_page_header_sp *)page_header;
6032                         if (page_index->subpage == sph->subpage) {
6033                                 page_len = scsi_2btoul(sph->page_length);
6034                                 break;
6035                         }
6036                 }
6037         }
6038
6039         /*
6040          * If we couldn't find the page, or if we don't have a mode select
6041          * handler for it, send back an error to the user.
6042          */
6043         if ((i >= CTL_NUM_MODE_PAGES)
6044          || (page_index->select_handler == NULL)) {
6045                 ctl_set_invalid_field(ctsio,
6046                                       /*sks_valid*/ 1,
6047                                       /*command*/ 0,
6048                                       /*field*/ *len_used,
6049                                       /*bit_valid*/ 0,
6050                                       /*bit*/ 0);
6051                 free(ctsio->kern_data_ptr, M_CTL);
6052                 ctl_done((union ctl_io *)ctsio);
6053                 return (CTL_RETVAL_COMPLETE);
6054         }
6055
6056         if (page_index->page_code & SMPH_SPF) {
6057                 page_len_offset = 2;
6058                 page_len_size = 2;
6059         } else {
6060                 page_len_size = 1;
6061                 page_len_offset = 1;
6062         }
6063
6064         /*
6065          * If the length the initiator gives us isn't the one we specify in
6066          * the mode page header, or if they didn't specify enough data in
6067          * the CDB to avoid truncating this page, kick out the request.
6068          */
6069         if (page_len != page_index->page_len - page_len_offset - page_len_size) {
6070                 ctl_set_invalid_field(ctsio,
6071                                       /*sks_valid*/ 1,
6072                                       /*command*/ 0,
6073                                       /*field*/ *len_used + page_len_offset,
6074                                       /*bit_valid*/ 0,
6075                                       /*bit*/ 0);
6076                 free(ctsio->kern_data_ptr, M_CTL);
6077                 ctl_done((union ctl_io *)ctsio);
6078                 return (CTL_RETVAL_COMPLETE);
6079         }
6080         if (*len_left < page_index->page_len) {
6081                 free(ctsio->kern_data_ptr, M_CTL);
6082                 ctl_set_param_len_error(ctsio);
6083                 ctl_done((union ctl_io *)ctsio);
6084                 return (CTL_RETVAL_COMPLETE);
6085         }
6086
6087         /*
6088          * Run through the mode page, checking to make sure that the bits
6089          * the user changed are actually legal for him to change.
6090          */
6091         for (i = 0; i < page_index->page_len; i++) {
6092                 uint8_t *user_byte, *change_mask, *current_byte;
6093                 int bad_bit;
6094                 int j;
6095
6096                 user_byte = (uint8_t *)page_header + i;
6097                 change_mask = page_index->page_data +
6098                               (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
6099                 current_byte = page_index->page_data +
6100                                (page_index->page_len * CTL_PAGE_CURRENT) + i;
6101
6102                 /*
6103                  * Check to see whether the user set any bits in this byte
6104                  * that he is not allowed to set.
6105                  */
6106                 if ((*user_byte & ~(*change_mask)) ==
6107                     (*current_byte & ~(*change_mask)))
6108                         continue;
6109
6110                 /*
6111                  * Go through bit by bit to determine which one is illegal.
6112                  */
6113                 bad_bit = 0;
6114                 for (j = 7; j >= 0; j--) {
6115                         if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
6116                             (((1 << i) & ~(*change_mask)) & *current_byte)) {
6117                                 bad_bit = i;
6118                                 break;
6119                         }
6120                 }
6121                 ctl_set_invalid_field(ctsio,
6122                                       /*sks_valid*/ 1,
6123                                       /*command*/ 0,
6124                                       /*field*/ *len_used + i,
6125                                       /*bit_valid*/ 1,
6126                                       /*bit*/ bad_bit);
6127                 free(ctsio->kern_data_ptr, M_CTL);
6128                 ctl_done((union ctl_io *)ctsio);
6129                 return (CTL_RETVAL_COMPLETE);
6130         }
6131
6132         /*
6133          * Decrement these before we call the page handler, since we may
6134          * end up getting called back one way or another before the handler
6135          * returns to this context.
6136          */
6137         *len_left -= page_index->page_len;
6138         *len_used += page_index->page_len;
6139
6140         retval = page_index->select_handler(ctsio, page_index,
6141                                             (uint8_t *)page_header);
6142
6143         /*
6144          * If the page handler returns CTL_RETVAL_QUEUED, then we need to
6145          * wait until this queued command completes to finish processing
6146          * the mode page.  If it returns anything other than
6147          * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
6148          * already set the sense information, freed the data pointer, and
6149          * completed the io for us.
6150          */
6151         if (retval != CTL_RETVAL_COMPLETE)
6152                 goto bailout_no_done;
6153
6154         /*
6155          * If the initiator sent us more than one page, parse the next one.
6156          */
6157         if (*len_left > 0)
6158                 goto do_next_page;
6159
6160         ctl_set_success(ctsio);
6161         free(ctsio->kern_data_ptr, M_CTL);
6162         ctl_done((union ctl_io *)ctsio);
6163
6164 bailout_no_done:
6165
6166         return (CTL_RETVAL_COMPLETE);
6167
6168 }
6169
6170 int
6171 ctl_mode_select(struct ctl_scsiio *ctsio)
6172 {
6173         int param_len, pf, sp;
6174         int header_size, bd_len;
6175         union ctl_modepage_info *modepage_info;
6176
6177         switch (ctsio->cdb[0]) {
6178         case MODE_SELECT_6: {
6179                 struct scsi_mode_select_6 *cdb;
6180
6181                 cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
6182
6183                 pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6184                 sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6185                 param_len = cdb->length;
6186                 header_size = sizeof(struct scsi_mode_header_6);
6187                 break;
6188         }
6189         case MODE_SELECT_10: {
6190                 struct scsi_mode_select_10 *cdb;
6191
6192                 cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
6193
6194                 pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6195                 sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6196                 param_len = scsi_2btoul(cdb->length);
6197                 header_size = sizeof(struct scsi_mode_header_10);
6198                 break;
6199         }
6200         default:
6201                 ctl_set_invalid_opcode(ctsio);
6202                 ctl_done((union ctl_io *)ctsio);
6203                 return (CTL_RETVAL_COMPLETE);
6204         }
6205
6206         /*
6207          * From SPC-3:
6208          * "A parameter list length of zero indicates that the Data-Out Buffer
6209          * shall be empty. This condition shall not be considered as an error."
6210          */
6211         if (param_len == 0) {
6212                 ctl_set_success(ctsio);
6213                 ctl_done((union ctl_io *)ctsio);
6214                 return (CTL_RETVAL_COMPLETE);
6215         }
6216
6217         /*
6218          * Since we'll hit this the first time through, prior to
6219          * allocation, we don't need to free a data buffer here.
6220          */
6221         if (param_len < header_size) {
6222                 ctl_set_param_len_error(ctsio);
6223                 ctl_done((union ctl_io *)ctsio);
6224                 return (CTL_RETVAL_COMPLETE);
6225         }
6226
6227         /*
6228          * Allocate the data buffer and grab the user's data.  In theory,
6229          * we shouldn't have to sanity check the parameter list length here
6230          * because the maximum size is 64K.  We should be able to malloc
6231          * that much without too many problems.
6232          */
6233         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6234                 ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6235                 ctsio->kern_data_len = param_len;
6236                 ctsio->kern_total_len = param_len;
6237                 ctsio->kern_data_resid = 0;
6238                 ctsio->kern_rel_offset = 0;
6239                 ctsio->kern_sg_entries = 0;
6240                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6241                 ctsio->be_move_done = ctl_config_move_done;
6242                 ctl_datamove((union ctl_io *)ctsio);
6243
6244                 return (CTL_RETVAL_COMPLETE);
6245         }
6246
6247         switch (ctsio->cdb[0]) {
6248         case MODE_SELECT_6: {
6249                 struct scsi_mode_header_6 *mh6;
6250
6251                 mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6252                 bd_len = mh6->blk_desc_len;
6253                 break;
6254         }
6255         case MODE_SELECT_10: {
6256                 struct scsi_mode_header_10 *mh10;
6257
6258                 mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6259                 bd_len = scsi_2btoul(mh10->blk_desc_len);
6260                 break;
6261         }
6262         default:
6263                 panic("%s: Invalid CDB type %#x", __func__, ctsio->cdb[0]);
6264         }
6265
6266         if (param_len < (header_size + bd_len)) {
6267                 free(ctsio->kern_data_ptr, M_CTL);
6268                 ctl_set_param_len_error(ctsio);
6269                 ctl_done((union ctl_io *)ctsio);
6270                 return (CTL_RETVAL_COMPLETE);
6271         }
6272
6273         /*
6274          * Set the IO_CONT flag, so that if this I/O gets passed to
6275          * ctl_config_write_done(), it'll get passed back to
6276          * ctl_do_mode_select() for further processing, or completion if
6277          * we're all done.
6278          */
6279         ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6280         ctsio->io_cont = ctl_do_mode_select;
6281
6282         modepage_info = (union ctl_modepage_info *)
6283                 ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6284         memset(modepage_info, 0, sizeof(*modepage_info));
6285         modepage_info->header.len_left = param_len - header_size - bd_len;
6286         modepage_info->header.len_used = header_size + bd_len;
6287
6288         return (ctl_do_mode_select((union ctl_io *)ctsio));
6289 }
6290
6291 int
6292 ctl_mode_sense(struct ctl_scsiio *ctsio)
6293 {
6294         struct ctl_lun *lun;
6295         int pc, page_code, dbd, llba, subpage;
6296         int alloc_len, page_len, header_len, total_len;
6297         struct scsi_mode_block_descr *block_desc;
6298         struct ctl_page_index *page_index;
6299
6300         dbd = 0;
6301         llba = 0;
6302         block_desc = NULL;
6303
6304         CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6305
6306         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6307         switch (ctsio->cdb[0]) {
6308         case MODE_SENSE_6: {
6309                 struct scsi_mode_sense_6 *cdb;
6310
6311                 cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6312
6313                 header_len = sizeof(struct scsi_mode_hdr_6);
6314                 if (cdb->byte2 & SMS_DBD)
6315                         dbd = 1;
6316                 else
6317                         header_len += sizeof(struct scsi_mode_block_descr);
6318
6319                 pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6320                 page_code = cdb->page & SMS_PAGE_CODE;
6321                 subpage = cdb->subpage;
6322                 alloc_len = cdb->length;
6323                 break;
6324         }
6325         case MODE_SENSE_10: {
6326                 struct scsi_mode_sense_10 *cdb;
6327
6328                 cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6329
6330                 header_len = sizeof(struct scsi_mode_hdr_10);
6331
6332                 if (cdb->byte2 & SMS_DBD)
6333                         dbd = 1;
6334                 else
6335                         header_len += sizeof(struct scsi_mode_block_descr);
6336                 if (cdb->byte2 & SMS10_LLBAA)
6337                         llba = 1;
6338                 pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6339                 page_code = cdb->page & SMS_PAGE_CODE;
6340                 subpage = cdb->subpage;
6341                 alloc_len = scsi_2btoul(cdb->length);
6342                 break;
6343         }
6344         default:
6345                 ctl_set_invalid_opcode(ctsio);
6346                 ctl_done((union ctl_io *)ctsio);
6347                 return (CTL_RETVAL_COMPLETE);
6348                 break; /* NOTREACHED */
6349         }
6350
6351         /*
6352          * We have to make a first pass through to calculate the size of
6353          * the pages that match the user's query.  Then we allocate enough
6354          * memory to hold it, and actually copy the data into the buffer.
6355          */
6356         switch (page_code) {
6357         case SMS_ALL_PAGES_PAGE: {
6358                 u_int i;
6359
6360                 page_len = 0;
6361
6362                 /*
6363                  * At the moment, values other than 0 and 0xff here are
6364                  * reserved according to SPC-3.
6365                  */
6366                 if ((subpage != SMS_SUBPAGE_PAGE_0)
6367                  && (subpage != SMS_SUBPAGE_ALL)) {
6368                         ctl_set_invalid_field(ctsio,
6369                                               /*sks_valid*/ 1,
6370                                               /*command*/ 1,
6371                                               /*field*/ 3,
6372                                               /*bit_valid*/ 0,
6373                                               /*bit*/ 0);
6374                         ctl_done((union ctl_io *)ctsio);
6375                         return (CTL_RETVAL_COMPLETE);
6376                 }
6377
6378                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6379                         page_index = &lun->mode_pages.index[i];
6380
6381                         /* Make sure the page is supported for this dev type */
6382                         if (lun->be_lun->lun_type == T_DIRECT &&
6383                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6384                                 continue;
6385                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6386                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6387                                 continue;
6388                         if (lun->be_lun->lun_type == T_CDROM &&
6389                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6390                                 continue;
6391
6392                         /*
6393                          * We don't use this subpage if the user didn't
6394                          * request all subpages.
6395                          */
6396                         if ((page_index->subpage != 0)
6397                          && (subpage == SMS_SUBPAGE_PAGE_0))
6398                                 continue;
6399
6400 #if 0
6401                         printf("found page %#x len %d\n",
6402                                page_index->page_code & SMPH_PC_MASK,
6403                                page_index->page_len);
6404 #endif
6405                         page_len += page_index->page_len;
6406                 }
6407                 break;
6408         }
6409         default: {
6410                 u_int i;
6411
6412                 page_len = 0;
6413
6414                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6415                         page_index = &lun->mode_pages.index[i];
6416
6417                         /* Make sure the page is supported for this dev type */
6418                         if (lun->be_lun->lun_type == T_DIRECT &&
6419                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6420                                 continue;
6421                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6422                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6423                                 continue;
6424                         if (lun->be_lun->lun_type == T_CDROM &&
6425                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6426                                 continue;
6427
6428                         /* Look for the right page code */
6429                         if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6430                                 continue;
6431
6432                         /* Look for the right subpage or the subpage wildcard*/
6433                         if ((page_index->subpage != subpage)
6434                          && (subpage != SMS_SUBPAGE_ALL))
6435                                 continue;
6436
6437 #if 0
6438                         printf("found page %#x len %d\n",
6439                                page_index->page_code & SMPH_PC_MASK,
6440                                page_index->page_len);
6441 #endif
6442
6443                         page_len += page_index->page_len;
6444                 }
6445
6446                 if (page_len == 0) {
6447                         ctl_set_invalid_field(ctsio,
6448                                               /*sks_valid*/ 1,
6449                                               /*command*/ 1,
6450                                               /*field*/ 2,
6451                                               /*bit_valid*/ 1,
6452                                               /*bit*/ 5);
6453                         ctl_done((union ctl_io *)ctsio);
6454                         return (CTL_RETVAL_COMPLETE);
6455                 }
6456                 break;
6457         }
6458         }
6459
6460         total_len = header_len + page_len;
6461 #if 0
6462         printf("header_len = %d, page_len = %d, total_len = %d\n",
6463                header_len, page_len, total_len);
6464 #endif
6465
6466         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6467         ctsio->kern_sg_entries = 0;
6468         ctsio->kern_data_resid = 0;
6469         ctsio->kern_rel_offset = 0;
6470         if (total_len < alloc_len) {
6471                 ctsio->residual = alloc_len - total_len;
6472                 ctsio->kern_data_len = total_len;
6473                 ctsio->kern_total_len = total_len;
6474         } else {
6475                 ctsio->residual = 0;
6476                 ctsio->kern_data_len = alloc_len;
6477                 ctsio->kern_total_len = alloc_len;
6478         }
6479
6480         switch (ctsio->cdb[0]) {
6481         case MODE_SENSE_6: {
6482                 struct scsi_mode_hdr_6 *header;
6483
6484                 header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
6485
6486                 header->datalen = MIN(total_len - 1, 254);
6487                 if (lun->be_lun->lun_type == T_DIRECT) {
6488                         header->dev_specific = 0x10; /* DPOFUA */
6489                         if ((lun->be_lun->flags & CTL_LUN_FLAG_READONLY) ||
6490                             (lun->MODE_CTRL.eca_and_aen & SCP_SWP) != 0)
6491                                 header->dev_specific |= 0x80; /* WP */
6492                 }
6493                 if (dbd)
6494                         header->block_descr_len = 0;
6495                 else
6496                         header->block_descr_len =
6497                                 sizeof(struct scsi_mode_block_descr);
6498                 block_desc = (struct scsi_mode_block_descr *)&header[1];
6499                 break;
6500         }
6501         case MODE_SENSE_10: {
6502                 struct scsi_mode_hdr_10 *header;
6503                 int datalen;
6504
6505                 header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
6506
6507                 datalen = MIN(total_len - 2, 65533);
6508                 scsi_ulto2b(datalen, header->datalen);
6509                 if (lun->be_lun->lun_type == T_DIRECT) {
6510                         header->dev_specific = 0x10; /* DPOFUA */
6511                         if ((lun->be_lun->flags & CTL_LUN_FLAG_READONLY) ||
6512                             (lun->MODE_CTRL.eca_and_aen & SCP_SWP) != 0)
6513                                 header->dev_specific |= 0x80; /* WP */
6514                 }
6515                 if (dbd)
6516                         scsi_ulto2b(0, header->block_descr_len);
6517                 else
6518                         scsi_ulto2b(sizeof(struct scsi_mode_block_descr),
6519                                     header->block_descr_len);
6520                 block_desc = (struct scsi_mode_block_descr *)&header[1];
6521                 break;
6522         }
6523         default:
6524                 panic("%s: Invalid CDB type %#x", __func__, ctsio->cdb[0]);
6525         }
6526
6527         /*
6528          * If we've got a disk, use its blocksize in the block
6529          * descriptor.  Otherwise, just set it to 0.
6530          */
6531         if (dbd == 0) {
6532                 if (lun->be_lun->lun_type == T_DIRECT)
6533                         scsi_ulto3b(lun->be_lun->blocksize,
6534                                     block_desc->block_len);
6535                 else
6536                         scsi_ulto3b(0, block_desc->block_len);
6537         }
6538
6539         switch (page_code) {
6540         case SMS_ALL_PAGES_PAGE: {
6541                 int i, data_used;
6542
6543                 data_used = header_len;
6544                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6545                         struct ctl_page_index *page_index;
6546
6547                         page_index = &lun->mode_pages.index[i];
6548                         if (lun->be_lun->lun_type == T_DIRECT &&
6549                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6550                                 continue;
6551                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6552                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6553                                 continue;
6554                         if (lun->be_lun->lun_type == T_CDROM &&
6555                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6556                                 continue;
6557
6558                         /*
6559                          * We don't use this subpage if the user didn't
6560                          * request all subpages.  We already checked (above)
6561                          * to make sure the user only specified a subpage
6562                          * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case.
6563                          */
6564                         if ((page_index->subpage != 0)
6565                          && (subpage == SMS_SUBPAGE_PAGE_0))
6566                                 continue;
6567
6568                         /*
6569                          * Call the handler, if it exists, to update the
6570                          * page to the latest values.
6571                          */
6572                         if (page_index->sense_handler != NULL)
6573                                 page_index->sense_handler(ctsio, page_index,pc);
6574
6575                         memcpy(ctsio->kern_data_ptr + data_used,
6576                                page_index->page_data +
6577                                (page_index->page_len * pc),
6578                                page_index->page_len);
6579                         data_used += page_index->page_len;
6580                 }
6581                 break;
6582         }
6583         default: {
6584                 int i, data_used;
6585
6586                 data_used = header_len;
6587
6588                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6589                         struct ctl_page_index *page_index;
6590
6591                         page_index = &lun->mode_pages.index[i];
6592
6593                         /* Look for the right page code */
6594                         if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6595                                 continue;
6596
6597                         /* Look for the right subpage or the subpage wildcard*/
6598                         if ((page_index->subpage != subpage)
6599                          && (subpage != SMS_SUBPAGE_ALL))
6600                                 continue;
6601
6602                         /* Make sure the page is supported for this dev type */
6603                         if (lun->be_lun->lun_type == T_DIRECT &&
6604                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6605                                 continue;
6606                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6607                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6608                                 continue;
6609                         if (lun->be_lun->lun_type == T_CDROM &&
6610                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6611                                 continue;
6612
6613                         /*
6614                          * Call the handler, if it exists, to update the
6615                          * page to the latest values.
6616                          */
6617                         if (page_index->sense_handler != NULL)
6618                                 page_index->sense_handler(ctsio, page_index,pc);
6619
6620                         memcpy(ctsio->kern_data_ptr + data_used,
6621                                page_index->page_data +
6622                                (page_index->page_len * pc),
6623                                page_index->page_len);
6624                         data_used += page_index->page_len;
6625                 }
6626                 break;
6627         }
6628         }
6629
6630         ctl_set_success(ctsio);
6631         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6632         ctsio->be_move_done = ctl_config_move_done;
6633         ctl_datamove((union ctl_io *)ctsio);
6634         return (CTL_RETVAL_COMPLETE);
6635 }
6636
6637 int
6638 ctl_lbp_log_sense_handler(struct ctl_scsiio *ctsio,
6639                                struct ctl_page_index *page_index,
6640                                int pc)
6641 {
6642         struct ctl_lun *lun;
6643         struct scsi_log_param_header *phdr;
6644         uint8_t *data;
6645         uint64_t val;
6646
6647         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6648         data = page_index->page_data;
6649
6650         if (lun->backend->lun_attr != NULL &&
6651             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "blocksavail"))
6652              != UINT64_MAX) {
6653                 phdr = (struct scsi_log_param_header *)data;
6654                 scsi_ulto2b(0x0001, phdr->param_code);
6655                 phdr->param_control = SLP_LBIN | SLP_LP;
6656                 phdr->param_len = 8;
6657                 data = (uint8_t *)(phdr + 1);
6658                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6659                 data[4] = 0x02; /* per-pool */
6660                 data += phdr->param_len;
6661         }
6662
6663         if (lun->backend->lun_attr != NULL &&
6664             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "blocksused"))
6665              != UINT64_MAX) {
6666                 phdr = (struct scsi_log_param_header *)data;
6667                 scsi_ulto2b(0x0002, phdr->param_code);
6668                 phdr->param_control = SLP_LBIN | SLP_LP;
6669                 phdr->param_len = 8;
6670                 data = (uint8_t *)(phdr + 1);
6671                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6672                 data[4] = 0x01; /* per-LUN */
6673                 data += phdr->param_len;
6674         }
6675
6676         if (lun->backend->lun_attr != NULL &&
6677             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "poolblocksavail"))
6678              != UINT64_MAX) {
6679                 phdr = (struct scsi_log_param_header *)data;
6680                 scsi_ulto2b(0x00f1, phdr->param_code);
6681                 phdr->param_control = SLP_LBIN | SLP_LP;
6682                 phdr->param_len = 8;
6683                 data = (uint8_t *)(phdr + 1);
6684                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6685                 data[4] = 0x02; /* per-pool */
6686                 data += phdr->param_len;
6687         }
6688
6689         if (lun->backend->lun_attr != NULL &&
6690             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "poolblocksused"))
6691              != UINT64_MAX) {
6692                 phdr = (struct scsi_log_param_header *)data;
6693                 scsi_ulto2b(0x00f2, phdr->param_code);
6694                 phdr->param_control = SLP_LBIN | SLP_LP;
6695                 phdr->param_len = 8;
6696                 data = (uint8_t *)(phdr + 1);
6697                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6698                 data[4] = 0x02; /* per-pool */
6699                 data += phdr->param_len;
6700         }
6701
6702         page_index->page_len = data - page_index->page_data;
6703         return (0);
6704 }
6705
6706 int
6707 ctl_sap_log_sense_handler(struct ctl_scsiio *ctsio,
6708                                struct ctl_page_index *page_index,
6709                                int pc)
6710 {
6711         struct ctl_lun *lun;
6712         struct stat_page *data;
6713         uint64_t rn, wn, rb, wb;
6714         struct bintime rt, wt;
6715         int i;
6716
6717         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6718         data = (struct stat_page *)page_index->page_data;
6719
6720         scsi_ulto2b(SLP_SAP, data->sap.hdr.param_code);
6721         data->sap.hdr.param_control = SLP_LBIN;
6722         data->sap.hdr.param_len = sizeof(struct scsi_log_stat_and_perf) -
6723             sizeof(struct scsi_log_param_header);
6724         rn = wn = rb = wb = 0;
6725         bintime_clear(&rt);
6726         bintime_clear(&wt);
6727         for (i = 0; i < CTL_MAX_PORTS; i++) {
6728                 rn += lun->stats.ports[i].operations[CTL_STATS_READ];
6729                 wn += lun->stats.ports[i].operations[CTL_STATS_WRITE];
6730                 rb += lun->stats.ports[i].bytes[CTL_STATS_READ];
6731                 wb += lun->stats.ports[i].bytes[CTL_STATS_WRITE];
6732                 bintime_add(&rt, &lun->stats.ports[i].time[CTL_STATS_READ]);
6733                 bintime_add(&wt, &lun->stats.ports[i].time[CTL_STATS_WRITE]);
6734         }
6735         scsi_u64to8b(rn, data->sap.read_num);
6736         scsi_u64to8b(wn, data->sap.write_num);
6737         if (lun->stats.blocksize > 0) {
6738                 scsi_u64to8b(wb / lun->stats.blocksize,
6739                     data->sap.recvieved_lba);
6740                 scsi_u64to8b(rb / lun->stats.blocksize,
6741                     data->sap.transmitted_lba);
6742         }
6743         scsi_u64to8b((uint64_t)rt.sec * 1000 + rt.frac / (UINT64_MAX / 1000),
6744             data->sap.read_int);
6745         scsi_u64to8b((uint64_t)wt.sec * 1000 + wt.frac / (UINT64_MAX / 1000),
6746             data->sap.write_int);
6747         scsi_u64to8b(0, data->sap.weighted_num);
6748         scsi_u64to8b(0, data->sap.weighted_int);
6749         scsi_ulto2b(SLP_IT, data->it.hdr.param_code);
6750         data->it.hdr.param_control = SLP_LBIN;
6751         data->it.hdr.param_len = sizeof(struct scsi_log_idle_time) -
6752             sizeof(struct scsi_log_param_header);
6753 #ifdef CTL_TIME_IO
6754         scsi_u64to8b(lun->idle_time / SBT_1MS, data->it.idle_int);
6755 #endif
6756         scsi_ulto2b(SLP_TI, data->ti.hdr.param_code);
6757         data->it.hdr.param_control = SLP_LBIN;
6758         data->ti.hdr.param_len = sizeof(struct scsi_log_time_interval) -
6759             sizeof(struct scsi_log_param_header);
6760         scsi_ulto4b(3, data->ti.exponent);
6761         scsi_ulto4b(1, data->ti.integer);
6762         return (0);
6763 }
6764
6765 int
6766 ctl_ie_log_sense_handler(struct ctl_scsiio *ctsio,
6767                                struct ctl_page_index *page_index,
6768                                int pc)
6769 {
6770         struct ctl_lun *lun;
6771         struct scsi_log_informational_exceptions *data;
6772
6773         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6774         data = (struct scsi_log_informational_exceptions *)page_index->page_data;
6775
6776         scsi_ulto2b(SLP_IE_GEN, data->hdr.param_code);
6777         data->hdr.param_control = SLP_LBIN;
6778         data->hdr.param_len = sizeof(struct scsi_log_informational_exceptions) -
6779             sizeof(struct scsi_log_param_header);
6780         data->ie_asc = lun->ie_asc;
6781         data->ie_ascq = lun->ie_ascq;
6782         data->temperature = 0xff;
6783         return (0);
6784 }
6785
6786 int
6787 ctl_log_sense(struct ctl_scsiio *ctsio)
6788 {
6789         struct ctl_lun *lun;
6790         int i, pc, page_code, subpage;
6791         int alloc_len, total_len;
6792         struct ctl_page_index *page_index;
6793         struct scsi_log_sense *cdb;
6794         struct scsi_log_header *header;
6795
6796         CTL_DEBUG_PRINT(("ctl_log_sense\n"));
6797
6798         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6799         cdb = (struct scsi_log_sense *)ctsio->cdb;
6800         pc = (cdb->page & SLS_PAGE_CTRL_MASK) >> 6;
6801         page_code = cdb->page & SLS_PAGE_CODE;
6802         subpage = cdb->subpage;
6803         alloc_len = scsi_2btoul(cdb->length);
6804
6805         page_index = NULL;
6806         for (i = 0; i < CTL_NUM_LOG_PAGES; i++) {
6807                 page_index = &lun->log_pages.index[i];
6808
6809                 /* Look for the right page code */
6810                 if ((page_index->page_code & SL_PAGE_CODE) != page_code)
6811                         continue;
6812
6813                 /* Look for the right subpage or the subpage wildcard*/
6814                 if (page_index->subpage != subpage)
6815                         continue;
6816
6817                 break;
6818         }
6819         if (i >= CTL_NUM_LOG_PAGES) {
6820                 ctl_set_invalid_field(ctsio,
6821                                       /*sks_valid*/ 1,
6822                                       /*command*/ 1,
6823                                       /*field*/ 2,
6824                                       /*bit_valid*/ 0,
6825                                       /*bit*/ 0);
6826                 ctl_done((union ctl_io *)ctsio);
6827                 return (CTL_RETVAL_COMPLETE);
6828         }
6829
6830         total_len = sizeof(struct scsi_log_header) + page_index->page_len;
6831
6832         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6833         ctsio->kern_sg_entries = 0;
6834         ctsio->kern_data_resid = 0;
6835         ctsio->kern_rel_offset = 0;
6836         if (total_len < alloc_len) {
6837                 ctsio->residual = alloc_len - total_len;
6838                 ctsio->kern_data_len = total_len;
6839                 ctsio->kern_total_len = total_len;
6840         } else {
6841                 ctsio->residual = 0;
6842                 ctsio->kern_data_len = alloc_len;
6843                 ctsio->kern_total_len = alloc_len;
6844         }
6845
6846         header = (struct scsi_log_header *)ctsio->kern_data_ptr;
6847         header->page = page_index->page_code;
6848         if (page_index->page_code == SLS_LOGICAL_BLOCK_PROVISIONING)
6849                 header->page |= SL_DS;
6850         if (page_index->subpage) {
6851                 header->page |= SL_SPF;
6852                 header->subpage = page_index->subpage;
6853         }
6854         scsi_ulto2b(page_index->page_len, header->datalen);
6855
6856         /*
6857          * Call the handler, if it exists, to update the
6858          * page to the latest values.
6859          */
6860         if (page_index->sense_handler != NULL)
6861                 page_index->sense_handler(ctsio, page_index, pc);
6862
6863         memcpy(header + 1, page_index->page_data, page_index->page_len);
6864
6865         ctl_set_success(ctsio);
6866         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6867         ctsio->be_move_done = ctl_config_move_done;
6868         ctl_datamove((union ctl_io *)ctsio);
6869         return (CTL_RETVAL_COMPLETE);
6870 }
6871
6872 int
6873 ctl_read_capacity(struct ctl_scsiio *ctsio)
6874 {
6875         struct scsi_read_capacity *cdb;
6876         struct scsi_read_capacity_data *data;
6877         struct ctl_lun *lun;
6878         uint32_t lba;
6879
6880         CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
6881
6882         cdb = (struct scsi_read_capacity *)ctsio->cdb;
6883
6884         lba = scsi_4btoul(cdb->addr);
6885         if (((cdb->pmi & SRC_PMI) == 0)
6886          && (lba != 0)) {
6887                 ctl_set_invalid_field(/*ctsio*/ ctsio,
6888                                       /*sks_valid*/ 1,
6889                                       /*command*/ 1,
6890                                       /*field*/ 2,
6891                                       /*bit_valid*/ 0,
6892                                       /*bit*/ 0);
6893                 ctl_done((union ctl_io *)ctsio);
6894                 return (CTL_RETVAL_COMPLETE);
6895         }
6896
6897         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6898
6899         ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
6900         data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
6901         ctsio->residual = 0;
6902         ctsio->kern_data_len = sizeof(*data);
6903         ctsio->kern_total_len = sizeof(*data);
6904         ctsio->kern_data_resid = 0;
6905         ctsio->kern_rel_offset = 0;
6906         ctsio->kern_sg_entries = 0;
6907
6908         /*
6909          * If the maximum LBA is greater than 0xfffffffe, the user must
6910          * issue a SERVICE ACTION IN (16) command, with the read capacity
6911          * serivce action set.
6912          */
6913         if (lun->be_lun->maxlba > 0xfffffffe)
6914                 scsi_ulto4b(0xffffffff, data->addr);
6915         else
6916                 scsi_ulto4b(lun->be_lun->maxlba, data->addr);
6917
6918         /*
6919          * XXX KDM this may not be 512 bytes...
6920          */
6921         scsi_ulto4b(lun->be_lun->blocksize, data->length);
6922
6923         ctl_set_success(ctsio);
6924         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6925         ctsio->be_move_done = ctl_config_move_done;
6926         ctl_datamove((union ctl_io *)ctsio);
6927         return (CTL_RETVAL_COMPLETE);
6928 }
6929
6930 int
6931 ctl_read_capacity_16(struct ctl_scsiio *ctsio)
6932 {
6933         struct scsi_read_capacity_16 *cdb;
6934         struct scsi_read_capacity_data_long *data;
6935         struct ctl_lun *lun;
6936         uint64_t lba;
6937         uint32_t alloc_len;
6938
6939         CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
6940
6941         cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
6942
6943         alloc_len = scsi_4btoul(cdb->alloc_len);
6944         lba = scsi_8btou64(cdb->addr);
6945
6946         if ((cdb->reladr & SRC16_PMI)
6947          && (lba != 0)) {
6948                 ctl_set_invalid_field(/*ctsio*/ ctsio,
6949                                       /*sks_valid*/ 1,
6950                                       /*command*/ 1,
6951                                       /*field*/ 2,
6952                                       /*bit_valid*/ 0,
6953                                       /*bit*/ 0);
6954                 ctl_done((union ctl_io *)ctsio);
6955                 return (CTL_RETVAL_COMPLETE);
6956         }
6957
6958         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6959
6960         ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
6961         data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
6962
6963         if (sizeof(*data) < alloc_len) {
6964                 ctsio->residual = alloc_len - sizeof(*data);
6965                 ctsio->kern_data_len = sizeof(*data);
6966                 ctsio->kern_total_len = sizeof(*data);
6967         } else {
6968                 ctsio->residual = 0;
6969                 ctsio->kern_data_len = alloc_len;
6970                 ctsio->kern_total_len = alloc_len;
6971         }
6972         ctsio->kern_data_resid = 0;
6973         ctsio->kern_rel_offset = 0;
6974         ctsio->kern_sg_entries = 0;
6975
6976         scsi_u64to8b(lun->be_lun->maxlba, data->addr);
6977         /* XXX KDM this may not be 512 bytes... */
6978         scsi_ulto4b(lun->be_lun->blocksize, data->length);
6979         data->prot_lbppbe = lun->be_lun->pblockexp & SRC16_LBPPBE;
6980         scsi_ulto2b(lun->be_lun->pblockoff & SRC16_LALBA_A, data->lalba_lbp);
6981         if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
6982                 data->lalba_lbp[0] |= SRC16_LBPME | SRC16_LBPRZ;
6983
6984         ctl_set_success(ctsio);
6985         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6986         ctsio->be_move_done = ctl_config_move_done;
6987         ctl_datamove((union ctl_io *)ctsio);
6988         return (CTL_RETVAL_COMPLETE);
6989 }
6990
6991 int
6992 ctl_get_lba_status(struct ctl_scsiio *ctsio)
6993 {
6994         struct scsi_get_lba_status *cdb;
6995         struct scsi_get_lba_status_data *data;
6996         struct ctl_lun *lun;
6997         struct ctl_lba_len_flags *lbalen;
6998         uint64_t lba;
6999         uint32_t alloc_len, total_len;
7000         int retval;
7001
7002         CTL_DEBUG_PRINT(("ctl_get_lba_status\n"));
7003
7004         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7005         cdb = (struct scsi_get_lba_status *)ctsio->cdb;
7006         lba = scsi_8btou64(cdb->addr);
7007         alloc_len = scsi_4btoul(cdb->alloc_len);
7008
7009         if (lba > lun->be_lun->maxlba) {
7010                 ctl_set_lba_out_of_range(ctsio, lba);
7011                 ctl_done((union ctl_io *)ctsio);
7012                 return (CTL_RETVAL_COMPLETE);
7013         }
7014
7015         total_len = sizeof(*data) + sizeof(data->descr[0]);
7016         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7017         data = (struct scsi_get_lba_status_data *)ctsio->kern_data_ptr;
7018
7019         if (total_len < alloc_len) {
7020                 ctsio->residual = alloc_len - total_len;
7021                 ctsio->kern_data_len = total_len;
7022                 ctsio->kern_total_len = total_len;
7023         } else {
7024                 ctsio->residual = 0;
7025                 ctsio->kern_data_len = alloc_len;
7026                 ctsio->kern_total_len = alloc_len;
7027         }
7028         ctsio->kern_data_resid = 0;
7029         ctsio->kern_rel_offset = 0;
7030         ctsio->kern_sg_entries = 0;
7031
7032         /* Fill dummy data in case backend can't tell anything. */
7033         scsi_ulto4b(4 + sizeof(data->descr[0]), data->length);
7034         scsi_u64to8b(lba, data->descr[0].addr);
7035         scsi_ulto4b(MIN(UINT32_MAX, lun->be_lun->maxlba + 1 - lba),
7036             data->descr[0].length);
7037         data->descr[0].status = 0; /* Mapped or unknown. */
7038
7039         ctl_set_success(ctsio);
7040         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7041         ctsio->be_move_done = ctl_config_move_done;
7042
7043         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
7044         lbalen->lba = lba;
7045         lbalen->len = total_len;
7046         lbalen->flags = 0;
7047         retval = lun->backend->config_read((union ctl_io *)ctsio);
7048         return (CTL_RETVAL_COMPLETE);
7049 }
7050
7051 int
7052 ctl_read_defect(struct ctl_scsiio *ctsio)
7053 {
7054         struct scsi_read_defect_data_10 *ccb10;
7055         struct scsi_read_defect_data_12 *ccb12;
7056         struct scsi_read_defect_data_hdr_10 *data10;
7057         struct scsi_read_defect_data_hdr_12 *data12;
7058         uint32_t alloc_len, data_len;
7059         uint8_t format;
7060
7061         CTL_DEBUG_PRINT(("ctl_read_defect\n"));
7062
7063         if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
7064                 ccb10 = (struct scsi_read_defect_data_10 *)&ctsio->cdb;
7065                 format = ccb10->format;
7066                 alloc_len = scsi_2btoul(ccb10->alloc_length);
7067                 data_len = sizeof(*data10);
7068         } else {
7069                 ccb12 = (struct scsi_read_defect_data_12 *)&ctsio->cdb;
7070                 format = ccb12->format;
7071                 alloc_len = scsi_4btoul(ccb12->alloc_length);
7072                 data_len = sizeof(*data12);
7073         }
7074         if (alloc_len == 0) {
7075                 ctl_set_success(ctsio);
7076                 ctl_done((union ctl_io *)ctsio);
7077                 return (CTL_RETVAL_COMPLETE);
7078         }
7079
7080         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
7081         if (data_len < alloc_len) {
7082                 ctsio->residual = alloc_len - data_len;
7083                 ctsio->kern_data_len = data_len;
7084                 ctsio->kern_total_len = data_len;
7085         } else {
7086                 ctsio->residual = 0;
7087                 ctsio->kern_data_len = alloc_len;
7088                 ctsio->kern_total_len = alloc_len;
7089         }
7090         ctsio->kern_data_resid = 0;
7091         ctsio->kern_rel_offset = 0;
7092         ctsio->kern_sg_entries = 0;
7093
7094         if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
7095                 data10 = (struct scsi_read_defect_data_hdr_10 *)
7096                     ctsio->kern_data_ptr;
7097                 data10->format = format;
7098                 scsi_ulto2b(0, data10->length);
7099         } else {
7100                 data12 = (struct scsi_read_defect_data_hdr_12 *)
7101                     ctsio->kern_data_ptr;
7102                 data12->format = format;
7103                 scsi_ulto2b(0, data12->generation);
7104                 scsi_ulto4b(0, data12->length);
7105         }
7106
7107         ctl_set_success(ctsio);
7108         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7109         ctsio->be_move_done = ctl_config_move_done;
7110         ctl_datamove((union ctl_io *)ctsio);
7111         return (CTL_RETVAL_COMPLETE);
7112 }
7113
7114 int
7115 ctl_report_tagret_port_groups(struct ctl_scsiio *ctsio)
7116 {
7117         struct scsi_maintenance_in *cdb;
7118         int retval;
7119         int alloc_len, ext, total_len = 0, g, pc, pg, ts, os;
7120         int num_ha_groups, num_target_ports, shared_group;
7121         struct ctl_lun *lun;
7122         struct ctl_softc *softc;
7123         struct ctl_port *port;
7124         struct scsi_target_group_data *rtg_ptr;
7125         struct scsi_target_group_data_extended *rtg_ext_ptr;
7126         struct scsi_target_port_group_descriptor *tpg_desc;
7127
7128         CTL_DEBUG_PRINT(("ctl_report_tagret_port_groups\n"));
7129
7130         cdb = (struct scsi_maintenance_in *)ctsio->cdb;
7131         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7132         softc = lun->ctl_softc;
7133
7134         retval = CTL_RETVAL_COMPLETE;
7135
7136         switch (cdb->byte2 & STG_PDF_MASK) {
7137         case STG_PDF_LENGTH:
7138                 ext = 0;
7139                 break;
7140         case STG_PDF_EXTENDED:
7141                 ext = 1;
7142                 break;
7143         default:
7144                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7145                                       /*sks_valid*/ 1,
7146                                       /*command*/ 1,
7147                                       /*field*/ 2,
7148                                       /*bit_valid*/ 1,
7149                                       /*bit*/ 5);
7150                 ctl_done((union ctl_io *)ctsio);
7151                 return(retval);
7152         }
7153
7154         num_target_ports = 0;
7155         shared_group = (softc->is_single != 0);
7156         mtx_lock(&softc->ctl_lock);
7157         STAILQ_FOREACH(port, &softc->port_list, links) {
7158                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7159                         continue;
7160                 if (ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
7161                         continue;
7162                 num_target_ports++;
7163                 if (port->status & CTL_PORT_STATUS_HA_SHARED)
7164                         shared_group = 1;
7165         }
7166         mtx_unlock(&softc->ctl_lock);
7167         num_ha_groups = (softc->is_single) ? 0 : NUM_HA_SHELVES;
7168
7169         if (ext)
7170                 total_len = sizeof(struct scsi_target_group_data_extended);
7171         else
7172                 total_len = sizeof(struct scsi_target_group_data);
7173         total_len += sizeof(struct scsi_target_port_group_descriptor) *
7174                 (shared_group + num_ha_groups) +
7175             sizeof(struct scsi_target_port_descriptor) * num_target_ports;
7176
7177         alloc_len = scsi_4btoul(cdb->length);
7178
7179         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7180
7181         ctsio->kern_sg_entries = 0;
7182
7183         if (total_len < alloc_len) {
7184                 ctsio->residual = alloc_len - total_len;
7185                 ctsio->kern_data_len = total_len;
7186                 ctsio->kern_total_len = total_len;
7187         } else {
7188                 ctsio->residual = 0;
7189                 ctsio->kern_data_len = alloc_len;
7190                 ctsio->kern_total_len = alloc_len;
7191         }
7192         ctsio->kern_data_resid = 0;
7193         ctsio->kern_rel_offset = 0;
7194
7195         if (ext) {
7196                 rtg_ext_ptr = (struct scsi_target_group_data_extended *)
7197                     ctsio->kern_data_ptr;
7198                 scsi_ulto4b(total_len - 4, rtg_ext_ptr->length);
7199                 rtg_ext_ptr->format_type = 0x10;
7200                 rtg_ext_ptr->implicit_transition_time = 0;
7201                 tpg_desc = &rtg_ext_ptr->groups[0];
7202         } else {
7203                 rtg_ptr = (struct scsi_target_group_data *)
7204                     ctsio->kern_data_ptr;
7205                 scsi_ulto4b(total_len - 4, rtg_ptr->length);
7206                 tpg_desc = &rtg_ptr->groups[0];
7207         }
7208
7209         mtx_lock(&softc->ctl_lock);
7210         pg = softc->port_min / softc->port_cnt;
7211         if (lun->flags & (CTL_LUN_PRIMARY_SC | CTL_LUN_PEER_SC_PRIMARY)) {
7212                 /* Some shelf is known to be primary. */
7213                 if (softc->ha_link == CTL_HA_LINK_OFFLINE)
7214                         os = TPG_ASYMMETRIC_ACCESS_UNAVAILABLE;
7215                 else if (softc->ha_link == CTL_HA_LINK_UNKNOWN)
7216                         os = TPG_ASYMMETRIC_ACCESS_TRANSITIONING;
7217                 else if (softc->ha_mode == CTL_HA_MODE_ACT_STBY)
7218                         os = TPG_ASYMMETRIC_ACCESS_STANDBY;
7219                 else
7220                         os = TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7221                 if (lun->flags & CTL_LUN_PRIMARY_SC) {
7222                         ts = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7223                 } else {
7224                         ts = os;
7225                         os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7226                 }
7227         } else {
7228                 /* No known primary shelf. */
7229                 if (softc->ha_link == CTL_HA_LINK_OFFLINE) {
7230                         ts = TPG_ASYMMETRIC_ACCESS_UNAVAILABLE;
7231                         os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7232                 } else if (softc->ha_link == CTL_HA_LINK_UNKNOWN) {
7233                         ts = TPG_ASYMMETRIC_ACCESS_TRANSITIONING;
7234                         os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7235                 } else {
7236                         ts = os = TPG_ASYMMETRIC_ACCESS_TRANSITIONING;
7237                 }
7238         }
7239         if (shared_group) {
7240                 tpg_desc->pref_state = ts;
7241                 tpg_desc->support = TPG_AO_SUP | TPG_AN_SUP | TPG_S_SUP |
7242                     TPG_U_SUP | TPG_T_SUP;
7243                 scsi_ulto2b(1, tpg_desc->target_port_group);
7244                 tpg_desc->status = TPG_IMPLICIT;
7245                 pc = 0;
7246                 STAILQ_FOREACH(port, &softc->port_list, links) {
7247                         if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7248                                 continue;
7249                         if (!softc->is_single &&
7250                             (port->status & CTL_PORT_STATUS_HA_SHARED) == 0)
7251                                 continue;
7252                         if (ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
7253                                 continue;
7254                         scsi_ulto2b(port->targ_port, tpg_desc->descriptors[pc].
7255                             relative_target_port_identifier);
7256                         pc++;
7257                 }
7258                 tpg_desc->target_port_count = pc;
7259                 tpg_desc = (struct scsi_target_port_group_descriptor *)
7260                     &tpg_desc->descriptors[pc];
7261         }
7262         for (g = 0; g < num_ha_groups; g++) {
7263                 tpg_desc->pref_state = (g == pg) ? ts : os;
7264                 tpg_desc->support = TPG_AO_SUP | TPG_AN_SUP | TPG_S_SUP |
7265                     TPG_U_SUP | TPG_T_SUP;
7266                 scsi_ulto2b(2 + g, tpg_desc->target_port_group);
7267                 tpg_desc->status = TPG_IMPLICIT;
7268                 pc = 0;
7269                 STAILQ_FOREACH(port, &softc->port_list, links) {
7270                         if (port->targ_port < g * softc->port_cnt ||
7271                             port->targ_port >= (g + 1) * softc->port_cnt)
7272                                 continue;
7273                         if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7274                                 continue;
7275                         if (port->status & CTL_PORT_STATUS_HA_SHARED)
7276                                 continue;
7277                         if (ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
7278                                 continue;
7279                         scsi_ulto2b(port->targ_port, tpg_desc->descriptors[pc].
7280                             relative_target_port_identifier);
7281                         pc++;
7282                 }
7283                 tpg_desc->target_port_count = pc;
7284                 tpg_desc = (struct scsi_target_port_group_descriptor *)
7285                     &tpg_desc->descriptors[pc];
7286         }
7287         mtx_unlock(&softc->ctl_lock);
7288
7289         ctl_set_success(ctsio);
7290         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7291         ctsio->be_move_done = ctl_config_move_done;
7292         ctl_datamove((union ctl_io *)ctsio);
7293         return(retval);
7294 }
7295
7296 int
7297 ctl_report_supported_opcodes(struct ctl_scsiio *ctsio)
7298 {
7299         struct ctl_lun *lun;
7300         struct scsi_report_supported_opcodes *cdb;
7301         const struct ctl_cmd_entry *entry, *sentry;
7302         struct scsi_report_supported_opcodes_all *all;
7303         struct scsi_report_supported_opcodes_descr *descr;
7304         struct scsi_report_supported_opcodes_one *one;
7305         int retval;
7306         int alloc_len, total_len;
7307         int opcode, service_action, i, j, num;
7308
7309         CTL_DEBUG_PRINT(("ctl_report_supported_opcodes\n"));
7310
7311         cdb = (struct scsi_report_supported_opcodes *)ctsio->cdb;
7312         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7313
7314         retval = CTL_RETVAL_COMPLETE;
7315
7316         opcode = cdb->requested_opcode;
7317         service_action = scsi_2btoul(cdb->requested_service_action);
7318         switch (cdb->options & RSO_OPTIONS_MASK) {
7319         case RSO_OPTIONS_ALL:
7320                 num = 0;
7321                 for (i = 0; i < 256; i++) {
7322                         entry = &ctl_cmd_table[i];
7323                         if (entry->flags & CTL_CMD_FLAG_SA5) {
7324                                 for (j = 0; j < 32; j++) {
7325                                         sentry = &((const struct ctl_cmd_entry *)
7326                                             entry->execute)[j];
7327                                         if (ctl_cmd_applicable(
7328                                             lun->be_lun->lun_type, sentry))
7329                                                 num++;
7330                                 }
7331                         } else {
7332                                 if (ctl_cmd_applicable(lun->be_lun->lun_type,
7333                                     entry))
7334                                         num++;
7335                         }
7336                 }
7337                 total_len = sizeof(struct scsi_report_supported_opcodes_all) +
7338                     num * sizeof(struct scsi_report_supported_opcodes_descr);
7339                 break;
7340         case RSO_OPTIONS_OC:
7341                 if (ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) {
7342                         ctl_set_invalid_field(/*ctsio*/ ctsio,
7343                                               /*sks_valid*/ 1,
7344                                               /*command*/ 1,
7345                                               /*field*/ 2,
7346                                               /*bit_valid*/ 1,
7347                                               /*bit*/ 2);
7348                         ctl_done((union ctl_io *)ctsio);
7349                         return (CTL_RETVAL_COMPLETE);
7350                 }
7351                 total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7352                 break;
7353         case RSO_OPTIONS_OC_SA:
7354                 if ((ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) == 0 ||
7355                     service_action >= 32) {
7356                         ctl_set_invalid_field(/*ctsio*/ ctsio,
7357                                               /*sks_valid*/ 1,
7358                                               /*command*/ 1,
7359                                               /*field*/ 2,
7360                                               /*bit_valid*/ 1,
7361                                               /*bit*/ 2);
7362                         ctl_done((union ctl_io *)ctsio);
7363                         return (CTL_RETVAL_COMPLETE);
7364                 }
7365                 total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7366                 break;
7367         default:
7368                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7369                                       /*sks_valid*/ 1,
7370                                       /*command*/ 1,
7371                                       /*field*/ 2,
7372                                       /*bit_valid*/ 1,
7373                                       /*bit*/ 2);
7374                 ctl_done((union ctl_io *)ctsio);
7375                 return (CTL_RETVAL_COMPLETE);
7376         }
7377
7378         alloc_len = scsi_4btoul(cdb->length);
7379
7380         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7381
7382         ctsio->kern_sg_entries = 0;
7383
7384         if (total_len < alloc_len) {
7385                 ctsio->residual = alloc_len - total_len;
7386                 ctsio->kern_data_len = total_len;
7387                 ctsio->kern_total_len = total_len;
7388         } else {
7389                 ctsio->residual = 0;
7390                 ctsio->kern_data_len = alloc_len;
7391                 ctsio->kern_total_len = alloc_len;
7392         }
7393         ctsio->kern_data_resid = 0;
7394         ctsio->kern_rel_offset = 0;
7395
7396         switch (cdb->options & RSO_OPTIONS_MASK) {
7397         case RSO_OPTIONS_ALL:
7398                 all = (struct scsi_report_supported_opcodes_all *)
7399                     ctsio->kern_data_ptr;
7400                 num = 0;
7401                 for (i = 0; i < 256; i++) {
7402                         entry = &ctl_cmd_table[i];
7403                         if (entry->flags & CTL_CMD_FLAG_SA5) {
7404                                 for (j = 0; j < 32; j++) {
7405                                         sentry = &((const struct ctl_cmd_entry *)
7406                                             entry->execute)[j];
7407                                         if (!ctl_cmd_applicable(
7408                                             lun->be_lun->lun_type, sentry))
7409                                                 continue;
7410                                         descr = &all->descr[num++];
7411                                         descr->opcode = i;
7412                                         scsi_ulto2b(j, descr->service_action);
7413                                         descr->flags = RSO_SERVACTV;
7414                                         scsi_ulto2b(sentry->length,
7415                                             descr->cdb_length);
7416                                 }
7417                         } else {
7418                                 if (!ctl_cmd_applicable(lun->be_lun->lun_type,
7419                                     entry))
7420                                         continue;
7421                                 descr = &all->descr[num++];
7422                                 descr->opcode = i;
7423                                 scsi_ulto2b(0, descr->service_action);
7424                                 descr->flags = 0;
7425                                 scsi_ulto2b(entry->length, descr->cdb_length);
7426                         }
7427                 }
7428                 scsi_ulto4b(
7429                     num * sizeof(struct scsi_report_supported_opcodes_descr),
7430                     all->length);
7431                 break;
7432         case RSO_OPTIONS_OC:
7433                 one = (struct scsi_report_supported_opcodes_one *)
7434                     ctsio->kern_data_ptr;
7435                 entry = &ctl_cmd_table[opcode];
7436                 goto fill_one;
7437         case RSO_OPTIONS_OC_SA:
7438                 one = (struct scsi_report_supported_opcodes_one *)
7439                     ctsio->kern_data_ptr;
7440                 entry = &ctl_cmd_table[opcode];
7441                 entry = &((const struct ctl_cmd_entry *)
7442                     entry->execute)[service_action];
7443 fill_one:
7444                 if (ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
7445                         one->support = 3;
7446                         scsi_ulto2b(entry->length, one->cdb_length);
7447                         one->cdb_usage[0] = opcode;
7448                         memcpy(&one->cdb_usage[1], entry->usage,
7449                             entry->length - 1);
7450                 } else
7451                         one->support = 1;
7452                 break;
7453         }
7454
7455         ctl_set_success(ctsio);
7456         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7457         ctsio->be_move_done = ctl_config_move_done;
7458         ctl_datamove((union ctl_io *)ctsio);
7459         return(retval);
7460 }
7461
7462 int
7463 ctl_report_supported_tmf(struct ctl_scsiio *ctsio)
7464 {
7465         struct scsi_report_supported_tmf *cdb;
7466         struct scsi_report_supported_tmf_data *data;
7467         int retval;
7468         int alloc_len, total_len;
7469
7470         CTL_DEBUG_PRINT(("ctl_report_supported_tmf\n"));
7471
7472         cdb = (struct scsi_report_supported_tmf *)ctsio->cdb;
7473
7474         retval = CTL_RETVAL_COMPLETE;
7475
7476         total_len = sizeof(struct scsi_report_supported_tmf_data);
7477         alloc_len = scsi_4btoul(cdb->length);
7478
7479         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7480
7481         ctsio->kern_sg_entries = 0;
7482
7483         if (total_len < alloc_len) {
7484                 ctsio->residual = alloc_len - total_len;
7485                 ctsio->kern_data_len = total_len;
7486                 ctsio->kern_total_len = total_len;
7487         } else {
7488                 ctsio->residual = 0;
7489                 ctsio->kern_data_len = alloc_len;
7490                 ctsio->kern_total_len = alloc_len;
7491         }
7492         ctsio->kern_data_resid = 0;
7493         ctsio->kern_rel_offset = 0;
7494
7495         data = (struct scsi_report_supported_tmf_data *)ctsio->kern_data_ptr;
7496         data->byte1 |= RST_ATS | RST_ATSS | RST_CTSS | RST_LURS | RST_QTS |
7497             RST_TRS;
7498         data->byte2 |= RST_QAES | RST_QTSS | RST_ITNRS;
7499
7500         ctl_set_success(ctsio);
7501         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7502         ctsio->be_move_done = ctl_config_move_done;
7503         ctl_datamove((union ctl_io *)ctsio);
7504         return (retval);
7505 }
7506
7507 int
7508 ctl_report_timestamp(struct ctl_scsiio *ctsio)
7509 {
7510         struct scsi_report_timestamp *cdb;
7511         struct scsi_report_timestamp_data *data;
7512         struct timeval tv;
7513         int64_t timestamp;
7514         int retval;
7515         int alloc_len, total_len;
7516
7517         CTL_DEBUG_PRINT(("ctl_report_timestamp\n"));
7518
7519         cdb = (struct scsi_report_timestamp *)ctsio->cdb;
7520
7521         retval = CTL_RETVAL_COMPLETE;
7522
7523         total_len = sizeof(struct scsi_report_timestamp_data);
7524         alloc_len = scsi_4btoul(cdb->length);
7525
7526         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7527
7528         ctsio->kern_sg_entries = 0;
7529
7530         if (total_len < alloc_len) {
7531                 ctsio->residual = alloc_len - total_len;
7532                 ctsio->kern_data_len = total_len;
7533                 ctsio->kern_total_len = total_len;
7534         } else {
7535                 ctsio->residual = 0;
7536                 ctsio->kern_data_len = alloc_len;
7537                 ctsio->kern_total_len = alloc_len;
7538         }
7539         ctsio->kern_data_resid = 0;
7540         ctsio->kern_rel_offset = 0;
7541
7542         data = (struct scsi_report_timestamp_data *)ctsio->kern_data_ptr;
7543         scsi_ulto2b(sizeof(*data) - 2, data->length);
7544         data->origin = RTS_ORIG_OUTSIDE;
7545         getmicrotime(&tv);
7546         timestamp = (int64_t)tv.tv_sec * 1000 + tv.tv_usec / 1000;
7547         scsi_ulto4b(timestamp >> 16, data->timestamp);
7548         scsi_ulto2b(timestamp & 0xffff, &data->timestamp[4]);
7549
7550         ctl_set_success(ctsio);
7551         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7552         ctsio->be_move_done = ctl_config_move_done;
7553         ctl_datamove((union ctl_io *)ctsio);
7554         return (retval);
7555 }
7556
7557 int
7558 ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7559 {
7560         struct scsi_per_res_in *cdb;
7561         int alloc_len, total_len = 0;
7562         /* struct scsi_per_res_in_rsrv in_data; */
7563         struct ctl_lun *lun;
7564         struct ctl_softc *softc;
7565         uint64_t key;
7566
7567         CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7568
7569         cdb = (struct scsi_per_res_in *)ctsio->cdb;
7570
7571         alloc_len = scsi_2btoul(cdb->length);
7572
7573         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7574         softc = lun->ctl_softc;
7575
7576 retry:
7577         mtx_lock(&lun->lun_lock);
7578         switch (cdb->action) {
7579         case SPRI_RK: /* read keys */
7580                 total_len = sizeof(struct scsi_per_res_in_keys) +
7581                         lun->pr_key_count *
7582                         sizeof(struct scsi_per_res_key);
7583                 break;
7584         case SPRI_RR: /* read reservation */
7585                 if (lun->flags & CTL_LUN_PR_RESERVED)
7586                         total_len = sizeof(struct scsi_per_res_in_rsrv);
7587                 else
7588                         total_len = sizeof(struct scsi_per_res_in_header);
7589                 break;
7590         case SPRI_RC: /* report capabilities */
7591                 total_len = sizeof(struct scsi_per_res_cap);
7592                 break;
7593         case SPRI_RS: /* read full status */
7594                 total_len = sizeof(struct scsi_per_res_in_header) +
7595                     (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7596                     lun->pr_key_count;
7597                 break;
7598         default:
7599                 panic("%s: Invalid PR type %#x", __func__, cdb->action);
7600         }
7601         mtx_unlock(&lun->lun_lock);
7602
7603         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7604
7605         if (total_len < alloc_len) {
7606                 ctsio->residual = alloc_len - total_len;
7607                 ctsio->kern_data_len = total_len;
7608                 ctsio->kern_total_len = total_len;
7609         } else {
7610                 ctsio->residual = 0;
7611                 ctsio->kern_data_len = alloc_len;
7612                 ctsio->kern_total_len = alloc_len;
7613         }
7614
7615         ctsio->kern_data_resid = 0;
7616         ctsio->kern_rel_offset = 0;
7617         ctsio->kern_sg_entries = 0;
7618
7619         mtx_lock(&lun->lun_lock);
7620         switch (cdb->action) {
7621         case SPRI_RK: { // read keys
7622         struct scsi_per_res_in_keys *res_keys;
7623                 int i, key_count;
7624
7625                 res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7626
7627                 /*
7628                  * We had to drop the lock to allocate our buffer, which
7629                  * leaves time for someone to come in with another
7630                  * persistent reservation.  (That is unlikely, though,
7631                  * since this should be the only persistent reservation
7632                  * command active right now.)
7633                  */
7634                 if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7635                     (lun->pr_key_count *
7636                      sizeof(struct scsi_per_res_key)))){
7637                         mtx_unlock(&lun->lun_lock);
7638                         free(ctsio->kern_data_ptr, M_CTL);
7639                         printf("%s: reservation length changed, retrying\n",
7640                                __func__);
7641                         goto retry;
7642                 }
7643
7644                 scsi_ulto4b(lun->pr_generation, res_keys->header.generation);
7645
7646                 scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7647                              lun->pr_key_count, res_keys->header.length);
7648
7649                 for (i = 0, key_count = 0; i < CTL_MAX_INITIATORS; i++) {
7650                         if ((key = ctl_get_prkey(lun, i)) == 0)
7651                                 continue;
7652
7653                         /*
7654                          * We used lun->pr_key_count to calculate the
7655                          * size to allocate.  If it turns out the number of
7656                          * initiators with the registered flag set is
7657                          * larger than that (i.e. they haven't been kept in
7658                          * sync), we've got a problem.
7659                          */
7660                         if (key_count >= lun->pr_key_count) {
7661                                 key_count++;
7662                                 continue;
7663                         }
7664                         scsi_u64to8b(key, res_keys->keys[key_count].key);
7665                         key_count++;
7666                 }
7667                 break;
7668         }
7669         case SPRI_RR: { // read reservation
7670                 struct scsi_per_res_in_rsrv *res;
7671                 int tmp_len, header_only;
7672
7673                 res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7674
7675                 scsi_ulto4b(lun->pr_generation, res->header.generation);
7676
7677                 if (lun->flags & CTL_LUN_PR_RESERVED)
7678                 {
7679                         tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7680                         scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7681                                     res->header.length);
7682                         header_only = 0;
7683                 } else {
7684                         tmp_len = sizeof(struct scsi_per_res_in_header);
7685                         scsi_ulto4b(0, res->header.length);
7686                         header_only = 1;
7687                 }
7688
7689                 /*
7690                  * We had to drop the lock to allocate our buffer, which
7691                  * leaves time for someone to come in with another
7692                  * persistent reservation.  (That is unlikely, though,
7693                  * since this should be the only persistent reservation
7694                  * command active right now.)
7695                  */
7696                 if (tmp_len != total_len) {
7697                         mtx_unlock(&lun->lun_lock);
7698                         free(ctsio->kern_data_ptr, M_CTL);
7699                         printf("%s: reservation status changed, retrying\n",
7700                                __func__);
7701                         goto retry;
7702                 }
7703
7704                 /*
7705                  * No reservation held, so we're done.
7706                  */
7707                 if (header_only != 0)
7708                         break;
7709
7710                 /*
7711                  * If the registration is an All Registrants type, the key
7712                  * is 0, since it doesn't really matter.
7713                  */
7714                 if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7715                         scsi_u64to8b(ctl_get_prkey(lun, lun->pr_res_idx),
7716                             res->data.reservation);
7717                 }
7718                 res->data.scopetype = lun->pr_res_type;
7719                 break;
7720         }
7721         case SPRI_RC:     //report capabilities
7722         {
7723                 struct scsi_per_res_cap *res_cap;
7724                 uint16_t type_mask;
7725
7726                 res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7727                 scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7728                 res_cap->flags1 = SPRI_CRH;
7729                 res_cap->flags2 = SPRI_TMV | SPRI_ALLOW_5;
7730                 type_mask = SPRI_TM_WR_EX_AR |
7731                             SPRI_TM_EX_AC_RO |
7732                             SPRI_TM_WR_EX_RO |
7733                             SPRI_TM_EX_AC |
7734                             SPRI_TM_WR_EX |
7735                             SPRI_TM_EX_AC_AR;
7736                 scsi_ulto2b(type_mask, res_cap->type_mask);
7737                 break;
7738         }
7739         case SPRI_RS: { // read full status
7740                 struct scsi_per_res_in_full *res_status;
7741                 struct scsi_per_res_in_full_desc *res_desc;
7742                 struct ctl_port *port;
7743                 int i, len;
7744
7745                 res_status = (struct scsi_per_res_in_full*)ctsio->kern_data_ptr;
7746
7747                 /*
7748                  * We had to drop the lock to allocate our buffer, which
7749                  * leaves time for someone to come in with another
7750                  * persistent reservation.  (That is unlikely, though,
7751                  * since this should be the only persistent reservation
7752                  * command active right now.)
7753                  */
7754                 if (total_len < (sizeof(struct scsi_per_res_in_header) +
7755                     (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7756                      lun->pr_key_count)){
7757                         mtx_unlock(&lun->lun_lock);
7758                         free(ctsio->kern_data_ptr, M_CTL);
7759                         printf("%s: reservation length changed, retrying\n",
7760                                __func__);
7761                         goto retry;
7762                 }
7763
7764                 scsi_ulto4b(lun->pr_generation, res_status->header.generation);
7765
7766                 res_desc = &res_status->desc[0];
7767                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
7768                         if ((key = ctl_get_prkey(lun, i)) == 0)
7769                                 continue;
7770
7771                         scsi_u64to8b(key, res_desc->res_key.key);
7772                         if ((lun->flags & CTL_LUN_PR_RESERVED) &&
7773                             (lun->pr_res_idx == i ||
7774                              lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS)) {
7775                                 res_desc->flags = SPRI_FULL_R_HOLDER;
7776                                 res_desc->scopetype = lun->pr_res_type;
7777                         }
7778                         scsi_ulto2b(i / CTL_MAX_INIT_PER_PORT,
7779                             res_desc->rel_trgt_port_id);
7780                         len = 0;
7781                         port = softc->ctl_ports[i / CTL_MAX_INIT_PER_PORT];
7782                         if (port != NULL)
7783                                 len = ctl_create_iid(port,
7784                                     i % CTL_MAX_INIT_PER_PORT,
7785                                     res_desc->transport_id);
7786                         scsi_ulto4b(len, res_desc->additional_length);
7787                         res_desc = (struct scsi_per_res_in_full_desc *)
7788                             &res_desc->transport_id[len];
7789                 }
7790                 scsi_ulto4b((uint8_t *)res_desc - (uint8_t *)&res_status->desc[0],
7791                     res_status->header.length);
7792                 break;
7793         }
7794         default:
7795                 panic("%s: Invalid PR type %#x", __func__, cdb->action);
7796         }
7797         mtx_unlock(&lun->lun_lock);
7798
7799         ctl_set_success(ctsio);
7800         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7801         ctsio->be_move_done = ctl_config_move_done;
7802         ctl_datamove((union ctl_io *)ctsio);
7803         return (CTL_RETVAL_COMPLETE);
7804 }
7805
7806 /*
7807  * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
7808  * it should return.
7809  */
7810 static int
7811 ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
7812                 uint64_t sa_res_key, uint8_t type, uint32_t residx,
7813                 struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
7814                 struct scsi_per_res_out_parms* param)
7815 {
7816         union ctl_ha_msg persis_io;
7817         int i;
7818
7819         mtx_lock(&lun->lun_lock);
7820         if (sa_res_key == 0) {
7821                 if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
7822                         /* validate scope and type */
7823                         if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7824                              SPR_LU_SCOPE) {
7825                                 mtx_unlock(&lun->lun_lock);
7826                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7827                                                       /*sks_valid*/ 1,
7828                                                       /*command*/ 1,
7829                                                       /*field*/ 2,
7830                                                       /*bit_valid*/ 1,
7831                                                       /*bit*/ 4);
7832                                 ctl_done((union ctl_io *)ctsio);
7833                                 return (1);
7834                         }
7835
7836                         if (type>8 || type==2 || type==4 || type==0) {
7837                                 mtx_unlock(&lun->lun_lock);
7838                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7839                                                       /*sks_valid*/ 1,
7840                                                       /*command*/ 1,
7841                                                       /*field*/ 2,
7842                                                       /*bit_valid*/ 1,
7843                                                       /*bit*/ 0);
7844                                 ctl_done((union ctl_io *)ctsio);
7845                                 return (1);
7846                         }
7847
7848                         /*
7849                          * Unregister everybody else and build UA for
7850                          * them
7851                          */
7852                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
7853                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
7854                                         continue;
7855
7856                                 ctl_clr_prkey(lun, i);
7857                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7858                         }
7859                         lun->pr_key_count = 1;
7860                         lun->pr_res_type = type;
7861                         if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
7862                             lun->pr_res_type != SPR_TYPE_EX_AC_AR)
7863                                 lun->pr_res_idx = residx;
7864                         lun->pr_generation++;
7865                         mtx_unlock(&lun->lun_lock);
7866
7867                         /* send msg to other side */
7868                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7869                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7870                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7871                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
7872                         persis_io.pr.pr_info.res_type = type;
7873                         memcpy(persis_io.pr.pr_info.sa_res_key,
7874                                param->serv_act_res_key,
7875                                sizeof(param->serv_act_res_key));
7876                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
7877                             sizeof(persis_io.pr), M_WAITOK);
7878                 } else {
7879                         /* not all registrants */
7880                         mtx_unlock(&lun->lun_lock);
7881                         free(ctsio->kern_data_ptr, M_CTL);
7882                         ctl_set_invalid_field(ctsio,
7883                                               /*sks_valid*/ 1,
7884                                               /*command*/ 0,
7885                                               /*field*/ 8,
7886                                               /*bit_valid*/ 0,
7887                                               /*bit*/ 0);
7888                         ctl_done((union ctl_io *)ctsio);
7889                         return (1);
7890                 }
7891         } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
7892                 || !(lun->flags & CTL_LUN_PR_RESERVED)) {
7893                 int found = 0;
7894
7895                 if (res_key == sa_res_key) {
7896                         /* special case */
7897                         /*
7898                          * The spec implies this is not good but doesn't
7899                          * say what to do. There are two choices either
7900                          * generate a res conflict or check condition
7901                          * with illegal field in parameter data. Since
7902                          * that is what is done when the sa_res_key is
7903                          * zero I'll take that approach since this has
7904                          * to do with the sa_res_key.
7905                          */
7906                         mtx_unlock(&lun->lun_lock);
7907                         free(ctsio->kern_data_ptr, M_CTL);
7908                         ctl_set_invalid_field(ctsio,
7909                                               /*sks_valid*/ 1,
7910                                               /*command*/ 0,
7911                                               /*field*/ 8,
7912                                               /*bit_valid*/ 0,
7913                                               /*bit*/ 0);
7914                         ctl_done((union ctl_io *)ctsio);
7915                         return (1);
7916                 }
7917
7918                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
7919                         if (ctl_get_prkey(lun, i) != sa_res_key)
7920                                 continue;
7921
7922                         found = 1;
7923                         ctl_clr_prkey(lun, i);
7924                         lun->pr_key_count--;
7925                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7926                 }
7927                 if (!found) {
7928                         mtx_unlock(&lun->lun_lock);
7929                         free(ctsio->kern_data_ptr, M_CTL);
7930                         ctl_set_reservation_conflict(ctsio);
7931                         ctl_done((union ctl_io *)ctsio);
7932                         return (CTL_RETVAL_COMPLETE);
7933                 }
7934                 lun->pr_generation++;
7935                 mtx_unlock(&lun->lun_lock);
7936
7937                 /* send msg to other side */
7938                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7939                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7940                 persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7941                 persis_io.pr.pr_info.residx = lun->pr_res_idx;
7942                 persis_io.pr.pr_info.res_type = type;
7943                 memcpy(persis_io.pr.pr_info.sa_res_key,
7944                        param->serv_act_res_key,
7945                        sizeof(param->serv_act_res_key));
7946                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
7947                     sizeof(persis_io.pr), M_WAITOK);
7948         } else {
7949                 /* Reserved but not all registrants */
7950                 /* sa_res_key is res holder */
7951                 if (sa_res_key == ctl_get_prkey(lun, lun->pr_res_idx)) {
7952                         /* validate scope and type */
7953                         if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7954                              SPR_LU_SCOPE) {
7955                                 mtx_unlock(&lun->lun_lock);
7956                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7957                                                       /*sks_valid*/ 1,
7958                                                       /*command*/ 1,
7959                                                       /*field*/ 2,
7960                                                       /*bit_valid*/ 1,
7961                                                       /*bit*/ 4);
7962                                 ctl_done((union ctl_io *)ctsio);
7963                                 return (1);
7964                         }
7965
7966                         if (type>8 || type==2 || type==4 || type==0) {
7967                                 mtx_unlock(&lun->lun_lock);
7968                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7969                                                       /*sks_valid*/ 1,
7970                                                       /*command*/ 1,
7971                                                       /*field*/ 2,
7972                                                       /*bit_valid*/ 1,
7973                                                       /*bit*/ 0);
7974                                 ctl_done((union ctl_io *)ctsio);
7975                                 return (1);
7976                         }
7977
7978                         /*
7979                          * Do the following:
7980                          * if sa_res_key != res_key remove all
7981                          * registrants w/sa_res_key and generate UA
7982                          * for these registrants(Registrations
7983                          * Preempted) if it wasn't an exclusive
7984                          * reservation generate UA(Reservations
7985                          * Preempted) for all other registered nexuses
7986                          * if the type has changed. Establish the new
7987                          * reservation and holder. If res_key and
7988                          * sa_res_key are the same do the above
7989                          * except don't unregister the res holder.
7990                          */
7991
7992                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
7993                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
7994                                         continue;
7995
7996                                 if (sa_res_key == ctl_get_prkey(lun, i)) {
7997                                         ctl_clr_prkey(lun, i);
7998                                         lun->pr_key_count--;
7999                                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8000                                 } else if (type != lun->pr_res_type &&
8001                                     (lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8002                                      lun->pr_res_type == SPR_TYPE_EX_AC_RO)) {
8003                                         ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8004                                 }
8005                         }
8006                         lun->pr_res_type = type;
8007                         if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
8008                             lun->pr_res_type != SPR_TYPE_EX_AC_AR)
8009                                 lun->pr_res_idx = residx;
8010                         else
8011                                 lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8012                         lun->pr_generation++;
8013                         mtx_unlock(&lun->lun_lock);
8014
8015                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8016                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8017                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8018                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8019                         persis_io.pr.pr_info.res_type = type;
8020                         memcpy(persis_io.pr.pr_info.sa_res_key,
8021                                param->serv_act_res_key,
8022                                sizeof(param->serv_act_res_key));
8023                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8024                             sizeof(persis_io.pr), M_WAITOK);
8025                 } else {
8026                         /*
8027                          * sa_res_key is not the res holder just
8028                          * remove registrants
8029                          */
8030                         int found=0;
8031
8032                         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8033                                 if (sa_res_key != ctl_get_prkey(lun, i))
8034                                         continue;
8035
8036                                 found = 1;
8037                                 ctl_clr_prkey(lun, i);
8038                                 lun->pr_key_count--;
8039                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8040                         }
8041
8042                         if (!found) {
8043                                 mtx_unlock(&lun->lun_lock);
8044                                 free(ctsio->kern_data_ptr, M_CTL);
8045                                 ctl_set_reservation_conflict(ctsio);
8046                                 ctl_done((union ctl_io *)ctsio);
8047                                 return (1);
8048                         }
8049                         lun->pr_generation++;
8050                         mtx_unlock(&lun->lun_lock);
8051
8052                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8053                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8054                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8055                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8056                         persis_io.pr.pr_info.res_type = type;
8057                         memcpy(persis_io.pr.pr_info.sa_res_key,
8058                                param->serv_act_res_key,
8059                                sizeof(param->serv_act_res_key));
8060                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8061                             sizeof(persis_io.pr), M_WAITOK);
8062                 }
8063         }
8064         return (0);
8065 }
8066
8067 static void
8068 ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
8069 {
8070         uint64_t sa_res_key;
8071         int i;
8072
8073         sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
8074
8075         if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8076          || lun->pr_res_idx == CTL_PR_NO_RESERVATION
8077          || sa_res_key != ctl_get_prkey(lun, lun->pr_res_idx)) {
8078                 if (sa_res_key == 0) {
8079                         /*
8080                          * Unregister everybody else and build UA for
8081                          * them
8082                          */
8083                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
8084                                 if (i == msg->pr.pr_info.residx ||
8085                                     ctl_get_prkey(lun, i) == 0)
8086                                         continue;
8087
8088                                 ctl_clr_prkey(lun, i);
8089                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8090                         }
8091
8092                         lun->pr_key_count = 1;
8093                         lun->pr_res_type = msg->pr.pr_info.res_type;
8094                         if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
8095                             lun->pr_res_type != SPR_TYPE_EX_AC_AR)
8096                                 lun->pr_res_idx = msg->pr.pr_info.residx;
8097                 } else {
8098                         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8099                                 if (sa_res_key == ctl_get_prkey(lun, i))
8100                                         continue;
8101
8102                                 ctl_clr_prkey(lun, i);
8103                                 lun->pr_key_count--;
8104                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8105                         }
8106                 }
8107         } else {
8108                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8109                         if (i == msg->pr.pr_info.residx ||
8110                             ctl_get_prkey(lun, i) == 0)
8111                                 continue;
8112
8113                         if (sa_res_key == ctl_get_prkey(lun, i)) {
8114                                 ctl_clr_prkey(lun, i);
8115                                 lun->pr_key_count--;
8116                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8117                         } else if (msg->pr.pr_info.res_type != lun->pr_res_type
8118                             && (lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8119                              lun->pr_res_type == SPR_TYPE_EX_AC_RO)) {
8120                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8121                         }
8122                 }
8123                 lun->pr_res_type = msg->pr.pr_info.res_type;
8124                 if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
8125                     lun->pr_res_type != SPR_TYPE_EX_AC_AR)
8126                         lun->pr_res_idx = msg->pr.pr_info.residx;
8127                 else
8128                         lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8129         }
8130         lun->pr_generation++;
8131
8132 }
8133
8134
8135 int
8136 ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
8137 {
8138         int retval;
8139         u_int32_t param_len;
8140         struct scsi_per_res_out *cdb;
8141         struct ctl_lun *lun;
8142         struct scsi_per_res_out_parms* param;
8143         struct ctl_softc *softc;
8144         uint32_t residx;
8145         uint64_t res_key, sa_res_key, key;
8146         uint8_t type;
8147         union ctl_ha_msg persis_io;
8148         int    i;
8149
8150         CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n"));
8151
8152         retval = CTL_RETVAL_COMPLETE;
8153
8154         cdb = (struct scsi_per_res_out *)ctsio->cdb;
8155         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8156         softc = lun->ctl_softc;
8157
8158         /*
8159          * We only support whole-LUN scope.  The scope & type are ignored for
8160          * register, register and ignore existing key and clear.
8161          * We sometimes ignore scope and type on preempts too!!
8162          * Verify reservation type here as well.
8163          */
8164         type = cdb->scope_type & SPR_TYPE_MASK;
8165         if ((cdb->action == SPRO_RESERVE)
8166          || (cdb->action == SPRO_RELEASE)) {
8167                 if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
8168                         ctl_set_invalid_field(/*ctsio*/ ctsio,
8169                                               /*sks_valid*/ 1,
8170                                               /*command*/ 1,
8171                                               /*field*/ 2,
8172                                               /*bit_valid*/ 1,
8173                                               /*bit*/ 4);
8174                         ctl_done((union ctl_io *)ctsio);
8175                         return (CTL_RETVAL_COMPLETE);
8176                 }
8177
8178                 if (type>8 || type==2 || type==4 || type==0) {
8179                         ctl_set_invalid_field(/*ctsio*/ ctsio,
8180                                               /*sks_valid*/ 1,
8181                                               /*command*/ 1,
8182                                               /*field*/ 2,
8183                                               /*bit_valid*/ 1,
8184                                               /*bit*/ 0);
8185                         ctl_done((union ctl_io *)ctsio);
8186                         return (CTL_RETVAL_COMPLETE);
8187                 }
8188         }
8189
8190         param_len = scsi_4btoul(cdb->length);
8191
8192         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
8193                 ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
8194                 ctsio->kern_data_len = param_len;
8195                 ctsio->kern_total_len = param_len;
8196                 ctsio->kern_data_resid = 0;
8197                 ctsio->kern_rel_offset = 0;
8198                 ctsio->kern_sg_entries = 0;
8199                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
8200                 ctsio->be_move_done = ctl_config_move_done;
8201                 ctl_datamove((union ctl_io *)ctsio);
8202
8203                 return (CTL_RETVAL_COMPLETE);
8204         }
8205
8206         param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
8207
8208         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
8209         res_key = scsi_8btou64(param->res_key.key);
8210         sa_res_key = scsi_8btou64(param->serv_act_res_key);
8211
8212         /*
8213          * Validate the reservation key here except for SPRO_REG_IGNO
8214          * This must be done for all other service actions
8215          */
8216         if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
8217                 mtx_lock(&lun->lun_lock);
8218                 if ((key = ctl_get_prkey(lun, residx)) != 0) {
8219                         if (res_key != key) {
8220                                 /*
8221                                  * The current key passed in doesn't match
8222                                  * the one the initiator previously
8223                                  * registered.
8224                                  */
8225                                 mtx_unlock(&lun->lun_lock);
8226                                 free(ctsio->kern_data_ptr, M_CTL);
8227                                 ctl_set_reservation_conflict(ctsio);
8228                                 ctl_done((union ctl_io *)ctsio);
8229                                 return (CTL_RETVAL_COMPLETE);
8230                         }
8231                 } else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
8232                         /*
8233                          * We are not registered
8234                          */
8235                         mtx_unlock(&lun->lun_lock);
8236                         free(ctsio->kern_data_ptr, M_CTL);
8237                         ctl_set_reservation_conflict(ctsio);
8238                         ctl_done((union ctl_io *)ctsio);
8239                         return (CTL_RETVAL_COMPLETE);
8240                 } else if (res_key != 0) {
8241                         /*
8242                          * We are not registered and trying to register but
8243                          * the register key isn't zero.
8244                          */
8245                         mtx_unlock(&lun->lun_lock);
8246                         free(ctsio->kern_data_ptr, M_CTL);
8247                         ctl_set_reservation_conflict(ctsio);
8248                         ctl_done((union ctl_io *)ctsio);
8249                         return (CTL_RETVAL_COMPLETE);
8250                 }
8251                 mtx_unlock(&lun->lun_lock);
8252         }
8253
8254         switch (cdb->action & SPRO_ACTION_MASK) {
8255         case SPRO_REGISTER:
8256         case SPRO_REG_IGNO: {
8257
8258 #if 0
8259                 printf("Registration received\n");
8260 #endif
8261
8262                 /*
8263                  * We don't support any of these options, as we report in
8264                  * the read capabilities request (see
8265                  * ctl_persistent_reserve_in(), above).
8266                  */
8267                 if ((param->flags & SPR_SPEC_I_PT)
8268                  || (param->flags & SPR_ALL_TG_PT)
8269                  || (param->flags & SPR_APTPL)) {
8270                         int bit_ptr;
8271
8272                         if (param->flags & SPR_APTPL)
8273                                 bit_ptr = 0;
8274                         else if (param->flags & SPR_ALL_TG_PT)
8275                                 bit_ptr = 2;
8276                         else /* SPR_SPEC_I_PT */
8277                                 bit_ptr = 3;
8278
8279                         free(ctsio->kern_data_ptr, M_CTL);
8280                         ctl_set_invalid_field(ctsio,
8281                                               /*sks_valid*/ 1,
8282                                               /*command*/ 0,
8283                                               /*field*/ 20,
8284                                               /*bit_valid*/ 1,
8285                                               /*bit*/ bit_ptr);
8286                         ctl_done((union ctl_io *)ctsio);
8287                         return (CTL_RETVAL_COMPLETE);
8288                 }
8289
8290                 mtx_lock(&lun->lun_lock);
8291
8292                 /*
8293                  * The initiator wants to clear the
8294                  * key/unregister.
8295                  */
8296                 if (sa_res_key == 0) {
8297                         if ((res_key == 0
8298                           && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
8299                          || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
8300                           && ctl_get_prkey(lun, residx) == 0)) {
8301                                 mtx_unlock(&lun->lun_lock);
8302                                 goto done;
8303                         }
8304
8305                         ctl_clr_prkey(lun, residx);
8306                         lun->pr_key_count--;
8307
8308                         if (residx == lun->pr_res_idx) {
8309                                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8310                                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8311
8312                                 if ((lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8313                                      lun->pr_res_type == SPR_TYPE_EX_AC_RO) &&
8314                                     lun->pr_key_count) {
8315                                         /*
8316                                          * If the reservation is a registrants
8317                                          * only type we need to generate a UA
8318                                          * for other registered inits.  The
8319                                          * sense code should be RESERVATIONS
8320                                          * RELEASED
8321                                          */
8322
8323                                         for (i = softc->init_min; i < softc->init_max; i++){
8324                                                 if (ctl_get_prkey(lun, i) == 0)
8325                                                         continue;
8326                                                 ctl_est_ua(lun, i,
8327                                                     CTL_UA_RES_RELEASE);
8328                                         }
8329                                 }
8330                                 lun->pr_res_type = 0;
8331                         } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8332                                 if (lun->pr_key_count==0) {
8333                                         lun->flags &= ~CTL_LUN_PR_RESERVED;
8334                                         lun->pr_res_type = 0;
8335                                         lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8336                                 }
8337                         }
8338                         lun->pr_generation++;
8339                         mtx_unlock(&lun->lun_lock);
8340
8341                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8342                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8343                         persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
8344                         persis_io.pr.pr_info.residx = residx;
8345                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8346                             sizeof(persis_io.pr), M_WAITOK);
8347                 } else /* sa_res_key != 0 */ {
8348
8349                         /*
8350                          * If we aren't registered currently then increment
8351                          * the key count and set the registered flag.
8352                          */
8353                         ctl_alloc_prkey(lun, residx);
8354                         if (ctl_get_prkey(lun, residx) == 0)
8355                                 lun->pr_key_count++;
8356                         ctl_set_prkey(lun, residx, sa_res_key);
8357                         lun->pr_generation++;
8358                         mtx_unlock(&lun->lun_lock);
8359
8360                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8361                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8362                         persis_io.pr.pr_info.action = CTL_PR_REG_KEY;
8363                         persis_io.pr.pr_info.residx = residx;
8364                         memcpy(persis_io.pr.pr_info.sa_res_key,
8365                                param->serv_act_res_key,
8366                                sizeof(param->serv_act_res_key));
8367                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8368                             sizeof(persis_io.pr), M_WAITOK);
8369                 }
8370
8371                 break;
8372         }
8373         case SPRO_RESERVE:
8374 #if 0
8375                 printf("Reserve executed type %d\n", type);
8376 #endif
8377                 mtx_lock(&lun->lun_lock);
8378                 if (lun->flags & CTL_LUN_PR_RESERVED) {
8379                         /*
8380                          * if this isn't the reservation holder and it's
8381                          * not a "all registrants" type or if the type is
8382                          * different then we have a conflict
8383                          */
8384                         if ((lun->pr_res_idx != residx
8385                           && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8386                          || lun->pr_res_type != type) {
8387                                 mtx_unlock(&lun->lun_lock);
8388                                 free(ctsio->kern_data_ptr, M_CTL);
8389                                 ctl_set_reservation_conflict(ctsio);
8390                                 ctl_done((union ctl_io *)ctsio);
8391                                 return (CTL_RETVAL_COMPLETE);
8392                         }
8393                         mtx_unlock(&lun->lun_lock);
8394                 } else /* create a reservation */ {
8395                         /*
8396                          * If it's not an "all registrants" type record
8397                          * reservation holder
8398                          */
8399                         if (type != SPR_TYPE_WR_EX_AR
8400                          && type != SPR_TYPE_EX_AC_AR)
8401                                 lun->pr_res_idx = residx; /* Res holder */
8402                         else
8403                                 lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8404
8405                         lun->flags |= CTL_LUN_PR_RESERVED;
8406                         lun->pr_res_type = type;
8407
8408                         mtx_unlock(&lun->lun_lock);
8409
8410                         /* send msg to other side */
8411                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8412                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8413                         persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8414                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8415                         persis_io.pr.pr_info.res_type = type;
8416                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8417                             sizeof(persis_io.pr), M_WAITOK);
8418                 }
8419                 break;
8420
8421         case SPRO_RELEASE:
8422                 mtx_lock(&lun->lun_lock);
8423                 if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8424                         /* No reservation exists return good status */
8425                         mtx_unlock(&lun->lun_lock);
8426                         goto done;
8427                 }
8428                 /*
8429                  * Is this nexus a reservation holder?
8430                  */
8431                 if (lun->pr_res_idx != residx
8432                  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8433                         /*
8434                          * not a res holder return good status but
8435                          * do nothing
8436                          */
8437                         mtx_unlock(&lun->lun_lock);
8438                         goto done;
8439                 }
8440
8441                 if (lun->pr_res_type != type) {
8442                         mtx_unlock(&lun->lun_lock);
8443                         free(ctsio->kern_data_ptr, M_CTL);
8444                         ctl_set_illegal_pr_release(ctsio);
8445                         ctl_done((union ctl_io *)ctsio);
8446                         return (CTL_RETVAL_COMPLETE);
8447                 }
8448
8449                 /* okay to release */
8450                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8451                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8452                 lun->pr_res_type = 0;
8453
8454                 /*
8455                  * If this isn't an exclusive access reservation and NUAR
8456                  * is not set, generate UA for all other registrants.
8457                  */
8458                 if (type != SPR_TYPE_EX_AC && type != SPR_TYPE_WR_EX &&
8459                     (lun->MODE_CTRL.queue_flags & SCP_NUAR) == 0) {
8460                         for (i = softc->init_min; i < softc->init_max; i++) {
8461                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
8462                                         continue;
8463                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8464                         }
8465                 }
8466                 mtx_unlock(&lun->lun_lock);
8467
8468                 /* Send msg to other side */
8469                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8470                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8471                 persis_io.pr.pr_info.action = CTL_PR_RELEASE;
8472                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8473                      sizeof(persis_io.pr), M_WAITOK);
8474                 break;
8475
8476         case SPRO_CLEAR:
8477                 /* send msg to other side */
8478
8479                 mtx_lock(&lun->lun_lock);
8480                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8481                 lun->pr_res_type = 0;
8482                 lun->pr_key_count = 0;
8483                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8484
8485                 ctl_clr_prkey(lun, residx);
8486                 for (i = 0; i < CTL_MAX_INITIATORS; i++)
8487                         if (ctl_get_prkey(lun, i) != 0) {
8488                                 ctl_clr_prkey(lun, i);
8489                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8490                         }
8491                 lun->pr_generation++;
8492                 mtx_unlock(&lun->lun_lock);
8493
8494                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8495                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8496                 persis_io.pr.pr_info.action = CTL_PR_CLEAR;
8497                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8498                      sizeof(persis_io.pr), M_WAITOK);
8499                 break;
8500
8501         case SPRO_PREEMPT:
8502         case SPRO_PRE_ABO: {
8503                 int nretval;
8504
8505                 nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8506                                           residx, ctsio, cdb, param);
8507                 if (nretval != 0)
8508                         return (CTL_RETVAL_COMPLETE);
8509                 break;
8510         }
8511         default:
8512                 panic("%s: Invalid PR type %#x", __func__, cdb->action);
8513         }
8514
8515 done:
8516         free(ctsio->kern_data_ptr, M_CTL);
8517         ctl_set_success(ctsio);
8518         ctl_done((union ctl_io *)ctsio);
8519
8520         return (retval);
8521 }
8522
8523 /*
8524  * This routine is for handling a message from the other SC pertaining to
8525  * persistent reserve out. All the error checking will have been done
8526  * so only perorming the action need be done here to keep the two
8527  * in sync.
8528  */
8529 static void
8530 ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg)
8531 {
8532         struct ctl_softc *softc = control_softc;
8533         struct ctl_lun *lun;
8534         int i;
8535         uint32_t residx, targ_lun;
8536
8537         targ_lun = msg->hdr.nexus.targ_mapped_lun;
8538         mtx_lock(&softc->ctl_lock);
8539         if ((targ_lun >= CTL_MAX_LUNS) ||
8540             ((lun = softc->ctl_luns[targ_lun]) == NULL)) {
8541                 mtx_unlock(&softc->ctl_lock);
8542                 return;
8543         }
8544         mtx_lock(&lun->lun_lock);
8545         mtx_unlock(&softc->ctl_lock);
8546         if (lun->flags & CTL_LUN_DISABLED) {
8547                 mtx_unlock(&lun->lun_lock);
8548                 return;
8549         }
8550         residx = ctl_get_initindex(&msg->hdr.nexus);
8551         switch(msg->pr.pr_info.action) {
8552         case CTL_PR_REG_KEY:
8553                 ctl_alloc_prkey(lun, msg->pr.pr_info.residx);
8554                 if (ctl_get_prkey(lun, msg->pr.pr_info.residx) == 0)
8555                         lun->pr_key_count++;
8556                 ctl_set_prkey(lun, msg->pr.pr_info.residx,
8557                     scsi_8btou64(msg->pr.pr_info.sa_res_key));
8558                 lun->pr_generation++;
8559                 break;
8560
8561         case CTL_PR_UNREG_KEY:
8562                 ctl_clr_prkey(lun, msg->pr.pr_info.residx);
8563                 lun->pr_key_count--;
8564
8565                 /* XXX Need to see if the reservation has been released */
8566                 /* if so do we need to generate UA? */
8567                 if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8568                         lun->flags &= ~CTL_LUN_PR_RESERVED;
8569                         lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8570
8571                         if ((lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8572                              lun->pr_res_type == SPR_TYPE_EX_AC_RO) &&
8573                             lun->pr_key_count) {
8574                                 /*
8575                                  * If the reservation is a registrants
8576                                  * only type we need to generate a UA
8577                                  * for other registered inits.  The
8578                                  * sense code should be RESERVATIONS
8579                                  * RELEASED
8580                                  */
8581
8582                                 for (i = softc->init_min; i < softc->init_max; i++) {
8583                                         if (ctl_get_prkey(lun, i) == 0)
8584                                                 continue;
8585
8586                                         ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8587                                 }
8588                         }
8589                         lun->pr_res_type = 0;
8590                 } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8591                         if (lun->pr_key_count==0) {
8592                                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8593                                 lun->pr_res_type = 0;
8594                                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8595                         }
8596                 }
8597                 lun->pr_generation++;
8598                 break;
8599
8600         case CTL_PR_RESERVE:
8601                 lun->flags |= CTL_LUN_PR_RESERVED;
8602                 lun->pr_res_type = msg->pr.pr_info.res_type;
8603                 lun->pr_res_idx = msg->pr.pr_info.residx;
8604
8605                 break;
8606
8607         case CTL_PR_RELEASE:
8608                 /*
8609                  * If this isn't an exclusive access reservation and NUAR
8610                  * is not set, generate UA for all other registrants.
8611                  */
8612                 if (lun->pr_res_type != SPR_TYPE_EX_AC &&
8613                     lun->pr_res_type != SPR_TYPE_WR_EX &&
8614                     (lun->MODE_CTRL.queue_flags & SCP_NUAR) == 0) {
8615                         for (i = softc->init_min; i < softc->init_max; i++)
8616                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
8617                                         continue;
8618                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8619                 }
8620
8621                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8622                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8623                 lun->pr_res_type = 0;
8624                 break;
8625
8626         case CTL_PR_PREEMPT:
8627                 ctl_pro_preempt_other(lun, msg);
8628                 break;
8629         case CTL_PR_CLEAR:
8630                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8631                 lun->pr_res_type = 0;
8632                 lun->pr_key_count = 0;
8633                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8634
8635                 for (i=0; i < CTL_MAX_INITIATORS; i++) {
8636                         if (ctl_get_prkey(lun, i) == 0)
8637                                 continue;
8638                         ctl_clr_prkey(lun, i);
8639                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8640                 }
8641                 lun->pr_generation++;
8642                 break;
8643         }
8644
8645         mtx_unlock(&lun->lun_lock);
8646 }
8647
8648 int
8649 ctl_read_write(struct ctl_scsiio *ctsio)
8650 {
8651         struct ctl_lun *lun;
8652         struct ctl_lba_len_flags *lbalen;
8653         uint64_t lba;
8654         uint32_t num_blocks;
8655         int flags, retval;
8656         int isread;
8657
8658         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8659
8660         CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
8661
8662         flags = 0;
8663         isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
8664               || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
8665         switch (ctsio->cdb[0]) {
8666         case READ_6:
8667         case WRITE_6: {
8668                 struct scsi_rw_6 *cdb;
8669
8670                 cdb = (struct scsi_rw_6 *)ctsio->cdb;
8671
8672                 lba = scsi_3btoul(cdb->addr);
8673                 /* only 5 bits are valid in the most significant address byte */
8674                 lba &= 0x1fffff;
8675                 num_blocks = cdb->length;
8676                 /*
8677                  * This is correct according to SBC-2.
8678                  */
8679                 if (num_blocks == 0)
8680                         num_blocks = 256;
8681                 break;
8682         }
8683         case READ_10:
8684         case WRITE_10: {
8685                 struct scsi_rw_10 *cdb;
8686
8687                 cdb = (struct scsi_rw_10 *)ctsio->cdb;
8688                 if (cdb->byte2 & SRW10_FUA)
8689                         flags |= CTL_LLF_FUA;
8690                 if (cdb->byte2 & SRW10_DPO)
8691                         flags |= CTL_LLF_DPO;
8692                 lba = scsi_4btoul(cdb->addr);
8693                 num_blocks = scsi_2btoul(cdb->length);
8694                 break;
8695         }
8696         case WRITE_VERIFY_10: {
8697                 struct scsi_write_verify_10 *cdb;
8698
8699                 cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
8700                 flags |= CTL_LLF_FUA;
8701                 if (cdb->byte2 & SWV_DPO)
8702                         flags |= CTL_LLF_DPO;
8703                 lba = scsi_4btoul(cdb->addr);
8704                 num_blocks = scsi_2btoul(cdb->length);
8705                 break;
8706         }
8707         case READ_12:
8708         case WRITE_12: {
8709                 struct scsi_rw_12 *cdb;
8710
8711                 cdb = (struct scsi_rw_12 *)ctsio->cdb;
8712                 if (cdb->byte2 & SRW12_FUA)
8713                         flags |= CTL_LLF_FUA;
8714                 if (cdb->byte2 & SRW12_DPO)
8715                         flags |= CTL_LLF_DPO;
8716                 lba = scsi_4btoul(cdb->addr);
8717                 num_blocks = scsi_4btoul(cdb->length);
8718                 break;
8719         }
8720         case WRITE_VERIFY_12: {
8721                 struct scsi_write_verify_12 *cdb;
8722
8723                 cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
8724                 flags |= CTL_LLF_FUA;
8725                 if (cdb->byte2 & SWV_DPO)
8726                         flags |= CTL_LLF_DPO;
8727                 lba = scsi_4btoul(cdb->addr);
8728                 num_blocks = scsi_4btoul(cdb->length);
8729                 break;
8730         }
8731         case READ_16:
8732         case WRITE_16: {
8733                 struct scsi_rw_16 *cdb;
8734
8735                 cdb = (struct scsi_rw_16 *)ctsio->cdb;
8736                 if (cdb->byte2 & SRW12_FUA)
8737                         flags |= CTL_LLF_FUA;
8738                 if (cdb->byte2 & SRW12_DPO)
8739                         flags |= CTL_LLF_DPO;
8740                 lba = scsi_8btou64(cdb->addr);
8741                 num_blocks = scsi_4btoul(cdb->length);
8742                 break;
8743         }
8744         case WRITE_ATOMIC_16: {
8745                 struct scsi_write_atomic_16 *cdb;
8746
8747                 if (lun->be_lun->atomicblock == 0) {
8748                         ctl_set_invalid_opcode(ctsio);
8749                         ctl_done((union ctl_io *)ctsio);
8750                         return (CTL_RETVAL_COMPLETE);
8751                 }
8752
8753                 cdb = (struct scsi_write_atomic_16 *)ctsio->cdb;
8754                 if (cdb->byte2 & SRW12_FUA)
8755                         flags |= CTL_LLF_FUA;
8756                 if (cdb->byte2 & SRW12_DPO)
8757                         flags |= CTL_LLF_DPO;
8758                 lba = scsi_8btou64(cdb->addr);
8759                 num_blocks = scsi_2btoul(cdb->length);
8760                 if (num_blocks > lun->be_lun->atomicblock) {
8761                         ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
8762                             /*command*/ 1, /*field*/ 12, /*bit_valid*/ 0,
8763                             /*bit*/ 0);
8764                         ctl_done((union ctl_io *)ctsio);
8765                         return (CTL_RETVAL_COMPLETE);
8766                 }
8767                 break;
8768         }
8769         case WRITE_VERIFY_16: {
8770                 struct scsi_write_verify_16 *cdb;
8771
8772                 cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
8773                 flags |= CTL_LLF_FUA;
8774                 if (cdb->byte2 & SWV_DPO)
8775                         flags |= CTL_LLF_DPO;
8776                 lba = scsi_8btou64(cdb->addr);
8777                 num_blocks = scsi_4btoul(cdb->length);
8778                 break;
8779         }
8780         default:
8781                 /*
8782                  * We got a command we don't support.  This shouldn't
8783                  * happen, commands should be filtered out above us.
8784                  */
8785                 ctl_set_invalid_opcode(ctsio);
8786                 ctl_done((union ctl_io *)ctsio);
8787
8788                 return (CTL_RETVAL_COMPLETE);
8789                 break; /* NOTREACHED */
8790         }
8791
8792         /*
8793          * The first check is to make sure we're in bounds, the second
8794          * check is to catch wrap-around problems.  If the lba + num blocks
8795          * is less than the lba, then we've wrapped around and the block
8796          * range is invalid anyway.
8797          */
8798         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8799          || ((lba + num_blocks) < lba)) {
8800                 ctl_set_lba_out_of_range(ctsio,
8801                     MAX(lba, lun->be_lun->maxlba + 1));
8802                 ctl_done((union ctl_io *)ctsio);
8803                 return (CTL_RETVAL_COMPLETE);
8804         }
8805
8806         /*
8807          * According to SBC-3, a transfer length of 0 is not an error.
8808          * Note that this cannot happen with WRITE(6) or READ(6), since 0
8809          * translates to 256 blocks for those commands.
8810          */
8811         if (num_blocks == 0) {
8812                 ctl_set_success(ctsio);
8813                 ctl_done((union ctl_io *)ctsio);
8814                 return (CTL_RETVAL_COMPLETE);
8815         }
8816
8817         /* Set FUA and/or DPO if caches are disabled. */
8818         if (isread) {
8819                 if ((lun->MODE_CACHING.flags1 & SCP_RCD) != 0)
8820                         flags |= CTL_LLF_FUA | CTL_LLF_DPO;
8821         } else {
8822                 if ((lun->MODE_CACHING.flags1 & SCP_WCE) == 0)
8823                         flags |= CTL_LLF_FUA;
8824         }
8825
8826         lbalen = (struct ctl_lba_len_flags *)
8827             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8828         lbalen->lba = lba;
8829         lbalen->len = num_blocks;
8830         lbalen->flags = (isread ? CTL_LLF_READ : CTL_LLF_WRITE) | flags;
8831
8832         ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
8833         ctsio->kern_rel_offset = 0;
8834
8835         CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
8836
8837         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8838         return (retval);
8839 }
8840
8841 static int
8842 ctl_cnw_cont(union ctl_io *io)
8843 {
8844         struct ctl_scsiio *ctsio;
8845         struct ctl_lun *lun;
8846         struct ctl_lba_len_flags *lbalen;
8847         int retval;
8848
8849         ctsio = &io->scsiio;
8850         ctsio->io_hdr.status = CTL_STATUS_NONE;
8851         ctsio->io_hdr.flags &= ~CTL_FLAG_IO_CONT;
8852         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8853         lbalen = (struct ctl_lba_len_flags *)
8854             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8855         lbalen->flags &= ~CTL_LLF_COMPARE;
8856         lbalen->flags |= CTL_LLF_WRITE;
8857
8858         CTL_DEBUG_PRINT(("ctl_cnw_cont: calling data_submit()\n"));
8859         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8860         return (retval);
8861 }
8862
8863 int
8864 ctl_cnw(struct ctl_scsiio *ctsio)
8865 {
8866         struct ctl_lun *lun;
8867         struct ctl_lba_len_flags *lbalen;
8868         uint64_t lba;
8869         uint32_t num_blocks;
8870         int flags, retval;
8871
8872         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8873
8874         CTL_DEBUG_PRINT(("ctl_cnw: command: %#x\n", ctsio->cdb[0]));
8875
8876         flags = 0;
8877         switch (ctsio->cdb[0]) {
8878         case COMPARE_AND_WRITE: {
8879                 struct scsi_compare_and_write *cdb;
8880
8881                 cdb = (struct scsi_compare_and_write *)ctsio->cdb;
8882                 if (cdb->byte2 & SRW10_FUA)
8883                         flags |= CTL_LLF_FUA;
8884                 if (cdb->byte2 & SRW10_DPO)
8885                         flags |= CTL_LLF_DPO;
8886                 lba = scsi_8btou64(cdb->addr);
8887                 num_blocks = cdb->length;
8888                 break;
8889         }
8890         default:
8891                 /*
8892                  * We got a command we don't support.  This shouldn't
8893                  * happen, commands should be filtered out above us.
8894                  */
8895                 ctl_set_invalid_opcode(ctsio);
8896                 ctl_done((union ctl_io *)ctsio);
8897
8898                 return (CTL_RETVAL_COMPLETE);
8899                 break; /* NOTREACHED */
8900         }
8901
8902         /*
8903          * The first check is to make sure we're in bounds, the second
8904          * check is to catch wrap-around problems.  If the lba + num blocks
8905          * is less than the lba, then we've wrapped around and the block
8906          * range is invalid anyway.
8907          */
8908         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8909          || ((lba + num_blocks) < lba)) {
8910                 ctl_set_lba_out_of_range(ctsio,
8911                     MAX(lba, lun->be_lun->maxlba + 1));
8912                 ctl_done((union ctl_io *)ctsio);
8913                 return (CTL_RETVAL_COMPLETE);
8914         }
8915
8916         /*
8917          * According to SBC-3, a transfer length of 0 is not an error.
8918          */
8919         if (num_blocks == 0) {
8920                 ctl_set_success(ctsio);
8921                 ctl_done((union ctl_io *)ctsio);
8922                 return (CTL_RETVAL_COMPLETE);
8923         }
8924
8925         /* Set FUA if write cache is disabled. */
8926         if ((lun->MODE_CACHING.flags1 & SCP_WCE) == 0)
8927                 flags |= CTL_LLF_FUA;
8928
8929         ctsio->kern_total_len = 2 * num_blocks * lun->be_lun->blocksize;
8930         ctsio->kern_rel_offset = 0;
8931
8932         /*
8933          * Set the IO_CONT flag, so that if this I/O gets passed to
8934          * ctl_data_submit_done(), it'll get passed back to
8935          * ctl_ctl_cnw_cont() for further processing.
8936          */
8937         ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
8938         ctsio->io_cont = ctl_cnw_cont;
8939
8940         lbalen = (struct ctl_lba_len_flags *)
8941             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8942         lbalen->lba = lba;
8943         lbalen->len = num_blocks;
8944         lbalen->flags = CTL_LLF_COMPARE | flags;
8945
8946         CTL_DEBUG_PRINT(("ctl_cnw: calling data_submit()\n"));
8947         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8948         return (retval);
8949 }
8950
8951 int
8952 ctl_verify(struct ctl_scsiio *ctsio)
8953 {
8954         struct ctl_lun *lun;
8955         struct ctl_lba_len_flags *lbalen;
8956         uint64_t lba;
8957         uint32_t num_blocks;
8958         int bytchk, flags;
8959         int retval;
8960
8961         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8962
8963         CTL_DEBUG_PRINT(("ctl_verify: command: %#x\n", ctsio->cdb[0]));
8964
8965         bytchk = 0;
8966         flags = CTL_LLF_FUA;
8967         switch (ctsio->cdb[0]) {
8968         case VERIFY_10: {
8969                 struct scsi_verify_10 *cdb;
8970
8971                 cdb = (struct scsi_verify_10 *)ctsio->cdb;
8972                 if (cdb->byte2 & SVFY_BYTCHK)
8973                         bytchk = 1;
8974                 if (cdb->byte2 & SVFY_DPO)
8975                         flags |= CTL_LLF_DPO;
8976                 lba = scsi_4btoul(cdb->addr);
8977                 num_blocks = scsi_2btoul(cdb->length);
8978                 break;
8979         }
8980         case VERIFY_12: {
8981                 struct scsi_verify_12 *cdb;
8982
8983                 cdb = (struct scsi_verify_12 *)ctsio->cdb;
8984                 if (cdb->byte2 & SVFY_BYTCHK)
8985                         bytchk = 1;
8986                 if (cdb->byte2 & SVFY_DPO)
8987                         flags |= CTL_LLF_DPO;
8988                 lba = scsi_4btoul(cdb->addr);
8989                 num_blocks = scsi_4btoul(cdb->length);
8990                 break;
8991         }
8992         case VERIFY_16: {
8993                 struct scsi_rw_16 *cdb;
8994
8995                 cdb = (struct scsi_rw_16 *)ctsio->cdb;
8996                 if (cdb->byte2 & SVFY_BYTCHK)
8997                         bytchk = 1;
8998                 if (cdb->byte2 & SVFY_DPO)
8999                         flags |= CTL_LLF_DPO;
9000                 lba = scsi_8btou64(cdb->addr);
9001                 num_blocks = scsi_4btoul(cdb->length);
9002                 break;
9003         }
9004         default:
9005                 /*
9006                  * We got a command we don't support.  This shouldn't
9007                  * happen, commands should be filtered out above us.
9008                  */
9009                 ctl_set_invalid_opcode(ctsio);
9010                 ctl_done((union ctl_io *)ctsio);
9011                 return (CTL_RETVAL_COMPLETE);
9012         }
9013
9014         /*
9015          * The first check is to make sure we're in bounds, the second
9016          * check is to catch wrap-around problems.  If the lba + num blocks
9017          * is less than the lba, then we've wrapped around and the block
9018          * range is invalid anyway.
9019          */
9020         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9021          || ((lba + num_blocks) < lba)) {
9022                 ctl_set_lba_out_of_range(ctsio,
9023                     MAX(lba, lun->be_lun->maxlba + 1));
9024                 ctl_done((union ctl_io *)ctsio);
9025                 return (CTL_RETVAL_COMPLETE);
9026         }
9027
9028         /*
9029          * According to SBC-3, a transfer length of 0 is not an error.
9030          */
9031         if (num_blocks == 0) {
9032                 ctl_set_success(ctsio);
9033                 ctl_done((union ctl_io *)ctsio);
9034                 return (CTL_RETVAL_COMPLETE);
9035         }
9036
9037         lbalen = (struct ctl_lba_len_flags *)
9038             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9039         lbalen->lba = lba;
9040         lbalen->len = num_blocks;
9041         if (bytchk) {
9042                 lbalen->flags = CTL_LLF_COMPARE | flags;
9043                 ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9044         } else {
9045                 lbalen->flags = CTL_LLF_VERIFY | flags;
9046                 ctsio->kern_total_len = 0;
9047         }
9048         ctsio->kern_rel_offset = 0;
9049
9050         CTL_DEBUG_PRINT(("ctl_verify: calling data_submit()\n"));
9051         retval = lun->backend->data_submit((union ctl_io *)ctsio);
9052         return (retval);
9053 }
9054
9055 int
9056 ctl_report_luns(struct ctl_scsiio *ctsio)
9057 {
9058         struct ctl_softc *softc;
9059         struct scsi_report_luns *cdb;
9060         struct scsi_report_luns_data *lun_data;
9061         struct ctl_lun *lun, *request_lun;
9062         struct ctl_port *port;
9063         int num_luns, retval;
9064         uint32_t alloc_len, lun_datalen;
9065         int num_filled;
9066         uint32_t initidx, targ_lun_id, lun_id;
9067
9068         retval = CTL_RETVAL_COMPLETE;
9069         cdb = (struct scsi_report_luns *)ctsio->cdb;
9070         port = ctl_io_port(&ctsio->io_hdr);
9071         softc = port->ctl_softc;
9072
9073         CTL_DEBUG_PRINT(("ctl_report_luns\n"));
9074
9075         mtx_lock(&softc->ctl_lock);
9076         num_luns = 0;
9077         for (targ_lun_id = 0; targ_lun_id < CTL_MAX_LUNS; targ_lun_id++) {
9078                 if (ctl_lun_map_from_port(port, targ_lun_id) < CTL_MAX_LUNS)
9079                         num_luns++;
9080         }
9081         mtx_unlock(&softc->ctl_lock);
9082
9083         switch (cdb->select_report) {
9084         case RPL_REPORT_DEFAULT:
9085         case RPL_REPORT_ALL:
9086         case RPL_REPORT_NONSUBSID:
9087                 break;
9088         case RPL_REPORT_WELLKNOWN:
9089         case RPL_REPORT_ADMIN:
9090         case RPL_REPORT_CONGLOM:
9091                 num_luns = 0;
9092                 break;
9093         default:
9094                 ctl_set_invalid_field(ctsio,
9095                                       /*sks_valid*/ 1,
9096                                       /*command*/ 1,
9097                                       /*field*/ 2,
9098                                       /*bit_valid*/ 0,
9099                                       /*bit*/ 0);
9100                 ctl_done((union ctl_io *)ctsio);
9101                 return (retval);
9102                 break; /* NOTREACHED */
9103         }
9104
9105         alloc_len = scsi_4btoul(cdb->length);
9106         /*
9107          * The initiator has to allocate at least 16 bytes for this request,
9108          * so he can at least get the header and the first LUN.  Otherwise
9109          * we reject the request (per SPC-3 rev 14, section 6.21).
9110          */
9111         if (alloc_len < (sizeof(struct scsi_report_luns_data) +
9112             sizeof(struct scsi_report_luns_lundata))) {
9113                 ctl_set_invalid_field(ctsio,
9114                                       /*sks_valid*/ 1,
9115                                       /*command*/ 1,
9116                                       /*field*/ 6,
9117                                       /*bit_valid*/ 0,
9118                                       /*bit*/ 0);
9119                 ctl_done((union ctl_io *)ctsio);
9120                 return (retval);
9121         }
9122
9123         request_lun = (struct ctl_lun *)
9124                 ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9125
9126         lun_datalen = sizeof(*lun_data) +
9127                 (num_luns * sizeof(struct scsi_report_luns_lundata));
9128
9129         ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO);
9130         lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
9131         ctsio->kern_sg_entries = 0;
9132
9133         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9134
9135         mtx_lock(&softc->ctl_lock);
9136         for (targ_lun_id = 0, num_filled = 0; targ_lun_id < CTL_MAX_LUNS && num_filled < num_luns; targ_lun_id++) {
9137                 lun_id = ctl_lun_map_from_port(port, targ_lun_id);
9138                 if (lun_id >= CTL_MAX_LUNS)
9139                         continue;
9140                 lun = softc->ctl_luns[lun_id];
9141                 if (lun == NULL)
9142                         continue;
9143
9144                 be64enc(lun_data->luns[num_filled++].lundata,
9145                     ctl_encode_lun(targ_lun_id));
9146
9147                 /*
9148                  * According to SPC-3, rev 14 section 6.21:
9149                  *
9150                  * "The execution of a REPORT LUNS command to any valid and
9151                  * installed logical unit shall clear the REPORTED LUNS DATA
9152                  * HAS CHANGED unit attention condition for all logical
9153                  * units of that target with respect to the requesting
9154                  * initiator. A valid and installed logical unit is one
9155                  * having a PERIPHERAL QUALIFIER of 000b in the standard
9156                  * INQUIRY data (see 6.4.2)."
9157                  *
9158                  * If request_lun is NULL, the LUN this report luns command
9159                  * was issued to is either disabled or doesn't exist. In that
9160                  * case, we shouldn't clear any pending lun change unit
9161                  * attention.
9162                  */
9163                 if (request_lun != NULL) {
9164                         mtx_lock(&lun->lun_lock);
9165                         ctl_clr_ua(lun, initidx, CTL_UA_LUN_CHANGE);
9166                         mtx_unlock(&lun->lun_lock);
9167                 }
9168         }
9169         mtx_unlock(&softc->ctl_lock);
9170
9171         /*
9172          * It's quite possible that we've returned fewer LUNs than we allocated
9173          * space for.  Trim it.
9174          */
9175         lun_datalen = sizeof(*lun_data) +
9176                 (num_filled * sizeof(struct scsi_report_luns_lundata));
9177
9178         if (lun_datalen < alloc_len) {
9179                 ctsio->residual = alloc_len - lun_datalen;
9180                 ctsio->kern_data_len = lun_datalen;
9181                 ctsio->kern_total_len = lun_datalen;
9182         } else {
9183                 ctsio->residual = 0;
9184                 ctsio->kern_data_len = alloc_len;
9185                 ctsio->kern_total_len = alloc_len;
9186         }
9187         ctsio->kern_data_resid = 0;
9188         ctsio->kern_rel_offset = 0;
9189         ctsio->kern_sg_entries = 0;
9190
9191         /*
9192          * We set this to the actual data length, regardless of how much
9193          * space we actually have to return results.  If the user looks at
9194          * this value, he'll know whether or not he allocated enough space
9195          * and reissue the command if necessary.  We don't support well
9196          * known logical units, so if the user asks for that, return none.
9197          */
9198         scsi_ulto4b(lun_datalen - 8, lun_data->length);
9199
9200         /*
9201          * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
9202          * this request.
9203          */
9204         ctl_set_success(ctsio);
9205         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9206         ctsio->be_move_done = ctl_config_move_done;
9207         ctl_datamove((union ctl_io *)ctsio);
9208         return (retval);
9209 }
9210
9211 int
9212 ctl_request_sense(struct ctl_scsiio *ctsio)
9213 {
9214         struct scsi_request_sense *cdb;
9215         struct scsi_sense_data *sense_ptr;
9216         struct ctl_softc *softc;
9217         struct ctl_lun *lun;
9218         uint32_t initidx;
9219         int have_error;
9220         scsi_sense_data_type sense_format;
9221         ctl_ua_type ua_type;
9222         uint8_t asc = 0, ascq = 0;
9223
9224         cdb = (struct scsi_request_sense *)ctsio->cdb;
9225
9226         softc = control_softc;
9227         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9228
9229         CTL_DEBUG_PRINT(("ctl_request_sense\n"));
9230
9231         /*
9232          * Determine which sense format the user wants.
9233          */
9234         if (cdb->byte2 & SRS_DESC)
9235                 sense_format = SSD_TYPE_DESC;
9236         else
9237                 sense_format = SSD_TYPE_FIXED;
9238
9239         ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
9240         sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
9241         ctsio->kern_sg_entries = 0;
9242
9243         /*
9244          * struct scsi_sense_data, which is currently set to 256 bytes, is
9245          * larger than the largest allowed value for the length field in the
9246          * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
9247          */
9248         ctsio->residual = 0;
9249         ctsio->kern_data_len = cdb->length;
9250         ctsio->kern_total_len = cdb->length;
9251
9252         ctsio->kern_data_resid = 0;
9253         ctsio->kern_rel_offset = 0;
9254         ctsio->kern_sg_entries = 0;
9255
9256         /*
9257          * If we don't have a LUN, we don't have any pending sense.
9258          */
9259         if (lun == NULL ||
9260             ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
9261              softc->ha_link < CTL_HA_LINK_UNKNOWN)) {
9262                 /* "Logical unit not supported" */
9263                 ctl_set_sense_data(sense_ptr, NULL, sense_format,
9264                     /*current_error*/ 1,
9265                     /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
9266                     /*asc*/ 0x25,
9267                     /*ascq*/ 0x00,
9268                     SSD_ELEM_NONE);
9269                 goto send;
9270         }
9271
9272         have_error = 0;
9273         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9274         /*
9275          * Check for pending sense, and then for pending unit attentions.
9276          * Pending sense gets returned first, then pending unit attentions.
9277          */
9278         mtx_lock(&lun->lun_lock);
9279 #ifdef CTL_WITH_CA
9280         if (ctl_is_set(lun->have_ca, initidx)) {
9281                 scsi_sense_data_type stored_format;
9282
9283                 /*
9284                  * Check to see which sense format was used for the stored
9285                  * sense data.
9286                  */
9287                 stored_format = scsi_sense_type(&lun->pending_sense[initidx]);
9288
9289                 /*
9290                  * If the user requested a different sense format than the
9291                  * one we stored, then we need to convert it to the other
9292                  * format.  If we're going from descriptor to fixed format
9293                  * sense data, we may lose things in translation, depending
9294                  * on what options were used.
9295                  *
9296                  * If the stored format is SSD_TYPE_NONE (i.e. invalid),
9297                  * for some reason we'll just copy it out as-is.
9298                  */
9299                 if ((stored_format == SSD_TYPE_FIXED)
9300                  && (sense_format == SSD_TYPE_DESC))
9301                         ctl_sense_to_desc((struct scsi_sense_data_fixed *)
9302                             &lun->pending_sense[initidx],
9303                             (struct scsi_sense_data_desc *)sense_ptr);
9304                 else if ((stored_format == SSD_TYPE_DESC)
9305                       && (sense_format == SSD_TYPE_FIXED))
9306                         ctl_sense_to_fixed((struct scsi_sense_data_desc *)
9307                             &lun->pending_sense[initidx],
9308                             (struct scsi_sense_data_fixed *)sense_ptr);
9309                 else
9310                         memcpy(sense_ptr, &lun->pending_sense[initidx],
9311                                MIN(sizeof(*sense_ptr),
9312                                sizeof(lun->pending_sense[initidx])));
9313
9314                 ctl_clear_mask(lun->have_ca, initidx);
9315                 have_error = 1;
9316         } else
9317 #endif
9318         if (have_error == 0) {
9319                 ua_type = ctl_build_ua(lun, initidx, sense_ptr, sense_format);
9320                 if (ua_type != CTL_UA_NONE)
9321                         have_error = 1;
9322         }
9323         if (have_error == 0) {
9324                 /*
9325                  * Report informational exception if have one and allowed.
9326                  */
9327                 if (lun->MODE_IE.mrie != SIEP_MRIE_NO) {
9328                         asc = lun->ie_asc;
9329                         ascq = lun->ie_ascq;
9330                 }
9331                 ctl_set_sense_data(sense_ptr, lun, sense_format,
9332                     /*current_error*/ 1,
9333                     /*sense_key*/ SSD_KEY_NO_SENSE,
9334                     /*asc*/ asc,
9335                     /*ascq*/ ascq,
9336                     SSD_ELEM_NONE);
9337         }
9338         mtx_unlock(&lun->lun_lock);
9339
9340 send:
9341         /*
9342          * We report the SCSI status as OK, since the status of the command
9343          * itself is OK.  We're reporting sense as parameter data.
9344          */
9345         ctl_set_success(ctsio);
9346         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9347         ctsio->be_move_done = ctl_config_move_done;
9348         ctl_datamove((union ctl_io *)ctsio);
9349         return (CTL_RETVAL_COMPLETE);
9350 }
9351
9352 int
9353 ctl_tur(struct ctl_scsiio *ctsio)
9354 {
9355
9356         CTL_DEBUG_PRINT(("ctl_tur\n"));
9357
9358         ctl_set_success(ctsio);
9359         ctl_done((union ctl_io *)ctsio);
9360
9361         return (CTL_RETVAL_COMPLETE);
9362 }
9363
9364 /*
9365  * SCSI VPD page 0x00, the Supported VPD Pages page.
9366  */
9367 static int
9368 ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
9369 {
9370         struct scsi_vpd_supported_pages *pages;
9371         int sup_page_size;
9372         struct ctl_lun *lun;
9373         int p;
9374
9375         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9376
9377         sup_page_size = sizeof(struct scsi_vpd_supported_pages) *
9378             SCSI_EVPD_NUM_SUPPORTED_PAGES;
9379         ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO);
9380         pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
9381         ctsio->kern_sg_entries = 0;
9382
9383         if (sup_page_size < alloc_len) {
9384                 ctsio->residual = alloc_len - sup_page_size;
9385                 ctsio->kern_data_len = sup_page_size;
9386                 ctsio->kern_total_len = sup_page_size;
9387         } else {
9388                 ctsio->residual = 0;
9389                 ctsio->kern_data_len = alloc_len;
9390                 ctsio->kern_total_len = alloc_len;
9391         }
9392         ctsio->kern_data_resid = 0;
9393         ctsio->kern_rel_offset = 0;
9394         ctsio->kern_sg_entries = 0;
9395
9396         /*
9397          * The control device is always connected.  The disk device, on the
9398          * other hand, may not be online all the time.  Need to change this
9399          * to figure out whether the disk device is actually online or not.
9400          */
9401         if (lun != NULL)
9402                 pages->device = (SID_QUAL_LU_CONNECTED << 5) |
9403                                 lun->be_lun->lun_type;
9404         else
9405                 pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9406
9407         p = 0;
9408         /* Supported VPD pages */
9409         pages->page_list[p++] = SVPD_SUPPORTED_PAGES;
9410         /* Serial Number */
9411         pages->page_list[p++] = SVPD_UNIT_SERIAL_NUMBER;
9412         /* Device Identification */
9413         pages->page_list[p++] = SVPD_DEVICE_ID;
9414         /* Extended INQUIRY Data */
9415         pages->page_list[p++] = SVPD_EXTENDED_INQUIRY_DATA;
9416         /* Mode Page Policy */
9417         pages->page_list[p++] = SVPD_MODE_PAGE_POLICY;
9418         /* SCSI Ports */
9419         pages->page_list[p++] = SVPD_SCSI_PORTS;
9420         /* Third-party Copy */
9421         pages->page_list[p++] = SVPD_SCSI_TPC;
9422         if (lun != NULL && lun->be_lun->lun_type == T_DIRECT) {
9423                 /* Block limits */
9424                 pages->page_list[p++] = SVPD_BLOCK_LIMITS;
9425                 /* Block Device Characteristics */
9426                 pages->page_list[p++] = SVPD_BDC;
9427                 /* Logical Block Provisioning */
9428                 pages->page_list[p++] = SVPD_LBP;
9429         }
9430         pages->length = p;
9431
9432         ctl_set_success(ctsio);
9433         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9434         ctsio->be_move_done = ctl_config_move_done;
9435         ctl_datamove((union ctl_io *)ctsio);
9436         return (CTL_RETVAL_COMPLETE);
9437 }
9438
9439 /*
9440  * SCSI VPD page 0x80, the Unit Serial Number page.
9441  */
9442 static int
9443 ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
9444 {
9445         struct scsi_vpd_unit_serial_number *sn_ptr;
9446         struct ctl_lun *lun;
9447         int data_len;
9448
9449         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9450
9451         data_len = 4 + CTL_SN_LEN;
9452         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9453         sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
9454         if (data_len < alloc_len) {
9455                 ctsio->residual = alloc_len - data_len;
9456                 ctsio->kern_data_len = data_len;
9457                 ctsio->kern_total_len = data_len;
9458         } else {
9459                 ctsio->residual = 0;
9460                 ctsio->kern_data_len = alloc_len;
9461                 ctsio->kern_total_len = alloc_len;
9462         }
9463         ctsio->kern_data_resid = 0;
9464         ctsio->kern_rel_offset = 0;
9465         ctsio->kern_sg_entries = 0;
9466
9467         /*
9468          * The control device is always connected.  The disk device, on the
9469          * other hand, may not be online all the time.  Need to change this
9470          * to figure out whether the disk device is actually online or not.
9471          */
9472         if (lun != NULL)
9473                 sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9474                                   lun->be_lun->lun_type;
9475         else
9476                 sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9477
9478         sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9479         sn_ptr->length = CTL_SN_LEN;
9480         /*
9481          * If we don't have a LUN, we just leave the serial number as
9482          * all spaces.
9483          */
9484         if (lun != NULL) {
9485                 strncpy((char *)sn_ptr->serial_num,
9486                         (char *)lun->be_lun->serial_num, CTL_SN_LEN);
9487         } else
9488                 memset(sn_ptr->serial_num, 0x20, CTL_SN_LEN);
9489
9490         ctl_set_success(ctsio);
9491         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9492         ctsio->be_move_done = ctl_config_move_done;
9493         ctl_datamove((union ctl_io *)ctsio);
9494         return (CTL_RETVAL_COMPLETE);
9495 }
9496
9497
9498 /*
9499  * SCSI VPD page 0x86, the Extended INQUIRY Data page.
9500  */
9501 static int
9502 ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len)
9503 {
9504         struct scsi_vpd_extended_inquiry_data *eid_ptr;
9505         struct ctl_lun *lun;
9506         int data_len;
9507
9508         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9509
9510         data_len = sizeof(struct scsi_vpd_extended_inquiry_data);
9511         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9512         eid_ptr = (struct scsi_vpd_extended_inquiry_data *)ctsio->kern_data_ptr;
9513         ctsio->kern_sg_entries = 0;
9514
9515         if (data_len < alloc_len) {
9516                 ctsio->residual = alloc_len - data_len;
9517                 ctsio->kern_data_len = data_len;
9518                 ctsio->kern_total_len = data_len;
9519         } else {
9520                 ctsio->residual = 0;
9521                 ctsio->kern_data_len = alloc_len;
9522                 ctsio->kern_total_len = alloc_len;
9523         }
9524         ctsio->kern_data_resid = 0;
9525         ctsio->kern_rel_offset = 0;
9526         ctsio->kern_sg_entries = 0;
9527
9528         /*
9529          * The control device is always connected.  The disk device, on the
9530          * other hand, may not be online all the time.
9531          */
9532         if (lun != NULL)
9533                 eid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9534                                      lun->be_lun->lun_type;
9535         else
9536                 eid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9537         eid_ptr->page_code = SVPD_EXTENDED_INQUIRY_DATA;
9538         scsi_ulto2b(data_len - 4, eid_ptr->page_length);
9539         /*
9540          * We support head of queue, ordered and simple tags.
9541          */
9542         eid_ptr->flags2 = SVPD_EID_HEADSUP | SVPD_EID_ORDSUP | SVPD_EID_SIMPSUP;
9543         /*
9544          * Volatile cache supported.
9545          */
9546         eid_ptr->flags3 = SVPD_EID_V_SUP;
9547
9548         /*
9549          * This means that we clear the REPORTED LUNS DATA HAS CHANGED unit
9550          * attention for a particular IT nexus on all LUNs once we report
9551          * it to that nexus once.  This bit is required as of SPC-4.
9552          */
9553         eid_ptr->flags4 = SVPD_EID_LUICLR;
9554
9555         /*
9556          * XXX KDM in order to correctly answer this, we would need
9557          * information from the SIM to determine how much sense data it
9558          * can send.  So this would really be a path inquiry field, most
9559          * likely.  This can be set to a maximum of 252 according to SPC-4,
9560          * but the hardware may or may not be able to support that much.
9561          * 0 just means that the maximum sense data length is not reported.
9562          */
9563         eid_ptr->max_sense_length = 0;
9564
9565         ctl_set_success(ctsio);
9566         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9567         ctsio->be_move_done = ctl_config_move_done;
9568         ctl_datamove((union ctl_io *)ctsio);
9569         return (CTL_RETVAL_COMPLETE);
9570 }
9571
9572 static int
9573 ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len)
9574 {
9575         struct scsi_vpd_mode_page_policy *mpp_ptr;
9576         struct ctl_lun *lun;
9577         int data_len;
9578
9579         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9580
9581         data_len = sizeof(struct scsi_vpd_mode_page_policy) +
9582             sizeof(struct scsi_vpd_mode_page_policy_descr);
9583
9584         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9585         mpp_ptr = (struct scsi_vpd_mode_page_policy *)ctsio->kern_data_ptr;
9586         ctsio->kern_sg_entries = 0;
9587
9588         if (data_len < alloc_len) {
9589                 ctsio->residual = alloc_len - data_len;
9590                 ctsio->kern_data_len = data_len;
9591                 ctsio->kern_total_len = data_len;
9592         } else {
9593                 ctsio->residual = 0;
9594                 ctsio->kern_data_len = alloc_len;
9595                 ctsio->kern_total_len = alloc_len;
9596         }
9597         ctsio->kern_data_resid = 0;
9598         ctsio->kern_rel_offset = 0;
9599         ctsio->kern_sg_entries = 0;
9600
9601         /*
9602          * The control device is always connected.  The disk device, on the
9603          * other hand, may not be online all the time.
9604          */
9605         if (lun != NULL)
9606                 mpp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9607                                      lun->be_lun->lun_type;
9608         else
9609                 mpp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9610         mpp_ptr->page_code = SVPD_MODE_PAGE_POLICY;
9611         scsi_ulto2b(data_len - 4, mpp_ptr->page_length);
9612         mpp_ptr->descr[0].page_code = 0x3f;
9613         mpp_ptr->descr[0].subpage_code = 0xff;
9614         mpp_ptr->descr[0].policy = SVPD_MPP_SHARED;
9615
9616         ctl_set_success(ctsio);
9617         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9618         ctsio->be_move_done = ctl_config_move_done;
9619         ctl_datamove((union ctl_io *)ctsio);
9620         return (CTL_RETVAL_COMPLETE);
9621 }
9622
9623 /*
9624  * SCSI VPD page 0x83, the Device Identification page.
9625  */
9626 static int
9627 ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
9628 {
9629         struct scsi_vpd_device_id *devid_ptr;
9630         struct scsi_vpd_id_descriptor *desc;
9631         struct ctl_softc *softc;
9632         struct ctl_lun *lun;
9633         struct ctl_port *port;
9634         int data_len, g;
9635         uint8_t proto;
9636
9637         softc = control_softc;
9638
9639         port = ctl_io_port(&ctsio->io_hdr);
9640         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9641
9642         data_len = sizeof(struct scsi_vpd_device_id) +
9643             sizeof(struct scsi_vpd_id_descriptor) +
9644                 sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
9645             sizeof(struct scsi_vpd_id_descriptor) +
9646                 sizeof(struct scsi_vpd_id_trgt_port_grp_id);
9647         if (lun && lun->lun_devid)
9648                 data_len += lun->lun_devid->len;
9649         if (port && port->port_devid)
9650                 data_len += port->port_devid->len;
9651         if (port && port->target_devid)
9652                 data_len += port->target_devid->len;
9653
9654         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9655         devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
9656         ctsio->kern_sg_entries = 0;
9657
9658         if (data_len < alloc_len) {
9659                 ctsio->residual = alloc_len - data_len;
9660                 ctsio->kern_data_len = data_len;
9661                 ctsio->kern_total_len = data_len;
9662         } else {
9663                 ctsio->residual = 0;
9664                 ctsio->kern_data_len = alloc_len;
9665                 ctsio->kern_total_len = alloc_len;
9666         }
9667         ctsio->kern_data_resid = 0;
9668         ctsio->kern_rel_offset = 0;
9669         ctsio->kern_sg_entries = 0;
9670
9671         /*
9672          * The control device is always connected.  The disk device, on the
9673          * other hand, may not be online all the time.
9674          */
9675         if (lun != NULL)
9676                 devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9677                                      lun->be_lun->lun_type;
9678         else
9679                 devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9680         devid_ptr->page_code = SVPD_DEVICE_ID;
9681         scsi_ulto2b(data_len - 4, devid_ptr->length);
9682
9683         if (port && port->port_type == CTL_PORT_FC)
9684                 proto = SCSI_PROTO_FC << 4;
9685         else if (port && port->port_type == CTL_PORT_ISCSI)
9686                 proto = SCSI_PROTO_ISCSI << 4;
9687         else
9688                 proto = SCSI_PROTO_SPI << 4;
9689         desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
9690
9691         /*
9692          * We're using a LUN association here.  i.e., this device ID is a
9693          * per-LUN identifier.
9694          */
9695         if (lun && lun->lun_devid) {
9696                 memcpy(desc, lun->lun_devid->data, lun->lun_devid->len);
9697                 desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9698                     lun->lun_devid->len);
9699         }
9700
9701         /*
9702          * This is for the WWPN which is a port association.
9703          */
9704         if (port && port->port_devid) {
9705                 memcpy(desc, port->port_devid->data, port->port_devid->len);
9706                 desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9707                     port->port_devid->len);
9708         }
9709
9710         /*
9711          * This is for the Relative Target Port(type 4h) identifier
9712          */
9713         desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9714         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9715             SVPD_ID_TYPE_RELTARG;
9716         desc->length = 4;
9717         scsi_ulto2b(ctsio->io_hdr.nexus.targ_port, &desc->identifier[2]);
9718         desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9719             sizeof(struct scsi_vpd_id_rel_trgt_port_id));
9720
9721         /*
9722          * This is for the Target Port Group(type 5h) identifier
9723          */
9724         desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9725         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9726             SVPD_ID_TYPE_TPORTGRP;
9727         desc->length = 4;
9728         if (softc->is_single ||
9729             (port && port->status & CTL_PORT_STATUS_HA_SHARED))
9730                 g = 1;
9731         else
9732                 g = 2 + ctsio->io_hdr.nexus.targ_port / softc->port_cnt;
9733         scsi_ulto2b(g, &desc->identifier[2]);
9734         desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9735             sizeof(struct scsi_vpd_id_trgt_port_grp_id));
9736
9737         /*
9738          * This is for the Target identifier
9739          */
9740         if (port && port->target_devid) {
9741                 memcpy(desc, port->target_devid->data, port->target_devid->len);
9742         }
9743
9744         ctl_set_success(ctsio);
9745         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9746         ctsio->be_move_done = ctl_config_move_done;
9747         ctl_datamove((union ctl_io *)ctsio);
9748         return (CTL_RETVAL_COMPLETE);
9749 }
9750
9751 static int
9752 ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio, int alloc_len)
9753 {
9754         struct ctl_softc *softc = control_softc;
9755         struct scsi_vpd_scsi_ports *sp;
9756         struct scsi_vpd_port_designation *pd;
9757         struct scsi_vpd_port_designation_cont *pdc;
9758         struct ctl_lun *lun;
9759         struct ctl_port *port;
9760         int data_len, num_target_ports, iid_len, id_len;
9761
9762         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9763
9764         num_target_ports = 0;
9765         iid_len = 0;
9766         id_len = 0;
9767         mtx_lock(&softc->ctl_lock);
9768         STAILQ_FOREACH(port, &softc->port_list, links) {
9769                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
9770                         continue;
9771                 if (lun != NULL &&
9772                     ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
9773                         continue;
9774                 num_target_ports++;
9775                 if (port->init_devid)
9776                         iid_len += port->init_devid->len;
9777                 if (port->port_devid)
9778                         id_len += port->port_devid->len;
9779         }
9780         mtx_unlock(&softc->ctl_lock);
9781
9782         data_len = sizeof(struct scsi_vpd_scsi_ports) +
9783             num_target_ports * (sizeof(struct scsi_vpd_port_designation) +
9784              sizeof(struct scsi_vpd_port_designation_cont)) + iid_len + id_len;
9785         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9786         sp = (struct scsi_vpd_scsi_ports *)ctsio->kern_data_ptr;
9787         ctsio->kern_sg_entries = 0;
9788
9789         if (data_len < alloc_len) {
9790                 ctsio->residual = alloc_len - data_len;
9791                 ctsio->kern_data_len = data_len;
9792                 ctsio->kern_total_len = data_len;
9793         } else {
9794                 ctsio->residual = 0;
9795                 ctsio->kern_data_len = alloc_len;
9796                 ctsio->kern_total_len = alloc_len;
9797         }
9798         ctsio->kern_data_resid = 0;
9799         ctsio->kern_rel_offset = 0;
9800         ctsio->kern_sg_entries = 0;
9801
9802         /*
9803          * The control device is always connected.  The disk device, on the
9804          * other hand, may not be online all the time.  Need to change this
9805          * to figure out whether the disk device is actually online or not.
9806          */
9807         if (lun != NULL)
9808                 sp->device = (SID_QUAL_LU_CONNECTED << 5) |
9809                                   lun->be_lun->lun_type;
9810         else
9811                 sp->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9812
9813         sp->page_code = SVPD_SCSI_PORTS;
9814         scsi_ulto2b(data_len - sizeof(struct scsi_vpd_scsi_ports),
9815             sp->page_length);
9816         pd = &sp->design[0];
9817
9818         mtx_lock(&softc->ctl_lock);
9819         STAILQ_FOREACH(port, &softc->port_list, links) {
9820                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
9821                         continue;
9822                 if (lun != NULL &&
9823                     ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
9824                         continue;
9825                 scsi_ulto2b(port->targ_port, pd->relative_port_id);
9826                 if (port->init_devid) {
9827                         iid_len = port->init_devid->len;
9828                         memcpy(pd->initiator_transportid,
9829                             port->init_devid->data, port->init_devid->len);
9830                 } else
9831                         iid_len = 0;
9832                 scsi_ulto2b(iid_len, pd->initiator_transportid_length);
9833                 pdc = (struct scsi_vpd_port_designation_cont *)
9834                     (&pd->initiator_transportid[iid_len]);
9835                 if (port->port_devid) {
9836                         id_len = port->port_devid->len;
9837                         memcpy(pdc->target_port_descriptors,
9838                             port->port_devid->data, port->port_devid->len);
9839                 } else
9840                         id_len = 0;
9841                 scsi_ulto2b(id_len, pdc->target_port_descriptors_length);
9842                 pd = (struct scsi_vpd_port_designation *)
9843                     ((uint8_t *)pdc->target_port_descriptors + id_len);
9844         }
9845         mtx_unlock(&softc->ctl_lock);
9846
9847         ctl_set_success(ctsio);
9848         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9849         ctsio->be_move_done = ctl_config_move_done;
9850         ctl_datamove((union ctl_io *)ctsio);
9851         return (CTL_RETVAL_COMPLETE);
9852 }
9853
9854 static int
9855 ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio, int alloc_len)
9856 {
9857         struct scsi_vpd_block_limits *bl_ptr;
9858         struct ctl_lun *lun;
9859         uint64_t ival;
9860
9861         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9862
9863         ctsio->kern_data_ptr = malloc(sizeof(*bl_ptr), M_CTL, M_WAITOK | M_ZERO);
9864         bl_ptr = (struct scsi_vpd_block_limits *)ctsio->kern_data_ptr;
9865         ctsio->kern_sg_entries = 0;
9866
9867         if (sizeof(*bl_ptr) < alloc_len) {
9868                 ctsio->residual = alloc_len - sizeof(*bl_ptr);
9869                 ctsio->kern_data_len = sizeof(*bl_ptr);
9870                 ctsio->kern_total_len = sizeof(*bl_ptr);
9871         } else {
9872                 ctsio->residual = 0;
9873                 ctsio->kern_data_len = alloc_len;
9874                 ctsio->kern_total_len = alloc_len;
9875         }
9876         ctsio->kern_data_resid = 0;
9877         ctsio->kern_rel_offset = 0;
9878         ctsio->kern_sg_entries = 0;
9879
9880         /*
9881          * The control device is always connected.  The disk device, on the
9882          * other hand, may not be online all the time.  Need to change this
9883          * to figure out whether the disk device is actually online or not.
9884          */
9885         if (lun != NULL)
9886                 bl_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9887                                   lun->be_lun->lun_type;
9888         else
9889                 bl_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9890
9891         bl_ptr->page_code = SVPD_BLOCK_LIMITS;
9892         scsi_ulto2b(sizeof(*bl_ptr) - 4, bl_ptr->page_length);
9893         bl_ptr->max_cmp_write_len = 0xff;
9894         scsi_ulto4b(0xffffffff, bl_ptr->max_txfer_len);
9895         if (lun != NULL) {
9896                 scsi_ulto4b(lun->be_lun->opttxferlen, bl_ptr->opt_txfer_len);
9897                 if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
9898                         ival = 0xffffffff;
9899                         ctl_get_opt_number(&lun->be_lun->options,
9900                             "unmap_max_lba", &ival);
9901                         scsi_ulto4b(ival, bl_ptr->max_unmap_lba_cnt);
9902                         ival = 0xffffffff;
9903                         ctl_get_opt_number(&lun->be_lun->options,
9904                             "unmap_max_descr", &ival);
9905                         scsi_ulto4b(ival, bl_ptr->max_unmap_blk_cnt);
9906                         if (lun->be_lun->ublockexp != 0) {
9907                                 scsi_ulto4b((1 << lun->be_lun->ublockexp),
9908                                     bl_ptr->opt_unmap_grain);
9909                                 scsi_ulto4b(0x80000000 | lun->be_lun->ublockoff,
9910                                     bl_ptr->unmap_grain_align);
9911                         }
9912                 }
9913                 scsi_ulto4b(lun->be_lun->atomicblock,
9914                     bl_ptr->max_atomic_transfer_length);
9915                 scsi_ulto4b(0, bl_ptr->atomic_alignment);
9916                 scsi_ulto4b(0, bl_ptr->atomic_transfer_length_granularity);
9917                 scsi_ulto4b(0, bl_ptr->max_atomic_transfer_length_with_atomic_boundary);
9918                 scsi_ulto4b(0, bl_ptr->max_atomic_boundary_size);
9919                 ival = UINT64_MAX;
9920                 ctl_get_opt_number(&lun->be_lun->options, "write_same_max_lba", &ival);
9921                 scsi_u64to8b(ival, bl_ptr->max_write_same_length);
9922         }
9923
9924         ctl_set_success(ctsio);
9925         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9926         ctsio->be_move_done = ctl_config_move_done;
9927         ctl_datamove((union ctl_io *)ctsio);
9928         return (CTL_RETVAL_COMPLETE);
9929 }
9930
9931 static int
9932 ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len)
9933 {
9934         struct scsi_vpd_block_device_characteristics *bdc_ptr;
9935         struct ctl_lun *lun;
9936         const char *value;
9937         u_int i;
9938
9939         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9940
9941         ctsio->kern_data_ptr = malloc(sizeof(*bdc_ptr), M_CTL, M_WAITOK | M_ZERO);
9942         bdc_ptr = (struct scsi_vpd_block_device_characteristics *)ctsio->kern_data_ptr;
9943         ctsio->kern_sg_entries = 0;
9944
9945         if (sizeof(*bdc_ptr) < alloc_len) {
9946                 ctsio->residual = alloc_len - sizeof(*bdc_ptr);
9947                 ctsio->kern_data_len = sizeof(*bdc_ptr);
9948                 ctsio->kern_total_len = sizeof(*bdc_ptr);
9949         } else {
9950                 ctsio->residual = 0;
9951                 ctsio->kern_data_len = alloc_len;
9952                 ctsio->kern_total_len = alloc_len;
9953         }
9954         ctsio->kern_data_resid = 0;
9955         ctsio->kern_rel_offset = 0;
9956         ctsio->kern_sg_entries = 0;
9957
9958         /*
9959          * The control device is always connected.  The disk device, on the
9960          * other hand, may not be online all the time.  Need to change this
9961          * to figure out whether the disk device is actually online or not.
9962          */
9963         if (lun != NULL)
9964                 bdc_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9965                                   lun->be_lun->lun_type;
9966         else
9967                 bdc_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9968         bdc_ptr->page_code = SVPD_BDC;
9969         scsi_ulto2b(sizeof(*bdc_ptr) - 4, bdc_ptr->page_length);
9970         if (lun != NULL &&
9971             (value = ctl_get_opt(&lun->be_lun->options, "rpm")) != NULL)
9972                 i = strtol(value, NULL, 0);
9973         else
9974                 i = CTL_DEFAULT_ROTATION_RATE;
9975         scsi_ulto2b(i, bdc_ptr->medium_rotation_rate);
9976         if (lun != NULL &&
9977             (value = ctl_get_opt(&lun->be_lun->options, "formfactor")) != NULL)
9978                 i = strtol(value, NULL, 0);
9979         else
9980                 i = 0;
9981         bdc_ptr->wab_wac_ff = (i & 0x0f);
9982         bdc_ptr->flags = SVPD_FUAB | SVPD_VBULS;
9983
9984         ctl_set_success(ctsio);
9985         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9986         ctsio->be_move_done = ctl_config_move_done;
9987         ctl_datamove((union ctl_io *)ctsio);
9988         return (CTL_RETVAL_COMPLETE);
9989 }
9990
9991 static int
9992 ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len)
9993 {
9994         struct scsi_vpd_logical_block_prov *lbp_ptr;
9995         struct ctl_lun *lun;
9996         const char *value;
9997
9998         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9999
10000         ctsio->kern_data_ptr = malloc(sizeof(*lbp_ptr), M_CTL, M_WAITOK | M_ZERO);
10001         lbp_ptr = (struct scsi_vpd_logical_block_prov *)ctsio->kern_data_ptr;
10002         ctsio->kern_sg_entries = 0;
10003
10004         if (sizeof(*lbp_ptr) < alloc_len) {
10005                 ctsio->residual = alloc_len - sizeof(*lbp_ptr);
10006                 ctsio->kern_data_len = sizeof(*lbp_ptr);
10007                 ctsio->kern_total_len = sizeof(*lbp_ptr);
10008         } else {
10009                 ctsio->residual = 0;
10010                 ctsio->kern_data_len = alloc_len;
10011                 ctsio->kern_total_len = alloc_len;
10012         }
10013         ctsio->kern_data_resid = 0;
10014         ctsio->kern_rel_offset = 0;
10015         ctsio->kern_sg_entries = 0;
10016
10017         /*
10018          * The control device is always connected.  The disk device, on the
10019          * other hand, may not be online all the time.  Need to change this
10020          * to figure out whether the disk device is actually online or not.
10021          */
10022         if (lun != NULL)
10023                 lbp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10024                                   lun->be_lun->lun_type;
10025         else
10026                 lbp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10027
10028         lbp_ptr->page_code = SVPD_LBP;
10029         scsi_ulto2b(sizeof(*lbp_ptr) - 4, lbp_ptr->page_length);
10030         lbp_ptr->threshold_exponent = CTL_LBP_EXPONENT;
10031         if (lun != NULL && lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
10032                 lbp_ptr->flags = SVPD_LBP_UNMAP | SVPD_LBP_WS16 |
10033                     SVPD_LBP_WS10 | SVPD_LBP_RZ | SVPD_LBP_ANC_SUP;
10034                 value = ctl_get_opt(&lun->be_lun->options, "provisioning_type");
10035                 if (value != NULL) {
10036                         if (strcmp(value, "resource") == 0)
10037                                 lbp_ptr->prov_type = SVPD_LBP_RESOURCE;
10038                         else if (strcmp(value, "thin") == 0)
10039                                 lbp_ptr->prov_type = SVPD_LBP_THIN;
10040                 } else
10041                         lbp_ptr->prov_type = SVPD_LBP_THIN;
10042         }
10043
10044         ctl_set_success(ctsio);
10045         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10046         ctsio->be_move_done = ctl_config_move_done;
10047         ctl_datamove((union ctl_io *)ctsio);
10048         return (CTL_RETVAL_COMPLETE);
10049 }
10050
10051 /*
10052  * INQUIRY with the EVPD bit set.
10053  */
10054 static int
10055 ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
10056 {
10057         struct ctl_lun *lun;
10058         struct scsi_inquiry *cdb;
10059         int alloc_len, retval;
10060
10061         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10062         cdb = (struct scsi_inquiry *)ctsio->cdb;
10063         alloc_len = scsi_2btoul(cdb->length);
10064
10065         switch (cdb->page_code) {
10066         case SVPD_SUPPORTED_PAGES:
10067                 retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
10068                 break;
10069         case SVPD_UNIT_SERIAL_NUMBER:
10070                 retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
10071                 break;
10072         case SVPD_DEVICE_ID:
10073                 retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
10074                 break;
10075         case SVPD_EXTENDED_INQUIRY_DATA:
10076                 retval = ctl_inquiry_evpd_eid(ctsio, alloc_len);
10077                 break;
10078         case SVPD_MODE_PAGE_POLICY:
10079                 retval = ctl_inquiry_evpd_mpp(ctsio, alloc_len);
10080                 break;
10081         case SVPD_SCSI_PORTS:
10082                 retval = ctl_inquiry_evpd_scsi_ports(ctsio, alloc_len);
10083                 break;
10084         case SVPD_SCSI_TPC:
10085                 retval = ctl_inquiry_evpd_tpc(ctsio, alloc_len);
10086                 break;
10087         case SVPD_BLOCK_LIMITS:
10088                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
10089                         goto err;
10090                 retval = ctl_inquiry_evpd_block_limits(ctsio, alloc_len);
10091                 break;
10092         case SVPD_BDC:
10093                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
10094                         goto err;
10095                 retval = ctl_inquiry_evpd_bdc(ctsio, alloc_len);
10096                 break;
10097         case SVPD_LBP:
10098                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
10099                         goto err;
10100                 retval = ctl_inquiry_evpd_lbp(ctsio, alloc_len);
10101                 break;
10102         default:
10103 err:
10104                 ctl_set_invalid_field(ctsio,
10105                                       /*sks_valid*/ 1,
10106                                       /*command*/ 1,
10107                                       /*field*/ 2,
10108                                       /*bit_valid*/ 0,
10109                                       /*bit*/ 0);
10110                 ctl_done((union ctl_io *)ctsio);
10111                 retval = CTL_RETVAL_COMPLETE;
10112                 break;
10113         }
10114
10115         return (retval);
10116 }
10117
10118 /*
10119  * Standard INQUIRY data.
10120  */
10121 static int
10122 ctl_inquiry_std(struct ctl_scsiio *ctsio)
10123 {
10124         struct scsi_inquiry_data *inq_ptr;
10125         struct scsi_inquiry *cdb;
10126         struct ctl_softc *softc = control_softc;
10127         struct ctl_port *port;
10128         struct ctl_lun *lun;
10129         char *val;
10130         uint32_t alloc_len, data_len;
10131         ctl_port_type port_type;
10132
10133         port = ctl_io_port(&ctsio->io_hdr);
10134         port_type = port->port_type;
10135         if (port_type == CTL_PORT_IOCTL || port_type == CTL_PORT_INTERNAL)
10136                 port_type = CTL_PORT_SCSI;
10137
10138         lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10139         cdb = (struct scsi_inquiry *)ctsio->cdb;
10140         alloc_len = scsi_2btoul(cdb->length);
10141
10142         /*
10143          * We malloc the full inquiry data size here and fill it
10144          * in.  If the user only asks for less, we'll give him
10145          * that much.
10146          */
10147         data_len = offsetof(struct scsi_inquiry_data, vendor_specific1);
10148         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10149         inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
10150         ctsio->kern_sg_entries = 0;
10151         ctsio->kern_data_resid = 0;
10152         ctsio->kern_rel_offset = 0;
10153
10154         if (data_len < alloc_len) {
10155                 ctsio->residual = alloc_len - data_len;
10156                 ctsio->kern_data_len = data_len;
10157                 ctsio->kern_total_len = data_len;
10158         } else {
10159                 ctsio->residual = 0;
10160                 ctsio->kern_data_len = alloc_len;
10161                 ctsio->kern_total_len = alloc_len;
10162         }
10163
10164         if (lun != NULL) {
10165                 if ((lun->flags & CTL_LUN_PRIMARY_SC) ||
10166                     softc->ha_link >= CTL_HA_LINK_UNKNOWN) {
10167                         inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10168                             lun->be_lun->lun_type;
10169                 } else {
10170                         inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) |
10171                             lun->be_lun->lun_type;
10172                 }
10173                 if (lun->flags & CTL_LUN_REMOVABLE)
10174                         inq_ptr->dev_qual2 |= SID_RMB;
10175         } else
10176                 inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE;
10177
10178         /* RMB in byte 2 is 0 */
10179         inq_ptr->version = SCSI_REV_SPC5;
10180
10181         /*
10182          * According to SAM-3, even if a device only supports a single
10183          * level of LUN addressing, it should still set the HISUP bit:
10184          *
10185          * 4.9.1 Logical unit numbers overview
10186          *
10187          * All logical unit number formats described in this standard are
10188          * hierarchical in structure even when only a single level in that
10189          * hierarchy is used. The HISUP bit shall be set to one in the
10190          * standard INQUIRY data (see SPC-2) when any logical unit number
10191          * format described in this standard is used.  Non-hierarchical
10192          * formats are outside the scope of this standard.
10193          *
10194          * Therefore we set the HiSup bit here.
10195          *
10196          * The response format is 2, per SPC-3.
10197          */
10198         inq_ptr->response_format = SID_HiSup | 2;
10199
10200         inq_ptr->additional_length = data_len -
10201             (offsetof(struct scsi_inquiry_data, additional_length) + 1);
10202         CTL_DEBUG_PRINT(("additional_length = %d\n",
10203                          inq_ptr->additional_length));
10204
10205         inq_ptr->spc3_flags = SPC3_SID_3PC | SPC3_SID_TPGS_IMPLICIT;
10206         if (port_type == CTL_PORT_SCSI)
10207                 inq_ptr->spc2_flags = SPC2_SID_ADDR16;
10208         inq_ptr->spc2_flags |= SPC2_SID_MultiP;
10209         inq_ptr->flags = SID_CmdQue;
10210         if (port_type == CTL_PORT_SCSI)
10211                 inq_ptr->flags |= SID_WBus16 | SID_Sync;
10212
10213         /*
10214          * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
10215          * We have 8 bytes for the vendor name, and 16 bytes for the device
10216          * name and 4 bytes for the revision.
10217          */
10218         if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10219             "vendor")) == NULL) {
10220                 strncpy(inq_ptr->vendor, CTL_VENDOR, sizeof(inq_ptr->vendor));
10221         } else {
10222                 memset(inq_ptr->vendor, ' ', sizeof(inq_ptr->vendor));
10223                 strncpy(inq_ptr->vendor, val,
10224                     min(sizeof(inq_ptr->vendor), strlen(val)));
10225         }
10226         if (lun == NULL) {
10227                 strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10228                     sizeof(inq_ptr->product));
10229         } else if ((val = ctl_get_opt(&lun->be_lun->options, "product")) == NULL) {
10230                 switch (lun->be_lun->lun_type) {
10231                 case T_DIRECT:
10232                         strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10233                             sizeof(inq_ptr->product));
10234                         break;
10235                 case T_PROCESSOR:
10236                         strncpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT,
10237                             sizeof(inq_ptr->product));
10238                         break;
10239                 case T_CDROM:
10240                         strncpy(inq_ptr->product, CTL_CDROM_PRODUCT,
10241                             sizeof(inq_ptr->product));
10242                         break;
10243                 default:
10244                         strncpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT,
10245                             sizeof(inq_ptr->product));
10246                         break;
10247                 }
10248         } else {
10249                 memset(inq_ptr->product, ' ', sizeof(inq_ptr->product));
10250                 strncpy(inq_ptr->product, val,
10251                     min(sizeof(inq_ptr->product), strlen(val)));
10252         }
10253
10254         /*
10255          * XXX make this a macro somewhere so it automatically gets
10256          * incremented when we make changes.
10257          */
10258         if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10259             "revision")) == NULL) {
10260                 strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
10261         } else {
10262                 memset(inq_ptr->revision, ' ', sizeof(inq_ptr->revision));
10263                 strncpy(inq_ptr->revision, val,
10264                     min(sizeof(inq_ptr->revision), strlen(val)));
10265         }
10266
10267         /*
10268          * For parallel SCSI, we support double transition and single
10269          * transition clocking.  We also support QAS (Quick Arbitration
10270          * and Selection) and Information Unit transfers on both the
10271          * control and array devices.
10272          */
10273         if (port_type == CTL_PORT_SCSI)
10274                 inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
10275                                     SID_SPI_IUS;
10276
10277         /* SAM-6 (no version claimed) */
10278         scsi_ulto2b(0x00C0, inq_ptr->version1);
10279         /* SPC-5 (no version claimed) */
10280         scsi_ulto2b(0x05C0, inq_ptr->version2);
10281         if (port_type == CTL_PORT_FC) {
10282                 /* FCP-2 ANSI INCITS.350:2003 */
10283                 scsi_ulto2b(0x0917, inq_ptr->version3);
10284         } else if (port_type == CTL_PORT_SCSI) {
10285                 /* SPI-4 ANSI INCITS.362:200x */
10286                 scsi_ulto2b(0x0B56, inq_ptr->version3);
10287         } else if (port_type == CTL_PORT_ISCSI) {
10288                 /* iSCSI (no version claimed) */
10289                 scsi_ulto2b(0x0960, inq_ptr->version3);
10290         } else if (port_type == CTL_PORT_SAS) {
10291                 /* SAS (no version claimed) */
10292                 scsi_ulto2b(0x0BE0, inq_ptr->version3);
10293         }
10294
10295         if (lun == NULL) {
10296                 /* SBC-4 (no version claimed) */
10297                 scsi_ulto2b(0x0600, inq_ptr->version4);
10298         } else {
10299                 switch (lun->be_lun->lun_type) {
10300                 case T_DIRECT:
10301                         /* SBC-4 (no version claimed) */
10302                         scsi_ulto2b(0x0600, inq_ptr->version4);
10303                         break;
10304                 case T_PROCESSOR:
10305                         break;
10306                 case T_CDROM:
10307                         /* MMC-6 (no version claimed) */
10308                         scsi_ulto2b(0x04E0, inq_ptr->version4);
10309                         break;
10310                 default:
10311                         break;
10312                 }
10313         }
10314
10315         ctl_set_success(ctsio);
10316         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10317         ctsio->be_move_done = ctl_config_move_done;
10318         ctl_datamove((union ctl_io *)ctsio);
10319         return (CTL_RETVAL_COMPLETE);
10320 }
10321
10322 int
10323 ctl_inquiry(struct ctl_scsiio *ctsio)
10324 {
10325         struct scsi_inquiry *cdb;
10326         int retval;
10327
10328         CTL_DEBUG_PRINT(("ctl_inquiry\n"));
10329
10330         cdb = (struct scsi_inquiry *)ctsio->cdb;
10331         if (cdb->byte2 & SI_EVPD)
10332                 retval = ctl_inquiry_evpd(ctsio);
10333         else if (cdb->page_code == 0)
10334                 retval = ctl_inquiry_std(ctsio);
10335         else {
10336                 ctl_set_invalid_field(ctsio,
10337                                       /*sks_valid*/ 1,
10338                                       /*command*/ 1,
10339                                       /*field*/ 2,
10340                                       /*bit_valid*/ 0,
10341                                       /*bit*/ 0);
10342                 ctl_done((union ctl_io *)ctsio);
10343                 return (CTL_RETVAL_COMPLETE);
10344         }
10345
10346         return (retval);
10347 }
10348
10349 int
10350 ctl_get_config(struct ctl_scsiio *ctsio)
10351 {
10352         struct scsi_get_config_header *hdr;
10353         struct scsi_get_config_feature *feature;
10354         struct scsi_get_config *cdb;
10355         struct ctl_lun *lun;
10356         uint32_t alloc_len, data_len;
10357         int rt, starting;
10358
10359         lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10360         cdb = (struct scsi_get_config *)ctsio->cdb;
10361         rt = (cdb->rt & SGC_RT_MASK);
10362         starting = scsi_2btoul(cdb->starting_feature);
10363         alloc_len = scsi_2btoul(cdb->length);
10364
10365         data_len = sizeof(struct scsi_get_config_header) +
10366             sizeof(struct scsi_get_config_feature) + 8 +
10367             sizeof(struct scsi_get_config_feature) + 8 +
10368             sizeof(struct scsi_get_config_feature) + 4 +
10369             sizeof(struct scsi_get_config_feature) + 4 +
10370             sizeof(struct scsi_get_config_feature) + 8 +
10371             sizeof(struct scsi_get_config_feature) +
10372             sizeof(struct scsi_get_config_feature) + 4 +
10373             sizeof(struct scsi_get_config_feature) + 4 +
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         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10379         ctsio->kern_sg_entries = 0;
10380         ctsio->kern_data_resid = 0;
10381         ctsio->kern_rel_offset = 0;
10382
10383         hdr = (struct scsi_get_config_header *)ctsio->kern_data_ptr;
10384         if (lun->flags & CTL_LUN_NO_MEDIA)
10385                 scsi_ulto2b(0x0000, hdr->current_profile);
10386         else
10387                 scsi_ulto2b(0x0010, hdr->current_profile);
10388         feature = (struct scsi_get_config_feature *)(hdr + 1);
10389
10390         if (starting > 0x003b)
10391                 goto done;
10392         if (starting > 0x003a)
10393                 goto f3b;
10394         if (starting > 0x002b)
10395                 goto f3a;
10396         if (starting > 0x002a)
10397                 goto f2b;
10398         if (starting > 0x001f)
10399                 goto f2a;
10400         if (starting > 0x001e)
10401                 goto f1f;
10402         if (starting > 0x001d)
10403                 goto f1e;
10404         if (starting > 0x0010)
10405                 goto f1d;
10406         if (starting > 0x0003)
10407                 goto f10;
10408         if (starting > 0x0002)
10409                 goto f3;
10410         if (starting > 0x0001)
10411                 goto f2;
10412         if (starting > 0x0000)
10413                 goto f1;
10414
10415         /* Profile List */
10416         scsi_ulto2b(0x0000, feature->feature_code);
10417         feature->flags = SGC_F_PERSISTENT | SGC_F_CURRENT;
10418         feature->add_length = 8;
10419         scsi_ulto2b(0x0008, &feature->feature_data[0]); /* CD-ROM */
10420         feature->feature_data[2] = 0x00;
10421         scsi_ulto2b(0x0010, &feature->feature_data[4]); /* DVD-ROM */
10422         feature->feature_data[6] = 0x01;
10423         feature = (struct scsi_get_config_feature *)
10424             &feature->feature_data[feature->add_length];
10425
10426 f1:     /* Core */
10427         scsi_ulto2b(0x0001, feature->feature_code);
10428         feature->flags = 0x08 | SGC_F_PERSISTENT | SGC_F_CURRENT;
10429         feature->add_length = 8;
10430         scsi_ulto4b(0x00000000, &feature->feature_data[0]);
10431         feature->feature_data[4] = 0x03;
10432         feature = (struct scsi_get_config_feature *)
10433             &feature->feature_data[feature->add_length];
10434
10435 f2:     /* Morphing */
10436         scsi_ulto2b(0x0002, feature->feature_code);
10437         feature->flags = 0x04 | SGC_F_PERSISTENT | SGC_F_CURRENT;
10438         feature->add_length = 4;
10439         feature->feature_data[0] = 0x02;
10440         feature = (struct scsi_get_config_feature *)
10441             &feature->feature_data[feature->add_length];
10442
10443 f3:     /* Removable Medium */
10444         scsi_ulto2b(0x0003, feature->feature_code);
10445         feature->flags = 0x04 | SGC_F_PERSISTENT | SGC_F_CURRENT;
10446         feature->add_length = 4;
10447         feature->feature_data[0] = 0x39;
10448         feature = (struct scsi_get_config_feature *)
10449             &feature->feature_data[feature->add_length];
10450
10451         if (rt == SGC_RT_CURRENT && (lun->flags & CTL_LUN_NO_MEDIA))
10452                 goto done;
10453
10454 f10:    /* Random Read */
10455         scsi_ulto2b(0x0010, feature->feature_code);
10456         feature->flags = 0x00;
10457         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10458                 feature->flags |= SGC_F_CURRENT;
10459         feature->add_length = 8;
10460         scsi_ulto4b(lun->be_lun->blocksize, &feature->feature_data[0]);
10461         scsi_ulto2b(1, &feature->feature_data[4]);
10462         feature->feature_data[6] = 0x00;
10463         feature = (struct scsi_get_config_feature *)
10464             &feature->feature_data[feature->add_length];
10465
10466 f1d:    /* Multi-Read */
10467         scsi_ulto2b(0x001D, feature->feature_code);
10468         feature->flags = 0x00;
10469         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10470                 feature->flags |= SGC_F_CURRENT;
10471         feature->add_length = 0;
10472         feature = (struct scsi_get_config_feature *)
10473             &feature->feature_data[feature->add_length];
10474
10475 f1e:    /* CD Read */
10476         scsi_ulto2b(0x001E, feature->feature_code);
10477         feature->flags = 0x00;
10478         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10479                 feature->flags |= SGC_F_CURRENT;
10480         feature->add_length = 4;
10481         feature->feature_data[0] = 0x00;
10482         feature = (struct scsi_get_config_feature *)
10483             &feature->feature_data[feature->add_length];
10484
10485 f1f:    /* DVD Read */
10486         scsi_ulto2b(0x001F, feature->feature_code);
10487         feature->flags = 0x08;
10488         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10489                 feature->flags |= SGC_F_CURRENT;
10490         feature->add_length = 4;
10491         feature->feature_data[0] = 0x01;
10492         feature->feature_data[2] = 0x03;
10493         feature = (struct scsi_get_config_feature *)
10494             &feature->feature_data[feature->add_length];
10495
10496 f2a:    /* DVD+RW */
10497         scsi_ulto2b(0x002A, feature->feature_code);
10498         feature->flags = 0x04;
10499         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10500                 feature->flags |= SGC_F_CURRENT;
10501         feature->add_length = 4;
10502         feature->feature_data[0] = 0x00;
10503         feature->feature_data[1] = 0x00;
10504         feature = (struct scsi_get_config_feature *)
10505             &feature->feature_data[feature->add_length];
10506
10507 f2b:    /* DVD+R */
10508         scsi_ulto2b(0x002B, feature->feature_code);
10509         feature->flags = 0x00;
10510         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10511                 feature->flags |= SGC_F_CURRENT;
10512         feature->add_length = 4;
10513         feature->feature_data[0] = 0x00;
10514         feature = (struct scsi_get_config_feature *)
10515             &feature->feature_data[feature->add_length];
10516
10517 f3a:    /* DVD+RW Dual Layer */
10518         scsi_ulto2b(0x003A, feature->feature_code);
10519         feature->flags = 0x00;
10520         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10521                 feature->flags |= SGC_F_CURRENT;
10522         feature->add_length = 4;
10523         feature->feature_data[0] = 0x00;
10524         feature->feature_data[1] = 0x00;
10525         feature = (struct scsi_get_config_feature *)
10526             &feature->feature_data[feature->add_length];
10527
10528 f3b:    /* DVD+R Dual Layer */
10529         scsi_ulto2b(0x003B, feature->feature_code);
10530         feature->flags = 0x00;
10531         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10532                 feature->flags |= SGC_F_CURRENT;
10533         feature->add_length = 4;
10534         feature->feature_data[0] = 0x00;
10535         feature = (struct scsi_get_config_feature *)
10536             &feature->feature_data[feature->add_length];
10537
10538 done:
10539         data_len = (uint8_t *)feature - (uint8_t *)hdr;
10540         if (rt == SGC_RT_SPECIFIC && data_len > 4) {
10541                 feature = (struct scsi_get_config_feature *)(hdr + 1);
10542                 if (scsi_2btoul(feature->feature_code) == starting)
10543                         feature = (struct scsi_get_config_feature *)
10544                             &feature->feature_data[feature->add_length];
10545                 data_len = (uint8_t *)feature - (uint8_t *)hdr;
10546         }
10547         scsi_ulto4b(data_len - 4, hdr->data_length);
10548         if (data_len < alloc_len) {
10549                 ctsio->residual = alloc_len - data_len;
10550                 ctsio->kern_data_len = data_len;
10551                 ctsio->kern_total_len = data_len;
10552         } else {
10553                 ctsio->residual = 0;
10554                 ctsio->kern_data_len = alloc_len;
10555                 ctsio->kern_total_len = alloc_len;
10556         }
10557
10558         ctl_set_success(ctsio);
10559         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10560         ctsio->be_move_done = ctl_config_move_done;
10561         ctl_datamove((union ctl_io *)ctsio);
10562         return (CTL_RETVAL_COMPLETE);
10563 }
10564
10565 int
10566 ctl_get_event_status(struct ctl_scsiio *ctsio)
10567 {
10568         struct scsi_get_event_status_header *hdr;
10569         struct scsi_get_event_status *cdb;
10570         struct ctl_lun *lun;
10571         uint32_t alloc_len, data_len;
10572         int notif_class;
10573
10574         lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
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         struct ctl_lun *lun;
10619         uint32_t alloc_len, data_len;
10620
10621         lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10622         cdb = (struct scsi_mechanism_status *)ctsio->cdb;
10623         alloc_len = scsi_2btoul(cdb->length);
10624
10625         data_len = sizeof(struct scsi_mechanism_status_header);
10626         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10627         ctsio->kern_sg_entries = 0;
10628         ctsio->kern_data_resid = 0;
10629         ctsio->kern_rel_offset = 0;
10630
10631         if (data_len < alloc_len) {
10632                 ctsio->residual = alloc_len - data_len;
10633                 ctsio->kern_data_len = data_len;
10634                 ctsio->kern_total_len = data_len;
10635         } else {
10636                 ctsio->residual = 0;
10637                 ctsio->kern_data_len = alloc_len;
10638                 ctsio->kern_total_len = alloc_len;
10639         }
10640
10641         hdr = (struct scsi_mechanism_status_header *)ctsio->kern_data_ptr;
10642         hdr->state1 = 0x00;
10643         hdr->state2 = 0xe0;
10644         scsi_ulto3b(0, hdr->lba);
10645         hdr->slots_num = 0;
10646         scsi_ulto2b(0, hdr->slots_length);
10647
10648         ctl_set_success(ctsio);
10649         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10650         ctsio->be_move_done = ctl_config_move_done;
10651         ctl_datamove((union ctl_io *)ctsio);
10652         return (CTL_RETVAL_COMPLETE);
10653 }
10654
10655 static void
10656 ctl_ultomsf(uint32_t lba, uint8_t *buf)
10657 {
10658
10659         lba += 150;
10660         buf[0] = 0;
10661         buf[1] = bin2bcd((lba / 75) / 60);
10662         buf[2] = bin2bcd((lba / 75) % 60);
10663         buf[3] = bin2bcd(lba % 75);
10664 }
10665
10666 int
10667 ctl_read_toc(struct ctl_scsiio *ctsio)
10668 {
10669         struct scsi_read_toc_hdr *hdr;
10670         struct scsi_read_toc_type01_descr *descr;
10671         struct scsi_read_toc *cdb;
10672         struct ctl_lun *lun;
10673         uint32_t alloc_len, data_len;
10674         int format, msf;
10675
10676         lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10677         cdb = (struct scsi_read_toc *)ctsio->cdb;
10678         msf = (cdb->byte2 & CD_MSF) != 0;
10679         format = cdb->format;
10680         alloc_len = scsi_2btoul(cdb->data_len);
10681
10682         data_len = sizeof(struct scsi_read_toc_hdr);
10683         if (format == 0)
10684                 data_len += 2 * sizeof(struct scsi_read_toc_type01_descr);
10685         else
10686                 data_len += sizeof(struct scsi_read_toc_type01_descr);
10687         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10688         ctsio->kern_sg_entries = 0;
10689         ctsio->kern_data_resid = 0;
10690         ctsio->kern_rel_offset = 0;
10691
10692         if (data_len < alloc_len) {
10693                 ctsio->residual = alloc_len - data_len;
10694                 ctsio->kern_data_len = data_len;
10695                 ctsio->kern_total_len = data_len;
10696         } else {
10697                 ctsio->residual = 0;
10698                 ctsio->kern_data_len = alloc_len;
10699                 ctsio->kern_total_len = alloc_len;
10700         }
10701
10702         hdr = (struct scsi_read_toc_hdr *)ctsio->kern_data_ptr;
10703         if (format == 0) {
10704                 scsi_ulto2b(0x12, hdr->data_length);
10705                 hdr->first = 1;
10706                 hdr->last = 1;
10707                 descr = (struct scsi_read_toc_type01_descr *)(hdr + 1);
10708                 descr->addr_ctl = 0x14;
10709                 descr->track_number = 1;
10710                 if (msf)
10711                         ctl_ultomsf(0, descr->track_start);
10712                 else
10713                         scsi_ulto4b(0, descr->track_start);
10714                 descr++;
10715                 descr->addr_ctl = 0x14;
10716                 descr->track_number = 0xaa;
10717                 if (msf)
10718                         ctl_ultomsf(lun->be_lun->maxlba+1, descr->track_start);
10719                 else
10720                         scsi_ulto4b(lun->be_lun->maxlba+1, descr->track_start);
10721         } else {
10722                 scsi_ulto2b(0x0a, hdr->data_length);
10723                 hdr->first = 1;
10724                 hdr->last = 1;
10725                 descr = (struct scsi_read_toc_type01_descr *)(hdr + 1);
10726                 descr->addr_ctl = 0x14;
10727                 descr->track_number = 1;
10728                 if (msf)
10729                         ctl_ultomsf(0, descr->track_start);
10730                 else
10731                         scsi_ulto4b(0, descr->track_start);
10732         }
10733
10734         ctl_set_success(ctsio);
10735         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10736         ctsio->be_move_done = ctl_config_move_done;
10737         ctl_datamove((union ctl_io *)ctsio);
10738         return (CTL_RETVAL_COMPLETE);
10739 }
10740
10741 /*
10742  * For known CDB types, parse the LBA and length.
10743  */
10744 static int
10745 ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len)
10746 {
10747         if (io->io_hdr.io_type != CTL_IO_SCSI)
10748                 return (1);
10749
10750         switch (io->scsiio.cdb[0]) {
10751         case COMPARE_AND_WRITE: {
10752                 struct scsi_compare_and_write *cdb;
10753
10754                 cdb = (struct scsi_compare_and_write *)io->scsiio.cdb;
10755
10756                 *lba = scsi_8btou64(cdb->addr);
10757                 *len = cdb->length;
10758                 break;
10759         }
10760         case READ_6:
10761         case WRITE_6: {
10762                 struct scsi_rw_6 *cdb;
10763
10764                 cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
10765
10766                 *lba = scsi_3btoul(cdb->addr);
10767                 /* only 5 bits are valid in the most significant address byte */
10768                 *lba &= 0x1fffff;
10769                 *len = cdb->length;
10770                 break;
10771         }
10772         case READ_10:
10773         case WRITE_10: {
10774                 struct scsi_rw_10 *cdb;
10775
10776                 cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
10777
10778                 *lba = scsi_4btoul(cdb->addr);
10779                 *len = scsi_2btoul(cdb->length);
10780                 break;
10781         }
10782         case WRITE_VERIFY_10: {
10783                 struct scsi_write_verify_10 *cdb;
10784
10785                 cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
10786
10787                 *lba = scsi_4btoul(cdb->addr);
10788                 *len = scsi_2btoul(cdb->length);
10789                 break;
10790         }
10791         case READ_12:
10792         case WRITE_12: {
10793                 struct scsi_rw_12 *cdb;
10794
10795                 cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
10796
10797                 *lba = scsi_4btoul(cdb->addr);
10798                 *len = scsi_4btoul(cdb->length);
10799                 break;
10800         }
10801         case WRITE_VERIFY_12: {
10802                 struct scsi_write_verify_12 *cdb;
10803
10804                 cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
10805
10806                 *lba = scsi_4btoul(cdb->addr);
10807                 *len = scsi_4btoul(cdb->length);
10808                 break;
10809         }
10810         case READ_16:
10811         case WRITE_16: {
10812                 struct scsi_rw_16 *cdb;
10813
10814                 cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
10815
10816                 *lba = scsi_8btou64(cdb->addr);
10817                 *len = scsi_4btoul(cdb->length);
10818                 break;
10819         }
10820         case WRITE_ATOMIC_16: {
10821                 struct scsi_write_atomic_16 *cdb;
10822
10823                 cdb = (struct scsi_write_atomic_16 *)io->scsiio.cdb;
10824
10825                 *lba = scsi_8btou64(cdb->addr);
10826                 *len = scsi_2btoul(cdb->length);
10827                 break;
10828         }
10829         case WRITE_VERIFY_16: {
10830                 struct scsi_write_verify_16 *cdb;
10831
10832                 cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
10833
10834                 *lba = scsi_8btou64(cdb->addr);
10835                 *len = scsi_4btoul(cdb->length);
10836                 break;
10837         }
10838         case WRITE_SAME_10: {
10839                 struct scsi_write_same_10 *cdb;
10840
10841                 cdb = (struct scsi_write_same_10 *)io->scsiio.cdb;
10842
10843                 *lba = scsi_4btoul(cdb->addr);
10844                 *len = scsi_2btoul(cdb->length);
10845                 break;
10846         }
10847         case WRITE_SAME_16: {
10848                 struct scsi_write_same_16 *cdb;
10849
10850                 cdb = (struct scsi_write_same_16 *)io->scsiio.cdb;
10851
10852                 *lba = scsi_8btou64(cdb->addr);
10853                 *len = scsi_4btoul(cdb->length);
10854                 break;
10855         }
10856         case VERIFY_10: {
10857                 struct scsi_verify_10 *cdb;
10858
10859                 cdb = (struct scsi_verify_10 *)io->scsiio.cdb;
10860
10861                 *lba = scsi_4btoul(cdb->addr);
10862                 *len = scsi_2btoul(cdb->length);
10863                 break;
10864         }
10865         case VERIFY_12: {
10866                 struct scsi_verify_12 *cdb;
10867
10868                 cdb = (struct scsi_verify_12 *)io->scsiio.cdb;
10869
10870                 *lba = scsi_4btoul(cdb->addr);
10871                 *len = scsi_4btoul(cdb->length);
10872                 break;
10873         }
10874         case VERIFY_16: {
10875                 struct scsi_verify_16 *cdb;
10876
10877                 cdb = (struct scsi_verify_16 *)io->scsiio.cdb;
10878
10879                 *lba = scsi_8btou64(cdb->addr);
10880                 *len = scsi_4btoul(cdb->length);
10881                 break;
10882         }
10883         case UNMAP: {
10884                 *lba = 0;
10885                 *len = UINT64_MAX;
10886                 break;
10887         }
10888         case SERVICE_ACTION_IN: {       /* GET LBA STATUS */
10889                 struct scsi_get_lba_status *cdb;
10890
10891                 cdb = (struct scsi_get_lba_status *)io->scsiio.cdb;
10892                 *lba = scsi_8btou64(cdb->addr);
10893                 *len = UINT32_MAX;
10894                 break;
10895         }
10896         default:
10897                 return (1);
10898                 break; /* NOTREACHED */
10899         }
10900
10901         return (0);
10902 }
10903
10904 static ctl_action
10905 ctl_extent_check_lba(uint64_t lba1, uint64_t len1, uint64_t lba2, uint64_t len2,
10906     bool seq)
10907 {
10908         uint64_t endlba1, endlba2;
10909
10910         endlba1 = lba1 + len1 - (seq ? 0 : 1);
10911         endlba2 = lba2 + len2 - 1;
10912
10913         if ((endlba1 < lba2) || (endlba2 < lba1))
10914                 return (CTL_ACTION_PASS);
10915         else
10916                 return (CTL_ACTION_BLOCK);
10917 }
10918
10919 static int
10920 ctl_extent_check_unmap(union ctl_io *io, uint64_t lba2, uint64_t len2)
10921 {
10922         struct ctl_ptr_len_flags *ptrlen;
10923         struct scsi_unmap_desc *buf, *end, *range;
10924         uint64_t lba;
10925         uint32_t len;
10926
10927         /* If not UNMAP -- go other way. */
10928         if (io->io_hdr.io_type != CTL_IO_SCSI ||
10929             io->scsiio.cdb[0] != UNMAP)
10930                 return (CTL_ACTION_ERROR);
10931
10932         /* If UNMAP without data -- block and wait for data. */
10933         ptrlen = (struct ctl_ptr_len_flags *)
10934             &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
10935         if ((io->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0 ||
10936             ptrlen->ptr == NULL)
10937                 return (CTL_ACTION_BLOCK);
10938
10939         /* UNMAP with data -- check for collision. */
10940         buf = (struct scsi_unmap_desc *)ptrlen->ptr;
10941         end = buf + ptrlen->len / sizeof(*buf);
10942         for (range = buf; range < end; range++) {
10943                 lba = scsi_8btou64(range->lba);
10944                 len = scsi_4btoul(range->length);
10945                 if ((lba < lba2 + len2) && (lba + len > lba2))
10946                         return (CTL_ACTION_BLOCK);
10947         }
10948         return (CTL_ACTION_PASS);
10949 }
10950
10951 static ctl_action
10952 ctl_extent_check(union ctl_io *io1, union ctl_io *io2, bool seq)
10953 {
10954         uint64_t lba1, lba2;
10955         uint64_t len1, len2;
10956         int retval;
10957
10958         if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
10959                 return (CTL_ACTION_ERROR);
10960
10961         retval = ctl_extent_check_unmap(io1, lba2, len2);
10962         if (retval != CTL_ACTION_ERROR)
10963                 return (retval);
10964
10965         if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
10966                 return (CTL_ACTION_ERROR);
10967
10968         if (io1->io_hdr.flags & CTL_FLAG_SERSEQ_DONE)
10969                 seq = FALSE;
10970         return (ctl_extent_check_lba(lba1, len1, lba2, len2, seq));
10971 }
10972
10973 static ctl_action
10974 ctl_extent_check_seq(union ctl_io *io1, union ctl_io *io2)
10975 {
10976         uint64_t lba1, lba2;
10977         uint64_t len1, len2;
10978
10979         if (io1->io_hdr.flags & CTL_FLAG_SERSEQ_DONE)
10980                 return (CTL_ACTION_PASS);
10981         if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
10982                 return (CTL_ACTION_ERROR);
10983         if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
10984                 return (CTL_ACTION_ERROR);
10985
10986         if (lba1 + len1 == lba2)
10987                 return (CTL_ACTION_BLOCK);
10988         return (CTL_ACTION_PASS);
10989 }
10990
10991 static ctl_action
10992 ctl_check_for_blockage(struct ctl_lun *lun, union ctl_io *pending_io,
10993     union ctl_io *ooa_io)
10994 {
10995         const struct ctl_cmd_entry *pending_entry, *ooa_entry;
10996         const ctl_serialize_action *serialize_row;
10997
10998         /*
10999          * The initiator attempted multiple untagged commands at the same
11000          * time.  Can't do that.
11001          */
11002         if ((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
11003          && (ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
11004          && ((pending_io->io_hdr.nexus.targ_port ==
11005               ooa_io->io_hdr.nexus.targ_port)
11006           && (pending_io->io_hdr.nexus.initid ==
11007               ooa_io->io_hdr.nexus.initid))
11008          && ((ooa_io->io_hdr.flags & (CTL_FLAG_ABORT |
11009               CTL_FLAG_STATUS_SENT)) == 0))
11010                 return (CTL_ACTION_OVERLAP);
11011
11012         /*
11013          * The initiator attempted to send multiple tagged commands with
11014          * the same ID.  (It's fine if different initiators have the same
11015          * tag ID.)
11016          *
11017          * Even if all of those conditions are true, we don't kill the I/O
11018          * if the command ahead of us has been aborted.  We won't end up
11019          * sending it to the FETD, and it's perfectly legal to resend a
11020          * command with the same tag number as long as the previous
11021          * instance of this tag number has been aborted somehow.
11022          */
11023         if ((pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
11024          && (ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
11025          && (pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
11026          && ((pending_io->io_hdr.nexus.targ_port ==
11027               ooa_io->io_hdr.nexus.targ_port)
11028           && (pending_io->io_hdr.nexus.initid ==
11029               ooa_io->io_hdr.nexus.initid))
11030          && ((ooa_io->io_hdr.flags & (CTL_FLAG_ABORT |
11031               CTL_FLAG_STATUS_SENT)) == 0))
11032                 return (CTL_ACTION_OVERLAP_TAG);
11033
11034         /*
11035          * If we get a head of queue tag, SAM-3 says that we should
11036          * immediately execute it.
11037          *
11038          * What happens if this command would normally block for some other
11039          * reason?  e.g. a request sense with a head of queue tag
11040          * immediately after a write.  Normally that would block, but this
11041          * will result in its getting executed immediately...
11042          *
11043          * We currently return "pass" instead of "skip", so we'll end up
11044          * going through the rest of the queue to check for overlapped tags.
11045          *
11046          * XXX KDM check for other types of blockage first??
11047          */
11048         if (pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
11049                 return (CTL_ACTION_PASS);
11050
11051         /*
11052          * Ordered tags have to block until all items ahead of them
11053          * have completed.  If we get called with an ordered tag, we always
11054          * block, if something else is ahead of us in the queue.
11055          */
11056         if (pending_io->scsiio.tag_type == CTL_TAG_ORDERED)
11057                 return (CTL_ACTION_BLOCK);
11058
11059         /*
11060          * Simple tags get blocked until all head of queue and ordered tags
11061          * ahead of them have completed.  I'm lumping untagged commands in
11062          * with simple tags here.  XXX KDM is that the right thing to do?
11063          */
11064         if (((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
11065           || (pending_io->scsiio.tag_type == CTL_TAG_SIMPLE))
11066          && ((ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
11067           || (ooa_io->scsiio.tag_type == CTL_TAG_ORDERED)))
11068                 return (CTL_ACTION_BLOCK);
11069
11070         pending_entry = ctl_get_cmd_entry(&pending_io->scsiio, NULL);
11071         KASSERT(pending_entry->seridx < CTL_SERIDX_COUNT,
11072             ("%s: Invalid seridx %d for pending CDB %02x %02x @ %p",
11073              __func__, pending_entry->seridx, pending_io->scsiio.cdb[0],
11074              pending_io->scsiio.cdb[1], pending_io));
11075         ooa_entry = ctl_get_cmd_entry(&ooa_io->scsiio, NULL);
11076         if (ooa_entry->seridx == CTL_SERIDX_INVLD)
11077                 return (CTL_ACTION_PASS); /* Unsupported command in OOA queue */
11078         KASSERT(ooa_entry->seridx < CTL_SERIDX_COUNT,
11079             ("%s: Invalid seridx %d for ooa CDB %02x %02x @ %p",
11080              __func__, ooa_entry->seridx, ooa_io->scsiio.cdb[0],
11081              ooa_io->scsiio.cdb[1], ooa_io));
11082
11083         serialize_row = ctl_serialize_table[ooa_entry->seridx];
11084
11085         switch (serialize_row[pending_entry->seridx]) {
11086         case CTL_SER_BLOCK:
11087                 return (CTL_ACTION_BLOCK);
11088         case CTL_SER_EXTENT:
11089                 return (ctl_extent_check(ooa_io, pending_io,
11090                     (lun->be_lun && lun->be_lun->serseq == CTL_LUN_SERSEQ_ON)));
11091         case CTL_SER_EXTENTOPT:
11092                 if ((lun->MODE_CTRL.queue_flags & SCP_QUEUE_ALG_MASK) !=
11093                     SCP_QUEUE_ALG_UNRESTRICTED)
11094                         return (ctl_extent_check(ooa_io, pending_io,
11095                             (lun->be_lun &&
11096                              lun->be_lun->serseq == CTL_LUN_SERSEQ_ON)));
11097                 return (CTL_ACTION_PASS);
11098         case CTL_SER_EXTENTSEQ:
11099                 if (lun->be_lun && lun->be_lun->serseq != CTL_LUN_SERSEQ_OFF)
11100                         return (ctl_extent_check_seq(ooa_io, pending_io));
11101                 return (CTL_ACTION_PASS);
11102         case CTL_SER_PASS:
11103                 return (CTL_ACTION_PASS);
11104         case CTL_SER_BLOCKOPT:
11105                 if ((lun->MODE_CTRL.queue_flags & SCP_QUEUE_ALG_MASK) !=
11106                     SCP_QUEUE_ALG_UNRESTRICTED)
11107                         return (CTL_ACTION_BLOCK);
11108                 return (CTL_ACTION_PASS);
11109         case CTL_SER_SKIP:
11110                 return (CTL_ACTION_SKIP);
11111         default:
11112                 panic("%s: Invalid serialization value %d for %d => %d",
11113                     __func__, serialize_row[pending_entry->seridx],
11114                     pending_entry->seridx, ooa_entry->seridx);
11115         }
11116
11117         return (CTL_ACTION_ERROR);
11118 }
11119
11120 /*
11121  * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
11122  * Assumptions:
11123  * - pending_io is generally either incoming, or on the blocked queue
11124  * - starting I/O is the I/O we want to start the check with.
11125  */
11126 static ctl_action
11127 ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
11128               union ctl_io *starting_io)
11129 {
11130         union ctl_io *ooa_io;
11131         ctl_action action;
11132
11133         mtx_assert(&lun->lun_lock, MA_OWNED);
11134
11135         /*
11136          * Run back along the OOA queue, starting with the current
11137          * blocked I/O and going through every I/O before it on the
11138          * queue.  If starting_io is NULL, we'll just end up returning
11139          * CTL_ACTION_PASS.
11140          */
11141         for (ooa_io = starting_io; ooa_io != NULL;
11142              ooa_io = (union ctl_io *)TAILQ_PREV(&ooa_io->io_hdr, ctl_ooaq,
11143              ooa_links)){
11144
11145                 /*
11146                  * This routine just checks to see whether
11147                  * cur_blocked is blocked by ooa_io, which is ahead
11148                  * of it in the queue.  It doesn't queue/dequeue
11149                  * cur_blocked.
11150                  */
11151                 action = ctl_check_for_blockage(lun, pending_io, ooa_io);
11152                 switch (action) {
11153                 case CTL_ACTION_BLOCK:
11154                 case CTL_ACTION_OVERLAP:
11155                 case CTL_ACTION_OVERLAP_TAG:
11156                 case CTL_ACTION_SKIP:
11157                 case CTL_ACTION_ERROR:
11158                         return (action);
11159                         break; /* NOTREACHED */
11160                 case CTL_ACTION_PASS:
11161                         break;
11162                 default:
11163                         panic("%s: Invalid action %d\n", __func__, action);
11164                 }
11165         }
11166
11167         return (CTL_ACTION_PASS);
11168 }
11169
11170 /*
11171  * Assumptions:
11172  * - An I/O has just completed, and has been removed from the per-LUN OOA
11173  *   queue, so some items on the blocked queue may now be unblocked.
11174  */
11175 static int
11176 ctl_check_blocked(struct ctl_lun *lun)
11177 {
11178         struct ctl_softc *softc = lun->ctl_softc;
11179         union ctl_io *cur_blocked, *next_blocked;
11180
11181         mtx_assert(&lun->lun_lock, MA_OWNED);
11182
11183         /*
11184          * Run forward from the head of the blocked queue, checking each
11185          * entry against the I/Os prior to it on the OOA queue to see if
11186          * there is still any blockage.
11187          *
11188          * We cannot use the TAILQ_FOREACH() macro, because it can't deal
11189          * with our removing a variable on it while it is traversing the
11190          * list.
11191          */
11192         for (cur_blocked = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue);
11193              cur_blocked != NULL; cur_blocked = next_blocked) {
11194                 union ctl_io *prev_ooa;
11195                 ctl_action action;
11196
11197                 next_blocked = (union ctl_io *)TAILQ_NEXT(&cur_blocked->io_hdr,
11198                                                           blocked_links);
11199
11200                 prev_ooa = (union ctl_io *)TAILQ_PREV(&cur_blocked->io_hdr,
11201                                                       ctl_ooaq, ooa_links);
11202
11203                 /*
11204                  * If cur_blocked happens to be the first item in the OOA
11205                  * queue now, prev_ooa will be NULL, and the action
11206                  * returned will just be CTL_ACTION_PASS.
11207                  */
11208                 action = ctl_check_ooa(lun, cur_blocked, prev_ooa);
11209
11210                 switch (action) {
11211                 case CTL_ACTION_BLOCK:
11212                         /* Nothing to do here, still blocked */
11213                         break;
11214                 case CTL_ACTION_OVERLAP:
11215                 case CTL_ACTION_OVERLAP_TAG:
11216                         /*
11217                          * This shouldn't happen!  In theory we've already
11218                          * checked this command for overlap...
11219                          */
11220                         break;
11221                 case CTL_ACTION_PASS:
11222                 case CTL_ACTION_SKIP: {
11223                         const struct ctl_cmd_entry *entry;
11224
11225                         /*
11226                          * The skip case shouldn't happen, this transaction
11227                          * should have never made it onto the blocked queue.
11228                          */
11229                         /*
11230                          * This I/O is no longer blocked, we can remove it
11231                          * from the blocked queue.  Since this is a TAILQ
11232                          * (doubly linked list), we can do O(1) removals
11233                          * from any place on the list.
11234                          */
11235                         TAILQ_REMOVE(&lun->blocked_queue, &cur_blocked->io_hdr,
11236                                      blocked_links);
11237                         cur_blocked->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
11238
11239                         if ((softc->ha_mode != CTL_HA_MODE_XFER) &&
11240                             (cur_blocked->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)){
11241                                 /*
11242                                  * Need to send IO back to original side to
11243                                  * run
11244                                  */
11245                                 union ctl_ha_msg msg_info;
11246
11247                                 cur_blocked->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11248                                 msg_info.hdr.original_sc =
11249                                         cur_blocked->io_hdr.original_sc;
11250                                 msg_info.hdr.serializing_sc = cur_blocked;
11251                                 msg_info.hdr.msg_type = CTL_MSG_R2R;
11252                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11253                                     sizeof(msg_info.hdr), M_NOWAIT);
11254                                 break;
11255                         }
11256                         entry = ctl_get_cmd_entry(&cur_blocked->scsiio, NULL);
11257
11258                         /*
11259                          * Check this I/O for LUN state changes that may
11260                          * have happened while this command was blocked.
11261                          * The LUN state may have been changed by a command
11262                          * ahead of us in the queue, so we need to re-check
11263                          * for any states that can be caused by SCSI
11264                          * commands.
11265                          */
11266                         if (ctl_scsiio_lun_check(lun, entry,
11267                                                  &cur_blocked->scsiio) == 0) {
11268                                 cur_blocked->io_hdr.flags |=
11269                                                       CTL_FLAG_IS_WAS_ON_RTR;
11270                                 ctl_enqueue_rtr(cur_blocked);
11271                         } else
11272                                 ctl_done(cur_blocked);
11273                         break;
11274                 }
11275                 default:
11276                         /*
11277                          * This probably shouldn't happen -- we shouldn't
11278                          * get CTL_ACTION_ERROR, or anything else.
11279                          */
11280                         break;
11281                 }
11282         }
11283
11284         return (CTL_RETVAL_COMPLETE);
11285 }
11286
11287 /*
11288  * This routine (with one exception) checks LUN flags that can be set by
11289  * commands ahead of us in the OOA queue.  These flags have to be checked
11290  * when a command initially comes in, and when we pull a command off the
11291  * blocked queue and are preparing to execute it.  The reason we have to
11292  * check these flags for commands on the blocked queue is that the LUN
11293  * state may have been changed by a command ahead of us while we're on the
11294  * blocked queue.
11295  *
11296  * Ordering is somewhat important with these checks, so please pay
11297  * careful attention to the placement of any new checks.
11298  */
11299 static int
11300 ctl_scsiio_lun_check(struct ctl_lun *lun,
11301     const struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
11302 {
11303         struct ctl_softc *softc = lun->ctl_softc;
11304         int retval;
11305         uint32_t residx;
11306
11307         retval = 0;
11308
11309         mtx_assert(&lun->lun_lock, MA_OWNED);
11310
11311         /*
11312          * If this shelf is a secondary shelf controller, we may have to
11313          * reject some commands disallowed by HA mode and link state.
11314          */
11315         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
11316                 if (softc->ha_link == CTL_HA_LINK_OFFLINE &&
11317                     (entry->flags & CTL_CMD_FLAG_OK_ON_UNAVAIL) == 0) {
11318                         ctl_set_lun_unavail(ctsio);
11319                         retval = 1;
11320                         goto bailout;
11321                 }
11322                 if ((lun->flags & CTL_LUN_PEER_SC_PRIMARY) == 0 &&
11323                     (entry->flags & CTL_CMD_FLAG_OK_ON_UNAVAIL) == 0) {
11324                         ctl_set_lun_transit(ctsio);
11325                         retval = 1;
11326                         goto bailout;
11327                 }
11328                 if (softc->ha_mode == CTL_HA_MODE_ACT_STBY &&
11329                     (entry->flags & CTL_CMD_FLAG_OK_ON_STANDBY) == 0) {
11330                         ctl_set_lun_standby(ctsio);
11331                         retval = 1;
11332                         goto bailout;
11333                 }
11334
11335                 /* The rest of checks are only done on executing side */
11336                 if (softc->ha_mode == CTL_HA_MODE_XFER)
11337                         goto bailout;
11338         }
11339
11340         if (entry->pattern & CTL_LUN_PAT_WRITE) {
11341                 if (lun->be_lun &&
11342                     lun->be_lun->flags & CTL_LUN_FLAG_READONLY) {
11343                         ctl_set_hw_write_protected(ctsio);
11344                         retval = 1;
11345                         goto bailout;
11346                 }
11347                 if ((lun->MODE_CTRL.eca_and_aen & SCP_SWP) != 0) {
11348                         ctl_set_sense(ctsio, /*current_error*/ 1,
11349                             /*sense_key*/ SSD_KEY_DATA_PROTECT,
11350                             /*asc*/ 0x27, /*ascq*/ 0x02, SSD_ELEM_NONE);
11351                         retval = 1;
11352                         goto bailout;
11353                 }
11354         }
11355
11356         /*
11357          * Check for a reservation conflict.  If this command isn't allowed
11358          * even on reserved LUNs, and if this initiator isn't the one who
11359          * reserved us, reject the command with a reservation conflict.
11360          */
11361         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11362         if ((lun->flags & CTL_LUN_RESERVED)
11363          && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
11364                 if (lun->res_idx != residx) {
11365                         ctl_set_reservation_conflict(ctsio);
11366                         retval = 1;
11367                         goto bailout;
11368                 }
11369         }
11370
11371         if ((lun->flags & CTL_LUN_PR_RESERVED) == 0 ||
11372             (entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV)) {
11373                 /* No reservation or command is allowed. */;
11374         } else if ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_WRESV) &&
11375             (lun->pr_res_type == SPR_TYPE_WR_EX ||
11376              lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
11377              lun->pr_res_type == SPR_TYPE_WR_EX_AR)) {
11378                 /* The command is allowed for Write Exclusive resv. */;
11379         } else {
11380                 /*
11381                  * if we aren't registered or it's a res holder type
11382                  * reservation and this isn't the res holder then set a
11383                  * conflict.
11384                  */
11385                 if (ctl_get_prkey(lun, residx) == 0 ||
11386                     (residx != lun->pr_res_idx && lun->pr_res_type < 4)) {
11387                         ctl_set_reservation_conflict(ctsio);
11388                         retval = 1;
11389                         goto bailout;
11390                 }
11391         }
11392
11393         if ((entry->flags & CTL_CMD_FLAG_OK_ON_NO_MEDIA) == 0) {
11394                 if (lun->flags & CTL_LUN_EJECTED)
11395                         ctl_set_lun_ejected(ctsio);
11396                 else if (lun->flags & CTL_LUN_NO_MEDIA) {
11397                         if (lun->flags & CTL_LUN_REMOVABLE)
11398                                 ctl_set_lun_no_media(ctsio);
11399                         else
11400                                 ctl_set_lun_int_reqd(ctsio);
11401                 } else if (lun->flags & CTL_LUN_STOPPED)
11402                         ctl_set_lun_stopped(ctsio);
11403                 else
11404                         goto bailout;
11405                 retval = 1;
11406                 goto bailout;
11407         }
11408
11409 bailout:
11410         return (retval);
11411 }
11412
11413 static void
11414 ctl_failover_io(union ctl_io *io, int have_lock)
11415 {
11416         ctl_set_busy(&io->scsiio);
11417         ctl_done(io);
11418 }
11419
11420 static void
11421 ctl_failover_lun(union ctl_io *rio)
11422 {
11423         struct ctl_softc *softc = control_softc;
11424         struct ctl_lun *lun;
11425         struct ctl_io_hdr *io, *next_io;
11426         uint32_t targ_lun;
11427
11428         targ_lun = rio->io_hdr.nexus.targ_mapped_lun;
11429         CTL_DEBUG_PRINT(("FAILOVER for lun %ju\n", targ_lun));
11430
11431         /* Find and lock the LUN. */
11432         mtx_lock(&softc->ctl_lock);
11433         if ((targ_lun < CTL_MAX_LUNS) &&
11434             ((lun = softc->ctl_luns[targ_lun]) != NULL)) {
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         } else {
11442                 mtx_unlock(&softc->ctl_lock);
11443                 return;
11444         }
11445
11446         if (softc->ha_mode == CTL_HA_MODE_XFER) {
11447                 TAILQ_FOREACH_SAFE(io, &lun->ooa_queue, ooa_links, next_io) {
11448                         /* We are master */
11449                         if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
11450                                 if (io->flags & CTL_FLAG_IO_ACTIVE) {
11451                                         io->flags |= CTL_FLAG_ABORT;
11452                                         io->flags |= CTL_FLAG_FAILOVER;
11453                                 } else { /* This can be only due to DATAMOVE */
11454                                         io->msg_type = CTL_MSG_DATAMOVE_DONE;
11455                                         io->flags &= ~CTL_FLAG_DMA_INPROG;
11456                                         io->flags |= CTL_FLAG_IO_ACTIVE;
11457                                         io->port_status = 31340;
11458                                         ctl_enqueue_isc((union ctl_io *)io);
11459                                 }
11460                         }
11461                         /* We are slave */
11462                         if (io->flags & CTL_FLAG_SENT_2OTHER_SC) {
11463                                 io->flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11464                                 if (io->flags & CTL_FLAG_IO_ACTIVE) {
11465                                         io->flags |= CTL_FLAG_FAILOVER;
11466                                 } else {
11467                                         ctl_set_busy(&((union ctl_io *)io)->
11468                                             scsiio);
11469                                         ctl_done((union ctl_io *)io);
11470                                 }
11471                         }
11472                 }
11473         } else { /* SERIALIZE modes */
11474                 TAILQ_FOREACH_SAFE(io, &lun->blocked_queue, blocked_links,
11475                     next_io) {
11476                         /* We are master */
11477                         if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
11478                                 TAILQ_REMOVE(&lun->blocked_queue, io,
11479                                     blocked_links);
11480                                 io->flags &= ~CTL_FLAG_BLOCKED;
11481                                 TAILQ_REMOVE(&lun->ooa_queue, io, ooa_links);
11482                                 ctl_free_io((union ctl_io *)io);
11483                         }
11484                 }
11485                 TAILQ_FOREACH_SAFE(io, &lun->ooa_queue, ooa_links, next_io) {
11486                         /* We are master */
11487                         if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
11488                                 TAILQ_REMOVE(&lun->ooa_queue, io, ooa_links);
11489                                 ctl_free_io((union ctl_io *)io);
11490                         }
11491                         /* We are slave */
11492                         if (io->flags & CTL_FLAG_SENT_2OTHER_SC) {
11493                                 io->flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11494                                 if (!(io->flags & CTL_FLAG_IO_ACTIVE)) {
11495                                         ctl_set_busy(&((union ctl_io *)io)->
11496                                             scsiio);
11497                                         ctl_done((union ctl_io *)io);
11498                                 }
11499                         }
11500                 }
11501                 ctl_check_blocked(lun);
11502         }
11503         mtx_unlock(&lun->lun_lock);
11504 }
11505
11506 static int
11507 ctl_scsiio_precheck(struct ctl_softc *softc, struct ctl_scsiio *ctsio)
11508 {
11509         struct ctl_lun *lun;
11510         const struct ctl_cmd_entry *entry;
11511         uint32_t initidx, targ_lun;
11512         int retval;
11513
11514         retval = 0;
11515
11516         lun = NULL;
11517
11518         targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
11519         if ((targ_lun < CTL_MAX_LUNS)
11520          && ((lun = softc->ctl_luns[targ_lun]) != NULL)) {
11521                 /*
11522                  * If the LUN is invalid, pretend that it doesn't exist.
11523                  * It will go away as soon as all pending I/O has been
11524                  * completed.
11525                  */
11526                 mtx_lock(&lun->lun_lock);
11527                 if (lun->flags & CTL_LUN_DISABLED) {
11528                         mtx_unlock(&lun->lun_lock);
11529                         lun = NULL;
11530                         ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
11531                         ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
11532                 } else {
11533                         ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = lun;
11534                         ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr =
11535                                 lun->be_lun;
11536
11537                         /*
11538                          * Every I/O goes into the OOA queue for a
11539                          * particular LUN, and stays there until completion.
11540                          */
11541 #ifdef CTL_TIME_IO
11542                         if (TAILQ_EMPTY(&lun->ooa_queue)) {
11543                                 lun->idle_time += getsbinuptime() -
11544                                     lun->last_busy;
11545                         }
11546 #endif
11547                         TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr,
11548                             ooa_links);
11549                 }
11550         } else {
11551                 ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
11552                 ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
11553         }
11554
11555         /* Get command entry and return error if it is unsuppotyed. */
11556         entry = ctl_validate_command(ctsio);
11557         if (entry == NULL) {
11558                 if (lun)
11559                         mtx_unlock(&lun->lun_lock);
11560                 return (retval);
11561         }
11562
11563         ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
11564         ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
11565
11566         /*
11567          * Check to see whether we can send this command to LUNs that don't
11568          * exist.  This should pretty much only be the case for inquiry
11569          * and request sense.  Further checks, below, really require having
11570          * a LUN, so we can't really check the command anymore.  Just put
11571          * it on the rtr queue.
11572          */
11573         if (lun == NULL) {
11574                 if (entry->flags & CTL_CMD_FLAG_OK_ON_NO_LUN) {
11575                         ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11576                         ctl_enqueue_rtr((union ctl_io *)ctsio);
11577                         return (retval);
11578                 }
11579
11580                 ctl_set_unsupported_lun(ctsio);
11581                 ctl_done((union ctl_io *)ctsio);
11582                 CTL_DEBUG_PRINT(("ctl_scsiio_precheck: bailing out due to invalid LUN\n"));
11583                 return (retval);
11584         } else {
11585                 /*
11586                  * Make sure we support this particular command on this LUN.
11587                  * e.g., we don't support writes to the control LUN.
11588                  */
11589                 if (!ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
11590                         mtx_unlock(&lun->lun_lock);
11591                         ctl_set_invalid_opcode(ctsio);
11592                         ctl_done((union ctl_io *)ctsio);
11593                         return (retval);
11594                 }
11595         }
11596
11597         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11598
11599 #ifdef CTL_WITH_CA
11600         /*
11601          * If we've got a request sense, it'll clear the contingent
11602          * allegiance condition.  Otherwise, if we have a CA condition for
11603          * this initiator, clear it, because it sent down a command other
11604          * than request sense.
11605          */
11606         if ((ctsio->cdb[0] != REQUEST_SENSE)
11607          && (ctl_is_set(lun->have_ca, initidx)))
11608                 ctl_clear_mask(lun->have_ca, initidx);
11609 #endif
11610
11611         /*
11612          * If the command has this flag set, it handles its own unit
11613          * attention reporting, we shouldn't do anything.  Otherwise we
11614          * check for any pending unit attentions, and send them back to the
11615          * initiator.  We only do this when a command initially comes in,
11616          * not when we pull it off the blocked queue.
11617          *
11618          * According to SAM-3, section 5.3.2, the order that things get
11619          * presented back to the host is basically unit attentions caused
11620          * by some sort of reset event, busy status, reservation conflicts
11621          * or task set full, and finally any other status.
11622          *
11623          * One issue here is that some of the unit attentions we report
11624          * don't fall into the "reset" category (e.g. "reported luns data
11625          * has changed").  So reporting it here, before the reservation
11626          * check, may be technically wrong.  I guess the only thing to do
11627          * would be to check for and report the reset events here, and then
11628          * check for the other unit attention types after we check for a
11629          * reservation conflict.
11630          *
11631          * XXX KDM need to fix this
11632          */
11633         if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
11634                 ctl_ua_type ua_type;
11635
11636                 ua_type = ctl_build_ua(lun, initidx, &ctsio->sense_data,
11637                     SSD_TYPE_NONE);
11638                 if (ua_type != CTL_UA_NONE) {
11639                         mtx_unlock(&lun->lun_lock);
11640                         ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
11641                         ctsio->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
11642                         ctsio->sense_len = SSD_FULL_SIZE;
11643                         ctl_done((union ctl_io *)ctsio);
11644                         return (retval);
11645                 }
11646         }
11647
11648
11649         if (ctl_scsiio_lun_check(lun, entry, ctsio) != 0) {
11650                 mtx_unlock(&lun->lun_lock);
11651                 ctl_done((union ctl_io *)ctsio);
11652                 return (retval);
11653         }
11654
11655         /*
11656          * XXX CHD this is where we want to send IO to other side if
11657          * this LUN is secondary on this SC. We will need to make a copy
11658          * of the IO and flag the IO on this side as SENT_2OTHER and the flag
11659          * the copy we send as FROM_OTHER.
11660          * We also need to stuff the address of the original IO so we can
11661          * find it easily. Something similar will need be done on the other
11662          * side so when we are done we can find the copy.
11663          */
11664         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
11665             (lun->flags & CTL_LUN_PEER_SC_PRIMARY) != 0 &&
11666             (entry->flags & CTL_CMD_FLAG_RUN_HERE) == 0) {
11667                 union ctl_ha_msg msg_info;
11668                 int isc_retval;
11669
11670                 ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11671                 ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11672                 mtx_unlock(&lun->lun_lock);
11673
11674                 msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
11675                 msg_info.hdr.original_sc = (union ctl_io *)ctsio;
11676                 msg_info.hdr.serializing_sc = NULL;
11677                 msg_info.hdr.nexus = ctsio->io_hdr.nexus;
11678                 msg_info.scsi.tag_num = ctsio->tag_num;
11679                 msg_info.scsi.tag_type = ctsio->tag_type;
11680                 msg_info.scsi.cdb_len = ctsio->cdb_len;
11681                 memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
11682
11683                 if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11684                     sizeof(msg_info.scsi) - sizeof(msg_info.scsi.sense_data),
11685                     M_WAITOK)) > CTL_HA_STATUS_SUCCESS) {
11686                         ctl_set_busy(ctsio);
11687                         ctl_done((union ctl_io *)ctsio);
11688                         return (retval);
11689                 }
11690                 return (retval);
11691         }
11692
11693         switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
11694                               (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr,
11695                               ctl_ooaq, ooa_links))) {
11696         case CTL_ACTION_BLOCK:
11697                 ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
11698                 TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
11699                                   blocked_links);
11700                 mtx_unlock(&lun->lun_lock);
11701                 return (retval);
11702         case CTL_ACTION_PASS:
11703         case CTL_ACTION_SKIP:
11704                 ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11705                 mtx_unlock(&lun->lun_lock);
11706                 ctl_enqueue_rtr((union ctl_io *)ctsio);
11707                 break;
11708         case CTL_ACTION_OVERLAP:
11709                 mtx_unlock(&lun->lun_lock);
11710                 ctl_set_overlapped_cmd(ctsio);
11711                 ctl_done((union ctl_io *)ctsio);
11712                 break;
11713         case CTL_ACTION_OVERLAP_TAG:
11714                 mtx_unlock(&lun->lun_lock);
11715                 ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
11716                 ctl_done((union ctl_io *)ctsio);
11717                 break;
11718         case CTL_ACTION_ERROR:
11719         default:
11720                 mtx_unlock(&lun->lun_lock);
11721                 ctl_set_internal_failure(ctsio,
11722                                          /*sks_valid*/ 0,
11723                                          /*retry_count*/ 0);
11724                 ctl_done((union ctl_io *)ctsio);
11725                 break;
11726         }
11727         return (retval);
11728 }
11729
11730 const struct ctl_cmd_entry *
11731 ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa)
11732 {
11733         const struct ctl_cmd_entry *entry;
11734         int service_action;
11735
11736         entry = &ctl_cmd_table[ctsio->cdb[0]];
11737         if (sa)
11738                 *sa = ((entry->flags & CTL_CMD_FLAG_SA5) != 0);
11739         if (entry->flags & CTL_CMD_FLAG_SA5) {
11740                 service_action = ctsio->cdb[1] & SERVICE_ACTION_MASK;
11741                 entry = &((const struct ctl_cmd_entry *)
11742                     entry->execute)[service_action];
11743         }
11744         return (entry);
11745 }
11746
11747 const struct ctl_cmd_entry *
11748 ctl_validate_command(struct ctl_scsiio *ctsio)
11749 {
11750         const struct ctl_cmd_entry *entry;
11751         int i, sa;
11752         uint8_t diff;
11753
11754         entry = ctl_get_cmd_entry(ctsio, &sa);
11755         if (entry->execute == NULL) {
11756                 if (sa)
11757                         ctl_set_invalid_field(ctsio,
11758                                               /*sks_valid*/ 1,
11759                                               /*command*/ 1,
11760                                               /*field*/ 1,
11761                                               /*bit_valid*/ 1,
11762                                               /*bit*/ 4);
11763                 else
11764                         ctl_set_invalid_opcode(ctsio);
11765                 ctl_done((union ctl_io *)ctsio);
11766                 return (NULL);
11767         }
11768         KASSERT(entry->length > 0,
11769             ("Not defined length for command 0x%02x/0x%02x",
11770              ctsio->cdb[0], ctsio->cdb[1]));
11771         for (i = 1; i < entry->length; i++) {
11772                 diff = ctsio->cdb[i] & ~entry->usage[i - 1];
11773                 if (diff == 0)
11774                         continue;
11775                 ctl_set_invalid_field(ctsio,
11776                                       /*sks_valid*/ 1,
11777                                       /*command*/ 1,
11778                                       /*field*/ i,
11779                                       /*bit_valid*/ 1,
11780                                       /*bit*/ fls(diff) - 1);
11781                 ctl_done((union ctl_io *)ctsio);
11782                 return (NULL);
11783         }
11784         return (entry);
11785 }
11786
11787 static int
11788 ctl_cmd_applicable(uint8_t lun_type, const struct ctl_cmd_entry *entry)
11789 {
11790
11791         switch (lun_type) {
11792         case T_DIRECT:
11793                 if ((entry->flags & CTL_CMD_FLAG_OK_ON_DIRECT) == 0)
11794                         return (0);
11795                 break;
11796         case T_PROCESSOR:
11797                 if ((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0)
11798                         return (0);
11799                 break;
11800         case T_CDROM:
11801                 if ((entry->flags & CTL_CMD_FLAG_OK_ON_CDROM) == 0)
11802                         return (0);
11803                 break;
11804         default:
11805                 return (0);
11806         }
11807         return (1);
11808 }
11809
11810 static int
11811 ctl_scsiio(struct ctl_scsiio *ctsio)
11812 {
11813         int retval;
11814         const struct ctl_cmd_entry *entry;
11815
11816         retval = CTL_RETVAL_COMPLETE;
11817
11818         CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
11819
11820         entry = ctl_get_cmd_entry(ctsio, NULL);
11821
11822         /*
11823          * If this I/O has been aborted, just send it straight to
11824          * ctl_done() without executing it.
11825          */
11826         if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
11827                 ctl_done((union ctl_io *)ctsio);
11828                 goto bailout;
11829         }
11830
11831         /*
11832          * All the checks should have been handled by ctl_scsiio_precheck().
11833          * We should be clear now to just execute the I/O.
11834          */
11835         retval = entry->execute(ctsio);
11836
11837 bailout:
11838         return (retval);
11839 }
11840
11841 /*
11842  * Since we only implement one target right now, a bus reset simply resets
11843  * our single target.
11844  */
11845 static int
11846 ctl_bus_reset(struct ctl_softc *softc, union ctl_io *io)
11847 {
11848         return(ctl_target_reset(softc, io, CTL_UA_BUS_RESET));
11849 }
11850
11851 static int
11852 ctl_target_reset(struct ctl_softc *softc, union ctl_io *io,
11853                  ctl_ua_type ua_type)
11854 {
11855         struct ctl_port *port;
11856         struct ctl_lun *lun;
11857         int retval;
11858
11859         if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11860                 union ctl_ha_msg msg_info;
11861
11862                 msg_info.hdr.nexus = io->io_hdr.nexus;
11863                 if (ua_type==CTL_UA_TARG_RESET)
11864                         msg_info.task.task_action = CTL_TASK_TARGET_RESET;
11865                 else
11866                         msg_info.task.task_action = CTL_TASK_BUS_RESET;
11867                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11868                 msg_info.hdr.original_sc = NULL;
11869                 msg_info.hdr.serializing_sc = NULL;
11870                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11871                     sizeof(msg_info.task), M_WAITOK);
11872         }
11873         retval = 0;
11874
11875         mtx_lock(&softc->ctl_lock);
11876         port = ctl_io_port(&io->io_hdr);
11877         STAILQ_FOREACH(lun, &softc->lun_list, links) {
11878                 if (port != NULL &&
11879                     ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
11880                         continue;
11881                 retval += ctl_do_lun_reset(lun, io, ua_type);
11882         }
11883         mtx_unlock(&softc->ctl_lock);
11884         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11885         return (retval);
11886 }
11887
11888 /*
11889  * The LUN should always be set.  The I/O is optional, and is used to
11890  * distinguish between I/Os sent by this initiator, and by other
11891  * initiators.  We set unit attention for initiators other than this one.
11892  * SAM-3 is vague on this point.  It does say that a unit attention should
11893  * be established for other initiators when a LUN is reset (see section
11894  * 5.7.3), but it doesn't specifically say that the unit attention should
11895  * be established for this particular initiator when a LUN is reset.  Here
11896  * is the relevant text, from SAM-3 rev 8:
11897  *
11898  * 5.7.2 When a SCSI initiator port aborts its own tasks
11899  *
11900  * When a SCSI initiator port causes its own task(s) to be aborted, no
11901  * notification that the task(s) have been aborted shall be returned to
11902  * the SCSI initiator port other than the completion response for the
11903  * command or task management function action that caused the task(s) to
11904  * be aborted and notification(s) associated with related effects of the
11905  * action (e.g., a reset unit attention condition).
11906  *
11907  * XXX KDM for now, we're setting unit attention for all initiators.
11908  */
11909 static int
11910 ctl_do_lun_reset(struct ctl_lun *lun, union ctl_io *io, ctl_ua_type ua_type)
11911 {
11912         union ctl_io *xio;
11913 #if 0
11914         uint32_t initidx;
11915 #endif
11916         int i;
11917
11918         mtx_lock(&lun->lun_lock);
11919         /*
11920          * Run through the OOA queue and abort each I/O.
11921          */
11922         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11923              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11924                 xio->io_hdr.flags |= CTL_FLAG_ABORT | CTL_FLAG_ABORT_STATUS;
11925         }
11926
11927         /*
11928          * This version sets unit attention for every
11929          */
11930 #if 0
11931         initidx = ctl_get_initindex(&io->io_hdr.nexus);
11932         ctl_est_ua_all(lun, initidx, ua_type);
11933 #else
11934         ctl_est_ua_all(lun, -1, ua_type);
11935 #endif
11936
11937         /*
11938          * A reset (any kind, really) clears reservations established with
11939          * RESERVE/RELEASE.  It does not clear reservations established
11940          * with PERSISTENT RESERVE OUT, but we don't support that at the
11941          * moment anyway.  See SPC-2, section 5.6.  SPC-3 doesn't address
11942          * reservations made with the RESERVE/RELEASE commands, because
11943          * those commands are obsolete in SPC-3.
11944          */
11945         lun->flags &= ~CTL_LUN_RESERVED;
11946
11947 #ifdef CTL_WITH_CA
11948         for (i = 0; i < CTL_MAX_INITIATORS; i++)
11949                 ctl_clear_mask(lun->have_ca, i);
11950 #endif
11951         lun->prevent_count = 0;
11952         for (i = 0; i < CTL_MAX_INITIATORS; i++)
11953                 ctl_clear_mask(lun->prevent, i);
11954         mtx_unlock(&lun->lun_lock);
11955
11956         return (0);
11957 }
11958
11959 static int
11960 ctl_lun_reset(struct ctl_softc *softc, union ctl_io *io)
11961 {
11962         struct ctl_lun *lun;
11963         uint32_t targ_lun;
11964         int retval;
11965
11966         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11967         mtx_lock(&softc->ctl_lock);
11968         if ((targ_lun >= CTL_MAX_LUNS) ||
11969             (lun = softc->ctl_luns[targ_lun]) == NULL) {
11970                 mtx_unlock(&softc->ctl_lock);
11971                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
11972                 return (1);
11973         }
11974         retval = ctl_do_lun_reset(lun, io, CTL_UA_LUN_RESET);
11975         mtx_unlock(&softc->ctl_lock);
11976         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11977
11978         if ((io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) == 0) {
11979                 union ctl_ha_msg msg_info;
11980
11981                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11982                 msg_info.hdr.nexus = io->io_hdr.nexus;
11983                 msg_info.task.task_action = CTL_TASK_LUN_RESET;
11984                 msg_info.hdr.original_sc = NULL;
11985                 msg_info.hdr.serializing_sc = NULL;
11986                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11987                     sizeof(msg_info.task), M_WAITOK);
11988         }
11989         return (retval);
11990 }
11991
11992 static void
11993 ctl_abort_tasks_lun(struct ctl_lun *lun, uint32_t targ_port, uint32_t init_id,
11994     int other_sc)
11995 {
11996         union ctl_io *xio;
11997
11998         mtx_assert(&lun->lun_lock, MA_OWNED);
11999
12000         /*
12001          * Run through the OOA queue and attempt to find the given I/O.
12002          * The target port, initiator ID, tag type and tag number have to
12003          * match the values that we got from the initiator.  If we have an
12004          * untagged command to abort, simply abort the first untagged command
12005          * we come to.  We only allow one untagged command at a time of course.
12006          */
12007         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12008              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12009
12010                 if ((targ_port == UINT32_MAX ||
12011                      targ_port == xio->io_hdr.nexus.targ_port) &&
12012                     (init_id == UINT32_MAX ||
12013                      init_id == xio->io_hdr.nexus.initid)) {
12014                         if (targ_port != xio->io_hdr.nexus.targ_port ||
12015                             init_id != xio->io_hdr.nexus.initid)
12016                                 xio->io_hdr.flags |= CTL_FLAG_ABORT_STATUS;
12017                         xio->io_hdr.flags |= CTL_FLAG_ABORT;
12018                         if (!other_sc && !(lun->flags & CTL_LUN_PRIMARY_SC)) {
12019                                 union ctl_ha_msg msg_info;
12020
12021                                 msg_info.hdr.nexus = xio->io_hdr.nexus;
12022                                 msg_info.task.task_action = CTL_TASK_ABORT_TASK;
12023                                 msg_info.task.tag_num = xio->scsiio.tag_num;
12024                                 msg_info.task.tag_type = xio->scsiio.tag_type;
12025                                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
12026                                 msg_info.hdr.original_sc = NULL;
12027                                 msg_info.hdr.serializing_sc = NULL;
12028                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
12029                                     sizeof(msg_info.task), M_NOWAIT);
12030                         }
12031                 }
12032         }
12033 }
12034
12035 static int
12036 ctl_abort_task_set(union ctl_io *io)
12037 {
12038         struct ctl_softc *softc = control_softc;
12039         struct ctl_lun *lun;
12040         uint32_t targ_lun;
12041
12042         /*
12043          * Look up the LUN.
12044          */
12045         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12046         mtx_lock(&softc->ctl_lock);
12047         if ((targ_lun >= CTL_MAX_LUNS) ||
12048             (lun = softc->ctl_luns[targ_lun]) == NULL) {
12049                 mtx_unlock(&softc->ctl_lock);
12050                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
12051                 return (1);
12052         }
12053
12054         mtx_lock(&lun->lun_lock);
12055         mtx_unlock(&softc->ctl_lock);
12056         if (io->taskio.task_action == CTL_TASK_ABORT_TASK_SET) {
12057                 ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
12058                     io->io_hdr.nexus.initid,
12059                     (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12060         } else { /* CTL_TASK_CLEAR_TASK_SET */
12061                 ctl_abort_tasks_lun(lun, UINT32_MAX, UINT32_MAX,
12062                     (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12063         }
12064         mtx_unlock(&lun->lun_lock);
12065         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12066         return (0);
12067 }
12068
12069 static int
12070 ctl_i_t_nexus_reset(union ctl_io *io)
12071 {
12072         struct ctl_softc *softc = control_softc;
12073         struct ctl_lun *lun;
12074         uint32_t initidx;
12075
12076         if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
12077                 union ctl_ha_msg msg_info;
12078
12079                 msg_info.hdr.nexus = io->io_hdr.nexus;
12080                 msg_info.task.task_action = CTL_TASK_I_T_NEXUS_RESET;
12081                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
12082                 msg_info.hdr.original_sc = NULL;
12083                 msg_info.hdr.serializing_sc = NULL;
12084                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
12085                     sizeof(msg_info.task), M_WAITOK);
12086         }
12087
12088         initidx = ctl_get_initindex(&io->io_hdr.nexus);
12089         mtx_lock(&softc->ctl_lock);
12090         STAILQ_FOREACH(lun, &softc->lun_list, links) {
12091                 mtx_lock(&lun->lun_lock);
12092                 ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
12093                     io->io_hdr.nexus.initid, 1);
12094 #ifdef CTL_WITH_CA
12095                 ctl_clear_mask(lun->have_ca, initidx);
12096 #endif
12097                 if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == initidx))
12098                         lun->flags &= ~CTL_LUN_RESERVED;
12099                 if (ctl_is_set(lun->prevent, initidx)) {
12100                         ctl_clear_mask(lun->prevent, initidx);
12101                         lun->prevent_count--;
12102                 }
12103                 ctl_est_ua(lun, initidx, CTL_UA_I_T_NEXUS_LOSS);
12104                 mtx_unlock(&lun->lun_lock);
12105         }
12106         mtx_unlock(&softc->ctl_lock);
12107         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12108         return (0);
12109 }
12110
12111 static int
12112 ctl_abort_task(union ctl_io *io)
12113 {
12114         union ctl_io *xio;
12115         struct ctl_lun *lun;
12116         struct ctl_softc *softc;
12117 #if 0
12118         struct sbuf sb;
12119         char printbuf[128];
12120 #endif
12121         int found;
12122         uint32_t targ_lun;
12123
12124         softc = control_softc;
12125         found = 0;
12126
12127         /*
12128          * Look up the LUN.
12129          */
12130         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12131         mtx_lock(&softc->ctl_lock);
12132         if ((targ_lun >= CTL_MAX_LUNS) ||
12133             (lun = softc->ctl_luns[targ_lun]) == NULL) {
12134                 mtx_unlock(&softc->ctl_lock);
12135                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
12136                 return (1);
12137         }
12138
12139 #if 0
12140         printf("ctl_abort_task: called for lun %lld, tag %d type %d\n",
12141                lun->lun, io->taskio.tag_num, io->taskio.tag_type);
12142 #endif
12143
12144         mtx_lock(&lun->lun_lock);
12145         mtx_unlock(&softc->ctl_lock);
12146         /*
12147          * Run through the OOA queue and attempt to find the given I/O.
12148          * The target port, initiator ID, tag type and tag number have to
12149          * match the values that we got from the initiator.  If we have an
12150          * untagged command to abort, simply abort the first untagged command
12151          * we come to.  We only allow one untagged command at a time of course.
12152          */
12153         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12154              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12155 #if 0
12156                 sbuf_new(&sb, printbuf, sizeof(printbuf), SBUF_FIXEDLEN);
12157
12158                 sbuf_printf(&sb, "LUN %lld tag %d type %d%s%s%s%s: ",
12159                             lun->lun, xio->scsiio.tag_num,
12160                             xio->scsiio.tag_type,
12161                             (xio->io_hdr.blocked_links.tqe_prev
12162                             == NULL) ? "" : " BLOCKED",
12163                             (xio->io_hdr.flags &
12164                             CTL_FLAG_DMA_INPROG) ? " DMA" : "",
12165                             (xio->io_hdr.flags &
12166                             CTL_FLAG_ABORT) ? " ABORT" : "",
12167                             (xio->io_hdr.flags &
12168                             CTL_FLAG_IS_WAS_ON_RTR ? " RTR" : ""));
12169                 ctl_scsi_command_string(&xio->scsiio, NULL, &sb);
12170                 sbuf_finish(&sb);
12171                 printf("%s\n", sbuf_data(&sb));
12172 #endif
12173
12174                 if ((xio->io_hdr.nexus.targ_port != io->io_hdr.nexus.targ_port)
12175                  || (xio->io_hdr.nexus.initid != io->io_hdr.nexus.initid)
12176                  || (xio->io_hdr.flags & CTL_FLAG_ABORT))
12177                         continue;
12178
12179                 /*
12180                  * If the abort says that the task is untagged, the
12181                  * task in the queue must be untagged.  Otherwise,
12182                  * we just check to see whether the tag numbers
12183                  * match.  This is because the QLogic firmware
12184                  * doesn't pass back the tag type in an abort
12185                  * request.
12186                  */
12187 #if 0
12188                 if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
12189                   && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
12190                  || (xio->scsiio.tag_num == io->taskio.tag_num))
12191 #endif
12192                 /*
12193                  * XXX KDM we've got problems with FC, because it
12194                  * doesn't send down a tag type with aborts.  So we
12195                  * can only really go by the tag number...
12196                  * This may cause problems with parallel SCSI.
12197                  * Need to figure that out!!
12198                  */
12199                 if (xio->scsiio.tag_num == io->taskio.tag_num) {
12200                         xio->io_hdr.flags |= CTL_FLAG_ABORT;
12201                         found = 1;
12202                         if ((io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) == 0 &&
12203                             !(lun->flags & CTL_LUN_PRIMARY_SC)) {
12204                                 union ctl_ha_msg msg_info;
12205
12206                                 msg_info.hdr.nexus = io->io_hdr.nexus;
12207                                 msg_info.task.task_action = CTL_TASK_ABORT_TASK;
12208                                 msg_info.task.tag_num = io->taskio.tag_num;
12209                                 msg_info.task.tag_type = io->taskio.tag_type;
12210                                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
12211                                 msg_info.hdr.original_sc = NULL;
12212                                 msg_info.hdr.serializing_sc = NULL;
12213 #if 0
12214                                 printf("Sent Abort to other side\n");
12215 #endif
12216                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
12217                                     sizeof(msg_info.task), M_NOWAIT);
12218                         }
12219 #if 0
12220                         printf("ctl_abort_task: found I/O to abort\n");
12221 #endif
12222                 }
12223         }
12224         mtx_unlock(&lun->lun_lock);
12225
12226         if (found == 0) {
12227                 /*
12228                  * This isn't really an error.  It's entirely possible for
12229                  * the abort and command completion to cross on the wire.
12230                  * This is more of an informative/diagnostic error.
12231                  */
12232 #if 0
12233                 printf("ctl_abort_task: ABORT sent for nonexistent I/O: "
12234                        "%u:%u:%u tag %d type %d\n",
12235                        io->io_hdr.nexus.initid,
12236                        io->io_hdr.nexus.targ_port,
12237                        io->io_hdr.nexus.targ_lun, io->taskio.tag_num,
12238                        io->taskio.tag_type);
12239 #endif
12240         }
12241         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12242         return (0);
12243 }
12244
12245 static int
12246 ctl_query_task(union ctl_io *io, int task_set)
12247 {
12248         union ctl_io *xio;
12249         struct ctl_lun *lun;
12250         struct ctl_softc *softc;
12251         int found = 0;
12252         uint32_t targ_lun;
12253
12254         softc = control_softc;
12255         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12256         mtx_lock(&softc->ctl_lock);
12257         if ((targ_lun >= CTL_MAX_LUNS) ||
12258             (lun = softc->ctl_luns[targ_lun]) == NULL) {
12259                 mtx_unlock(&softc->ctl_lock);
12260                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
12261                 return (1);
12262         }
12263         mtx_lock(&lun->lun_lock);
12264         mtx_unlock(&softc->ctl_lock);
12265         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12266              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12267
12268                 if ((xio->io_hdr.nexus.targ_port != io->io_hdr.nexus.targ_port)
12269                  || (xio->io_hdr.nexus.initid != io->io_hdr.nexus.initid)
12270                  || (xio->io_hdr.flags & CTL_FLAG_ABORT))
12271                         continue;
12272
12273                 if (task_set || xio->scsiio.tag_num == io->taskio.tag_num) {
12274                         found = 1;
12275                         break;
12276                 }
12277         }
12278         mtx_unlock(&lun->lun_lock);
12279         if (found)
12280                 io->taskio.task_status = CTL_TASK_FUNCTION_SUCCEEDED;
12281         else
12282                 io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12283         return (0);
12284 }
12285
12286 static int
12287 ctl_query_async_event(union ctl_io *io)
12288 {
12289         struct ctl_lun *lun;
12290         struct ctl_softc *softc;
12291         ctl_ua_type ua;
12292         uint32_t targ_lun, initidx;
12293
12294         softc = control_softc;
12295         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12296         mtx_lock(&softc->ctl_lock);
12297         if ((targ_lun >= CTL_MAX_LUNS) ||
12298             (lun = softc->ctl_luns[targ_lun]) == NULL) {
12299                 mtx_unlock(&softc->ctl_lock);
12300                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
12301                 return (1);
12302         }
12303         mtx_lock(&lun->lun_lock);
12304         mtx_unlock(&softc->ctl_lock);
12305         initidx = ctl_get_initindex(&io->io_hdr.nexus);
12306         ua = ctl_build_qae(lun, initidx, io->taskio.task_resp);
12307         mtx_unlock(&lun->lun_lock);
12308         if (ua != CTL_UA_NONE)
12309                 io->taskio.task_status = CTL_TASK_FUNCTION_SUCCEEDED;
12310         else
12311                 io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12312         return (0);
12313 }
12314
12315 static void
12316 ctl_run_task(union ctl_io *io)
12317 {
12318         struct ctl_softc *softc = control_softc;
12319         int retval = 1;
12320
12321         CTL_DEBUG_PRINT(("ctl_run_task\n"));
12322         KASSERT(io->io_hdr.io_type == CTL_IO_TASK,
12323             ("ctl_run_task: Unextected io_type %d\n", io->io_hdr.io_type));
12324         io->taskio.task_status = CTL_TASK_FUNCTION_NOT_SUPPORTED;
12325         bzero(io->taskio.task_resp, sizeof(io->taskio.task_resp));
12326         switch (io->taskio.task_action) {
12327         case CTL_TASK_ABORT_TASK:
12328                 retval = ctl_abort_task(io);
12329                 break;
12330         case CTL_TASK_ABORT_TASK_SET:
12331         case CTL_TASK_CLEAR_TASK_SET:
12332                 retval = ctl_abort_task_set(io);
12333                 break;
12334         case CTL_TASK_CLEAR_ACA:
12335                 break;
12336         case CTL_TASK_I_T_NEXUS_RESET:
12337                 retval = ctl_i_t_nexus_reset(io);
12338                 break;
12339         case CTL_TASK_LUN_RESET:
12340                 retval = ctl_lun_reset(softc, io);
12341                 break;
12342         case CTL_TASK_TARGET_RESET:
12343                 retval = ctl_target_reset(softc, io, CTL_UA_TARG_RESET);
12344                 break;
12345         case CTL_TASK_BUS_RESET:
12346                 retval = ctl_bus_reset(softc, io);
12347                 break;
12348         case CTL_TASK_PORT_LOGIN:
12349                 break;
12350         case CTL_TASK_PORT_LOGOUT:
12351                 break;
12352         case CTL_TASK_QUERY_TASK:
12353                 retval = ctl_query_task(io, 0);
12354                 break;
12355         case CTL_TASK_QUERY_TASK_SET:
12356                 retval = ctl_query_task(io, 1);
12357                 break;
12358         case CTL_TASK_QUERY_ASYNC_EVENT:
12359                 retval = ctl_query_async_event(io);
12360                 break;
12361         default:
12362                 printf("%s: got unknown task management event %d\n",
12363                        __func__, io->taskio.task_action);
12364                 break;
12365         }
12366         if (retval == 0)
12367                 io->io_hdr.status = CTL_SUCCESS;
12368         else
12369                 io->io_hdr.status = CTL_ERROR;
12370         ctl_done(io);
12371 }
12372
12373 /*
12374  * For HA operation.  Handle commands that come in from the other
12375  * controller.
12376  */
12377 static void
12378 ctl_handle_isc(union ctl_io *io)
12379 {
12380         int free_io;
12381         struct ctl_lun *lun;
12382         struct ctl_softc *softc = control_softc;
12383         uint32_t targ_lun;
12384
12385         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12386         lun = softc->ctl_luns[targ_lun];
12387
12388         switch (io->io_hdr.msg_type) {
12389         case CTL_MSG_SERIALIZE:
12390                 free_io = ctl_serialize_other_sc_cmd(&io->scsiio);
12391                 break;
12392         case CTL_MSG_R2R: {
12393                 const struct ctl_cmd_entry *entry;
12394
12395                 /*
12396                  * This is only used in SER_ONLY mode.
12397                  */
12398                 free_io = 0;
12399                 entry = ctl_get_cmd_entry(&io->scsiio, NULL);
12400                 mtx_lock(&lun->lun_lock);
12401                 if (ctl_scsiio_lun_check(lun,
12402                     entry, (struct ctl_scsiio *)io) != 0) {
12403                         mtx_unlock(&lun->lun_lock);
12404                         ctl_done(io);
12405                         break;
12406                 }
12407                 io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
12408                 mtx_unlock(&lun->lun_lock);
12409                 ctl_enqueue_rtr(io);
12410                 break;
12411         }
12412         case CTL_MSG_FINISH_IO:
12413                 if (softc->ha_mode == CTL_HA_MODE_XFER) {
12414                         free_io = 0;
12415                         ctl_done(io);
12416                 } else {
12417                         free_io = 1;
12418                         mtx_lock(&lun->lun_lock);
12419                         TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr,
12420                                      ooa_links);
12421                         ctl_check_blocked(lun);
12422                         mtx_unlock(&lun->lun_lock);
12423                 }
12424                 break;
12425         case CTL_MSG_PERS_ACTION:
12426                 ctl_hndl_per_res_out_on_other_sc(
12427                         (union ctl_ha_msg *)&io->presio.pr_msg);
12428                 free_io = 1;
12429                 break;
12430         case CTL_MSG_BAD_JUJU:
12431                 free_io = 0;
12432                 ctl_done(io);
12433                 break;
12434         case CTL_MSG_DATAMOVE:
12435                 /* Only used in XFER mode */
12436                 free_io = 0;
12437                 ctl_datamove_remote(io);
12438                 break;
12439         case CTL_MSG_DATAMOVE_DONE:
12440                 /* Only used in XFER mode */
12441                 free_io = 0;
12442                 io->scsiio.be_move_done(io);
12443                 break;
12444         case CTL_MSG_FAILOVER:
12445                 ctl_failover_lun(io);
12446                 free_io = 1;
12447                 break;
12448         default:
12449                 free_io = 1;
12450                 printf("%s: Invalid message type %d\n",
12451                        __func__, io->io_hdr.msg_type);
12452                 break;
12453         }
12454         if (free_io)
12455                 ctl_free_io(io);
12456
12457 }
12458
12459
12460 /*
12461  * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
12462  * there is no match.
12463  */
12464 static ctl_lun_error_pattern
12465 ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
12466 {
12467         const struct ctl_cmd_entry *entry;
12468         ctl_lun_error_pattern filtered_pattern, pattern;
12469
12470         pattern = desc->error_pattern;
12471
12472         /*
12473          * XXX KDM we need more data passed into this function to match a
12474          * custom pattern, and we actually need to implement custom pattern
12475          * matching.
12476          */
12477         if (pattern & CTL_LUN_PAT_CMD)
12478                 return (CTL_LUN_PAT_CMD);
12479
12480         if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
12481                 return (CTL_LUN_PAT_ANY);
12482
12483         entry = ctl_get_cmd_entry(ctsio, NULL);
12484
12485         filtered_pattern = entry->pattern & pattern;
12486
12487         /*
12488          * If the user requested specific flags in the pattern (e.g.
12489          * CTL_LUN_PAT_RANGE), make sure the command supports all of those
12490          * flags.
12491          *
12492          * If the user did not specify any flags, it doesn't matter whether
12493          * or not the command supports the flags.
12494          */
12495         if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
12496              (pattern & ~CTL_LUN_PAT_MASK))
12497                 return (CTL_LUN_PAT_NONE);
12498
12499         /*
12500          * If the user asked for a range check, see if the requested LBA
12501          * range overlaps with this command's LBA range.
12502          */
12503         if (filtered_pattern & CTL_LUN_PAT_RANGE) {
12504                 uint64_t lba1;
12505                 uint64_t len1;
12506                 ctl_action action;
12507                 int retval;
12508
12509                 retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
12510                 if (retval != 0)
12511                         return (CTL_LUN_PAT_NONE);
12512
12513                 action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
12514                                               desc->lba_range.len, FALSE);
12515                 /*
12516                  * A "pass" means that the LBA ranges don't overlap, so
12517                  * this doesn't match the user's range criteria.
12518                  */
12519                 if (action == CTL_ACTION_PASS)
12520                         return (CTL_LUN_PAT_NONE);
12521         }
12522
12523         return (filtered_pattern);
12524 }
12525
12526 static void
12527 ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
12528 {
12529         struct ctl_error_desc *desc, *desc2;
12530
12531         mtx_assert(&lun->lun_lock, MA_OWNED);
12532
12533         STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
12534                 ctl_lun_error_pattern pattern;
12535                 /*
12536                  * Check to see whether this particular command matches
12537                  * the pattern in the descriptor.
12538                  */
12539                 pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
12540                 if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
12541                         continue;
12542
12543                 switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
12544                 case CTL_LUN_INJ_ABORTED:
12545                         ctl_set_aborted(&io->scsiio);
12546                         break;
12547                 case CTL_LUN_INJ_MEDIUM_ERR:
12548                         ctl_set_medium_error(&io->scsiio,
12549                             (io->io_hdr.flags & CTL_FLAG_DATA_MASK) !=
12550                              CTL_FLAG_DATA_OUT);
12551                         break;
12552                 case CTL_LUN_INJ_UA:
12553                         /* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
12554                          * OCCURRED */
12555                         ctl_set_ua(&io->scsiio, 0x29, 0x00);
12556                         break;
12557                 case CTL_LUN_INJ_CUSTOM:
12558                         /*
12559                          * We're assuming the user knows what he is doing.
12560                          * Just copy the sense information without doing
12561                          * checks.
12562                          */
12563                         bcopy(&desc->custom_sense, &io->scsiio.sense_data,
12564                               MIN(sizeof(desc->custom_sense),
12565                                   sizeof(io->scsiio.sense_data)));
12566                         io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
12567                         io->scsiio.sense_len = SSD_FULL_SIZE;
12568                         io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
12569                         break;
12570                 case CTL_LUN_INJ_NONE:
12571                 default:
12572                         /*
12573                          * If this is an error injection type we don't know
12574                          * about, clear the continuous flag (if it is set)
12575                          * so it will get deleted below.
12576                          */
12577                         desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
12578                         break;
12579                 }
12580                 /*
12581                  * By default, each error injection action is a one-shot
12582                  */
12583                 if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
12584                         continue;
12585
12586                 STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
12587
12588                 free(desc, M_CTL);
12589         }
12590 }
12591
12592 #ifdef CTL_IO_DELAY
12593 static void
12594 ctl_datamove_timer_wakeup(void *arg)
12595 {
12596         union ctl_io *io;
12597
12598         io = (union ctl_io *)arg;
12599
12600         ctl_datamove(io);
12601 }
12602 #endif /* CTL_IO_DELAY */
12603
12604 void
12605 ctl_datamove(union ctl_io *io)
12606 {
12607         struct ctl_lun *lun;
12608         void (*fe_datamove)(union ctl_io *io);
12609
12610         mtx_assert(&control_softc->ctl_lock, MA_NOTOWNED);
12611
12612         CTL_DEBUG_PRINT(("ctl_datamove\n"));
12613
12614         lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12615 #ifdef CTL_TIME_IO
12616         if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12617                 char str[256];
12618                 char path_str[64];
12619                 struct sbuf sb;
12620
12621                 ctl_scsi_path_string(io, path_str, sizeof(path_str));
12622                 sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12623
12624                 sbuf_cat(&sb, path_str);
12625                 switch (io->io_hdr.io_type) {
12626                 case CTL_IO_SCSI:
12627                         ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12628                         sbuf_printf(&sb, "\n");
12629                         sbuf_cat(&sb, path_str);
12630                         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12631                                     io->scsiio.tag_num, io->scsiio.tag_type);
12632                         break;
12633                 case CTL_IO_TASK:
12634                         sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12635                                     "Tag Type: %d\n", io->taskio.task_action,
12636                                     io->taskio.tag_num, io->taskio.tag_type);
12637                         break;
12638                 default:
12639                         panic("%s: Invalid CTL I/O type %d\n",
12640                             __func__, io->io_hdr.io_type);
12641                 }
12642                 sbuf_cat(&sb, path_str);
12643                 sbuf_printf(&sb, "ctl_datamove: %jd seconds\n",
12644                             (intmax_t)time_uptime - io->io_hdr.start_time);
12645                 sbuf_finish(&sb);
12646                 printf("%s", sbuf_data(&sb));
12647         }
12648 #endif /* CTL_TIME_IO */
12649
12650 #ifdef CTL_IO_DELAY
12651         if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
12652                 io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
12653         } else {
12654                 if ((lun != NULL)
12655                  && (lun->delay_info.datamove_delay > 0)) {
12656
12657                         callout_init(&io->io_hdr.delay_callout, /*mpsafe*/ 1);
12658                         io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
12659                         callout_reset(&io->io_hdr.delay_callout,
12660                                       lun->delay_info.datamove_delay * hz,
12661                                       ctl_datamove_timer_wakeup, io);
12662                         if (lun->delay_info.datamove_type ==
12663                             CTL_DELAY_TYPE_ONESHOT)
12664                                 lun->delay_info.datamove_delay = 0;
12665                         return;
12666                 }
12667         }
12668 #endif
12669
12670         /*
12671          * This command has been aborted.  Set the port status, so we fail
12672          * the data move.
12673          */
12674         if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12675                 printf("ctl_datamove: tag 0x%04x on (%u:%u:%u) aborted\n",
12676                        io->scsiio.tag_num, io->io_hdr.nexus.initid,
12677                        io->io_hdr.nexus.targ_port,
12678                        io->io_hdr.nexus.targ_lun);
12679                 io->io_hdr.port_status = 31337;
12680                 /*
12681                  * Note that the backend, in this case, will get the
12682                  * callback in its context.  In other cases it may get
12683                  * called in the frontend's interrupt thread context.
12684                  */
12685                 io->scsiio.be_move_done(io);
12686                 return;
12687         }
12688
12689         /* Don't confuse frontend with zero length data move. */
12690         if (io->scsiio.kern_data_len == 0) {
12691                 io->scsiio.be_move_done(io);
12692                 return;
12693         }
12694
12695         fe_datamove = ctl_io_port(&io->io_hdr)->fe_datamove;
12696         fe_datamove(io);
12697 }
12698
12699 static void
12700 ctl_send_datamove_done(union ctl_io *io, int have_lock)
12701 {
12702         union ctl_ha_msg msg;
12703 #ifdef CTL_TIME_IO
12704         struct bintime cur_bt;
12705 #endif
12706
12707         memset(&msg, 0, sizeof(msg));
12708         msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
12709         msg.hdr.original_sc = io;
12710         msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
12711         msg.hdr.nexus = io->io_hdr.nexus;
12712         msg.hdr.status = io->io_hdr.status;
12713         msg.scsi.tag_num = io->scsiio.tag_num;
12714         msg.scsi.tag_type = io->scsiio.tag_type;
12715         msg.scsi.scsi_status = io->scsiio.scsi_status;
12716         memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
12717                io->scsiio.sense_len);
12718         msg.scsi.sense_len = io->scsiio.sense_len;
12719         msg.scsi.sense_residual = io->scsiio.sense_residual;
12720         msg.scsi.fetd_status = io->io_hdr.port_status;
12721         msg.scsi.residual = io->scsiio.residual;
12722         io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12723         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12724                 ctl_failover_io(io, /*have_lock*/ have_lock);
12725                 return;
12726         }
12727         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12728             sizeof(msg.scsi) - sizeof(msg.scsi.sense_data) +
12729             msg.scsi.sense_len, M_WAITOK);
12730
12731 #ifdef CTL_TIME_IO
12732         getbinuptime(&cur_bt);
12733         bintime_sub(&cur_bt, &io->io_hdr.dma_start_bt);
12734         bintime_add(&io->io_hdr.dma_bt, &cur_bt);
12735 #endif
12736         io->io_hdr.num_dmas++;
12737 }
12738
12739 /*
12740  * The DMA to the remote side is done, now we need to tell the other side
12741  * we're done so it can continue with its data movement.
12742  */
12743 static void
12744 ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
12745 {
12746         union ctl_io *io;
12747         uint32_t i;
12748
12749         io = rq->context;
12750
12751         if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12752                 printf("%s: ISC DMA write failed with error %d", __func__,
12753                        rq->ret);
12754                 ctl_set_internal_failure(&io->scsiio,
12755                                          /*sks_valid*/ 1,
12756                                          /*retry_count*/ rq->ret);
12757         }
12758
12759         ctl_dt_req_free(rq);
12760
12761         for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12762                 free(io->io_hdr.local_sglist[i].addr, M_CTL);
12763         free(io->io_hdr.remote_sglist, M_CTL);
12764         io->io_hdr.remote_sglist = NULL;
12765         io->io_hdr.local_sglist = NULL;
12766
12767         /*
12768          * The data is in local and remote memory, so now we need to send
12769          * status (good or back) back to the other side.
12770          */
12771         ctl_send_datamove_done(io, /*have_lock*/ 0);
12772 }
12773
12774 /*
12775  * We've moved the data from the host/controller into local memory.  Now we
12776  * need to push it over to the remote controller's memory.
12777  */
12778 static int
12779 ctl_datamove_remote_dm_write_cb(union ctl_io *io)
12780 {
12781         int retval;
12782
12783         retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
12784                                           ctl_datamove_remote_write_cb);
12785         return (retval);
12786 }
12787
12788 static void
12789 ctl_datamove_remote_write(union ctl_io *io)
12790 {
12791         int retval;
12792         void (*fe_datamove)(union ctl_io *io);
12793
12794         /*
12795          * - Get the data from the host/HBA into local memory.
12796          * - DMA memory from the local controller to the remote controller.
12797          * - Send status back to the remote controller.
12798          */
12799
12800         retval = ctl_datamove_remote_sgl_setup(io);
12801         if (retval != 0)
12802                 return;
12803
12804         /* Switch the pointer over so the FETD knows what to do */
12805         io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12806
12807         /*
12808          * Use a custom move done callback, since we need to send completion
12809          * back to the other controller, not to the backend on this side.
12810          */
12811         io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
12812
12813         fe_datamove = ctl_io_port(&io->io_hdr)->fe_datamove;
12814         fe_datamove(io);
12815 }
12816
12817 static int
12818 ctl_datamove_remote_dm_read_cb(union ctl_io *io)
12819 {
12820 #if 0
12821         char str[256];
12822         char path_str[64];
12823         struct sbuf sb;
12824 #endif
12825         uint32_t i;
12826
12827         for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12828                 free(io->io_hdr.local_sglist[i].addr, M_CTL);
12829         free(io->io_hdr.remote_sglist, M_CTL);
12830         io->io_hdr.remote_sglist = NULL;
12831         io->io_hdr.local_sglist = NULL;
12832
12833 #if 0
12834         scsi_path_string(io, path_str, sizeof(path_str));
12835         sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12836         sbuf_cat(&sb, path_str);
12837         scsi_command_string(&io->scsiio, NULL, &sb);
12838         sbuf_printf(&sb, "\n");
12839         sbuf_cat(&sb, path_str);
12840         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12841                     io->scsiio.tag_num, io->scsiio.tag_type);
12842         sbuf_cat(&sb, path_str);
12843         sbuf_printf(&sb, "%s: flags %#x, status %#x\n", __func__,
12844                     io->io_hdr.flags, io->io_hdr.status);
12845         sbuf_finish(&sb);
12846         printk("%s", sbuf_data(&sb));
12847 #endif
12848
12849
12850         /*
12851          * The read is done, now we need to send status (good or bad) back
12852          * to the other side.
12853          */
12854         ctl_send_datamove_done(io, /*have_lock*/ 0);
12855
12856         return (0);
12857 }
12858
12859 static void
12860 ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
12861 {
12862         union ctl_io *io;
12863         void (*fe_datamove)(union ctl_io *io);
12864
12865         io = rq->context;
12866
12867         if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12868                 printf("%s: ISC DMA read failed with error %d\n", __func__,
12869                        rq->ret);
12870                 ctl_set_internal_failure(&io->scsiio,
12871                                          /*sks_valid*/ 1,
12872                                          /*retry_count*/ rq->ret);
12873         }
12874
12875         ctl_dt_req_free(rq);
12876
12877         /* Switch the pointer over so the FETD knows what to do */
12878         io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12879
12880         /*
12881          * Use a custom move done callback, since we need to send completion
12882          * back to the other controller, not to the backend on this side.
12883          */
12884         io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
12885
12886         /* XXX KDM add checks like the ones in ctl_datamove? */
12887
12888         fe_datamove = ctl_io_port(&io->io_hdr)->fe_datamove;
12889         fe_datamove(io);
12890 }
12891
12892 static int
12893 ctl_datamove_remote_sgl_setup(union ctl_io *io)
12894 {
12895         struct ctl_sg_entry *local_sglist;
12896         uint32_t len_to_go;
12897         int retval;
12898         int i;
12899
12900         retval = 0;
12901         local_sglist = io->io_hdr.local_sglist;
12902         len_to_go = io->scsiio.kern_data_len;
12903
12904         /*
12905          * The difficult thing here is that the size of the various
12906          * S/G segments may be different than the size from the
12907          * remote controller.  That'll make it harder when DMAing
12908          * the data back to the other side.
12909          */
12910         for (i = 0; len_to_go > 0; i++) {
12911                 local_sglist[i].len = MIN(len_to_go, CTL_HA_DATAMOVE_SEGMENT);
12912                 local_sglist[i].addr =
12913                     malloc(local_sglist[i].len, M_CTL, M_WAITOK);
12914
12915                 len_to_go -= local_sglist[i].len;
12916         }
12917         /*
12918          * Reset the number of S/G entries accordingly.  The original
12919          * number of S/G entries is available in rem_sg_entries.
12920          */
12921         io->scsiio.kern_sg_entries = i;
12922
12923 #if 0
12924         printf("%s: kern_sg_entries = %d\n", __func__,
12925                io->scsiio.kern_sg_entries);
12926         for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12927                 printf("%s: sg[%d] = %p, %lu\n", __func__, i,
12928                        local_sglist[i].addr, local_sglist[i].len);
12929 #endif
12930
12931         return (retval);
12932 }
12933
12934 static int
12935 ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
12936                          ctl_ha_dt_cb callback)
12937 {
12938         struct ctl_ha_dt_req *rq;
12939         struct ctl_sg_entry *remote_sglist, *local_sglist;
12940         uint32_t local_used, remote_used, total_used;
12941         int i, j, isc_ret;
12942
12943         rq = ctl_dt_req_alloc();
12944
12945         /*
12946          * If we failed to allocate the request, and if the DMA didn't fail
12947          * anyway, set busy status.  This is just a resource allocation
12948          * failure.
12949          */
12950         if ((rq == NULL)
12951          && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
12952              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS))
12953                 ctl_set_busy(&io->scsiio);
12954
12955         if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
12956             (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS) {
12957
12958                 if (rq != NULL)
12959                         ctl_dt_req_free(rq);
12960
12961                 /*
12962                  * The data move failed.  We need to return status back
12963                  * to the other controller.  No point in trying to DMA
12964                  * data to the remote controller.
12965                  */
12966
12967                 ctl_send_datamove_done(io, /*have_lock*/ 0);
12968
12969                 return (1);
12970         }
12971
12972         local_sglist = io->io_hdr.local_sglist;
12973         remote_sglist = io->io_hdr.remote_sglist;
12974         local_used = 0;
12975         remote_used = 0;
12976         total_used = 0;
12977
12978         /*
12979          * Pull/push the data over the wire from/to the other controller.
12980          * This takes into account the possibility that the local and
12981          * remote sglists may not be identical in terms of the size of
12982          * the elements and the number of elements.
12983          *
12984          * One fundamental assumption here is that the length allocated for
12985          * both the local and remote sglists is identical.  Otherwise, we've
12986          * essentially got a coding error of some sort.
12987          */
12988         isc_ret = CTL_HA_STATUS_SUCCESS;
12989         for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
12990                 uint32_t cur_len;
12991                 uint8_t *tmp_ptr;
12992
12993                 rq->command = command;
12994                 rq->context = io;
12995
12996                 /*
12997                  * Both pointers should be aligned.  But it is possible
12998                  * that the allocation length is not.  They should both
12999                  * also have enough slack left over at the end, though,
13000                  * to round up to the next 8 byte boundary.
13001                  */
13002                 cur_len = MIN(local_sglist[i].len - local_used,
13003                               remote_sglist[j].len - remote_used);
13004                 rq->size = cur_len;
13005
13006                 tmp_ptr = (uint8_t *)local_sglist[i].addr;
13007                 tmp_ptr += local_used;
13008
13009 #if 0
13010                 /* Use physical addresses when talking to ISC hardware */
13011                 if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
13012                         /* XXX KDM use busdma */
13013                         rq->local = vtophys(tmp_ptr);
13014                 } else
13015                         rq->local = tmp_ptr;
13016 #else
13017                 KASSERT((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0,
13018                     ("HA does not support BUS_ADDR"));
13019                 rq->local = tmp_ptr;
13020 #endif
13021
13022                 tmp_ptr = (uint8_t *)remote_sglist[j].addr;
13023                 tmp_ptr += remote_used;
13024                 rq->remote = tmp_ptr;
13025
13026                 rq->callback = NULL;
13027
13028                 local_used += cur_len;
13029                 if (local_used >= local_sglist[i].len) {
13030                         i++;
13031                         local_used = 0;
13032                 }
13033
13034                 remote_used += cur_len;
13035                 if (remote_used >= remote_sglist[j].len) {
13036                         j++;
13037                         remote_used = 0;
13038                 }
13039                 total_used += cur_len;
13040
13041                 if (total_used >= io->scsiio.kern_data_len)
13042                         rq->callback = callback;
13043
13044 #if 0
13045                 printf("%s: %s: local %p remote %p size %d\n", __func__,
13046                        (command == CTL_HA_DT_CMD_WRITE) ? "WRITE" : "READ",
13047                        rq->local, rq->remote, rq->size);
13048 #endif
13049
13050                 isc_ret = ctl_dt_single(rq);
13051                 if (isc_ret > CTL_HA_STATUS_SUCCESS)
13052                         break;
13053         }
13054         if (isc_ret != CTL_HA_STATUS_WAIT) {
13055                 rq->ret = isc_ret;
13056                 callback(rq);
13057         }
13058
13059         return (0);
13060 }
13061
13062 static void
13063 ctl_datamove_remote_read(union ctl_io *io)
13064 {
13065         int retval;
13066         uint32_t i;
13067
13068         /*
13069          * This will send an error to the other controller in the case of a
13070          * failure.
13071          */
13072         retval = ctl_datamove_remote_sgl_setup(io);
13073         if (retval != 0)
13074                 return;
13075
13076         retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
13077                                           ctl_datamove_remote_read_cb);
13078         if (retval != 0) {
13079                 /*
13080                  * Make sure we free memory if there was an error..  The
13081                  * ctl_datamove_remote_xfer() function will send the
13082                  * datamove done message, or call the callback with an
13083                  * error if there is a problem.
13084                  */
13085                 for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13086                         free(io->io_hdr.local_sglist[i].addr, M_CTL);
13087                 free(io->io_hdr.remote_sglist, M_CTL);
13088                 io->io_hdr.remote_sglist = NULL;
13089                 io->io_hdr.local_sglist = NULL;
13090         }
13091 }
13092
13093 /*
13094  * Process a datamove request from the other controller.  This is used for
13095  * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
13096  * first.  Once that is complete, the data gets DMAed into the remote
13097  * controller's memory.  For reads, we DMA from the remote controller's
13098  * memory into our memory first, and then move it out to the FETD.
13099  */
13100 static void
13101 ctl_datamove_remote(union ctl_io *io)
13102 {
13103
13104         mtx_assert(&control_softc->ctl_lock, MA_NOTOWNED);
13105
13106         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
13107                 ctl_failover_io(io, /*have_lock*/ 0);
13108                 return;
13109         }
13110
13111         /*
13112          * Note that we look for an aborted I/O here, but don't do some of
13113          * the other checks that ctl_datamove() normally does.
13114          * We don't need to run the datamove delay code, since that should
13115          * have been done if need be on the other controller.
13116          */
13117         if (io->io_hdr.flags & CTL_FLAG_ABORT) {
13118                 printf("%s: tag 0x%04x on (%u:%u:%u) aborted\n", __func__,
13119                        io->scsiio.tag_num, io->io_hdr.nexus.initid,
13120                        io->io_hdr.nexus.targ_port,
13121                        io->io_hdr.nexus.targ_lun);
13122                 io->io_hdr.port_status = 31338;
13123                 ctl_send_datamove_done(io, /*have_lock*/ 0);
13124                 return;
13125         }
13126
13127         if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT)
13128                 ctl_datamove_remote_write(io);
13129         else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN)
13130                 ctl_datamove_remote_read(io);
13131         else {
13132                 io->io_hdr.port_status = 31339;
13133                 ctl_send_datamove_done(io, /*have_lock*/ 0);
13134         }
13135 }
13136
13137 static void
13138 ctl_process_done(union ctl_io *io)
13139 {
13140         struct ctl_lun *lun;
13141         struct ctl_softc *softc = control_softc;
13142         void (*fe_done)(union ctl_io *io);
13143         union ctl_ha_msg msg;
13144         uint32_t targ_port = io->io_hdr.nexus.targ_port;
13145
13146         CTL_DEBUG_PRINT(("ctl_process_done\n"));
13147         fe_done = softc->ctl_ports[targ_port]->fe_done;
13148
13149 #ifdef CTL_TIME_IO
13150         if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
13151                 char str[256];
13152                 char path_str[64];
13153                 struct sbuf sb;
13154
13155                 ctl_scsi_path_string(io, path_str, sizeof(path_str));
13156                 sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
13157
13158                 sbuf_cat(&sb, path_str);
13159                 switch (io->io_hdr.io_type) {
13160                 case CTL_IO_SCSI:
13161                         ctl_scsi_command_string(&io->scsiio, NULL, &sb);
13162                         sbuf_printf(&sb, "\n");
13163                         sbuf_cat(&sb, path_str);
13164                         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
13165                                     io->scsiio.tag_num, io->scsiio.tag_type);
13166                         break;
13167                 case CTL_IO_TASK:
13168                         sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
13169                                     "Tag Type: %d\n", io->taskio.task_action,
13170                                     io->taskio.tag_num, io->taskio.tag_type);
13171                         break;
13172                 default:
13173                         panic("%s: Invalid CTL I/O type %d\n",
13174                             __func__, io->io_hdr.io_type);
13175                 }
13176                 sbuf_cat(&sb, path_str);
13177                 sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
13178                             (intmax_t)time_uptime - io->io_hdr.start_time);
13179                 sbuf_finish(&sb);
13180                 printf("%s", sbuf_data(&sb));
13181         }
13182 #endif /* CTL_TIME_IO */
13183
13184         switch (io->io_hdr.io_type) {
13185         case CTL_IO_SCSI:
13186                 break;
13187         case CTL_IO_TASK:
13188                 if (ctl_debug & CTL_DEBUG_INFO)
13189                         ctl_io_error_print(io, NULL);
13190                 fe_done(io);
13191                 return;
13192         default:
13193                 panic("%s: Invalid CTL I/O type %d\n",
13194                     __func__, io->io_hdr.io_type);
13195         }
13196
13197         lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13198         if (lun == NULL) {
13199                 CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
13200                                  io->io_hdr.nexus.targ_mapped_lun));
13201                 goto bailout;
13202         }
13203
13204         mtx_lock(&lun->lun_lock);
13205
13206         /*
13207          * Check to see if we have any informational exception and status
13208          * of this command can be modified to report it in form of either
13209          * RECOVERED ERROR or NO SENSE, depending on MRIE mode page field.
13210          */
13211         if (lun->ie_reported == 0 && lun->ie_asc != 0 &&
13212             io->io_hdr.status == CTL_SUCCESS &&
13213             (io->io_hdr.flags & CTL_FLAG_STATUS_SENT) == 0) {
13214                 uint8_t mrie = lun->MODE_IE.mrie;
13215                 uint8_t per = ((lun->MODE_RWER.byte3 & SMS_RWER_PER) ||
13216                     (lun->MODE_VER.byte3 & SMS_VER_PER));
13217                 if (((mrie == SIEP_MRIE_REC_COND && per) ||
13218                      mrie == SIEP_MRIE_REC_UNCOND ||
13219                      mrie == SIEP_MRIE_NO_SENSE) &&
13220                     (ctl_get_cmd_entry(&io->scsiio, NULL)->flags &
13221                      CTL_CMD_FLAG_NO_SENSE) == 0) {
13222                         ctl_set_sense(&io->scsiio,
13223                               /*current_error*/ 1,
13224                               /*sense_key*/ (mrie == SIEP_MRIE_NO_SENSE) ?
13225                                 SSD_KEY_NO_SENSE : SSD_KEY_RECOVERED_ERROR,
13226                               /*asc*/ lun->ie_asc,
13227                               /*ascq*/ lun->ie_ascq,
13228                               SSD_ELEM_NONE);
13229                         lun->ie_reported = 1;
13230                 }
13231         } else if (lun->ie_reported < 0)
13232                 lun->ie_reported = 0;
13233
13234         /*
13235          * Check to see if we have any errors to inject here.  We only
13236          * inject errors for commands that don't already have errors set.
13237          */
13238         if (!STAILQ_EMPTY(&lun->error_list) &&
13239             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS) &&
13240             ((io->io_hdr.flags & CTL_FLAG_STATUS_SENT) == 0))
13241                 ctl_inject_error(lun, io);
13242
13243         /*
13244          * XXX KDM how do we treat commands that aren't completed
13245          * successfully?
13246          *
13247          * XXX KDM should we also track I/O latency?
13248          */
13249         if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS &&
13250             io->io_hdr.io_type == CTL_IO_SCSI) {
13251 #ifdef CTL_TIME_IO
13252                 struct bintime cur_bt;
13253 #endif
13254                 int type;
13255
13256                 if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13257                     CTL_FLAG_DATA_IN)
13258                         type = CTL_STATS_READ;
13259                 else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13260                     CTL_FLAG_DATA_OUT)
13261                         type = CTL_STATS_WRITE;
13262                 else
13263                         type = CTL_STATS_NO_IO;
13264
13265                 lun->stats.ports[targ_port].bytes[type] +=
13266                     io->scsiio.kern_total_len;
13267                 lun->stats.ports[targ_port].operations[type]++;
13268 #ifdef CTL_TIME_IO
13269                 bintime_add(&lun->stats.ports[targ_port].dma_time[type],
13270                    &io->io_hdr.dma_bt);
13271                 getbinuptime(&cur_bt);
13272                 bintime_sub(&cur_bt, &io->io_hdr.start_bt);
13273                 bintime_add(&lun->stats.ports[targ_port].time[type], &cur_bt);
13274 #endif
13275                 lun->stats.ports[targ_port].num_dmas[type] +=
13276                     io->io_hdr.num_dmas;
13277         }
13278
13279         /*
13280          * Remove this from the OOA queue.
13281          */
13282         TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
13283 #ifdef CTL_TIME_IO
13284         if (TAILQ_EMPTY(&lun->ooa_queue))
13285                 lun->last_busy = getsbinuptime();
13286 #endif
13287
13288         /*
13289          * Run through the blocked queue on this LUN and see if anything
13290          * has become unblocked, now that this transaction is done.
13291          */
13292         ctl_check_blocked(lun);
13293
13294         /*
13295          * If the LUN has been invalidated, free it if there is nothing
13296          * left on its OOA queue.
13297          */
13298         if ((lun->flags & CTL_LUN_INVALID)
13299          && TAILQ_EMPTY(&lun->ooa_queue)) {
13300                 mtx_unlock(&lun->lun_lock);
13301                 mtx_lock(&softc->ctl_lock);
13302                 ctl_free_lun(lun);
13303                 mtx_unlock(&softc->ctl_lock);
13304         } else
13305                 mtx_unlock(&lun->lun_lock);
13306
13307 bailout:
13308
13309         /*
13310          * If this command has been aborted, make sure we set the status
13311          * properly.  The FETD is responsible for freeing the I/O and doing
13312          * whatever it needs to do to clean up its state.
13313          */
13314         if (io->io_hdr.flags & CTL_FLAG_ABORT)
13315                 ctl_set_task_aborted(&io->scsiio);
13316
13317         /*
13318          * If enabled, print command error status.
13319          */
13320         if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS &&
13321             (ctl_debug & CTL_DEBUG_INFO) != 0)
13322                 ctl_io_error_print(io, NULL);
13323
13324         /*
13325          * Tell the FETD or the other shelf controller we're done with this
13326          * command.  Note that only SCSI commands get to this point.  Task
13327          * management commands are completed above.
13328          */
13329         if ((softc->ha_mode != CTL_HA_MODE_XFER) &&
13330             (io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)) {
13331                 memset(&msg, 0, sizeof(msg));
13332                 msg.hdr.msg_type = CTL_MSG_FINISH_IO;
13333                 msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
13334                 msg.hdr.nexus = io->io_hdr.nexus;
13335                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13336                     sizeof(msg.scsi) - sizeof(msg.scsi.sense_data),
13337                     M_WAITOK);
13338         }
13339
13340         fe_done(io);
13341 }
13342
13343 #ifdef CTL_WITH_CA
13344 /*
13345  * Front end should call this if it doesn't do autosense.  When the request
13346  * sense comes back in from the initiator, we'll dequeue this and send it.
13347  */
13348 int
13349 ctl_queue_sense(union ctl_io *io)
13350 {
13351         struct ctl_lun *lun;
13352         struct ctl_port *port;
13353         struct ctl_softc *softc;
13354         uint32_t initidx, targ_lun;
13355
13356         softc = control_softc;
13357
13358         CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
13359
13360         /*
13361          * LUN lookup will likely move to the ctl_work_thread() once we
13362          * have our new queueing infrastructure (that doesn't put things on
13363          * a per-LUN queue initially).  That is so that we can handle
13364          * things like an INQUIRY to a LUN that we don't have enabled.  We
13365          * can't deal with that right now.
13366          */
13367         mtx_lock(&softc->ctl_lock);
13368
13369         /*
13370          * If we don't have a LUN for this, just toss the sense
13371          * information.
13372          */
13373         port = ctl_io_port(&ctsio->io_hdr);
13374         targ_lun = ctl_lun_map_from_port(port, io->io_hdr.nexus.targ_lun);
13375         if ((targ_lun < CTL_MAX_LUNS)
13376          && (softc->ctl_luns[targ_lun] != NULL))
13377                 lun = softc->ctl_luns[targ_lun];
13378         else
13379                 goto bailout;
13380
13381         initidx = ctl_get_initindex(&io->io_hdr.nexus);
13382
13383         mtx_lock(&lun->lun_lock);
13384         /*
13385          * Already have CA set for this LUN...toss the sense information.
13386          */
13387         if (ctl_is_set(lun->have_ca, initidx)) {
13388                 mtx_unlock(&lun->lun_lock);
13389                 goto bailout;
13390         }
13391
13392         memcpy(&lun->pending_sense[initidx], &io->scsiio.sense_data,
13393                MIN(sizeof(lun->pending_sense[initidx]),
13394                sizeof(io->scsiio.sense_data)));
13395         ctl_set_mask(lun->have_ca, initidx);
13396         mtx_unlock(&lun->lun_lock);
13397
13398 bailout:
13399         mtx_unlock(&softc->ctl_lock);
13400
13401         ctl_free_io(io);
13402
13403         return (CTL_RETVAL_COMPLETE);
13404 }
13405 #endif
13406
13407 /*
13408  * Primary command inlet from frontend ports.  All SCSI and task I/O
13409  * requests must go through this function.
13410  */
13411 int
13412 ctl_queue(union ctl_io *io)
13413 {
13414         struct ctl_port *port;
13415
13416         CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
13417
13418 #ifdef CTL_TIME_IO
13419         io->io_hdr.start_time = time_uptime;
13420         getbinuptime(&io->io_hdr.start_bt);
13421 #endif /* CTL_TIME_IO */
13422
13423         /* Map FE-specific LUN ID into global one. */
13424         port = ctl_io_port(&io->io_hdr);
13425         io->io_hdr.nexus.targ_mapped_lun =
13426             ctl_lun_map_from_port(port, io->io_hdr.nexus.targ_lun);
13427
13428         switch (io->io_hdr.io_type) {
13429         case CTL_IO_SCSI:
13430         case CTL_IO_TASK:
13431                 if (ctl_debug & CTL_DEBUG_CDB)
13432                         ctl_io_print(io);
13433                 ctl_enqueue_incoming(io);
13434                 break;
13435         default:
13436                 printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
13437                 return (EINVAL);
13438         }
13439
13440         return (CTL_RETVAL_COMPLETE);
13441 }
13442
13443 #ifdef CTL_IO_DELAY
13444 static void
13445 ctl_done_timer_wakeup(void *arg)
13446 {
13447         union ctl_io *io;
13448
13449         io = (union ctl_io *)arg;
13450         ctl_done(io);
13451 }
13452 #endif /* CTL_IO_DELAY */
13453
13454 void
13455 ctl_serseq_done(union ctl_io *io)
13456 {
13457         struct ctl_lun *lun;
13458
13459         lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13460         if (lun->be_lun == NULL ||
13461             lun->be_lun->serseq == CTL_LUN_SERSEQ_OFF)
13462                 return;
13463         mtx_lock(&lun->lun_lock);
13464         io->io_hdr.flags |= CTL_FLAG_SERSEQ_DONE;
13465         ctl_check_blocked(lun);
13466         mtx_unlock(&lun->lun_lock);
13467 }
13468
13469 void
13470 ctl_done(union ctl_io *io)
13471 {
13472
13473         /*
13474          * Enable this to catch duplicate completion issues.
13475          */
13476 #if 0
13477         if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
13478                 printf("%s: type %d msg %d cdb %x iptl: "
13479                        "%u:%u:%u tag 0x%04x "
13480                        "flag %#x status %x\n",
13481                         __func__,
13482                         io->io_hdr.io_type,
13483                         io->io_hdr.msg_type,
13484                         io->scsiio.cdb[0],
13485                         io->io_hdr.nexus.initid,
13486                         io->io_hdr.nexus.targ_port,
13487                         io->io_hdr.nexus.targ_lun,
13488                         (io->io_hdr.io_type ==
13489                         CTL_IO_TASK) ?
13490                         io->taskio.tag_num :
13491                         io->scsiio.tag_num,
13492                         io->io_hdr.flags,
13493                         io->io_hdr.status);
13494         } else
13495                 io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
13496 #endif
13497
13498         /*
13499          * This is an internal copy of an I/O, and should not go through
13500          * the normal done processing logic.
13501          */
13502         if (io->io_hdr.flags & CTL_FLAG_INT_COPY)
13503                 return;
13504
13505 #ifdef CTL_IO_DELAY
13506         if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
13507                 io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
13508         } else {
13509                 struct ctl_lun *lun;
13510
13511                 lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13512
13513                 if ((lun != NULL)
13514                  && (lun->delay_info.done_delay > 0)) {
13515
13516                         callout_init(&io->io_hdr.delay_callout, /*mpsafe*/ 1);
13517                         io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
13518                         callout_reset(&io->io_hdr.delay_callout,
13519                                       lun->delay_info.done_delay * hz,
13520                                       ctl_done_timer_wakeup, io);
13521                         if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
13522                                 lun->delay_info.done_delay = 0;
13523                         return;
13524                 }
13525         }
13526 #endif /* CTL_IO_DELAY */
13527
13528         ctl_enqueue_done(io);
13529 }
13530
13531 static void
13532 ctl_work_thread(void *arg)
13533 {
13534         struct ctl_thread *thr = (struct ctl_thread *)arg;
13535         struct ctl_softc *softc = thr->ctl_softc;
13536         union ctl_io *io;
13537         int retval;
13538
13539         CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
13540
13541         for (;;) {
13542                 /*
13543                  * We handle the queues in this order:
13544                  * - ISC
13545                  * - done queue (to free up resources, unblock other commands)
13546                  * - RtR queue
13547                  * - incoming queue
13548                  *
13549                  * If those queues are empty, we break out of the loop and
13550                  * go to sleep.
13551                  */
13552                 mtx_lock(&thr->queue_lock);
13553                 io = (union ctl_io *)STAILQ_FIRST(&thr->isc_queue);
13554                 if (io != NULL) {
13555                         STAILQ_REMOVE_HEAD(&thr->isc_queue, links);
13556                         mtx_unlock(&thr->queue_lock);
13557                         ctl_handle_isc(io);
13558                         continue;
13559                 }
13560                 io = (union ctl_io *)STAILQ_FIRST(&thr->done_queue);
13561                 if (io != NULL) {
13562                         STAILQ_REMOVE_HEAD(&thr->done_queue, links);
13563                         /* clear any blocked commands, call fe_done */
13564                         mtx_unlock(&thr->queue_lock);
13565                         ctl_process_done(io);
13566                         continue;
13567                 }
13568                 io = (union ctl_io *)STAILQ_FIRST(&thr->incoming_queue);
13569                 if (io != NULL) {
13570                         STAILQ_REMOVE_HEAD(&thr->incoming_queue, links);
13571                         mtx_unlock(&thr->queue_lock);
13572                         if (io->io_hdr.io_type == CTL_IO_TASK)
13573                                 ctl_run_task(io);
13574                         else
13575                                 ctl_scsiio_precheck(softc, &io->scsiio);
13576                         continue;
13577                 }
13578                 io = (union ctl_io *)STAILQ_FIRST(&thr->rtr_queue);
13579                 if (io != NULL) {
13580                         STAILQ_REMOVE_HEAD(&thr->rtr_queue, links);
13581                         mtx_unlock(&thr->queue_lock);
13582                         retval = ctl_scsiio(&io->scsiio);
13583                         if (retval != CTL_RETVAL_COMPLETE)
13584                                 CTL_DEBUG_PRINT(("ctl_scsiio failed\n"));
13585                         continue;
13586                 }
13587
13588                 /* Sleep until we have something to do. */
13589                 mtx_sleep(thr, &thr->queue_lock, PDROP | PRIBIO, "-", 0);
13590         }
13591 }
13592
13593 static void
13594 ctl_lun_thread(void *arg)
13595 {
13596         struct ctl_softc *softc = (struct ctl_softc *)arg;
13597         struct ctl_be_lun *be_lun;
13598
13599         CTL_DEBUG_PRINT(("ctl_lun_thread starting\n"));
13600
13601         for (;;) {
13602                 mtx_lock(&softc->ctl_lock);
13603                 be_lun = STAILQ_FIRST(&softc->pending_lun_queue);
13604                 if (be_lun != NULL) {
13605                         STAILQ_REMOVE_HEAD(&softc->pending_lun_queue, links);
13606                         mtx_unlock(&softc->ctl_lock);
13607                         ctl_create_lun(be_lun);
13608                         continue;
13609                 }
13610
13611                 /* Sleep until we have something to do. */
13612                 mtx_sleep(&softc->pending_lun_queue, &softc->ctl_lock,
13613                     PDROP | PRIBIO, "-", 0);
13614         }
13615 }
13616
13617 static void
13618 ctl_thresh_thread(void *arg)
13619 {
13620         struct ctl_softc *softc = (struct ctl_softc *)arg;
13621         struct ctl_lun *lun;
13622         struct ctl_logical_block_provisioning_page *page;
13623         const char *attr;
13624         union ctl_ha_msg msg;
13625         uint64_t thres, val;
13626         int i, e, set;
13627
13628         CTL_DEBUG_PRINT(("ctl_thresh_thread starting\n"));
13629
13630         for (;;) {
13631                 mtx_lock(&softc->ctl_lock);
13632                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
13633                         if ((lun->flags & CTL_LUN_DISABLED) ||
13634                             (lun->flags & CTL_LUN_NO_MEDIA) ||
13635                             lun->backend->lun_attr == NULL)
13636                                 continue;
13637                         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
13638                             softc->ha_mode == CTL_HA_MODE_XFER)
13639                                 continue;
13640                         if ((lun->MODE_RWER.byte8 & SMS_RWER_LBPERE) == 0)
13641                                 continue;
13642                         e = 0;
13643                         page = &lun->MODE_LBP;
13644                         for (i = 0; i < CTL_NUM_LBP_THRESH; i++) {
13645                                 if ((page->descr[i].flags & SLBPPD_ENABLED) == 0)
13646                                         continue;
13647                                 thres = scsi_4btoul(page->descr[i].count);
13648                                 thres <<= CTL_LBP_EXPONENT;
13649                                 switch (page->descr[i].resource) {
13650                                 case 0x01:
13651                                         attr = "blocksavail";
13652                                         break;
13653                                 case 0x02:
13654                                         attr = "blocksused";
13655                                         break;
13656                                 case 0xf1:
13657                                         attr = "poolblocksavail";
13658                                         break;
13659                                 case 0xf2:
13660                                         attr = "poolblocksused";
13661                                         break;
13662                                 default:
13663                                         continue;
13664                                 }
13665                                 mtx_unlock(&softc->ctl_lock); // XXX
13666                                 val = lun->backend->lun_attr(
13667                                     lun->be_lun->be_lun, attr);
13668                                 mtx_lock(&softc->ctl_lock);
13669                                 if (val == UINT64_MAX)
13670                                         continue;
13671                                 if ((page->descr[i].flags & SLBPPD_ARMING_MASK)
13672                                     == SLBPPD_ARMING_INC)
13673                                         e = (val >= thres);
13674                                 else
13675                                         e = (val <= thres);
13676                                 if (e)
13677                                         break;
13678                         }
13679                         mtx_lock(&lun->lun_lock);
13680                         if (e) {
13681                                 scsi_u64to8b((uint8_t *)&page->descr[i] -
13682                                     (uint8_t *)page, lun->ua_tpt_info);
13683                                 if (lun->lasttpt == 0 ||
13684                                     time_uptime - lun->lasttpt >= CTL_LBP_UA_PERIOD) {
13685                                         lun->lasttpt = time_uptime;
13686                                         ctl_est_ua_all(lun, -1, CTL_UA_THIN_PROV_THRES);
13687                                         set = 1;
13688                                 } else
13689                                         set = 0;
13690                         } else {
13691                                 lun->lasttpt = 0;
13692                                 ctl_clr_ua_all(lun, -1, CTL_UA_THIN_PROV_THRES);
13693                                 set = -1;
13694                         }
13695                         mtx_unlock(&lun->lun_lock);
13696                         if (set != 0 &&
13697                             lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
13698                                 /* Send msg to other side. */
13699                                 bzero(&msg.ua, sizeof(msg.ua));
13700                                 msg.hdr.msg_type = CTL_MSG_UA;
13701                                 msg.hdr.nexus.initid = -1;
13702                                 msg.hdr.nexus.targ_port = -1;
13703                                 msg.hdr.nexus.targ_lun = lun->lun;
13704                                 msg.hdr.nexus.targ_mapped_lun = lun->lun;
13705                                 msg.ua.ua_all = 1;
13706                                 msg.ua.ua_set = (set > 0);
13707                                 msg.ua.ua_type = CTL_UA_THIN_PROV_THRES;
13708                                 memcpy(msg.ua.ua_info, lun->ua_tpt_info, 8);
13709                                 mtx_unlock(&softc->ctl_lock); // XXX
13710                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13711                                     sizeof(msg.ua), M_WAITOK);
13712                                 mtx_lock(&softc->ctl_lock);
13713                         }
13714                 }
13715                 mtx_unlock(&softc->ctl_lock);
13716                 pause("-", CTL_LBP_PERIOD * hz);
13717         }
13718 }
13719
13720 static void
13721 ctl_enqueue_incoming(union ctl_io *io)
13722 {
13723         struct ctl_softc *softc = control_softc;
13724         struct ctl_thread *thr;
13725         u_int idx;
13726
13727         idx = (io->io_hdr.nexus.targ_port * 127 +
13728                io->io_hdr.nexus.initid) % worker_threads;
13729         thr = &softc->threads[idx];
13730         mtx_lock(&thr->queue_lock);
13731         STAILQ_INSERT_TAIL(&thr->incoming_queue, &io->io_hdr, links);
13732         mtx_unlock(&thr->queue_lock);
13733         wakeup(thr);
13734 }
13735
13736 static void
13737 ctl_enqueue_rtr(union ctl_io *io)
13738 {
13739         struct ctl_softc *softc = control_softc;
13740         struct ctl_thread *thr;
13741
13742         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13743         mtx_lock(&thr->queue_lock);
13744         STAILQ_INSERT_TAIL(&thr->rtr_queue, &io->io_hdr, links);
13745         mtx_unlock(&thr->queue_lock);
13746         wakeup(thr);
13747 }
13748
13749 static void
13750 ctl_enqueue_done(union ctl_io *io)
13751 {
13752         struct ctl_softc *softc = control_softc;
13753         struct ctl_thread *thr;
13754
13755         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13756         mtx_lock(&thr->queue_lock);
13757         STAILQ_INSERT_TAIL(&thr->done_queue, &io->io_hdr, links);
13758         mtx_unlock(&thr->queue_lock);
13759         wakeup(thr);
13760 }
13761
13762 static void
13763 ctl_enqueue_isc(union ctl_io *io)
13764 {
13765         struct ctl_softc *softc = control_softc;
13766         struct ctl_thread *thr;
13767
13768         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13769         mtx_lock(&thr->queue_lock);
13770         STAILQ_INSERT_TAIL(&thr->isc_queue, &io->io_hdr, links);
13771         mtx_unlock(&thr->queue_lock);
13772         wakeup(thr);
13773 }
13774
13775 /*
13776  *  vim: ts=8
13777  */