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*/0,
 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                 c = str[i];
4366                 if (isdigit(c))
4367                         c -= '0';
4368                 else if (isalpha(c))
4369                         c -= isupper(c) ? 'A' - 10 : 'a' - 10;
4370                 else
4371                         break;
4372                 if (c >= 16)
4373                         break;
4374                 if ((i & 1) == 0)
4375                         buf[i / 2] |= (c << 4);
4376                 else
4377                         buf[i / 2] |= c;
4378         }
4379         return ((i + 1) / 2);
4380 }
4381
4382 /*
4383  * LUN allocation.
4384  *
4385  * Requirements:
4386  * - caller allocates and zeros LUN storage, or passes in a NULL LUN if he
4387  *   wants us to allocate the LUN and he can block.
4388  * - ctl_softc is always set
4389  * - be_lun is set if the LUN has a backend (needed for disk LUNs)
4390  *
4391  * Returns 0 for success, non-zero (errno) for failure.
4392  */
4393 static int
4394 ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *ctl_lun,
4395               struct ctl_be_lun *const be_lun)
4396 {
4397         struct ctl_lun *nlun, *lun;
4398         struct scsi_vpd_id_descriptor *desc;
4399         struct scsi_vpd_id_t10 *t10id;
4400         const char *eui, *naa, *scsiname, *vendor, *value;
4401         int lun_number, i, lun_malloced;
4402         int devidlen, idlen1, idlen2 = 0, len;
4403
4404         if (be_lun == NULL)
4405                 return (EINVAL);
4406
4407         /*
4408          * We currently only support Direct Access or Processor LUN types.
4409          */
4410         switch (be_lun->lun_type) {
4411         case T_DIRECT:
4412         case T_PROCESSOR:
4413         case T_CDROM:
4414                 break;
4415         case T_SEQUENTIAL:
4416         case T_CHANGER:
4417         default:
4418                 be_lun->lun_config_status(be_lun->be_lun,
4419                                           CTL_LUN_CONFIG_FAILURE);
4420                 break;
4421         }
4422         if (ctl_lun == NULL) {
4423                 lun = malloc(sizeof(*lun), M_CTL, M_WAITOK);
4424                 lun_malloced = 1;
4425         } else {
4426                 lun_malloced = 0;
4427                 lun = ctl_lun;
4428         }
4429
4430         memset(lun, 0, sizeof(*lun));
4431         if (lun_malloced)
4432                 lun->flags = CTL_LUN_MALLOCED;
4433
4434         /* Generate LUN ID. */
4435         devidlen = max(CTL_DEVID_MIN_LEN,
4436             strnlen(be_lun->device_id, CTL_DEVID_LEN));
4437         idlen1 = sizeof(*t10id) + devidlen;
4438         len = sizeof(struct scsi_vpd_id_descriptor) + idlen1;
4439         scsiname = ctl_get_opt(&be_lun->options, "scsiname");
4440         if (scsiname != NULL) {
4441                 idlen2 = roundup2(strlen(scsiname) + 1, 4);
4442                 len += sizeof(struct scsi_vpd_id_descriptor) + idlen2;
4443         }
4444         eui = ctl_get_opt(&be_lun->options, "eui");
4445         if (eui != NULL) {
4446                 len += sizeof(struct scsi_vpd_id_descriptor) + 16;
4447         }
4448         naa = ctl_get_opt(&be_lun->options, "naa");
4449         if (naa != NULL) {
4450                 len += sizeof(struct scsi_vpd_id_descriptor) + 16;
4451         }
4452         lun->lun_devid = malloc(sizeof(struct ctl_devid) + len,
4453             M_CTL, M_WAITOK | M_ZERO);
4454         desc = (struct scsi_vpd_id_descriptor *)lun->lun_devid->data;
4455         desc->proto_codeset = SVPD_ID_CODESET_ASCII;
4456         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_T10;
4457         desc->length = idlen1;
4458         t10id = (struct scsi_vpd_id_t10 *)&desc->identifier[0];
4459         memset(t10id->vendor, ' ', sizeof(t10id->vendor));
4460         if ((vendor = ctl_get_opt(&be_lun->options, "vendor")) == NULL) {
4461                 strncpy((char *)t10id->vendor, CTL_VENDOR, sizeof(t10id->vendor));
4462         } else {
4463                 strncpy(t10id->vendor, vendor,
4464                     min(sizeof(t10id->vendor), strlen(vendor)));
4465         }
4466         strncpy((char *)t10id->vendor_spec_id,
4467             (char *)be_lun->device_id, devidlen);
4468         if (scsiname != NULL) {
4469                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4470                     desc->length);
4471                 desc->proto_codeset = SVPD_ID_CODESET_UTF8;
4472                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4473                     SVPD_ID_TYPE_SCSI_NAME;
4474                 desc->length = idlen2;
4475                 strlcpy(desc->identifier, scsiname, idlen2);
4476         }
4477         if (eui != NULL) {
4478                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4479                     desc->length);
4480                 desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4481                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4482                     SVPD_ID_TYPE_EUI64;
4483                 desc->length = hex2bin(eui, desc->identifier, 16);
4484                 desc->length = desc->length > 12 ? 16 :
4485                     (desc->length > 8 ? 12 : 8);
4486                 len -= 16 - desc->length;
4487         }
4488         if (naa != NULL) {
4489                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4490                     desc->length);
4491                 desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4492                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4493                     SVPD_ID_TYPE_NAA;
4494                 desc->length = hex2bin(naa, desc->identifier, 16);
4495                 desc->length = desc->length > 8 ? 16 : 8;
4496                 len -= 16 - desc->length;
4497         }
4498         lun->lun_devid->len = len;
4499
4500         mtx_lock(&ctl_softc->ctl_lock);
4501         /*
4502          * See if the caller requested a particular LUN number.  If so, see
4503          * if it is available.  Otherwise, allocate the first available LUN.
4504          */
4505         if (be_lun->flags & CTL_LUN_FLAG_ID_REQ) {
4506                 if ((be_lun->req_lun_id > (CTL_MAX_LUNS - 1))
4507                  || (ctl_is_set(ctl_softc->ctl_lun_mask, be_lun->req_lun_id))) {
4508                         mtx_unlock(&ctl_softc->ctl_lock);
4509                         if (be_lun->req_lun_id > (CTL_MAX_LUNS - 1)) {
4510                                 printf("ctl: requested LUN ID %d is higher "
4511                                        "than CTL_MAX_LUNS - 1 (%d)\n",
4512                                        be_lun->req_lun_id, CTL_MAX_LUNS - 1);
4513                         } else {
4514                                 /*
4515                                  * XXX KDM return an error, or just assign
4516                                  * another LUN ID in this case??
4517                                  */
4518                                 printf("ctl: requested LUN ID %d is already "
4519                                        "in use\n", be_lun->req_lun_id);
4520                         }
4521                         if (lun->flags & CTL_LUN_MALLOCED)
4522                                 free(lun, M_CTL);
4523                         be_lun->lun_config_status(be_lun->be_lun,
4524                                                   CTL_LUN_CONFIG_FAILURE);
4525                         return (ENOSPC);
4526                 }
4527                 lun_number = be_lun->req_lun_id;
4528         } else {
4529                 lun_number = ctl_ffz(ctl_softc->ctl_lun_mask, 0, CTL_MAX_LUNS);
4530                 if (lun_number == -1) {
4531                         mtx_unlock(&ctl_softc->ctl_lock);
4532                         printf("ctl: can't allocate LUN, out of LUNs\n");
4533                         if (lun->flags & CTL_LUN_MALLOCED)
4534                                 free(lun, M_CTL);
4535                         be_lun->lun_config_status(be_lun->be_lun,
4536                                                   CTL_LUN_CONFIG_FAILURE);
4537                         return (ENOSPC);
4538                 }
4539         }
4540         ctl_set_mask(ctl_softc->ctl_lun_mask, lun_number);
4541
4542         mtx_init(&lun->lun_lock, "CTL LUN", NULL, MTX_DEF);
4543         lun->lun = lun_number;
4544         lun->be_lun = be_lun;
4545         /*
4546          * The processor LUN is always enabled.  Disk LUNs come on line
4547          * disabled, and must be enabled by the backend.
4548          */
4549         lun->flags |= CTL_LUN_DISABLED;
4550         lun->backend = be_lun->be;
4551         be_lun->ctl_lun = lun;
4552         be_lun->lun_id = lun_number;
4553         atomic_add_int(&be_lun->be->num_luns, 1);
4554         if (be_lun->flags & CTL_LUN_FLAG_EJECTED)
4555                 lun->flags |= CTL_LUN_EJECTED;
4556         if (be_lun->flags & CTL_LUN_FLAG_NO_MEDIA)
4557                 lun->flags |= CTL_LUN_NO_MEDIA;
4558         if (be_lun->flags & CTL_LUN_FLAG_STOPPED)
4559                 lun->flags |= CTL_LUN_STOPPED;
4560
4561         if (be_lun->flags & CTL_LUN_FLAG_PRIMARY)
4562                 lun->flags |= CTL_LUN_PRIMARY_SC;
4563
4564         value = ctl_get_opt(&be_lun->options, "removable");
4565         if (value != NULL) {
4566                 if (strcmp(value, "on") == 0)
4567                         lun->flags |= CTL_LUN_REMOVABLE;
4568         } else if (be_lun->lun_type == T_CDROM)
4569                 lun->flags |= CTL_LUN_REMOVABLE;
4570
4571         lun->ctl_softc = ctl_softc;
4572 #ifdef CTL_TIME_IO
4573         lun->last_busy = getsbinuptime();
4574 #endif
4575         TAILQ_INIT(&lun->ooa_queue);
4576         TAILQ_INIT(&lun->blocked_queue);
4577         STAILQ_INIT(&lun->error_list);
4578         lun->ie_reported = 1;
4579         callout_init_mtx(&lun->ie_callout, &lun->lun_lock, 0);
4580         ctl_tpc_lun_init(lun);
4581
4582         /*
4583          * Initialize the mode and log page index.
4584          */
4585         ctl_init_page_index(lun);
4586         ctl_init_log_page_index(lun);
4587
4588         /*
4589          * Now, before we insert this lun on the lun list, set the lun
4590          * inventory changed UA for all other luns.
4591          */
4592         STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) {
4593                 mtx_lock(&nlun->lun_lock);
4594                 ctl_est_ua_all(nlun, -1, CTL_UA_LUN_CHANGE);
4595                 mtx_unlock(&nlun->lun_lock);
4596         }
4597
4598         STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links);
4599
4600         ctl_softc->ctl_luns[lun_number] = lun;
4601
4602         ctl_softc->num_luns++;
4603
4604         /* Setup statistics gathering */
4605         lun->stats.device_type = be_lun->lun_type;
4606         lun->stats.lun_number = lun_number;
4607         lun->stats.blocksize = be_lun->blocksize;
4608         if (be_lun->blocksize == 0)
4609                 lun->stats.flags = CTL_LUN_STATS_NO_BLOCKSIZE;
4610         for (i = 0;i < CTL_MAX_PORTS;i++)
4611                 lun->stats.ports[i].targ_port = i;
4612
4613         mtx_unlock(&ctl_softc->ctl_lock);
4614
4615         lun->be_lun->lun_config_status(lun->be_lun->be_lun, CTL_LUN_CONFIG_OK);
4616         return (0);
4617 }
4618
4619 /*
4620  * Delete a LUN.
4621  * Assumptions:
4622  * - LUN has already been marked invalid and any pending I/O has been taken
4623  *   care of.
4624  */
4625 static int
4626 ctl_free_lun(struct ctl_lun *lun)
4627 {
4628         struct ctl_softc *softc;
4629         struct ctl_lun *nlun;
4630         int i;
4631
4632         softc = lun->ctl_softc;
4633
4634         mtx_assert(&softc->ctl_lock, MA_OWNED);
4635
4636         STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links);
4637
4638         ctl_clear_mask(softc->ctl_lun_mask, lun->lun);
4639
4640         softc->ctl_luns[lun->lun] = NULL;
4641
4642         if (!TAILQ_EMPTY(&lun->ooa_queue))
4643                 panic("Freeing a LUN %p with outstanding I/O!!\n", lun);
4644
4645         softc->num_luns--;
4646
4647         /*
4648          * Tell the backend to free resources, if this LUN has a backend.
4649          */
4650         atomic_subtract_int(&lun->be_lun->be->num_luns, 1);
4651         lun->be_lun->lun_shutdown(lun->be_lun->be_lun);
4652
4653         lun->ie_reportcnt = UINT32_MAX;
4654         callout_drain(&lun->ie_callout);
4655
4656         ctl_tpc_lun_shutdown(lun);
4657         mtx_destroy(&lun->lun_lock);
4658         free(lun->lun_devid, M_CTL);
4659         for (i = 0; i < CTL_MAX_PORTS; i++)
4660                 free(lun->pending_ua[i], M_CTL);
4661         for (i = 0; i < CTL_MAX_PORTS; i++)
4662                 free(lun->pr_keys[i], M_CTL);
4663         free(lun->write_buffer, M_CTL);
4664         if (lun->flags & CTL_LUN_MALLOCED)
4665                 free(lun, M_CTL);
4666
4667         STAILQ_FOREACH(nlun, &softc->lun_list, links) {
4668                 mtx_lock(&nlun->lun_lock);
4669                 ctl_est_ua_all(nlun, -1, CTL_UA_LUN_CHANGE);
4670                 mtx_unlock(&nlun->lun_lock);
4671         }
4672
4673         return (0);
4674 }
4675
4676 static void
4677 ctl_create_lun(struct ctl_be_lun *be_lun)
4678 {
4679
4680         /*
4681          * ctl_alloc_lun() should handle all potential failure cases.
4682          */
4683         ctl_alloc_lun(control_softc, NULL, be_lun);
4684 }
4685
4686 int
4687 ctl_add_lun(struct ctl_be_lun *be_lun)
4688 {
4689         struct ctl_softc *softc = control_softc;
4690
4691         mtx_lock(&softc->ctl_lock);
4692         STAILQ_INSERT_TAIL(&softc->pending_lun_queue, be_lun, links);
4693         mtx_unlock(&softc->ctl_lock);
4694         wakeup(&softc->pending_lun_queue);
4695
4696         return (0);
4697 }
4698
4699 int
4700 ctl_enable_lun(struct ctl_be_lun *be_lun)
4701 {
4702         struct ctl_softc *softc;
4703         struct ctl_port *port, *nport;
4704         struct ctl_lun *lun;
4705         int retval;
4706
4707         lun = (struct ctl_lun *)be_lun->ctl_lun;
4708         softc = lun->ctl_softc;
4709
4710         mtx_lock(&softc->ctl_lock);
4711         mtx_lock(&lun->lun_lock);
4712         if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4713                 /*
4714                  * eh?  Why did we get called if the LUN is already
4715                  * enabled?
4716                  */
4717                 mtx_unlock(&lun->lun_lock);
4718                 mtx_unlock(&softc->ctl_lock);
4719                 return (0);
4720         }
4721         lun->flags &= ~CTL_LUN_DISABLED;
4722         mtx_unlock(&lun->lun_lock);
4723
4724         STAILQ_FOREACH_SAFE(port, &softc->port_list, links, nport) {
4725                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0 ||
4726                     port->lun_map != NULL || port->lun_enable == NULL)
4727                         continue;
4728
4729                 /*
4730                  * Drop the lock while we call the FETD's enable routine.
4731                  * This can lead to a callback into CTL (at least in the
4732                  * case of the internal initiator frontend.
4733                  */
4734                 mtx_unlock(&softc->ctl_lock);
4735                 retval = port->lun_enable(port->targ_lun_arg, lun->lun);
4736                 mtx_lock(&softc->ctl_lock);
4737                 if (retval != 0) {
4738                         printf("%s: FETD %s port %d returned error "
4739                                "%d for lun_enable on lun %jd\n",
4740                                __func__, port->port_name, port->targ_port,
4741                                retval, (intmax_t)lun->lun);
4742                 }
4743         }
4744
4745         mtx_unlock(&softc->ctl_lock);
4746         ctl_isc_announce_lun(lun);
4747
4748         return (0);
4749 }
4750
4751 int
4752 ctl_disable_lun(struct ctl_be_lun *be_lun)
4753 {
4754         struct ctl_softc *softc;
4755         struct ctl_port *port;
4756         struct ctl_lun *lun;
4757         int retval;
4758
4759         lun = (struct ctl_lun *)be_lun->ctl_lun;
4760         softc = lun->ctl_softc;
4761
4762         mtx_lock(&softc->ctl_lock);
4763         mtx_lock(&lun->lun_lock);
4764         if (lun->flags & CTL_LUN_DISABLED) {
4765                 mtx_unlock(&lun->lun_lock);
4766                 mtx_unlock(&softc->ctl_lock);
4767                 return (0);
4768         }
4769         lun->flags |= CTL_LUN_DISABLED;
4770         mtx_unlock(&lun->lun_lock);
4771
4772         STAILQ_FOREACH(port, &softc->port_list, links) {
4773                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0 ||
4774                     port->lun_map != NULL || port->lun_disable == NULL)
4775                         continue;
4776
4777                 /*
4778                  * Drop the lock before we call the frontend's disable
4779                  * routine, to avoid lock order reversals.
4780                  *
4781                  * XXX KDM what happens if the frontend list changes while
4782                  * we're traversing it?  It's unlikely, but should be handled.
4783                  */
4784                 mtx_unlock(&softc->ctl_lock);
4785                 retval = port->lun_disable(port->targ_lun_arg, lun->lun);
4786                 mtx_lock(&softc->ctl_lock);
4787                 if (retval != 0) {
4788                         printf("%s: FETD %s port %d returned error "
4789                                "%d for lun_disable on lun %jd\n",
4790                                __func__, port->port_name, port->targ_port,
4791                                retval, (intmax_t)lun->lun);
4792                 }
4793         }
4794
4795         mtx_unlock(&softc->ctl_lock);
4796         ctl_isc_announce_lun(lun);
4797
4798         return (0);
4799 }
4800
4801 int
4802 ctl_start_lun(struct ctl_be_lun *be_lun)
4803 {
4804         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4805
4806         mtx_lock(&lun->lun_lock);
4807         lun->flags &= ~CTL_LUN_STOPPED;
4808         mtx_unlock(&lun->lun_lock);
4809         return (0);
4810 }
4811
4812 int
4813 ctl_stop_lun(struct ctl_be_lun *be_lun)
4814 {
4815         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4816
4817         mtx_lock(&lun->lun_lock);
4818         lun->flags |= CTL_LUN_STOPPED;
4819         mtx_unlock(&lun->lun_lock);
4820         return (0);
4821 }
4822
4823 int
4824 ctl_lun_no_media(struct ctl_be_lun *be_lun)
4825 {
4826         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4827
4828         mtx_lock(&lun->lun_lock);
4829         lun->flags |= CTL_LUN_NO_MEDIA;
4830         mtx_unlock(&lun->lun_lock);
4831         return (0);
4832 }
4833
4834 int
4835 ctl_lun_has_media(struct ctl_be_lun *be_lun)
4836 {
4837         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4838         union ctl_ha_msg msg;
4839
4840         mtx_lock(&lun->lun_lock);
4841         lun->flags &= ~(CTL_LUN_NO_MEDIA | CTL_LUN_EJECTED);
4842         if (lun->flags & CTL_LUN_REMOVABLE)
4843                 ctl_est_ua_all(lun, -1, CTL_UA_MEDIUM_CHANGE);
4844         mtx_unlock(&lun->lun_lock);
4845         if ((lun->flags & CTL_LUN_REMOVABLE) &&
4846             lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
4847                 bzero(&msg.ua, sizeof(msg.ua));
4848                 msg.hdr.msg_type = CTL_MSG_UA;
4849                 msg.hdr.nexus.initid = -1;
4850                 msg.hdr.nexus.targ_port = -1;
4851                 msg.hdr.nexus.targ_lun = lun->lun;
4852                 msg.hdr.nexus.targ_mapped_lun = lun->lun;
4853                 msg.ua.ua_all = 1;
4854                 msg.ua.ua_set = 1;
4855                 msg.ua.ua_type = CTL_UA_MEDIUM_CHANGE;
4856                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg.ua),
4857                     M_WAITOK);
4858         }
4859         return (0);
4860 }
4861
4862 int
4863 ctl_lun_ejected(struct ctl_be_lun *be_lun)
4864 {
4865         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4866
4867         mtx_lock(&lun->lun_lock);
4868         lun->flags |= CTL_LUN_EJECTED;
4869         mtx_unlock(&lun->lun_lock);
4870         return (0);
4871 }
4872
4873 int
4874 ctl_lun_primary(struct ctl_be_lun *be_lun)
4875 {
4876         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4877
4878         mtx_lock(&lun->lun_lock);
4879         lun->flags |= CTL_LUN_PRIMARY_SC;
4880         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
4881         mtx_unlock(&lun->lun_lock);
4882         ctl_isc_announce_lun(lun);
4883         return (0);
4884 }
4885
4886 int
4887 ctl_lun_secondary(struct ctl_be_lun *be_lun)
4888 {
4889         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4890
4891         mtx_lock(&lun->lun_lock);
4892         lun->flags &= ~CTL_LUN_PRIMARY_SC;
4893         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
4894         mtx_unlock(&lun->lun_lock);
4895         ctl_isc_announce_lun(lun);
4896         return (0);
4897 }
4898
4899 int
4900 ctl_invalidate_lun(struct ctl_be_lun *be_lun)
4901 {
4902         struct ctl_softc *softc;
4903         struct ctl_lun *lun;
4904
4905         lun = (struct ctl_lun *)be_lun->ctl_lun;
4906         softc = lun->ctl_softc;
4907
4908         mtx_lock(&lun->lun_lock);
4909
4910         /*
4911          * The LUN needs to be disabled before it can be marked invalid.
4912          */
4913         if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4914                 mtx_unlock(&lun->lun_lock);
4915                 return (-1);
4916         }
4917         /*
4918          * Mark the LUN invalid.
4919          */
4920         lun->flags |= CTL_LUN_INVALID;
4921
4922         /*
4923          * If there is nothing in the OOA queue, go ahead and free the LUN.
4924          * If we have something in the OOA queue, we'll free it when the
4925          * last I/O completes.
4926          */
4927         if (TAILQ_EMPTY(&lun->ooa_queue)) {
4928                 mtx_unlock(&lun->lun_lock);
4929                 mtx_lock(&softc->ctl_lock);
4930                 ctl_free_lun(lun);
4931                 mtx_unlock(&softc->ctl_lock);
4932         } else
4933                 mtx_unlock(&lun->lun_lock);
4934
4935         return (0);
4936 }
4937
4938 void
4939 ctl_lun_capacity_changed(struct ctl_be_lun *be_lun)
4940 {
4941         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4942         union ctl_ha_msg msg;
4943
4944         mtx_lock(&lun->lun_lock);
4945         ctl_est_ua_all(lun, -1, CTL_UA_CAPACITY_CHANGE);
4946         mtx_unlock(&lun->lun_lock);
4947         if (lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
4948                 /* Send msg to other side. */
4949                 bzero(&msg.ua, sizeof(msg.ua));
4950                 msg.hdr.msg_type = CTL_MSG_UA;
4951                 msg.hdr.nexus.initid = -1;
4952                 msg.hdr.nexus.targ_port = -1;
4953                 msg.hdr.nexus.targ_lun = lun->lun;
4954                 msg.hdr.nexus.targ_mapped_lun = lun->lun;
4955                 msg.ua.ua_all = 1;
4956                 msg.ua.ua_set = 1;
4957                 msg.ua.ua_type = CTL_UA_CAPACITY_CHANGE;
4958                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg.ua),
4959                     M_WAITOK);
4960         }
4961 }
4962
4963 /*
4964  * Backend "memory move is complete" callback for requests that never
4965  * make it down to say RAIDCore's configuration code.
4966  */
4967 int
4968 ctl_config_move_done(union ctl_io *io)
4969 {
4970         int retval;
4971
4972         CTL_DEBUG_PRINT(("ctl_config_move_done\n"));
4973         KASSERT(io->io_hdr.io_type == CTL_IO_SCSI,
4974             ("Config I/O type isn't CTL_IO_SCSI (%d)!", io->io_hdr.io_type));
4975
4976         if ((io->io_hdr.port_status != 0) &&
4977             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
4978              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
4979                 /*
4980                  * For hardware error sense keys, the sense key
4981                  * specific value is defined to be a retry count,
4982                  * but we use it to pass back an internal FETD
4983                  * error code.  XXX KDM  Hopefully the FETD is only
4984                  * using 16 bits for an error code, since that's
4985                  * all the space we have in the sks field.
4986                  */
4987                 ctl_set_internal_failure(&io->scsiio,
4988                                          /*sks_valid*/ 1,
4989                                          /*retry_count*/
4990                                          io->io_hdr.port_status);
4991         }
4992
4993         if (ctl_debug & CTL_DEBUG_CDB_DATA)
4994                 ctl_data_print(io);
4995         if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN) ||
4996             ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
4997              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS) ||
4998             ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) {
4999                 /*
5000                  * XXX KDM just assuming a single pointer here, and not a
5001                  * S/G list.  If we start using S/G lists for config data,
5002                  * we'll need to know how to clean them up here as well.
5003                  */
5004                 if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5005                         free(io->scsiio.kern_data_ptr, M_CTL);
5006                 ctl_done(io);
5007                 retval = CTL_RETVAL_COMPLETE;
5008         } else {
5009                 /*
5010                  * XXX KDM now we need to continue data movement.  Some
5011                  * options:
5012                  * - call ctl_scsiio() again?  We don't do this for data
5013                  *   writes, because for those at least we know ahead of
5014                  *   time where the write will go and how long it is.  For
5015                  *   config writes, though, that information is largely
5016                  *   contained within the write itself, thus we need to
5017                  *   parse out the data again.
5018                  *
5019                  * - Call some other function once the data is in?
5020                  */
5021
5022                 /*
5023                  * XXX KDM call ctl_scsiio() again for now, and check flag
5024                  * bits to see whether we're allocated or not.
5025                  */
5026                 retval = ctl_scsiio(&io->scsiio);
5027         }
5028         return (retval);
5029 }
5030
5031 /*
5032  * This gets called by a backend driver when it is done with a
5033  * data_submit method.
5034  */
5035 void
5036 ctl_data_submit_done(union ctl_io *io)
5037 {
5038         /*
5039          * If the IO_CONT flag is set, we need to call the supplied
5040          * function to continue processing the I/O, instead of completing
5041          * the I/O just yet.
5042          *
5043          * If there is an error, though, we don't want to keep processing.
5044          * Instead, just send status back to the initiator.
5045          */
5046         if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5047             (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5048             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5049              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5050                 io->scsiio.io_cont(io);
5051                 return;
5052         }
5053         ctl_done(io);
5054 }
5055
5056 /*
5057  * This gets called by a backend driver when it is done with a
5058  * configuration write.
5059  */
5060 void
5061 ctl_config_write_done(union ctl_io *io)
5062 {
5063         uint8_t *buf;
5064
5065         /*
5066          * If the IO_CONT flag is set, we need to call the supplied
5067          * function to continue processing the I/O, instead of completing
5068          * the I/O just yet.
5069          *
5070          * If there is an error, though, we don't want to keep processing.
5071          * Instead, just send status back to the initiator.
5072          */
5073         if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5074             (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5075             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5076              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5077                 io->scsiio.io_cont(io);
5078                 return;
5079         }
5080         /*
5081          * Since a configuration write can be done for commands that actually
5082          * have data allocated, like write buffer, and commands that have
5083          * no data, like start/stop unit, we need to check here.
5084          */
5085         if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5086                 buf = io->scsiio.kern_data_ptr;
5087         else
5088                 buf = NULL;
5089         ctl_done(io);
5090         if (buf)
5091                 free(buf, M_CTL);
5092 }
5093
5094 void
5095 ctl_config_read_done(union ctl_io *io)
5096 {
5097         uint8_t *buf;
5098
5099         /*
5100          * If there is some error -- we are done, skip data transfer.
5101          */
5102         if ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0 ||
5103             ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
5104              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) {
5105                 if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5106                         buf = io->scsiio.kern_data_ptr;
5107                 else
5108                         buf = NULL;
5109                 ctl_done(io);
5110                 if (buf)
5111                         free(buf, M_CTL);
5112                 return;
5113         }
5114
5115         /*
5116          * If the IO_CONT flag is set, we need to call the supplied
5117          * function to continue processing the I/O, instead of completing
5118          * the I/O just yet.
5119          */
5120         if (io->io_hdr.flags & CTL_FLAG_IO_CONT) {
5121                 io->scsiio.io_cont(io);
5122                 return;
5123         }
5124
5125         ctl_datamove(io);
5126 }
5127
5128 /*
5129  * SCSI release command.
5130  */
5131 int
5132 ctl_scsi_release(struct ctl_scsiio *ctsio)
5133 {
5134         struct ctl_lun *lun;
5135         uint32_t residx;
5136
5137         CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
5138
5139         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5140         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5141
5142         /*
5143          * XXX KDM right now, we only support LUN reservation.  We don't
5144          * support 3rd party reservations, or extent reservations, which
5145          * might actually need the parameter list.  If we've gotten this
5146          * far, we've got a LUN reservation.  Anything else got kicked out
5147          * above.  So, according to SPC, ignore the length.
5148          */
5149
5150         mtx_lock(&lun->lun_lock);
5151
5152         /*
5153          * According to SPC, it is not an error for an intiator to attempt
5154          * to release a reservation on a LUN that isn't reserved, or that
5155          * is reserved by another initiator.  The reservation can only be
5156          * released, though, by the initiator who made it or by one of
5157          * several reset type events.
5158          */
5159         if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == residx))
5160                         lun->flags &= ~CTL_LUN_RESERVED;
5161
5162         mtx_unlock(&lun->lun_lock);
5163
5164         ctl_set_success(ctsio);
5165         ctl_done((union ctl_io *)ctsio);
5166         return (CTL_RETVAL_COMPLETE);
5167 }
5168
5169 int
5170 ctl_scsi_reserve(struct ctl_scsiio *ctsio)
5171 {
5172         struct ctl_lun *lun;
5173         uint32_t residx;
5174
5175         CTL_DEBUG_PRINT(("ctl_reserve\n"));
5176
5177         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5178         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5179
5180         /*
5181          * XXX KDM right now, we only support LUN reservation.  We don't
5182          * support 3rd party reservations, or extent reservations, which
5183          * might actually need the parameter list.  If we've gotten this
5184          * far, we've got a LUN reservation.  Anything else got kicked out
5185          * above.  So, according to SPC, ignore the length.
5186          */
5187
5188         mtx_lock(&lun->lun_lock);
5189         if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx != residx)) {
5190                 ctl_set_reservation_conflict(ctsio);
5191                 goto bailout;
5192         }
5193
5194         /* SPC-3 exceptions to SPC-2 RESERVE and RELEASE behavior. */
5195         if (lun->flags & CTL_LUN_PR_RESERVED) {
5196                 ctl_set_success(ctsio);
5197                 goto bailout;
5198         }
5199
5200         lun->flags |= CTL_LUN_RESERVED;
5201         lun->res_idx = residx;
5202         ctl_set_success(ctsio);
5203
5204 bailout:
5205         mtx_unlock(&lun->lun_lock);
5206         ctl_done((union ctl_io *)ctsio);
5207         return (CTL_RETVAL_COMPLETE);
5208 }
5209
5210 int
5211 ctl_start_stop(struct ctl_scsiio *ctsio)
5212 {
5213         struct scsi_start_stop_unit *cdb;
5214         struct ctl_lun *lun;
5215         int retval;
5216
5217         CTL_DEBUG_PRINT(("ctl_start_stop\n"));
5218
5219         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5220         cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
5221
5222         if ((cdb->how & SSS_PC_MASK) == 0) {
5223                 if ((lun->flags & CTL_LUN_PR_RESERVED) &&
5224                     (cdb->how & SSS_START) == 0) {
5225                         uint32_t residx;
5226
5227                         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5228                         if (ctl_get_prkey(lun, residx) == 0 ||
5229                             (lun->pr_res_idx != residx && lun->pr_res_type < 4)) {
5230
5231                                 ctl_set_reservation_conflict(ctsio);
5232                                 ctl_done((union ctl_io *)ctsio);
5233                                 return (CTL_RETVAL_COMPLETE);
5234                         }
5235                 }
5236
5237                 if ((cdb->how & SSS_LOEJ) &&
5238                     (lun->flags & CTL_LUN_REMOVABLE) == 0) {
5239                         ctl_set_invalid_field(ctsio,
5240                                               /*sks_valid*/ 1,
5241                                               /*command*/ 1,
5242                                               /*field*/ 4,
5243                                               /*bit_valid*/ 1,
5244                                               /*bit*/ 1);
5245                         ctl_done((union ctl_io *)ctsio);
5246                         return (CTL_RETVAL_COMPLETE);
5247                 }
5248
5249                 if ((cdb->how & SSS_START) == 0 && (cdb->how & SSS_LOEJ) &&
5250                     lun->prevent_count > 0) {
5251                         /* "Medium removal prevented" */
5252                         ctl_set_sense(ctsio, /*current_error*/ 1,
5253                             /*sense_key*/(lun->flags & CTL_LUN_NO_MEDIA) ?
5254                              SSD_KEY_NOT_READY : SSD_KEY_ILLEGAL_REQUEST,
5255                             /*asc*/ 0x53, /*ascq*/ 0x02, SSD_ELEM_NONE);
5256                         ctl_done((union ctl_io *)ctsio);
5257                         return (CTL_RETVAL_COMPLETE);
5258                 }
5259         }
5260
5261         retval = lun->backend->config_write((union ctl_io *)ctsio);
5262         return (retval);
5263 }
5264
5265 int
5266 ctl_prevent_allow(struct ctl_scsiio *ctsio)
5267 {
5268         struct ctl_lun *lun;
5269         struct scsi_prevent *cdb;
5270         int retval;
5271         uint32_t initidx;
5272
5273         CTL_DEBUG_PRINT(("ctl_prevent_allow\n"));
5274
5275         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5276         cdb = (struct scsi_prevent *)ctsio->cdb;
5277
5278         if ((lun->flags & CTL_LUN_REMOVABLE) == 0) {
5279                 ctl_set_invalid_opcode(ctsio);
5280                 ctl_done((union ctl_io *)ctsio);
5281                 return (CTL_RETVAL_COMPLETE);
5282         }
5283
5284         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5285         mtx_lock(&lun->lun_lock);
5286         if ((cdb->how & PR_PREVENT) &&
5287             ctl_is_set(lun->prevent, initidx) == 0) {
5288                 ctl_set_mask(lun->prevent, initidx);
5289                 lun->prevent_count++;
5290         } else if ((cdb->how & PR_PREVENT) == 0 &&
5291             ctl_is_set(lun->prevent, initidx)) {
5292                 ctl_clear_mask(lun->prevent, initidx);
5293                 lun->prevent_count--;
5294         }
5295         mtx_unlock(&lun->lun_lock);
5296         retval = lun->backend->config_write((union ctl_io *)ctsio);
5297         return (retval);
5298 }
5299
5300 /*
5301  * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
5302  * we don't really do anything with the LBA and length fields if the user
5303  * passes them in.  Instead we'll just flush out the cache for the entire
5304  * LUN.
5305  */
5306 int
5307 ctl_sync_cache(struct ctl_scsiio *ctsio)
5308 {
5309         struct ctl_lun *lun;
5310         struct ctl_softc *softc;
5311         struct ctl_lba_len_flags *lbalen;
5312         uint64_t starting_lba;
5313         uint32_t block_count;
5314         int retval;
5315         uint8_t byte2;
5316
5317         CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
5318
5319         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5320         softc = lun->ctl_softc;
5321         retval = 0;
5322
5323         switch (ctsio->cdb[0]) {
5324         case SYNCHRONIZE_CACHE: {
5325                 struct scsi_sync_cache *cdb;
5326                 cdb = (struct scsi_sync_cache *)ctsio->cdb;
5327
5328                 starting_lba = scsi_4btoul(cdb->begin_lba);
5329                 block_count = scsi_2btoul(cdb->lb_count);
5330                 byte2 = cdb->byte2;
5331                 break;
5332         }
5333         case SYNCHRONIZE_CACHE_16: {
5334                 struct scsi_sync_cache_16 *cdb;
5335                 cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5336
5337                 starting_lba = scsi_8btou64(cdb->begin_lba);
5338                 block_count = scsi_4btoul(cdb->lb_count);
5339                 byte2 = cdb->byte2;
5340                 break;
5341         }
5342         default:
5343                 ctl_set_invalid_opcode(ctsio);
5344                 ctl_done((union ctl_io *)ctsio);
5345                 goto bailout;
5346                 break; /* NOTREACHED */
5347         }
5348
5349         /*
5350          * We check the LBA and length, but don't do anything with them.
5351          * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5352          * get flushed.  This check will just help satisfy anyone who wants
5353          * to see an error for an out of range LBA.
5354          */
5355         if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5356                 ctl_set_lba_out_of_range(ctsio,
5357                     MAX(starting_lba, lun->be_lun->maxlba + 1));
5358                 ctl_done((union ctl_io *)ctsio);
5359                 goto bailout;
5360         }
5361
5362         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5363         lbalen->lba = starting_lba;
5364         lbalen->len = block_count;
5365         lbalen->flags = byte2;
5366         retval = lun->backend->config_write((union ctl_io *)ctsio);
5367
5368 bailout:
5369         return (retval);
5370 }
5371
5372 int
5373 ctl_format(struct ctl_scsiio *ctsio)
5374 {
5375         struct scsi_format *cdb;
5376         struct ctl_lun *lun;
5377         int length, defect_list_len;
5378
5379         CTL_DEBUG_PRINT(("ctl_format\n"));
5380
5381         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5382
5383         cdb = (struct scsi_format *)ctsio->cdb;
5384
5385         length = 0;
5386         if (cdb->byte2 & SF_FMTDATA) {
5387                 if (cdb->byte2 & SF_LONGLIST)
5388                         length = sizeof(struct scsi_format_header_long);
5389                 else
5390                         length = sizeof(struct scsi_format_header_short);
5391         }
5392
5393         if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5394          && (length > 0)) {
5395                 ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5396                 ctsio->kern_data_len = length;
5397                 ctsio->kern_total_len = length;
5398                 ctsio->kern_data_resid = 0;
5399                 ctsio->kern_rel_offset = 0;
5400                 ctsio->kern_sg_entries = 0;
5401                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5402                 ctsio->be_move_done = ctl_config_move_done;
5403                 ctl_datamove((union ctl_io *)ctsio);
5404
5405                 return (CTL_RETVAL_COMPLETE);
5406         }
5407
5408         defect_list_len = 0;
5409
5410         if (cdb->byte2 & SF_FMTDATA) {
5411                 if (cdb->byte2 & SF_LONGLIST) {
5412                         struct scsi_format_header_long *header;
5413
5414                         header = (struct scsi_format_header_long *)
5415                                 ctsio->kern_data_ptr;
5416
5417                         defect_list_len = scsi_4btoul(header->defect_list_len);
5418                         if (defect_list_len != 0) {
5419                                 ctl_set_invalid_field(ctsio,
5420                                                       /*sks_valid*/ 1,
5421                                                       /*command*/ 0,
5422                                                       /*field*/ 2,
5423                                                       /*bit_valid*/ 0,
5424                                                       /*bit*/ 0);
5425                                 goto bailout;
5426                         }
5427                 } else {
5428                         struct scsi_format_header_short *header;
5429
5430                         header = (struct scsi_format_header_short *)
5431                                 ctsio->kern_data_ptr;
5432
5433                         defect_list_len = scsi_2btoul(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                 }
5444         }
5445
5446         ctl_set_success(ctsio);
5447 bailout:
5448
5449         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5450                 free(ctsio->kern_data_ptr, M_CTL);
5451                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5452         }
5453
5454         ctl_done((union ctl_io *)ctsio);
5455         return (CTL_RETVAL_COMPLETE);
5456 }
5457
5458 int
5459 ctl_read_buffer(struct ctl_scsiio *ctsio)
5460 {
5461         struct ctl_lun *lun;
5462         uint64_t buffer_offset;
5463         uint32_t len;
5464         uint8_t byte2;
5465         static uint8_t descr[4];
5466         static uint8_t echo_descr[4] = { 0 };
5467
5468         CTL_DEBUG_PRINT(("ctl_read_buffer\n"));
5469         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5470         switch (ctsio->cdb[0]) {
5471         case READ_BUFFER: {
5472                 struct scsi_read_buffer *cdb;
5473
5474                 cdb = (struct scsi_read_buffer *)ctsio->cdb;
5475                 buffer_offset = scsi_3btoul(cdb->offset);
5476                 len = scsi_3btoul(cdb->length);
5477                 byte2 = cdb->byte2;
5478                 break;
5479         }
5480         case READ_BUFFER_16: {
5481                 struct scsi_read_buffer_16 *cdb;
5482
5483                 cdb = (struct scsi_read_buffer_16 *)ctsio->cdb;
5484                 buffer_offset = scsi_8btou64(cdb->offset);
5485                 len = scsi_4btoul(cdb->length);
5486                 byte2 = cdb->byte2;
5487                 break;
5488         }
5489         default: /* This shouldn't happen. */
5490                 ctl_set_invalid_opcode(ctsio);
5491                 ctl_done((union ctl_io *)ctsio);
5492                 return (CTL_RETVAL_COMPLETE);
5493         }
5494
5495         if ((byte2 & RWB_MODE) != RWB_MODE_DATA &&
5496             (byte2 & RWB_MODE) != RWB_MODE_ECHO_DESCR &&
5497             (byte2 & RWB_MODE) != RWB_MODE_DESCR) {
5498                 ctl_set_invalid_field(ctsio,
5499                                       /*sks_valid*/ 1,
5500                                       /*command*/ 1,
5501                                       /*field*/ 1,
5502                                       /*bit_valid*/ 1,
5503                                       /*bit*/ 4);
5504                 ctl_done((union ctl_io *)ctsio);
5505                 return (CTL_RETVAL_COMPLETE);
5506         }
5507
5508         if (buffer_offset > CTL_WRITE_BUFFER_SIZE ||
5509             buffer_offset + len > CTL_WRITE_BUFFER_SIZE) {
5510                 ctl_set_invalid_field(ctsio,
5511                                       /*sks_valid*/ 1,
5512                                       /*command*/ 1,
5513                                       /*field*/ 6,
5514                                       /*bit_valid*/ 0,
5515                                       /*bit*/ 0);
5516                 ctl_done((union ctl_io *)ctsio);
5517                 return (CTL_RETVAL_COMPLETE);
5518         }
5519
5520         if ((byte2 & RWB_MODE) == RWB_MODE_DESCR) {
5521                 descr[0] = 0;
5522                 scsi_ulto3b(CTL_WRITE_BUFFER_SIZE, &descr[1]);
5523                 ctsio->kern_data_ptr = descr;
5524                 len = min(len, sizeof(descr));
5525         } else if ((byte2 & RWB_MODE) == RWB_MODE_ECHO_DESCR) {
5526                 ctsio->kern_data_ptr = echo_descr;
5527                 len = min(len, sizeof(echo_descr));
5528         } else {
5529                 if (lun->write_buffer == NULL) {
5530                         lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE,
5531                             M_CTL, M_WAITOK);
5532                 }
5533                 ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5534         }
5535         ctsio->kern_data_len = len;
5536         ctsio->kern_total_len = len;
5537         ctsio->kern_data_resid = 0;
5538         ctsio->kern_rel_offset = 0;
5539         ctsio->kern_sg_entries = 0;
5540         ctl_set_success(ctsio);
5541         ctsio->be_move_done = ctl_config_move_done;
5542         ctl_datamove((union ctl_io *)ctsio);
5543         return (CTL_RETVAL_COMPLETE);
5544 }
5545
5546 int
5547 ctl_write_buffer(struct ctl_scsiio *ctsio)
5548 {
5549         struct scsi_write_buffer *cdb;
5550         struct ctl_lun *lun;
5551         int buffer_offset, len;
5552
5553         CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5554
5555         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5556         cdb = (struct scsi_write_buffer *)ctsio->cdb;
5557
5558         if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA) {
5559                 ctl_set_invalid_field(ctsio,
5560                                       /*sks_valid*/ 1,
5561                                       /*command*/ 1,
5562                                       /*field*/ 1,
5563                                       /*bit_valid*/ 1,
5564                                       /*bit*/ 4);
5565                 ctl_done((union ctl_io *)ctsio);
5566                 return (CTL_RETVAL_COMPLETE);
5567         }
5568
5569         len = scsi_3btoul(cdb->length);
5570         buffer_offset = scsi_3btoul(cdb->offset);
5571
5572         if (buffer_offset + len > CTL_WRITE_BUFFER_SIZE) {
5573                 ctl_set_invalid_field(ctsio,
5574                                       /*sks_valid*/ 1,
5575                                       /*command*/ 1,
5576                                       /*field*/ 6,
5577                                       /*bit_valid*/ 0,
5578                                       /*bit*/ 0);
5579                 ctl_done((union ctl_io *)ctsio);
5580                 return (CTL_RETVAL_COMPLETE);
5581         }
5582
5583         /*
5584          * If we've got a kernel request that hasn't been malloced yet,
5585          * malloc it and tell the caller the data buffer is here.
5586          */
5587         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5588                 if (lun->write_buffer == NULL) {
5589                         lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE,
5590                             M_CTL, M_WAITOK);
5591                 }
5592                 ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5593                 ctsio->kern_data_len = len;
5594                 ctsio->kern_total_len = len;
5595                 ctsio->kern_data_resid = 0;
5596                 ctsio->kern_rel_offset = 0;
5597                 ctsio->kern_sg_entries = 0;
5598                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5599                 ctsio->be_move_done = ctl_config_move_done;
5600                 ctl_datamove((union ctl_io *)ctsio);
5601
5602                 return (CTL_RETVAL_COMPLETE);
5603         }
5604
5605         ctl_set_success(ctsio);
5606         ctl_done((union ctl_io *)ctsio);
5607         return (CTL_RETVAL_COMPLETE);
5608 }
5609
5610 int
5611 ctl_write_same(struct ctl_scsiio *ctsio)
5612 {
5613         struct ctl_lun *lun;
5614         struct ctl_lba_len_flags *lbalen;
5615         uint64_t lba;
5616         uint32_t num_blocks;
5617         int len, retval;
5618         uint8_t byte2;
5619
5620         CTL_DEBUG_PRINT(("ctl_write_same\n"));
5621
5622         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5623
5624         switch (ctsio->cdb[0]) {
5625         case WRITE_SAME_10: {
5626                 struct scsi_write_same_10 *cdb;
5627
5628                 cdb = (struct scsi_write_same_10 *)ctsio->cdb;
5629
5630                 lba = scsi_4btoul(cdb->addr);
5631                 num_blocks = scsi_2btoul(cdb->length);
5632                 byte2 = cdb->byte2;
5633                 break;
5634         }
5635         case WRITE_SAME_16: {
5636                 struct scsi_write_same_16 *cdb;
5637
5638                 cdb = (struct scsi_write_same_16 *)ctsio->cdb;
5639
5640                 lba = scsi_8btou64(cdb->addr);
5641                 num_blocks = scsi_4btoul(cdb->length);
5642                 byte2 = cdb->byte2;
5643                 break;
5644         }
5645         default:
5646                 /*
5647                  * We got a command we don't support.  This shouldn't
5648                  * happen, commands should be filtered out above us.
5649                  */
5650                 ctl_set_invalid_opcode(ctsio);
5651                 ctl_done((union ctl_io *)ctsio);
5652
5653                 return (CTL_RETVAL_COMPLETE);
5654                 break; /* NOTREACHED */
5655         }
5656
5657         /* ANCHOR flag can be used only together with UNMAP */
5658         if ((byte2 & SWS_UNMAP) == 0 && (byte2 & SWS_ANCHOR) != 0) {
5659                 ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
5660                     /*command*/ 1, /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 0);
5661                 ctl_done((union ctl_io *)ctsio);
5662                 return (CTL_RETVAL_COMPLETE);
5663         }
5664
5665         /*
5666          * The first check is to make sure we're in bounds, the second
5667          * check is to catch wrap-around problems.  If the lba + num blocks
5668          * is less than the lba, then we've wrapped around and the block
5669          * range is invalid anyway.
5670          */
5671         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5672          || ((lba + num_blocks) < lba)) {
5673                 ctl_set_lba_out_of_range(ctsio,
5674                     MAX(lba, lun->be_lun->maxlba + 1));
5675                 ctl_done((union ctl_io *)ctsio);
5676                 return (CTL_RETVAL_COMPLETE);
5677         }
5678
5679         /* Zero number of blocks means "to the last logical block" */
5680         if (num_blocks == 0) {
5681                 if ((lun->be_lun->maxlba + 1) - lba > UINT32_MAX) {
5682                         ctl_set_invalid_field(ctsio,
5683                                               /*sks_valid*/ 0,
5684                                               /*command*/ 1,
5685                                               /*field*/ 0,
5686                                               /*bit_valid*/ 0,
5687                                               /*bit*/ 0);
5688                         ctl_done((union ctl_io *)ctsio);
5689                         return (CTL_RETVAL_COMPLETE);
5690                 }
5691                 num_blocks = (lun->be_lun->maxlba + 1) - lba;
5692         }
5693
5694         len = lun->be_lun->blocksize;
5695
5696         /*
5697          * If we've got a kernel request that hasn't been malloced yet,
5698          * malloc it and tell the caller the data buffer is here.
5699          */
5700         if ((byte2 & SWS_NDOB) == 0 &&
5701             (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5702                 ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);
5703                 ctsio->kern_data_len = len;
5704                 ctsio->kern_total_len = len;
5705                 ctsio->kern_data_resid = 0;
5706                 ctsio->kern_rel_offset = 0;
5707                 ctsio->kern_sg_entries = 0;
5708                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5709                 ctsio->be_move_done = ctl_config_move_done;
5710                 ctl_datamove((union ctl_io *)ctsio);
5711
5712                 return (CTL_RETVAL_COMPLETE);
5713         }
5714
5715         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5716         lbalen->lba = lba;
5717         lbalen->len = num_blocks;
5718         lbalen->flags = byte2;
5719         retval = lun->backend->config_write((union ctl_io *)ctsio);
5720
5721         return (retval);
5722 }
5723
5724 int
5725 ctl_unmap(struct ctl_scsiio *ctsio)
5726 {
5727         struct ctl_lun *lun;
5728         struct scsi_unmap *cdb;
5729         struct ctl_ptr_len_flags *ptrlen;
5730         struct scsi_unmap_header *hdr;
5731         struct scsi_unmap_desc *buf, *end, *endnz, *range;
5732         uint64_t lba;
5733         uint32_t num_blocks;
5734         int len, retval;
5735         uint8_t byte2;
5736
5737         CTL_DEBUG_PRINT(("ctl_unmap\n"));
5738
5739         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5740         cdb = (struct scsi_unmap *)ctsio->cdb;
5741
5742         len = scsi_2btoul(cdb->length);
5743         byte2 = cdb->byte2;
5744
5745         /*
5746          * If we've got a kernel request that hasn't been malloced yet,
5747          * malloc it and tell the caller the data buffer is here.
5748          */
5749         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5750                 ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);
5751                 ctsio->kern_data_len = len;
5752                 ctsio->kern_total_len = len;
5753                 ctsio->kern_data_resid = 0;
5754                 ctsio->kern_rel_offset = 0;
5755                 ctsio->kern_sg_entries = 0;
5756                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5757                 ctsio->be_move_done = ctl_config_move_done;
5758                 ctl_datamove((union ctl_io *)ctsio);
5759
5760                 return (CTL_RETVAL_COMPLETE);
5761         }
5762
5763         len = ctsio->kern_total_len - ctsio->kern_data_resid;
5764         hdr = (struct scsi_unmap_header *)ctsio->kern_data_ptr;
5765         if (len < sizeof (*hdr) ||
5766             len < (scsi_2btoul(hdr->length) + sizeof(hdr->length)) ||
5767             len < (scsi_2btoul(hdr->desc_length) + sizeof (*hdr)) ||
5768             scsi_2btoul(hdr->desc_length) % sizeof(*buf) != 0) {
5769                 ctl_set_invalid_field(ctsio,
5770                                       /*sks_valid*/ 0,
5771                                       /*command*/ 0,
5772                                       /*field*/ 0,
5773                                       /*bit_valid*/ 0,
5774                                       /*bit*/ 0);
5775                 goto done;
5776         }
5777         len = scsi_2btoul(hdr->desc_length);
5778         buf = (struct scsi_unmap_desc *)(hdr + 1);
5779         end = buf + len / sizeof(*buf);
5780
5781         endnz = buf;
5782         for (range = buf; range < end; range++) {
5783                 lba = scsi_8btou64(range->lba);
5784                 num_blocks = scsi_4btoul(range->length);
5785                 if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5786                  || ((lba + num_blocks) < lba)) {
5787                         ctl_set_lba_out_of_range(ctsio,
5788                             MAX(lba, lun->be_lun->maxlba + 1));
5789                         ctl_done((union ctl_io *)ctsio);
5790                         return (CTL_RETVAL_COMPLETE);
5791                 }
5792                 if (num_blocks != 0)
5793                         endnz = range + 1;
5794         }
5795
5796         /*
5797          * Block backend can not handle zero last range.
5798          * Filter it out and return if there is nothing left.
5799          */
5800         len = (uint8_t *)endnz - (uint8_t *)buf;
5801         if (len == 0) {
5802                 ctl_set_success(ctsio);
5803                 goto done;
5804         }
5805
5806         mtx_lock(&lun->lun_lock);
5807         ptrlen = (struct ctl_ptr_len_flags *)
5808             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5809         ptrlen->ptr = (void *)buf;
5810         ptrlen->len = len;
5811         ptrlen->flags = byte2;
5812         ctl_check_blocked(lun);
5813         mtx_unlock(&lun->lun_lock);
5814
5815         retval = lun->backend->config_write((union ctl_io *)ctsio);
5816         return (retval);
5817
5818 done:
5819         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5820                 free(ctsio->kern_data_ptr, M_CTL);
5821                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5822         }
5823         ctl_done((union ctl_io *)ctsio);
5824         return (CTL_RETVAL_COMPLETE);
5825 }
5826
5827 int
5828 ctl_default_page_handler(struct ctl_scsiio *ctsio,
5829                          struct ctl_page_index *page_index, uint8_t *page_ptr)
5830 {
5831         struct ctl_lun *lun;
5832         uint8_t *current_cp, *saved_cp;
5833         int set_ua;
5834         uint32_t initidx;
5835
5836         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5837         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5838         set_ua = 0;
5839
5840         current_cp = (page_index->page_data + (page_index->page_len *
5841             CTL_PAGE_CURRENT));
5842         saved_cp = (page_index->page_data + (page_index->page_len *
5843             CTL_PAGE_SAVED));
5844
5845         mtx_lock(&lun->lun_lock);
5846         if (memcmp(current_cp, page_ptr, page_index->page_len)) {
5847                 memcpy(current_cp, page_ptr, page_index->page_len);
5848                 memcpy(saved_cp, page_ptr, page_index->page_len);
5849                 set_ua = 1;
5850         }
5851         if (set_ua != 0)
5852                 ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
5853         mtx_unlock(&lun->lun_lock);
5854         if (set_ua) {
5855                 ctl_isc_announce_mode(lun,
5856                     ctl_get_initindex(&ctsio->io_hdr.nexus),
5857                     page_index->page_code, page_index->subpage);
5858         }
5859         return (CTL_RETVAL_COMPLETE);
5860 }
5861
5862 static void
5863 ctl_ie_timer(void *arg)
5864 {
5865         struct ctl_lun *lun = arg;
5866         uint64_t t;
5867
5868         if (lun->ie_asc == 0)
5869                 return;
5870
5871         if (lun->MODE_IE.mrie == SIEP_MRIE_UA)
5872                 ctl_est_ua_all(lun, -1, CTL_UA_IE);
5873         else
5874                 lun->ie_reported = 0;
5875
5876         if (lun->ie_reportcnt < scsi_4btoul(lun->MODE_IE.report_count)) {
5877                 lun->ie_reportcnt++;
5878                 t = scsi_4btoul(lun->MODE_IE.interval_timer);
5879                 if (t == 0 || t == UINT32_MAX)
5880                         t = 3000;  /* 5 min */
5881                 callout_schedule(&lun->ie_callout, t * hz / 10);
5882         }
5883 }
5884
5885 int
5886 ctl_ie_page_handler(struct ctl_scsiio *ctsio,
5887                          struct ctl_page_index *page_index, uint8_t *page_ptr)
5888 {
5889         struct scsi_info_exceptions_page *pg;
5890         struct ctl_lun *lun;
5891         uint64_t t;
5892
5893         (void)ctl_default_page_handler(ctsio, page_index, page_ptr);
5894
5895         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5896         pg = (struct scsi_info_exceptions_page *)page_ptr;
5897         mtx_lock(&lun->lun_lock);
5898         if (pg->info_flags & SIEP_FLAGS_TEST) {
5899                 lun->ie_asc = 0x5d;
5900                 lun->ie_ascq = 0xff;
5901                 if (pg->mrie == SIEP_MRIE_UA) {
5902                         ctl_est_ua_all(lun, -1, CTL_UA_IE);
5903                         lun->ie_reported = 1;
5904                 } else {
5905                         ctl_clr_ua_all(lun, -1, CTL_UA_IE);
5906                         lun->ie_reported = -1;
5907                 }
5908                 lun->ie_reportcnt = 1;
5909                 if (lun->ie_reportcnt < scsi_4btoul(pg->report_count)) {
5910                         lun->ie_reportcnt++;
5911                         t = scsi_4btoul(pg->interval_timer);
5912                         if (t == 0 || t == UINT32_MAX)
5913                                 t = 3000;  /* 5 min */
5914                         callout_reset(&lun->ie_callout, t * hz / 10,
5915                             ctl_ie_timer, lun);
5916                 }
5917         } else {
5918                 lun->ie_asc = 0;
5919                 lun->ie_ascq = 0;
5920                 lun->ie_reported = 1;
5921                 ctl_clr_ua_all(lun, -1, CTL_UA_IE);
5922                 lun->ie_reportcnt = UINT32_MAX;
5923                 callout_stop(&lun->ie_callout);
5924         }
5925         mtx_unlock(&lun->lun_lock);
5926         return (CTL_RETVAL_COMPLETE);
5927 }
5928
5929 static int
5930 ctl_do_mode_select(union ctl_io *io)
5931 {
5932         struct scsi_mode_page_header *page_header;
5933         struct ctl_page_index *page_index;
5934         struct ctl_scsiio *ctsio;
5935         int page_len, page_len_offset, page_len_size;
5936         union ctl_modepage_info *modepage_info;
5937         struct ctl_lun *lun;
5938         int *len_left, *len_used;
5939         int retval, i;
5940
5941         ctsio = &io->scsiio;
5942         page_index = NULL;
5943         page_len = 0;
5944         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5945
5946         modepage_info = (union ctl_modepage_info *)
5947                 ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
5948         len_left = &modepage_info->header.len_left;
5949         len_used = &modepage_info->header.len_used;
5950
5951 do_next_page:
5952
5953         page_header = (struct scsi_mode_page_header *)
5954                 (ctsio->kern_data_ptr + *len_used);
5955
5956         if (*len_left == 0) {
5957                 free(ctsio->kern_data_ptr, M_CTL);
5958                 ctl_set_success(ctsio);
5959                 ctl_done((union ctl_io *)ctsio);
5960                 return (CTL_RETVAL_COMPLETE);
5961         } else if (*len_left < sizeof(struct scsi_mode_page_header)) {
5962
5963                 free(ctsio->kern_data_ptr, M_CTL);
5964                 ctl_set_param_len_error(ctsio);
5965                 ctl_done((union ctl_io *)ctsio);
5966                 return (CTL_RETVAL_COMPLETE);
5967
5968         } else if ((page_header->page_code & SMPH_SPF)
5969                 && (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
5970
5971                 free(ctsio->kern_data_ptr, M_CTL);
5972                 ctl_set_param_len_error(ctsio);
5973                 ctl_done((union ctl_io *)ctsio);
5974                 return (CTL_RETVAL_COMPLETE);
5975         }
5976
5977
5978         /*
5979          * XXX KDM should we do something with the block descriptor?
5980          */
5981         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
5982                 page_index = &lun->mode_pages.index[i];
5983                 if (lun->be_lun->lun_type == T_DIRECT &&
5984                     (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
5985                         continue;
5986                 if (lun->be_lun->lun_type == T_PROCESSOR &&
5987                     (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
5988                         continue;
5989                 if (lun->be_lun->lun_type == T_CDROM &&
5990                     (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
5991                         continue;
5992
5993                 if ((page_index->page_code & SMPH_PC_MASK) !=
5994                     (page_header->page_code & SMPH_PC_MASK))
5995                         continue;
5996
5997                 /*
5998                  * If neither page has a subpage code, then we've got a
5999                  * match.
6000                  */
6001                 if (((page_index->page_code & SMPH_SPF) == 0)
6002                  && ((page_header->page_code & SMPH_SPF) == 0)) {
6003                         page_len = page_header->page_length;
6004                         break;
6005                 }
6006
6007                 /*
6008                  * If both pages have subpages, then the subpage numbers
6009                  * have to match.
6010                  */
6011                 if ((page_index->page_code & SMPH_SPF)
6012                   && (page_header->page_code & SMPH_SPF)) {
6013                         struct scsi_mode_page_header_sp *sph;
6014
6015                         sph = (struct scsi_mode_page_header_sp *)page_header;
6016                         if (page_index->subpage == sph->subpage) {
6017                                 page_len = scsi_2btoul(sph->page_length);
6018                                 break;
6019                         }
6020                 }
6021         }
6022
6023         /*
6024          * If we couldn't find the page, or if we don't have a mode select
6025          * handler for it, send back an error to the user.
6026          */
6027         if ((i >= CTL_NUM_MODE_PAGES)
6028          || (page_index->select_handler == NULL)) {
6029                 ctl_set_invalid_field(ctsio,
6030                                       /*sks_valid*/ 1,
6031                                       /*command*/ 0,
6032                                       /*field*/ *len_used,
6033                                       /*bit_valid*/ 0,
6034                                       /*bit*/ 0);
6035                 free(ctsio->kern_data_ptr, M_CTL);
6036                 ctl_done((union ctl_io *)ctsio);
6037                 return (CTL_RETVAL_COMPLETE);
6038         }
6039
6040         if (page_index->page_code & SMPH_SPF) {
6041                 page_len_offset = 2;
6042                 page_len_size = 2;
6043         } else {
6044                 page_len_size = 1;
6045                 page_len_offset = 1;
6046         }
6047
6048         /*
6049          * If the length the initiator gives us isn't the one we specify in
6050          * the mode page header, or if they didn't specify enough data in
6051          * the CDB to avoid truncating this page, kick out the request.
6052          */
6053         if ((page_len != (page_index->page_len - page_len_offset -
6054                           page_len_size))
6055          || (*len_left < page_index->page_len)) {
6056
6057
6058                 ctl_set_invalid_field(ctsio,
6059                                       /*sks_valid*/ 1,
6060                                       /*command*/ 0,
6061                                       /*field*/ *len_used + page_len_offset,
6062                                       /*bit_valid*/ 0,
6063                                       /*bit*/ 0);
6064                 free(ctsio->kern_data_ptr, M_CTL);
6065                 ctl_done((union ctl_io *)ctsio);
6066                 return (CTL_RETVAL_COMPLETE);
6067         }
6068
6069         /*
6070          * Run through the mode page, checking to make sure that the bits
6071          * the user changed are actually legal for him to change.
6072          */
6073         for (i = 0; i < page_index->page_len; i++) {
6074                 uint8_t *user_byte, *change_mask, *current_byte;
6075                 int bad_bit;
6076                 int j;
6077
6078                 user_byte = (uint8_t *)page_header + i;
6079                 change_mask = page_index->page_data +
6080                               (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
6081                 current_byte = page_index->page_data +
6082                                (page_index->page_len * CTL_PAGE_CURRENT) + i;
6083
6084                 /*
6085                  * Check to see whether the user set any bits in this byte
6086                  * that he is not allowed to set.
6087                  */
6088                 if ((*user_byte & ~(*change_mask)) ==
6089                     (*current_byte & ~(*change_mask)))
6090                         continue;
6091
6092                 /*
6093                  * Go through bit by bit to determine which one is illegal.
6094                  */
6095                 bad_bit = 0;
6096                 for (j = 7; j >= 0; j--) {
6097                         if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
6098                             (((1 << i) & ~(*change_mask)) & *current_byte)) {
6099                                 bad_bit = i;
6100                                 break;
6101                         }
6102                 }
6103                 ctl_set_invalid_field(ctsio,
6104                                       /*sks_valid*/ 1,
6105                                       /*command*/ 0,
6106                                       /*field*/ *len_used + i,
6107                                       /*bit_valid*/ 1,
6108                                       /*bit*/ bad_bit);
6109                 free(ctsio->kern_data_ptr, M_CTL);
6110                 ctl_done((union ctl_io *)ctsio);
6111                 return (CTL_RETVAL_COMPLETE);
6112         }
6113
6114         /*
6115          * Decrement these before we call the page handler, since we may
6116          * end up getting called back one way or another before the handler
6117          * returns to this context.
6118          */
6119         *len_left -= page_index->page_len;
6120         *len_used += page_index->page_len;
6121
6122         retval = page_index->select_handler(ctsio, page_index,
6123                                             (uint8_t *)page_header);
6124
6125         /*
6126          * If the page handler returns CTL_RETVAL_QUEUED, then we need to
6127          * wait until this queued command completes to finish processing
6128          * the mode page.  If it returns anything other than
6129          * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
6130          * already set the sense information, freed the data pointer, and
6131          * completed the io for us.
6132          */
6133         if (retval != CTL_RETVAL_COMPLETE)
6134                 goto bailout_no_done;
6135
6136         /*
6137          * If the initiator sent us more than one page, parse the next one.
6138          */
6139         if (*len_left > 0)
6140                 goto do_next_page;
6141
6142         ctl_set_success(ctsio);
6143         free(ctsio->kern_data_ptr, M_CTL);
6144         ctl_done((union ctl_io *)ctsio);
6145
6146 bailout_no_done:
6147
6148         return (CTL_RETVAL_COMPLETE);
6149
6150 }
6151
6152 int
6153 ctl_mode_select(struct ctl_scsiio *ctsio)
6154 {
6155         int param_len, pf, sp;
6156         int header_size, bd_len;
6157         union ctl_modepage_info *modepage_info;
6158
6159         switch (ctsio->cdb[0]) {
6160         case MODE_SELECT_6: {
6161                 struct scsi_mode_select_6 *cdb;
6162
6163                 cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
6164
6165                 pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6166                 sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6167                 param_len = cdb->length;
6168                 header_size = sizeof(struct scsi_mode_header_6);
6169                 break;
6170         }
6171         case MODE_SELECT_10: {
6172                 struct scsi_mode_select_10 *cdb;
6173
6174                 cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
6175
6176                 pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6177                 sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6178                 param_len = scsi_2btoul(cdb->length);
6179                 header_size = sizeof(struct scsi_mode_header_10);
6180                 break;
6181         }
6182         default:
6183                 ctl_set_invalid_opcode(ctsio);
6184                 ctl_done((union ctl_io *)ctsio);
6185                 return (CTL_RETVAL_COMPLETE);
6186         }
6187
6188         /*
6189          * From SPC-3:
6190          * "A parameter list length of zero indicates that the Data-Out Buffer
6191          * shall be empty. This condition shall not be considered as an error."
6192          */
6193         if (param_len == 0) {
6194                 ctl_set_success(ctsio);
6195                 ctl_done((union ctl_io *)ctsio);
6196                 return (CTL_RETVAL_COMPLETE);
6197         }
6198
6199         /*
6200          * Since we'll hit this the first time through, prior to
6201          * allocation, we don't need to free a data buffer here.
6202          */
6203         if (param_len < header_size) {
6204                 ctl_set_param_len_error(ctsio);
6205                 ctl_done((union ctl_io *)ctsio);
6206                 return (CTL_RETVAL_COMPLETE);
6207         }
6208
6209         /*
6210          * Allocate the data buffer and grab the user's data.  In theory,
6211          * we shouldn't have to sanity check the parameter list length here
6212          * because the maximum size is 64K.  We should be able to malloc
6213          * that much without too many problems.
6214          */
6215         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6216                 ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6217                 ctsio->kern_data_len = param_len;
6218                 ctsio->kern_total_len = param_len;
6219                 ctsio->kern_data_resid = 0;
6220                 ctsio->kern_rel_offset = 0;
6221                 ctsio->kern_sg_entries = 0;
6222                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6223                 ctsio->be_move_done = ctl_config_move_done;
6224                 ctl_datamove((union ctl_io *)ctsio);
6225
6226                 return (CTL_RETVAL_COMPLETE);
6227         }
6228
6229         switch (ctsio->cdb[0]) {
6230         case MODE_SELECT_6: {
6231                 struct scsi_mode_header_6 *mh6;
6232
6233                 mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6234                 bd_len = mh6->blk_desc_len;
6235                 break;
6236         }
6237         case MODE_SELECT_10: {
6238                 struct scsi_mode_header_10 *mh10;
6239
6240                 mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6241                 bd_len = scsi_2btoul(mh10->blk_desc_len);
6242                 break;
6243         }
6244         default:
6245                 panic("%s: Invalid CDB type %#x", __func__, ctsio->cdb[0]);
6246         }
6247
6248         if (param_len < (header_size + bd_len)) {
6249                 free(ctsio->kern_data_ptr, M_CTL);
6250                 ctl_set_param_len_error(ctsio);
6251                 ctl_done((union ctl_io *)ctsio);
6252                 return (CTL_RETVAL_COMPLETE);
6253         }
6254
6255         /*
6256          * Set the IO_CONT flag, so that if this I/O gets passed to
6257          * ctl_config_write_done(), it'll get passed back to
6258          * ctl_do_mode_select() for further processing, or completion if
6259          * we're all done.
6260          */
6261         ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6262         ctsio->io_cont = ctl_do_mode_select;
6263
6264         modepage_info = (union ctl_modepage_info *)
6265                 ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6266         memset(modepage_info, 0, sizeof(*modepage_info));
6267         modepage_info->header.len_left = param_len - header_size - bd_len;
6268         modepage_info->header.len_used = header_size + bd_len;
6269
6270         return (ctl_do_mode_select((union ctl_io *)ctsio));
6271 }
6272
6273 int
6274 ctl_mode_sense(struct ctl_scsiio *ctsio)
6275 {
6276         struct ctl_lun *lun;
6277         int pc, page_code, dbd, llba, subpage;
6278         int alloc_len, page_len, header_len, total_len;
6279         struct scsi_mode_block_descr *block_desc;
6280         struct ctl_page_index *page_index;
6281
6282         dbd = 0;
6283         llba = 0;
6284         block_desc = NULL;
6285
6286         CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6287
6288         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6289         switch (ctsio->cdb[0]) {
6290         case MODE_SENSE_6: {
6291                 struct scsi_mode_sense_6 *cdb;
6292
6293                 cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6294
6295                 header_len = sizeof(struct scsi_mode_hdr_6);
6296                 if (cdb->byte2 & SMS_DBD)
6297                         dbd = 1;
6298                 else
6299                         header_len += sizeof(struct scsi_mode_block_descr);
6300
6301                 pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6302                 page_code = cdb->page & SMS_PAGE_CODE;
6303                 subpage = cdb->subpage;
6304                 alloc_len = cdb->length;
6305                 break;
6306         }
6307         case MODE_SENSE_10: {
6308                 struct scsi_mode_sense_10 *cdb;
6309
6310                 cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6311
6312                 header_len = sizeof(struct scsi_mode_hdr_10);
6313
6314                 if (cdb->byte2 & SMS_DBD)
6315                         dbd = 1;
6316                 else
6317                         header_len += sizeof(struct scsi_mode_block_descr);
6318                 if (cdb->byte2 & SMS10_LLBAA)
6319                         llba = 1;
6320                 pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6321                 page_code = cdb->page & SMS_PAGE_CODE;
6322                 subpage = cdb->subpage;
6323                 alloc_len = scsi_2btoul(cdb->length);
6324                 break;
6325         }
6326         default:
6327                 ctl_set_invalid_opcode(ctsio);
6328                 ctl_done((union ctl_io *)ctsio);
6329                 return (CTL_RETVAL_COMPLETE);
6330                 break; /* NOTREACHED */
6331         }
6332
6333         /*
6334          * We have to make a first pass through to calculate the size of
6335          * the pages that match the user's query.  Then we allocate enough
6336          * memory to hold it, and actually copy the data into the buffer.
6337          */
6338         switch (page_code) {
6339         case SMS_ALL_PAGES_PAGE: {
6340                 u_int i;
6341
6342                 page_len = 0;
6343
6344                 /*
6345                  * At the moment, values other than 0 and 0xff here are
6346                  * reserved according to SPC-3.
6347                  */
6348                 if ((subpage != SMS_SUBPAGE_PAGE_0)
6349                  && (subpage != SMS_SUBPAGE_ALL)) {
6350                         ctl_set_invalid_field(ctsio,
6351                                               /*sks_valid*/ 1,
6352                                               /*command*/ 1,
6353                                               /*field*/ 3,
6354                                               /*bit_valid*/ 0,
6355                                               /*bit*/ 0);
6356                         ctl_done((union ctl_io *)ctsio);
6357                         return (CTL_RETVAL_COMPLETE);
6358                 }
6359
6360                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6361                         page_index = &lun->mode_pages.index[i];
6362
6363                         /* Make sure the page is supported for this dev type */
6364                         if (lun->be_lun->lun_type == T_DIRECT &&
6365                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6366                                 continue;
6367                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6368                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6369                                 continue;
6370                         if (lun->be_lun->lun_type == T_CDROM &&
6371                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6372                                 continue;
6373
6374                         /*
6375                          * We don't use this subpage if the user didn't
6376                          * request all subpages.
6377                          */
6378                         if ((page_index->subpage != 0)
6379                          && (subpage == SMS_SUBPAGE_PAGE_0))
6380                                 continue;
6381
6382 #if 0
6383                         printf("found page %#x len %d\n",
6384                                page_index->page_code & SMPH_PC_MASK,
6385                                page_index->page_len);
6386 #endif
6387                         page_len += page_index->page_len;
6388                 }
6389                 break;
6390         }
6391         default: {
6392                 u_int i;
6393
6394                 page_len = 0;
6395
6396                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6397                         page_index = &lun->mode_pages.index[i];
6398
6399                         /* Make sure the page is supported for this dev type */
6400                         if (lun->be_lun->lun_type == T_DIRECT &&
6401                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6402                                 continue;
6403                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6404                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6405                                 continue;
6406                         if (lun->be_lun->lun_type == T_CDROM &&
6407                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6408                                 continue;
6409
6410                         /* Look for the right page code */
6411                         if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6412                                 continue;
6413
6414                         /* Look for the right subpage or the subpage wildcard*/
6415                         if ((page_index->subpage != subpage)
6416                          && (subpage != SMS_SUBPAGE_ALL))
6417                                 continue;
6418
6419 #if 0
6420                         printf("found page %#x len %d\n",
6421                                page_index->page_code & SMPH_PC_MASK,
6422                                page_index->page_len);
6423 #endif
6424
6425                         page_len += page_index->page_len;
6426                 }
6427
6428                 if (page_len == 0) {
6429                         ctl_set_invalid_field(ctsio,
6430                                               /*sks_valid*/ 1,
6431                                               /*command*/ 1,
6432                                               /*field*/ 2,
6433                                               /*bit_valid*/ 1,
6434                                               /*bit*/ 5);
6435                         ctl_done((union ctl_io *)ctsio);
6436                         return (CTL_RETVAL_COMPLETE);
6437                 }
6438                 break;
6439         }
6440         }
6441
6442         total_len = header_len + page_len;
6443 #if 0
6444         printf("header_len = %d, page_len = %d, total_len = %d\n",
6445                header_len, page_len, total_len);
6446 #endif
6447
6448         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6449         ctsio->kern_sg_entries = 0;
6450         ctsio->kern_data_resid = 0;
6451         ctsio->kern_rel_offset = 0;
6452         if (total_len < alloc_len) {
6453                 ctsio->residual = alloc_len - total_len;
6454                 ctsio->kern_data_len = total_len;
6455                 ctsio->kern_total_len = total_len;
6456         } else {
6457                 ctsio->residual = 0;
6458                 ctsio->kern_data_len = alloc_len;
6459                 ctsio->kern_total_len = alloc_len;
6460         }
6461
6462         switch (ctsio->cdb[0]) {
6463         case MODE_SENSE_6: {
6464                 struct scsi_mode_hdr_6 *header;
6465
6466                 header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
6467
6468                 header->datalen = MIN(total_len - 1, 254);
6469                 if (lun->be_lun->lun_type == T_DIRECT) {
6470                         header->dev_specific = 0x10; /* DPOFUA */
6471                         if ((lun->be_lun->flags & CTL_LUN_FLAG_READONLY) ||
6472                             (lun->MODE_CTRL.eca_and_aen & SCP_SWP) != 0)
6473                                 header->dev_specific |= 0x80; /* WP */
6474                 }
6475                 if (dbd)
6476                         header->block_descr_len = 0;
6477                 else
6478                         header->block_descr_len =
6479                                 sizeof(struct scsi_mode_block_descr);
6480                 block_desc = (struct scsi_mode_block_descr *)&header[1];
6481                 break;
6482         }
6483         case MODE_SENSE_10: {
6484                 struct scsi_mode_hdr_10 *header;
6485                 int datalen;
6486
6487                 header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
6488
6489                 datalen = MIN(total_len - 2, 65533);
6490                 scsi_ulto2b(datalen, header->datalen);
6491                 if (lun->be_lun->lun_type == T_DIRECT) {
6492                         header->dev_specific = 0x10; /* DPOFUA */
6493                         if ((lun->be_lun->flags & CTL_LUN_FLAG_READONLY) ||
6494                             (lun->MODE_CTRL.eca_and_aen & SCP_SWP) != 0)
6495                                 header->dev_specific |= 0x80; /* WP */
6496                 }
6497                 if (dbd)
6498                         scsi_ulto2b(0, header->block_descr_len);
6499                 else
6500                         scsi_ulto2b(sizeof(struct scsi_mode_block_descr),
6501                                     header->block_descr_len);
6502                 block_desc = (struct scsi_mode_block_descr *)&header[1];
6503                 break;
6504         }
6505         default:
6506                 panic("%s: Invalid CDB type %#x", __func__, ctsio->cdb[0]);
6507         }
6508
6509         /*
6510          * If we've got a disk, use its blocksize in the block
6511          * descriptor.  Otherwise, just set it to 0.
6512          */
6513         if (dbd == 0) {
6514                 if (lun->be_lun->lun_type == T_DIRECT)
6515                         scsi_ulto3b(lun->be_lun->blocksize,
6516                                     block_desc->block_len);
6517                 else
6518                         scsi_ulto3b(0, block_desc->block_len);
6519         }
6520
6521         switch (page_code) {
6522         case SMS_ALL_PAGES_PAGE: {
6523                 int i, data_used;
6524
6525                 data_used = header_len;
6526                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6527                         struct ctl_page_index *page_index;
6528
6529                         page_index = &lun->mode_pages.index[i];
6530                         if (lun->be_lun->lun_type == T_DIRECT &&
6531                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6532                                 continue;
6533                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6534                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6535                                 continue;
6536                         if (lun->be_lun->lun_type == T_CDROM &&
6537                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6538                                 continue;
6539
6540                         /*
6541                          * We don't use this subpage if the user didn't
6542                          * request all subpages.  We already checked (above)
6543                          * to make sure the user only specified a subpage
6544                          * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case.
6545                          */
6546                         if ((page_index->subpage != 0)
6547                          && (subpage == SMS_SUBPAGE_PAGE_0))
6548                                 continue;
6549
6550                         /*
6551                          * Call the handler, if it exists, to update the
6552                          * page to the latest values.
6553                          */
6554                         if (page_index->sense_handler != NULL)
6555                                 page_index->sense_handler(ctsio, page_index,pc);
6556
6557                         memcpy(ctsio->kern_data_ptr + data_used,
6558                                page_index->page_data +
6559                                (page_index->page_len * pc),
6560                                page_index->page_len);
6561                         data_used += page_index->page_len;
6562                 }
6563                 break;
6564         }
6565         default: {
6566                 int i, data_used;
6567
6568                 data_used = header_len;
6569
6570                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6571                         struct ctl_page_index *page_index;
6572
6573                         page_index = &lun->mode_pages.index[i];
6574
6575                         /* Look for the right page code */
6576                         if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6577                                 continue;
6578
6579                         /* Look for the right subpage or the subpage wildcard*/
6580                         if ((page_index->subpage != subpage)
6581                          && (subpage != SMS_SUBPAGE_ALL))
6582                                 continue;
6583
6584                         /* Make sure the page is supported for this dev type */
6585                         if (lun->be_lun->lun_type == T_DIRECT &&
6586                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6587                                 continue;
6588                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6589                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6590                                 continue;
6591                         if (lun->be_lun->lun_type == T_CDROM &&
6592                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6593                                 continue;
6594
6595                         /*
6596                          * Call the handler, if it exists, to update the
6597                          * page to the latest values.
6598                          */
6599                         if (page_index->sense_handler != NULL)
6600                                 page_index->sense_handler(ctsio, page_index,pc);
6601
6602                         memcpy(ctsio->kern_data_ptr + data_used,
6603                                page_index->page_data +
6604                                (page_index->page_len * pc),
6605                                page_index->page_len);
6606                         data_used += page_index->page_len;
6607                 }
6608                 break;
6609         }
6610         }
6611
6612         ctl_set_success(ctsio);
6613         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6614         ctsio->be_move_done = ctl_config_move_done;
6615         ctl_datamove((union ctl_io *)ctsio);
6616         return (CTL_RETVAL_COMPLETE);
6617 }
6618
6619 int
6620 ctl_lbp_log_sense_handler(struct ctl_scsiio *ctsio,
6621                                struct ctl_page_index *page_index,
6622                                int pc)
6623 {
6624         struct ctl_lun *lun;
6625         struct scsi_log_param_header *phdr;
6626         uint8_t *data;
6627         uint64_t val;
6628
6629         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6630         data = page_index->page_data;
6631
6632         if (lun->backend->lun_attr != NULL &&
6633             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "blocksavail"))
6634              != UINT64_MAX) {
6635                 phdr = (struct scsi_log_param_header *)data;
6636                 scsi_ulto2b(0x0001, phdr->param_code);
6637                 phdr->param_control = SLP_LBIN | SLP_LP;
6638                 phdr->param_len = 8;
6639                 data = (uint8_t *)(phdr + 1);
6640                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6641                 data[4] = 0x02; /* per-pool */
6642                 data += phdr->param_len;
6643         }
6644
6645         if (lun->backend->lun_attr != NULL &&
6646             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "blocksused"))
6647              != UINT64_MAX) {
6648                 phdr = (struct scsi_log_param_header *)data;
6649                 scsi_ulto2b(0x0002, phdr->param_code);
6650                 phdr->param_control = SLP_LBIN | SLP_LP;
6651                 phdr->param_len = 8;
6652                 data = (uint8_t *)(phdr + 1);
6653                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6654                 data[4] = 0x01; /* per-LUN */
6655                 data += phdr->param_len;
6656         }
6657
6658         if (lun->backend->lun_attr != NULL &&
6659             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "poolblocksavail"))
6660              != UINT64_MAX) {
6661                 phdr = (struct scsi_log_param_header *)data;
6662                 scsi_ulto2b(0x00f1, phdr->param_code);
6663                 phdr->param_control = SLP_LBIN | SLP_LP;
6664                 phdr->param_len = 8;
6665                 data = (uint8_t *)(phdr + 1);
6666                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6667                 data[4] = 0x02; /* per-pool */
6668                 data += phdr->param_len;
6669         }
6670
6671         if (lun->backend->lun_attr != NULL &&
6672             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "poolblocksused"))
6673              != UINT64_MAX) {
6674                 phdr = (struct scsi_log_param_header *)data;
6675                 scsi_ulto2b(0x00f2, phdr->param_code);
6676                 phdr->param_control = SLP_LBIN | SLP_LP;
6677                 phdr->param_len = 8;
6678                 data = (uint8_t *)(phdr + 1);
6679                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6680                 data[4] = 0x02; /* per-pool */
6681                 data += phdr->param_len;
6682         }
6683
6684         page_index->page_len = data - page_index->page_data;
6685         return (0);
6686 }
6687
6688 int
6689 ctl_sap_log_sense_handler(struct ctl_scsiio *ctsio,
6690                                struct ctl_page_index *page_index,
6691                                int pc)
6692 {
6693         struct ctl_lun *lun;
6694         struct stat_page *data;
6695         uint64_t rn, wn, rb, wb;
6696         struct bintime rt, wt;
6697         int i;
6698
6699         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6700         data = (struct stat_page *)page_index->page_data;
6701
6702         scsi_ulto2b(SLP_SAP, data->sap.hdr.param_code);
6703         data->sap.hdr.param_control = SLP_LBIN;
6704         data->sap.hdr.param_len = sizeof(struct scsi_log_stat_and_perf) -
6705             sizeof(struct scsi_log_param_header);
6706         rn = wn = rb = wb = 0;
6707         bintime_clear(&rt);
6708         bintime_clear(&wt);
6709         for (i = 0; i < CTL_MAX_PORTS; i++) {
6710                 rn += lun->stats.ports[i].operations[CTL_STATS_READ];
6711                 wn += lun->stats.ports[i].operations[CTL_STATS_WRITE];
6712                 rb += lun->stats.ports[i].bytes[CTL_STATS_READ];
6713                 wb += lun->stats.ports[i].bytes[CTL_STATS_WRITE];
6714                 bintime_add(&rt, &lun->stats.ports[i].time[CTL_STATS_READ]);
6715                 bintime_add(&wt, &lun->stats.ports[i].time[CTL_STATS_WRITE]);
6716         }
6717         scsi_u64to8b(rn, data->sap.read_num);
6718         scsi_u64to8b(wn, data->sap.write_num);
6719         if (lun->stats.blocksize > 0) {
6720                 scsi_u64to8b(wb / lun->stats.blocksize,
6721                     data->sap.recvieved_lba);
6722                 scsi_u64to8b(rb / lun->stats.blocksize,
6723                     data->sap.transmitted_lba);
6724         }
6725         scsi_u64to8b((uint64_t)rt.sec * 1000 + rt.frac / (UINT64_MAX / 1000),
6726             data->sap.read_int);
6727         scsi_u64to8b((uint64_t)wt.sec * 1000 + wt.frac / (UINT64_MAX / 1000),
6728             data->sap.write_int);
6729         scsi_u64to8b(0, data->sap.weighted_num);
6730         scsi_u64to8b(0, data->sap.weighted_int);
6731         scsi_ulto2b(SLP_IT, data->it.hdr.param_code);
6732         data->it.hdr.param_control = SLP_LBIN;
6733         data->it.hdr.param_len = sizeof(struct scsi_log_idle_time) -
6734             sizeof(struct scsi_log_param_header);
6735 #ifdef CTL_TIME_IO
6736         scsi_u64to8b(lun->idle_time / SBT_1MS, data->it.idle_int);
6737 #endif
6738         scsi_ulto2b(SLP_TI, data->ti.hdr.param_code);
6739         data->it.hdr.param_control = SLP_LBIN;
6740         data->ti.hdr.param_len = sizeof(struct scsi_log_time_interval) -
6741             sizeof(struct scsi_log_param_header);
6742         scsi_ulto4b(3, data->ti.exponent);
6743         scsi_ulto4b(1, data->ti.integer);
6744         return (0);
6745 }
6746
6747 int
6748 ctl_ie_log_sense_handler(struct ctl_scsiio *ctsio,
6749                                struct ctl_page_index *page_index,
6750                                int pc)
6751 {
6752         struct ctl_lun *lun;
6753         struct scsi_log_informational_exceptions *data;
6754
6755         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6756         data = (struct scsi_log_informational_exceptions *)page_index->page_data;
6757
6758         scsi_ulto2b(SLP_IE_GEN, data->hdr.param_code);
6759         data->hdr.param_control = SLP_LBIN;
6760         data->hdr.param_len = sizeof(struct scsi_log_informational_exceptions) -
6761             sizeof(struct scsi_log_param_header);
6762         data->ie_asc = lun->ie_asc;
6763         data->ie_ascq = lun->ie_ascq;
6764         data->temperature = 0xff;
6765         return (0);
6766 }
6767
6768 int
6769 ctl_log_sense(struct ctl_scsiio *ctsio)
6770 {
6771         struct ctl_lun *lun;
6772         int i, pc, page_code, subpage;
6773         int alloc_len, total_len;
6774         struct ctl_page_index *page_index;
6775         struct scsi_log_sense *cdb;
6776         struct scsi_log_header *header;
6777
6778         CTL_DEBUG_PRINT(("ctl_log_sense\n"));
6779
6780         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6781         cdb = (struct scsi_log_sense *)ctsio->cdb;
6782         pc = (cdb->page & SLS_PAGE_CTRL_MASK) >> 6;
6783         page_code = cdb->page & SLS_PAGE_CODE;
6784         subpage = cdb->subpage;
6785         alloc_len = scsi_2btoul(cdb->length);
6786
6787         page_index = NULL;
6788         for (i = 0; i < CTL_NUM_LOG_PAGES; i++) {
6789                 page_index = &lun->log_pages.index[i];
6790
6791                 /* Look for the right page code */
6792                 if ((page_index->page_code & SL_PAGE_CODE) != page_code)
6793                         continue;
6794
6795                 /* Look for the right subpage or the subpage wildcard*/
6796                 if (page_index->subpage != subpage)
6797                         continue;
6798
6799                 break;
6800         }
6801         if (i >= CTL_NUM_LOG_PAGES) {
6802                 ctl_set_invalid_field(ctsio,
6803                                       /*sks_valid*/ 1,
6804                                       /*command*/ 1,
6805                                       /*field*/ 2,
6806                                       /*bit_valid*/ 0,
6807                                       /*bit*/ 0);
6808                 ctl_done((union ctl_io *)ctsio);
6809                 return (CTL_RETVAL_COMPLETE);
6810         }
6811
6812         total_len = sizeof(struct scsi_log_header) + page_index->page_len;
6813
6814         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6815         ctsio->kern_sg_entries = 0;
6816         ctsio->kern_data_resid = 0;
6817         ctsio->kern_rel_offset = 0;
6818         if (total_len < alloc_len) {
6819                 ctsio->residual = alloc_len - total_len;
6820                 ctsio->kern_data_len = total_len;
6821                 ctsio->kern_total_len = total_len;
6822         } else {
6823                 ctsio->residual = 0;
6824                 ctsio->kern_data_len = alloc_len;
6825                 ctsio->kern_total_len = alloc_len;
6826         }
6827
6828         header = (struct scsi_log_header *)ctsio->kern_data_ptr;
6829         header->page = page_index->page_code;
6830         if (page_index->page_code == SLS_LOGICAL_BLOCK_PROVISIONING)
6831                 header->page |= SL_DS;
6832         if (page_index->subpage) {
6833                 header->page |= SL_SPF;
6834                 header->subpage = page_index->subpage;
6835         }
6836         scsi_ulto2b(page_index->page_len, header->datalen);
6837
6838         /*
6839          * Call the handler, if it exists, to update the
6840          * page to the latest values.
6841          */
6842         if (page_index->sense_handler != NULL)
6843                 page_index->sense_handler(ctsio, page_index, pc);
6844
6845         memcpy(header + 1, page_index->page_data, page_index->page_len);
6846
6847         ctl_set_success(ctsio);
6848         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6849         ctsio->be_move_done = ctl_config_move_done;
6850         ctl_datamove((union ctl_io *)ctsio);
6851         return (CTL_RETVAL_COMPLETE);
6852 }
6853
6854 int
6855 ctl_read_capacity(struct ctl_scsiio *ctsio)
6856 {
6857         struct scsi_read_capacity *cdb;
6858         struct scsi_read_capacity_data *data;
6859         struct ctl_lun *lun;
6860         uint32_t lba;
6861
6862         CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
6863
6864         cdb = (struct scsi_read_capacity *)ctsio->cdb;
6865
6866         lba = scsi_4btoul(cdb->addr);
6867         if (((cdb->pmi & SRC_PMI) == 0)
6868          && (lba != 0)) {
6869                 ctl_set_invalid_field(/*ctsio*/ ctsio,
6870                                       /*sks_valid*/ 1,
6871                                       /*command*/ 1,
6872                                       /*field*/ 2,
6873                                       /*bit_valid*/ 0,
6874                                       /*bit*/ 0);
6875                 ctl_done((union ctl_io *)ctsio);
6876                 return (CTL_RETVAL_COMPLETE);
6877         }
6878
6879         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6880
6881         ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
6882         data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
6883         ctsio->residual = 0;
6884         ctsio->kern_data_len = sizeof(*data);
6885         ctsio->kern_total_len = sizeof(*data);
6886         ctsio->kern_data_resid = 0;
6887         ctsio->kern_rel_offset = 0;
6888         ctsio->kern_sg_entries = 0;
6889
6890         /*
6891          * If the maximum LBA is greater than 0xfffffffe, the user must
6892          * issue a SERVICE ACTION IN (16) command, with the read capacity
6893          * serivce action set.
6894          */
6895         if (lun->be_lun->maxlba > 0xfffffffe)
6896                 scsi_ulto4b(0xffffffff, data->addr);
6897         else
6898                 scsi_ulto4b(lun->be_lun->maxlba, data->addr);
6899
6900         /*
6901          * XXX KDM this may not be 512 bytes...
6902          */
6903         scsi_ulto4b(lun->be_lun->blocksize, data->length);
6904
6905         ctl_set_success(ctsio);
6906         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6907         ctsio->be_move_done = ctl_config_move_done;
6908         ctl_datamove((union ctl_io *)ctsio);
6909         return (CTL_RETVAL_COMPLETE);
6910 }
6911
6912 int
6913 ctl_read_capacity_16(struct ctl_scsiio *ctsio)
6914 {
6915         struct scsi_read_capacity_16 *cdb;
6916         struct scsi_read_capacity_data_long *data;
6917         struct ctl_lun *lun;
6918         uint64_t lba;
6919         uint32_t alloc_len;
6920
6921         CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
6922
6923         cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
6924
6925         alloc_len = scsi_4btoul(cdb->alloc_len);
6926         lba = scsi_8btou64(cdb->addr);
6927
6928         if ((cdb->reladr & SRC16_PMI)
6929          && (lba != 0)) {
6930                 ctl_set_invalid_field(/*ctsio*/ ctsio,
6931                                       /*sks_valid*/ 1,
6932                                       /*command*/ 1,
6933                                       /*field*/ 2,
6934                                       /*bit_valid*/ 0,
6935                                       /*bit*/ 0);
6936                 ctl_done((union ctl_io *)ctsio);
6937                 return (CTL_RETVAL_COMPLETE);
6938         }
6939
6940         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6941
6942         ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
6943         data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
6944
6945         if (sizeof(*data) < alloc_len) {
6946                 ctsio->residual = alloc_len - sizeof(*data);
6947                 ctsio->kern_data_len = sizeof(*data);
6948                 ctsio->kern_total_len = sizeof(*data);
6949         } else {
6950                 ctsio->residual = 0;
6951                 ctsio->kern_data_len = alloc_len;
6952                 ctsio->kern_total_len = alloc_len;
6953         }
6954         ctsio->kern_data_resid = 0;
6955         ctsio->kern_rel_offset = 0;
6956         ctsio->kern_sg_entries = 0;
6957
6958         scsi_u64to8b(lun->be_lun->maxlba, data->addr);
6959         /* XXX KDM this may not be 512 bytes... */
6960         scsi_ulto4b(lun->be_lun->blocksize, data->length);
6961         data->prot_lbppbe = lun->be_lun->pblockexp & SRC16_LBPPBE;
6962         scsi_ulto2b(lun->be_lun->pblockoff & SRC16_LALBA_A, data->lalba_lbp);
6963         if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
6964                 data->lalba_lbp[0] |= SRC16_LBPME | SRC16_LBPRZ;
6965
6966         ctl_set_success(ctsio);
6967         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6968         ctsio->be_move_done = ctl_config_move_done;
6969         ctl_datamove((union ctl_io *)ctsio);
6970         return (CTL_RETVAL_COMPLETE);
6971 }
6972
6973 int
6974 ctl_get_lba_status(struct ctl_scsiio *ctsio)
6975 {
6976         struct scsi_get_lba_status *cdb;
6977         struct scsi_get_lba_status_data *data;
6978         struct ctl_lun *lun;
6979         struct ctl_lba_len_flags *lbalen;
6980         uint64_t lba;
6981         uint32_t alloc_len, total_len;
6982         int retval;
6983
6984         CTL_DEBUG_PRINT(("ctl_get_lba_status\n"));
6985
6986         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6987         cdb = (struct scsi_get_lba_status *)ctsio->cdb;
6988         lba = scsi_8btou64(cdb->addr);
6989         alloc_len = scsi_4btoul(cdb->alloc_len);
6990
6991         if (lba > lun->be_lun->maxlba) {
6992                 ctl_set_lba_out_of_range(ctsio, lba);
6993                 ctl_done((union ctl_io *)ctsio);
6994                 return (CTL_RETVAL_COMPLETE);
6995         }
6996
6997         total_len = sizeof(*data) + sizeof(data->descr[0]);
6998         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6999         data = (struct scsi_get_lba_status_data *)ctsio->kern_data_ptr;
7000
7001         if (total_len < alloc_len) {
7002                 ctsio->residual = alloc_len - total_len;
7003                 ctsio->kern_data_len = total_len;
7004                 ctsio->kern_total_len = total_len;
7005         } else {
7006                 ctsio->residual = 0;
7007                 ctsio->kern_data_len = alloc_len;
7008                 ctsio->kern_total_len = alloc_len;
7009         }
7010         ctsio->kern_data_resid = 0;
7011         ctsio->kern_rel_offset = 0;
7012         ctsio->kern_sg_entries = 0;
7013
7014         /* Fill dummy data in case backend can't tell anything. */
7015         scsi_ulto4b(4 + sizeof(data->descr[0]), data->length);
7016         scsi_u64to8b(lba, data->descr[0].addr);
7017         scsi_ulto4b(MIN(UINT32_MAX, lun->be_lun->maxlba + 1 - lba),
7018             data->descr[0].length);
7019         data->descr[0].status = 0; /* Mapped or unknown. */
7020
7021         ctl_set_success(ctsio);
7022         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7023         ctsio->be_move_done = ctl_config_move_done;
7024
7025         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
7026         lbalen->lba = lba;
7027         lbalen->len = total_len;
7028         lbalen->flags = 0;
7029         retval = lun->backend->config_read((union ctl_io *)ctsio);
7030         return (CTL_RETVAL_COMPLETE);
7031 }
7032
7033 int
7034 ctl_read_defect(struct ctl_scsiio *ctsio)
7035 {
7036         struct scsi_read_defect_data_10 *ccb10;
7037         struct scsi_read_defect_data_12 *ccb12;
7038         struct scsi_read_defect_data_hdr_10 *data10;
7039         struct scsi_read_defect_data_hdr_12 *data12;
7040         uint32_t alloc_len, data_len;
7041         uint8_t format;
7042
7043         CTL_DEBUG_PRINT(("ctl_read_defect\n"));
7044
7045         if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
7046                 ccb10 = (struct scsi_read_defect_data_10 *)&ctsio->cdb;
7047                 format = ccb10->format;
7048                 alloc_len = scsi_2btoul(ccb10->alloc_length);
7049                 data_len = sizeof(*data10);
7050         } else {
7051                 ccb12 = (struct scsi_read_defect_data_12 *)&ctsio->cdb;
7052                 format = ccb12->format;
7053                 alloc_len = scsi_4btoul(ccb12->alloc_length);
7054                 data_len = sizeof(*data12);
7055         }
7056         if (alloc_len == 0) {
7057                 ctl_set_success(ctsio);
7058                 ctl_done((union ctl_io *)ctsio);
7059                 return (CTL_RETVAL_COMPLETE);
7060         }
7061
7062         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
7063         if (data_len < alloc_len) {
7064                 ctsio->residual = alloc_len - data_len;
7065                 ctsio->kern_data_len = data_len;
7066                 ctsio->kern_total_len = data_len;
7067         } else {
7068                 ctsio->residual = 0;
7069                 ctsio->kern_data_len = alloc_len;
7070                 ctsio->kern_total_len = alloc_len;
7071         }
7072         ctsio->kern_data_resid = 0;
7073         ctsio->kern_rel_offset = 0;
7074         ctsio->kern_sg_entries = 0;
7075
7076         if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
7077                 data10 = (struct scsi_read_defect_data_hdr_10 *)
7078                     ctsio->kern_data_ptr;
7079                 data10->format = format;
7080                 scsi_ulto2b(0, data10->length);
7081         } else {
7082                 data12 = (struct scsi_read_defect_data_hdr_12 *)
7083                     ctsio->kern_data_ptr;
7084                 data12->format = format;
7085                 scsi_ulto2b(0, data12->generation);
7086                 scsi_ulto4b(0, data12->length);
7087         }
7088
7089         ctl_set_success(ctsio);
7090         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7091         ctsio->be_move_done = ctl_config_move_done;
7092         ctl_datamove((union ctl_io *)ctsio);
7093         return (CTL_RETVAL_COMPLETE);
7094 }
7095
7096 int
7097 ctl_report_tagret_port_groups(struct ctl_scsiio *ctsio)
7098 {
7099         struct scsi_maintenance_in *cdb;
7100         int retval;
7101         int alloc_len, ext, total_len = 0, g, pc, pg, ts, os;
7102         int num_ha_groups, num_target_ports, shared_group;
7103         struct ctl_lun *lun;
7104         struct ctl_softc *softc;
7105         struct ctl_port *port;
7106         struct scsi_target_group_data *rtg_ptr;
7107         struct scsi_target_group_data_extended *rtg_ext_ptr;
7108         struct scsi_target_port_group_descriptor *tpg_desc;
7109
7110         CTL_DEBUG_PRINT(("ctl_report_tagret_port_groups\n"));
7111
7112         cdb = (struct scsi_maintenance_in *)ctsio->cdb;
7113         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7114         softc = lun->ctl_softc;
7115
7116         retval = CTL_RETVAL_COMPLETE;
7117
7118         switch (cdb->byte2 & STG_PDF_MASK) {
7119         case STG_PDF_LENGTH:
7120                 ext = 0;
7121                 break;
7122         case STG_PDF_EXTENDED:
7123                 ext = 1;
7124                 break;
7125         default:
7126                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7127                                       /*sks_valid*/ 1,
7128                                       /*command*/ 1,
7129                                       /*field*/ 2,
7130                                       /*bit_valid*/ 1,
7131                                       /*bit*/ 5);
7132                 ctl_done((union ctl_io *)ctsio);
7133                 return(retval);
7134         }
7135
7136         num_target_ports = 0;
7137         shared_group = (softc->is_single != 0);
7138         mtx_lock(&softc->ctl_lock);
7139         STAILQ_FOREACH(port, &softc->port_list, links) {
7140                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7141                         continue;
7142                 if (ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
7143                         continue;
7144                 num_target_ports++;
7145                 if (port->status & CTL_PORT_STATUS_HA_SHARED)
7146                         shared_group = 1;
7147         }
7148         mtx_unlock(&softc->ctl_lock);
7149         num_ha_groups = (softc->is_single) ? 0 : NUM_HA_SHELVES;
7150
7151         if (ext)
7152                 total_len = sizeof(struct scsi_target_group_data_extended);
7153         else
7154                 total_len = sizeof(struct scsi_target_group_data);
7155         total_len += sizeof(struct scsi_target_port_group_descriptor) *
7156                 (shared_group + num_ha_groups) +
7157             sizeof(struct scsi_target_port_descriptor) * num_target_ports;
7158
7159         alloc_len = scsi_4btoul(cdb->length);
7160
7161         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7162
7163         ctsio->kern_sg_entries = 0;
7164
7165         if (total_len < alloc_len) {
7166                 ctsio->residual = alloc_len - total_len;
7167                 ctsio->kern_data_len = total_len;
7168                 ctsio->kern_total_len = total_len;
7169         } else {
7170                 ctsio->residual = 0;
7171                 ctsio->kern_data_len = alloc_len;
7172                 ctsio->kern_total_len = alloc_len;
7173         }
7174         ctsio->kern_data_resid = 0;
7175         ctsio->kern_rel_offset = 0;
7176
7177         if (ext) {
7178                 rtg_ext_ptr = (struct scsi_target_group_data_extended *)
7179                     ctsio->kern_data_ptr;
7180                 scsi_ulto4b(total_len - 4, rtg_ext_ptr->length);
7181                 rtg_ext_ptr->format_type = 0x10;
7182                 rtg_ext_ptr->implicit_transition_time = 0;
7183                 tpg_desc = &rtg_ext_ptr->groups[0];
7184         } else {
7185                 rtg_ptr = (struct scsi_target_group_data *)
7186                     ctsio->kern_data_ptr;
7187                 scsi_ulto4b(total_len - 4, rtg_ptr->length);
7188                 tpg_desc = &rtg_ptr->groups[0];
7189         }
7190
7191         mtx_lock(&softc->ctl_lock);
7192         pg = softc->port_min / softc->port_cnt;
7193         if (lun->flags & (CTL_LUN_PRIMARY_SC | CTL_LUN_PEER_SC_PRIMARY)) {
7194                 /* Some shelf is known to be primary. */
7195                 if (softc->ha_link == CTL_HA_LINK_OFFLINE)
7196                         os = TPG_ASYMMETRIC_ACCESS_UNAVAILABLE;
7197                 else if (softc->ha_link == CTL_HA_LINK_UNKNOWN)
7198                         os = TPG_ASYMMETRIC_ACCESS_TRANSITIONING;
7199                 else if (softc->ha_mode == CTL_HA_MODE_ACT_STBY)
7200                         os = TPG_ASYMMETRIC_ACCESS_STANDBY;
7201                 else
7202                         os = TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7203                 if (lun->flags & CTL_LUN_PRIMARY_SC) {
7204                         ts = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7205                 } else {
7206                         ts = os;
7207                         os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7208                 }
7209         } else {
7210                 /* No known primary shelf. */
7211                 if (softc->ha_link == CTL_HA_LINK_OFFLINE) {
7212                         ts = TPG_ASYMMETRIC_ACCESS_UNAVAILABLE;
7213                         os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7214                 } else if (softc->ha_link == CTL_HA_LINK_UNKNOWN) {
7215                         ts = TPG_ASYMMETRIC_ACCESS_TRANSITIONING;
7216                         os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7217                 } else {
7218                         ts = os = TPG_ASYMMETRIC_ACCESS_TRANSITIONING;
7219                 }
7220         }
7221         if (shared_group) {
7222                 tpg_desc->pref_state = ts;
7223                 tpg_desc->support = TPG_AO_SUP | TPG_AN_SUP | TPG_S_SUP |
7224                     TPG_U_SUP | TPG_T_SUP;
7225                 scsi_ulto2b(1, tpg_desc->target_port_group);
7226                 tpg_desc->status = TPG_IMPLICIT;
7227                 pc = 0;
7228                 STAILQ_FOREACH(port, &softc->port_list, links) {
7229                         if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7230                                 continue;
7231                         if (!softc->is_single &&
7232                             (port->status & CTL_PORT_STATUS_HA_SHARED) == 0)
7233                                 continue;
7234                         if (ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
7235                                 continue;
7236                         scsi_ulto2b(port->targ_port, tpg_desc->descriptors[pc].
7237                             relative_target_port_identifier);
7238                         pc++;
7239                 }
7240                 tpg_desc->target_port_count = pc;
7241                 tpg_desc = (struct scsi_target_port_group_descriptor *)
7242                     &tpg_desc->descriptors[pc];
7243         }
7244         for (g = 0; g < num_ha_groups; g++) {
7245                 tpg_desc->pref_state = (g == pg) ? ts : os;
7246                 tpg_desc->support = TPG_AO_SUP | TPG_AN_SUP | TPG_S_SUP |
7247                     TPG_U_SUP | TPG_T_SUP;
7248                 scsi_ulto2b(2 + g, tpg_desc->target_port_group);
7249                 tpg_desc->status = TPG_IMPLICIT;
7250                 pc = 0;
7251                 STAILQ_FOREACH(port, &softc->port_list, links) {
7252                         if (port->targ_port < g * softc->port_cnt ||
7253                             port->targ_port >= (g + 1) * softc->port_cnt)
7254                                 continue;
7255                         if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7256                                 continue;
7257                         if (port->status & CTL_PORT_STATUS_HA_SHARED)
7258                                 continue;
7259                         if (ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
7260                                 continue;
7261                         scsi_ulto2b(port->targ_port, tpg_desc->descriptors[pc].
7262                             relative_target_port_identifier);
7263                         pc++;
7264                 }
7265                 tpg_desc->target_port_count = pc;
7266                 tpg_desc = (struct scsi_target_port_group_descriptor *)
7267                     &tpg_desc->descriptors[pc];
7268         }
7269         mtx_unlock(&softc->ctl_lock);
7270
7271         ctl_set_success(ctsio);
7272         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7273         ctsio->be_move_done = ctl_config_move_done;
7274         ctl_datamove((union ctl_io *)ctsio);
7275         return(retval);
7276 }
7277
7278 int
7279 ctl_report_supported_opcodes(struct ctl_scsiio *ctsio)
7280 {
7281         struct ctl_lun *lun;
7282         struct scsi_report_supported_opcodes *cdb;
7283         const struct ctl_cmd_entry *entry, *sentry;
7284         struct scsi_report_supported_opcodes_all *all;
7285         struct scsi_report_supported_opcodes_descr *descr;
7286         struct scsi_report_supported_opcodes_one *one;
7287         int retval;
7288         int alloc_len, total_len;
7289         int opcode, service_action, i, j, num;
7290
7291         CTL_DEBUG_PRINT(("ctl_report_supported_opcodes\n"));
7292
7293         cdb = (struct scsi_report_supported_opcodes *)ctsio->cdb;
7294         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7295
7296         retval = CTL_RETVAL_COMPLETE;
7297
7298         opcode = cdb->requested_opcode;
7299         service_action = scsi_2btoul(cdb->requested_service_action);
7300         switch (cdb->options & RSO_OPTIONS_MASK) {
7301         case RSO_OPTIONS_ALL:
7302                 num = 0;
7303                 for (i = 0; i < 256; i++) {
7304                         entry = &ctl_cmd_table[i];
7305                         if (entry->flags & CTL_CMD_FLAG_SA5) {
7306                                 for (j = 0; j < 32; j++) {
7307                                         sentry = &((const struct ctl_cmd_entry *)
7308                                             entry->execute)[j];
7309                                         if (ctl_cmd_applicable(
7310                                             lun->be_lun->lun_type, sentry))
7311                                                 num++;
7312                                 }
7313                         } else {
7314                                 if (ctl_cmd_applicable(lun->be_lun->lun_type,
7315                                     entry))
7316                                         num++;
7317                         }
7318                 }
7319                 total_len = sizeof(struct scsi_report_supported_opcodes_all) +
7320                     num * sizeof(struct scsi_report_supported_opcodes_descr);
7321                 break;
7322         case RSO_OPTIONS_OC:
7323                 if (ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) {
7324                         ctl_set_invalid_field(/*ctsio*/ ctsio,
7325                                               /*sks_valid*/ 1,
7326                                               /*command*/ 1,
7327                                               /*field*/ 2,
7328                                               /*bit_valid*/ 1,
7329                                               /*bit*/ 2);
7330                         ctl_done((union ctl_io *)ctsio);
7331                         return (CTL_RETVAL_COMPLETE);
7332                 }
7333                 total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7334                 break;
7335         case RSO_OPTIONS_OC_SA:
7336                 if ((ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) == 0 ||
7337                     service_action >= 32) {
7338                         ctl_set_invalid_field(/*ctsio*/ ctsio,
7339                                               /*sks_valid*/ 1,
7340                                               /*command*/ 1,
7341                                               /*field*/ 2,
7342                                               /*bit_valid*/ 1,
7343                                               /*bit*/ 2);
7344                         ctl_done((union ctl_io *)ctsio);
7345                         return (CTL_RETVAL_COMPLETE);
7346                 }
7347                 total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7348                 break;
7349         default:
7350                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7351                                       /*sks_valid*/ 1,
7352                                       /*command*/ 1,
7353                                       /*field*/ 2,
7354                                       /*bit_valid*/ 1,
7355                                       /*bit*/ 2);
7356                 ctl_done((union ctl_io *)ctsio);
7357                 return (CTL_RETVAL_COMPLETE);
7358         }
7359
7360         alloc_len = scsi_4btoul(cdb->length);
7361
7362         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7363
7364         ctsio->kern_sg_entries = 0;
7365
7366         if (total_len < alloc_len) {
7367                 ctsio->residual = alloc_len - total_len;
7368                 ctsio->kern_data_len = total_len;
7369                 ctsio->kern_total_len = total_len;
7370         } else {
7371                 ctsio->residual = 0;
7372                 ctsio->kern_data_len = alloc_len;
7373                 ctsio->kern_total_len = alloc_len;
7374         }
7375         ctsio->kern_data_resid = 0;
7376         ctsio->kern_rel_offset = 0;
7377
7378         switch (cdb->options & RSO_OPTIONS_MASK) {
7379         case RSO_OPTIONS_ALL:
7380                 all = (struct scsi_report_supported_opcodes_all *)
7381                     ctsio->kern_data_ptr;
7382                 num = 0;
7383                 for (i = 0; i < 256; i++) {
7384                         entry = &ctl_cmd_table[i];
7385                         if (entry->flags & CTL_CMD_FLAG_SA5) {
7386                                 for (j = 0; j < 32; j++) {
7387                                         sentry = &((const struct ctl_cmd_entry *)
7388                                             entry->execute)[j];
7389                                         if (!ctl_cmd_applicable(
7390                                             lun->be_lun->lun_type, sentry))
7391                                                 continue;
7392                                         descr = &all->descr[num++];
7393                                         descr->opcode = i;
7394                                         scsi_ulto2b(j, descr->service_action);
7395                                         descr->flags = RSO_SERVACTV;
7396                                         scsi_ulto2b(sentry->length,
7397                                             descr->cdb_length);
7398                                 }
7399                         } else {
7400                                 if (!ctl_cmd_applicable(lun->be_lun->lun_type,
7401                                     entry))
7402                                         continue;
7403                                 descr = &all->descr[num++];
7404                                 descr->opcode = i;
7405                                 scsi_ulto2b(0, descr->service_action);
7406                                 descr->flags = 0;
7407                                 scsi_ulto2b(entry->length, descr->cdb_length);
7408                         }
7409                 }
7410                 scsi_ulto4b(
7411                     num * sizeof(struct scsi_report_supported_opcodes_descr),
7412                     all->length);
7413                 break;
7414         case RSO_OPTIONS_OC:
7415                 one = (struct scsi_report_supported_opcodes_one *)
7416                     ctsio->kern_data_ptr;
7417                 entry = &ctl_cmd_table[opcode];
7418                 goto fill_one;
7419         case RSO_OPTIONS_OC_SA:
7420                 one = (struct scsi_report_supported_opcodes_one *)
7421                     ctsio->kern_data_ptr;
7422                 entry = &ctl_cmd_table[opcode];
7423                 entry = &((const struct ctl_cmd_entry *)
7424                     entry->execute)[service_action];
7425 fill_one:
7426                 if (ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
7427                         one->support = 3;
7428                         scsi_ulto2b(entry->length, one->cdb_length);
7429                         one->cdb_usage[0] = opcode;
7430                         memcpy(&one->cdb_usage[1], entry->usage,
7431                             entry->length - 1);
7432                 } else
7433                         one->support = 1;
7434                 break;
7435         }
7436
7437         ctl_set_success(ctsio);
7438         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7439         ctsio->be_move_done = ctl_config_move_done;
7440         ctl_datamove((union ctl_io *)ctsio);
7441         return(retval);
7442 }
7443
7444 int
7445 ctl_report_supported_tmf(struct ctl_scsiio *ctsio)
7446 {
7447         struct scsi_report_supported_tmf *cdb;
7448         struct scsi_report_supported_tmf_data *data;
7449         int retval;
7450         int alloc_len, total_len;
7451
7452         CTL_DEBUG_PRINT(("ctl_report_supported_tmf\n"));
7453
7454         cdb = (struct scsi_report_supported_tmf *)ctsio->cdb;
7455
7456         retval = CTL_RETVAL_COMPLETE;
7457
7458         total_len = sizeof(struct scsi_report_supported_tmf_data);
7459         alloc_len = scsi_4btoul(cdb->length);
7460
7461         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7462
7463         ctsio->kern_sg_entries = 0;
7464
7465         if (total_len < alloc_len) {
7466                 ctsio->residual = alloc_len - total_len;
7467                 ctsio->kern_data_len = total_len;
7468                 ctsio->kern_total_len = total_len;
7469         } else {
7470                 ctsio->residual = 0;
7471                 ctsio->kern_data_len = alloc_len;
7472                 ctsio->kern_total_len = alloc_len;
7473         }
7474         ctsio->kern_data_resid = 0;
7475         ctsio->kern_rel_offset = 0;
7476
7477         data = (struct scsi_report_supported_tmf_data *)ctsio->kern_data_ptr;
7478         data->byte1 |= RST_ATS | RST_ATSS | RST_CTSS | RST_LURS | RST_QTS |
7479             RST_TRS;
7480         data->byte2 |= RST_QAES | RST_QTSS | RST_ITNRS;
7481
7482         ctl_set_success(ctsio);
7483         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7484         ctsio->be_move_done = ctl_config_move_done;
7485         ctl_datamove((union ctl_io *)ctsio);
7486         return (retval);
7487 }
7488
7489 int
7490 ctl_report_timestamp(struct ctl_scsiio *ctsio)
7491 {
7492         struct scsi_report_timestamp *cdb;
7493         struct scsi_report_timestamp_data *data;
7494         struct timeval tv;
7495         int64_t timestamp;
7496         int retval;
7497         int alloc_len, total_len;
7498
7499         CTL_DEBUG_PRINT(("ctl_report_timestamp\n"));
7500
7501         cdb = (struct scsi_report_timestamp *)ctsio->cdb;
7502
7503         retval = CTL_RETVAL_COMPLETE;
7504
7505         total_len = sizeof(struct scsi_report_timestamp_data);
7506         alloc_len = scsi_4btoul(cdb->length);
7507
7508         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7509
7510         ctsio->kern_sg_entries = 0;
7511
7512         if (total_len < alloc_len) {
7513                 ctsio->residual = alloc_len - total_len;
7514                 ctsio->kern_data_len = total_len;
7515                 ctsio->kern_total_len = total_len;
7516         } else {
7517                 ctsio->residual = 0;
7518                 ctsio->kern_data_len = alloc_len;
7519                 ctsio->kern_total_len = alloc_len;
7520         }
7521         ctsio->kern_data_resid = 0;
7522         ctsio->kern_rel_offset = 0;
7523
7524         data = (struct scsi_report_timestamp_data *)ctsio->kern_data_ptr;
7525         scsi_ulto2b(sizeof(*data) - 2, data->length);
7526         data->origin = RTS_ORIG_OUTSIDE;
7527         getmicrotime(&tv);
7528         timestamp = (int64_t)tv.tv_sec * 1000 + tv.tv_usec / 1000;
7529         scsi_ulto4b(timestamp >> 16, data->timestamp);
7530         scsi_ulto2b(timestamp & 0xffff, &data->timestamp[4]);
7531
7532         ctl_set_success(ctsio);
7533         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7534         ctsio->be_move_done = ctl_config_move_done;
7535         ctl_datamove((union ctl_io *)ctsio);
7536         return (retval);
7537 }
7538
7539 int
7540 ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7541 {
7542         struct scsi_per_res_in *cdb;
7543         int alloc_len, total_len = 0;
7544         /* struct scsi_per_res_in_rsrv in_data; */
7545         struct ctl_lun *lun;
7546         struct ctl_softc *softc;
7547         uint64_t key;
7548
7549         CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7550
7551         cdb = (struct scsi_per_res_in *)ctsio->cdb;
7552
7553         alloc_len = scsi_2btoul(cdb->length);
7554
7555         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7556         softc = lun->ctl_softc;
7557
7558 retry:
7559         mtx_lock(&lun->lun_lock);
7560         switch (cdb->action) {
7561         case SPRI_RK: /* read keys */
7562                 total_len = sizeof(struct scsi_per_res_in_keys) +
7563                         lun->pr_key_count *
7564                         sizeof(struct scsi_per_res_key);
7565                 break;
7566         case SPRI_RR: /* read reservation */
7567                 if (lun->flags & CTL_LUN_PR_RESERVED)
7568                         total_len = sizeof(struct scsi_per_res_in_rsrv);
7569                 else
7570                         total_len = sizeof(struct scsi_per_res_in_header);
7571                 break;
7572         case SPRI_RC: /* report capabilities */
7573                 total_len = sizeof(struct scsi_per_res_cap);
7574                 break;
7575         case SPRI_RS: /* read full status */
7576                 total_len = sizeof(struct scsi_per_res_in_header) +
7577                     (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7578                     lun->pr_key_count;
7579                 break;
7580         default:
7581                 panic("%s: Invalid PR type %#x", __func__, cdb->action);
7582         }
7583         mtx_unlock(&lun->lun_lock);
7584
7585         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7586
7587         if (total_len < alloc_len) {
7588                 ctsio->residual = alloc_len - total_len;
7589                 ctsio->kern_data_len = total_len;
7590                 ctsio->kern_total_len = total_len;
7591         } else {
7592                 ctsio->residual = 0;
7593                 ctsio->kern_data_len = alloc_len;
7594                 ctsio->kern_total_len = alloc_len;
7595         }
7596
7597         ctsio->kern_data_resid = 0;
7598         ctsio->kern_rel_offset = 0;
7599         ctsio->kern_sg_entries = 0;
7600
7601         mtx_lock(&lun->lun_lock);
7602         switch (cdb->action) {
7603         case SPRI_RK: { // read keys
7604         struct scsi_per_res_in_keys *res_keys;
7605                 int i, key_count;
7606
7607                 res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7608
7609                 /*
7610                  * We had to drop the lock to allocate our buffer, which
7611                  * leaves time for someone to come in with another
7612                  * persistent reservation.  (That is unlikely, though,
7613                  * since this should be the only persistent reservation
7614                  * command active right now.)
7615                  */
7616                 if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7617                     (lun->pr_key_count *
7618                      sizeof(struct scsi_per_res_key)))){
7619                         mtx_unlock(&lun->lun_lock);
7620                         free(ctsio->kern_data_ptr, M_CTL);
7621                         printf("%s: reservation length changed, retrying\n",
7622                                __func__);
7623                         goto retry;
7624                 }
7625
7626                 scsi_ulto4b(lun->pr_generation, res_keys->header.generation);
7627
7628                 scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7629                              lun->pr_key_count, res_keys->header.length);
7630
7631                 for (i = 0, key_count = 0; i < CTL_MAX_INITIATORS; i++) {
7632                         if ((key = ctl_get_prkey(lun, i)) == 0)
7633                                 continue;
7634
7635                         /*
7636                          * We used lun->pr_key_count to calculate the
7637                          * size to allocate.  If it turns out the number of
7638                          * initiators with the registered flag set is
7639                          * larger than that (i.e. they haven't been kept in
7640                          * sync), we've got a problem.
7641                          */
7642                         if (key_count >= lun->pr_key_count) {
7643                                 key_count++;
7644                                 continue;
7645                         }
7646                         scsi_u64to8b(key, res_keys->keys[key_count].key);
7647                         key_count++;
7648                 }
7649                 break;
7650         }
7651         case SPRI_RR: { // read reservation
7652                 struct scsi_per_res_in_rsrv *res;
7653                 int tmp_len, header_only;
7654
7655                 res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7656
7657                 scsi_ulto4b(lun->pr_generation, res->header.generation);
7658
7659                 if (lun->flags & CTL_LUN_PR_RESERVED)
7660                 {
7661                         tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7662                         scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7663                                     res->header.length);
7664                         header_only = 0;
7665                 } else {
7666                         tmp_len = sizeof(struct scsi_per_res_in_header);
7667                         scsi_ulto4b(0, res->header.length);
7668                         header_only = 1;
7669                 }
7670
7671                 /*
7672                  * We had to drop the lock to allocate our buffer, which
7673                  * leaves time for someone to come in with another
7674                  * persistent reservation.  (That is unlikely, though,
7675                  * since this should be the only persistent reservation
7676                  * command active right now.)
7677                  */
7678                 if (tmp_len != total_len) {
7679                         mtx_unlock(&lun->lun_lock);
7680                         free(ctsio->kern_data_ptr, M_CTL);
7681                         printf("%s: reservation status changed, retrying\n",
7682                                __func__);
7683                         goto retry;
7684                 }
7685
7686                 /*
7687                  * No reservation held, so we're done.
7688                  */
7689                 if (header_only != 0)
7690                         break;
7691
7692                 /*
7693                  * If the registration is an All Registrants type, the key
7694                  * is 0, since it doesn't really matter.
7695                  */
7696                 if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7697                         scsi_u64to8b(ctl_get_prkey(lun, lun->pr_res_idx),
7698                             res->data.reservation);
7699                 }
7700                 res->data.scopetype = lun->pr_res_type;
7701                 break;
7702         }
7703         case SPRI_RC:     //report capabilities
7704         {
7705                 struct scsi_per_res_cap *res_cap;
7706                 uint16_t type_mask;
7707
7708                 res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7709                 scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7710                 res_cap->flags1 = SPRI_CRH;
7711                 res_cap->flags2 = SPRI_TMV | SPRI_ALLOW_5;
7712                 type_mask = SPRI_TM_WR_EX_AR |
7713                             SPRI_TM_EX_AC_RO |
7714                             SPRI_TM_WR_EX_RO |
7715                             SPRI_TM_EX_AC |
7716                             SPRI_TM_WR_EX |
7717                             SPRI_TM_EX_AC_AR;
7718                 scsi_ulto2b(type_mask, res_cap->type_mask);
7719                 break;
7720         }
7721         case SPRI_RS: { // read full status
7722                 struct scsi_per_res_in_full *res_status;
7723                 struct scsi_per_res_in_full_desc *res_desc;
7724                 struct ctl_port *port;
7725                 int i, len;
7726
7727                 res_status = (struct scsi_per_res_in_full*)ctsio->kern_data_ptr;
7728
7729                 /*
7730                  * We had to drop the lock to allocate our buffer, which
7731                  * leaves time for someone to come in with another
7732                  * persistent reservation.  (That is unlikely, though,
7733                  * since this should be the only persistent reservation
7734                  * command active right now.)
7735                  */
7736                 if (total_len < (sizeof(struct scsi_per_res_in_header) +
7737                     (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7738                      lun->pr_key_count)){
7739                         mtx_unlock(&lun->lun_lock);
7740                         free(ctsio->kern_data_ptr, M_CTL);
7741                         printf("%s: reservation length changed, retrying\n",
7742                                __func__);
7743                         goto retry;
7744                 }
7745
7746                 scsi_ulto4b(lun->pr_generation, res_status->header.generation);
7747
7748                 res_desc = &res_status->desc[0];
7749                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
7750                         if ((key = ctl_get_prkey(lun, i)) == 0)
7751                                 continue;
7752
7753                         scsi_u64to8b(key, res_desc->res_key.key);
7754                         if ((lun->flags & CTL_LUN_PR_RESERVED) &&
7755                             (lun->pr_res_idx == i ||
7756                              lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS)) {
7757                                 res_desc->flags = SPRI_FULL_R_HOLDER;
7758                                 res_desc->scopetype = lun->pr_res_type;
7759                         }
7760                         scsi_ulto2b(i / CTL_MAX_INIT_PER_PORT,
7761                             res_desc->rel_trgt_port_id);
7762                         len = 0;
7763                         port = softc->ctl_ports[i / CTL_MAX_INIT_PER_PORT];
7764                         if (port != NULL)
7765                                 len = ctl_create_iid(port,
7766                                     i % CTL_MAX_INIT_PER_PORT,
7767                                     res_desc->transport_id);
7768                         scsi_ulto4b(len, res_desc->additional_length);
7769                         res_desc = (struct scsi_per_res_in_full_desc *)
7770                             &res_desc->transport_id[len];
7771                 }
7772                 scsi_ulto4b((uint8_t *)res_desc - (uint8_t *)&res_status->desc[0],
7773                     res_status->header.length);
7774                 break;
7775         }
7776         default:
7777                 panic("%s: Invalid PR type %#x", __func__, cdb->action);
7778         }
7779         mtx_unlock(&lun->lun_lock);
7780
7781         ctl_set_success(ctsio);
7782         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7783         ctsio->be_move_done = ctl_config_move_done;
7784         ctl_datamove((union ctl_io *)ctsio);
7785         return (CTL_RETVAL_COMPLETE);
7786 }
7787
7788 /*
7789  * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
7790  * it should return.
7791  */
7792 static int
7793 ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
7794                 uint64_t sa_res_key, uint8_t type, uint32_t residx,
7795                 struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
7796                 struct scsi_per_res_out_parms* param)
7797 {
7798         union ctl_ha_msg persis_io;
7799         int i;
7800
7801         mtx_lock(&lun->lun_lock);
7802         if (sa_res_key == 0) {
7803                 if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
7804                         /* validate scope and type */
7805                         if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7806                              SPR_LU_SCOPE) {
7807                                 mtx_unlock(&lun->lun_lock);
7808                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7809                                                       /*sks_valid*/ 1,
7810                                                       /*command*/ 1,
7811                                                       /*field*/ 2,
7812                                                       /*bit_valid*/ 1,
7813                                                       /*bit*/ 4);
7814                                 ctl_done((union ctl_io *)ctsio);
7815                                 return (1);
7816                         }
7817
7818                         if (type>8 || type==2 || type==4 || type==0) {
7819                                 mtx_unlock(&lun->lun_lock);
7820                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7821                                                       /*sks_valid*/ 1,
7822                                                       /*command*/ 1,
7823                                                       /*field*/ 2,
7824                                                       /*bit_valid*/ 1,
7825                                                       /*bit*/ 0);
7826                                 ctl_done((union ctl_io *)ctsio);
7827                                 return (1);
7828                         }
7829
7830                         /*
7831                          * Unregister everybody else and build UA for
7832                          * them
7833                          */
7834                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
7835                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
7836                                         continue;
7837
7838                                 ctl_clr_prkey(lun, i);
7839                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7840                         }
7841                         lun->pr_key_count = 1;
7842                         lun->pr_res_type = type;
7843                         if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
7844                             lun->pr_res_type != SPR_TYPE_EX_AC_AR)
7845                                 lun->pr_res_idx = residx;
7846                         lun->pr_generation++;
7847                         mtx_unlock(&lun->lun_lock);
7848
7849                         /* send msg to other side */
7850                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7851                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7852                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7853                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
7854                         persis_io.pr.pr_info.res_type = type;
7855                         memcpy(persis_io.pr.pr_info.sa_res_key,
7856                                param->serv_act_res_key,
7857                                sizeof(param->serv_act_res_key));
7858                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
7859                             sizeof(persis_io.pr), M_WAITOK);
7860                 } else {
7861                         /* not all registrants */
7862                         mtx_unlock(&lun->lun_lock);
7863                         free(ctsio->kern_data_ptr, M_CTL);
7864                         ctl_set_invalid_field(ctsio,
7865                                               /*sks_valid*/ 1,
7866                                               /*command*/ 0,
7867                                               /*field*/ 8,
7868                                               /*bit_valid*/ 0,
7869                                               /*bit*/ 0);
7870                         ctl_done((union ctl_io *)ctsio);
7871                         return (1);
7872                 }
7873         } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
7874                 || !(lun->flags & CTL_LUN_PR_RESERVED)) {
7875                 int found = 0;
7876
7877                 if (res_key == sa_res_key) {
7878                         /* special case */
7879                         /*
7880                          * The spec implies this is not good but doesn't
7881                          * say what to do. There are two choices either
7882                          * generate a res conflict or check condition
7883                          * with illegal field in parameter data. Since
7884                          * that is what is done when the sa_res_key is
7885                          * zero I'll take that approach since this has
7886                          * to do with the sa_res_key.
7887                          */
7888                         mtx_unlock(&lun->lun_lock);
7889                         free(ctsio->kern_data_ptr, M_CTL);
7890                         ctl_set_invalid_field(ctsio,
7891                                               /*sks_valid*/ 1,
7892                                               /*command*/ 0,
7893                                               /*field*/ 8,
7894                                               /*bit_valid*/ 0,
7895                                               /*bit*/ 0);
7896                         ctl_done((union ctl_io *)ctsio);
7897                         return (1);
7898                 }
7899
7900                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
7901                         if (ctl_get_prkey(lun, i) != sa_res_key)
7902                                 continue;
7903
7904                         found = 1;
7905                         ctl_clr_prkey(lun, i);
7906                         lun->pr_key_count--;
7907                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7908                 }
7909                 if (!found) {
7910                         mtx_unlock(&lun->lun_lock);
7911                         free(ctsio->kern_data_ptr, M_CTL);
7912                         ctl_set_reservation_conflict(ctsio);
7913                         ctl_done((union ctl_io *)ctsio);
7914                         return (CTL_RETVAL_COMPLETE);
7915                 }
7916                 lun->pr_generation++;
7917                 mtx_unlock(&lun->lun_lock);
7918
7919                 /* send msg to other side */
7920                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7921                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7922                 persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7923                 persis_io.pr.pr_info.residx = lun->pr_res_idx;
7924                 persis_io.pr.pr_info.res_type = type;
7925                 memcpy(persis_io.pr.pr_info.sa_res_key,
7926                        param->serv_act_res_key,
7927                        sizeof(param->serv_act_res_key));
7928                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
7929                     sizeof(persis_io.pr), M_WAITOK);
7930         } else {
7931                 /* Reserved but not all registrants */
7932                 /* sa_res_key is res holder */
7933                 if (sa_res_key == ctl_get_prkey(lun, lun->pr_res_idx)) {
7934                         /* validate scope and type */
7935                         if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7936                              SPR_LU_SCOPE) {
7937                                 mtx_unlock(&lun->lun_lock);
7938                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7939                                                       /*sks_valid*/ 1,
7940                                                       /*command*/ 1,
7941                                                       /*field*/ 2,
7942                                                       /*bit_valid*/ 1,
7943                                                       /*bit*/ 4);
7944                                 ctl_done((union ctl_io *)ctsio);
7945                                 return (1);
7946                         }
7947
7948                         if (type>8 || type==2 || type==4 || type==0) {
7949                                 mtx_unlock(&lun->lun_lock);
7950                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7951                                                       /*sks_valid*/ 1,
7952                                                       /*command*/ 1,
7953                                                       /*field*/ 2,
7954                                                       /*bit_valid*/ 1,
7955                                                       /*bit*/ 0);
7956                                 ctl_done((union ctl_io *)ctsio);
7957                                 return (1);
7958                         }
7959
7960                         /*
7961                          * Do the following:
7962                          * if sa_res_key != res_key remove all
7963                          * registrants w/sa_res_key and generate UA
7964                          * for these registrants(Registrations
7965                          * Preempted) if it wasn't an exclusive
7966                          * reservation generate UA(Reservations
7967                          * Preempted) for all other registered nexuses
7968                          * if the type has changed. Establish the new
7969                          * reservation and holder. If res_key and
7970                          * sa_res_key are the same do the above
7971                          * except don't unregister the res holder.
7972                          */
7973
7974                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
7975                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
7976                                         continue;
7977
7978                                 if (sa_res_key == ctl_get_prkey(lun, i)) {
7979                                         ctl_clr_prkey(lun, i);
7980                                         lun->pr_key_count--;
7981                                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7982                                 } else if (type != lun->pr_res_type &&
7983                                     (lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
7984                                      lun->pr_res_type == SPR_TYPE_EX_AC_RO)) {
7985                                         ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
7986                                 }
7987                         }
7988                         lun->pr_res_type = type;
7989                         if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
7990                             lun->pr_res_type != SPR_TYPE_EX_AC_AR)
7991                                 lun->pr_res_idx = residx;
7992                         else
7993                                 lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
7994                         lun->pr_generation++;
7995                         mtx_unlock(&lun->lun_lock);
7996
7997                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7998                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7999                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8000                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8001                         persis_io.pr.pr_info.res_type = type;
8002                         memcpy(persis_io.pr.pr_info.sa_res_key,
8003                                param->serv_act_res_key,
8004                                sizeof(param->serv_act_res_key));
8005                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8006                             sizeof(persis_io.pr), M_WAITOK);
8007                 } else {
8008                         /*
8009                          * sa_res_key is not the res holder just
8010                          * remove registrants
8011                          */
8012                         int found=0;
8013
8014                         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8015                                 if (sa_res_key != ctl_get_prkey(lun, i))
8016                                         continue;
8017
8018                                 found = 1;
8019                                 ctl_clr_prkey(lun, i);
8020                                 lun->pr_key_count--;
8021                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8022                         }
8023
8024                         if (!found) {
8025                                 mtx_unlock(&lun->lun_lock);
8026                                 free(ctsio->kern_data_ptr, M_CTL);
8027                                 ctl_set_reservation_conflict(ctsio);
8028                                 ctl_done((union ctl_io *)ctsio);
8029                                 return (1);
8030                         }
8031                         lun->pr_generation++;
8032                         mtx_unlock(&lun->lun_lock);
8033
8034                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8035                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8036                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8037                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8038                         persis_io.pr.pr_info.res_type = type;
8039                         memcpy(persis_io.pr.pr_info.sa_res_key,
8040                                param->serv_act_res_key,
8041                                sizeof(param->serv_act_res_key));
8042                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8043                             sizeof(persis_io.pr), M_WAITOK);
8044                 }
8045         }
8046         return (0);
8047 }
8048
8049 static void
8050 ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
8051 {
8052         uint64_t sa_res_key;
8053         int i;
8054
8055         sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
8056
8057         if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8058          || lun->pr_res_idx == CTL_PR_NO_RESERVATION
8059          || sa_res_key != ctl_get_prkey(lun, lun->pr_res_idx)) {
8060                 if (sa_res_key == 0) {
8061                         /*
8062                          * Unregister everybody else and build UA for
8063                          * them
8064                          */
8065                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
8066                                 if (i == msg->pr.pr_info.residx ||
8067                                     ctl_get_prkey(lun, i) == 0)
8068                                         continue;
8069
8070                                 ctl_clr_prkey(lun, i);
8071                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8072                         }
8073
8074                         lun->pr_key_count = 1;
8075                         lun->pr_res_type = msg->pr.pr_info.res_type;
8076                         if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
8077                             lun->pr_res_type != SPR_TYPE_EX_AC_AR)
8078                                 lun->pr_res_idx = msg->pr.pr_info.residx;
8079                 } else {
8080                         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8081                                 if (sa_res_key == ctl_get_prkey(lun, i))
8082                                         continue;
8083
8084                                 ctl_clr_prkey(lun, i);
8085                                 lun->pr_key_count--;
8086                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8087                         }
8088                 }
8089         } else {
8090                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8091                         if (i == msg->pr.pr_info.residx ||
8092                             ctl_get_prkey(lun, i) == 0)
8093                                 continue;
8094
8095                         if (sa_res_key == ctl_get_prkey(lun, i)) {
8096                                 ctl_clr_prkey(lun, i);
8097                                 lun->pr_key_count--;
8098                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8099                         } else if (msg->pr.pr_info.res_type != lun->pr_res_type
8100                             && (lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8101                              lun->pr_res_type == SPR_TYPE_EX_AC_RO)) {
8102                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8103                         }
8104                 }
8105                 lun->pr_res_type = msg->pr.pr_info.res_type;
8106                 if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
8107                     lun->pr_res_type != SPR_TYPE_EX_AC_AR)
8108                         lun->pr_res_idx = msg->pr.pr_info.residx;
8109                 else
8110                         lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8111         }
8112         lun->pr_generation++;
8113
8114 }
8115
8116
8117 int
8118 ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
8119 {
8120         int retval;
8121         u_int32_t param_len;
8122         struct scsi_per_res_out *cdb;
8123         struct ctl_lun *lun;
8124         struct scsi_per_res_out_parms* param;
8125         struct ctl_softc *softc;
8126         uint32_t residx;
8127         uint64_t res_key, sa_res_key, key;
8128         uint8_t type;
8129         union ctl_ha_msg persis_io;
8130         int    i;
8131
8132         CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n"));
8133
8134         retval = CTL_RETVAL_COMPLETE;
8135
8136         cdb = (struct scsi_per_res_out *)ctsio->cdb;
8137         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8138         softc = lun->ctl_softc;
8139
8140         /*
8141          * We only support whole-LUN scope.  The scope & type are ignored for
8142          * register, register and ignore existing key and clear.
8143          * We sometimes ignore scope and type on preempts too!!
8144          * Verify reservation type here as well.
8145          */
8146         type = cdb->scope_type & SPR_TYPE_MASK;
8147         if ((cdb->action == SPRO_RESERVE)
8148          || (cdb->action == SPRO_RELEASE)) {
8149                 if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
8150                         ctl_set_invalid_field(/*ctsio*/ ctsio,
8151                                               /*sks_valid*/ 1,
8152                                               /*command*/ 1,
8153                                               /*field*/ 2,
8154                                               /*bit_valid*/ 1,
8155                                               /*bit*/ 4);
8156                         ctl_done((union ctl_io *)ctsio);
8157                         return (CTL_RETVAL_COMPLETE);
8158                 }
8159
8160                 if (type>8 || type==2 || type==4 || type==0) {
8161                         ctl_set_invalid_field(/*ctsio*/ ctsio,
8162                                               /*sks_valid*/ 1,
8163                                               /*command*/ 1,
8164                                               /*field*/ 2,
8165                                               /*bit_valid*/ 1,
8166                                               /*bit*/ 0);
8167                         ctl_done((union ctl_io *)ctsio);
8168                         return (CTL_RETVAL_COMPLETE);
8169                 }
8170         }
8171
8172         param_len = scsi_4btoul(cdb->length);
8173
8174         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
8175                 ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
8176                 ctsio->kern_data_len = param_len;
8177                 ctsio->kern_total_len = param_len;
8178                 ctsio->kern_data_resid = 0;
8179                 ctsio->kern_rel_offset = 0;
8180                 ctsio->kern_sg_entries = 0;
8181                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
8182                 ctsio->be_move_done = ctl_config_move_done;
8183                 ctl_datamove((union ctl_io *)ctsio);
8184
8185                 return (CTL_RETVAL_COMPLETE);
8186         }
8187
8188         param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
8189
8190         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
8191         res_key = scsi_8btou64(param->res_key.key);
8192         sa_res_key = scsi_8btou64(param->serv_act_res_key);
8193
8194         /*
8195          * Validate the reservation key here except for SPRO_REG_IGNO
8196          * This must be done for all other service actions
8197          */
8198         if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
8199                 mtx_lock(&lun->lun_lock);
8200                 if ((key = ctl_get_prkey(lun, residx)) != 0) {
8201                         if (res_key != key) {
8202                                 /*
8203                                  * The current key passed in doesn't match
8204                                  * the one the initiator previously
8205                                  * registered.
8206                                  */
8207                                 mtx_unlock(&lun->lun_lock);
8208                                 free(ctsio->kern_data_ptr, M_CTL);
8209                                 ctl_set_reservation_conflict(ctsio);
8210                                 ctl_done((union ctl_io *)ctsio);
8211                                 return (CTL_RETVAL_COMPLETE);
8212                         }
8213                 } else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
8214                         /*
8215                          * We are not registered
8216                          */
8217                         mtx_unlock(&lun->lun_lock);
8218                         free(ctsio->kern_data_ptr, M_CTL);
8219                         ctl_set_reservation_conflict(ctsio);
8220                         ctl_done((union ctl_io *)ctsio);
8221                         return (CTL_RETVAL_COMPLETE);
8222                 } else if (res_key != 0) {
8223                         /*
8224                          * We are not registered and trying to register but
8225                          * the register key isn't zero.
8226                          */
8227                         mtx_unlock(&lun->lun_lock);
8228                         free(ctsio->kern_data_ptr, M_CTL);
8229                         ctl_set_reservation_conflict(ctsio);
8230                         ctl_done((union ctl_io *)ctsio);
8231                         return (CTL_RETVAL_COMPLETE);
8232                 }
8233                 mtx_unlock(&lun->lun_lock);
8234         }
8235
8236         switch (cdb->action & SPRO_ACTION_MASK) {
8237         case SPRO_REGISTER:
8238         case SPRO_REG_IGNO: {
8239
8240 #if 0
8241                 printf("Registration received\n");
8242 #endif
8243
8244                 /*
8245                  * We don't support any of these options, as we report in
8246                  * the read capabilities request (see
8247                  * ctl_persistent_reserve_in(), above).
8248                  */
8249                 if ((param->flags & SPR_SPEC_I_PT)
8250                  || (param->flags & SPR_ALL_TG_PT)
8251                  || (param->flags & SPR_APTPL)) {
8252                         int bit_ptr;
8253
8254                         if (param->flags & SPR_APTPL)
8255                                 bit_ptr = 0;
8256                         else if (param->flags & SPR_ALL_TG_PT)
8257                                 bit_ptr = 2;
8258                         else /* SPR_SPEC_I_PT */
8259                                 bit_ptr = 3;
8260
8261                         free(ctsio->kern_data_ptr, M_CTL);
8262                         ctl_set_invalid_field(ctsio,
8263                                               /*sks_valid*/ 1,
8264                                               /*command*/ 0,
8265                                               /*field*/ 20,
8266                                               /*bit_valid*/ 1,
8267                                               /*bit*/ bit_ptr);
8268                         ctl_done((union ctl_io *)ctsio);
8269                         return (CTL_RETVAL_COMPLETE);
8270                 }
8271
8272                 mtx_lock(&lun->lun_lock);
8273
8274                 /*
8275                  * The initiator wants to clear the
8276                  * key/unregister.
8277                  */
8278                 if (sa_res_key == 0) {
8279                         if ((res_key == 0
8280                           && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
8281                          || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
8282                           && ctl_get_prkey(lun, residx) == 0)) {
8283                                 mtx_unlock(&lun->lun_lock);
8284                                 goto done;
8285                         }
8286
8287                         ctl_clr_prkey(lun, residx);
8288                         lun->pr_key_count--;
8289
8290                         if (residx == lun->pr_res_idx) {
8291                                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8292                                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8293
8294                                 if ((lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8295                                      lun->pr_res_type == SPR_TYPE_EX_AC_RO) &&
8296                                     lun->pr_key_count) {
8297                                         /*
8298                                          * If the reservation is a registrants
8299                                          * only type we need to generate a UA
8300                                          * for other registered inits.  The
8301                                          * sense code should be RESERVATIONS
8302                                          * RELEASED
8303                                          */
8304
8305                                         for (i = softc->init_min; i < softc->init_max; i++){
8306                                                 if (ctl_get_prkey(lun, i) == 0)
8307                                                         continue;
8308                                                 ctl_est_ua(lun, i,
8309                                                     CTL_UA_RES_RELEASE);
8310                                         }
8311                                 }
8312                                 lun->pr_res_type = 0;
8313                         } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8314                                 if (lun->pr_key_count==0) {
8315                                         lun->flags &= ~CTL_LUN_PR_RESERVED;
8316                                         lun->pr_res_type = 0;
8317                                         lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8318                                 }
8319                         }
8320                         lun->pr_generation++;
8321                         mtx_unlock(&lun->lun_lock);
8322
8323                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8324                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8325                         persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
8326                         persis_io.pr.pr_info.residx = residx;
8327                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8328                             sizeof(persis_io.pr), M_WAITOK);
8329                 } else /* sa_res_key != 0 */ {
8330
8331                         /*
8332                          * If we aren't registered currently then increment
8333                          * the key count and set the registered flag.
8334                          */
8335                         ctl_alloc_prkey(lun, residx);
8336                         if (ctl_get_prkey(lun, residx) == 0)
8337                                 lun->pr_key_count++;
8338                         ctl_set_prkey(lun, residx, sa_res_key);
8339                         lun->pr_generation++;
8340                         mtx_unlock(&lun->lun_lock);
8341
8342                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8343                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8344                         persis_io.pr.pr_info.action = CTL_PR_REG_KEY;
8345                         persis_io.pr.pr_info.residx = residx;
8346                         memcpy(persis_io.pr.pr_info.sa_res_key,
8347                                param->serv_act_res_key,
8348                                sizeof(param->serv_act_res_key));
8349                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8350                             sizeof(persis_io.pr), M_WAITOK);
8351                 }
8352
8353                 break;
8354         }
8355         case SPRO_RESERVE:
8356 #if 0
8357                 printf("Reserve executed type %d\n", type);
8358 #endif
8359                 mtx_lock(&lun->lun_lock);
8360                 if (lun->flags & CTL_LUN_PR_RESERVED) {
8361                         /*
8362                          * if this isn't the reservation holder and it's
8363                          * not a "all registrants" type or if the type is
8364                          * different then we have a conflict
8365                          */
8366                         if ((lun->pr_res_idx != residx
8367                           && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8368                          || lun->pr_res_type != type) {
8369                                 mtx_unlock(&lun->lun_lock);
8370                                 free(ctsio->kern_data_ptr, M_CTL);
8371                                 ctl_set_reservation_conflict(ctsio);
8372                                 ctl_done((union ctl_io *)ctsio);
8373                                 return (CTL_RETVAL_COMPLETE);
8374                         }
8375                         mtx_unlock(&lun->lun_lock);
8376                 } else /* create a reservation */ {
8377                         /*
8378                          * If it's not an "all registrants" type record
8379                          * reservation holder
8380                          */
8381                         if (type != SPR_TYPE_WR_EX_AR
8382                          && type != SPR_TYPE_EX_AC_AR)
8383                                 lun->pr_res_idx = residx; /* Res holder */
8384                         else
8385                                 lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8386
8387                         lun->flags |= CTL_LUN_PR_RESERVED;
8388                         lun->pr_res_type = type;
8389
8390                         mtx_unlock(&lun->lun_lock);
8391
8392                         /* send msg to other side */
8393                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8394                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8395                         persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8396                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8397                         persis_io.pr.pr_info.res_type = type;
8398                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8399                             sizeof(persis_io.pr), M_WAITOK);
8400                 }
8401                 break;
8402
8403         case SPRO_RELEASE:
8404                 mtx_lock(&lun->lun_lock);
8405                 if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8406                         /* No reservation exists return good status */
8407                         mtx_unlock(&lun->lun_lock);
8408                         goto done;
8409                 }
8410                 /*
8411                  * Is this nexus a reservation holder?
8412                  */
8413                 if (lun->pr_res_idx != residx
8414                  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8415                         /*
8416                          * not a res holder return good status but
8417                          * do nothing
8418                          */
8419                         mtx_unlock(&lun->lun_lock);
8420                         goto done;
8421                 }
8422
8423                 if (lun->pr_res_type != type) {
8424                         mtx_unlock(&lun->lun_lock);
8425                         free(ctsio->kern_data_ptr, M_CTL);
8426                         ctl_set_illegal_pr_release(ctsio);
8427                         ctl_done((union ctl_io *)ctsio);
8428                         return (CTL_RETVAL_COMPLETE);
8429                 }
8430
8431                 /* okay to release */
8432                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8433                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8434                 lun->pr_res_type = 0;
8435
8436                 /*
8437                  * If this isn't an exclusive access reservation and NUAR
8438                  * is not set, generate UA for all other registrants.
8439                  */
8440                 if (type != SPR_TYPE_EX_AC && type != SPR_TYPE_WR_EX &&
8441                     (lun->MODE_CTRL.queue_flags & SCP_NUAR) == 0) {
8442                         for (i = softc->init_min; i < softc->init_max; i++) {
8443                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
8444                                         continue;
8445                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8446                         }
8447                 }
8448                 mtx_unlock(&lun->lun_lock);
8449
8450                 /* Send msg to other side */
8451                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8452                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8453                 persis_io.pr.pr_info.action = CTL_PR_RELEASE;
8454                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8455                      sizeof(persis_io.pr), M_WAITOK);
8456                 break;
8457
8458         case SPRO_CLEAR:
8459                 /* send msg to other side */
8460
8461                 mtx_lock(&lun->lun_lock);
8462                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8463                 lun->pr_res_type = 0;
8464                 lun->pr_key_count = 0;
8465                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8466
8467                 ctl_clr_prkey(lun, residx);
8468                 for (i = 0; i < CTL_MAX_INITIATORS; i++)
8469                         if (ctl_get_prkey(lun, i) != 0) {
8470                                 ctl_clr_prkey(lun, i);
8471                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8472                         }
8473                 lun->pr_generation++;
8474                 mtx_unlock(&lun->lun_lock);
8475
8476                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8477                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8478                 persis_io.pr.pr_info.action = CTL_PR_CLEAR;
8479                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8480                      sizeof(persis_io.pr), M_WAITOK);
8481                 break;
8482
8483         case SPRO_PREEMPT:
8484         case SPRO_PRE_ABO: {
8485                 int nretval;
8486
8487                 nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8488                                           residx, ctsio, cdb, param);
8489                 if (nretval != 0)
8490                         return (CTL_RETVAL_COMPLETE);
8491                 break;
8492         }
8493         default:
8494                 panic("%s: Invalid PR type %#x", __func__, cdb->action);
8495         }
8496
8497 done:
8498         free(ctsio->kern_data_ptr, M_CTL);
8499         ctl_set_success(ctsio);
8500         ctl_done((union ctl_io *)ctsio);
8501
8502         return (retval);
8503 }
8504
8505 /*
8506  * This routine is for handling a message from the other SC pertaining to
8507  * persistent reserve out. All the error checking will have been done
8508  * so only perorming the action need be done here to keep the two
8509  * in sync.
8510  */
8511 static void
8512 ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg)
8513 {
8514         struct ctl_softc *softc = control_softc;
8515         struct ctl_lun *lun;
8516         int i;
8517         uint32_t residx, targ_lun;
8518
8519         targ_lun = msg->hdr.nexus.targ_mapped_lun;
8520         mtx_lock(&softc->ctl_lock);
8521         if ((targ_lun >= CTL_MAX_LUNS) ||
8522             ((lun = softc->ctl_luns[targ_lun]) == NULL)) {
8523                 mtx_unlock(&softc->ctl_lock);
8524                 return;
8525         }
8526         mtx_lock(&lun->lun_lock);
8527         mtx_unlock(&softc->ctl_lock);
8528         if (lun->flags & CTL_LUN_DISABLED) {
8529                 mtx_unlock(&lun->lun_lock);
8530                 return;
8531         }
8532         residx = ctl_get_initindex(&msg->hdr.nexus);
8533         switch(msg->pr.pr_info.action) {
8534         case CTL_PR_REG_KEY:
8535                 ctl_alloc_prkey(lun, msg->pr.pr_info.residx);
8536                 if (ctl_get_prkey(lun, msg->pr.pr_info.residx) == 0)
8537                         lun->pr_key_count++;
8538                 ctl_set_prkey(lun, msg->pr.pr_info.residx,
8539                     scsi_8btou64(msg->pr.pr_info.sa_res_key));
8540                 lun->pr_generation++;
8541                 break;
8542
8543         case CTL_PR_UNREG_KEY:
8544                 ctl_clr_prkey(lun, msg->pr.pr_info.residx);
8545                 lun->pr_key_count--;
8546
8547                 /* XXX Need to see if the reservation has been released */
8548                 /* if so do we need to generate UA? */
8549                 if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8550                         lun->flags &= ~CTL_LUN_PR_RESERVED;
8551                         lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8552
8553                         if ((lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8554                              lun->pr_res_type == SPR_TYPE_EX_AC_RO) &&
8555                             lun->pr_key_count) {
8556                                 /*
8557                                  * If the reservation is a registrants
8558                                  * only type we need to generate a UA
8559                                  * for other registered inits.  The
8560                                  * sense code should be RESERVATIONS
8561                                  * RELEASED
8562                                  */
8563
8564                                 for (i = softc->init_min; i < softc->init_max; i++) {
8565                                         if (ctl_get_prkey(lun, i) == 0)
8566                                                 continue;
8567
8568                                         ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8569                                 }
8570                         }
8571                         lun->pr_res_type = 0;
8572                 } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8573                         if (lun->pr_key_count==0) {
8574                                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8575                                 lun->pr_res_type = 0;
8576                                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8577                         }
8578                 }
8579                 lun->pr_generation++;
8580                 break;
8581
8582         case CTL_PR_RESERVE:
8583                 lun->flags |= CTL_LUN_PR_RESERVED;
8584                 lun->pr_res_type = msg->pr.pr_info.res_type;
8585                 lun->pr_res_idx = msg->pr.pr_info.residx;
8586
8587                 break;
8588
8589         case CTL_PR_RELEASE:
8590                 /*
8591                  * If this isn't an exclusive access reservation and NUAR
8592                  * is not set, generate UA for all other registrants.
8593                  */
8594                 if (lun->pr_res_type != SPR_TYPE_EX_AC &&
8595                     lun->pr_res_type != SPR_TYPE_WR_EX &&
8596                     (lun->MODE_CTRL.queue_flags & SCP_NUAR) == 0) {
8597                         for (i = softc->init_min; i < softc->init_max; i++)
8598                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
8599                                         continue;
8600                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8601                 }
8602
8603                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8604                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8605                 lun->pr_res_type = 0;
8606                 break;
8607
8608         case CTL_PR_PREEMPT:
8609                 ctl_pro_preempt_other(lun, msg);
8610                 break;
8611         case CTL_PR_CLEAR:
8612                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8613                 lun->pr_res_type = 0;
8614                 lun->pr_key_count = 0;
8615                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8616
8617                 for (i=0; i < CTL_MAX_INITIATORS; i++) {
8618                         if (ctl_get_prkey(lun, i) == 0)
8619                                 continue;
8620                         ctl_clr_prkey(lun, i);
8621                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8622                 }
8623                 lun->pr_generation++;
8624                 break;
8625         }
8626
8627         mtx_unlock(&lun->lun_lock);
8628 }
8629
8630 int
8631 ctl_read_write(struct ctl_scsiio *ctsio)
8632 {
8633         struct ctl_lun *lun;
8634         struct ctl_lba_len_flags *lbalen;
8635         uint64_t lba;
8636         uint32_t num_blocks;
8637         int flags, retval;
8638         int isread;
8639
8640         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8641
8642         CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
8643
8644         flags = 0;
8645         isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
8646               || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
8647         switch (ctsio->cdb[0]) {
8648         case READ_6:
8649         case WRITE_6: {
8650                 struct scsi_rw_6 *cdb;
8651
8652                 cdb = (struct scsi_rw_6 *)ctsio->cdb;
8653
8654                 lba = scsi_3btoul(cdb->addr);
8655                 /* only 5 bits are valid in the most significant address byte */
8656                 lba &= 0x1fffff;
8657                 num_blocks = cdb->length;
8658                 /*
8659                  * This is correct according to SBC-2.
8660                  */
8661                 if (num_blocks == 0)
8662                         num_blocks = 256;
8663                 break;
8664         }
8665         case READ_10:
8666         case WRITE_10: {
8667                 struct scsi_rw_10 *cdb;
8668
8669                 cdb = (struct scsi_rw_10 *)ctsio->cdb;
8670                 if (cdb->byte2 & SRW10_FUA)
8671                         flags |= CTL_LLF_FUA;
8672                 if (cdb->byte2 & SRW10_DPO)
8673                         flags |= CTL_LLF_DPO;
8674                 lba = scsi_4btoul(cdb->addr);
8675                 num_blocks = scsi_2btoul(cdb->length);
8676                 break;
8677         }
8678         case WRITE_VERIFY_10: {
8679                 struct scsi_write_verify_10 *cdb;
8680
8681                 cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
8682                 flags |= CTL_LLF_FUA;
8683                 if (cdb->byte2 & SWV_DPO)
8684                         flags |= CTL_LLF_DPO;
8685                 lba = scsi_4btoul(cdb->addr);
8686                 num_blocks = scsi_2btoul(cdb->length);
8687                 break;
8688         }
8689         case READ_12:
8690         case WRITE_12: {
8691                 struct scsi_rw_12 *cdb;
8692
8693                 cdb = (struct scsi_rw_12 *)ctsio->cdb;
8694                 if (cdb->byte2 & SRW12_FUA)
8695                         flags |= CTL_LLF_FUA;
8696                 if (cdb->byte2 & SRW12_DPO)
8697                         flags |= CTL_LLF_DPO;
8698                 lba = scsi_4btoul(cdb->addr);
8699                 num_blocks = scsi_4btoul(cdb->length);
8700                 break;
8701         }
8702         case WRITE_VERIFY_12: {
8703                 struct scsi_write_verify_12 *cdb;
8704
8705                 cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
8706                 flags |= CTL_LLF_FUA;
8707                 if (cdb->byte2 & SWV_DPO)
8708                         flags |= CTL_LLF_DPO;
8709                 lba = scsi_4btoul(cdb->addr);
8710                 num_blocks = scsi_4btoul(cdb->length);
8711                 break;
8712         }
8713         case READ_16:
8714         case WRITE_16: {
8715                 struct scsi_rw_16 *cdb;
8716
8717                 cdb = (struct scsi_rw_16 *)ctsio->cdb;
8718                 if (cdb->byte2 & SRW12_FUA)
8719                         flags |= CTL_LLF_FUA;
8720                 if (cdb->byte2 & SRW12_DPO)
8721                         flags |= CTL_LLF_DPO;
8722                 lba = scsi_8btou64(cdb->addr);
8723                 num_blocks = scsi_4btoul(cdb->length);
8724                 break;
8725         }
8726         case WRITE_ATOMIC_16: {
8727                 struct scsi_write_atomic_16 *cdb;
8728
8729                 if (lun->be_lun->atomicblock == 0) {
8730                         ctl_set_invalid_opcode(ctsio);
8731                         ctl_done((union ctl_io *)ctsio);
8732                         return (CTL_RETVAL_COMPLETE);
8733                 }
8734
8735                 cdb = (struct scsi_write_atomic_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_2btoul(cdb->length);
8742                 if (num_blocks > lun->be_lun->atomicblock) {
8743                         ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
8744                             /*command*/ 1, /*field*/ 12, /*bit_valid*/ 0,
8745                             /*bit*/ 0);
8746                         ctl_done((union ctl_io *)ctsio);
8747                         return (CTL_RETVAL_COMPLETE);
8748                 }
8749                 break;
8750         }
8751         case WRITE_VERIFY_16: {
8752                 struct scsi_write_verify_16 *cdb;
8753
8754                 cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
8755                 flags |= CTL_LLF_FUA;
8756                 if (cdb->byte2 & SWV_DPO)
8757                         flags |= CTL_LLF_DPO;
8758                 lba = scsi_8btou64(cdb->addr);
8759                 num_blocks = scsi_4btoul(cdb->length);
8760                 break;
8761         }
8762         default:
8763                 /*
8764                  * We got a command we don't support.  This shouldn't
8765                  * happen, commands should be filtered out above us.
8766                  */
8767                 ctl_set_invalid_opcode(ctsio);
8768                 ctl_done((union ctl_io *)ctsio);
8769
8770                 return (CTL_RETVAL_COMPLETE);
8771                 break; /* NOTREACHED */
8772         }
8773
8774         /*
8775          * The first check is to make sure we're in bounds, the second
8776          * check is to catch wrap-around problems.  If the lba + num blocks
8777          * is less than the lba, then we've wrapped around and the block
8778          * range is invalid anyway.
8779          */
8780         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8781          || ((lba + num_blocks) < lba)) {
8782                 ctl_set_lba_out_of_range(ctsio,
8783                     MAX(lba, lun->be_lun->maxlba + 1));
8784                 ctl_done((union ctl_io *)ctsio);
8785                 return (CTL_RETVAL_COMPLETE);
8786         }
8787
8788         /*
8789          * According to SBC-3, a transfer length of 0 is not an error.
8790          * Note that this cannot happen with WRITE(6) or READ(6), since 0
8791          * translates to 256 blocks for those commands.
8792          */
8793         if (num_blocks == 0) {
8794                 ctl_set_success(ctsio);
8795                 ctl_done((union ctl_io *)ctsio);
8796                 return (CTL_RETVAL_COMPLETE);
8797         }
8798
8799         /* Set FUA and/or DPO if caches are disabled. */
8800         if (isread) {
8801                 if ((lun->MODE_CACHING.flags1 & SCP_RCD) != 0)
8802                         flags |= CTL_LLF_FUA | CTL_LLF_DPO;
8803         } else {
8804                 if ((lun->MODE_CACHING.flags1 & SCP_WCE) == 0)
8805                         flags |= CTL_LLF_FUA;
8806         }
8807
8808         lbalen = (struct ctl_lba_len_flags *)
8809             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8810         lbalen->lba = lba;
8811         lbalen->len = num_blocks;
8812         lbalen->flags = (isread ? CTL_LLF_READ : CTL_LLF_WRITE) | flags;
8813
8814         ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
8815         ctsio->kern_rel_offset = 0;
8816
8817         CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
8818
8819         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8820         return (retval);
8821 }
8822
8823 static int
8824 ctl_cnw_cont(union ctl_io *io)
8825 {
8826         struct ctl_scsiio *ctsio;
8827         struct ctl_lun *lun;
8828         struct ctl_lba_len_flags *lbalen;
8829         int retval;
8830
8831         ctsio = &io->scsiio;
8832         ctsio->io_hdr.status = CTL_STATUS_NONE;
8833         ctsio->io_hdr.flags &= ~CTL_FLAG_IO_CONT;
8834         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8835         lbalen = (struct ctl_lba_len_flags *)
8836             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8837         lbalen->flags &= ~CTL_LLF_COMPARE;
8838         lbalen->flags |= CTL_LLF_WRITE;
8839
8840         CTL_DEBUG_PRINT(("ctl_cnw_cont: calling data_submit()\n"));
8841         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8842         return (retval);
8843 }
8844
8845 int
8846 ctl_cnw(struct ctl_scsiio *ctsio)
8847 {
8848         struct ctl_lun *lun;
8849         struct ctl_lba_len_flags *lbalen;
8850         uint64_t lba;
8851         uint32_t num_blocks;
8852         int flags, retval;
8853
8854         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8855
8856         CTL_DEBUG_PRINT(("ctl_cnw: command: %#x\n", ctsio->cdb[0]));
8857
8858         flags = 0;
8859         switch (ctsio->cdb[0]) {
8860         case COMPARE_AND_WRITE: {
8861                 struct scsi_compare_and_write *cdb;
8862
8863                 cdb = (struct scsi_compare_and_write *)ctsio->cdb;
8864                 if (cdb->byte2 & SRW10_FUA)
8865                         flags |= CTL_LLF_FUA;
8866                 if (cdb->byte2 & SRW10_DPO)
8867                         flags |= CTL_LLF_DPO;
8868                 lba = scsi_8btou64(cdb->addr);
8869                 num_blocks = cdb->length;
8870                 break;
8871         }
8872         default:
8873                 /*
8874                  * We got a command we don't support.  This shouldn't
8875                  * happen, commands should be filtered out above us.
8876                  */
8877                 ctl_set_invalid_opcode(ctsio);
8878                 ctl_done((union ctl_io *)ctsio);
8879
8880                 return (CTL_RETVAL_COMPLETE);
8881                 break; /* NOTREACHED */
8882         }
8883
8884         /*
8885          * The first check is to make sure we're in bounds, the second
8886          * check is to catch wrap-around problems.  If the lba + num blocks
8887          * is less than the lba, then we've wrapped around and the block
8888          * range is invalid anyway.
8889          */
8890         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8891          || ((lba + num_blocks) < lba)) {
8892                 ctl_set_lba_out_of_range(ctsio,
8893                     MAX(lba, lun->be_lun->maxlba + 1));
8894                 ctl_done((union ctl_io *)ctsio);
8895                 return (CTL_RETVAL_COMPLETE);
8896         }
8897
8898         /*
8899          * According to SBC-3, a transfer length of 0 is not an error.
8900          */
8901         if (num_blocks == 0) {
8902                 ctl_set_success(ctsio);
8903                 ctl_done((union ctl_io *)ctsio);
8904                 return (CTL_RETVAL_COMPLETE);
8905         }
8906
8907         /* Set FUA if write cache is disabled. */
8908         if ((lun->MODE_CACHING.flags1 & SCP_WCE) == 0)
8909                 flags |= CTL_LLF_FUA;
8910
8911         ctsio->kern_total_len = 2 * num_blocks * lun->be_lun->blocksize;
8912         ctsio->kern_rel_offset = 0;
8913
8914         /*
8915          * Set the IO_CONT flag, so that if this I/O gets passed to
8916          * ctl_data_submit_done(), it'll get passed back to
8917          * ctl_ctl_cnw_cont() for further processing.
8918          */
8919         ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
8920         ctsio->io_cont = ctl_cnw_cont;
8921
8922         lbalen = (struct ctl_lba_len_flags *)
8923             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8924         lbalen->lba = lba;
8925         lbalen->len = num_blocks;
8926         lbalen->flags = CTL_LLF_COMPARE | flags;
8927
8928         CTL_DEBUG_PRINT(("ctl_cnw: calling data_submit()\n"));
8929         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8930         return (retval);
8931 }
8932
8933 int
8934 ctl_verify(struct ctl_scsiio *ctsio)
8935 {
8936         struct ctl_lun *lun;
8937         struct ctl_lba_len_flags *lbalen;
8938         uint64_t lba;
8939         uint32_t num_blocks;
8940         int bytchk, flags;
8941         int retval;
8942
8943         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8944
8945         CTL_DEBUG_PRINT(("ctl_verify: command: %#x\n", ctsio->cdb[0]));
8946
8947         bytchk = 0;
8948         flags = CTL_LLF_FUA;
8949         switch (ctsio->cdb[0]) {
8950         case VERIFY_10: {
8951                 struct scsi_verify_10 *cdb;
8952
8953                 cdb = (struct scsi_verify_10 *)ctsio->cdb;
8954                 if (cdb->byte2 & SVFY_BYTCHK)
8955                         bytchk = 1;
8956                 if (cdb->byte2 & SVFY_DPO)
8957                         flags |= CTL_LLF_DPO;
8958                 lba = scsi_4btoul(cdb->addr);
8959                 num_blocks = scsi_2btoul(cdb->length);
8960                 break;
8961         }
8962         case VERIFY_12: {
8963                 struct scsi_verify_12 *cdb;
8964
8965                 cdb = (struct scsi_verify_12 *)ctsio->cdb;
8966                 if (cdb->byte2 & SVFY_BYTCHK)
8967                         bytchk = 1;
8968                 if (cdb->byte2 & SVFY_DPO)
8969                         flags |= CTL_LLF_DPO;
8970                 lba = scsi_4btoul(cdb->addr);
8971                 num_blocks = scsi_4btoul(cdb->length);
8972                 break;
8973         }
8974         case VERIFY_16: {
8975                 struct scsi_rw_16 *cdb;
8976
8977                 cdb = (struct scsi_rw_16 *)ctsio->cdb;
8978                 if (cdb->byte2 & SVFY_BYTCHK)
8979                         bytchk = 1;
8980                 if (cdb->byte2 & SVFY_DPO)
8981                         flags |= CTL_LLF_DPO;
8982                 lba = scsi_8btou64(cdb->addr);
8983                 num_blocks = scsi_4btoul(cdb->length);
8984                 break;
8985         }
8986         default:
8987                 /*
8988                  * We got a command we don't support.  This shouldn't
8989                  * happen, commands should be filtered out above us.
8990                  */
8991                 ctl_set_invalid_opcode(ctsio);
8992                 ctl_done((union ctl_io *)ctsio);
8993                 return (CTL_RETVAL_COMPLETE);
8994         }
8995
8996         /*
8997          * The first check is to make sure we're in bounds, the second
8998          * check is to catch wrap-around problems.  If the lba + num blocks
8999          * is less than the lba, then we've wrapped around and the block
9000          * range is invalid anyway.
9001          */
9002         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9003          || ((lba + num_blocks) < lba)) {
9004                 ctl_set_lba_out_of_range(ctsio,
9005                     MAX(lba, lun->be_lun->maxlba + 1));
9006                 ctl_done((union ctl_io *)ctsio);
9007                 return (CTL_RETVAL_COMPLETE);
9008         }
9009
9010         /*
9011          * According to SBC-3, a transfer length of 0 is not an error.
9012          */
9013         if (num_blocks == 0) {
9014                 ctl_set_success(ctsio);
9015                 ctl_done((union ctl_io *)ctsio);
9016                 return (CTL_RETVAL_COMPLETE);
9017         }
9018
9019         lbalen = (struct ctl_lba_len_flags *)
9020             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9021         lbalen->lba = lba;
9022         lbalen->len = num_blocks;
9023         if (bytchk) {
9024                 lbalen->flags = CTL_LLF_COMPARE | flags;
9025                 ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9026         } else {
9027                 lbalen->flags = CTL_LLF_VERIFY | flags;
9028                 ctsio->kern_total_len = 0;
9029         }
9030         ctsio->kern_rel_offset = 0;
9031
9032         CTL_DEBUG_PRINT(("ctl_verify: calling data_submit()\n"));
9033         retval = lun->backend->data_submit((union ctl_io *)ctsio);
9034         return (retval);
9035 }
9036
9037 int
9038 ctl_report_luns(struct ctl_scsiio *ctsio)
9039 {
9040         struct ctl_softc *softc;
9041         struct scsi_report_luns *cdb;
9042         struct scsi_report_luns_data *lun_data;
9043         struct ctl_lun *lun, *request_lun;
9044         struct ctl_port *port;
9045         int num_luns, retval;
9046         uint32_t alloc_len, lun_datalen;
9047         int num_filled;
9048         uint32_t initidx, targ_lun_id, lun_id;
9049
9050         retval = CTL_RETVAL_COMPLETE;
9051         cdb = (struct scsi_report_luns *)ctsio->cdb;
9052         port = ctl_io_port(&ctsio->io_hdr);
9053         softc = port->ctl_softc;
9054
9055         CTL_DEBUG_PRINT(("ctl_report_luns\n"));
9056
9057         mtx_lock(&softc->ctl_lock);
9058         num_luns = 0;
9059         for (targ_lun_id = 0; targ_lun_id < CTL_MAX_LUNS; targ_lun_id++) {
9060                 if (ctl_lun_map_from_port(port, targ_lun_id) < CTL_MAX_LUNS)
9061                         num_luns++;
9062         }
9063         mtx_unlock(&softc->ctl_lock);
9064
9065         switch (cdb->select_report) {
9066         case RPL_REPORT_DEFAULT:
9067         case RPL_REPORT_ALL:
9068         case RPL_REPORT_NONSUBSID:
9069                 break;
9070         case RPL_REPORT_WELLKNOWN:
9071         case RPL_REPORT_ADMIN:
9072         case RPL_REPORT_CONGLOM:
9073                 num_luns = 0;
9074                 break;
9075         default:
9076                 ctl_set_invalid_field(ctsio,
9077                                       /*sks_valid*/ 1,
9078                                       /*command*/ 1,
9079                                       /*field*/ 2,
9080                                       /*bit_valid*/ 0,
9081                                       /*bit*/ 0);
9082                 ctl_done((union ctl_io *)ctsio);
9083                 return (retval);
9084                 break; /* NOTREACHED */
9085         }
9086
9087         alloc_len = scsi_4btoul(cdb->length);
9088         /*
9089          * The initiator has to allocate at least 16 bytes for this request,
9090          * so he can at least get the header and the first LUN.  Otherwise
9091          * we reject the request (per SPC-3 rev 14, section 6.21).
9092          */
9093         if (alloc_len < (sizeof(struct scsi_report_luns_data) +
9094             sizeof(struct scsi_report_luns_lundata))) {
9095                 ctl_set_invalid_field(ctsio,
9096                                       /*sks_valid*/ 1,
9097                                       /*command*/ 1,
9098                                       /*field*/ 6,
9099                                       /*bit_valid*/ 0,
9100                                       /*bit*/ 0);
9101                 ctl_done((union ctl_io *)ctsio);
9102                 return (retval);
9103         }
9104
9105         request_lun = (struct ctl_lun *)
9106                 ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9107
9108         lun_datalen = sizeof(*lun_data) +
9109                 (num_luns * sizeof(struct scsi_report_luns_lundata));
9110
9111         ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO);
9112         lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
9113         ctsio->kern_sg_entries = 0;
9114
9115         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9116
9117         mtx_lock(&softc->ctl_lock);
9118         for (targ_lun_id = 0, num_filled = 0; targ_lun_id < CTL_MAX_LUNS && num_filled < num_luns; targ_lun_id++) {
9119                 lun_id = ctl_lun_map_from_port(port, targ_lun_id);
9120                 if (lun_id >= CTL_MAX_LUNS)
9121                         continue;
9122                 lun = softc->ctl_luns[lun_id];
9123                 if (lun == NULL)
9124                         continue;
9125
9126                 be64enc(lun_data->luns[num_filled++].lundata,
9127                     ctl_encode_lun(targ_lun_id));
9128
9129                 /*
9130                  * According to SPC-3, rev 14 section 6.21:
9131                  *
9132                  * "The execution of a REPORT LUNS command to any valid and
9133                  * installed logical unit shall clear the REPORTED LUNS DATA
9134                  * HAS CHANGED unit attention condition for all logical
9135                  * units of that target with respect to the requesting
9136                  * initiator. A valid and installed logical unit is one
9137                  * having a PERIPHERAL QUALIFIER of 000b in the standard
9138                  * INQUIRY data (see 6.4.2)."
9139                  *
9140                  * If request_lun is NULL, the LUN this report luns command
9141                  * was issued to is either disabled or doesn't exist. In that
9142                  * case, we shouldn't clear any pending lun change unit
9143                  * attention.
9144                  */
9145                 if (request_lun != NULL) {
9146                         mtx_lock(&lun->lun_lock);
9147                         ctl_clr_ua(lun, initidx, CTL_UA_LUN_CHANGE);
9148                         mtx_unlock(&lun->lun_lock);
9149                 }
9150         }
9151         mtx_unlock(&softc->ctl_lock);
9152
9153         /*
9154          * It's quite possible that we've returned fewer LUNs than we allocated
9155          * space for.  Trim it.
9156          */
9157         lun_datalen = sizeof(*lun_data) +
9158                 (num_filled * sizeof(struct scsi_report_luns_lundata));
9159
9160         if (lun_datalen < alloc_len) {
9161                 ctsio->residual = alloc_len - lun_datalen;
9162                 ctsio->kern_data_len = lun_datalen;
9163                 ctsio->kern_total_len = lun_datalen;
9164         } else {
9165                 ctsio->residual = 0;
9166                 ctsio->kern_data_len = alloc_len;
9167                 ctsio->kern_total_len = alloc_len;
9168         }
9169         ctsio->kern_data_resid = 0;
9170         ctsio->kern_rel_offset = 0;
9171         ctsio->kern_sg_entries = 0;
9172
9173         /*
9174          * We set this to the actual data length, regardless of how much
9175          * space we actually have to return results.  If the user looks at
9176          * this value, he'll know whether or not he allocated enough space
9177          * and reissue the command if necessary.  We don't support well
9178          * known logical units, so if the user asks for that, return none.
9179          */
9180         scsi_ulto4b(lun_datalen - 8, lun_data->length);
9181
9182         /*
9183          * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
9184          * this request.
9185          */
9186         ctl_set_success(ctsio);
9187         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9188         ctsio->be_move_done = ctl_config_move_done;
9189         ctl_datamove((union ctl_io *)ctsio);
9190         return (retval);
9191 }
9192
9193 int
9194 ctl_request_sense(struct ctl_scsiio *ctsio)
9195 {
9196         struct scsi_request_sense *cdb;
9197         struct scsi_sense_data *sense_ptr;
9198         struct ctl_softc *softc;
9199         struct ctl_lun *lun;
9200         uint32_t initidx;
9201         int have_error;
9202         scsi_sense_data_type sense_format;
9203         ctl_ua_type ua_type;
9204         uint8_t asc = 0, ascq = 0;
9205
9206         cdb = (struct scsi_request_sense *)ctsio->cdb;
9207
9208         softc = control_softc;
9209         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9210
9211         CTL_DEBUG_PRINT(("ctl_request_sense\n"));
9212
9213         /*
9214          * Determine which sense format the user wants.
9215          */
9216         if (cdb->byte2 & SRS_DESC)
9217                 sense_format = SSD_TYPE_DESC;
9218         else
9219                 sense_format = SSD_TYPE_FIXED;
9220
9221         ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
9222         sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
9223         ctsio->kern_sg_entries = 0;
9224
9225         /*
9226          * struct scsi_sense_data, which is currently set to 256 bytes, is
9227          * larger than the largest allowed value for the length field in the
9228          * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
9229          */
9230         ctsio->residual = 0;
9231         ctsio->kern_data_len = cdb->length;
9232         ctsio->kern_total_len = cdb->length;
9233
9234         ctsio->kern_data_resid = 0;
9235         ctsio->kern_rel_offset = 0;
9236         ctsio->kern_sg_entries = 0;
9237
9238         /*
9239          * If we don't have a LUN, we don't have any pending sense.
9240          */
9241         if (lun == NULL ||
9242             ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
9243              softc->ha_link < CTL_HA_LINK_UNKNOWN)) {
9244                 /* "Logical unit not supported" */
9245                 ctl_set_sense_data(sense_ptr, NULL, sense_format,
9246                     /*current_error*/ 1,
9247                     /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
9248                     /*asc*/ 0x25,
9249                     /*ascq*/ 0x00,
9250                     SSD_ELEM_NONE);
9251                 goto send;
9252         }
9253
9254         have_error = 0;
9255         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9256         /*
9257          * Check for pending sense, and then for pending unit attentions.
9258          * Pending sense gets returned first, then pending unit attentions.
9259          */
9260         mtx_lock(&lun->lun_lock);
9261 #ifdef CTL_WITH_CA
9262         if (ctl_is_set(lun->have_ca, initidx)) {
9263                 scsi_sense_data_type stored_format;
9264
9265                 /*
9266                  * Check to see which sense format was used for the stored
9267                  * sense data.
9268                  */
9269                 stored_format = scsi_sense_type(&lun->pending_sense[initidx]);
9270
9271                 /*
9272                  * If the user requested a different sense format than the
9273                  * one we stored, then we need to convert it to the other
9274                  * format.  If we're going from descriptor to fixed format
9275                  * sense data, we may lose things in translation, depending
9276                  * on what options were used.
9277                  *
9278                  * If the stored format is SSD_TYPE_NONE (i.e. invalid),
9279                  * for some reason we'll just copy it out as-is.
9280                  */
9281                 if ((stored_format == SSD_TYPE_FIXED)
9282                  && (sense_format == SSD_TYPE_DESC))
9283                         ctl_sense_to_desc((struct scsi_sense_data_fixed *)
9284                             &lun->pending_sense[initidx],
9285                             (struct scsi_sense_data_desc *)sense_ptr);
9286                 else if ((stored_format == SSD_TYPE_DESC)
9287                       && (sense_format == SSD_TYPE_FIXED))
9288                         ctl_sense_to_fixed((struct scsi_sense_data_desc *)
9289                             &lun->pending_sense[initidx],
9290                             (struct scsi_sense_data_fixed *)sense_ptr);
9291                 else
9292                         memcpy(sense_ptr, &lun->pending_sense[initidx],
9293                                MIN(sizeof(*sense_ptr),
9294                                sizeof(lun->pending_sense[initidx])));
9295
9296                 ctl_clear_mask(lun->have_ca, initidx);
9297                 have_error = 1;
9298         } else
9299 #endif
9300         if (have_error == 0) {
9301                 ua_type = ctl_build_ua(lun, initidx, sense_ptr, sense_format);
9302                 if (ua_type != CTL_UA_NONE)
9303                         have_error = 1;
9304                 if (ua_type == CTL_UA_LUN_CHANGE) {
9305                         mtx_unlock(&lun->lun_lock);
9306                         mtx_lock(&softc->ctl_lock);
9307                         ctl_clr_ua_allluns(softc, initidx, ua_type);
9308                         mtx_unlock(&softc->ctl_lock);
9309                         mtx_lock(&lun->lun_lock);
9310                 }
9311         }
9312         if (have_error == 0) {
9313                 /*
9314                  * Report informational exception if have one and allowed.
9315                  */
9316                 if (lun->MODE_IE.mrie != SIEP_MRIE_NO) {
9317                         asc = lun->ie_asc;
9318                         ascq = lun->ie_ascq;
9319                 }
9320                 ctl_set_sense_data(sense_ptr, lun, sense_format,
9321                     /*current_error*/ 1,
9322                     /*sense_key*/ SSD_KEY_NO_SENSE,
9323                     /*asc*/ asc,
9324                     /*ascq*/ ascq,
9325                     SSD_ELEM_NONE);
9326         }
9327         mtx_unlock(&lun->lun_lock);
9328
9329 send:
9330         /*
9331          * We report the SCSI status as OK, since the status of the command
9332          * itself is OK.  We're reporting sense as parameter data.
9333          */
9334         ctl_set_success(ctsio);
9335         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9336         ctsio->be_move_done = ctl_config_move_done;
9337         ctl_datamove((union ctl_io *)ctsio);
9338         return (CTL_RETVAL_COMPLETE);
9339 }
9340
9341 int
9342 ctl_tur(struct ctl_scsiio *ctsio)
9343 {
9344
9345         CTL_DEBUG_PRINT(("ctl_tur\n"));
9346
9347         ctl_set_success(ctsio);
9348         ctl_done((union ctl_io *)ctsio);
9349
9350         return (CTL_RETVAL_COMPLETE);
9351 }
9352
9353 /*
9354  * SCSI VPD page 0x00, the Supported VPD Pages page.
9355  */
9356 static int
9357 ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
9358 {
9359         struct scsi_vpd_supported_pages *pages;
9360         int sup_page_size;
9361         struct ctl_lun *lun;
9362         int p;
9363
9364         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9365
9366         sup_page_size = sizeof(struct scsi_vpd_supported_pages) *
9367             SCSI_EVPD_NUM_SUPPORTED_PAGES;
9368         ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO);
9369         pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
9370         ctsio->kern_sg_entries = 0;
9371
9372         if (sup_page_size < alloc_len) {
9373                 ctsio->residual = alloc_len - sup_page_size;
9374                 ctsio->kern_data_len = sup_page_size;
9375                 ctsio->kern_total_len = sup_page_size;
9376         } else {
9377                 ctsio->residual = 0;
9378                 ctsio->kern_data_len = alloc_len;
9379                 ctsio->kern_total_len = alloc_len;
9380         }
9381         ctsio->kern_data_resid = 0;
9382         ctsio->kern_rel_offset = 0;
9383         ctsio->kern_sg_entries = 0;
9384
9385         /*
9386          * The control device is always connected.  The disk device, on the
9387          * other hand, may not be online all the time.  Need to change this
9388          * to figure out whether the disk device is actually online or not.
9389          */
9390         if (lun != NULL)
9391                 pages->device = (SID_QUAL_LU_CONNECTED << 5) |
9392                                 lun->be_lun->lun_type;
9393         else
9394                 pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9395
9396         p = 0;
9397         /* Supported VPD pages */
9398         pages->page_list[p++] = SVPD_SUPPORTED_PAGES;
9399         /* Serial Number */
9400         pages->page_list[p++] = SVPD_UNIT_SERIAL_NUMBER;
9401         /* Device Identification */
9402         pages->page_list[p++] = SVPD_DEVICE_ID;
9403         /* Extended INQUIRY Data */
9404         pages->page_list[p++] = SVPD_EXTENDED_INQUIRY_DATA;
9405         /* Mode Page Policy */
9406         pages->page_list[p++] = SVPD_MODE_PAGE_POLICY;
9407         /* SCSI Ports */
9408         pages->page_list[p++] = SVPD_SCSI_PORTS;
9409         /* Third-party Copy */
9410         pages->page_list[p++] = SVPD_SCSI_TPC;
9411         if (lun != NULL && lun->be_lun->lun_type == T_DIRECT) {
9412                 /* Block limits */
9413                 pages->page_list[p++] = SVPD_BLOCK_LIMITS;
9414                 /* Block Device Characteristics */
9415                 pages->page_list[p++] = SVPD_BDC;
9416                 /* Logical Block Provisioning */
9417                 pages->page_list[p++] = SVPD_LBP;
9418         }
9419         pages->length = p;
9420
9421         ctl_set_success(ctsio);
9422         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9423         ctsio->be_move_done = ctl_config_move_done;
9424         ctl_datamove((union ctl_io *)ctsio);
9425         return (CTL_RETVAL_COMPLETE);
9426 }
9427
9428 /*
9429  * SCSI VPD page 0x80, the Unit Serial Number page.
9430  */
9431 static int
9432 ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
9433 {
9434         struct scsi_vpd_unit_serial_number *sn_ptr;
9435         struct ctl_lun *lun;
9436         int data_len;
9437
9438         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9439
9440         data_len = 4 + CTL_SN_LEN;
9441         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9442         sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
9443         if (data_len < alloc_len) {
9444                 ctsio->residual = alloc_len - data_len;
9445                 ctsio->kern_data_len = data_len;
9446                 ctsio->kern_total_len = data_len;
9447         } else {
9448                 ctsio->residual = 0;
9449                 ctsio->kern_data_len = alloc_len;
9450                 ctsio->kern_total_len = alloc_len;
9451         }
9452         ctsio->kern_data_resid = 0;
9453         ctsio->kern_rel_offset = 0;
9454         ctsio->kern_sg_entries = 0;
9455
9456         /*
9457          * The control device is always connected.  The disk device, on the
9458          * other hand, may not be online all the time.  Need to change this
9459          * to figure out whether the disk device is actually online or not.
9460          */
9461         if (lun != NULL)
9462                 sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9463                                   lun->be_lun->lun_type;
9464         else
9465                 sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9466
9467         sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9468         sn_ptr->length = CTL_SN_LEN;
9469         /*
9470          * If we don't have a LUN, we just leave the serial number as
9471          * all spaces.
9472          */
9473         if (lun != NULL) {
9474                 strncpy((char *)sn_ptr->serial_num,
9475                         (char *)lun->be_lun->serial_num, CTL_SN_LEN);
9476         } else
9477                 memset(sn_ptr->serial_num, 0x20, CTL_SN_LEN);
9478
9479         ctl_set_success(ctsio);
9480         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9481         ctsio->be_move_done = ctl_config_move_done;
9482         ctl_datamove((union ctl_io *)ctsio);
9483         return (CTL_RETVAL_COMPLETE);
9484 }
9485
9486
9487 /*
9488  * SCSI VPD page 0x86, the Extended INQUIRY Data page.
9489  */
9490 static int
9491 ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len)
9492 {
9493         struct scsi_vpd_extended_inquiry_data *eid_ptr;
9494         struct ctl_lun *lun;
9495         int data_len;
9496
9497         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9498
9499         data_len = sizeof(struct scsi_vpd_extended_inquiry_data);
9500         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9501         eid_ptr = (struct scsi_vpd_extended_inquiry_data *)ctsio->kern_data_ptr;
9502         ctsio->kern_sg_entries = 0;
9503
9504         if (data_len < alloc_len) {
9505                 ctsio->residual = alloc_len - data_len;
9506                 ctsio->kern_data_len = data_len;
9507                 ctsio->kern_total_len = data_len;
9508         } else {
9509                 ctsio->residual = 0;
9510                 ctsio->kern_data_len = alloc_len;
9511                 ctsio->kern_total_len = alloc_len;
9512         }
9513         ctsio->kern_data_resid = 0;
9514         ctsio->kern_rel_offset = 0;
9515         ctsio->kern_sg_entries = 0;
9516
9517         /*
9518          * The control device is always connected.  The disk device, on the
9519          * other hand, may not be online all the time.
9520          */
9521         if (lun != NULL)
9522                 eid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9523                                      lun->be_lun->lun_type;
9524         else
9525                 eid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9526         eid_ptr->page_code = SVPD_EXTENDED_INQUIRY_DATA;
9527         scsi_ulto2b(data_len - 4, eid_ptr->page_length);
9528         /*
9529          * We support head of queue, ordered and simple tags.
9530          */
9531         eid_ptr->flags2 = SVPD_EID_HEADSUP | SVPD_EID_ORDSUP | SVPD_EID_SIMPSUP;
9532         /*
9533          * Volatile cache supported.
9534          */
9535         eid_ptr->flags3 = SVPD_EID_V_SUP;
9536
9537         /*
9538          * This means that we clear the REPORTED LUNS DATA HAS CHANGED unit
9539          * attention for a particular IT nexus on all LUNs once we report
9540          * it to that nexus once.  This bit is required as of SPC-4.
9541          */
9542         eid_ptr->flags4 = SVPD_EID_LUICLR;
9543
9544         /*
9545          * XXX KDM in order to correctly answer this, we would need
9546          * information from the SIM to determine how much sense data it
9547          * can send.  So this would really be a path inquiry field, most
9548          * likely.  This can be set to a maximum of 252 according to SPC-4,
9549          * but the hardware may or may not be able to support that much.
9550          * 0 just means that the maximum sense data length is not reported.
9551          */
9552         eid_ptr->max_sense_length = 0;
9553
9554         ctl_set_success(ctsio);
9555         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9556         ctsio->be_move_done = ctl_config_move_done;
9557         ctl_datamove((union ctl_io *)ctsio);
9558         return (CTL_RETVAL_COMPLETE);
9559 }
9560
9561 static int
9562 ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len)
9563 {
9564         struct scsi_vpd_mode_page_policy *mpp_ptr;
9565         struct ctl_lun *lun;
9566         int data_len;
9567
9568         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9569
9570         data_len = sizeof(struct scsi_vpd_mode_page_policy) +
9571             sizeof(struct scsi_vpd_mode_page_policy_descr);
9572
9573         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9574         mpp_ptr = (struct scsi_vpd_mode_page_policy *)ctsio->kern_data_ptr;
9575         ctsio->kern_sg_entries = 0;
9576
9577         if (data_len < alloc_len) {
9578                 ctsio->residual = alloc_len - data_len;
9579                 ctsio->kern_data_len = data_len;
9580                 ctsio->kern_total_len = data_len;
9581         } else {
9582                 ctsio->residual = 0;
9583                 ctsio->kern_data_len = alloc_len;
9584                 ctsio->kern_total_len = alloc_len;
9585         }
9586         ctsio->kern_data_resid = 0;
9587         ctsio->kern_rel_offset = 0;
9588         ctsio->kern_sg_entries = 0;
9589
9590         /*
9591          * The control device is always connected.  The disk device, on the
9592          * other hand, may not be online all the time.
9593          */
9594         if (lun != NULL)
9595                 mpp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9596                                      lun->be_lun->lun_type;
9597         else
9598                 mpp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9599         mpp_ptr->page_code = SVPD_MODE_PAGE_POLICY;
9600         scsi_ulto2b(data_len - 4, mpp_ptr->page_length);
9601         mpp_ptr->descr[0].page_code = 0x3f;
9602         mpp_ptr->descr[0].subpage_code = 0xff;
9603         mpp_ptr->descr[0].policy = SVPD_MPP_SHARED;
9604
9605         ctl_set_success(ctsio);
9606         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9607         ctsio->be_move_done = ctl_config_move_done;
9608         ctl_datamove((union ctl_io *)ctsio);
9609         return (CTL_RETVAL_COMPLETE);
9610 }
9611
9612 /*
9613  * SCSI VPD page 0x83, the Device Identification page.
9614  */
9615 static int
9616 ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
9617 {
9618         struct scsi_vpd_device_id *devid_ptr;
9619         struct scsi_vpd_id_descriptor *desc;
9620         struct ctl_softc *softc;
9621         struct ctl_lun *lun;
9622         struct ctl_port *port;
9623         int data_len, g;
9624         uint8_t proto;
9625
9626         softc = control_softc;
9627
9628         port = ctl_io_port(&ctsio->io_hdr);
9629         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9630
9631         data_len = sizeof(struct scsi_vpd_device_id) +
9632             sizeof(struct scsi_vpd_id_descriptor) +
9633                 sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
9634             sizeof(struct scsi_vpd_id_descriptor) +
9635                 sizeof(struct scsi_vpd_id_trgt_port_grp_id);
9636         if (lun && lun->lun_devid)
9637                 data_len += lun->lun_devid->len;
9638         if (port && port->port_devid)
9639                 data_len += port->port_devid->len;
9640         if (port && port->target_devid)
9641                 data_len += port->target_devid->len;
9642
9643         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9644         devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
9645         ctsio->kern_sg_entries = 0;
9646
9647         if (data_len < alloc_len) {
9648                 ctsio->residual = alloc_len - data_len;
9649                 ctsio->kern_data_len = data_len;
9650                 ctsio->kern_total_len = data_len;
9651         } else {
9652                 ctsio->residual = 0;
9653                 ctsio->kern_data_len = alloc_len;
9654                 ctsio->kern_total_len = alloc_len;
9655         }
9656         ctsio->kern_data_resid = 0;
9657         ctsio->kern_rel_offset = 0;
9658         ctsio->kern_sg_entries = 0;
9659
9660         /*
9661          * The control device is always connected.  The disk device, on the
9662          * other hand, may not be online all the time.
9663          */
9664         if (lun != NULL)
9665                 devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9666                                      lun->be_lun->lun_type;
9667         else
9668                 devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9669         devid_ptr->page_code = SVPD_DEVICE_ID;
9670         scsi_ulto2b(data_len - 4, devid_ptr->length);
9671
9672         if (port && port->port_type == CTL_PORT_FC)
9673                 proto = SCSI_PROTO_FC << 4;
9674         else if (port && port->port_type == CTL_PORT_ISCSI)
9675                 proto = SCSI_PROTO_ISCSI << 4;
9676         else
9677                 proto = SCSI_PROTO_SPI << 4;
9678         desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
9679
9680         /*
9681          * We're using a LUN association here.  i.e., this device ID is a
9682          * per-LUN identifier.
9683          */
9684         if (lun && lun->lun_devid) {
9685                 memcpy(desc, lun->lun_devid->data, lun->lun_devid->len);
9686                 desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9687                     lun->lun_devid->len);
9688         }
9689
9690         /*
9691          * This is for the WWPN which is a port association.
9692          */
9693         if (port && port->port_devid) {
9694                 memcpy(desc, port->port_devid->data, port->port_devid->len);
9695                 desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9696                     port->port_devid->len);
9697         }
9698
9699         /*
9700          * This is for the Relative Target Port(type 4h) identifier
9701          */
9702         desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9703         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9704             SVPD_ID_TYPE_RELTARG;
9705         desc->length = 4;
9706         scsi_ulto2b(ctsio->io_hdr.nexus.targ_port, &desc->identifier[2]);
9707         desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9708             sizeof(struct scsi_vpd_id_rel_trgt_port_id));
9709
9710         /*
9711          * This is for the Target Port Group(type 5h) 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_TPORTGRP;
9716         desc->length = 4;
9717         if (softc->is_single ||
9718             (port && port->status & CTL_PORT_STATUS_HA_SHARED))
9719                 g = 1;
9720         else
9721                 g = 2 + ctsio->io_hdr.nexus.targ_port / softc->port_cnt;
9722         scsi_ulto2b(g, &desc->identifier[2]);
9723         desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9724             sizeof(struct scsi_vpd_id_trgt_port_grp_id));
9725
9726         /*
9727          * This is for the Target identifier
9728          */
9729         if (port && port->target_devid) {
9730                 memcpy(desc, port->target_devid->data, port->target_devid->len);
9731         }
9732
9733         ctl_set_success(ctsio);
9734         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9735         ctsio->be_move_done = ctl_config_move_done;
9736         ctl_datamove((union ctl_io *)ctsio);
9737         return (CTL_RETVAL_COMPLETE);
9738 }
9739
9740 static int
9741 ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio, int alloc_len)
9742 {
9743         struct ctl_softc *softc = control_softc;
9744         struct scsi_vpd_scsi_ports *sp;
9745         struct scsi_vpd_port_designation *pd;
9746         struct scsi_vpd_port_designation_cont *pdc;
9747         struct ctl_lun *lun;
9748         struct ctl_port *port;
9749         int data_len, num_target_ports, iid_len, id_len;
9750
9751         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9752
9753         num_target_ports = 0;
9754         iid_len = 0;
9755         id_len = 0;
9756         mtx_lock(&softc->ctl_lock);
9757         STAILQ_FOREACH(port, &softc->port_list, links) {
9758                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
9759                         continue;
9760                 if (lun != NULL &&
9761                     ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
9762                         continue;
9763                 num_target_ports++;
9764                 if (port->init_devid)
9765                         iid_len += port->init_devid->len;
9766                 if (port->port_devid)
9767                         id_len += port->port_devid->len;
9768         }
9769         mtx_unlock(&softc->ctl_lock);
9770
9771         data_len = sizeof(struct scsi_vpd_scsi_ports) +
9772             num_target_ports * (sizeof(struct scsi_vpd_port_designation) +
9773              sizeof(struct scsi_vpd_port_designation_cont)) + iid_len + id_len;
9774         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9775         sp = (struct scsi_vpd_scsi_ports *)ctsio->kern_data_ptr;
9776         ctsio->kern_sg_entries = 0;
9777
9778         if (data_len < alloc_len) {
9779                 ctsio->residual = alloc_len - data_len;
9780                 ctsio->kern_data_len = data_len;
9781                 ctsio->kern_total_len = data_len;
9782         } else {
9783                 ctsio->residual = 0;
9784                 ctsio->kern_data_len = alloc_len;
9785                 ctsio->kern_total_len = alloc_len;
9786         }
9787         ctsio->kern_data_resid = 0;
9788         ctsio->kern_rel_offset = 0;
9789         ctsio->kern_sg_entries = 0;
9790
9791         /*
9792          * The control device is always connected.  The disk device, on the
9793          * other hand, may not be online all the time.  Need to change this
9794          * to figure out whether the disk device is actually online or not.
9795          */
9796         if (lun != NULL)
9797                 sp->device = (SID_QUAL_LU_CONNECTED << 5) |
9798                                   lun->be_lun->lun_type;
9799         else
9800                 sp->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9801
9802         sp->page_code = SVPD_SCSI_PORTS;
9803         scsi_ulto2b(data_len - sizeof(struct scsi_vpd_scsi_ports),
9804             sp->page_length);
9805         pd = &sp->design[0];
9806
9807         mtx_lock(&softc->ctl_lock);
9808         STAILQ_FOREACH(port, &softc->port_list, links) {
9809                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
9810                         continue;
9811                 if (lun != NULL &&
9812                     ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
9813                         continue;
9814                 scsi_ulto2b(port->targ_port, pd->relative_port_id);
9815                 if (port->init_devid) {
9816                         iid_len = port->init_devid->len;
9817                         memcpy(pd->initiator_transportid,
9818                             port->init_devid->data, port->init_devid->len);
9819                 } else
9820                         iid_len = 0;
9821                 scsi_ulto2b(iid_len, pd->initiator_transportid_length);
9822                 pdc = (struct scsi_vpd_port_designation_cont *)
9823                     (&pd->initiator_transportid[iid_len]);
9824                 if (port->port_devid) {
9825                         id_len = port->port_devid->len;
9826                         memcpy(pdc->target_port_descriptors,
9827                             port->port_devid->data, port->port_devid->len);
9828                 } else
9829                         id_len = 0;
9830                 scsi_ulto2b(id_len, pdc->target_port_descriptors_length);
9831                 pd = (struct scsi_vpd_port_designation *)
9832                     ((uint8_t *)pdc->target_port_descriptors + id_len);
9833         }
9834         mtx_unlock(&softc->ctl_lock);
9835
9836         ctl_set_success(ctsio);
9837         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9838         ctsio->be_move_done = ctl_config_move_done;
9839         ctl_datamove((union ctl_io *)ctsio);
9840         return (CTL_RETVAL_COMPLETE);
9841 }
9842
9843 static int
9844 ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio, int alloc_len)
9845 {
9846         struct scsi_vpd_block_limits *bl_ptr;
9847         struct ctl_lun *lun;
9848         uint64_t ival;
9849
9850         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9851
9852         ctsio->kern_data_ptr = malloc(sizeof(*bl_ptr), M_CTL, M_WAITOK | M_ZERO);
9853         bl_ptr = (struct scsi_vpd_block_limits *)ctsio->kern_data_ptr;
9854         ctsio->kern_sg_entries = 0;
9855
9856         if (sizeof(*bl_ptr) < alloc_len) {
9857                 ctsio->residual = alloc_len - sizeof(*bl_ptr);
9858                 ctsio->kern_data_len = sizeof(*bl_ptr);
9859                 ctsio->kern_total_len = sizeof(*bl_ptr);
9860         } else {
9861                 ctsio->residual = 0;
9862                 ctsio->kern_data_len = alloc_len;
9863                 ctsio->kern_total_len = alloc_len;
9864         }
9865         ctsio->kern_data_resid = 0;
9866         ctsio->kern_rel_offset = 0;
9867         ctsio->kern_sg_entries = 0;
9868
9869         /*
9870          * The control device is always connected.  The disk device, on the
9871          * other hand, may not be online all the time.  Need to change this
9872          * to figure out whether the disk device is actually online or not.
9873          */
9874         if (lun != NULL)
9875                 bl_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9876                                   lun->be_lun->lun_type;
9877         else
9878                 bl_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9879
9880         bl_ptr->page_code = SVPD_BLOCK_LIMITS;
9881         scsi_ulto2b(sizeof(*bl_ptr) - 4, bl_ptr->page_length);
9882         bl_ptr->max_cmp_write_len = 0xff;
9883         scsi_ulto4b(0xffffffff, bl_ptr->max_txfer_len);
9884         if (lun != NULL) {
9885                 scsi_ulto4b(lun->be_lun->opttxferlen, bl_ptr->opt_txfer_len);
9886                 if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
9887                         ival = 0xffffffff;
9888                         ctl_get_opt_number(&lun->be_lun->options,
9889                             "unmap_max_lba", &ival);
9890                         scsi_ulto4b(ival, bl_ptr->max_unmap_lba_cnt);
9891                         ival = 0xffffffff;
9892                         ctl_get_opt_number(&lun->be_lun->options,
9893                             "unmap_max_descr", &ival);
9894                         scsi_ulto4b(ival, bl_ptr->max_unmap_blk_cnt);
9895                         if (lun->be_lun->ublockexp != 0) {
9896                                 scsi_ulto4b((1 << lun->be_lun->ublockexp),
9897                                     bl_ptr->opt_unmap_grain);
9898                                 scsi_ulto4b(0x80000000 | lun->be_lun->ublockoff,
9899                                     bl_ptr->unmap_grain_align);
9900                         }
9901                 }
9902                 scsi_ulto4b(lun->be_lun->atomicblock,
9903                     bl_ptr->max_atomic_transfer_length);
9904                 scsi_ulto4b(0, bl_ptr->atomic_alignment);
9905                 scsi_ulto4b(0, bl_ptr->atomic_transfer_length_granularity);
9906                 scsi_ulto4b(0, bl_ptr->max_atomic_transfer_length_with_atomic_boundary);
9907                 scsi_ulto4b(0, bl_ptr->max_atomic_boundary_size);
9908                 ival = UINT64_MAX;
9909                 ctl_get_opt_number(&lun->be_lun->options, "write_same_max_lba", &ival);
9910                 scsi_u64to8b(ival, bl_ptr->max_write_same_length);
9911         }
9912
9913         ctl_set_success(ctsio);
9914         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9915         ctsio->be_move_done = ctl_config_move_done;
9916         ctl_datamove((union ctl_io *)ctsio);
9917         return (CTL_RETVAL_COMPLETE);
9918 }
9919
9920 static int
9921 ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len)
9922 {
9923         struct scsi_vpd_block_device_characteristics *bdc_ptr;
9924         struct ctl_lun *lun;
9925         const char *value;
9926         u_int i;
9927
9928         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9929
9930         ctsio->kern_data_ptr = malloc(sizeof(*bdc_ptr), M_CTL, M_WAITOK | M_ZERO);
9931         bdc_ptr = (struct scsi_vpd_block_device_characteristics *)ctsio->kern_data_ptr;
9932         ctsio->kern_sg_entries = 0;
9933
9934         if (sizeof(*bdc_ptr) < alloc_len) {
9935                 ctsio->residual = alloc_len - sizeof(*bdc_ptr);
9936                 ctsio->kern_data_len = sizeof(*bdc_ptr);
9937                 ctsio->kern_total_len = sizeof(*bdc_ptr);
9938         } else {
9939                 ctsio->residual = 0;
9940                 ctsio->kern_data_len = alloc_len;
9941                 ctsio->kern_total_len = alloc_len;
9942         }
9943         ctsio->kern_data_resid = 0;
9944         ctsio->kern_rel_offset = 0;
9945         ctsio->kern_sg_entries = 0;
9946
9947         /*
9948          * The control device is always connected.  The disk device, on the
9949          * other hand, may not be online all the time.  Need to change this
9950          * to figure out whether the disk device is actually online or not.
9951          */
9952         if (lun != NULL)
9953                 bdc_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9954                                   lun->be_lun->lun_type;
9955         else
9956                 bdc_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9957         bdc_ptr->page_code = SVPD_BDC;
9958         scsi_ulto2b(sizeof(*bdc_ptr) - 4, bdc_ptr->page_length);
9959         if (lun != NULL &&
9960             (value = ctl_get_opt(&lun->be_lun->options, "rpm")) != NULL)
9961                 i = strtol(value, NULL, 0);
9962         else
9963                 i = CTL_DEFAULT_ROTATION_RATE;
9964         scsi_ulto2b(i, bdc_ptr->medium_rotation_rate);
9965         if (lun != NULL &&
9966             (value = ctl_get_opt(&lun->be_lun->options, "formfactor")) != NULL)
9967                 i = strtol(value, NULL, 0);
9968         else
9969                 i = 0;
9970         bdc_ptr->wab_wac_ff = (i & 0x0f);
9971         bdc_ptr->flags = SVPD_FUAB | SVPD_VBULS;
9972
9973         ctl_set_success(ctsio);
9974         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9975         ctsio->be_move_done = ctl_config_move_done;
9976         ctl_datamove((union ctl_io *)ctsio);
9977         return (CTL_RETVAL_COMPLETE);
9978 }
9979
9980 static int
9981 ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len)
9982 {
9983         struct scsi_vpd_logical_block_prov *lbp_ptr;
9984         struct ctl_lun *lun;
9985         const char *value;
9986
9987         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9988
9989         ctsio->kern_data_ptr = malloc(sizeof(*lbp_ptr), M_CTL, M_WAITOK | M_ZERO);
9990         lbp_ptr = (struct scsi_vpd_logical_block_prov *)ctsio->kern_data_ptr;
9991         ctsio->kern_sg_entries = 0;
9992
9993         if (sizeof(*lbp_ptr) < alloc_len) {
9994                 ctsio->residual = alloc_len - sizeof(*lbp_ptr);
9995                 ctsio->kern_data_len = sizeof(*lbp_ptr);
9996                 ctsio->kern_total_len = sizeof(*lbp_ptr);
9997         } else {
9998                 ctsio->residual = 0;
9999                 ctsio->kern_data_len = alloc_len;
10000                 ctsio->kern_total_len = alloc_len;
10001         }
10002         ctsio->kern_data_resid = 0;
10003         ctsio->kern_rel_offset = 0;
10004         ctsio->kern_sg_entries = 0;
10005
10006         /*
10007          * The control device is always connected.  The disk device, on the
10008          * other hand, may not be online all the time.  Need to change this
10009          * to figure out whether the disk device is actually online or not.
10010          */
10011         if (lun != NULL)
10012                 lbp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10013                                   lun->be_lun->lun_type;
10014         else
10015                 lbp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10016
10017         lbp_ptr->page_code = SVPD_LBP;
10018         scsi_ulto2b(sizeof(*lbp_ptr) - 4, lbp_ptr->page_length);
10019         lbp_ptr->threshold_exponent = CTL_LBP_EXPONENT;
10020         if (lun != NULL && lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
10021                 lbp_ptr->flags = SVPD_LBP_UNMAP | SVPD_LBP_WS16 |
10022                     SVPD_LBP_WS10 | SVPD_LBP_RZ | SVPD_LBP_ANC_SUP;
10023                 value = ctl_get_opt(&lun->be_lun->options, "provisioning_type");
10024                 if (value != NULL) {
10025                         if (strcmp(value, "resource") == 0)
10026                                 lbp_ptr->prov_type = SVPD_LBP_RESOURCE;
10027                         else if (strcmp(value, "thin") == 0)
10028                                 lbp_ptr->prov_type = SVPD_LBP_THIN;
10029                 } else
10030                         lbp_ptr->prov_type = SVPD_LBP_THIN;
10031         }
10032
10033         ctl_set_success(ctsio);
10034         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10035         ctsio->be_move_done = ctl_config_move_done;
10036         ctl_datamove((union ctl_io *)ctsio);
10037         return (CTL_RETVAL_COMPLETE);
10038 }
10039
10040 /*
10041  * INQUIRY with the EVPD bit set.
10042  */
10043 static int
10044 ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
10045 {
10046         struct ctl_lun *lun;
10047         struct scsi_inquiry *cdb;
10048         int alloc_len, retval;
10049
10050         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10051         cdb = (struct scsi_inquiry *)ctsio->cdb;
10052         alloc_len = scsi_2btoul(cdb->length);
10053
10054         switch (cdb->page_code) {
10055         case SVPD_SUPPORTED_PAGES:
10056                 retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
10057                 break;
10058         case SVPD_UNIT_SERIAL_NUMBER:
10059                 retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
10060                 break;
10061         case SVPD_DEVICE_ID:
10062                 retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
10063                 break;
10064         case SVPD_EXTENDED_INQUIRY_DATA:
10065                 retval = ctl_inquiry_evpd_eid(ctsio, alloc_len);
10066                 break;
10067         case SVPD_MODE_PAGE_POLICY:
10068                 retval = ctl_inquiry_evpd_mpp(ctsio, alloc_len);
10069                 break;
10070         case SVPD_SCSI_PORTS:
10071                 retval = ctl_inquiry_evpd_scsi_ports(ctsio, alloc_len);
10072                 break;
10073         case SVPD_SCSI_TPC:
10074                 retval = ctl_inquiry_evpd_tpc(ctsio, alloc_len);
10075                 break;
10076         case SVPD_BLOCK_LIMITS:
10077                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
10078                         goto err;
10079                 retval = ctl_inquiry_evpd_block_limits(ctsio, alloc_len);
10080                 break;
10081         case SVPD_BDC:
10082                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
10083                         goto err;
10084                 retval = ctl_inquiry_evpd_bdc(ctsio, alloc_len);
10085                 break;
10086         case SVPD_LBP:
10087                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
10088                         goto err;
10089                 retval = ctl_inquiry_evpd_lbp(ctsio, alloc_len);
10090                 break;
10091         default:
10092 err:
10093                 ctl_set_invalid_field(ctsio,
10094                                       /*sks_valid*/ 1,
10095                                       /*command*/ 1,
10096                                       /*field*/ 2,
10097                                       /*bit_valid*/ 0,
10098                                       /*bit*/ 0);
10099                 ctl_done((union ctl_io *)ctsio);
10100                 retval = CTL_RETVAL_COMPLETE;
10101                 break;
10102         }
10103
10104         return (retval);
10105 }
10106
10107 /*
10108  * Standard INQUIRY data.
10109  */
10110 static int
10111 ctl_inquiry_std(struct ctl_scsiio *ctsio)
10112 {
10113         struct scsi_inquiry_data *inq_ptr;
10114         struct scsi_inquiry *cdb;
10115         struct ctl_softc *softc = control_softc;
10116         struct ctl_port *port;
10117         struct ctl_lun *lun;
10118         char *val;
10119         uint32_t alloc_len, data_len;
10120         ctl_port_type port_type;
10121
10122         port = ctl_io_port(&ctsio->io_hdr);
10123         port_type = port->port_type;
10124         if (port_type == CTL_PORT_IOCTL || port_type == CTL_PORT_INTERNAL)
10125                 port_type = CTL_PORT_SCSI;
10126
10127         lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10128         cdb = (struct scsi_inquiry *)ctsio->cdb;
10129         alloc_len = scsi_2btoul(cdb->length);
10130
10131         /*
10132          * We malloc the full inquiry data size here and fill it
10133          * in.  If the user only asks for less, we'll give him
10134          * that much.
10135          */
10136         data_len = offsetof(struct scsi_inquiry_data, vendor_specific1);
10137         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10138         inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
10139         ctsio->kern_sg_entries = 0;
10140         ctsio->kern_data_resid = 0;
10141         ctsio->kern_rel_offset = 0;
10142
10143         if (data_len < alloc_len) {
10144                 ctsio->residual = alloc_len - data_len;
10145                 ctsio->kern_data_len = data_len;
10146                 ctsio->kern_total_len = data_len;
10147         } else {
10148                 ctsio->residual = 0;
10149                 ctsio->kern_data_len = alloc_len;
10150                 ctsio->kern_total_len = alloc_len;
10151         }
10152
10153         if (lun != NULL) {
10154                 if ((lun->flags & CTL_LUN_PRIMARY_SC) ||
10155                     softc->ha_link >= CTL_HA_LINK_UNKNOWN) {
10156                         inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10157                             lun->be_lun->lun_type;
10158                 } else {
10159                         inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) |
10160                             lun->be_lun->lun_type;
10161                 }
10162                 if (lun->flags & CTL_LUN_REMOVABLE)
10163                         inq_ptr->dev_qual2 |= SID_RMB;
10164         } else
10165                 inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE;
10166
10167         /* RMB in byte 2 is 0 */
10168         inq_ptr->version = SCSI_REV_SPC5;
10169
10170         /*
10171          * According to SAM-3, even if a device only supports a single
10172          * level of LUN addressing, it should still set the HISUP bit:
10173          *
10174          * 4.9.1 Logical unit numbers overview
10175          *
10176          * All logical unit number formats described in this standard are
10177          * hierarchical in structure even when only a single level in that
10178          * hierarchy is used. The HISUP bit shall be set to one in the
10179          * standard INQUIRY data (see SPC-2) when any logical unit number
10180          * format described in this standard is used.  Non-hierarchical
10181          * formats are outside the scope of this standard.
10182          *
10183          * Therefore we set the HiSup bit here.
10184          *
10185          * The response format is 2, per SPC-3.
10186          */
10187         inq_ptr->response_format = SID_HiSup | 2;
10188
10189         inq_ptr->additional_length = data_len -
10190             (offsetof(struct scsi_inquiry_data, additional_length) + 1);
10191         CTL_DEBUG_PRINT(("additional_length = %d\n",
10192                          inq_ptr->additional_length));
10193
10194         inq_ptr->spc3_flags = SPC3_SID_3PC | SPC3_SID_TPGS_IMPLICIT;
10195         if (port_type == CTL_PORT_SCSI)
10196                 inq_ptr->spc2_flags = SPC2_SID_ADDR16;
10197         inq_ptr->spc2_flags |= SPC2_SID_MultiP;
10198         inq_ptr->flags = SID_CmdQue;
10199         if (port_type == CTL_PORT_SCSI)
10200                 inq_ptr->flags |= SID_WBus16 | SID_Sync;
10201
10202         /*
10203          * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
10204          * We have 8 bytes for the vendor name, and 16 bytes for the device
10205          * name and 4 bytes for the revision.
10206          */
10207         if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10208             "vendor")) == NULL) {
10209                 strncpy(inq_ptr->vendor, CTL_VENDOR, sizeof(inq_ptr->vendor));
10210         } else {
10211                 memset(inq_ptr->vendor, ' ', sizeof(inq_ptr->vendor));
10212                 strncpy(inq_ptr->vendor, val,
10213                     min(sizeof(inq_ptr->vendor), strlen(val)));
10214         }
10215         if (lun == NULL) {
10216                 strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10217                     sizeof(inq_ptr->product));
10218         } else if ((val = ctl_get_opt(&lun->be_lun->options, "product")) == NULL) {
10219                 switch (lun->be_lun->lun_type) {
10220                 case T_DIRECT:
10221                         strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10222                             sizeof(inq_ptr->product));
10223                         break;
10224                 case T_PROCESSOR:
10225                         strncpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT,
10226                             sizeof(inq_ptr->product));
10227                         break;
10228                 case T_CDROM:
10229                         strncpy(inq_ptr->product, CTL_CDROM_PRODUCT,
10230                             sizeof(inq_ptr->product));
10231                         break;
10232                 default:
10233                         strncpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT,
10234                             sizeof(inq_ptr->product));
10235                         break;
10236                 }
10237         } else {
10238                 memset(inq_ptr->product, ' ', sizeof(inq_ptr->product));
10239                 strncpy(inq_ptr->product, val,
10240                     min(sizeof(inq_ptr->product), strlen(val)));
10241         }
10242
10243         /*
10244          * XXX make this a macro somewhere so it automatically gets
10245          * incremented when we make changes.
10246          */
10247         if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10248             "revision")) == NULL) {
10249                 strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
10250         } else {
10251                 memset(inq_ptr->revision, ' ', sizeof(inq_ptr->revision));
10252                 strncpy(inq_ptr->revision, val,
10253                     min(sizeof(inq_ptr->revision), strlen(val)));
10254         }
10255
10256         /*
10257          * For parallel SCSI, we support double transition and single
10258          * transition clocking.  We also support QAS (Quick Arbitration
10259          * and Selection) and Information Unit transfers on both the
10260          * control and array devices.
10261          */
10262         if (port_type == CTL_PORT_SCSI)
10263                 inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
10264                                     SID_SPI_IUS;
10265
10266         /* SAM-6 (no version claimed) */
10267         scsi_ulto2b(0x00C0, inq_ptr->version1);
10268         /* SPC-5 (no version claimed) */
10269         scsi_ulto2b(0x05C0, inq_ptr->version2);
10270         if (port_type == CTL_PORT_FC) {
10271                 /* FCP-2 ANSI INCITS.350:2003 */
10272                 scsi_ulto2b(0x0917, inq_ptr->version3);
10273         } else if (port_type == CTL_PORT_SCSI) {
10274                 /* SPI-4 ANSI INCITS.362:200x */
10275                 scsi_ulto2b(0x0B56, inq_ptr->version3);
10276         } else if (port_type == CTL_PORT_ISCSI) {
10277                 /* iSCSI (no version claimed) */
10278                 scsi_ulto2b(0x0960, inq_ptr->version3);
10279         } else if (port_type == CTL_PORT_SAS) {
10280                 /* SAS (no version claimed) */
10281                 scsi_ulto2b(0x0BE0, inq_ptr->version3);
10282         }
10283
10284         if (lun == NULL) {
10285                 /* SBC-4 (no version claimed) */
10286                 scsi_ulto2b(0x0600, inq_ptr->version4);
10287         } else {
10288                 switch (lun->be_lun->lun_type) {
10289                 case T_DIRECT:
10290                         /* SBC-4 (no version claimed) */
10291                         scsi_ulto2b(0x0600, inq_ptr->version4);
10292                         break;
10293                 case T_PROCESSOR:
10294                         break;
10295                 case T_CDROM:
10296                         /* MMC-6 (no version claimed) */
10297                         scsi_ulto2b(0x04E0, inq_ptr->version4);
10298                         break;
10299                 default:
10300                         break;
10301                 }
10302         }
10303
10304         ctl_set_success(ctsio);
10305         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10306         ctsio->be_move_done = ctl_config_move_done;
10307         ctl_datamove((union ctl_io *)ctsio);
10308         return (CTL_RETVAL_COMPLETE);
10309 }
10310
10311 int
10312 ctl_inquiry(struct ctl_scsiio *ctsio)
10313 {
10314         struct scsi_inquiry *cdb;
10315         int retval;
10316
10317         CTL_DEBUG_PRINT(("ctl_inquiry\n"));
10318
10319         cdb = (struct scsi_inquiry *)ctsio->cdb;
10320         if (cdb->byte2 & SI_EVPD)
10321                 retval = ctl_inquiry_evpd(ctsio);
10322         else if (cdb->page_code == 0)
10323                 retval = ctl_inquiry_std(ctsio);
10324         else {
10325                 ctl_set_invalid_field(ctsio,
10326                                       /*sks_valid*/ 1,
10327                                       /*command*/ 1,
10328                                       /*field*/ 2,
10329                                       /*bit_valid*/ 0,
10330                                       /*bit*/ 0);
10331                 ctl_done((union ctl_io *)ctsio);
10332                 return (CTL_RETVAL_COMPLETE);
10333         }
10334
10335         return (retval);
10336 }
10337
10338 int
10339 ctl_get_config(struct ctl_scsiio *ctsio)
10340 {
10341         struct scsi_get_config_header *hdr;
10342         struct scsi_get_config_feature *feature;
10343         struct scsi_get_config *cdb;
10344         struct ctl_lun *lun;
10345         uint32_t alloc_len, data_len;
10346         int rt, starting;
10347
10348         lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10349         cdb = (struct scsi_get_config *)ctsio->cdb;
10350         rt = (cdb->rt & SGC_RT_MASK);
10351         starting = scsi_2btoul(cdb->starting_feature);
10352         alloc_len = scsi_2btoul(cdb->length);
10353
10354         data_len = sizeof(struct scsi_get_config_header) +
10355             sizeof(struct scsi_get_config_feature) + 8 +
10356             sizeof(struct scsi_get_config_feature) + 8 +
10357             sizeof(struct scsi_get_config_feature) + 4 +
10358             sizeof(struct scsi_get_config_feature) + 4 +
10359             sizeof(struct scsi_get_config_feature) + 8 +
10360             sizeof(struct scsi_get_config_feature) +
10361             sizeof(struct scsi_get_config_feature) + 4 +
10362             sizeof(struct scsi_get_config_feature) + 4 +
10363             sizeof(struct scsi_get_config_feature) + 4 +
10364             sizeof(struct scsi_get_config_feature) + 4 +
10365             sizeof(struct scsi_get_config_feature) + 4 +
10366             sizeof(struct scsi_get_config_feature) + 4;
10367         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10368         ctsio->kern_sg_entries = 0;
10369         ctsio->kern_data_resid = 0;
10370         ctsio->kern_rel_offset = 0;
10371
10372         hdr = (struct scsi_get_config_header *)ctsio->kern_data_ptr;
10373         if (lun->flags & CTL_LUN_NO_MEDIA)
10374                 scsi_ulto2b(0x0000, hdr->current_profile);
10375         else
10376                 scsi_ulto2b(0x0010, hdr->current_profile);
10377         feature = (struct scsi_get_config_feature *)(hdr + 1);
10378
10379         if (starting > 0x003b)
10380                 goto done;
10381         if (starting > 0x003a)
10382                 goto f3b;
10383         if (starting > 0x002b)
10384                 goto f3a;
10385         if (starting > 0x002a)
10386                 goto f2b;
10387         if (starting > 0x001f)
10388                 goto f2a;
10389         if (starting > 0x001e)
10390                 goto f1f;
10391         if (starting > 0x001d)
10392                 goto f1e;
10393         if (starting > 0x0010)
10394                 goto f1d;
10395         if (starting > 0x0003)
10396                 goto f10;
10397         if (starting > 0x0002)
10398                 goto f3;
10399         if (starting > 0x0001)
10400                 goto f2;
10401         if (starting > 0x0000)
10402                 goto f1;
10403
10404         /* Profile List */
10405         scsi_ulto2b(0x0000, feature->feature_code);
10406         feature->flags = SGC_F_PERSISTENT | SGC_F_CURRENT;
10407         feature->add_length = 8;
10408         scsi_ulto2b(0x0008, &feature->feature_data[0]); /* CD-ROM */
10409         feature->feature_data[2] = 0x00;
10410         scsi_ulto2b(0x0010, &feature->feature_data[4]); /* DVD-ROM */
10411         feature->feature_data[6] = 0x01;
10412         feature = (struct scsi_get_config_feature *)
10413             &feature->feature_data[feature->add_length];
10414
10415 f1:     /* Core */
10416         scsi_ulto2b(0x0001, feature->feature_code);
10417         feature->flags = 0x08 | SGC_F_PERSISTENT | SGC_F_CURRENT;
10418         feature->add_length = 8;
10419         scsi_ulto4b(0x00000000, &feature->feature_data[0]);
10420         feature->feature_data[4] = 0x03;
10421         feature = (struct scsi_get_config_feature *)
10422             &feature->feature_data[feature->add_length];
10423
10424 f2:     /* Morphing */
10425         scsi_ulto2b(0x0002, feature->feature_code);
10426         feature->flags = 0x04 | SGC_F_PERSISTENT | SGC_F_CURRENT;
10427         feature->add_length = 4;
10428         feature->feature_data[0] = 0x02;
10429         feature = (struct scsi_get_config_feature *)
10430             &feature->feature_data[feature->add_length];
10431
10432 f3:     /* Removable Medium */
10433         scsi_ulto2b(0x0003, feature->feature_code);
10434         feature->flags = 0x04 | SGC_F_PERSISTENT | SGC_F_CURRENT;
10435         feature->add_length = 4;
10436         feature->feature_data[0] = 0x39;
10437         feature = (struct scsi_get_config_feature *)
10438             &feature->feature_data[feature->add_length];
10439
10440         if (rt == SGC_RT_CURRENT && (lun->flags & CTL_LUN_NO_MEDIA))
10441                 goto done;
10442
10443 f10:    /* Random Read */
10444         scsi_ulto2b(0x0010, feature->feature_code);
10445         feature->flags = 0x00;
10446         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10447                 feature->flags |= SGC_F_CURRENT;
10448         feature->add_length = 8;
10449         scsi_ulto4b(lun->be_lun->blocksize, &feature->feature_data[0]);
10450         scsi_ulto2b(1, &feature->feature_data[4]);
10451         feature->feature_data[6] = 0x00;
10452         feature = (struct scsi_get_config_feature *)
10453             &feature->feature_data[feature->add_length];
10454
10455 f1d:    /* Multi-Read */
10456         scsi_ulto2b(0x001D, feature->feature_code);
10457         feature->flags = 0x00;
10458         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10459                 feature->flags |= SGC_F_CURRENT;
10460         feature->add_length = 0;
10461         feature = (struct scsi_get_config_feature *)
10462             &feature->feature_data[feature->add_length];
10463
10464 f1e:    /* CD Read */
10465         scsi_ulto2b(0x001E, feature->feature_code);
10466         feature->flags = 0x00;
10467         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10468                 feature->flags |= SGC_F_CURRENT;
10469         feature->add_length = 4;
10470         feature->feature_data[0] = 0x00;
10471         feature = (struct scsi_get_config_feature *)
10472             &feature->feature_data[feature->add_length];
10473
10474 f1f:    /* DVD Read */
10475         scsi_ulto2b(0x001F, feature->feature_code);
10476         feature->flags = 0x08;
10477         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10478                 feature->flags |= SGC_F_CURRENT;
10479         feature->add_length = 4;
10480         feature->feature_data[0] = 0x01;
10481         feature->feature_data[2] = 0x03;
10482         feature = (struct scsi_get_config_feature *)
10483             &feature->feature_data[feature->add_length];
10484
10485 f2a:    /* DVD+RW */
10486         scsi_ulto2b(0x002A, feature->feature_code);
10487         feature->flags = 0x04;
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] = 0x00;
10492         feature->feature_data[1] = 0x00;
10493         feature = (struct scsi_get_config_feature *)
10494             &feature->feature_data[feature->add_length];
10495
10496 f2b:    /* DVD+R */
10497         scsi_ulto2b(0x002B, feature->feature_code);
10498         feature->flags = 0x00;
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 = (struct scsi_get_config_feature *)
10504             &feature->feature_data[feature->add_length];
10505
10506 f3a:    /* DVD+RW Dual Layer */
10507         scsi_ulto2b(0x003A, feature->feature_code);
10508         feature->flags = 0x00;
10509         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10510                 feature->flags |= SGC_F_CURRENT;
10511         feature->add_length = 4;
10512         feature->feature_data[0] = 0x00;
10513         feature->feature_data[1] = 0x00;
10514         feature = (struct scsi_get_config_feature *)
10515             &feature->feature_data[feature->add_length];
10516
10517 f3b:    /* DVD+R Dual Layer */
10518         scsi_ulto2b(0x003B, 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 = (struct scsi_get_config_feature *)
10525             &feature->feature_data[feature->add_length];
10526
10527 done:
10528         data_len = (uint8_t *)feature - (uint8_t *)hdr;
10529         if (rt == SGC_RT_SPECIFIC && data_len > 4) {
10530                 feature = (struct scsi_get_config_feature *)(hdr + 1);
10531                 if (scsi_2btoul(feature->feature_code) == starting)
10532                         feature = (struct scsi_get_config_feature *)
10533                             &feature->feature_data[feature->add_length];
10534                 data_len = (uint8_t *)feature - (uint8_t *)hdr;
10535         }
10536         scsi_ulto4b(data_len - 4, hdr->data_length);
10537         if (data_len < alloc_len) {
10538                 ctsio->residual = alloc_len - data_len;
10539                 ctsio->kern_data_len = data_len;
10540                 ctsio->kern_total_len = data_len;
10541         } else {
10542                 ctsio->residual = 0;
10543                 ctsio->kern_data_len = alloc_len;
10544                 ctsio->kern_total_len = alloc_len;
10545         }
10546
10547         ctl_set_success(ctsio);
10548         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10549         ctsio->be_move_done = ctl_config_move_done;
10550         ctl_datamove((union ctl_io *)ctsio);
10551         return (CTL_RETVAL_COMPLETE);
10552 }
10553
10554 int
10555 ctl_get_event_status(struct ctl_scsiio *ctsio)
10556 {
10557         struct scsi_get_event_status_header *hdr;
10558         struct scsi_get_event_status *cdb;
10559         struct ctl_lun *lun;
10560         uint32_t alloc_len, data_len;
10561         int notif_class;
10562
10563         lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10564         cdb = (struct scsi_get_event_status *)ctsio->cdb;
10565         if ((cdb->byte2 & SGESN_POLLED) == 0) {
10566                 ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 1,
10567                     /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 0);
10568                 ctl_done((union ctl_io *)ctsio);
10569                 return (CTL_RETVAL_COMPLETE);
10570         }
10571         notif_class = cdb->notif_class;
10572         alloc_len = scsi_2btoul(cdb->length);
10573
10574         data_len = sizeof(struct scsi_get_event_status_header);
10575         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10576         ctsio->kern_sg_entries = 0;
10577         ctsio->kern_data_resid = 0;
10578         ctsio->kern_rel_offset = 0;
10579
10580         if (data_len < alloc_len) {
10581                 ctsio->residual = alloc_len - data_len;
10582                 ctsio->kern_data_len = data_len;
10583                 ctsio->kern_total_len = data_len;
10584         } else {
10585                 ctsio->residual = 0;
10586                 ctsio->kern_data_len = alloc_len;
10587                 ctsio->kern_total_len = alloc_len;
10588         }
10589
10590         hdr = (struct scsi_get_event_status_header *)ctsio->kern_data_ptr;
10591         scsi_ulto2b(0, hdr->descr_length);
10592         hdr->nea_class = SGESN_NEA;
10593         hdr->supported_class = 0;
10594
10595         ctl_set_success(ctsio);
10596         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10597         ctsio->be_move_done = ctl_config_move_done;
10598         ctl_datamove((union ctl_io *)ctsio);
10599         return (CTL_RETVAL_COMPLETE);
10600 }
10601
10602 int
10603 ctl_mechanism_status(struct ctl_scsiio *ctsio)
10604 {
10605         struct scsi_mechanism_status_header *hdr;
10606         struct scsi_mechanism_status *cdb;
10607         struct ctl_lun *lun;
10608         uint32_t alloc_len, data_len;
10609
10610         lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10611         cdb = (struct scsi_mechanism_status *)ctsio->cdb;
10612         alloc_len = scsi_2btoul(cdb->length);
10613
10614         data_len = sizeof(struct scsi_mechanism_status_header);
10615         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10616         ctsio->kern_sg_entries = 0;
10617         ctsio->kern_data_resid = 0;
10618         ctsio->kern_rel_offset = 0;
10619
10620         if (data_len < alloc_len) {
10621                 ctsio->residual = alloc_len - data_len;
10622                 ctsio->kern_data_len = data_len;
10623                 ctsio->kern_total_len = data_len;
10624         } else {
10625                 ctsio->residual = 0;
10626                 ctsio->kern_data_len = alloc_len;
10627                 ctsio->kern_total_len = alloc_len;
10628         }
10629
10630         hdr = (struct scsi_mechanism_status_header *)ctsio->kern_data_ptr;
10631         hdr->state1 = 0x00;
10632         hdr->state2 = 0xe0;
10633         scsi_ulto3b(0, hdr->lba);
10634         hdr->slots_num = 0;
10635         scsi_ulto2b(0, hdr->slots_length);
10636
10637         ctl_set_success(ctsio);
10638         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10639         ctsio->be_move_done = ctl_config_move_done;
10640         ctl_datamove((union ctl_io *)ctsio);
10641         return (CTL_RETVAL_COMPLETE);
10642 }
10643
10644 static void
10645 ctl_ultomsf(uint32_t lba, uint8_t *buf)
10646 {
10647
10648         lba += 150;
10649         buf[0] = 0;
10650         buf[1] = bin2bcd((lba / 75) / 60);
10651         buf[2] = bin2bcd((lba / 75) % 60);
10652         buf[3] = bin2bcd(lba % 75);
10653 }
10654
10655 int
10656 ctl_read_toc(struct ctl_scsiio *ctsio)
10657 {
10658         struct scsi_read_toc_hdr *hdr;
10659         struct scsi_read_toc_type01_descr *descr;
10660         struct scsi_read_toc *cdb;
10661         struct ctl_lun *lun;
10662         uint32_t alloc_len, data_len;
10663         int format, msf;
10664
10665         lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10666         cdb = (struct scsi_read_toc *)ctsio->cdb;
10667         msf = (cdb->byte2 & CD_MSF) != 0;
10668         format = cdb->format;
10669         alloc_len = scsi_2btoul(cdb->data_len);
10670
10671         data_len = sizeof(struct scsi_read_toc_hdr);
10672         if (format == 0)
10673                 data_len += 2 * sizeof(struct scsi_read_toc_type01_descr);
10674         else
10675                 data_len += sizeof(struct scsi_read_toc_type01_descr);
10676         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10677         ctsio->kern_sg_entries = 0;
10678         ctsio->kern_data_resid = 0;
10679         ctsio->kern_rel_offset = 0;
10680
10681         if (data_len < alloc_len) {
10682                 ctsio->residual = alloc_len - data_len;
10683                 ctsio->kern_data_len = data_len;
10684                 ctsio->kern_total_len = data_len;
10685         } else {
10686                 ctsio->residual = 0;
10687                 ctsio->kern_data_len = alloc_len;
10688                 ctsio->kern_total_len = alloc_len;
10689         }
10690
10691         hdr = (struct scsi_read_toc_hdr *)ctsio->kern_data_ptr;
10692         if (format == 0) {
10693                 scsi_ulto2b(0x12, hdr->data_length);
10694                 hdr->first = 1;
10695                 hdr->last = 1;
10696                 descr = (struct scsi_read_toc_type01_descr *)(hdr + 1);
10697                 descr->addr_ctl = 0x14;
10698                 descr->track_number = 1;
10699                 if (msf)
10700                         ctl_ultomsf(0, descr->track_start);
10701                 else
10702                         scsi_ulto4b(0, descr->track_start);
10703                 descr++;
10704                 descr->addr_ctl = 0x14;
10705                 descr->track_number = 0xaa;
10706                 if (msf)
10707                         ctl_ultomsf(lun->be_lun->maxlba+1, descr->track_start);
10708                 else
10709                         scsi_ulto4b(lun->be_lun->maxlba+1, descr->track_start);
10710         } else {
10711                 scsi_ulto2b(0x0a, hdr->data_length);
10712                 hdr->first = 1;
10713                 hdr->last = 1;
10714                 descr = (struct scsi_read_toc_type01_descr *)(hdr + 1);
10715                 descr->addr_ctl = 0x14;
10716                 descr->track_number = 1;
10717                 if (msf)
10718                         ctl_ultomsf(0, descr->track_start);
10719                 else
10720                         scsi_ulto4b(0, descr->track_start);
10721         }
10722
10723         ctl_set_success(ctsio);
10724         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10725         ctsio->be_move_done = ctl_config_move_done;
10726         ctl_datamove((union ctl_io *)ctsio);
10727         return (CTL_RETVAL_COMPLETE);
10728 }
10729
10730 /*
10731  * For known CDB types, parse the LBA and length.
10732  */
10733 static int
10734 ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len)
10735 {
10736         if (io->io_hdr.io_type != CTL_IO_SCSI)
10737                 return (1);
10738
10739         switch (io->scsiio.cdb[0]) {
10740         case COMPARE_AND_WRITE: {
10741                 struct scsi_compare_and_write *cdb;
10742
10743                 cdb = (struct scsi_compare_and_write *)io->scsiio.cdb;
10744
10745                 *lba = scsi_8btou64(cdb->addr);
10746                 *len = cdb->length;
10747                 break;
10748         }
10749         case READ_6:
10750         case WRITE_6: {
10751                 struct scsi_rw_6 *cdb;
10752
10753                 cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
10754
10755                 *lba = scsi_3btoul(cdb->addr);
10756                 /* only 5 bits are valid in the most significant address byte */
10757                 *lba &= 0x1fffff;
10758                 *len = cdb->length;
10759                 break;
10760         }
10761         case READ_10:
10762         case WRITE_10: {
10763                 struct scsi_rw_10 *cdb;
10764
10765                 cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
10766
10767                 *lba = scsi_4btoul(cdb->addr);
10768                 *len = scsi_2btoul(cdb->length);
10769                 break;
10770         }
10771         case WRITE_VERIFY_10: {
10772                 struct scsi_write_verify_10 *cdb;
10773
10774                 cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
10775
10776                 *lba = scsi_4btoul(cdb->addr);
10777                 *len = scsi_2btoul(cdb->length);
10778                 break;
10779         }
10780         case READ_12:
10781         case WRITE_12: {
10782                 struct scsi_rw_12 *cdb;
10783
10784                 cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
10785
10786                 *lba = scsi_4btoul(cdb->addr);
10787                 *len = scsi_4btoul(cdb->length);
10788                 break;
10789         }
10790         case WRITE_VERIFY_12: {
10791                 struct scsi_write_verify_12 *cdb;
10792
10793                 cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
10794
10795                 *lba = scsi_4btoul(cdb->addr);
10796                 *len = scsi_4btoul(cdb->length);
10797                 break;
10798         }
10799         case READ_16:
10800         case WRITE_16: {
10801                 struct scsi_rw_16 *cdb;
10802
10803                 cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
10804
10805                 *lba = scsi_8btou64(cdb->addr);
10806                 *len = scsi_4btoul(cdb->length);
10807                 break;
10808         }
10809         case WRITE_ATOMIC_16: {
10810                 struct scsi_write_atomic_16 *cdb;
10811
10812                 cdb = (struct scsi_write_atomic_16 *)io->scsiio.cdb;
10813
10814                 *lba = scsi_8btou64(cdb->addr);
10815                 *len = scsi_2btoul(cdb->length);
10816                 break;
10817         }
10818         case WRITE_VERIFY_16: {
10819                 struct scsi_write_verify_16 *cdb;
10820
10821                 cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
10822
10823                 *lba = scsi_8btou64(cdb->addr);
10824                 *len = scsi_4btoul(cdb->length);
10825                 break;
10826         }
10827         case WRITE_SAME_10: {
10828                 struct scsi_write_same_10 *cdb;
10829
10830                 cdb = (struct scsi_write_same_10 *)io->scsiio.cdb;
10831
10832                 *lba = scsi_4btoul(cdb->addr);
10833                 *len = scsi_2btoul(cdb->length);
10834                 break;
10835         }
10836         case WRITE_SAME_16: {
10837                 struct scsi_write_same_16 *cdb;
10838
10839                 cdb = (struct scsi_write_same_16 *)io->scsiio.cdb;
10840
10841                 *lba = scsi_8btou64(cdb->addr);
10842                 *len = scsi_4btoul(cdb->length);
10843                 break;
10844         }
10845         case VERIFY_10: {
10846                 struct scsi_verify_10 *cdb;
10847
10848                 cdb = (struct scsi_verify_10 *)io->scsiio.cdb;
10849
10850                 *lba = scsi_4btoul(cdb->addr);
10851                 *len = scsi_2btoul(cdb->length);
10852                 break;
10853         }
10854         case VERIFY_12: {
10855                 struct scsi_verify_12 *cdb;
10856
10857                 cdb = (struct scsi_verify_12 *)io->scsiio.cdb;
10858
10859                 *lba = scsi_4btoul(cdb->addr);
10860                 *len = scsi_4btoul(cdb->length);
10861                 break;
10862         }
10863         case VERIFY_16: {
10864                 struct scsi_verify_16 *cdb;
10865
10866                 cdb = (struct scsi_verify_16 *)io->scsiio.cdb;
10867
10868                 *lba = scsi_8btou64(cdb->addr);
10869                 *len = scsi_4btoul(cdb->length);
10870                 break;
10871         }
10872         case UNMAP: {
10873                 *lba = 0;
10874                 *len = UINT64_MAX;
10875                 break;
10876         }
10877         case SERVICE_ACTION_IN: {       /* GET LBA STATUS */
10878                 struct scsi_get_lba_status *cdb;
10879
10880                 cdb = (struct scsi_get_lba_status *)io->scsiio.cdb;
10881                 *lba = scsi_8btou64(cdb->addr);
10882                 *len = UINT32_MAX;
10883                 break;
10884         }
10885         default:
10886                 return (1);
10887                 break; /* NOTREACHED */
10888         }
10889
10890         return (0);
10891 }
10892
10893 static ctl_action
10894 ctl_extent_check_lba(uint64_t lba1, uint64_t len1, uint64_t lba2, uint64_t len2,
10895     bool seq)
10896 {
10897         uint64_t endlba1, endlba2;
10898
10899         endlba1 = lba1 + len1 - (seq ? 0 : 1);
10900         endlba2 = lba2 + len2 - 1;
10901
10902         if ((endlba1 < lba2) || (endlba2 < lba1))
10903                 return (CTL_ACTION_PASS);
10904         else
10905                 return (CTL_ACTION_BLOCK);
10906 }
10907
10908 static int
10909 ctl_extent_check_unmap(union ctl_io *io, uint64_t lba2, uint64_t len2)
10910 {
10911         struct ctl_ptr_len_flags *ptrlen;
10912         struct scsi_unmap_desc *buf, *end, *range;
10913         uint64_t lba;
10914         uint32_t len;
10915
10916         /* If not UNMAP -- go other way. */
10917         if (io->io_hdr.io_type != CTL_IO_SCSI ||
10918             io->scsiio.cdb[0] != UNMAP)
10919                 return (CTL_ACTION_ERROR);
10920
10921         /* If UNMAP without data -- block and wait for data. */
10922         ptrlen = (struct ctl_ptr_len_flags *)
10923             &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
10924         if ((io->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0 ||
10925             ptrlen->ptr == NULL)
10926                 return (CTL_ACTION_BLOCK);
10927
10928         /* UNMAP with data -- check for collision. */
10929         buf = (struct scsi_unmap_desc *)ptrlen->ptr;
10930         end = buf + ptrlen->len / sizeof(*buf);
10931         for (range = buf; range < end; range++) {
10932                 lba = scsi_8btou64(range->lba);
10933                 len = scsi_4btoul(range->length);
10934                 if ((lba < lba2 + len2) && (lba + len > lba2))
10935                         return (CTL_ACTION_BLOCK);
10936         }
10937         return (CTL_ACTION_PASS);
10938 }
10939
10940 static ctl_action
10941 ctl_extent_check(union ctl_io *io1, union ctl_io *io2, bool seq)
10942 {
10943         uint64_t lba1, lba2;
10944         uint64_t len1, len2;
10945         int retval;
10946
10947         if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
10948                 return (CTL_ACTION_ERROR);
10949
10950         retval = ctl_extent_check_unmap(io1, lba2, len2);
10951         if (retval != CTL_ACTION_ERROR)
10952                 return (retval);
10953
10954         if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
10955                 return (CTL_ACTION_ERROR);
10956
10957         if (io1->io_hdr.flags & CTL_FLAG_SERSEQ_DONE)
10958                 seq = FALSE;
10959         return (ctl_extent_check_lba(lba1, len1, lba2, len2, seq));
10960 }
10961
10962 static ctl_action
10963 ctl_extent_check_seq(union ctl_io *io1, union ctl_io *io2)
10964 {
10965         uint64_t lba1, lba2;
10966         uint64_t len1, len2;
10967
10968         if (io1->io_hdr.flags & CTL_FLAG_SERSEQ_DONE)
10969                 return (CTL_ACTION_PASS);
10970         if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
10971                 return (CTL_ACTION_ERROR);
10972         if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
10973                 return (CTL_ACTION_ERROR);
10974
10975         if (lba1 + len1 == lba2)
10976                 return (CTL_ACTION_BLOCK);
10977         return (CTL_ACTION_PASS);
10978 }
10979
10980 static ctl_action
10981 ctl_check_for_blockage(struct ctl_lun *lun, union ctl_io *pending_io,
10982     union ctl_io *ooa_io)
10983 {
10984         const struct ctl_cmd_entry *pending_entry, *ooa_entry;
10985         const ctl_serialize_action *serialize_row;
10986
10987         /*
10988          * The initiator attempted multiple untagged commands at the same
10989          * time.  Can't do that.
10990          */
10991         if ((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10992          && (ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10993          && ((pending_io->io_hdr.nexus.targ_port ==
10994               ooa_io->io_hdr.nexus.targ_port)
10995           && (pending_io->io_hdr.nexus.initid ==
10996               ooa_io->io_hdr.nexus.initid))
10997          && ((ooa_io->io_hdr.flags & (CTL_FLAG_ABORT |
10998               CTL_FLAG_STATUS_SENT)) == 0))
10999                 return (CTL_ACTION_OVERLAP);
11000
11001         /*
11002          * The initiator attempted to send multiple tagged commands with
11003          * the same ID.  (It's fine if different initiators have the same
11004          * tag ID.)
11005          *
11006          * Even if all of those conditions are true, we don't kill the I/O
11007          * if the command ahead of us has been aborted.  We won't end up
11008          * sending it to the FETD, and it's perfectly legal to resend a
11009          * command with the same tag number as long as the previous
11010          * instance of this tag number has been aborted somehow.
11011          */
11012         if ((pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
11013          && (ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
11014          && (pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
11015          && ((pending_io->io_hdr.nexus.targ_port ==
11016               ooa_io->io_hdr.nexus.targ_port)
11017           && (pending_io->io_hdr.nexus.initid ==
11018               ooa_io->io_hdr.nexus.initid))
11019          && ((ooa_io->io_hdr.flags & (CTL_FLAG_ABORT |
11020               CTL_FLAG_STATUS_SENT)) == 0))
11021                 return (CTL_ACTION_OVERLAP_TAG);
11022
11023         /*
11024          * If we get a head of queue tag, SAM-3 says that we should
11025          * immediately execute it.
11026          *
11027          * What happens if this command would normally block for some other
11028          * reason?  e.g. a request sense with a head of queue tag
11029          * immediately after a write.  Normally that would block, but this
11030          * will result in its getting executed immediately...
11031          *
11032          * We currently return "pass" instead of "skip", so we'll end up
11033          * going through the rest of the queue to check for overlapped tags.
11034          *
11035          * XXX KDM check for other types of blockage first??
11036          */
11037         if (pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
11038                 return (CTL_ACTION_PASS);
11039
11040         /*
11041          * Ordered tags have to block until all items ahead of them
11042          * have completed.  If we get called with an ordered tag, we always
11043          * block, if something else is ahead of us in the queue.
11044          */
11045         if (pending_io->scsiio.tag_type == CTL_TAG_ORDERED)
11046                 return (CTL_ACTION_BLOCK);
11047
11048         /*
11049          * Simple tags get blocked until all head of queue and ordered tags
11050          * ahead of them have completed.  I'm lumping untagged commands in
11051          * with simple tags here.  XXX KDM is that the right thing to do?
11052          */
11053         if (((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
11054           || (pending_io->scsiio.tag_type == CTL_TAG_SIMPLE))
11055          && ((ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
11056           || (ooa_io->scsiio.tag_type == CTL_TAG_ORDERED)))
11057                 return (CTL_ACTION_BLOCK);
11058
11059         pending_entry = ctl_get_cmd_entry(&pending_io->scsiio, NULL);
11060         KASSERT(pending_entry->seridx < CTL_SERIDX_COUNT,
11061             ("%s: Invalid seridx %d for pending CDB %02x %02x @ %p",
11062              __func__, pending_entry->seridx, pending_io->scsiio.cdb[0],
11063              pending_io->scsiio.cdb[1], pending_io));
11064         ooa_entry = ctl_get_cmd_entry(&ooa_io->scsiio, NULL);
11065         if (ooa_entry->seridx == CTL_SERIDX_INVLD)
11066                 return (CTL_ACTION_PASS); /* Unsupported command in OOA queue */
11067         KASSERT(ooa_entry->seridx < CTL_SERIDX_COUNT,
11068             ("%s: Invalid seridx %d for ooa CDB %02x %02x @ %p",
11069              __func__, ooa_entry->seridx, ooa_io->scsiio.cdb[0],
11070              ooa_io->scsiio.cdb[1], ooa_io));
11071
11072         serialize_row = ctl_serialize_table[ooa_entry->seridx];
11073
11074         switch (serialize_row[pending_entry->seridx]) {
11075         case CTL_SER_BLOCK:
11076                 return (CTL_ACTION_BLOCK);
11077         case CTL_SER_EXTENT:
11078                 return (ctl_extent_check(ooa_io, pending_io,
11079                     (lun->be_lun && lun->be_lun->serseq == CTL_LUN_SERSEQ_ON)));
11080         case CTL_SER_EXTENTOPT:
11081                 if ((lun->MODE_CTRL.queue_flags & SCP_QUEUE_ALG_MASK) !=
11082                     SCP_QUEUE_ALG_UNRESTRICTED)
11083                         return (ctl_extent_check(ooa_io, pending_io,
11084                             (lun->be_lun &&
11085                              lun->be_lun->serseq == CTL_LUN_SERSEQ_ON)));
11086                 return (CTL_ACTION_PASS);
11087         case CTL_SER_EXTENTSEQ:
11088                 if (lun->be_lun && lun->be_lun->serseq != CTL_LUN_SERSEQ_OFF)
11089                         return (ctl_extent_check_seq(ooa_io, pending_io));
11090                 return (CTL_ACTION_PASS);
11091         case CTL_SER_PASS:
11092                 return (CTL_ACTION_PASS);
11093         case CTL_SER_BLOCKOPT:
11094                 if ((lun->MODE_CTRL.queue_flags & SCP_QUEUE_ALG_MASK) !=
11095                     SCP_QUEUE_ALG_UNRESTRICTED)
11096                         return (CTL_ACTION_BLOCK);
11097                 return (CTL_ACTION_PASS);
11098         case CTL_SER_SKIP:
11099                 return (CTL_ACTION_SKIP);
11100         default:
11101                 panic("%s: Invalid serialization value %d for %d => %d",
11102                     __func__, serialize_row[pending_entry->seridx],
11103                     pending_entry->seridx, ooa_entry->seridx);
11104         }
11105
11106         return (CTL_ACTION_ERROR);
11107 }
11108
11109 /*
11110  * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
11111  * Assumptions:
11112  * - pending_io is generally either incoming, or on the blocked queue
11113  * - starting I/O is the I/O we want to start the check with.
11114  */
11115 static ctl_action
11116 ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
11117               union ctl_io *starting_io)
11118 {
11119         union ctl_io *ooa_io;
11120         ctl_action action;
11121
11122         mtx_assert(&lun->lun_lock, MA_OWNED);
11123
11124         /*
11125          * Run back along the OOA queue, starting with the current
11126          * blocked I/O and going through every I/O before it on the
11127          * queue.  If starting_io is NULL, we'll just end up returning
11128          * CTL_ACTION_PASS.
11129          */
11130         for (ooa_io = starting_io; ooa_io != NULL;
11131              ooa_io = (union ctl_io *)TAILQ_PREV(&ooa_io->io_hdr, ctl_ooaq,
11132              ooa_links)){
11133
11134                 /*
11135                  * This routine just checks to see whether
11136                  * cur_blocked is blocked by ooa_io, which is ahead
11137                  * of it in the queue.  It doesn't queue/dequeue
11138                  * cur_blocked.
11139                  */
11140                 action = ctl_check_for_blockage(lun, pending_io, ooa_io);
11141                 switch (action) {
11142                 case CTL_ACTION_BLOCK:
11143                 case CTL_ACTION_OVERLAP:
11144                 case CTL_ACTION_OVERLAP_TAG:
11145                 case CTL_ACTION_SKIP:
11146                 case CTL_ACTION_ERROR:
11147                         return (action);
11148                         break; /* NOTREACHED */
11149                 case CTL_ACTION_PASS:
11150                         break;
11151                 default:
11152                         panic("%s: Invalid action %d\n", __func__, action);
11153                 }
11154         }
11155
11156         return (CTL_ACTION_PASS);
11157 }
11158
11159 /*
11160  * Assumptions:
11161  * - An I/O has just completed, and has been removed from the per-LUN OOA
11162  *   queue, so some items on the blocked queue may now be unblocked.
11163  */
11164 static int
11165 ctl_check_blocked(struct ctl_lun *lun)
11166 {
11167         struct ctl_softc *softc = lun->ctl_softc;
11168         union ctl_io *cur_blocked, *next_blocked;
11169
11170         mtx_assert(&lun->lun_lock, MA_OWNED);
11171
11172         /*
11173          * Run forward from the head of the blocked queue, checking each
11174          * entry against the I/Os prior to it on the OOA queue to see if
11175          * there is still any blockage.
11176          *
11177          * We cannot use the TAILQ_FOREACH() macro, because it can't deal
11178          * with our removing a variable on it while it is traversing the
11179          * list.
11180          */
11181         for (cur_blocked = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue);
11182              cur_blocked != NULL; cur_blocked = next_blocked) {
11183                 union ctl_io *prev_ooa;
11184                 ctl_action action;
11185
11186                 next_blocked = (union ctl_io *)TAILQ_NEXT(&cur_blocked->io_hdr,
11187                                                           blocked_links);
11188
11189                 prev_ooa = (union ctl_io *)TAILQ_PREV(&cur_blocked->io_hdr,
11190                                                       ctl_ooaq, ooa_links);
11191
11192                 /*
11193                  * If cur_blocked happens to be the first item in the OOA
11194                  * queue now, prev_ooa will be NULL, and the action
11195                  * returned will just be CTL_ACTION_PASS.
11196                  */
11197                 action = ctl_check_ooa(lun, cur_blocked, prev_ooa);
11198
11199                 switch (action) {
11200                 case CTL_ACTION_BLOCK:
11201                         /* Nothing to do here, still blocked */
11202                         break;
11203                 case CTL_ACTION_OVERLAP:
11204                 case CTL_ACTION_OVERLAP_TAG:
11205                         /*
11206                          * This shouldn't happen!  In theory we've already
11207                          * checked this command for overlap...
11208                          */
11209                         break;
11210                 case CTL_ACTION_PASS:
11211                 case CTL_ACTION_SKIP: {
11212                         const struct ctl_cmd_entry *entry;
11213
11214                         /*
11215                          * The skip case shouldn't happen, this transaction
11216                          * should have never made it onto the blocked queue.
11217                          */
11218                         /*
11219                          * This I/O is no longer blocked, we can remove it
11220                          * from the blocked queue.  Since this is a TAILQ
11221                          * (doubly linked list), we can do O(1) removals
11222                          * from any place on the list.
11223                          */
11224                         TAILQ_REMOVE(&lun->blocked_queue, &cur_blocked->io_hdr,
11225                                      blocked_links);
11226                         cur_blocked->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
11227
11228                         if ((softc->ha_mode != CTL_HA_MODE_XFER) &&
11229                             (cur_blocked->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)){
11230                                 /*
11231                                  * Need to send IO back to original side to
11232                                  * run
11233                                  */
11234                                 union ctl_ha_msg msg_info;
11235
11236                                 cur_blocked->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11237                                 msg_info.hdr.original_sc =
11238                                         cur_blocked->io_hdr.original_sc;
11239                                 msg_info.hdr.serializing_sc = cur_blocked;
11240                                 msg_info.hdr.msg_type = CTL_MSG_R2R;
11241                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11242                                     sizeof(msg_info.hdr), M_NOWAIT);
11243                                 break;
11244                         }
11245                         entry = ctl_get_cmd_entry(&cur_blocked->scsiio, NULL);
11246
11247                         /*
11248                          * Check this I/O for LUN state changes that may
11249                          * have happened while this command was blocked.
11250                          * The LUN state may have been changed by a command
11251                          * ahead of us in the queue, so we need to re-check
11252                          * for any states that can be caused by SCSI
11253                          * commands.
11254                          */
11255                         if (ctl_scsiio_lun_check(lun, entry,
11256                                                  &cur_blocked->scsiio) == 0) {
11257                                 cur_blocked->io_hdr.flags |=
11258                                                       CTL_FLAG_IS_WAS_ON_RTR;
11259                                 ctl_enqueue_rtr(cur_blocked);
11260                         } else
11261                                 ctl_done(cur_blocked);
11262                         break;
11263                 }
11264                 default:
11265                         /*
11266                          * This probably shouldn't happen -- we shouldn't
11267                          * get CTL_ACTION_ERROR, or anything else.
11268                          */
11269                         break;
11270                 }
11271         }
11272
11273         return (CTL_RETVAL_COMPLETE);
11274 }
11275
11276 /*
11277  * This routine (with one exception) checks LUN flags that can be set by
11278  * commands ahead of us in the OOA queue.  These flags have to be checked
11279  * when a command initially comes in, and when we pull a command off the
11280  * blocked queue and are preparing to execute it.  The reason we have to
11281  * check these flags for commands on the blocked queue is that the LUN
11282  * state may have been changed by a command ahead of us while we're on the
11283  * blocked queue.
11284  *
11285  * Ordering is somewhat important with these checks, so please pay
11286  * careful attention to the placement of any new checks.
11287  */
11288 static int
11289 ctl_scsiio_lun_check(struct ctl_lun *lun,
11290     const struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
11291 {
11292         struct ctl_softc *softc = lun->ctl_softc;
11293         int retval;
11294         uint32_t residx;
11295
11296         retval = 0;
11297
11298         mtx_assert(&lun->lun_lock, MA_OWNED);
11299
11300         /*
11301          * If this shelf is a secondary shelf controller, we may have to
11302          * reject some commands disallowed by HA mode and link state.
11303          */
11304         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
11305                 if (softc->ha_link == CTL_HA_LINK_OFFLINE &&
11306                     (entry->flags & CTL_CMD_FLAG_OK_ON_UNAVAIL) == 0) {
11307                         ctl_set_lun_unavail(ctsio);
11308                         retval = 1;
11309                         goto bailout;
11310                 }
11311                 if ((lun->flags & CTL_LUN_PEER_SC_PRIMARY) == 0 &&
11312                     (entry->flags & CTL_CMD_FLAG_OK_ON_UNAVAIL) == 0) {
11313                         ctl_set_lun_transit(ctsio);
11314                         retval = 1;
11315                         goto bailout;
11316                 }
11317                 if (softc->ha_mode == CTL_HA_MODE_ACT_STBY &&
11318                     (entry->flags & CTL_CMD_FLAG_OK_ON_STANDBY) == 0) {
11319                         ctl_set_lun_standby(ctsio);
11320                         retval = 1;
11321                         goto bailout;
11322                 }
11323
11324                 /* The rest of checks are only done on executing side */
11325                 if (softc->ha_mode == CTL_HA_MODE_XFER)
11326                         goto bailout;
11327         }
11328
11329         if (entry->pattern & CTL_LUN_PAT_WRITE) {
11330                 if (lun->be_lun &&
11331                     lun->be_lun->flags & CTL_LUN_FLAG_READONLY) {
11332                         ctl_set_hw_write_protected(ctsio);
11333                         retval = 1;
11334                         goto bailout;
11335                 }
11336                 if ((lun->MODE_CTRL.eca_and_aen & SCP_SWP) != 0) {
11337                         ctl_set_sense(ctsio, /*current_error*/ 1,
11338                             /*sense_key*/ SSD_KEY_DATA_PROTECT,
11339                             /*asc*/ 0x27, /*ascq*/ 0x02, SSD_ELEM_NONE);
11340                         retval = 1;
11341                         goto bailout;
11342                 }
11343         }
11344
11345         /*
11346          * Check for a reservation conflict.  If this command isn't allowed
11347          * even on reserved LUNs, and if this initiator isn't the one who
11348          * reserved us, reject the command with a reservation conflict.
11349          */
11350         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11351         if ((lun->flags & CTL_LUN_RESERVED)
11352          && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
11353                 if (lun->res_idx != residx) {
11354                         ctl_set_reservation_conflict(ctsio);
11355                         retval = 1;
11356                         goto bailout;
11357                 }
11358         }
11359
11360         if ((lun->flags & CTL_LUN_PR_RESERVED) == 0 ||
11361             (entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV)) {
11362                 /* No reservation or command is allowed. */;
11363         } else if ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_WRESV) &&
11364             (lun->pr_res_type == SPR_TYPE_WR_EX ||
11365              lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
11366              lun->pr_res_type == SPR_TYPE_WR_EX_AR)) {
11367                 /* The command is allowed for Write Exclusive resv. */;
11368         } else {
11369                 /*
11370                  * if we aren't registered or it's a res holder type
11371                  * reservation and this isn't the res holder then set a
11372                  * conflict.
11373                  */
11374                 if (ctl_get_prkey(lun, residx) == 0 ||
11375                     (residx != lun->pr_res_idx && lun->pr_res_type < 4)) {
11376                         ctl_set_reservation_conflict(ctsio);
11377                         retval = 1;
11378                         goto bailout;
11379                 }
11380         }
11381
11382         if ((entry->flags & CTL_CMD_FLAG_OK_ON_NO_MEDIA) == 0) {
11383                 if (lun->flags & CTL_LUN_EJECTED)
11384                         ctl_set_lun_ejected(ctsio);
11385                 else if (lun->flags & CTL_LUN_NO_MEDIA) {
11386                         if (lun->flags & CTL_LUN_REMOVABLE)
11387                                 ctl_set_lun_no_media(ctsio);
11388                         else
11389                                 ctl_set_lun_int_reqd(ctsio);
11390                 } else if (lun->flags & CTL_LUN_STOPPED)
11391                         ctl_set_lun_stopped(ctsio);
11392                 else
11393                         goto bailout;
11394                 retval = 1;
11395                 goto bailout;
11396         }
11397
11398 bailout:
11399         return (retval);
11400 }
11401
11402 static void
11403 ctl_failover_io(union ctl_io *io, int have_lock)
11404 {
11405         ctl_set_busy(&io->scsiio);
11406         ctl_done(io);
11407 }
11408
11409 static void
11410 ctl_failover_lun(union ctl_io *rio)
11411 {
11412         struct ctl_softc *softc = control_softc;
11413         struct ctl_lun *lun;
11414         struct ctl_io_hdr *io, *next_io;
11415         uint32_t targ_lun;
11416
11417         targ_lun = rio->io_hdr.nexus.targ_mapped_lun;
11418         CTL_DEBUG_PRINT(("FAILOVER for lun %ju\n", targ_lun));
11419
11420         /* Find and lock the LUN. */
11421         mtx_lock(&softc->ctl_lock);
11422         if ((targ_lun < CTL_MAX_LUNS) &&
11423             ((lun = softc->ctl_luns[targ_lun]) != NULL)) {
11424                 mtx_lock(&lun->lun_lock);
11425                 mtx_unlock(&softc->ctl_lock);
11426                 if (lun->flags & CTL_LUN_DISABLED) {
11427                         mtx_unlock(&lun->lun_lock);
11428                         return;
11429                 }
11430         } else {
11431                 mtx_unlock(&softc->ctl_lock);
11432                 return;
11433         }
11434
11435         if (softc->ha_mode == CTL_HA_MODE_XFER) {
11436                 TAILQ_FOREACH_SAFE(io, &lun->ooa_queue, ooa_links, next_io) {
11437                         /* We are master */
11438                         if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
11439                                 if (io->flags & CTL_FLAG_IO_ACTIVE) {
11440                                         io->flags |= CTL_FLAG_ABORT;
11441                                         io->flags |= CTL_FLAG_FAILOVER;
11442                                 } else { /* This can be only due to DATAMOVE */
11443                                         io->msg_type = CTL_MSG_DATAMOVE_DONE;
11444                                         io->flags &= ~CTL_FLAG_DMA_INPROG;
11445                                         io->flags |= CTL_FLAG_IO_ACTIVE;
11446                                         io->port_status = 31340;
11447                                         ctl_enqueue_isc((union ctl_io *)io);
11448                                 }
11449                         }
11450                         /* We are slave */
11451                         if (io->flags & CTL_FLAG_SENT_2OTHER_SC) {
11452                                 io->flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11453                                 if (io->flags & CTL_FLAG_IO_ACTIVE) {
11454                                         io->flags |= CTL_FLAG_FAILOVER;
11455                                 } else {
11456                                         ctl_set_busy(&((union ctl_io *)io)->
11457                                             scsiio);
11458                                         ctl_done((union ctl_io *)io);
11459                                 }
11460                         }
11461                 }
11462         } else { /* SERIALIZE modes */
11463                 TAILQ_FOREACH_SAFE(io, &lun->blocked_queue, blocked_links,
11464                     next_io) {
11465                         /* We are master */
11466                         if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
11467                                 TAILQ_REMOVE(&lun->blocked_queue, io,
11468                                     blocked_links);
11469                                 io->flags &= ~CTL_FLAG_BLOCKED;
11470                                 TAILQ_REMOVE(&lun->ooa_queue, io, ooa_links);
11471                                 ctl_free_io((union ctl_io *)io);
11472                         }
11473                 }
11474                 TAILQ_FOREACH_SAFE(io, &lun->ooa_queue, ooa_links, next_io) {
11475                         /* We are master */
11476                         if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
11477                                 TAILQ_REMOVE(&lun->ooa_queue, io, ooa_links);
11478                                 ctl_free_io((union ctl_io *)io);
11479                         }
11480                         /* We are slave */
11481                         if (io->flags & CTL_FLAG_SENT_2OTHER_SC) {
11482                                 io->flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11483                                 if (!(io->flags & CTL_FLAG_IO_ACTIVE)) {
11484                                         ctl_set_busy(&((union ctl_io *)io)->
11485                                             scsiio);
11486                                         ctl_done((union ctl_io *)io);
11487                                 }
11488                         }
11489                 }
11490                 ctl_check_blocked(lun);
11491         }
11492         mtx_unlock(&lun->lun_lock);
11493 }
11494
11495 static int
11496 ctl_scsiio_precheck(struct ctl_softc *softc, struct ctl_scsiio *ctsio)
11497 {
11498         struct ctl_lun *lun;
11499         const struct ctl_cmd_entry *entry;
11500         uint32_t initidx, targ_lun;
11501         int retval;
11502
11503         retval = 0;
11504
11505         lun = NULL;
11506
11507         targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
11508         if ((targ_lun < CTL_MAX_LUNS)
11509          && ((lun = softc->ctl_luns[targ_lun]) != NULL)) {
11510                 /*
11511                  * If the LUN is invalid, pretend that it doesn't exist.
11512                  * It will go away as soon as all pending I/O has been
11513                  * completed.
11514                  */
11515                 mtx_lock(&lun->lun_lock);
11516                 if (lun->flags & CTL_LUN_DISABLED) {
11517                         mtx_unlock(&lun->lun_lock);
11518                         lun = NULL;
11519                         ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
11520                         ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
11521                 } else {
11522                         ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = lun;
11523                         ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr =
11524                                 lun->be_lun;
11525
11526                         /*
11527                          * Every I/O goes into the OOA queue for a
11528                          * particular LUN, and stays there until completion.
11529                          */
11530 #ifdef CTL_TIME_IO
11531                         if (TAILQ_EMPTY(&lun->ooa_queue)) {
11532                                 lun->idle_time += getsbinuptime() -
11533                                     lun->last_busy;
11534                         }
11535 #endif
11536                         TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr,
11537                             ooa_links);
11538                 }
11539         } else {
11540                 ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
11541                 ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
11542         }
11543
11544         /* Get command entry and return error if it is unsuppotyed. */
11545         entry = ctl_validate_command(ctsio);
11546         if (entry == NULL) {
11547                 if (lun)
11548                         mtx_unlock(&lun->lun_lock);
11549                 return (retval);
11550         }
11551
11552         ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
11553         ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
11554
11555         /*
11556          * Check to see whether we can send this command to LUNs that don't
11557          * exist.  This should pretty much only be the case for inquiry
11558          * and request sense.  Further checks, below, really require having
11559          * a LUN, so we can't really check the command anymore.  Just put
11560          * it on the rtr queue.
11561          */
11562         if (lun == NULL) {
11563                 if (entry->flags & CTL_CMD_FLAG_OK_ON_NO_LUN) {
11564                         ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11565                         ctl_enqueue_rtr((union ctl_io *)ctsio);
11566                         return (retval);
11567                 }
11568
11569                 ctl_set_unsupported_lun(ctsio);
11570                 ctl_done((union ctl_io *)ctsio);
11571                 CTL_DEBUG_PRINT(("ctl_scsiio_precheck: bailing out due to invalid LUN\n"));
11572                 return (retval);
11573         } else {
11574                 /*
11575                  * Make sure we support this particular command on this LUN.
11576                  * e.g., we don't support writes to the control LUN.
11577                  */
11578                 if (!ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
11579                         mtx_unlock(&lun->lun_lock);
11580                         ctl_set_invalid_opcode(ctsio);
11581                         ctl_done((union ctl_io *)ctsio);
11582                         return (retval);
11583                 }
11584         }
11585
11586         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11587
11588 #ifdef CTL_WITH_CA
11589         /*
11590          * If we've got a request sense, it'll clear the contingent
11591          * allegiance condition.  Otherwise, if we have a CA condition for
11592          * this initiator, clear it, because it sent down a command other
11593          * than request sense.
11594          */
11595         if ((ctsio->cdb[0] != REQUEST_SENSE)
11596          && (ctl_is_set(lun->have_ca, initidx)))
11597                 ctl_clear_mask(lun->have_ca, initidx);
11598 #endif
11599
11600         /*
11601          * If the command has this flag set, it handles its own unit
11602          * attention reporting, we shouldn't do anything.  Otherwise we
11603          * check for any pending unit attentions, and send them back to the
11604          * initiator.  We only do this when a command initially comes in,
11605          * not when we pull it off the blocked queue.
11606          *
11607          * According to SAM-3, section 5.3.2, the order that things get
11608          * presented back to the host is basically unit attentions caused
11609          * by some sort of reset event, busy status, reservation conflicts
11610          * or task set full, and finally any other status.
11611          *
11612          * One issue here is that some of the unit attentions we report
11613          * don't fall into the "reset" category (e.g. "reported luns data
11614          * has changed").  So reporting it here, before the reservation
11615          * check, may be technically wrong.  I guess the only thing to do
11616          * would be to check for and report the reset events here, and then
11617          * check for the other unit attention types after we check for a
11618          * reservation conflict.
11619          *
11620          * XXX KDM need to fix this
11621          */
11622         if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
11623                 ctl_ua_type ua_type;
11624
11625                 ua_type = ctl_build_ua(lun, initidx, &ctsio->sense_data,
11626                     SSD_TYPE_NONE);
11627                 if (ua_type != CTL_UA_NONE) {
11628                         mtx_unlock(&lun->lun_lock);
11629                         ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
11630                         ctsio->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
11631                         ctsio->sense_len = SSD_FULL_SIZE;
11632                         ctl_done((union ctl_io *)ctsio);
11633                         return (retval);
11634                 }
11635         }
11636
11637
11638         if (ctl_scsiio_lun_check(lun, entry, ctsio) != 0) {
11639                 mtx_unlock(&lun->lun_lock);
11640                 ctl_done((union ctl_io *)ctsio);
11641                 return (retval);
11642         }
11643
11644         /*
11645          * XXX CHD this is where we want to send IO to other side if
11646          * this LUN is secondary on this SC. We will need to make a copy
11647          * of the IO and flag the IO on this side as SENT_2OTHER and the flag
11648          * the copy we send as FROM_OTHER.
11649          * We also need to stuff the address of the original IO so we can
11650          * find it easily. Something similar will need be done on the other
11651          * side so when we are done we can find the copy.
11652          */
11653         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
11654             (lun->flags & CTL_LUN_PEER_SC_PRIMARY) != 0 &&
11655             (entry->flags & CTL_CMD_FLAG_RUN_HERE) == 0) {
11656                 union ctl_ha_msg msg_info;
11657                 int isc_retval;
11658
11659                 ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11660                 ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11661                 mtx_unlock(&lun->lun_lock);
11662
11663                 msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
11664                 msg_info.hdr.original_sc = (union ctl_io *)ctsio;
11665                 msg_info.hdr.serializing_sc = NULL;
11666                 msg_info.hdr.nexus = ctsio->io_hdr.nexus;
11667                 msg_info.scsi.tag_num = ctsio->tag_num;
11668                 msg_info.scsi.tag_type = ctsio->tag_type;
11669                 msg_info.scsi.cdb_len = ctsio->cdb_len;
11670                 memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
11671
11672                 if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11673                     sizeof(msg_info.scsi) - sizeof(msg_info.scsi.sense_data),
11674                     M_WAITOK)) > CTL_HA_STATUS_SUCCESS) {
11675                         ctl_set_busy(ctsio);
11676                         ctl_done((union ctl_io *)ctsio);
11677                         return (retval);
11678                 }
11679                 return (retval);
11680         }
11681
11682         switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
11683                               (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr,
11684                               ctl_ooaq, ooa_links))) {
11685         case CTL_ACTION_BLOCK:
11686                 ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
11687                 TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
11688                                   blocked_links);
11689                 mtx_unlock(&lun->lun_lock);
11690                 return (retval);
11691         case CTL_ACTION_PASS:
11692         case CTL_ACTION_SKIP:
11693                 ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11694                 mtx_unlock(&lun->lun_lock);
11695                 ctl_enqueue_rtr((union ctl_io *)ctsio);
11696                 break;
11697         case CTL_ACTION_OVERLAP:
11698                 mtx_unlock(&lun->lun_lock);
11699                 ctl_set_overlapped_cmd(ctsio);
11700                 ctl_done((union ctl_io *)ctsio);
11701                 break;
11702         case CTL_ACTION_OVERLAP_TAG:
11703                 mtx_unlock(&lun->lun_lock);
11704                 ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
11705                 ctl_done((union ctl_io *)ctsio);
11706                 break;
11707         case CTL_ACTION_ERROR:
11708         default:
11709                 mtx_unlock(&lun->lun_lock);
11710                 ctl_set_internal_failure(ctsio,
11711                                          /*sks_valid*/ 0,
11712                                          /*retry_count*/ 0);
11713                 ctl_done((union ctl_io *)ctsio);
11714                 break;
11715         }
11716         return (retval);
11717 }
11718
11719 const struct ctl_cmd_entry *
11720 ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa)
11721 {
11722         const struct ctl_cmd_entry *entry;
11723         int service_action;
11724
11725         entry = &ctl_cmd_table[ctsio->cdb[0]];
11726         if (sa)
11727                 *sa = ((entry->flags & CTL_CMD_FLAG_SA5) != 0);
11728         if (entry->flags & CTL_CMD_FLAG_SA5) {
11729                 service_action = ctsio->cdb[1] & SERVICE_ACTION_MASK;
11730                 entry = &((const struct ctl_cmd_entry *)
11731                     entry->execute)[service_action];
11732         }
11733         return (entry);
11734 }
11735
11736 const struct ctl_cmd_entry *
11737 ctl_validate_command(struct ctl_scsiio *ctsio)
11738 {
11739         const struct ctl_cmd_entry *entry;
11740         int i, sa;
11741         uint8_t diff;
11742
11743         entry = ctl_get_cmd_entry(ctsio, &sa);
11744         if (entry->execute == NULL) {
11745                 if (sa)
11746                         ctl_set_invalid_field(ctsio,
11747                                               /*sks_valid*/ 1,
11748                                               /*command*/ 1,
11749                                               /*field*/ 1,
11750                                               /*bit_valid*/ 1,
11751                                               /*bit*/ 4);
11752                 else
11753                         ctl_set_invalid_opcode(ctsio);
11754                 ctl_done((union ctl_io *)ctsio);
11755                 return (NULL);
11756         }
11757         KASSERT(entry->length > 0,
11758             ("Not defined length for command 0x%02x/0x%02x",
11759              ctsio->cdb[0], ctsio->cdb[1]));
11760         for (i = 1; i < entry->length; i++) {
11761                 diff = ctsio->cdb[i] & ~entry->usage[i - 1];
11762                 if (diff == 0)
11763                         continue;
11764                 ctl_set_invalid_field(ctsio,
11765                                       /*sks_valid*/ 1,
11766                                       /*command*/ 1,
11767                                       /*field*/ i,
11768                                       /*bit_valid*/ 1,
11769                                       /*bit*/ fls(diff) - 1);
11770                 ctl_done((union ctl_io *)ctsio);
11771                 return (NULL);
11772         }
11773         return (entry);
11774 }
11775
11776 static int
11777 ctl_cmd_applicable(uint8_t lun_type, const struct ctl_cmd_entry *entry)
11778 {
11779
11780         switch (lun_type) {
11781         case T_DIRECT:
11782                 if ((entry->flags & CTL_CMD_FLAG_OK_ON_DIRECT) == 0)
11783                         return (0);
11784                 break;
11785         case T_PROCESSOR:
11786                 if ((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0)
11787                         return (0);
11788                 break;
11789         case T_CDROM:
11790                 if ((entry->flags & CTL_CMD_FLAG_OK_ON_CDROM) == 0)
11791                         return (0);
11792                 break;
11793         default:
11794                 return (0);
11795         }
11796         return (1);
11797 }
11798
11799 static int
11800 ctl_scsiio(struct ctl_scsiio *ctsio)
11801 {
11802         int retval;
11803         const struct ctl_cmd_entry *entry;
11804
11805         retval = CTL_RETVAL_COMPLETE;
11806
11807         CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
11808
11809         entry = ctl_get_cmd_entry(ctsio, NULL);
11810
11811         /*
11812          * If this I/O has been aborted, just send it straight to
11813          * ctl_done() without executing it.
11814          */
11815         if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
11816                 ctl_done((union ctl_io *)ctsio);
11817                 goto bailout;
11818         }
11819
11820         /*
11821          * All the checks should have been handled by ctl_scsiio_precheck().
11822          * We should be clear now to just execute the I/O.
11823          */
11824         retval = entry->execute(ctsio);
11825
11826 bailout:
11827         return (retval);
11828 }
11829
11830 /*
11831  * Since we only implement one target right now, a bus reset simply resets
11832  * our single target.
11833  */
11834 static int
11835 ctl_bus_reset(struct ctl_softc *softc, union ctl_io *io)
11836 {
11837         return(ctl_target_reset(softc, io, CTL_UA_BUS_RESET));
11838 }
11839
11840 static int
11841 ctl_target_reset(struct ctl_softc *softc, union ctl_io *io,
11842                  ctl_ua_type ua_type)
11843 {
11844         struct ctl_port *port;
11845         struct ctl_lun *lun;
11846         int retval;
11847
11848         if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11849                 union ctl_ha_msg msg_info;
11850
11851                 msg_info.hdr.nexus = io->io_hdr.nexus;
11852                 if (ua_type==CTL_UA_TARG_RESET)
11853                         msg_info.task.task_action = CTL_TASK_TARGET_RESET;
11854                 else
11855                         msg_info.task.task_action = CTL_TASK_BUS_RESET;
11856                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11857                 msg_info.hdr.original_sc = NULL;
11858                 msg_info.hdr.serializing_sc = NULL;
11859                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11860                     sizeof(msg_info.task), M_WAITOK);
11861         }
11862         retval = 0;
11863
11864         mtx_lock(&softc->ctl_lock);
11865         port = ctl_io_port(&io->io_hdr);
11866         STAILQ_FOREACH(lun, &softc->lun_list, links) {
11867                 if (port != NULL &&
11868                     ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
11869                         continue;
11870                 retval += ctl_do_lun_reset(lun, io, ua_type);
11871         }
11872         mtx_unlock(&softc->ctl_lock);
11873         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11874         return (retval);
11875 }
11876
11877 /*
11878  * The LUN should always be set.  The I/O is optional, and is used to
11879  * distinguish between I/Os sent by this initiator, and by other
11880  * initiators.  We set unit attention for initiators other than this one.
11881  * SAM-3 is vague on this point.  It does say that a unit attention should
11882  * be established for other initiators when a LUN is reset (see section
11883  * 5.7.3), but it doesn't specifically say that the unit attention should
11884  * be established for this particular initiator when a LUN is reset.  Here
11885  * is the relevant text, from SAM-3 rev 8:
11886  *
11887  * 5.7.2 When a SCSI initiator port aborts its own tasks
11888  *
11889  * When a SCSI initiator port causes its own task(s) to be aborted, no
11890  * notification that the task(s) have been aborted shall be returned to
11891  * the SCSI initiator port other than the completion response for the
11892  * command or task management function action that caused the task(s) to
11893  * be aborted and notification(s) associated with related effects of the
11894  * action (e.g., a reset unit attention condition).
11895  *
11896  * XXX KDM for now, we're setting unit attention for all initiators.
11897  */
11898 static int
11899 ctl_do_lun_reset(struct ctl_lun *lun, union ctl_io *io, ctl_ua_type ua_type)
11900 {
11901         union ctl_io *xio;
11902 #if 0
11903         uint32_t initidx;
11904 #endif
11905         int i;
11906
11907         mtx_lock(&lun->lun_lock);
11908         /*
11909          * Run through the OOA queue and abort each I/O.
11910          */
11911         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11912              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11913                 xio->io_hdr.flags |= CTL_FLAG_ABORT | CTL_FLAG_ABORT_STATUS;
11914         }
11915
11916         /*
11917          * This version sets unit attention for every
11918          */
11919 #if 0
11920         initidx = ctl_get_initindex(&io->io_hdr.nexus);
11921         ctl_est_ua_all(lun, initidx, ua_type);
11922 #else
11923         ctl_est_ua_all(lun, -1, ua_type);
11924 #endif
11925
11926         /*
11927          * A reset (any kind, really) clears reservations established with
11928          * RESERVE/RELEASE.  It does not clear reservations established
11929          * with PERSISTENT RESERVE OUT, but we don't support that at the
11930          * moment anyway.  See SPC-2, section 5.6.  SPC-3 doesn't address
11931          * reservations made with the RESERVE/RELEASE commands, because
11932          * those commands are obsolete in SPC-3.
11933          */
11934         lun->flags &= ~CTL_LUN_RESERVED;
11935
11936 #ifdef CTL_WITH_CA
11937         for (i = 0; i < CTL_MAX_INITIATORS; i++)
11938                 ctl_clear_mask(lun->have_ca, i);
11939 #endif
11940         lun->prevent_count = 0;
11941         for (i = 0; i < CTL_MAX_INITIATORS; i++)
11942                 ctl_clear_mask(lun->prevent, i);
11943         mtx_unlock(&lun->lun_lock);
11944
11945         return (0);
11946 }
11947
11948 static int
11949 ctl_lun_reset(struct ctl_softc *softc, union ctl_io *io)
11950 {
11951         struct ctl_lun *lun;
11952         uint32_t targ_lun;
11953         int retval;
11954
11955         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11956         mtx_lock(&softc->ctl_lock);
11957         if ((targ_lun >= CTL_MAX_LUNS) ||
11958             (lun = softc->ctl_luns[targ_lun]) == NULL) {
11959                 mtx_unlock(&softc->ctl_lock);
11960                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
11961                 return (1);
11962         }
11963         retval = ctl_do_lun_reset(lun, io, CTL_UA_LUN_RESET);
11964         mtx_unlock(&softc->ctl_lock);
11965         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11966
11967         if ((io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) == 0) {
11968                 union ctl_ha_msg msg_info;
11969
11970                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11971                 msg_info.hdr.nexus = io->io_hdr.nexus;
11972                 msg_info.task.task_action = CTL_TASK_LUN_RESET;
11973                 msg_info.hdr.original_sc = NULL;
11974                 msg_info.hdr.serializing_sc = NULL;
11975                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11976                     sizeof(msg_info.task), M_WAITOK);
11977         }
11978         return (retval);
11979 }
11980
11981 static void
11982 ctl_abort_tasks_lun(struct ctl_lun *lun, uint32_t targ_port, uint32_t init_id,
11983     int other_sc)
11984 {
11985         union ctl_io *xio;
11986
11987         mtx_assert(&lun->lun_lock, MA_OWNED);
11988
11989         /*
11990          * Run through the OOA queue and attempt to find the given I/O.
11991          * The target port, initiator ID, tag type and tag number have to
11992          * match the values that we got from the initiator.  If we have an
11993          * untagged command to abort, simply abort the first untagged command
11994          * we come to.  We only allow one untagged command at a time of course.
11995          */
11996         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11997              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11998
11999                 if ((targ_port == UINT32_MAX ||
12000                      targ_port == xio->io_hdr.nexus.targ_port) &&
12001                     (init_id == UINT32_MAX ||
12002                      init_id == xio->io_hdr.nexus.initid)) {
12003                         if (targ_port != xio->io_hdr.nexus.targ_port ||
12004                             init_id != xio->io_hdr.nexus.initid)
12005                                 xio->io_hdr.flags |= CTL_FLAG_ABORT_STATUS;
12006                         xio->io_hdr.flags |= CTL_FLAG_ABORT;
12007                         if (!other_sc && !(lun->flags & CTL_LUN_PRIMARY_SC)) {
12008                                 union ctl_ha_msg msg_info;
12009
12010                                 msg_info.hdr.nexus = xio->io_hdr.nexus;
12011                                 msg_info.task.task_action = CTL_TASK_ABORT_TASK;
12012                                 msg_info.task.tag_num = xio->scsiio.tag_num;
12013                                 msg_info.task.tag_type = xio->scsiio.tag_type;
12014                                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
12015                                 msg_info.hdr.original_sc = NULL;
12016                                 msg_info.hdr.serializing_sc = NULL;
12017                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
12018                                     sizeof(msg_info.task), M_NOWAIT);
12019                         }
12020                 }
12021         }
12022 }
12023
12024 static int
12025 ctl_abort_task_set(union ctl_io *io)
12026 {
12027         struct ctl_softc *softc = control_softc;
12028         struct ctl_lun *lun;
12029         uint32_t targ_lun;
12030
12031         /*
12032          * Look up the LUN.
12033          */
12034         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12035         mtx_lock(&softc->ctl_lock);
12036         if ((targ_lun >= CTL_MAX_LUNS) ||
12037             (lun = softc->ctl_luns[targ_lun]) == NULL) {
12038                 mtx_unlock(&softc->ctl_lock);
12039                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
12040                 return (1);
12041         }
12042
12043         mtx_lock(&lun->lun_lock);
12044         mtx_unlock(&softc->ctl_lock);
12045         if (io->taskio.task_action == CTL_TASK_ABORT_TASK_SET) {
12046                 ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
12047                     io->io_hdr.nexus.initid,
12048                     (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12049         } else { /* CTL_TASK_CLEAR_TASK_SET */
12050                 ctl_abort_tasks_lun(lun, UINT32_MAX, UINT32_MAX,
12051                     (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12052         }
12053         mtx_unlock(&lun->lun_lock);
12054         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12055         return (0);
12056 }
12057
12058 static int
12059 ctl_i_t_nexus_reset(union ctl_io *io)
12060 {
12061         struct ctl_softc *softc = control_softc;
12062         struct ctl_lun *lun;
12063         uint32_t initidx;
12064
12065         if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
12066                 union ctl_ha_msg msg_info;
12067
12068                 msg_info.hdr.nexus = io->io_hdr.nexus;
12069                 msg_info.task.task_action = CTL_TASK_I_T_NEXUS_RESET;
12070                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
12071                 msg_info.hdr.original_sc = NULL;
12072                 msg_info.hdr.serializing_sc = NULL;
12073                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
12074                     sizeof(msg_info.task), M_WAITOK);
12075         }
12076
12077         initidx = ctl_get_initindex(&io->io_hdr.nexus);
12078         mtx_lock(&softc->ctl_lock);
12079         STAILQ_FOREACH(lun, &softc->lun_list, links) {
12080                 mtx_lock(&lun->lun_lock);
12081                 ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
12082                     io->io_hdr.nexus.initid, 1);
12083 #ifdef CTL_WITH_CA
12084                 ctl_clear_mask(lun->have_ca, initidx);
12085 #endif
12086                 if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == initidx))
12087                         lun->flags &= ~CTL_LUN_RESERVED;
12088                 if (ctl_is_set(lun->prevent, initidx)) {
12089                         ctl_clear_mask(lun->prevent, initidx);
12090                         lun->prevent_count--;
12091                 }
12092                 ctl_est_ua(lun, initidx, CTL_UA_I_T_NEXUS_LOSS);
12093                 mtx_unlock(&lun->lun_lock);
12094         }
12095         mtx_unlock(&softc->ctl_lock);
12096         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12097         return (0);
12098 }
12099
12100 static int
12101 ctl_abort_task(union ctl_io *io)
12102 {
12103         union ctl_io *xio;
12104         struct ctl_lun *lun;
12105         struct ctl_softc *softc;
12106 #if 0
12107         struct sbuf sb;
12108         char printbuf[128];
12109 #endif
12110         int found;
12111         uint32_t targ_lun;
12112
12113         softc = control_softc;
12114         found = 0;
12115
12116         /*
12117          * Look up the LUN.
12118          */
12119         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12120         mtx_lock(&softc->ctl_lock);
12121         if ((targ_lun >= CTL_MAX_LUNS) ||
12122             (lun = softc->ctl_luns[targ_lun]) == NULL) {
12123                 mtx_unlock(&softc->ctl_lock);
12124                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
12125                 return (1);
12126         }
12127
12128 #if 0
12129         printf("ctl_abort_task: called for lun %lld, tag %d type %d\n",
12130                lun->lun, io->taskio.tag_num, io->taskio.tag_type);
12131 #endif
12132
12133         mtx_lock(&lun->lun_lock);
12134         mtx_unlock(&softc->ctl_lock);
12135         /*
12136          * Run through the OOA queue and attempt to find the given I/O.
12137          * The target port, initiator ID, tag type and tag number have to
12138          * match the values that we got from the initiator.  If we have an
12139          * untagged command to abort, simply abort the first untagged command
12140          * we come to.  We only allow one untagged command at a time of course.
12141          */
12142         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12143              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12144 #if 0
12145                 sbuf_new(&sb, printbuf, sizeof(printbuf), SBUF_FIXEDLEN);
12146
12147                 sbuf_printf(&sb, "LUN %lld tag %d type %d%s%s%s%s: ",
12148                             lun->lun, xio->scsiio.tag_num,
12149                             xio->scsiio.tag_type,
12150                             (xio->io_hdr.blocked_links.tqe_prev
12151                             == NULL) ? "" : " BLOCKED",
12152                             (xio->io_hdr.flags &
12153                             CTL_FLAG_DMA_INPROG) ? " DMA" : "",
12154                             (xio->io_hdr.flags &
12155                             CTL_FLAG_ABORT) ? " ABORT" : "",
12156                             (xio->io_hdr.flags &
12157                             CTL_FLAG_IS_WAS_ON_RTR ? " RTR" : ""));
12158                 ctl_scsi_command_string(&xio->scsiio, NULL, &sb);
12159                 sbuf_finish(&sb);
12160                 printf("%s\n", sbuf_data(&sb));
12161 #endif
12162
12163                 if ((xio->io_hdr.nexus.targ_port != io->io_hdr.nexus.targ_port)
12164                  || (xio->io_hdr.nexus.initid != io->io_hdr.nexus.initid)
12165                  || (xio->io_hdr.flags & CTL_FLAG_ABORT))
12166                         continue;
12167
12168                 /*
12169                  * If the abort says that the task is untagged, the
12170                  * task in the queue must be untagged.  Otherwise,
12171                  * we just check to see whether the tag numbers
12172                  * match.  This is because the QLogic firmware
12173                  * doesn't pass back the tag type in an abort
12174                  * request.
12175                  */
12176 #if 0
12177                 if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
12178                   && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
12179                  || (xio->scsiio.tag_num == io->taskio.tag_num))
12180 #endif
12181                 /*
12182                  * XXX KDM we've got problems with FC, because it
12183                  * doesn't send down a tag type with aborts.  So we
12184                  * can only really go by the tag number...
12185                  * This may cause problems with parallel SCSI.
12186                  * Need to figure that out!!
12187                  */
12188                 if (xio->scsiio.tag_num == io->taskio.tag_num) {
12189                         xio->io_hdr.flags |= CTL_FLAG_ABORT;
12190                         found = 1;
12191                         if ((io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) == 0 &&
12192                             !(lun->flags & CTL_LUN_PRIMARY_SC)) {
12193                                 union ctl_ha_msg msg_info;
12194
12195                                 msg_info.hdr.nexus = io->io_hdr.nexus;
12196                                 msg_info.task.task_action = CTL_TASK_ABORT_TASK;
12197                                 msg_info.task.tag_num = io->taskio.tag_num;
12198                                 msg_info.task.tag_type = io->taskio.tag_type;
12199                                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
12200                                 msg_info.hdr.original_sc = NULL;
12201                                 msg_info.hdr.serializing_sc = NULL;
12202 #if 0
12203                                 printf("Sent Abort to other side\n");
12204 #endif
12205                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
12206                                     sizeof(msg_info.task), M_NOWAIT);
12207                         }
12208 #if 0
12209                         printf("ctl_abort_task: found I/O to abort\n");
12210 #endif
12211                 }
12212         }
12213         mtx_unlock(&lun->lun_lock);
12214
12215         if (found == 0) {
12216                 /*
12217                  * This isn't really an error.  It's entirely possible for
12218                  * the abort and command completion to cross on the wire.
12219                  * This is more of an informative/diagnostic error.
12220                  */
12221 #if 0
12222                 printf("ctl_abort_task: ABORT sent for nonexistent I/O: "
12223                        "%u:%u:%u tag %d type %d\n",
12224                        io->io_hdr.nexus.initid,
12225                        io->io_hdr.nexus.targ_port,
12226                        io->io_hdr.nexus.targ_lun, io->taskio.tag_num,
12227                        io->taskio.tag_type);
12228 #endif
12229         }
12230         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12231         return (0);
12232 }
12233
12234 static int
12235 ctl_query_task(union ctl_io *io, int task_set)
12236 {
12237         union ctl_io *xio;
12238         struct ctl_lun *lun;
12239         struct ctl_softc *softc;
12240         int found = 0;
12241         uint32_t targ_lun;
12242
12243         softc = control_softc;
12244         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12245         mtx_lock(&softc->ctl_lock);
12246         if ((targ_lun >= CTL_MAX_LUNS) ||
12247             (lun = softc->ctl_luns[targ_lun]) == NULL) {
12248                 mtx_unlock(&softc->ctl_lock);
12249                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
12250                 return (1);
12251         }
12252         mtx_lock(&lun->lun_lock);
12253         mtx_unlock(&softc->ctl_lock);
12254         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12255              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12256
12257                 if ((xio->io_hdr.nexus.targ_port != io->io_hdr.nexus.targ_port)
12258                  || (xio->io_hdr.nexus.initid != io->io_hdr.nexus.initid)
12259                  || (xio->io_hdr.flags & CTL_FLAG_ABORT))
12260                         continue;
12261
12262                 if (task_set || xio->scsiio.tag_num == io->taskio.tag_num) {
12263                         found = 1;
12264                         break;
12265                 }
12266         }
12267         mtx_unlock(&lun->lun_lock);
12268         if (found)
12269                 io->taskio.task_status = CTL_TASK_FUNCTION_SUCCEEDED;
12270         else
12271                 io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12272         return (0);
12273 }
12274
12275 static int
12276 ctl_query_async_event(union ctl_io *io)
12277 {
12278         struct ctl_lun *lun;
12279         struct ctl_softc *softc;
12280         ctl_ua_type ua;
12281         uint32_t targ_lun, initidx;
12282
12283         softc = control_softc;
12284         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12285         mtx_lock(&softc->ctl_lock);
12286         if ((targ_lun >= CTL_MAX_LUNS) ||
12287             (lun = softc->ctl_luns[targ_lun]) == NULL) {
12288                 mtx_unlock(&softc->ctl_lock);
12289                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
12290                 return (1);
12291         }
12292         mtx_lock(&lun->lun_lock);
12293         mtx_unlock(&softc->ctl_lock);
12294         initidx = ctl_get_initindex(&io->io_hdr.nexus);
12295         ua = ctl_build_qae(lun, initidx, io->taskio.task_resp);
12296         mtx_unlock(&lun->lun_lock);
12297         if (ua != CTL_UA_NONE)
12298                 io->taskio.task_status = CTL_TASK_FUNCTION_SUCCEEDED;
12299         else
12300                 io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12301         return (0);
12302 }
12303
12304 static void
12305 ctl_run_task(union ctl_io *io)
12306 {
12307         struct ctl_softc *softc = control_softc;
12308         int retval = 1;
12309
12310         CTL_DEBUG_PRINT(("ctl_run_task\n"));
12311         KASSERT(io->io_hdr.io_type == CTL_IO_TASK,
12312             ("ctl_run_task: Unextected io_type %d\n", io->io_hdr.io_type));
12313         io->taskio.task_status = CTL_TASK_FUNCTION_NOT_SUPPORTED;
12314         bzero(io->taskio.task_resp, sizeof(io->taskio.task_resp));
12315         switch (io->taskio.task_action) {
12316         case CTL_TASK_ABORT_TASK:
12317                 retval = ctl_abort_task(io);
12318                 break;
12319         case CTL_TASK_ABORT_TASK_SET:
12320         case CTL_TASK_CLEAR_TASK_SET:
12321                 retval = ctl_abort_task_set(io);
12322                 break;
12323         case CTL_TASK_CLEAR_ACA:
12324                 break;
12325         case CTL_TASK_I_T_NEXUS_RESET:
12326                 retval = ctl_i_t_nexus_reset(io);
12327                 break;
12328         case CTL_TASK_LUN_RESET:
12329                 retval = ctl_lun_reset(softc, io);
12330                 break;
12331         case CTL_TASK_TARGET_RESET:
12332                 retval = ctl_target_reset(softc, io, CTL_UA_TARG_RESET);
12333                 break;
12334         case CTL_TASK_BUS_RESET:
12335                 retval = ctl_bus_reset(softc, io);
12336                 break;
12337         case CTL_TASK_PORT_LOGIN:
12338                 break;
12339         case CTL_TASK_PORT_LOGOUT:
12340                 break;
12341         case CTL_TASK_QUERY_TASK:
12342                 retval = ctl_query_task(io, 0);
12343                 break;
12344         case CTL_TASK_QUERY_TASK_SET:
12345                 retval = ctl_query_task(io, 1);
12346                 break;
12347         case CTL_TASK_QUERY_ASYNC_EVENT:
12348                 retval = ctl_query_async_event(io);
12349                 break;
12350         default:
12351                 printf("%s: got unknown task management event %d\n",
12352                        __func__, io->taskio.task_action);
12353                 break;
12354         }
12355         if (retval == 0)
12356                 io->io_hdr.status = CTL_SUCCESS;
12357         else
12358                 io->io_hdr.status = CTL_ERROR;
12359         ctl_done(io);
12360 }
12361
12362 /*
12363  * For HA operation.  Handle commands that come in from the other
12364  * controller.
12365  */
12366 static void
12367 ctl_handle_isc(union ctl_io *io)
12368 {
12369         int free_io;
12370         struct ctl_lun *lun;
12371         struct ctl_softc *softc = control_softc;
12372         uint32_t targ_lun;
12373
12374         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12375         lun = softc->ctl_luns[targ_lun];
12376
12377         switch (io->io_hdr.msg_type) {
12378         case CTL_MSG_SERIALIZE:
12379                 free_io = ctl_serialize_other_sc_cmd(&io->scsiio);
12380                 break;
12381         case CTL_MSG_R2R: {
12382                 const struct ctl_cmd_entry *entry;
12383
12384                 /*
12385                  * This is only used in SER_ONLY mode.
12386                  */
12387                 free_io = 0;
12388                 entry = ctl_get_cmd_entry(&io->scsiio, NULL);
12389                 mtx_lock(&lun->lun_lock);
12390                 if (ctl_scsiio_lun_check(lun,
12391                     entry, (struct ctl_scsiio *)io) != 0) {
12392                         mtx_unlock(&lun->lun_lock);
12393                         ctl_done(io);
12394                         break;
12395                 }
12396                 io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
12397                 mtx_unlock(&lun->lun_lock);
12398                 ctl_enqueue_rtr(io);
12399                 break;
12400         }
12401         case CTL_MSG_FINISH_IO:
12402                 if (softc->ha_mode == CTL_HA_MODE_XFER) {
12403                         free_io = 0;
12404                         ctl_done(io);
12405                 } else {
12406                         free_io = 1;
12407                         mtx_lock(&lun->lun_lock);
12408                         TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr,
12409                                      ooa_links);
12410                         ctl_check_blocked(lun);
12411                         mtx_unlock(&lun->lun_lock);
12412                 }
12413                 break;
12414         case CTL_MSG_PERS_ACTION:
12415                 ctl_hndl_per_res_out_on_other_sc(
12416                         (union ctl_ha_msg *)&io->presio.pr_msg);
12417                 free_io = 1;
12418                 break;
12419         case CTL_MSG_BAD_JUJU:
12420                 free_io = 0;
12421                 ctl_done(io);
12422                 break;
12423         case CTL_MSG_DATAMOVE:
12424                 /* Only used in XFER mode */
12425                 free_io = 0;
12426                 ctl_datamove_remote(io);
12427                 break;
12428         case CTL_MSG_DATAMOVE_DONE:
12429                 /* Only used in XFER mode */
12430                 free_io = 0;
12431                 io->scsiio.be_move_done(io);
12432                 break;
12433         case CTL_MSG_FAILOVER:
12434                 ctl_failover_lun(io);
12435                 free_io = 1;
12436                 break;
12437         default:
12438                 free_io = 1;
12439                 printf("%s: Invalid message type %d\n",
12440                        __func__, io->io_hdr.msg_type);
12441                 break;
12442         }
12443         if (free_io)
12444                 ctl_free_io(io);
12445
12446 }
12447
12448
12449 /*
12450  * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
12451  * there is no match.
12452  */
12453 static ctl_lun_error_pattern
12454 ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
12455 {
12456         const struct ctl_cmd_entry *entry;
12457         ctl_lun_error_pattern filtered_pattern, pattern;
12458
12459         pattern = desc->error_pattern;
12460
12461         /*
12462          * XXX KDM we need more data passed into this function to match a
12463          * custom pattern, and we actually need to implement custom pattern
12464          * matching.
12465          */
12466         if (pattern & CTL_LUN_PAT_CMD)
12467                 return (CTL_LUN_PAT_CMD);
12468
12469         if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
12470                 return (CTL_LUN_PAT_ANY);
12471
12472         entry = ctl_get_cmd_entry(ctsio, NULL);
12473
12474         filtered_pattern = entry->pattern & pattern;
12475
12476         /*
12477          * If the user requested specific flags in the pattern (e.g.
12478          * CTL_LUN_PAT_RANGE), make sure the command supports all of those
12479          * flags.
12480          *
12481          * If the user did not specify any flags, it doesn't matter whether
12482          * or not the command supports the flags.
12483          */
12484         if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
12485              (pattern & ~CTL_LUN_PAT_MASK))
12486                 return (CTL_LUN_PAT_NONE);
12487
12488         /*
12489          * If the user asked for a range check, see if the requested LBA
12490          * range overlaps with this command's LBA range.
12491          */
12492         if (filtered_pattern & CTL_LUN_PAT_RANGE) {
12493                 uint64_t lba1;
12494                 uint64_t len1;
12495                 ctl_action action;
12496                 int retval;
12497
12498                 retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
12499                 if (retval != 0)
12500                         return (CTL_LUN_PAT_NONE);
12501
12502                 action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
12503                                               desc->lba_range.len, FALSE);
12504                 /*
12505                  * A "pass" means that the LBA ranges don't overlap, so
12506                  * this doesn't match the user's range criteria.
12507                  */
12508                 if (action == CTL_ACTION_PASS)
12509                         return (CTL_LUN_PAT_NONE);
12510         }
12511
12512         return (filtered_pattern);
12513 }
12514
12515 static void
12516 ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
12517 {
12518         struct ctl_error_desc *desc, *desc2;
12519
12520         mtx_assert(&lun->lun_lock, MA_OWNED);
12521
12522         STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
12523                 ctl_lun_error_pattern pattern;
12524                 /*
12525                  * Check to see whether this particular command matches
12526                  * the pattern in the descriptor.
12527                  */
12528                 pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
12529                 if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
12530                         continue;
12531
12532                 switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
12533                 case CTL_LUN_INJ_ABORTED:
12534                         ctl_set_aborted(&io->scsiio);
12535                         break;
12536                 case CTL_LUN_INJ_MEDIUM_ERR:
12537                         ctl_set_medium_error(&io->scsiio,
12538                             (io->io_hdr.flags & CTL_FLAG_DATA_MASK) !=
12539                              CTL_FLAG_DATA_OUT);
12540                         break;
12541                 case CTL_LUN_INJ_UA:
12542                         /* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
12543                          * OCCURRED */
12544                         ctl_set_ua(&io->scsiio, 0x29, 0x00);
12545                         break;
12546                 case CTL_LUN_INJ_CUSTOM:
12547                         /*
12548                          * We're assuming the user knows what he is doing.
12549                          * Just copy the sense information without doing
12550                          * checks.
12551                          */
12552                         bcopy(&desc->custom_sense, &io->scsiio.sense_data,
12553                               MIN(sizeof(desc->custom_sense),
12554                                   sizeof(io->scsiio.sense_data)));
12555                         io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
12556                         io->scsiio.sense_len = SSD_FULL_SIZE;
12557                         io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
12558                         break;
12559                 case CTL_LUN_INJ_NONE:
12560                 default:
12561                         /*
12562                          * If this is an error injection type we don't know
12563                          * about, clear the continuous flag (if it is set)
12564                          * so it will get deleted below.
12565                          */
12566                         desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
12567                         break;
12568                 }
12569                 /*
12570                  * By default, each error injection action is a one-shot
12571                  */
12572                 if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
12573                         continue;
12574
12575                 STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
12576
12577                 free(desc, M_CTL);
12578         }
12579 }
12580
12581 #ifdef CTL_IO_DELAY
12582 static void
12583 ctl_datamove_timer_wakeup(void *arg)
12584 {
12585         union ctl_io *io;
12586
12587         io = (union ctl_io *)arg;
12588
12589         ctl_datamove(io);
12590 }
12591 #endif /* CTL_IO_DELAY */
12592
12593 void
12594 ctl_datamove(union ctl_io *io)
12595 {
12596         struct ctl_lun *lun;
12597         void (*fe_datamove)(union ctl_io *io);
12598
12599         mtx_assert(&control_softc->ctl_lock, MA_NOTOWNED);
12600
12601         CTL_DEBUG_PRINT(("ctl_datamove\n"));
12602
12603         lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12604 #ifdef CTL_TIME_IO
12605         if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12606                 char str[256];
12607                 char path_str[64];
12608                 struct sbuf sb;
12609
12610                 ctl_scsi_path_string(io, path_str, sizeof(path_str));
12611                 sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12612
12613                 sbuf_cat(&sb, path_str);
12614                 switch (io->io_hdr.io_type) {
12615                 case CTL_IO_SCSI:
12616                         ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12617                         sbuf_printf(&sb, "\n");
12618                         sbuf_cat(&sb, path_str);
12619                         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12620                                     io->scsiio.tag_num, io->scsiio.tag_type);
12621                         break;
12622                 case CTL_IO_TASK:
12623                         sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12624                                     "Tag Type: %d\n", io->taskio.task_action,
12625                                     io->taskio.tag_num, io->taskio.tag_type);
12626                         break;
12627                 default:
12628                         panic("%s: Invalid CTL I/O type %d\n",
12629                             __func__, io->io_hdr.io_type);
12630                 }
12631                 sbuf_cat(&sb, path_str);
12632                 sbuf_printf(&sb, "ctl_datamove: %jd seconds\n",
12633                             (intmax_t)time_uptime - io->io_hdr.start_time);
12634                 sbuf_finish(&sb);
12635                 printf("%s", sbuf_data(&sb));
12636         }
12637 #endif /* CTL_TIME_IO */
12638
12639 #ifdef CTL_IO_DELAY
12640         if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
12641                 io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
12642         } else {
12643                 if ((lun != NULL)
12644                  && (lun->delay_info.datamove_delay > 0)) {
12645
12646                         callout_init(&io->io_hdr.delay_callout, /*mpsafe*/ 1);
12647                         io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
12648                         callout_reset(&io->io_hdr.delay_callout,
12649                                       lun->delay_info.datamove_delay * hz,
12650                                       ctl_datamove_timer_wakeup, io);
12651                         if (lun->delay_info.datamove_type ==
12652                             CTL_DELAY_TYPE_ONESHOT)
12653                                 lun->delay_info.datamove_delay = 0;
12654                         return;
12655                 }
12656         }
12657 #endif
12658
12659         /*
12660          * This command has been aborted.  Set the port status, so we fail
12661          * the data move.
12662          */
12663         if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12664                 printf("ctl_datamove: tag 0x%04x on (%u:%u:%u) aborted\n",
12665                        io->scsiio.tag_num, io->io_hdr.nexus.initid,
12666                        io->io_hdr.nexus.targ_port,
12667                        io->io_hdr.nexus.targ_lun);
12668                 io->io_hdr.port_status = 31337;
12669                 /*
12670                  * Note that the backend, in this case, will get the
12671                  * callback in its context.  In other cases it may get
12672                  * called in the frontend's interrupt thread context.
12673                  */
12674                 io->scsiio.be_move_done(io);
12675                 return;
12676         }
12677
12678         /* Don't confuse frontend with zero length data move. */
12679         if (io->scsiio.kern_data_len == 0) {
12680                 io->scsiio.be_move_done(io);
12681                 return;
12682         }
12683
12684         fe_datamove = ctl_io_port(&io->io_hdr)->fe_datamove;
12685         fe_datamove(io);
12686 }
12687
12688 static void
12689 ctl_send_datamove_done(union ctl_io *io, int have_lock)
12690 {
12691         union ctl_ha_msg msg;
12692 #ifdef CTL_TIME_IO
12693         struct bintime cur_bt;
12694 #endif
12695
12696         memset(&msg, 0, sizeof(msg));
12697         msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
12698         msg.hdr.original_sc = io;
12699         msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
12700         msg.hdr.nexus = io->io_hdr.nexus;
12701         msg.hdr.status = io->io_hdr.status;
12702         msg.scsi.tag_num = io->scsiio.tag_num;
12703         msg.scsi.tag_type = io->scsiio.tag_type;
12704         msg.scsi.scsi_status = io->scsiio.scsi_status;
12705         memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
12706                io->scsiio.sense_len);
12707         msg.scsi.sense_len = io->scsiio.sense_len;
12708         msg.scsi.sense_residual = io->scsiio.sense_residual;
12709         msg.scsi.fetd_status = io->io_hdr.port_status;
12710         msg.scsi.residual = io->scsiio.residual;
12711         io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12712         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12713                 ctl_failover_io(io, /*have_lock*/ have_lock);
12714                 return;
12715         }
12716         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12717             sizeof(msg.scsi) - sizeof(msg.scsi.sense_data) +
12718             msg.scsi.sense_len, M_WAITOK);
12719
12720 #ifdef CTL_TIME_IO
12721         getbinuptime(&cur_bt);
12722         bintime_sub(&cur_bt, &io->io_hdr.dma_start_bt);
12723         bintime_add(&io->io_hdr.dma_bt, &cur_bt);
12724 #endif
12725         io->io_hdr.num_dmas++;
12726 }
12727
12728 /*
12729  * The DMA to the remote side is done, now we need to tell the other side
12730  * we're done so it can continue with its data movement.
12731  */
12732 static void
12733 ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
12734 {
12735         union ctl_io *io;
12736         uint32_t i;
12737
12738         io = rq->context;
12739
12740         if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12741                 printf("%s: ISC DMA write failed with error %d", __func__,
12742                        rq->ret);
12743                 ctl_set_internal_failure(&io->scsiio,
12744                                          /*sks_valid*/ 1,
12745                                          /*retry_count*/ rq->ret);
12746         }
12747
12748         ctl_dt_req_free(rq);
12749
12750         for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12751                 free(io->io_hdr.local_sglist[i].addr, M_CTL);
12752         free(io->io_hdr.remote_sglist, M_CTL);
12753         io->io_hdr.remote_sglist = NULL;
12754         io->io_hdr.local_sglist = NULL;
12755
12756         /*
12757          * The data is in local and remote memory, so now we need to send
12758          * status (good or back) back to the other side.
12759          */
12760         ctl_send_datamove_done(io, /*have_lock*/ 0);
12761 }
12762
12763 /*
12764  * We've moved the data from the host/controller into local memory.  Now we
12765  * need to push it over to the remote controller's memory.
12766  */
12767 static int
12768 ctl_datamove_remote_dm_write_cb(union ctl_io *io)
12769 {
12770         int retval;
12771
12772         retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
12773                                           ctl_datamove_remote_write_cb);
12774         return (retval);
12775 }
12776
12777 static void
12778 ctl_datamove_remote_write(union ctl_io *io)
12779 {
12780         int retval;
12781         void (*fe_datamove)(union ctl_io *io);
12782
12783         /*
12784          * - Get the data from the host/HBA into local memory.
12785          * - DMA memory from the local controller to the remote controller.
12786          * - Send status back to the remote controller.
12787          */
12788
12789         retval = ctl_datamove_remote_sgl_setup(io);
12790         if (retval != 0)
12791                 return;
12792
12793         /* Switch the pointer over so the FETD knows what to do */
12794         io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12795
12796         /*
12797          * Use a custom move done callback, since we need to send completion
12798          * back to the other controller, not to the backend on this side.
12799          */
12800         io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
12801
12802         fe_datamove = ctl_io_port(&io->io_hdr)->fe_datamove;
12803         fe_datamove(io);
12804 }
12805
12806 static int
12807 ctl_datamove_remote_dm_read_cb(union ctl_io *io)
12808 {
12809 #if 0
12810         char str[256];
12811         char path_str[64];
12812         struct sbuf sb;
12813 #endif
12814         uint32_t i;
12815
12816         for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12817                 free(io->io_hdr.local_sglist[i].addr, M_CTL);
12818         free(io->io_hdr.remote_sglist, M_CTL);
12819         io->io_hdr.remote_sglist = NULL;
12820         io->io_hdr.local_sglist = NULL;
12821
12822 #if 0
12823         scsi_path_string(io, path_str, sizeof(path_str));
12824         sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12825         sbuf_cat(&sb, path_str);
12826         scsi_command_string(&io->scsiio, NULL, &sb);
12827         sbuf_printf(&sb, "\n");
12828         sbuf_cat(&sb, path_str);
12829         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12830                     io->scsiio.tag_num, io->scsiio.tag_type);
12831         sbuf_cat(&sb, path_str);
12832         sbuf_printf(&sb, "%s: flags %#x, status %#x\n", __func__,
12833                     io->io_hdr.flags, io->io_hdr.status);
12834         sbuf_finish(&sb);
12835         printk("%s", sbuf_data(&sb));
12836 #endif
12837
12838
12839         /*
12840          * The read is done, now we need to send status (good or bad) back
12841          * to the other side.
12842          */
12843         ctl_send_datamove_done(io, /*have_lock*/ 0);
12844
12845         return (0);
12846 }
12847
12848 static void
12849 ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
12850 {
12851         union ctl_io *io;
12852         void (*fe_datamove)(union ctl_io *io);
12853
12854         io = rq->context;
12855
12856         if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12857                 printf("%s: ISC DMA read failed with error %d\n", __func__,
12858                        rq->ret);
12859                 ctl_set_internal_failure(&io->scsiio,
12860                                          /*sks_valid*/ 1,
12861                                          /*retry_count*/ rq->ret);
12862         }
12863
12864         ctl_dt_req_free(rq);
12865
12866         /* Switch the pointer over so the FETD knows what to do */
12867         io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12868
12869         /*
12870          * Use a custom move done callback, since we need to send completion
12871          * back to the other controller, not to the backend on this side.
12872          */
12873         io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
12874
12875         /* XXX KDM add checks like the ones in ctl_datamove? */
12876
12877         fe_datamove = ctl_io_port(&io->io_hdr)->fe_datamove;
12878         fe_datamove(io);
12879 }
12880
12881 static int
12882 ctl_datamove_remote_sgl_setup(union ctl_io *io)
12883 {
12884         struct ctl_sg_entry *local_sglist;
12885         uint32_t len_to_go;
12886         int retval;
12887         int i;
12888
12889         retval = 0;
12890         local_sglist = io->io_hdr.local_sglist;
12891         len_to_go = io->scsiio.kern_data_len;
12892
12893         /*
12894          * The difficult thing here is that the size of the various
12895          * S/G segments may be different than the size from the
12896          * remote controller.  That'll make it harder when DMAing
12897          * the data back to the other side.
12898          */
12899         for (i = 0; len_to_go > 0; i++) {
12900                 local_sglist[i].len = MIN(len_to_go, CTL_HA_DATAMOVE_SEGMENT);
12901                 local_sglist[i].addr =
12902                     malloc(local_sglist[i].len, M_CTL, M_WAITOK);
12903
12904                 len_to_go -= local_sglist[i].len;
12905         }
12906         /*
12907          * Reset the number of S/G entries accordingly.  The original
12908          * number of S/G entries is available in rem_sg_entries.
12909          */
12910         io->scsiio.kern_sg_entries = i;
12911
12912 #if 0
12913         printf("%s: kern_sg_entries = %d\n", __func__,
12914                io->scsiio.kern_sg_entries);
12915         for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12916                 printf("%s: sg[%d] = %p, %lu\n", __func__, i,
12917                        local_sglist[i].addr, local_sglist[i].len);
12918 #endif
12919
12920         return (retval);
12921 }
12922
12923 static int
12924 ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
12925                          ctl_ha_dt_cb callback)
12926 {
12927         struct ctl_ha_dt_req *rq;
12928         struct ctl_sg_entry *remote_sglist, *local_sglist;
12929         uint32_t local_used, remote_used, total_used;
12930         int i, j, isc_ret;
12931
12932         rq = ctl_dt_req_alloc();
12933
12934         /*
12935          * If we failed to allocate the request, and if the DMA didn't fail
12936          * anyway, set busy status.  This is just a resource allocation
12937          * failure.
12938          */
12939         if ((rq == NULL)
12940          && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
12941              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS))
12942                 ctl_set_busy(&io->scsiio);
12943
12944         if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
12945             (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS) {
12946
12947                 if (rq != NULL)
12948                         ctl_dt_req_free(rq);
12949
12950                 /*
12951                  * The data move failed.  We need to return status back
12952                  * to the other controller.  No point in trying to DMA
12953                  * data to the remote controller.
12954                  */
12955
12956                 ctl_send_datamove_done(io, /*have_lock*/ 0);
12957
12958                 return (1);
12959         }
12960
12961         local_sglist = io->io_hdr.local_sglist;
12962         remote_sglist = io->io_hdr.remote_sglist;
12963         local_used = 0;
12964         remote_used = 0;
12965         total_used = 0;
12966
12967         /*
12968          * Pull/push the data over the wire from/to the other controller.
12969          * This takes into account the possibility that the local and
12970          * remote sglists may not be identical in terms of the size of
12971          * the elements and the number of elements.
12972          *
12973          * One fundamental assumption here is that the length allocated for
12974          * both the local and remote sglists is identical.  Otherwise, we've
12975          * essentially got a coding error of some sort.
12976          */
12977         isc_ret = CTL_HA_STATUS_SUCCESS;
12978         for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
12979                 uint32_t cur_len;
12980                 uint8_t *tmp_ptr;
12981
12982                 rq->command = command;
12983                 rq->context = io;
12984
12985                 /*
12986                  * Both pointers should be aligned.  But it is possible
12987                  * that the allocation length is not.  They should both
12988                  * also have enough slack left over at the end, though,
12989                  * to round up to the next 8 byte boundary.
12990                  */
12991                 cur_len = MIN(local_sglist[i].len - local_used,
12992                               remote_sglist[j].len - remote_used);
12993                 rq->size = cur_len;
12994
12995                 tmp_ptr = (uint8_t *)local_sglist[i].addr;
12996                 tmp_ptr += local_used;
12997
12998 #if 0
12999                 /* Use physical addresses when talking to ISC hardware */
13000                 if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
13001                         /* XXX KDM use busdma */
13002                         rq->local = vtophys(tmp_ptr);
13003                 } else
13004                         rq->local = tmp_ptr;
13005 #else
13006                 KASSERT((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0,
13007                     ("HA does not support BUS_ADDR"));
13008                 rq->local = tmp_ptr;
13009 #endif
13010
13011                 tmp_ptr = (uint8_t *)remote_sglist[j].addr;
13012                 tmp_ptr += remote_used;
13013                 rq->remote = tmp_ptr;
13014
13015                 rq->callback = NULL;
13016
13017                 local_used += cur_len;
13018                 if (local_used >= local_sglist[i].len) {
13019                         i++;
13020                         local_used = 0;
13021                 }
13022
13023                 remote_used += cur_len;
13024                 if (remote_used >= remote_sglist[j].len) {
13025                         j++;
13026                         remote_used = 0;
13027                 }
13028                 total_used += cur_len;
13029
13030                 if (total_used >= io->scsiio.kern_data_len)
13031                         rq->callback = callback;
13032
13033 #if 0
13034                 printf("%s: %s: local %p remote %p size %d\n", __func__,
13035                        (command == CTL_HA_DT_CMD_WRITE) ? "WRITE" : "READ",
13036                        rq->local, rq->remote, rq->size);
13037 #endif
13038
13039                 isc_ret = ctl_dt_single(rq);
13040                 if (isc_ret > CTL_HA_STATUS_SUCCESS)
13041                         break;
13042         }
13043         if (isc_ret != CTL_HA_STATUS_WAIT) {
13044                 rq->ret = isc_ret;
13045                 callback(rq);
13046         }
13047
13048         return (0);
13049 }
13050
13051 static void
13052 ctl_datamove_remote_read(union ctl_io *io)
13053 {
13054         int retval;
13055         uint32_t i;
13056
13057         /*
13058          * This will send an error to the other controller in the case of a
13059          * failure.
13060          */
13061         retval = ctl_datamove_remote_sgl_setup(io);
13062         if (retval != 0)
13063                 return;
13064
13065         retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
13066                                           ctl_datamove_remote_read_cb);
13067         if (retval != 0) {
13068                 /*
13069                  * Make sure we free memory if there was an error..  The
13070                  * ctl_datamove_remote_xfer() function will send the
13071                  * datamove done message, or call the callback with an
13072                  * error if there is a problem.
13073                  */
13074                 for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13075                         free(io->io_hdr.local_sglist[i].addr, M_CTL);
13076                 free(io->io_hdr.remote_sglist, M_CTL);
13077                 io->io_hdr.remote_sglist = NULL;
13078                 io->io_hdr.local_sglist = NULL;
13079         }
13080 }
13081
13082 /*
13083  * Process a datamove request from the other controller.  This is used for
13084  * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
13085  * first.  Once that is complete, the data gets DMAed into the remote
13086  * controller's memory.  For reads, we DMA from the remote controller's
13087  * memory into our memory first, and then move it out to the FETD.
13088  */
13089 static void
13090 ctl_datamove_remote(union ctl_io *io)
13091 {
13092
13093         mtx_assert(&control_softc->ctl_lock, MA_NOTOWNED);
13094
13095         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
13096                 ctl_failover_io(io, /*have_lock*/ 0);
13097                 return;
13098         }
13099
13100         /*
13101          * Note that we look for an aborted I/O here, but don't do some of
13102          * the other checks that ctl_datamove() normally does.
13103          * We don't need to run the datamove delay code, since that should
13104          * have been done if need be on the other controller.
13105          */
13106         if (io->io_hdr.flags & CTL_FLAG_ABORT) {
13107                 printf("%s: tag 0x%04x on (%u:%u:%u) aborted\n", __func__,
13108                        io->scsiio.tag_num, io->io_hdr.nexus.initid,
13109                        io->io_hdr.nexus.targ_port,
13110                        io->io_hdr.nexus.targ_lun);
13111                 io->io_hdr.port_status = 31338;
13112                 ctl_send_datamove_done(io, /*have_lock*/ 0);
13113                 return;
13114         }
13115
13116         if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT)
13117                 ctl_datamove_remote_write(io);
13118         else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN)
13119                 ctl_datamove_remote_read(io);
13120         else {
13121                 io->io_hdr.port_status = 31339;
13122                 ctl_send_datamove_done(io, /*have_lock*/ 0);
13123         }
13124 }
13125
13126 static void
13127 ctl_process_done(union ctl_io *io)
13128 {
13129         struct ctl_lun *lun;
13130         struct ctl_softc *softc = control_softc;
13131         void (*fe_done)(union ctl_io *io);
13132         union ctl_ha_msg msg;
13133         uint32_t targ_port = io->io_hdr.nexus.targ_port;
13134
13135         CTL_DEBUG_PRINT(("ctl_process_done\n"));
13136         fe_done = softc->ctl_ports[targ_port]->fe_done;
13137
13138 #ifdef CTL_TIME_IO
13139         if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
13140                 char str[256];
13141                 char path_str[64];
13142                 struct sbuf sb;
13143
13144                 ctl_scsi_path_string(io, path_str, sizeof(path_str));
13145                 sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
13146
13147                 sbuf_cat(&sb, path_str);
13148                 switch (io->io_hdr.io_type) {
13149                 case CTL_IO_SCSI:
13150                         ctl_scsi_command_string(&io->scsiio, NULL, &sb);
13151                         sbuf_printf(&sb, "\n");
13152                         sbuf_cat(&sb, path_str);
13153                         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
13154                                     io->scsiio.tag_num, io->scsiio.tag_type);
13155                         break;
13156                 case CTL_IO_TASK:
13157                         sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
13158                                     "Tag Type: %d\n", io->taskio.task_action,
13159                                     io->taskio.tag_num, io->taskio.tag_type);
13160                         break;
13161                 default:
13162                         panic("%s: Invalid CTL I/O type %d\n",
13163                             __func__, io->io_hdr.io_type);
13164                 }
13165                 sbuf_cat(&sb, path_str);
13166                 sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
13167                             (intmax_t)time_uptime - io->io_hdr.start_time);
13168                 sbuf_finish(&sb);
13169                 printf("%s", sbuf_data(&sb));
13170         }
13171 #endif /* CTL_TIME_IO */
13172
13173         switch (io->io_hdr.io_type) {
13174         case CTL_IO_SCSI:
13175                 break;
13176         case CTL_IO_TASK:
13177                 if (ctl_debug & CTL_DEBUG_INFO)
13178                         ctl_io_error_print(io, NULL);
13179                 fe_done(io);
13180                 return;
13181         default:
13182                 panic("%s: Invalid CTL I/O type %d\n",
13183                     __func__, io->io_hdr.io_type);
13184         }
13185
13186         lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13187         if (lun == NULL) {
13188                 CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
13189                                  io->io_hdr.nexus.targ_mapped_lun));
13190                 goto bailout;
13191         }
13192
13193         mtx_lock(&lun->lun_lock);
13194
13195         /*
13196          * Check to see if we have any informational exception and status
13197          * of this command can be modified to report it in form of either
13198          * RECOVERED ERROR or NO SENSE, depending on MRIE mode page field.
13199          */
13200         if (lun->ie_reported == 0 && lun->ie_asc != 0 &&
13201             io->io_hdr.status == CTL_SUCCESS &&
13202             (io->io_hdr.flags & CTL_FLAG_STATUS_SENT) == 0) {
13203                 uint8_t mrie = lun->MODE_IE.mrie;
13204                 uint8_t per = ((lun->MODE_RWER.byte3 & SMS_RWER_PER) ||
13205                     (lun->MODE_VER.byte3 & SMS_VER_PER));
13206                 if (((mrie == SIEP_MRIE_REC_COND && per) ||
13207                      mrie == SIEP_MRIE_REC_UNCOND ||
13208                      mrie == SIEP_MRIE_NO_SENSE) &&
13209                     (ctl_get_cmd_entry(&io->scsiio, NULL)->flags &
13210                      CTL_CMD_FLAG_NO_SENSE) == 0) {
13211                         ctl_set_sense(&io->scsiio,
13212                               /*current_error*/ 1,
13213                               /*sense_key*/ (mrie == SIEP_MRIE_NO_SENSE) ?
13214                                 SSD_KEY_NO_SENSE : SSD_KEY_RECOVERED_ERROR,
13215                               /*asc*/ lun->ie_asc,
13216                               /*ascq*/ lun->ie_ascq,
13217                               SSD_ELEM_NONE);
13218                         lun->ie_reported = 1;
13219                 }
13220         } else if (lun->ie_reported < 0)
13221                 lun->ie_reported = 0;
13222
13223         /*
13224          * Check to see if we have any errors to inject here.  We only
13225          * inject errors for commands that don't already have errors set.
13226          */
13227         if (!STAILQ_EMPTY(&lun->error_list) &&
13228             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS) &&
13229             ((io->io_hdr.flags & CTL_FLAG_STATUS_SENT) == 0))
13230                 ctl_inject_error(lun, io);
13231
13232         /*
13233          * XXX KDM how do we treat commands that aren't completed
13234          * successfully?
13235          *
13236          * XXX KDM should we also track I/O latency?
13237          */
13238         if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS &&
13239             io->io_hdr.io_type == CTL_IO_SCSI) {
13240 #ifdef CTL_TIME_IO
13241                 struct bintime cur_bt;
13242 #endif
13243                 int type;
13244
13245                 if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13246                     CTL_FLAG_DATA_IN)
13247                         type = CTL_STATS_READ;
13248                 else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13249                     CTL_FLAG_DATA_OUT)
13250                         type = CTL_STATS_WRITE;
13251                 else
13252                         type = CTL_STATS_NO_IO;
13253
13254                 lun->stats.ports[targ_port].bytes[type] +=
13255                     io->scsiio.kern_total_len;
13256                 lun->stats.ports[targ_port].operations[type]++;
13257 #ifdef CTL_TIME_IO
13258                 bintime_add(&lun->stats.ports[targ_port].dma_time[type],
13259                    &io->io_hdr.dma_bt);
13260                 getbinuptime(&cur_bt);
13261                 bintime_sub(&cur_bt, &io->io_hdr.start_bt);
13262                 bintime_add(&lun->stats.ports[targ_port].time[type], &cur_bt);
13263 #endif
13264                 lun->stats.ports[targ_port].num_dmas[type] +=
13265                     io->io_hdr.num_dmas;
13266         }
13267
13268         /*
13269          * Remove this from the OOA queue.
13270          */
13271         TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
13272 #ifdef CTL_TIME_IO
13273         if (TAILQ_EMPTY(&lun->ooa_queue))
13274                 lun->last_busy = getsbinuptime();
13275 #endif
13276
13277         /*
13278          * Run through the blocked queue on this LUN and see if anything
13279          * has become unblocked, now that this transaction is done.
13280          */
13281         ctl_check_blocked(lun);
13282
13283         /*
13284          * If the LUN has been invalidated, free it if there is nothing
13285          * left on its OOA queue.
13286          */
13287         if ((lun->flags & CTL_LUN_INVALID)
13288          && TAILQ_EMPTY(&lun->ooa_queue)) {
13289                 mtx_unlock(&lun->lun_lock);
13290                 mtx_lock(&softc->ctl_lock);
13291                 ctl_free_lun(lun);
13292                 mtx_unlock(&softc->ctl_lock);
13293         } else
13294                 mtx_unlock(&lun->lun_lock);
13295
13296 bailout:
13297
13298         /*
13299          * If this command has been aborted, make sure we set the status
13300          * properly.  The FETD is responsible for freeing the I/O and doing
13301          * whatever it needs to do to clean up its state.
13302          */
13303         if (io->io_hdr.flags & CTL_FLAG_ABORT)
13304                 ctl_set_task_aborted(&io->scsiio);
13305
13306         /*
13307          * If enabled, print command error status.
13308          */
13309         if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS &&
13310             (ctl_debug & CTL_DEBUG_INFO) != 0)
13311                 ctl_io_error_print(io, NULL);
13312
13313         /*
13314          * Tell the FETD or the other shelf controller we're done with this
13315          * command.  Note that only SCSI commands get to this point.  Task
13316          * management commands are completed above.
13317          */
13318         if ((softc->ha_mode != CTL_HA_MODE_XFER) &&
13319             (io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)) {
13320                 memset(&msg, 0, sizeof(msg));
13321                 msg.hdr.msg_type = CTL_MSG_FINISH_IO;
13322                 msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
13323                 msg.hdr.nexus = io->io_hdr.nexus;
13324                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13325                     sizeof(msg.scsi) - sizeof(msg.scsi.sense_data),
13326                     M_WAITOK);
13327         }
13328
13329         fe_done(io);
13330 }
13331
13332 #ifdef CTL_WITH_CA
13333 /*
13334  * Front end should call this if it doesn't do autosense.  When the request
13335  * sense comes back in from the initiator, we'll dequeue this and send it.
13336  */
13337 int
13338 ctl_queue_sense(union ctl_io *io)
13339 {
13340         struct ctl_lun *lun;
13341         struct ctl_port *port;
13342         struct ctl_softc *softc;
13343         uint32_t initidx, targ_lun;
13344
13345         softc = control_softc;
13346
13347         CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
13348
13349         /*
13350          * LUN lookup will likely move to the ctl_work_thread() once we
13351          * have our new queueing infrastructure (that doesn't put things on
13352          * a per-LUN queue initially).  That is so that we can handle
13353          * things like an INQUIRY to a LUN that we don't have enabled.  We
13354          * can't deal with that right now.
13355          */
13356         mtx_lock(&softc->ctl_lock);
13357
13358         /*
13359          * If we don't have a LUN for this, just toss the sense
13360          * information.
13361          */
13362         port = ctl_io_port(&ctsio->io_hdr);
13363         targ_lun = ctl_lun_map_from_port(port, io->io_hdr.nexus.targ_lun);
13364         if ((targ_lun < CTL_MAX_LUNS)
13365          && (softc->ctl_luns[targ_lun] != NULL))
13366                 lun = softc->ctl_luns[targ_lun];
13367         else
13368                 goto bailout;
13369
13370         initidx = ctl_get_initindex(&io->io_hdr.nexus);
13371
13372         mtx_lock(&lun->lun_lock);
13373         /*
13374          * Already have CA set for this LUN...toss the sense information.
13375          */
13376         if (ctl_is_set(lun->have_ca, initidx)) {
13377                 mtx_unlock(&lun->lun_lock);
13378                 goto bailout;
13379         }
13380
13381         memcpy(&lun->pending_sense[initidx], &io->scsiio.sense_data,
13382                MIN(sizeof(lun->pending_sense[initidx]),
13383                sizeof(io->scsiio.sense_data)));
13384         ctl_set_mask(lun->have_ca, initidx);
13385         mtx_unlock(&lun->lun_lock);
13386
13387 bailout:
13388         mtx_unlock(&softc->ctl_lock);
13389
13390         ctl_free_io(io);
13391
13392         return (CTL_RETVAL_COMPLETE);
13393 }
13394 #endif
13395
13396 /*
13397  * Primary command inlet from frontend ports.  All SCSI and task I/O
13398  * requests must go through this function.
13399  */
13400 int
13401 ctl_queue(union ctl_io *io)
13402 {
13403         struct ctl_port *port;
13404
13405         CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
13406
13407 #ifdef CTL_TIME_IO
13408         io->io_hdr.start_time = time_uptime;
13409         getbinuptime(&io->io_hdr.start_bt);
13410 #endif /* CTL_TIME_IO */
13411
13412         /* Map FE-specific LUN ID into global one. */
13413         port = ctl_io_port(&io->io_hdr);
13414         io->io_hdr.nexus.targ_mapped_lun =
13415             ctl_lun_map_from_port(port, io->io_hdr.nexus.targ_lun);
13416
13417         switch (io->io_hdr.io_type) {
13418         case CTL_IO_SCSI:
13419         case CTL_IO_TASK:
13420                 if (ctl_debug & CTL_DEBUG_CDB)
13421                         ctl_io_print(io);
13422                 ctl_enqueue_incoming(io);
13423                 break;
13424         default:
13425                 printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
13426                 return (EINVAL);
13427         }
13428
13429         return (CTL_RETVAL_COMPLETE);
13430 }
13431
13432 #ifdef CTL_IO_DELAY
13433 static void
13434 ctl_done_timer_wakeup(void *arg)
13435 {
13436         union ctl_io *io;
13437
13438         io = (union ctl_io *)arg;
13439         ctl_done(io);
13440 }
13441 #endif /* CTL_IO_DELAY */
13442
13443 void
13444 ctl_serseq_done(union ctl_io *io)
13445 {
13446         struct ctl_lun *lun;
13447
13448         lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13449         if (lun->be_lun == NULL ||
13450             lun->be_lun->serseq == CTL_LUN_SERSEQ_OFF)
13451                 return;
13452         mtx_lock(&lun->lun_lock);
13453         io->io_hdr.flags |= CTL_FLAG_SERSEQ_DONE;
13454         ctl_check_blocked(lun);
13455         mtx_unlock(&lun->lun_lock);
13456 }
13457
13458 void
13459 ctl_done(union ctl_io *io)
13460 {
13461
13462         /*
13463          * Enable this to catch duplicate completion issues.
13464          */
13465 #if 0
13466         if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
13467                 printf("%s: type %d msg %d cdb %x iptl: "
13468                        "%u:%u:%u tag 0x%04x "
13469                        "flag %#x status %x\n",
13470                         __func__,
13471                         io->io_hdr.io_type,
13472                         io->io_hdr.msg_type,
13473                         io->scsiio.cdb[0],
13474                         io->io_hdr.nexus.initid,
13475                         io->io_hdr.nexus.targ_port,
13476                         io->io_hdr.nexus.targ_lun,
13477                         (io->io_hdr.io_type ==
13478                         CTL_IO_TASK) ?
13479                         io->taskio.tag_num :
13480                         io->scsiio.tag_num,
13481                         io->io_hdr.flags,
13482                         io->io_hdr.status);
13483         } else
13484                 io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
13485 #endif
13486
13487         /*
13488          * This is an internal copy of an I/O, and should not go through
13489          * the normal done processing logic.
13490          */
13491         if (io->io_hdr.flags & CTL_FLAG_INT_COPY)
13492                 return;
13493
13494 #ifdef CTL_IO_DELAY
13495         if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
13496                 io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
13497         } else {
13498                 struct ctl_lun *lun;
13499
13500                 lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13501
13502                 if ((lun != NULL)
13503                  && (lun->delay_info.done_delay > 0)) {
13504
13505                         callout_init(&io->io_hdr.delay_callout, /*mpsafe*/ 1);
13506                         io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
13507                         callout_reset(&io->io_hdr.delay_callout,
13508                                       lun->delay_info.done_delay * hz,
13509                                       ctl_done_timer_wakeup, io);
13510                         if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
13511                                 lun->delay_info.done_delay = 0;
13512                         return;
13513                 }
13514         }
13515 #endif /* CTL_IO_DELAY */
13516
13517         ctl_enqueue_done(io);
13518 }
13519
13520 static void
13521 ctl_work_thread(void *arg)
13522 {
13523         struct ctl_thread *thr = (struct ctl_thread *)arg;
13524         struct ctl_softc *softc = thr->ctl_softc;
13525         union ctl_io *io;
13526         int retval;
13527
13528         CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
13529
13530         for (;;) {
13531                 /*
13532                  * We handle the queues in this order:
13533                  * - ISC
13534                  * - done queue (to free up resources, unblock other commands)
13535                  * - RtR queue
13536                  * - incoming queue
13537                  *
13538                  * If those queues are empty, we break out of the loop and
13539                  * go to sleep.
13540                  */
13541                 mtx_lock(&thr->queue_lock);
13542                 io = (union ctl_io *)STAILQ_FIRST(&thr->isc_queue);
13543                 if (io != NULL) {
13544                         STAILQ_REMOVE_HEAD(&thr->isc_queue, links);
13545                         mtx_unlock(&thr->queue_lock);
13546                         ctl_handle_isc(io);
13547                         continue;
13548                 }
13549                 io = (union ctl_io *)STAILQ_FIRST(&thr->done_queue);
13550                 if (io != NULL) {
13551                         STAILQ_REMOVE_HEAD(&thr->done_queue, links);
13552                         /* clear any blocked commands, call fe_done */
13553                         mtx_unlock(&thr->queue_lock);
13554                         ctl_process_done(io);
13555                         continue;
13556                 }
13557                 io = (union ctl_io *)STAILQ_FIRST(&thr->incoming_queue);
13558                 if (io != NULL) {
13559                         STAILQ_REMOVE_HEAD(&thr->incoming_queue, links);
13560                         mtx_unlock(&thr->queue_lock);
13561                         if (io->io_hdr.io_type == CTL_IO_TASK)
13562                                 ctl_run_task(io);
13563                         else
13564                                 ctl_scsiio_precheck(softc, &io->scsiio);
13565                         continue;
13566                 }
13567                 io = (union ctl_io *)STAILQ_FIRST(&thr->rtr_queue);
13568                 if (io != NULL) {
13569                         STAILQ_REMOVE_HEAD(&thr->rtr_queue, links);
13570                         mtx_unlock(&thr->queue_lock);
13571                         retval = ctl_scsiio(&io->scsiio);
13572                         if (retval != CTL_RETVAL_COMPLETE)
13573                                 CTL_DEBUG_PRINT(("ctl_scsiio failed\n"));
13574                         continue;
13575                 }
13576
13577                 /* Sleep until we have something to do. */
13578                 mtx_sleep(thr, &thr->queue_lock, PDROP | PRIBIO, "-", 0);
13579         }
13580 }
13581
13582 static void
13583 ctl_lun_thread(void *arg)
13584 {
13585         struct ctl_softc *softc = (struct ctl_softc *)arg;
13586         struct ctl_be_lun *be_lun;
13587
13588         CTL_DEBUG_PRINT(("ctl_lun_thread starting\n"));
13589
13590         for (;;) {
13591                 mtx_lock(&softc->ctl_lock);
13592                 be_lun = STAILQ_FIRST(&softc->pending_lun_queue);
13593                 if (be_lun != NULL) {
13594                         STAILQ_REMOVE_HEAD(&softc->pending_lun_queue, links);
13595                         mtx_unlock(&softc->ctl_lock);
13596                         ctl_create_lun(be_lun);
13597                         continue;
13598                 }
13599
13600                 /* Sleep until we have something to do. */
13601                 mtx_sleep(&softc->pending_lun_queue, &softc->ctl_lock,
13602                     PDROP | PRIBIO, "-", 0);
13603         }
13604 }
13605
13606 static void
13607 ctl_thresh_thread(void *arg)
13608 {
13609         struct ctl_softc *softc = (struct ctl_softc *)arg;
13610         struct ctl_lun *lun;
13611         struct ctl_logical_block_provisioning_page *page;
13612         const char *attr;
13613         union ctl_ha_msg msg;
13614         uint64_t thres, val;
13615         int i, e, set;
13616
13617         CTL_DEBUG_PRINT(("ctl_thresh_thread starting\n"));
13618
13619         for (;;) {
13620                 mtx_lock(&softc->ctl_lock);
13621                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
13622                         if ((lun->flags & CTL_LUN_DISABLED) ||
13623                             (lun->flags & CTL_LUN_NO_MEDIA) ||
13624                             lun->backend->lun_attr == NULL)
13625                                 continue;
13626                         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
13627                             softc->ha_mode == CTL_HA_MODE_XFER)
13628                                 continue;
13629                         if ((lun->MODE_RWER.byte8 & SMS_RWER_LBPERE) == 0)
13630                                 continue;
13631                         e = 0;
13632                         page = &lun->MODE_LBP;
13633                         for (i = 0; i < CTL_NUM_LBP_THRESH; i++) {
13634                                 if ((page->descr[i].flags & SLBPPD_ENABLED) == 0)
13635                                         continue;
13636                                 thres = scsi_4btoul(page->descr[i].count);
13637                                 thres <<= CTL_LBP_EXPONENT;
13638                                 switch (page->descr[i].resource) {
13639                                 case 0x01:
13640                                         attr = "blocksavail";
13641                                         break;
13642                                 case 0x02:
13643                                         attr = "blocksused";
13644                                         break;
13645                                 case 0xf1:
13646                                         attr = "poolblocksavail";
13647                                         break;
13648                                 case 0xf2:
13649                                         attr = "poolblocksused";
13650                                         break;
13651                                 default:
13652                                         continue;
13653                                 }
13654                                 mtx_unlock(&softc->ctl_lock); // XXX
13655                                 val = lun->backend->lun_attr(
13656                                     lun->be_lun->be_lun, attr);
13657                                 mtx_lock(&softc->ctl_lock);
13658                                 if (val == UINT64_MAX)
13659                                         continue;
13660                                 if ((page->descr[i].flags & SLBPPD_ARMING_MASK)
13661                                     == SLBPPD_ARMING_INC)
13662                                         e = (val >= thres);
13663                                 else
13664                                         e = (val <= thres);
13665                                 if (e)
13666                                         break;
13667                         }
13668                         mtx_lock(&lun->lun_lock);
13669                         if (e) {
13670                                 scsi_u64to8b((uint8_t *)&page->descr[i] -
13671                                     (uint8_t *)page, lun->ua_tpt_info);
13672                                 if (lun->lasttpt == 0 ||
13673                                     time_uptime - lun->lasttpt >= CTL_LBP_UA_PERIOD) {
13674                                         lun->lasttpt = time_uptime;
13675                                         ctl_est_ua_all(lun, -1, CTL_UA_THIN_PROV_THRES);
13676                                         set = 1;
13677                                 } else
13678                                         set = 0;
13679                         } else {
13680                                 lun->lasttpt = 0;
13681                                 ctl_clr_ua_all(lun, -1, CTL_UA_THIN_PROV_THRES);
13682                                 set = -1;
13683                         }
13684                         mtx_unlock(&lun->lun_lock);
13685                         if (set != 0 &&
13686                             lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
13687                                 /* Send msg to other side. */
13688                                 bzero(&msg.ua, sizeof(msg.ua));
13689                                 msg.hdr.msg_type = CTL_MSG_UA;
13690                                 msg.hdr.nexus.initid = -1;
13691                                 msg.hdr.nexus.targ_port = -1;
13692                                 msg.hdr.nexus.targ_lun = lun->lun;
13693                                 msg.hdr.nexus.targ_mapped_lun = lun->lun;
13694                                 msg.ua.ua_all = 1;
13695                                 msg.ua.ua_set = (set > 0);
13696                                 msg.ua.ua_type = CTL_UA_THIN_PROV_THRES;
13697                                 memcpy(msg.ua.ua_info, lun->ua_tpt_info, 8);
13698                                 mtx_unlock(&softc->ctl_lock); // XXX
13699                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13700                                     sizeof(msg.ua), M_WAITOK);
13701                                 mtx_lock(&softc->ctl_lock);
13702                         }
13703                 }
13704                 mtx_unlock(&softc->ctl_lock);
13705                 pause("-", CTL_LBP_PERIOD * hz);
13706         }
13707 }
13708
13709 static void
13710 ctl_enqueue_incoming(union ctl_io *io)
13711 {
13712         struct ctl_softc *softc = control_softc;
13713         struct ctl_thread *thr;
13714         u_int idx;
13715
13716         idx = (io->io_hdr.nexus.targ_port * 127 +
13717                io->io_hdr.nexus.initid) % worker_threads;
13718         thr = &softc->threads[idx];
13719         mtx_lock(&thr->queue_lock);
13720         STAILQ_INSERT_TAIL(&thr->incoming_queue, &io->io_hdr, links);
13721         mtx_unlock(&thr->queue_lock);
13722         wakeup(thr);
13723 }
13724
13725 static void
13726 ctl_enqueue_rtr(union ctl_io *io)
13727 {
13728         struct ctl_softc *softc = control_softc;
13729         struct ctl_thread *thr;
13730
13731         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13732         mtx_lock(&thr->queue_lock);
13733         STAILQ_INSERT_TAIL(&thr->rtr_queue, &io->io_hdr, links);
13734         mtx_unlock(&thr->queue_lock);
13735         wakeup(thr);
13736 }
13737
13738 static void
13739 ctl_enqueue_done(union ctl_io *io)
13740 {
13741         struct ctl_softc *softc = control_softc;
13742         struct ctl_thread *thr;
13743
13744         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13745         mtx_lock(&thr->queue_lock);
13746         STAILQ_INSERT_TAIL(&thr->done_queue, &io->io_hdr, links);
13747         mtx_unlock(&thr->queue_lock);
13748         wakeup(thr);
13749 }
13750
13751 static void
13752 ctl_enqueue_isc(union ctl_io *io)
13753 {
13754         struct ctl_softc *softc = control_softc;
13755         struct ctl_thread *thr;
13756
13757         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13758         mtx_lock(&thr->queue_lock);
13759         STAILQ_INSERT_TAIL(&thr->isc_queue, &io->io_hdr, links);
13760         mtx_unlock(&thr->queue_lock);
13761         wakeup(thr);
13762 }
13763
13764 /*
13765  *  vim: ts=8
13766  */