sys/cam/ctl/ctl.c

   1 /*-
   2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
   3  *
   4  * Copyright (c) 2003-2009 Silicon Graphics International Corp.
   5  * Copyright (c) 2012 The FreeBSD Foundation
   6  * Copyright (c) 2014-2017 Alexander Motin <mav@FreeBSD.org>
   7  * Copyright (c) 2017 Jakub Wojciech Klama <jceel@FreeBSD.org>
   8  * Copyright (c) 2018 Marcelo Araujo <araujo@FreeBSD.org>
   9  * All rights reserved.
  10  *
  11  * Portions of this software were developed by Edward Tomasz Napierala
  12  * under sponsorship from the FreeBSD Foundation.
  13  *
  14  * Redistribution and use in source and binary forms, with or without
  15  * modification, are permitted provided that the following conditions
  16  * are met:
  17  * 1. Redistributions of source code must retain the above copyright
  18  *    notice, this list of conditions, and the following disclaimer,
  19  *    without modification.
  20  * 2. Redistributions in binary form must reproduce at minimum a disclaimer
  21  *    substantially similar to the "NO WARRANTY" disclaimer below
  22  *    ("Disclaimer") and any redistribution must be conditioned upon
  23  *    including a substantially similar Disclaimer requirement for further
  24  *    binary redistribution.
  25  *
  26  * NO WARRANTY
  27  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  28  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  29  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR
  30  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  31  * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  32  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  33  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  34  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
  35  * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
  36  * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
  37  * POSSIBILITY OF SUCH DAMAGES.
  38  *
  39  * $Id$
  40  */
  41 /*
  42  * CAM Target Layer, a SCSI device emulation subsystem.
  43  *
  44  * Author: Ken Merry <ken@FreeBSD.org>
  45  */
  46
  47 #include <sys/cdefs.h>
  48 __FBSDID("$FreeBSD$");
  49
  50 #include <sys/param.h>
  51 #include <sys/systm.h>
  52 #include <sys/ctype.h>
  53 #include <sys/kernel.h>
  54 #include <sys/types.h>
  55 #include <sys/kthread.h>
  56 #include <sys/bio.h>
  57 #include <sys/fcntl.h>
  58 #include <sys/lock.h>
  59 #include <sys/module.h>
  60 #include <sys/mutex.h>
  61 #include <sys/condvar.h>
  62 #include <sys/malloc.h>
  63 #include <sys/conf.h>
  64 #include <sys/ioccom.h>
  65 #include <sys/queue.h>
  66 #include <sys/sbuf.h>
  67 #include <sys/smp.h>
  68 #include <sys/endian.h>
  69 #include <sys/sysctl.h>
  70 #include <sys/nv.h>
  71 #include <sys/dnv.h>
  72 #include <vm/uma.h>
  73
  74 #include <cam/cam.h>
  75 #include <cam/scsi/scsi_all.h>
  76 #include <cam/scsi/scsi_cd.h>
  77 #include <cam/scsi/scsi_da.h>
  78 #include <cam/ctl/ctl_io.h>
  79 #include <cam/ctl/ctl.h>
  80 #include <cam/ctl/ctl_frontend.h>
  81 #include <cam/ctl/ctl_util.h>
  82 #include <cam/ctl/ctl_backend.h>
  83 #include <cam/ctl/ctl_ioctl.h>
  84 #include <cam/ctl/ctl_ha.h>
  85 #include <cam/ctl/ctl_private.h>
  86 #include <cam/ctl/ctl_debug.h>
  87 #include <cam/ctl/ctl_scsi_all.h>
  88 #include <cam/ctl/ctl_error.h>
  89
  90 struct ctl_softc *control_softc = NULL;
  91
  92 /*
  93  * Template mode pages.
  94  */
  95
  96 /*
  97  * Note that these are default values only.  The actual values will be
  98  * filled in when the user does a mode sense.
  99  */
 100 const static struct scsi_da_rw_recovery_page rw_er_page_default = {
 101         /*page_code*/SMS_RW_ERROR_RECOVERY_PAGE,
 102         /*page_length*/sizeof(struct scsi_da_rw_recovery_page) - 2,
 103         /*byte3*/SMS_RWER_AWRE|SMS_RWER_ARRE,
 104         /*read_retry_count*/0,
 105         /*correction_span*/0,
 106         /*head_offset_count*/0,
 107         /*data_strobe_offset_cnt*/0,
 108         /*byte8*/SMS_RWER_LBPERE,
 109         /*write_retry_count*/0,
 110         /*reserved2*/0,
 111         /*recovery_time_limit*/{0, 0},
 112 };
 113
 114 const static struct scsi_da_rw_recovery_page rw_er_page_changeable = {
 115         /*page_code*/SMS_RW_ERROR_RECOVERY_PAGE,
 116         /*page_length*/sizeof(struct scsi_da_rw_recovery_page) - 2,
 117         /*byte3*/SMS_RWER_PER,
 118         /*read_retry_count*/0,
 119         /*correction_span*/0,
 120         /*head_offset_count*/0,
 121         /*data_strobe_offset_cnt*/0,
 122         /*byte8*/SMS_RWER_LBPERE,
 123         /*write_retry_count*/0,
 124         /*reserved2*/0,
 125         /*recovery_time_limit*/{0, 0},
 126 };
 127
 128 const static struct scsi_format_page format_page_default = {
 129         /*page_code*/SMS_FORMAT_DEVICE_PAGE,
 130         /*page_length*/sizeof(struct scsi_format_page) - 2,
 131         /*tracks_per_zone*/ {0, 0},
 132         /*alt_sectors_per_zone*/ {0, 0},
 133         /*alt_tracks_per_zone*/ {0, 0},
 134         /*alt_tracks_per_lun*/ {0, 0},
 135         /*sectors_per_track*/ {(CTL_DEFAULT_SECTORS_PER_TRACK >> 8) & 0xff,
 136                                 CTL_DEFAULT_SECTORS_PER_TRACK & 0xff},
 137         /*bytes_per_sector*/ {0, 0},
 138         /*interleave*/ {0, 0},
 139         /*track_skew*/ {0, 0},
 140         /*cylinder_skew*/ {0, 0},
 141         /*flags*/ SFP_HSEC,
 142         /*reserved*/ {0, 0, 0}
 143 };
 144
 145 const static struct scsi_format_page format_page_changeable = {
 146         /*page_code*/SMS_FORMAT_DEVICE_PAGE,
 147         /*page_length*/sizeof(struct scsi_format_page) - 2,
 148         /*tracks_per_zone*/ {0, 0},
 149         /*alt_sectors_per_zone*/ {0, 0},
 150         /*alt_tracks_per_zone*/ {0, 0},
 151         /*alt_tracks_per_lun*/ {0, 0},
 152         /*sectors_per_track*/ {0, 0},
 153         /*bytes_per_sector*/ {0, 0},
 154         /*interleave*/ {0, 0},
 155         /*track_skew*/ {0, 0},
 156         /*cylinder_skew*/ {0, 0},
 157         /*flags*/ 0,
 158         /*reserved*/ {0, 0, 0}
 159 };
 160
 161 const static struct scsi_rigid_disk_page rigid_disk_page_default = {
 162         /*page_code*/SMS_RIGID_DISK_PAGE,
 163         /*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
 164         /*cylinders*/ {0, 0, 0},
 165         /*heads*/ CTL_DEFAULT_HEADS,
 166         /*start_write_precomp*/ {0, 0, 0},
 167         /*start_reduced_current*/ {0, 0, 0},
 168         /*step_rate*/ {0, 0},
 169         /*landing_zone_cylinder*/ {0, 0, 0},
 170         /*rpl*/ SRDP_RPL_DISABLED,
 171         /*rotational_offset*/ 0,
 172         /*reserved1*/ 0,
 173         /*rotation_rate*/ {(CTL_DEFAULT_ROTATION_RATE >> 8) & 0xff,
 174                            CTL_DEFAULT_ROTATION_RATE & 0xff},
 175         /*reserved2*/ {0, 0}
 176 };
 177
 178 const static struct scsi_rigid_disk_page rigid_disk_page_changeable = {
 179         /*page_code*/SMS_RIGID_DISK_PAGE,
 180         /*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
 181         /*cylinders*/ {0, 0, 0},
 182         /*heads*/ 0,
 183         /*start_write_precomp*/ {0, 0, 0},
 184         /*start_reduced_current*/ {0, 0, 0},
 185         /*step_rate*/ {0, 0},
 186         /*landing_zone_cylinder*/ {0, 0, 0},
 187         /*rpl*/ 0,
 188         /*rotational_offset*/ 0,
 189         /*reserved1*/ 0,
 190         /*rotation_rate*/ {0, 0},
 191         /*reserved2*/ {0, 0}
 192 };
 193
 194 const static struct scsi_da_verify_recovery_page verify_er_page_default = {
 195         /*page_code*/SMS_VERIFY_ERROR_RECOVERY_PAGE,
 196         /*page_length*/sizeof(struct scsi_da_verify_recovery_page) - 2,
 197         /*byte3*/0,
 198         /*read_retry_count*/0,
 199         /*reserved*/{ 0, 0, 0, 0, 0, 0 },
 200         /*recovery_time_limit*/{0, 0},
 201 };
 202
 203 const static struct scsi_da_verify_recovery_page verify_er_page_changeable = {
 204         /*page_code*/SMS_VERIFY_ERROR_RECOVERY_PAGE,
 205         /*page_length*/sizeof(struct scsi_da_verify_recovery_page) - 2,
 206         /*byte3*/SMS_VER_PER,
 207         /*read_retry_count*/0,
 208         /*reserved*/{ 0, 0, 0, 0, 0, 0 },
 209         /*recovery_time_limit*/{0, 0},
 210 };
 211
 212 const static struct scsi_caching_page caching_page_default = {
 213         /*page_code*/SMS_CACHING_PAGE,
 214         /*page_length*/sizeof(struct scsi_caching_page) - 2,
 215         /*flags1*/ SCP_DISC | SCP_WCE,
 216         /*ret_priority*/ 0,
 217         /*disable_pf_transfer_len*/ {0xff, 0xff},
 218         /*min_prefetch*/ {0, 0},
 219         /*max_prefetch*/ {0xff, 0xff},
 220         /*max_pf_ceiling*/ {0xff, 0xff},
 221         /*flags2*/ 0,
 222         /*cache_segments*/ 0,
 223         /*cache_seg_size*/ {0, 0},
 224         /*reserved*/ 0,
 225         /*non_cache_seg_size*/ {0, 0, 0}
 226 };
 227
 228 const static struct scsi_caching_page caching_page_changeable = {
 229         /*page_code*/SMS_CACHING_PAGE,
 230         /*page_length*/sizeof(struct scsi_caching_page) - 2,
 231         /*flags1*/ SCP_WCE | SCP_RCD,
 232         /*ret_priority*/ 0,
 233         /*disable_pf_transfer_len*/ {0, 0},
 234         /*min_prefetch*/ {0, 0},
 235         /*max_prefetch*/ {0, 0},
 236         /*max_pf_ceiling*/ {0, 0},
 237         /*flags2*/ 0,
 238         /*cache_segments*/ 0,
 239         /*cache_seg_size*/ {0, 0},
 240         /*reserved*/ 0,
 241         /*non_cache_seg_size*/ {0, 0, 0}
 242 };
 243
 244 const static struct scsi_control_page control_page_default = {
 245         /*page_code*/SMS_CONTROL_MODE_PAGE,
 246         /*page_length*/sizeof(struct scsi_control_page) - 2,
 247         /*rlec*/0,
 248         /*queue_flags*/SCP_QUEUE_ALG_RESTRICTED,
 249         /*eca_and_aen*/0,
 250         /*flags4*/SCP_TAS,
 251         /*aen_holdoff_period*/{0, 0},
 252         /*busy_timeout_period*/{0, 0},
 253         /*extended_selftest_completion_time*/{0, 0}
 254 };
 255
 256 const static struct scsi_control_page control_page_changeable = {
 257         /*page_code*/SMS_CONTROL_MODE_PAGE,
 258         /*page_length*/sizeof(struct scsi_control_page) - 2,
 259         /*rlec*/SCP_DSENSE,
 260         /*queue_flags*/SCP_QUEUE_ALG_MASK | SCP_NUAR,
 261         /*eca_and_aen*/SCP_SWP,
 262         /*flags4*/0,
 263         /*aen_holdoff_period*/{0, 0},
 264         /*busy_timeout_period*/{0, 0},
 265         /*extended_selftest_completion_time*/{0, 0}
 266 };
 267
 268 #define CTL_CEM_LEN     (sizeof(struct scsi_control_ext_page) - 4)
 269
 270 const static struct scsi_control_ext_page control_ext_page_default = {
 271         /*page_code*/SMS_CONTROL_MODE_PAGE | SMPH_SPF,
 272         /*subpage_code*/0x01,
 273         /*page_length*/{CTL_CEM_LEN >> 8, CTL_CEM_LEN},
 274         /*flags*/0,
 275         /*prio*/0,
 276         /*max_sense*/0
 277 };
 278
 279 const static struct scsi_control_ext_page control_ext_page_changeable = {
 280         /*page_code*/SMS_CONTROL_MODE_PAGE | SMPH_SPF,
 281         /*subpage_code*/0x01,
 282         /*page_length*/{CTL_CEM_LEN >> 8, CTL_CEM_LEN},
 283         /*flags*/0,
 284         /*prio*/0,
 285         /*max_sense*/0xff
 286 };
 287
 288 const static struct scsi_info_exceptions_page ie_page_default = {
 289         /*page_code*/SMS_INFO_EXCEPTIONS_PAGE,
 290         /*page_length*/sizeof(struct scsi_info_exceptions_page) - 2,
 291         /*info_flags*/SIEP_FLAGS_EWASC,
 292         /*mrie*/SIEP_MRIE_NO,
 293         /*interval_timer*/{0, 0, 0, 0},
 294         /*report_count*/{0, 0, 0, 1}
 295 };
 296
 297 const static struct scsi_info_exceptions_page ie_page_changeable = {
 298         /*page_code*/SMS_INFO_EXCEPTIONS_PAGE,
 299         /*page_length*/sizeof(struct scsi_info_exceptions_page) - 2,
 300         /*info_flags*/SIEP_FLAGS_EWASC | SIEP_FLAGS_DEXCPT | SIEP_FLAGS_TEST |
 301             SIEP_FLAGS_LOGERR,
 302         /*mrie*/0x0f,
 303         /*interval_timer*/{0xff, 0xff, 0xff, 0xff},
 304         /*report_count*/{0xff, 0xff, 0xff, 0xff}
 305 };
 306
 307 #define CTL_LBPM_LEN    (sizeof(struct ctl_logical_block_provisioning_page) - 4)
 308
 309 const static struct ctl_logical_block_provisioning_page lbp_page_default = {{
 310         /*page_code*/SMS_INFO_EXCEPTIONS_PAGE | SMPH_SPF,
 311         /*subpage_code*/0x02,
 312         /*page_length*/{CTL_LBPM_LEN >> 8, CTL_LBPM_LEN},
 313         /*flags*/0,
 314         /*reserved*/{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
 315         /*descr*/{}},
 316         {{/*flags*/0,
 317           /*resource*/0x01,
 318           /*reserved*/{0, 0},
 319           /*count*/{0, 0, 0, 0}},
 320          {/*flags*/0,
 321           /*resource*/0x02,
 322           /*reserved*/{0, 0},
 323           /*count*/{0, 0, 0, 0}},
 324          {/*flags*/0,
 325           /*resource*/0xf1,
 326           /*reserved*/{0, 0},
 327           /*count*/{0, 0, 0, 0}},
 328          {/*flags*/0,
 329           /*resource*/0xf2,
 330           /*reserved*/{0, 0},
 331           /*count*/{0, 0, 0, 0}}
 332         }
 333 };
 334
 335 const static struct ctl_logical_block_provisioning_page lbp_page_changeable = {{
 336         /*page_code*/SMS_INFO_EXCEPTIONS_PAGE | SMPH_SPF,
 337         /*subpage_code*/0x02,
 338         /*page_length*/{CTL_LBPM_LEN >> 8, CTL_LBPM_LEN},
 339         /*flags*/SLBPP_SITUA,
 340         /*reserved*/{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
 341         /*descr*/{}},
 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          {/*flags*/0,
 355           /*resource*/0,
 356           /*reserved*/{0, 0},
 357           /*count*/{0, 0, 0, 0}}
 358         }
 359 };
 360
 361 const static struct scsi_cddvd_capabilities_page cddvd_page_default = {
 362         /*page_code*/SMS_CDDVD_CAPS_PAGE,
 363         /*page_length*/sizeof(struct scsi_cddvd_capabilities_page) - 2,
 364         /*caps1*/0x3f,
 365         /*caps2*/0x00,
 366         /*caps3*/0xf0,
 367         /*caps4*/0x00,
 368         /*caps5*/0x29,
 369         /*caps6*/0x00,
 370         /*obsolete*/{0, 0},
 371         /*nvol_levels*/{0, 0},
 372         /*buffer_size*/{8, 0},
 373         /*obsolete2*/{0, 0},
 374         /*reserved*/0,
 375         /*digital*/0,
 376         /*obsolete3*/0,
 377         /*copy_management*/0,
 378         /*reserved2*/0,
 379         /*rotation_control*/0,
 380         /*cur_write_speed*/0,
 381         /*num_speed_descr*/0,
 382 };
 383
 384 const static struct scsi_cddvd_capabilities_page cddvd_page_changeable = {
 385         /*page_code*/SMS_CDDVD_CAPS_PAGE,
 386         /*page_length*/sizeof(struct scsi_cddvd_capabilities_page) - 2,
 387         /*caps1*/0,
 388         /*caps2*/0,
 389         /*caps3*/0,
 390         /*caps4*/0,
 391         /*caps5*/0,
 392         /*caps6*/0,
 393         /*obsolete*/{0, 0},
 394         /*nvol_levels*/{0, 0},
 395         /*buffer_size*/{0, 0},
 396         /*obsolete2*/{0, 0},
 397         /*reserved*/0,
 398         /*digital*/0,
 399         /*obsolete3*/0,
 400         /*copy_management*/0,
 401         /*reserved2*/0,
 402         /*rotation_control*/0,
 403         /*cur_write_speed*/0,
 404         /*num_speed_descr*/0,
 405 };
 406
 407 SYSCTL_NODE(_kern_cam, OID_AUTO, ctl, CTLFLAG_RD, 0, "CAM Target Layer");
 408 static int worker_threads = -1;
 409 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, worker_threads, CTLFLAG_RDTUN,
 410     &worker_threads, 1, "Number of worker threads");
 411 static int ctl_debug = CTL_DEBUG_NONE;
 412 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, debug, CTLFLAG_RWTUN,
 413     &ctl_debug, 0, "Enabled debug flags");
 414 static int ctl_lun_map_size = 1024;
 415 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, lun_map_size, CTLFLAG_RWTUN,
 416     &ctl_lun_map_size, 0, "Size of per-port LUN map (max LUN + 1)");
 417 #ifdef  CTL_TIME_IO
 418 static int ctl_time_io_secs = CTL_TIME_IO_DEFAULT_SECS;
 419 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, time_io_secs, CTLFLAG_RWTUN,
 420     &ctl_time_io_secs, 0, "Log requests taking more seconds");
 421 #endif
 422
 423 /*
 424  * Maximum number of LUNs we support.  MUST be a power of 2.
 425  */
 426 #define CTL_DEFAULT_MAX_LUNS    1024
 427 static int ctl_max_luns = CTL_DEFAULT_MAX_LUNS;
 428 TUNABLE_INT("kern.cam.ctl.max_luns", &ctl_max_luns);
 429 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, max_luns, CTLFLAG_RDTUN,
 430     &ctl_max_luns, CTL_DEFAULT_MAX_LUNS, "Maximum number of LUNs");
 431
 432 /*
 433  * Maximum number of ports registered at one time.
 434  */
 435 #define CTL_DEFAULT_MAX_PORTS           256
 436 static int ctl_max_ports = CTL_DEFAULT_MAX_PORTS;
 437 TUNABLE_INT("kern.cam.ctl.max_ports", &ctl_max_ports);
 438 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, max_ports, CTLFLAG_RDTUN,
 439     &ctl_max_ports, CTL_DEFAULT_MAX_LUNS, "Maximum number of ports");
 440
 441 /*
 442  * Maximum number of initiators we support.
 443  */
 444 #define CTL_MAX_INITIATORS      (CTL_MAX_INIT_PER_PORT * ctl_max_ports)
 445
 446 /*
 447  * Supported pages (0x00), Serial number (0x80), Device ID (0x83),
 448  * Extended INQUIRY Data (0x86), Mode Page Policy (0x87),
 449  * SCSI Ports (0x88), Third-party Copy (0x8F), Block limits (0xB0),
 450  * Block Device Characteristics (0xB1) and Logical Block Provisioning (0xB2)
 451  */
 452 #define SCSI_EVPD_NUM_SUPPORTED_PAGES   10
 453
 454 static void ctl_isc_event_handler(ctl_ha_channel chanel, ctl_ha_event event,
 455                                   int param);
 456 static void ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest);
 457 static void ctl_copy_sense_data_back(union ctl_io *src, union ctl_ha_msg *dest);
 458 static int ctl_init(void);
 459 static int ctl_shutdown(void);
 460 static int ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td);
 461 static int ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td);
 462 static void ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio);
 463 static void ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
 464                               struct ctl_ooa *ooa_hdr,
 465                               struct ctl_ooa_entry *kern_entries);
 466 static int ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
 467                      struct thread *td);
 468 static int ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
 469                          struct ctl_be_lun *be_lun);
 470 static int ctl_free_lun(struct ctl_lun *lun);
 471 static void ctl_create_lun(struct ctl_be_lun *be_lun);
 472
 473 static int ctl_do_mode_select(union ctl_io *io);
 474 static int ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun,
 475                            uint64_t res_key, uint64_t sa_res_key,
 476                            uint8_t type, uint32_t residx,
 477                            struct ctl_scsiio *ctsio,
 478                            struct scsi_per_res_out *cdb,
 479                            struct scsi_per_res_out_parms* param);
 480 static void ctl_pro_preempt_other(struct ctl_lun *lun,
 481                                   union ctl_ha_msg *msg);
 482 static void ctl_hndl_per_res_out_on_other_sc(union ctl_io *io);
 483 static int ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len);
 484 static int ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len);
 485 static int ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len);
 486 static int ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len);
 487 static int ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len);
 488 static int ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio,
 489                                          int alloc_len);
 490 static int ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio,
 491                                          int alloc_len);
 492 static int ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len);
 493 static int ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len);
 494 static int ctl_inquiry_evpd(struct ctl_scsiio *ctsio);
 495 static int ctl_inquiry_std(struct ctl_scsiio *ctsio);
 496 static int ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len);
 497 static ctl_action ctl_extent_check(union ctl_io *io1, union ctl_io *io2,
 498     bool seq);
 499 static ctl_action ctl_extent_check_seq(union ctl_io *io1, union ctl_io *io2);
 500 static ctl_action ctl_check_for_blockage(struct ctl_lun *lun,
 501     union ctl_io *pending_io, union ctl_io *ooa_io);
 502 static ctl_action ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
 503                                 union ctl_io *starting_io);
 504 static int ctl_check_blocked(struct ctl_lun *lun);
 505 static int ctl_scsiio_lun_check(struct ctl_lun *lun,
 506                                 const struct ctl_cmd_entry *entry,
 507                                 struct ctl_scsiio *ctsio);
 508 static void ctl_failover_lun(union ctl_io *io);
 509 static int ctl_scsiio_precheck(struct ctl_softc *ctl_softc,
 510                                struct ctl_scsiio *ctsio);
 511 static int ctl_scsiio(struct ctl_scsiio *ctsio);
 512
 513 static int ctl_target_reset(union ctl_io *io);
 514 static void ctl_do_lun_reset(struct ctl_lun *lun, uint32_t initidx,
 515                          ctl_ua_type ua_type);
 516 static int ctl_lun_reset(union ctl_io *io);
 517 static int ctl_abort_task(union ctl_io *io);
 518 static int ctl_abort_task_set(union ctl_io *io);
 519 static int ctl_query_task(union ctl_io *io, int task_set);
 520 static void ctl_i_t_nexus_loss(struct ctl_softc *softc, uint32_t initidx,
 521                               ctl_ua_type ua_type);
 522 static int ctl_i_t_nexus_reset(union ctl_io *io);
 523 static int ctl_query_async_event(union ctl_io *io);
 524 static void ctl_run_task(union ctl_io *io);
 525 #ifdef CTL_IO_DELAY
 526 static void ctl_datamove_timer_wakeup(void *arg);
 527 static void ctl_done_timer_wakeup(void *arg);
 528 #endif /* CTL_IO_DELAY */
 529
 530 static void ctl_send_datamove_done(union ctl_io *io, int have_lock);
 531 static void ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq);
 532 static int ctl_datamove_remote_dm_write_cb(union ctl_io *io);
 533 static void ctl_datamove_remote_write(union ctl_io *io);
 534 static int ctl_datamove_remote_dm_read_cb(union ctl_io *io);
 535 static void ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq);
 536 static int ctl_datamove_remote_sgl_setup(union ctl_io *io);
 537 static int ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
 538                                     ctl_ha_dt_cb callback);
 539 static void ctl_datamove_remote_read(union ctl_io *io);
 540 static void ctl_datamove_remote(union ctl_io *io);
 541 static void ctl_process_done(union ctl_io *io);
 542 static void ctl_lun_thread(void *arg);
 543 static void ctl_thresh_thread(void *arg);
 544 static void ctl_work_thread(void *arg);
 545 static void ctl_enqueue_incoming(union ctl_io *io);
 546 static void ctl_enqueue_rtr(union ctl_io *io);
 547 static void ctl_enqueue_done(union ctl_io *io);
 548 static void ctl_enqueue_isc(union ctl_io *io);
 549 static const struct ctl_cmd_entry *
 550     ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa);
 551 static const struct ctl_cmd_entry *
 552     ctl_validate_command(struct ctl_scsiio *ctsio);
 553 static int ctl_cmd_applicable(uint8_t lun_type,
 554     const struct ctl_cmd_entry *entry);
 555 static int ctl_ha_init(void);
 556 static int ctl_ha_shutdown(void);
 557
 558 static uint64_t ctl_get_prkey(struct ctl_lun *lun, uint32_t residx);
 559 static void ctl_clr_prkey(struct ctl_lun *lun, uint32_t residx);
 560 static void ctl_alloc_prkey(struct ctl_lun *lun, uint32_t residx);
 561 static void ctl_set_prkey(struct ctl_lun *lun, uint32_t residx, uint64_t key);
 562
 563 /*
 564  * Load the serialization table.  This isn't very pretty, but is probably
 565  * the easiest way to do it.
 566  */
 567 #include "ctl_ser_table.c"
 568
 569 /*
 570  * We only need to define open, close and ioctl routines for this driver.
 571  */
 572 static struct cdevsw ctl_cdevsw = {
 573         .d_version =    D_VERSION,
 574         .d_flags =      0,
 575         .d_open =       ctl_open,
 576         .d_close =      ctl_close,
 577         .d_ioctl =      ctl_ioctl,
 578         .d_name =       "ctl",
 579 };
 580
 581
 582 MALLOC_DEFINE(M_CTL, "ctlmem", "Memory used for CTL");
 583
 584 static int ctl_module_event_handler(module_t, int /*modeventtype_t*/, void *);
 585
 586 static moduledata_t ctl_moduledata = {
 587         "ctl",
 588         ctl_module_event_handler,
 589         NULL
 590 };
 591
 592 DECLARE_MODULE(ctl, ctl_moduledata, SI_SUB_CONFIGURE, SI_ORDER_THIRD);
 593 MODULE_VERSION(ctl, 1);
 594
 595 static struct ctl_frontend ha_frontend =
 596 {
 597         .name = "ha",
 598         .init = ctl_ha_init,
 599         .shutdown = ctl_ha_shutdown,
 600 };
 601
 602 static int
 603 ctl_ha_init(void)
 604 {
 605         struct ctl_softc *softc = control_softc;
 606
 607         if (ctl_pool_create(softc, "othersc", CTL_POOL_ENTRIES_OTHER_SC,
 608                             &softc->othersc_pool) != 0)
 609                 return (ENOMEM);
 610         if (ctl_ha_msg_init(softc) != CTL_HA_STATUS_SUCCESS) {
 611                 ctl_pool_free(softc->othersc_pool);
 612                 return (EIO);
 613         }
 614         if (ctl_ha_msg_register(CTL_HA_CHAN_CTL, ctl_isc_event_handler)
 615             != CTL_HA_STATUS_SUCCESS) {
 616                 ctl_ha_msg_destroy(softc);
 617                 ctl_pool_free(softc->othersc_pool);
 618                 return (EIO);
 619         }
 620         return (0);
 621 };
 622
 623 static int
 624 ctl_ha_shutdown(void)
 625 {
 626         struct ctl_softc *softc = control_softc;
 627         struct ctl_port *port;
 628
 629         ctl_ha_msg_shutdown(softc);
 630         if (ctl_ha_msg_deregister(CTL_HA_CHAN_CTL) != CTL_HA_STATUS_SUCCESS)
 631                 return (EIO);
 632         if (ctl_ha_msg_destroy(softc) != CTL_HA_STATUS_SUCCESS)
 633                 return (EIO);
 634         ctl_pool_free(softc->othersc_pool);
 635         while ((port = STAILQ_FIRST(&ha_frontend.port_list)) != NULL) {
 636                 ctl_port_deregister(port);
 637                 free(port->port_name, M_CTL);
 638                 free(port, M_CTL);
 639         }
 640         return (0);
 641 };
 642
 643 static void
 644 ctl_ha_datamove(union ctl_io *io)
 645 {
 646         struct ctl_lun *lun = CTL_LUN(io);
 647         struct ctl_sg_entry *sgl;
 648         union ctl_ha_msg msg;
 649         uint32_t sg_entries_sent;
 650         int do_sg_copy, i, j;
 651
 652         memset(&msg.dt, 0, sizeof(msg.dt));
 653         msg.hdr.msg_type = CTL_MSG_DATAMOVE;
 654         msg.hdr.original_sc = io->io_hdr.remote_io;
 655         msg.hdr.serializing_sc = io;
 656         msg.hdr.nexus = io->io_hdr.nexus;
 657         msg.hdr.status = io->io_hdr.status;
 658         msg.dt.flags = io->io_hdr.flags;
 659
 660         /*
 661          * We convert everything into a S/G list here.  We can't
 662          * pass by reference, only by value between controllers.
 663          * So we can't pass a pointer to the S/G list, only as many
 664          * S/G entries as we can fit in here.  If it's possible for
 665          * us to get more than CTL_HA_MAX_SG_ENTRIES S/G entries,
 666          * then we need to break this up into multiple transfers.
 667          */
 668         if (io->scsiio.kern_sg_entries == 0) {
 669                 msg.dt.kern_sg_entries = 1;
 670 #if 0
 671                 if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
 672                         msg.dt.sg_list[0].addr = io->scsiio.kern_data_ptr;
 673                 } else {
 674                         /* XXX KDM use busdma here! */
 675                         msg.dt.sg_list[0].addr =
 676                             (void *)vtophys(io->scsiio.kern_data_ptr);
 677                 }
 678 #else
 679                 KASSERT((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0,
 680                     ("HA does not support BUS_ADDR"));
 681                 msg.dt.sg_list[0].addr = io->scsiio.kern_data_ptr;
 682 #endif
 683                 msg.dt.sg_list[0].len = io->scsiio.kern_data_len;
 684                 do_sg_copy = 0;
 685         } else {
 686                 msg.dt.kern_sg_entries = io->scsiio.kern_sg_entries;
 687                 do_sg_copy = 1;
 688         }
 689
 690         msg.dt.kern_data_len = io->scsiio.kern_data_len;
 691         msg.dt.kern_total_len = io->scsiio.kern_total_len;
 692         msg.dt.kern_data_resid = io->scsiio.kern_data_resid;
 693         msg.dt.kern_rel_offset = io->scsiio.kern_rel_offset;
 694         msg.dt.sg_sequence = 0;
 695
 696         /*
 697          * Loop until we've sent all of the S/G entries.  On the
 698          * other end, we'll recompose these S/G entries into one
 699          * contiguous list before processing.
 700          */
 701         for (sg_entries_sent = 0; sg_entries_sent < msg.dt.kern_sg_entries;
 702             msg.dt.sg_sequence++) {
 703                 msg.dt.cur_sg_entries = MIN((sizeof(msg.dt.sg_list) /
 704                     sizeof(msg.dt.sg_list[0])),
 705                     msg.dt.kern_sg_entries - sg_entries_sent);
 706                 if (do_sg_copy != 0) {
 707                         sgl = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr;
 708                         for (i = sg_entries_sent, j = 0;
 709                              i < msg.dt.cur_sg_entries; i++, j++) {
 710 #if 0
 711                                 if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
 712                                         msg.dt.sg_list[j].addr = sgl[i].addr;
 713                                 } else {
 714                                         /* XXX KDM use busdma here! */
 715                                         msg.dt.sg_list[j].addr =
 716                                             (void *)vtophys(sgl[i].addr);
 717                                 }
 718 #else
 719                                 KASSERT((io->io_hdr.flags &
 720                                     CTL_FLAG_BUS_ADDR) == 0,
 721                                     ("HA does not support BUS_ADDR"));
 722                                 msg.dt.sg_list[j].addr = sgl[i].addr;
 723 #endif
 724                                 msg.dt.sg_list[j].len = sgl[i].len;
 725                         }
 726                 }
 727
 728                 sg_entries_sent += msg.dt.cur_sg_entries;
 729                 msg.dt.sg_last = (sg_entries_sent >= msg.dt.kern_sg_entries);
 730                 if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
 731                     sizeof(msg.dt) - sizeof(msg.dt.sg_list) +
 732                     sizeof(struct ctl_sg_entry) * msg.dt.cur_sg_entries,
 733                     M_WAITOK) > CTL_HA_STATUS_SUCCESS) {
 734                         io->io_hdr.port_status = 31341;
 735                         io->scsiio.be_move_done(io);
 736                         return;
 737                 }
 738                 msg.dt.sent_sg_entries = sg_entries_sent;
 739         }
 740
 741         /*
 742          * Officially handover the request from us to peer.
 743          * If failover has just happened, then we must return error.
 744          * If failover happen just after, then it is not our problem.
 745          */
 746         if (lun)
 747                 mtx_lock(&lun->lun_lock);
 748         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
 749                 if (lun)
 750                         mtx_unlock(&lun->lun_lock);
 751                 io->io_hdr.port_status = 31342;
 752                 io->scsiio.be_move_done(io);
 753                 return;
 754         }
 755         io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
 756         io->io_hdr.flags |= CTL_FLAG_DMA_INPROG;
 757         if (lun)
 758                 mtx_unlock(&lun->lun_lock);
 759 }
 760
 761 static void
 762 ctl_ha_done(union ctl_io *io)
 763 {
 764         union ctl_ha_msg msg;
 765
 766         if (io->io_hdr.io_type == CTL_IO_SCSI) {
 767                 memset(&msg, 0, sizeof(msg));
 768                 msg.hdr.msg_type = CTL_MSG_FINISH_IO;
 769                 msg.hdr.original_sc = io->io_hdr.remote_io;
 770                 msg.hdr.nexus = io->io_hdr.nexus;
 771                 msg.hdr.status = io->io_hdr.status;
 772                 msg.scsi.scsi_status = io->scsiio.scsi_status;
 773                 msg.scsi.tag_num = io->scsiio.tag_num;
 774                 msg.scsi.tag_type = io->scsiio.tag_type;
 775                 msg.scsi.sense_len = io->scsiio.sense_len;
 776                 memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
 777                     io->scsiio.sense_len);
 778                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
 779                     sizeof(msg.scsi) - sizeof(msg.scsi.sense_data) +
 780                     msg.scsi.sense_len, M_WAITOK);
 781         }
 782         ctl_free_io(io);
 783 }
 784
 785 static void
 786 ctl_isc_handler_finish_xfer(struct ctl_softc *ctl_softc,
 787                             union ctl_ha_msg *msg_info)
 788 {
 789         struct ctl_scsiio *ctsio;
 790
 791         if (msg_info->hdr.original_sc == NULL) {
 792                 printf("%s: original_sc == NULL!\n", __func__);
 793                 /* XXX KDM now what? */
 794                 return;
 795         }
 796
 797         ctsio = &msg_info->hdr.original_sc->scsiio;
 798         ctsio->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
 799         ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
 800         ctsio->io_hdr.status = msg_info->hdr.status;
 801         ctsio->scsi_status = msg_info->scsi.scsi_status;
 802         ctsio->sense_len = msg_info->scsi.sense_len;
 803         memcpy(&ctsio->sense_data, &msg_info->scsi.sense_data,
 804                msg_info->scsi.sense_len);
 805         ctl_enqueue_isc((union ctl_io *)ctsio);
 806 }
 807
 808 static void
 809 ctl_isc_handler_finish_ser_only(struct ctl_softc *ctl_softc,
 810                                 union ctl_ha_msg *msg_info)
 811 {
 812         struct ctl_scsiio *ctsio;
 813
 814         if (msg_info->hdr.serializing_sc == NULL) {
 815                 printf("%s: serializing_sc == NULL!\n", __func__);
 816                 /* XXX KDM now what? */
 817                 return;
 818         }
 819
 820         ctsio = &msg_info->hdr.serializing_sc->scsiio;
 821         ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
 822         ctl_enqueue_isc((union ctl_io *)ctsio);
 823 }
 824
 825 void
 826 ctl_isc_announce_lun(struct ctl_lun *lun)
 827 {
 828         struct ctl_softc *softc = lun->ctl_softc;
 829         union ctl_ha_msg *msg;
 830         struct ctl_ha_msg_lun_pr_key pr_key;
 831         int i, k;
 832
 833         if (softc->ha_link != CTL_HA_LINK_ONLINE)
 834                 return;
 835         mtx_lock(&lun->lun_lock);
 836         i = sizeof(msg->lun);
 837         if (lun->lun_devid)
 838                 i += lun->lun_devid->len;
 839         i += sizeof(pr_key) * lun->pr_key_count;
 840 alloc:
 841         mtx_unlock(&lun->lun_lock);
 842         msg = malloc(i, M_CTL, M_WAITOK);
 843         mtx_lock(&lun->lun_lock);
 844         k = sizeof(msg->lun);
 845         if (lun->lun_devid)
 846                 k += lun->lun_devid->len;
 847         k += sizeof(pr_key) * lun->pr_key_count;
 848         if (i < k) {
 849                 free(msg, M_CTL);
 850                 i = k;
 851                 goto alloc;
 852         }
 853         bzero(&msg->lun, sizeof(msg->lun));
 854         msg->hdr.msg_type = CTL_MSG_LUN_SYNC;
 855         msg->hdr.nexus.targ_lun = lun->lun;
 856         msg->hdr.nexus.targ_mapped_lun = lun->lun;
 857         msg->lun.flags = lun->flags;
 858         msg->lun.pr_generation = lun->pr_generation;
 859         msg->lun.pr_res_idx = lun->pr_res_idx;
 860         msg->lun.pr_res_type = lun->pr_res_type;
 861         msg->lun.pr_key_count = lun->pr_key_count;
 862         i = 0;
 863         if (lun->lun_devid) {
 864                 msg->lun.lun_devid_len = lun->lun_devid->len;
 865                 memcpy(&msg->lun.data[i], lun->lun_devid->data,
 866                     msg->lun.lun_devid_len);
 867                 i += msg->lun.lun_devid_len;
 868         }
 869         for (k = 0; k < CTL_MAX_INITIATORS; k++) {
 870                 if ((pr_key.pr_key = ctl_get_prkey(lun, k)) == 0)
 871                         continue;
 872                 pr_key.pr_iid = k;
 873                 memcpy(&msg->lun.data[i], &pr_key, sizeof(pr_key));
 874                 i += sizeof(pr_key);
 875         }
 876         mtx_unlock(&lun->lun_lock);
 877         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->port, sizeof(msg->port) + i,
 878             M_WAITOK);
 879         free(msg, M_CTL);
 880
 881         if (lun->flags & CTL_LUN_PRIMARY_SC) {
 882                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
 883                         ctl_isc_announce_mode(lun, -1,
 884                             lun->mode_pages.index[i].page_code & SMPH_PC_MASK,
 885                             lun->mode_pages.index[i].subpage);
 886                 }
 887         }
 888 }
 889
 890 void
 891 ctl_isc_announce_port(struct ctl_port *port)
 892 {
 893         struct ctl_softc *softc = port->ctl_softc;
 894         union ctl_ha_msg *msg;
 895         int i;
 896
 897         if (port->targ_port < softc->port_min ||
 898             port->targ_port >= softc->port_max ||
 899             softc->ha_link != CTL_HA_LINK_ONLINE)
 900                 return;
 901         i = sizeof(msg->port) + strlen(port->port_name) + 1;
 902         if (port->lun_map)
 903                 i += port->lun_map_size * sizeof(uint32_t);
 904         if (port->port_devid)
 905                 i += port->port_devid->len;
 906         if (port->target_devid)
 907                 i += port->target_devid->len;
 908         if (port->init_devid)
 909                 i += port->init_devid->len;
 910         msg = malloc(i, M_CTL, M_WAITOK);
 911         bzero(&msg->port, sizeof(msg->port));
 912         msg->hdr.msg_type = CTL_MSG_PORT_SYNC;
 913         msg->hdr.nexus.targ_port = port->targ_port;
 914         msg->port.port_type = port->port_type;
 915         msg->port.physical_port = port->physical_port;
 916         msg->port.virtual_port = port->virtual_port;
 917         msg->port.status = port->status;
 918         i = 0;
 919         msg->port.name_len = sprintf(&msg->port.data[i],
 920             "%d:%s", softc->ha_id, port->port_name) + 1;
 921         i += msg->port.name_len;
 922         if (port->lun_map) {
 923                 msg->port.lun_map_len = port->lun_map_size * sizeof(uint32_t);
 924                 memcpy(&msg->port.data[i], port->lun_map,
 925                     msg->port.lun_map_len);
 926                 i += msg->port.lun_map_len;
 927         }
 928         if (port->port_devid) {
 929                 msg->port.port_devid_len = port->port_devid->len;
 930                 memcpy(&msg->port.data[i], port->port_devid->data,
 931                     msg->port.port_devid_len);
 932                 i += msg->port.port_devid_len;
 933         }
 934         if (port->target_devid) {
 935                 msg->port.target_devid_len = port->target_devid->len;
 936                 memcpy(&msg->port.data[i], port->target_devid->data,
 937                     msg->port.target_devid_len);
 938                 i += msg->port.target_devid_len;
 939         }
 940         if (port->init_devid) {
 941                 msg->port.init_devid_len = port->init_devid->len;
 942                 memcpy(&msg->port.data[i], port->init_devid->data,
 943                     msg->port.init_devid_len);
 944                 i += msg->port.init_devid_len;
 945         }
 946         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->port, sizeof(msg->port) + i,
 947             M_WAITOK);
 948         free(msg, M_CTL);
 949 }
 950
 951 void
 952 ctl_isc_announce_iid(struct ctl_port *port, int iid)
 953 {
 954         struct ctl_softc *softc = port->ctl_softc;
 955         union ctl_ha_msg *msg;
 956         int i, l;
 957
 958         if (port->targ_port < softc->port_min ||
 959             port->targ_port >= softc->port_max ||
 960             softc->ha_link != CTL_HA_LINK_ONLINE)
 961                 return;
 962         mtx_lock(&softc->ctl_lock);
 963         i = sizeof(msg->iid);
 964         l = 0;
 965         if (port->wwpn_iid[iid].name)
 966                 l = strlen(port->wwpn_iid[iid].name) + 1;
 967         i += l;
 968         msg = malloc(i, M_CTL, M_NOWAIT);
 969         if (msg == NULL) {
 970                 mtx_unlock(&softc->ctl_lock);
 971                 return;
 972         }
 973         bzero(&msg->iid, sizeof(msg->iid));
 974         msg->hdr.msg_type = CTL_MSG_IID_SYNC;
 975         msg->hdr.nexus.targ_port = port->targ_port;
 976         msg->hdr.nexus.initid = iid;
 977         msg->iid.in_use = port->wwpn_iid[iid].in_use;
 978         msg->iid.name_len = l;
 979         msg->iid.wwpn = port->wwpn_iid[iid].wwpn;
 980         if (port->wwpn_iid[iid].name)
 981                 strlcpy(msg->iid.data, port->wwpn_iid[iid].name, l);
 982         mtx_unlock(&softc->ctl_lock);
 983         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->iid, i, M_NOWAIT);
 984         free(msg, M_CTL);
 985 }
 986
 987 void
 988 ctl_isc_announce_mode(struct ctl_lun *lun, uint32_t initidx,
 989     uint8_t page, uint8_t subpage)
 990 {
 991         struct ctl_softc *softc = lun->ctl_softc;
 992         union ctl_ha_msg msg;
 993         u_int i;
 994
 995         if (softc->ha_link != CTL_HA_LINK_ONLINE)
 996                 return;
 997         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
 998                 if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) ==
 999                     page && lun->mode_pages.index[i].subpage == subpage)
1000                         break;
1001         }
1002         if (i == CTL_NUM_MODE_PAGES)
1003                 return;
1004
1005         /* Don't try to replicate pages not present on this device. */
1006         if (lun->mode_pages.index[i].page_data == NULL)
1007                 return;
1008
1009         bzero(&msg.mode, sizeof(msg.mode));
1010         msg.hdr.msg_type = CTL_MSG_MODE_SYNC;
1011         msg.hdr.nexus.targ_port = initidx / CTL_MAX_INIT_PER_PORT;
1012         msg.hdr.nexus.initid = initidx % CTL_MAX_INIT_PER_PORT;
1013         msg.hdr.nexus.targ_lun = lun->lun;
1014         msg.hdr.nexus.targ_mapped_lun = lun->lun;
1015         msg.mode.page_code = page;
1016         msg.mode.subpage = subpage;
1017         msg.mode.page_len = lun->mode_pages.index[i].page_len;
1018         memcpy(msg.mode.data, lun->mode_pages.index[i].page_data,
1019             msg.mode.page_len);
1020         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg.mode, sizeof(msg.mode),
1021             M_WAITOK);
1022 }
1023
1024 static void
1025 ctl_isc_ha_link_up(struct ctl_softc *softc)
1026 {
1027         struct ctl_port *port;
1028         struct ctl_lun *lun;
1029         union ctl_ha_msg msg;
1030         int i;
1031
1032         /* Announce this node parameters to peer for validation. */
1033         msg.login.msg_type = CTL_MSG_LOGIN;
1034         msg.login.version = CTL_HA_VERSION;
1035         msg.login.ha_mode = softc->ha_mode;
1036         msg.login.ha_id = softc->ha_id;
1037         msg.login.max_luns = ctl_max_luns;
1038         msg.login.max_ports = ctl_max_ports;
1039         msg.login.max_init_per_port = CTL_MAX_INIT_PER_PORT;
1040         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg.login, sizeof(msg.login),
1041             M_WAITOK);
1042
1043         STAILQ_FOREACH(port, &softc->port_list, links) {
1044                 ctl_isc_announce_port(port);
1045                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1046                         if (port->wwpn_iid[i].in_use)
1047                                 ctl_isc_announce_iid(port, i);
1048                 }
1049         }
1050         STAILQ_FOREACH(lun, &softc->lun_list, links)
1051                 ctl_isc_announce_lun(lun);
1052 }
1053
1054 static void
1055 ctl_isc_ha_link_down(struct ctl_softc *softc)
1056 {
1057         struct ctl_port *port;
1058         struct ctl_lun *lun;
1059         union ctl_io *io;
1060         int i;
1061
1062         mtx_lock(&softc->ctl_lock);
1063         STAILQ_FOREACH(lun, &softc->lun_list, links) {
1064                 mtx_lock(&lun->lun_lock);
1065                 if (lun->flags & CTL_LUN_PEER_SC_PRIMARY) {
1066                         lun->flags &= ~CTL_LUN_PEER_SC_PRIMARY;
1067                         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
1068                 }
1069                 mtx_unlock(&lun->lun_lock);
1070
1071                 mtx_unlock(&softc->ctl_lock);
1072                 io = ctl_alloc_io(softc->othersc_pool);
1073                 mtx_lock(&softc->ctl_lock);
1074                 ctl_zero_io(io);
1075                 io->io_hdr.msg_type = CTL_MSG_FAILOVER;
1076                 io->io_hdr.nexus.targ_mapped_lun = lun->lun;
1077                 ctl_enqueue_isc(io);
1078         }
1079
1080         STAILQ_FOREACH(port, &softc->port_list, links) {
1081                 if (port->targ_port >= softc->port_min &&
1082                     port->targ_port < softc->port_max)
1083                         continue;
1084                 port->status &= ~CTL_PORT_STATUS_ONLINE;
1085                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1086                         port->wwpn_iid[i].in_use = 0;
1087                         free(port->wwpn_iid[i].name, M_CTL);
1088                         port->wwpn_iid[i].name = NULL;
1089                 }
1090         }
1091         mtx_unlock(&softc->ctl_lock);
1092 }
1093
1094 static void
1095 ctl_isc_ua(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1096 {
1097         struct ctl_lun *lun;
1098         uint32_t iid = ctl_get_initindex(&msg->hdr.nexus);
1099
1100         mtx_lock(&softc->ctl_lock);
1101         if (msg->hdr.nexus.targ_mapped_lun >= ctl_max_luns ||
1102             (lun = softc->ctl_luns[msg->hdr.nexus.targ_mapped_lun]) == NULL) {
1103                 mtx_unlock(&softc->ctl_lock);
1104                 return;
1105         }
1106         mtx_lock(&lun->lun_lock);
1107         mtx_unlock(&softc->ctl_lock);
1108         if (msg->ua.ua_type == CTL_UA_THIN_PROV_THRES && msg->ua.ua_set)
1109                 memcpy(lun->ua_tpt_info, msg->ua.ua_info, 8);
1110         if (msg->ua.ua_all) {
1111                 if (msg->ua.ua_set)
1112                         ctl_est_ua_all(lun, iid, msg->ua.ua_type);
1113                 else
1114                         ctl_clr_ua_all(lun, iid, msg->ua.ua_type);
1115         } else {
1116                 if (msg->ua.ua_set)
1117                         ctl_est_ua(lun, iid, msg->ua.ua_type);
1118                 else
1119                         ctl_clr_ua(lun, iid, msg->ua.ua_type);
1120         }
1121         mtx_unlock(&lun->lun_lock);
1122 }
1123
1124 static void
1125 ctl_isc_lun_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1126 {
1127         struct ctl_lun *lun;
1128         struct ctl_ha_msg_lun_pr_key pr_key;
1129         int i, k;
1130         ctl_lun_flags oflags;
1131         uint32_t targ_lun;
1132
1133         targ_lun = msg->hdr.nexus.targ_mapped_lun;
1134         mtx_lock(&softc->ctl_lock);
1135         if (targ_lun >= ctl_max_luns ||
1136             (lun = softc->ctl_luns[targ_lun]) == NULL) {
1137                 mtx_unlock(&softc->ctl_lock);
1138                 return;
1139         }
1140         mtx_lock(&lun->lun_lock);
1141         mtx_unlock(&softc->ctl_lock);
1142         if (lun->flags & CTL_LUN_DISABLED) {
1143                 mtx_unlock(&lun->lun_lock);
1144                 return;
1145         }
1146         i = (lun->lun_devid != NULL) ? lun->lun_devid->len : 0;
1147         if (msg->lun.lun_devid_len != i || (i > 0 &&
1148             memcmp(&msg->lun.data[0], lun->lun_devid->data, i) != 0)) {
1149                 mtx_unlock(&lun->lun_lock);
1150                 printf("%s: Received conflicting HA LUN %d\n",
1151                     __func__, targ_lun);
1152                 return;
1153         } else {
1154                 /* Record whether peer is primary. */
1155                 oflags = lun->flags;
1156                 if ((msg->lun.flags & CTL_LUN_PRIMARY_SC) &&
1157                     (msg->lun.flags & CTL_LUN_DISABLED) == 0)
1158                         lun->flags |= CTL_LUN_PEER_SC_PRIMARY;
1159                 else
1160                         lun->flags &= ~CTL_LUN_PEER_SC_PRIMARY;
1161                 if (oflags != lun->flags)
1162                         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
1163
1164                 /* If peer is primary and we are not -- use data */
1165                 if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
1166                     (lun->flags & CTL_LUN_PEER_SC_PRIMARY)) {
1167                         lun->pr_generation = msg->lun.pr_generation;
1168                         lun->pr_res_idx = msg->lun.pr_res_idx;
1169                         lun->pr_res_type = msg->lun.pr_res_type;
1170                         lun->pr_key_count = msg->lun.pr_key_count;
1171                         for (k = 0; k < CTL_MAX_INITIATORS; k++)
1172                                 ctl_clr_prkey(lun, k);
1173                         for (k = 0; k < msg->lun.pr_key_count; k++) {
1174                                 memcpy(&pr_key, &msg->lun.data[i],
1175                                     sizeof(pr_key));
1176                                 ctl_alloc_prkey(lun, pr_key.pr_iid);
1177                                 ctl_set_prkey(lun, pr_key.pr_iid,
1178                                     pr_key.pr_key);
1179                                 i += sizeof(pr_key);
1180                         }
1181                 }
1182
1183                 mtx_unlock(&lun->lun_lock);
1184                 CTL_DEBUG_PRINT(("%s: Known LUN %d, peer is %s\n",
1185                     __func__, targ_lun,
1186                     (msg->lun.flags & CTL_LUN_PRIMARY_SC) ?
1187                     "primary" : "secondary"));
1188
1189                 /* If we are primary but peer doesn't know -- notify */
1190                 if ((lun->flags & CTL_LUN_PRIMARY_SC) &&
1191                     (msg->lun.flags & CTL_LUN_PEER_SC_PRIMARY) == 0)
1192                         ctl_isc_announce_lun(lun);
1193         }
1194 }
1195
1196 static void
1197 ctl_isc_port_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1198 {
1199         struct ctl_port *port;
1200         struct ctl_lun *lun;
1201         int i, new;
1202
1203         port = softc->ctl_ports[msg->hdr.nexus.targ_port];
1204         if (port == NULL) {
1205                 CTL_DEBUG_PRINT(("%s: New port %d\n", __func__,
1206                     msg->hdr.nexus.targ_port));
1207                 new = 1;
1208                 port = malloc(sizeof(*port), M_CTL, M_WAITOK | M_ZERO);
1209                 port->frontend = &ha_frontend;
1210                 port->targ_port = msg->hdr.nexus.targ_port;
1211                 port->fe_datamove = ctl_ha_datamove;
1212                 port->fe_done = ctl_ha_done;
1213         } else if (port->frontend == &ha_frontend) {
1214                 CTL_DEBUG_PRINT(("%s: Updated port %d\n", __func__,
1215                     msg->hdr.nexus.targ_port));
1216                 new = 0;
1217         } else {
1218                 printf("%s: Received conflicting HA port %d\n",
1219                     __func__, msg->hdr.nexus.targ_port);
1220                 return;
1221         }
1222         port->port_type = msg->port.port_type;
1223         port->physical_port = msg->port.physical_port;
1224         port->virtual_port = msg->port.virtual_port;
1225         port->status = msg->port.status;
1226         i = 0;
1227         free(port->port_name, M_CTL);
1228         port->port_name = strndup(&msg->port.data[i], msg->port.name_len,
1229             M_CTL);
1230         i += msg->port.name_len;
1231         if (msg->port.lun_map_len != 0) {
1232                 if (port->lun_map == NULL ||
1233                     port->lun_map_size * sizeof(uint32_t) <
1234                     msg->port.lun_map_len) {
1235                         port->lun_map_size = 0;
1236                         free(port->lun_map, M_CTL);
1237                         port->lun_map = malloc(msg->port.lun_map_len,
1238                             M_CTL, M_WAITOK);
1239                 }
1240                 memcpy(port->lun_map, &msg->port.data[i], msg->port.lun_map_len);
1241                 port->lun_map_size = msg->port.lun_map_len / sizeof(uint32_t);
1242                 i += msg->port.lun_map_len;
1243         } else {
1244                 port->lun_map_size = 0;
1245                 free(port->lun_map, M_CTL);
1246                 port->lun_map = NULL;
1247         }
1248         if (msg->port.port_devid_len != 0) {
1249                 if (port->port_devid == NULL ||
1250                     port->port_devid->len < msg->port.port_devid_len) {
1251                         free(port->port_devid, M_CTL);
1252                         port->port_devid = malloc(sizeof(struct ctl_devid) +
1253                             msg->port.port_devid_len, M_CTL, M_WAITOK);
1254                 }
1255                 memcpy(port->port_devid->data, &msg->port.data[i],
1256                     msg->port.port_devid_len);
1257                 port->port_devid->len = msg->port.port_devid_len;
1258                 i += msg->port.port_devid_len;
1259         } else {
1260                 free(port->port_devid, M_CTL);
1261                 port->port_devid = NULL;
1262         }
1263         if (msg->port.target_devid_len != 0) {
1264                 if (port->target_devid == NULL ||
1265                     port->target_devid->len < msg->port.target_devid_len) {
1266                         free(port->target_devid, M_CTL);
1267                         port->target_devid = malloc(sizeof(struct ctl_devid) +
1268                             msg->port.target_devid_len, M_CTL, M_WAITOK);
1269                 }
1270                 memcpy(port->target_devid->data, &msg->port.data[i],
1271                     msg->port.target_devid_len);
1272                 port->target_devid->len = msg->port.target_devid_len;
1273                 i += msg->port.target_devid_len;
1274         } else {
1275                 free(port->target_devid, M_CTL);
1276                 port->target_devid = NULL;
1277         }
1278         if (msg->port.init_devid_len != 0) {
1279                 if (port->init_devid == NULL ||
1280                     port->init_devid->len < msg->port.init_devid_len) {
1281                         free(port->init_devid, M_CTL);
1282                         port->init_devid = malloc(sizeof(struct ctl_devid) +
1283                             msg->port.init_devid_len, M_CTL, M_WAITOK);
1284                 }
1285                 memcpy(port->init_devid->data, &msg->port.data[i],
1286                     msg->port.init_devid_len);
1287                 port->init_devid->len = msg->port.init_devid_len;
1288                 i += msg->port.init_devid_len;
1289         } else {
1290                 free(port->init_devid, M_CTL);
1291                 port->init_devid = NULL;
1292         }
1293         if (new) {
1294                 if (ctl_port_register(port) != 0) {
1295                         printf("%s: ctl_port_register() failed with error\n",
1296                             __func__);
1297                 }
1298         }
1299         mtx_lock(&softc->ctl_lock);
1300         STAILQ_FOREACH(lun, &softc->lun_list, links) {
1301                 if (ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
1302                         continue;
1303                 mtx_lock(&lun->lun_lock);
1304                 ctl_est_ua_all(lun, -1, CTL_UA_INQ_CHANGE);
1305                 mtx_unlock(&lun->lun_lock);
1306         }
1307         mtx_unlock(&softc->ctl_lock);
1308 }
1309
1310 static void
1311 ctl_isc_iid_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1312 {
1313         struct ctl_port *port;
1314         int iid;
1315
1316         port = softc->ctl_ports[msg->hdr.nexus.targ_port];
1317         if (port == NULL) {
1318                 printf("%s: Received IID for unknown port %d\n",
1319                     __func__, msg->hdr.nexus.targ_port);
1320                 return;
1321         }
1322         iid = msg->hdr.nexus.initid;
1323         if (port->wwpn_iid[iid].in_use != 0 &&
1324             msg->iid.in_use == 0)
1325                 ctl_i_t_nexus_loss(softc, iid, CTL_UA_POWERON);
1326         port->wwpn_iid[iid].in_use = msg->iid.in_use;
1327         port->wwpn_iid[iid].wwpn = msg->iid.wwpn;
1328         free(port->wwpn_iid[iid].name, M_CTL);
1329         if (msg->iid.name_len) {
1330                 port->wwpn_iid[iid].name = strndup(&msg->iid.data[0],
1331                     msg->iid.name_len, M_CTL);
1332         } else
1333                 port->wwpn_iid[iid].name = NULL;
1334 }
1335
1336 static void
1337 ctl_isc_login(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1338 {
1339
1340         if (msg->login.version != CTL_HA_VERSION) {
1341                 printf("CTL HA peers have different versions %d != %d\n",
1342                     msg->login.version, CTL_HA_VERSION);
1343                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1344                 return;
1345         }
1346         if (msg->login.ha_mode != softc->ha_mode) {
1347                 printf("CTL HA peers have different ha_mode %d != %d\n",
1348                     msg->login.ha_mode, softc->ha_mode);
1349                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1350                 return;
1351         }
1352         if (msg->login.ha_id == softc->ha_id) {
1353                 printf("CTL HA peers have same ha_id %d\n", msg->login.ha_id);
1354                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1355                 return;
1356         }
1357         if (msg->login.max_luns != ctl_max_luns ||
1358             msg->login.max_ports != ctl_max_ports ||
1359             msg->login.max_init_per_port != CTL_MAX_INIT_PER_PORT) {
1360                 printf("CTL HA peers have different limits\n");
1361                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1362                 return;
1363         }
1364 }
1365
1366 static void
1367 ctl_isc_mode_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1368 {
1369         struct ctl_lun *lun;
1370         u_int i;
1371         uint32_t initidx, targ_lun;
1372
1373         targ_lun = msg->hdr.nexus.targ_mapped_lun;
1374         mtx_lock(&softc->ctl_lock);
1375         if (targ_lun >= ctl_max_luns ||
1376             (lun = softc->ctl_luns[targ_lun]) == NULL) {
1377                 mtx_unlock(&softc->ctl_lock);
1378                 return;
1379         }
1380         mtx_lock(&lun->lun_lock);
1381         mtx_unlock(&softc->ctl_lock);
1382         if (lun->flags & CTL_LUN_DISABLED) {
1383                 mtx_unlock(&lun->lun_lock);
1384                 return;
1385         }
1386         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
1387                 if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) ==
1388                     msg->mode.page_code &&
1389                     lun->mode_pages.index[i].subpage == msg->mode.subpage)
1390                         break;
1391         }
1392         if (i == CTL_NUM_MODE_PAGES) {
1393                 mtx_unlock(&lun->lun_lock);
1394                 return;
1395         }
1396         memcpy(lun->mode_pages.index[i].page_data, msg->mode.data,
1397             lun->mode_pages.index[i].page_len);
1398         initidx = ctl_get_initindex(&msg->hdr.nexus);
1399         if (initidx != -1)
1400                 ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
1401         mtx_unlock(&lun->lun_lock);
1402 }
1403
1404 /*
1405  * ISC (Inter Shelf Communication) event handler.  Events from the HA
1406  * subsystem come in here.
1407  */
1408 static void
1409 ctl_isc_event_handler(ctl_ha_channel channel, ctl_ha_event event, int param)
1410 {
1411         struct ctl_softc *softc = control_softc;
1412         union ctl_io *io;
1413         struct ctl_prio *presio;
1414         ctl_ha_status isc_status;
1415
1416         CTL_DEBUG_PRINT(("CTL: Isc Msg event %d\n", event));
1417         if (event == CTL_HA_EVT_MSG_RECV) {
1418                 union ctl_ha_msg *msg, msgbuf;
1419
1420                 if (param > sizeof(msgbuf))
1421                         msg = malloc(param, M_CTL, M_WAITOK);
1422                 else
1423                         msg = &msgbuf;
1424                 isc_status = ctl_ha_msg_recv(CTL_HA_CHAN_CTL, msg, param,
1425                     M_WAITOK);
1426                 if (isc_status != CTL_HA_STATUS_SUCCESS) {
1427                         printf("%s: Error receiving message: %d\n",
1428                             __func__, isc_status);
1429                         if (msg != &msgbuf)
1430                                 free(msg, M_CTL);
1431                         return;
1432                 }
1433
1434                 CTL_DEBUG_PRINT(("CTL: msg_type %d\n", msg->msg_type));
1435                 switch (msg->hdr.msg_type) {
1436                 case CTL_MSG_SERIALIZE:
1437                         io = ctl_alloc_io(softc->othersc_pool);
1438                         ctl_zero_io(io);
1439                         // populate ctsio from msg
1440                         io->io_hdr.io_type = CTL_IO_SCSI;
1441                         io->io_hdr.msg_type = CTL_MSG_SERIALIZE;
1442                         io->io_hdr.remote_io = msg->hdr.original_sc;
1443                         io->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC |
1444                                             CTL_FLAG_IO_ACTIVE;
1445                         /*
1446                          * If we're in serialization-only mode, we don't
1447                          * want to go through full done processing.  Thus
1448                          * the COPY flag.
1449                          *
1450                          * XXX KDM add another flag that is more specific.
1451                          */
1452                         if (softc->ha_mode != CTL_HA_MODE_XFER)
1453                                 io->io_hdr.flags |= CTL_FLAG_INT_COPY;
1454                         io->io_hdr.nexus = msg->hdr.nexus;
1455                         io->scsiio.tag_num = msg->scsi.tag_num;
1456                         io->scsiio.tag_type = msg->scsi.tag_type;
1457 #ifdef CTL_TIME_IO
1458                         io->io_hdr.start_time = time_uptime;
1459                         getbinuptime(&io->io_hdr.start_bt);
1460 #endif /* CTL_TIME_IO */
1461                         io->scsiio.cdb_len = msg->scsi.cdb_len;
1462                         memcpy(io->scsiio.cdb, msg->scsi.cdb,
1463                                CTL_MAX_CDBLEN);
1464                         if (softc->ha_mode == CTL_HA_MODE_XFER) {
1465                                 const struct ctl_cmd_entry *entry;
1466
1467                                 entry = ctl_get_cmd_entry(&io->scsiio, NULL);
1468                                 io->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
1469                                 io->io_hdr.flags |=
1470                                         entry->flags & CTL_FLAG_DATA_MASK;
1471                         }
1472                         ctl_enqueue_isc(io);
1473                         break;
1474
1475                 /* Performed on the Originating SC, XFER mode only */
1476                 case CTL_MSG_DATAMOVE: {
1477                         struct ctl_sg_entry *sgl;
1478                         int i, j;
1479
1480                         io = msg->hdr.original_sc;
1481                         if (io == NULL) {
1482                                 printf("%s: original_sc == NULL!\n", __func__);
1483                                 /* XXX KDM do something here */
1484                                 break;
1485                         }
1486                         io->io_hdr.msg_type = CTL_MSG_DATAMOVE;
1487                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1488                         /*
1489                          * Keep track of this, we need to send it back over
1490                          * when the datamove is complete.
1491                          */
1492                         io->io_hdr.remote_io = msg->hdr.serializing_sc;
1493                         if (msg->hdr.status == CTL_SUCCESS)
1494                                 io->io_hdr.status = msg->hdr.status;
1495
1496                         if (msg->dt.sg_sequence == 0) {
1497 #ifdef CTL_TIME_IO
1498                                 getbinuptime(&io->io_hdr.dma_start_bt);
1499 #endif
1500                                 i = msg->dt.kern_sg_entries +
1501                                     msg->dt.kern_data_len /
1502                                     CTL_HA_DATAMOVE_SEGMENT + 1;
1503                                 sgl = malloc(sizeof(*sgl) * i, M_CTL,
1504                                     M_WAITOK | M_ZERO);
1505                                 CTL_RSGL(io) = sgl;
1506                                 CTL_LSGL(io) = &sgl[msg->dt.kern_sg_entries];
1507
1508                                 io->scsiio.kern_data_ptr = (uint8_t *)sgl;
1509
1510                                 io->scsiio.kern_sg_entries =
1511                                         msg->dt.kern_sg_entries;
1512                                 io->scsiio.rem_sg_entries =
1513                                         msg->dt.kern_sg_entries;
1514                                 io->scsiio.kern_data_len =
1515                                         msg->dt.kern_data_len;
1516                                 io->scsiio.kern_total_len =
1517                                         msg->dt.kern_total_len;
1518                                 io->scsiio.kern_data_resid =
1519                                         msg->dt.kern_data_resid;
1520                                 io->scsiio.kern_rel_offset =
1521                                         msg->dt.kern_rel_offset;
1522                                 io->io_hdr.flags &= ~CTL_FLAG_BUS_ADDR;
1523                                 io->io_hdr.flags |= msg->dt.flags &
1524                                     CTL_FLAG_BUS_ADDR;
1525                         } else
1526                                 sgl = (struct ctl_sg_entry *)
1527                                         io->scsiio.kern_data_ptr;
1528
1529                         for (i = msg->dt.sent_sg_entries, j = 0;
1530                              i < (msg->dt.sent_sg_entries +
1531                              msg->dt.cur_sg_entries); i++, j++) {
1532                                 sgl[i].addr = msg->dt.sg_list[j].addr;
1533                                 sgl[i].len = msg->dt.sg_list[j].len;
1534                         }
1535
1536                         /*
1537                          * If this is the last piece of the I/O, we've got
1538                          * the full S/G list.  Queue processing in the thread.
1539                          * Otherwise wait for the next piece.
1540                          */
1541                         if (msg->dt.sg_last != 0)
1542                                 ctl_enqueue_isc(io);
1543                         break;
1544                 }
1545                 /* Performed on the Serializing (primary) SC, XFER mode only */
1546                 case CTL_MSG_DATAMOVE_DONE: {
1547                         if (msg->hdr.serializing_sc == NULL) {
1548                                 printf("%s: serializing_sc == NULL!\n",
1549                                        __func__);
1550                                 /* XXX KDM now what? */
1551                                 break;
1552                         }
1553                         /*
1554                          * We grab the sense information here in case
1555                          * there was a failure, so we can return status
1556                          * back to the initiator.
1557                          */
1558                         io = msg->hdr.serializing_sc;
1559                         io->io_hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
1560                         io->io_hdr.flags &= ~CTL_FLAG_DMA_INPROG;
1561                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1562                         io->io_hdr.port_status = msg->scsi.port_status;
1563                         io->scsiio.kern_data_resid = msg->scsi.kern_data_resid;
1564                         if (msg->hdr.status != CTL_STATUS_NONE) {
1565                                 io->io_hdr.status = msg->hdr.status;
1566                                 io->scsiio.scsi_status = msg->scsi.scsi_status;
1567                                 io->scsiio.sense_len = msg->scsi.sense_len;
1568                                 memcpy(&io->scsiio.sense_data,
1569                                     &msg->scsi.sense_data,
1570                                     msg->scsi.sense_len);
1571                                 if (msg->hdr.status == CTL_SUCCESS)
1572                                         io->io_hdr.flags |= CTL_FLAG_STATUS_SENT;
1573                         }
1574                         ctl_enqueue_isc(io);
1575                         break;
1576                 }
1577
1578                 /* Preformed on Originating SC, SER_ONLY mode */
1579                 case CTL_MSG_R2R:
1580                         io = msg->hdr.original_sc;
1581                         if (io == NULL) {
1582                                 printf("%s: original_sc == NULL!\n",
1583                                     __func__);
1584                                 break;
1585                         }
1586                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1587                         io->io_hdr.msg_type = CTL_MSG_R2R;
1588                         io->io_hdr.remote_io = msg->hdr.serializing_sc;
1589                         ctl_enqueue_isc(io);
1590                         break;
1591
1592                 /*
1593                  * Performed on Serializing(i.e. primary SC) SC in SER_ONLY
1594                  * mode.
1595                  * Performed on the Originating (i.e. secondary) SC in XFER
1596                  * mode
1597                  */
1598                 case CTL_MSG_FINISH_IO:
1599                         if (softc->ha_mode == CTL_HA_MODE_XFER)
1600                                 ctl_isc_handler_finish_xfer(softc, msg);
1601                         else
1602                                 ctl_isc_handler_finish_ser_only(softc, msg);
1603                         break;
1604
1605                 /* Preformed on Originating SC */
1606                 case CTL_MSG_BAD_JUJU:
1607                         io = msg->hdr.original_sc;
1608                         if (io == NULL) {
1609                                 printf("%s: Bad JUJU!, original_sc is NULL!\n",
1610                                        __func__);
1611                                 break;
1612                         }
1613                         ctl_copy_sense_data(msg, io);
1614                         /*
1615                          * IO should have already been cleaned up on other
1616                          * SC so clear this flag so we won't send a message
1617                          * back to finish the IO there.
1618                          */
1619                         io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
1620                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1621
1622                         /* io = msg->hdr.serializing_sc; */
1623                         io->io_hdr.msg_type = CTL_MSG_BAD_JUJU;
1624                         ctl_enqueue_isc(io);
1625                         break;
1626
1627                 /* Handle resets sent from the other side */
1628                 case CTL_MSG_MANAGE_TASKS: {
1629                         struct ctl_taskio *taskio;
1630                         taskio = (struct ctl_taskio *)ctl_alloc_io(
1631                             softc->othersc_pool);
1632                         ctl_zero_io((union ctl_io *)taskio);
1633                         taskio->io_hdr.io_type = CTL_IO_TASK;
1634                         taskio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
1635                         taskio->io_hdr.nexus = msg->hdr.nexus;
1636                         taskio->task_action = msg->task.task_action;
1637                         taskio->tag_num = msg->task.tag_num;
1638                         taskio->tag_type = msg->task.tag_type;
1639 #ifdef CTL_TIME_IO
1640                         taskio->io_hdr.start_time = time_uptime;
1641                         getbinuptime(&taskio->io_hdr.start_bt);
1642 #endif /* CTL_TIME_IO */
1643                         ctl_run_task((union ctl_io *)taskio);
1644                         break;
1645                 }
1646                 /* Persistent Reserve action which needs attention */
1647                 case CTL_MSG_PERS_ACTION:
1648                         presio = (struct ctl_prio *)ctl_alloc_io(
1649                             softc->othersc_pool);
1650                         ctl_zero_io((union ctl_io *)presio);
1651                         presio->io_hdr.msg_type = CTL_MSG_PERS_ACTION;
1652                         presio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
1653                         presio->io_hdr.nexus = msg->hdr.nexus;
1654                         presio->pr_msg = msg->pr;
1655                         ctl_enqueue_isc((union ctl_io *)presio);
1656                         break;
1657                 case CTL_MSG_UA:
1658                         ctl_isc_ua(softc, msg, param);
1659                         break;
1660                 case CTL_MSG_PORT_SYNC:
1661                         ctl_isc_port_sync(softc, msg, param);
1662                         break;
1663                 case CTL_MSG_LUN_SYNC:
1664                         ctl_isc_lun_sync(softc, msg, param);
1665                         break;
1666                 case CTL_MSG_IID_SYNC:
1667                         ctl_isc_iid_sync(softc, msg, param);
1668                         break;
1669                 case CTL_MSG_LOGIN:
1670                         ctl_isc_login(softc, msg, param);
1671                         break;
1672                 case CTL_MSG_MODE_SYNC:
1673                         ctl_isc_mode_sync(softc, msg, param);
1674                         break;
1675                 default:
1676                         printf("Received HA message of unknown type %d\n",
1677                             msg->hdr.msg_type);
1678                         ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1679                         break;
1680                 }
1681                 if (msg != &msgbuf)
1682                         free(msg, M_CTL);
1683         } else if (event == CTL_HA_EVT_LINK_CHANGE) {
1684                 printf("CTL: HA link status changed from %d to %d\n",
1685                     softc->ha_link, param);
1686                 if (param == softc->ha_link)
1687                         return;
1688                 if (softc->ha_link == CTL_HA_LINK_ONLINE) {
1689                         softc->ha_link = param;
1690                         ctl_isc_ha_link_down(softc);
1691                 } else {
1692                         softc->ha_link = param;
1693                         if (softc->ha_link == CTL_HA_LINK_ONLINE)
1694                                 ctl_isc_ha_link_up(softc);
1695                 }
1696                 return;
1697         } else {
1698                 printf("ctl_isc_event_handler: Unknown event %d\n", event);
1699                 return;
1700         }
1701 }
1702
1703 static void
1704 ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest)
1705 {
1706
1707         memcpy(&dest->scsiio.sense_data, &src->scsi.sense_data,
1708             src->scsi.sense_len);
1709         dest->scsiio.scsi_status = src->scsi.scsi_status;
1710         dest->scsiio.sense_len = src->scsi.sense_len;
1711         dest->io_hdr.status = src->hdr.status;
1712 }
1713
1714 static void
1715 ctl_copy_sense_data_back(union ctl_io *src, union ctl_ha_msg *dest)
1716 {
1717
1718         memcpy(&dest->scsi.sense_data, &src->scsiio.sense_data,
1719             src->scsiio.sense_len);
1720         dest->scsi.scsi_status = src->scsiio.scsi_status;
1721         dest->scsi.sense_len = src->scsiio.sense_len;
1722         dest->hdr.status = src->io_hdr.status;
1723 }
1724
1725 void
1726 ctl_est_ua(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua)
1727 {
1728         struct ctl_softc *softc = lun->ctl_softc;
1729         ctl_ua_type *pu;
1730
1731         if (initidx < softc->init_min || initidx >= softc->init_max)
1732                 return;
1733         mtx_assert(&lun->lun_lock, MA_OWNED);
1734         pu = lun->pending_ua[initidx / CTL_MAX_INIT_PER_PORT];
1735         if (pu == NULL)
1736                 return;
1737         pu[initidx % CTL_MAX_INIT_PER_PORT] |= ua;
1738 }
1739
1740 void
1741 ctl_est_ua_port(struct ctl_lun *lun, int port, uint32_t except, ctl_ua_type ua)
1742 {
1743         int i;
1744
1745         mtx_assert(&lun->lun_lock, MA_OWNED);
1746         if (lun->pending_ua[port] == NULL)
1747                 return;
1748         for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1749                 if (port * CTL_MAX_INIT_PER_PORT + i == except)
1750                         continue;
1751                 lun->pending_ua[port][i] |= ua;
1752         }
1753 }
1754
1755 void
1756 ctl_est_ua_all(struct ctl_lun *lun, uint32_t except, ctl_ua_type ua)
1757 {
1758         struct ctl_softc *softc = lun->ctl_softc;
1759         int i;
1760
1761         mtx_assert(&lun->lun_lock, MA_OWNED);
1762         for (i = softc->port_min; i < softc->port_max; i++)
1763                 ctl_est_ua_port(lun, i, except, ua);
1764 }
1765
1766 void
1767 ctl_clr_ua(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua)
1768 {
1769         struct ctl_softc *softc = lun->ctl_softc;
1770         ctl_ua_type *pu;
1771
1772         if (initidx < softc->init_min || initidx >= softc->init_max)
1773                 return;
1774         mtx_assert(&lun->lun_lock, MA_OWNED);
1775         pu = lun->pending_ua[initidx / CTL_MAX_INIT_PER_PORT];
1776         if (pu == NULL)
1777                 return;
1778         pu[initidx % CTL_MAX_INIT_PER_PORT] &= ~ua;
1779 }
1780
1781 void
1782 ctl_clr_ua_all(struct ctl_lun *lun, uint32_t except, ctl_ua_type ua)
1783 {
1784         struct ctl_softc *softc = lun->ctl_softc;
1785         int i, j;
1786
1787         mtx_assert(&lun->lun_lock, MA_OWNED);
1788         for (i = softc->port_min; i < softc->port_max; i++) {
1789                 if (lun->pending_ua[i] == NULL)
1790                         continue;
1791                 for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
1792                         if (i * CTL_MAX_INIT_PER_PORT + j == except)
1793                                 continue;
1794                         lun->pending_ua[i][j] &= ~ua;
1795                 }
1796         }
1797 }
1798
1799 void
1800 ctl_clr_ua_allluns(struct ctl_softc *ctl_softc, uint32_t initidx,
1801     ctl_ua_type ua_type)
1802 {
1803         struct ctl_lun *lun;
1804
1805         mtx_assert(&ctl_softc->ctl_lock, MA_OWNED);
1806         STAILQ_FOREACH(lun, &ctl_softc->lun_list, links) {
1807                 mtx_lock(&lun->lun_lock);
1808                 ctl_clr_ua(lun, initidx, ua_type);
1809                 mtx_unlock(&lun->lun_lock);
1810         }
1811 }
1812
1813 static int
1814 ctl_ha_role_sysctl(SYSCTL_HANDLER_ARGS)
1815 {
1816         struct ctl_softc *softc = (struct ctl_softc *)arg1;
1817         struct ctl_lun *lun;
1818         struct ctl_lun_req ireq;
1819         int error, value;
1820
1821         value = (softc->flags & CTL_FLAG_ACTIVE_SHELF) ? 0 : 1;
1822         error = sysctl_handle_int(oidp, &value, 0, req);
1823         if ((error != 0) || (req->newptr == NULL))
1824                 return (error);
1825
1826         mtx_lock(&softc->ctl_lock);
1827         if (value == 0)
1828                 softc->flags |= CTL_FLAG_ACTIVE_SHELF;
1829         else
1830                 softc->flags &= ~CTL_FLAG_ACTIVE_SHELF;
1831         STAILQ_FOREACH(lun, &softc->lun_list, links) {
1832                 mtx_unlock(&softc->ctl_lock);
1833                 bzero(&ireq, sizeof(ireq));
1834                 ireq.reqtype = CTL_LUNREQ_MODIFY;
1835                 ireq.reqdata.modify.lun_id = lun->lun;
1836                 lun->backend->ioctl(NULL, CTL_LUN_REQ, (caddr_t)&ireq, 0,
1837                     curthread);
1838                 if (ireq.status != CTL_LUN_OK) {
1839                         printf("%s: CTL_LUNREQ_MODIFY returned %d '%s'\n",
1840                             __func__, ireq.status, ireq.error_str);
1841                 }
1842                 mtx_lock(&softc->ctl_lock);
1843         }
1844         mtx_unlock(&softc->ctl_lock);
1845         return (0);
1846 }
1847
1848 static int
1849 ctl_init(void)
1850 {
1851         struct make_dev_args args;
1852         struct ctl_softc *softc;
1853         int i, error;
1854
1855         softc = control_softc = malloc(sizeof(*control_softc), M_DEVBUF,
1856                                M_WAITOK | M_ZERO);
1857
1858         make_dev_args_init(&args);
1859         args.mda_devsw = &ctl_cdevsw;
1860         args.mda_uid = UID_ROOT;
1861         args.mda_gid = GID_OPERATOR;
1862         args.mda_mode = 0600;
1863         args.mda_si_drv1 = softc;
1864         args.mda_si_drv2 = NULL;
1865         error = make_dev_s(&args, &softc->dev, "cam/ctl");
1866         if (error != 0) {
1867                 free(softc, M_DEVBUF);
1868                 control_softc = NULL;
1869                 return (error);
1870         }
1871
1872         sysctl_ctx_init(&softc->sysctl_ctx);
1873         softc->sysctl_tree = SYSCTL_ADD_NODE(&softc->sysctl_ctx,
1874                 SYSCTL_STATIC_CHILDREN(_kern_cam), OID_AUTO, "ctl",
1875                 CTLFLAG_RD, 0, "CAM Target Layer");
1876
1877         if (softc->sysctl_tree == NULL) {
1878                 printf("%s: unable to allocate sysctl tree\n", __func__);
1879                 destroy_dev(softc->dev);
1880                 free(softc, M_DEVBUF);
1881                 control_softc = NULL;
1882                 return (ENOMEM);
1883         }
1884
1885         mtx_init(&softc->ctl_lock, "CTL mutex", NULL, MTX_DEF);
1886         softc->io_zone = uma_zcreate("CTL IO", sizeof(union ctl_io),
1887             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1888         softc->flags = 0;
1889
1890         SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1891             OID_AUTO, "ha_mode", CTLFLAG_RDTUN, (int *)&softc->ha_mode, 0,
1892             "HA mode (0 - act/stby, 1 - serialize only, 2 - xfer)");
1893
1894         if (ctl_max_luns <= 0 || powerof2(ctl_max_luns) == 0) {
1895                 printf("Bad value %d for kern.cam.ctl.max_luns, must be a power of two, using %d\n",
1896                     ctl_max_luns, CTL_DEFAULT_MAX_LUNS);
1897                 ctl_max_luns = CTL_DEFAULT_MAX_LUNS;
1898         }
1899         softc->ctl_luns = malloc(sizeof(struct ctl_lun *) * ctl_max_luns,
1900             M_DEVBUF, M_WAITOK | M_ZERO);
1901         softc->ctl_lun_mask = malloc(sizeof(uint32_t) *
1902             ((ctl_max_luns + 31) / 32), M_DEVBUF, M_WAITOK | M_ZERO);
1903         if (ctl_max_ports <= 0 || powerof2(ctl_max_ports) == 0) {
1904                 printf("Bad value %d for kern.cam.ctl.max_ports, must be a power of two, using %d\n",
1905                     ctl_max_ports, CTL_DEFAULT_MAX_PORTS);
1906                 ctl_max_ports = CTL_DEFAULT_MAX_PORTS;
1907         }
1908         softc->ctl_port_mask = malloc(sizeof(uint32_t) *
1909           ((ctl_max_ports + 31) / 32), M_DEVBUF, M_WAITOK | M_ZERO);
1910         softc->ctl_ports = malloc(sizeof(struct ctl_port *) * ctl_max_ports,
1911              M_DEVBUF, M_WAITOK | M_ZERO);
1912
1913
1914         /*
1915          * In Copan's HA scheme, the "master" and "slave" roles are
1916          * figured out through the slot the controller is in.  Although it
1917          * is an active/active system, someone has to be in charge.
1918          */
1919         SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1920             OID_AUTO, "ha_id", CTLFLAG_RDTUN, &softc->ha_id, 0,
1921             "HA head ID (0 - no HA)");
1922         if (softc->ha_id == 0 || softc->ha_id > NUM_HA_SHELVES) {
1923                 softc->flags |= CTL_FLAG_ACTIVE_SHELF;
1924                 softc->is_single = 1;
1925                 softc->port_cnt = ctl_max_ports;
1926                 softc->port_min = 0;
1927         } else {
1928                 softc->port_cnt = ctl_max_ports / NUM_HA_SHELVES;
1929                 softc->port_min = (softc->ha_id - 1) * softc->port_cnt;
1930         }
1931         softc->port_max = softc->port_min + softc->port_cnt;
1932         softc->init_min = softc->port_min * CTL_MAX_INIT_PER_PORT;
1933         softc->init_max = softc->port_max * CTL_MAX_INIT_PER_PORT;
1934
1935         SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1936             OID_AUTO, "ha_link", CTLFLAG_RD, (int *)&softc->ha_link, 0,
1937             "HA link state (0 - offline, 1 - unknown, 2 - online)");
1938
1939         STAILQ_INIT(&softc->lun_list);
1940         STAILQ_INIT(&softc->pending_lun_queue);
1941         STAILQ_INIT(&softc->fe_list);
1942         STAILQ_INIT(&softc->port_list);
1943         STAILQ_INIT(&softc->be_list);
1944         ctl_tpc_init(softc);
1945
1946         if (worker_threads <= 0)
1947                 worker_threads = max(1, mp_ncpus / 4);
1948         if (worker_threads > CTL_MAX_THREADS)
1949                 worker_threads = CTL_MAX_THREADS;
1950
1951         for (i = 0; i < worker_threads; i++) {
1952                 struct ctl_thread *thr = &softc->threads[i];
1953
1954                 mtx_init(&thr->queue_lock, "CTL queue mutex", NULL, MTX_DEF);
1955                 thr->ctl_softc = softc;
1956                 STAILQ_INIT(&thr->incoming_queue);
1957                 STAILQ_INIT(&thr->rtr_queue);
1958                 STAILQ_INIT(&thr->done_queue);
1959                 STAILQ_INIT(&thr->isc_queue);
1960
1961                 error = kproc_kthread_add(ctl_work_thread, thr,
1962                     &softc->ctl_proc, &thr->thread, 0, 0, "ctl", "work%d", i);
1963                 if (error != 0) {
1964                         printf("error creating CTL work thread!\n");
1965                         return (error);
1966                 }
1967         }
1968         error = kproc_kthread_add(ctl_lun_thread, softc,
1969             &softc->ctl_proc, &softc->lun_thread, 0, 0, "ctl", "lun");
1970         if (error != 0) {
1971                 printf("error creating CTL lun thread!\n");
1972                 return (error);
1973         }
1974         error = kproc_kthread_add(ctl_thresh_thread, softc,
1975             &softc->ctl_proc, &softc->thresh_thread, 0, 0, "ctl", "thresh");
1976         if (error != 0) {
1977                 printf("error creating CTL threshold thread!\n");
1978                 return (error);
1979         }
1980
1981         SYSCTL_ADD_PROC(&softc->sysctl_ctx,SYSCTL_CHILDREN(softc->sysctl_tree),
1982             OID_AUTO, "ha_role", CTLTYPE_INT | CTLFLAG_RWTUN,
1983             softc, 0, ctl_ha_role_sysctl, "I", "HA role for this head");
1984
1985         if (softc->is_single == 0) {
1986                 if (ctl_frontend_register(&ha_frontend) != 0)
1987                         softc->is_single = 1;
1988         }
1989         return (0);
1990 }
1991
1992 static int
1993 ctl_shutdown(void)
1994 {
1995         struct ctl_softc *softc = control_softc;
1996         int i;
1997
1998         if (softc->is_single == 0)
1999                 ctl_frontend_deregister(&ha_frontend);
2000
2001         destroy_dev(softc->dev);
2002
2003         /* Shutdown CTL threads. */
2004         softc->shutdown = 1;
2005         for (i = 0; i < worker_threads; i++) {
2006                 struct ctl_thread *thr = &softc->threads[i];
2007                 while (thr->thread != NULL) {
2008                         wakeup(thr);
2009                         if (thr->thread != NULL)
2010                                 pause("CTL thr shutdown", 1);
2011                 }
2012                 mtx_destroy(&thr->queue_lock);
2013         }
2014         while (softc->lun_thread != NULL) {
2015                 wakeup(&softc->pending_lun_queue);
2016                 if (softc->lun_thread != NULL)
2017                         pause("CTL thr shutdown", 1);
2018         }
2019         while (softc->thresh_thread != NULL) {
2020                 wakeup(softc->thresh_thread);
2021                 if (softc->thresh_thread != NULL)
2022                         pause("CTL thr shutdown", 1);
2023         }
2024
2025         ctl_tpc_shutdown(softc);
2026         uma_zdestroy(softc->io_zone);
2027         mtx_destroy(&softc->ctl_lock);
2028
2029         free(softc->ctl_luns, M_DEVBUF);
2030         free(softc->ctl_lun_mask, M_DEVBUF);
2031         free(softc->ctl_port_mask, M_DEVBUF);
2032         free(softc->ctl_ports, M_DEVBUF);
2033
2034         sysctl_ctx_free(&softc->sysctl_ctx);
2035
2036         free(softc, M_DEVBUF);
2037         control_softc = NULL;
2038         return (0);
2039 }
2040
2041 static int
2042 ctl_module_event_handler(module_t mod, int what, void *arg)
2043 {
2044
2045         switch (what) {
2046         case MOD_LOAD:
2047                 return (ctl_init());
2048         case MOD_UNLOAD:
2049                 return (ctl_shutdown());
2050         default:
2051                 return (EOPNOTSUPP);
2052         }
2053 }
2054
2055 /*
2056  * XXX KDM should we do some access checks here?  Bump a reference count to
2057  * prevent a CTL module from being unloaded while someone has it open?
2058  */
2059 static int
2060 ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td)
2061 {
2062         return (0);
2063 }
2064
2065 static int
2066 ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td)
2067 {
2068         return (0);
2069 }
2070
2071 /*
2072  * Remove an initiator by port number and initiator ID.
2073  * Returns 0 for success, -1 for failure.
2074  */
2075 int
2076 ctl_remove_initiator(struct ctl_port *port, int iid)
2077 {
2078         struct ctl_softc *softc = port->ctl_softc;
2079         int last;
2080
2081         mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
2082
2083         if (iid > CTL_MAX_INIT_PER_PORT) {
2084                 printf("%s: initiator ID %u > maximun %u!\n",
2085                        __func__, iid, CTL_MAX_INIT_PER_PORT);
2086                 return (-1);
2087         }
2088
2089         mtx_lock(&softc->ctl_lock);
2090         last = (--port->wwpn_iid[iid].in_use == 0);
2091         port->wwpn_iid[iid].last_use = time_uptime;
2092         mtx_unlock(&softc->ctl_lock);
2093         if (last)
2094                 ctl_i_t_nexus_loss(softc, iid, CTL_UA_POWERON);
2095         ctl_isc_announce_iid(port, iid);
2096
2097         return (0);
2098 }
2099
2100 /*
2101  * Add an initiator to the initiator map.
2102  * Returns iid for success, < 0 for failure.
2103  */
2104 int
2105 ctl_add_initiator(struct ctl_port *port, int iid, uint64_t wwpn, char *name)
2106 {
2107         struct ctl_softc *softc = port->ctl_softc;
2108         time_t best_time;
2109         int i, best;
2110
2111         mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
2112
2113         if (iid >= CTL_MAX_INIT_PER_PORT) {
2114                 printf("%s: WWPN %#jx initiator ID %u > maximum %u!\n",
2115                        __func__, wwpn, iid, CTL_MAX_INIT_PER_PORT);
2116                 free(name, M_CTL);
2117                 return (-1);
2118         }
2119
2120         mtx_lock(&softc->ctl_lock);
2121
2122         if (iid < 0 && (wwpn != 0 || name != NULL)) {
2123                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
2124                         if (wwpn != 0 && wwpn == port->wwpn_iid[i].wwpn) {
2125                                 iid = i;
2126                                 break;
2127                         }
2128                         if (name != NULL && port->wwpn_iid[i].name != NULL &&
2129                             strcmp(name, port->wwpn_iid[i].name) == 0) {
2130                                 iid = i;
2131                                 break;
2132                         }
2133                 }
2134         }
2135
2136         if (iid < 0) {
2137                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
2138                         if (port->wwpn_iid[i].in_use == 0 &&
2139                             port->wwpn_iid[i].wwpn == 0 &&
2140                             port->wwpn_iid[i].name == NULL) {
2141                                 iid = i;
2142                                 break;
2143                         }
2144                 }
2145         }
2146
2147         if (iid < 0) {
2148                 best = -1;
2149                 best_time = INT32_MAX;
2150                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
2151                         if (port->wwpn_iid[i].in_use == 0) {
2152                                 if (port->wwpn_iid[i].last_use < best_time) {
2153                                         best = i;
2154                                         best_time = port->wwpn_iid[i].last_use;
2155                                 }
2156                         }
2157                 }
2158                 iid = best;
2159         }
2160
2161         if (iid < 0) {
2162                 mtx_unlock(&softc->ctl_lock);
2163                 free(name, M_CTL);
2164                 return (-2);
2165         }
2166
2167         if (port->wwpn_iid[iid].in_use > 0 && (wwpn != 0 || name != NULL)) {
2168                 /*
2169                  * This is not an error yet.
2170                  */
2171                 if (wwpn != 0 && wwpn == port->wwpn_iid[iid].wwpn) {
2172 #if 0
2173                         printf("%s: port %d iid %u WWPN %#jx arrived"
2174                             " again\n", __func__, port->targ_port,
2175                             iid, (uintmax_t)wwpn);
2176 #endif
2177                         goto take;
2178                 }
2179                 if (name != NULL && port->wwpn_iid[iid].name != NULL &&
2180                     strcmp(name, port->wwpn_iid[iid].name) == 0) {
2181 #if 0
2182                         printf("%s: port %d iid %u name '%s' arrived"
2183                             " again\n", __func__, port->targ_port,
2184                             iid, name);
2185 #endif
2186                         goto take;
2187                 }
2188
2189                 /*
2190                  * This is an error, but what do we do about it?  The
2191                  * driver is telling us we have a new WWPN for this
2192                  * initiator ID, so we pretty much need to use it.
2193                  */
2194                 printf("%s: port %d iid %u WWPN %#jx '%s' arrived,"
2195                     " but WWPN %#jx '%s' is still at that address\n",
2196                     __func__, port->targ_port, iid, wwpn, name,
2197                     (uintmax_t)port->wwpn_iid[iid].wwpn,
2198                     port->wwpn_iid[iid].name);
2199         }
2200 take:
2201         free(port->wwpn_iid[iid].name, M_CTL);
2202         port->wwpn_iid[iid].name = name;
2203         port->wwpn_iid[iid].wwpn = wwpn;
2204         port->wwpn_iid[iid].in_use++;
2205         mtx_unlock(&softc->ctl_lock);
2206         ctl_isc_announce_iid(port, iid);
2207
2208         return (iid);
2209 }
2210
2211 static int
2212 ctl_create_iid(struct ctl_port *port, int iid, uint8_t *buf)
2213 {
2214         int len;
2215
2216         switch (port->port_type) {
2217         case CTL_PORT_FC:
2218         {
2219                 struct scsi_transportid_fcp *id =
2220                     (struct scsi_transportid_fcp *)buf;
2221                 if (port->wwpn_iid[iid].wwpn == 0)
2222                         return (0);
2223                 memset(id, 0, sizeof(*id));
2224                 id->format_protocol = SCSI_PROTO_FC;
2225                 scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->n_port_name);
2226                 return (sizeof(*id));
2227         }
2228         case CTL_PORT_ISCSI:
2229         {
2230                 struct scsi_transportid_iscsi_port *id =
2231                     (struct scsi_transportid_iscsi_port *)buf;
2232                 if (port->wwpn_iid[iid].name == NULL)
2233                         return (0);
2234                 memset(id, 0, 256);
2235                 id->format_protocol = SCSI_TRN_ISCSI_FORMAT_PORT |
2236                     SCSI_PROTO_ISCSI;
2237                 len = strlcpy(id->iscsi_name, port->wwpn_iid[iid].name, 252) + 1;
2238                 len = roundup2(min(len, 252), 4);
2239                 scsi_ulto2b(len, id->additional_length);
2240                 return (sizeof(*id) + len);
2241         }
2242         case CTL_PORT_SAS:
2243         {
2244                 struct scsi_transportid_sas *id =
2245                     (struct scsi_transportid_sas *)buf;
2246                 if (port->wwpn_iid[iid].wwpn == 0)
2247                         return (0);
2248                 memset(id, 0, sizeof(*id));
2249                 id->format_protocol = SCSI_PROTO_SAS;
2250                 scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->sas_address);
2251                 return (sizeof(*id));
2252         }
2253         default:
2254         {
2255                 struct scsi_transportid_spi *id =
2256                     (struct scsi_transportid_spi *)buf;
2257                 memset(id, 0, sizeof(*id));
2258                 id->format_protocol = SCSI_PROTO_SPI;
2259                 scsi_ulto2b(iid, id->scsi_addr);
2260                 scsi_ulto2b(port->targ_port, id->rel_trgt_port_id);
2261                 return (sizeof(*id));
2262         }
2263         }
2264 }
2265
2266 /*
2267  * Serialize a command that went down the "wrong" side, and so was sent to
2268  * this controller for execution.  The logic is a little different than the
2269  * standard case in ctl_scsiio_precheck().  Errors in this case need to get
2270  * sent back to the other side, but in the success case, we execute the
2271  * command on this side (XFER mode) or tell the other side to execute it
2272  * (SER_ONLY mode).
2273  */
2274 static void
2275 ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio)
2276 {
2277         struct ctl_softc *softc = CTL_SOFTC(ctsio);
2278         struct ctl_port *port = CTL_PORT(ctsio);
2279         union ctl_ha_msg msg_info;
2280         struct ctl_lun *lun;
2281         const struct ctl_cmd_entry *entry;
2282         uint32_t targ_lun;
2283
2284         targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
2285
2286         /* Make sure that we know about this port. */
2287         if (port == NULL || (port->status & CTL_PORT_STATUS_ONLINE) == 0) {
2288                 ctl_set_internal_failure(ctsio, /*sks_valid*/ 0,
2289                                          /*retry_count*/ 1);
2290                 goto badjuju;
2291         }
2292
2293         /* Make sure that we know about this LUN. */
2294         mtx_lock(&softc->ctl_lock);
2295         if (targ_lun >= ctl_max_luns ||
2296             (lun = softc->ctl_luns[targ_lun]) == NULL) {
2297                 mtx_unlock(&softc->ctl_lock);
2298
2299                 /*
2300                  * The other node would not send this request to us unless
2301                  * received announce that we are primary node for this LUN.
2302                  * If this LUN does not exist now, it is probably result of
2303                  * a race, so respond to initiator in the most opaque way.
2304                  */
2305                 ctl_set_busy(ctsio);
2306                 goto badjuju;
2307         }
2308         mtx_lock(&lun->lun_lock);
2309         mtx_unlock(&softc->ctl_lock);
2310
2311         /*
2312          * If the LUN is invalid, pretend that it doesn't exist.
2313          * It will go away as soon as all pending I/Os completed.
2314          */
2315         if (lun->flags & CTL_LUN_DISABLED) {
2316                 mtx_unlock(&lun->lun_lock);
2317                 ctl_set_busy(ctsio);
2318                 goto badjuju;
2319         }
2320
2321         entry = ctl_get_cmd_entry(ctsio, NULL);
2322         if (ctl_scsiio_lun_check(lun, entry, ctsio) != 0) {
2323                 mtx_unlock(&lun->lun_lock);
2324                 goto badjuju;
2325         }
2326
2327         CTL_LUN(ctsio) = lun;
2328         CTL_BACKEND_LUN(ctsio) = lun->be_lun;
2329
2330         /*
2331          * Every I/O goes into the OOA queue for a
2332          * particular LUN, and stays there until completion.
2333          */
2334 #ifdef CTL_TIME_IO
2335         if (TAILQ_EMPTY(&lun->ooa_queue))
2336                 lun->idle_time += getsbinuptime() - lun->last_busy;
2337 #endif
2338         TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2339
2340         switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
2341                 (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr, ctl_ooaq,
2342                  ooa_links))) {
2343         case CTL_ACTION_BLOCK:
2344                 ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
2345                 TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
2346                                   blocked_links);
2347                 mtx_unlock(&lun->lun_lock);
2348                 break;
2349         case CTL_ACTION_PASS:
2350         case CTL_ACTION_SKIP:
2351                 if (softc->ha_mode == CTL_HA_MODE_XFER) {
2352                         ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
2353                         ctl_enqueue_rtr((union ctl_io *)ctsio);
2354                         mtx_unlock(&lun->lun_lock);
2355                 } else {
2356                         ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
2357                         mtx_unlock(&lun->lun_lock);
2358
2359                         /* send msg back to other side */
2360                         msg_info.hdr.original_sc = ctsio->io_hdr.remote_io;
2361                         msg_info.hdr.serializing_sc = (union ctl_io *)ctsio;
2362                         msg_info.hdr.msg_type = CTL_MSG_R2R;
2363                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
2364                             sizeof(msg_info.hdr), M_WAITOK);
2365                 }
2366                 break;
2367         case CTL_ACTION_OVERLAP:
2368                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2369                 mtx_unlock(&lun->lun_lock);
2370                 ctl_set_overlapped_cmd(ctsio);
2371                 goto badjuju;
2372         case CTL_ACTION_OVERLAP_TAG:
2373                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2374                 mtx_unlock(&lun->lun_lock);
2375                 ctl_set_overlapped_tag(ctsio, ctsio->tag_num);
2376                 goto badjuju;
2377         case CTL_ACTION_ERROR:
2378         default:
2379                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2380                 mtx_unlock(&lun->lun_lock);
2381
2382                 ctl_set_internal_failure(ctsio, /*sks_valid*/ 0,
2383                                          /*retry_count*/ 0);
2384 badjuju:
2385                 ctl_copy_sense_data_back((union ctl_io *)ctsio, &msg_info);
2386                 msg_info.hdr.original_sc = ctsio->io_hdr.remote_io;
2387                 msg_info.hdr.serializing_sc = NULL;
2388                 msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
2389                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
2390                     sizeof(msg_info.scsi), M_WAITOK);
2391                 ctl_free_io((union ctl_io *)ctsio);
2392                 break;
2393         }
2394 }
2395
2396 /*
2397  * Returns 0 for success, errno for failure.
2398  */
2399 static void
2400 ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
2401                    struct ctl_ooa *ooa_hdr, struct ctl_ooa_entry *kern_entries)
2402 {
2403         union ctl_io *io;
2404
2405         mtx_lock(&lun->lun_lock);
2406         for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); (io != NULL);
2407              (*cur_fill_num)++, io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2408              ooa_links)) {
2409                 struct ctl_ooa_entry *entry;
2410
2411                 /*
2412                  * If we've got more than we can fit, just count the
2413                  * remaining entries.
2414                  */
2415                 if (*cur_fill_num >= ooa_hdr->alloc_num)
2416                         continue;
2417
2418                 entry = &kern_entries[*cur_fill_num];
2419
2420                 entry->tag_num = io->scsiio.tag_num;
2421                 entry->lun_num = lun->lun;
2422 #ifdef CTL_TIME_IO
2423                 entry->start_bt = io->io_hdr.start_bt;
2424 #endif
2425                 bcopy(io->scsiio.cdb, entry->cdb, io->scsiio.cdb_len);
2426                 entry->cdb_len = io->scsiio.cdb_len;
2427                 if (io->io_hdr.flags & CTL_FLAG_BLOCKED)
2428                         entry->cmd_flags |= CTL_OOACMD_FLAG_BLOCKED;
2429
2430                 if (io->io_hdr.flags & CTL_FLAG_DMA_INPROG)
2431                         entry->cmd_flags |= CTL_OOACMD_FLAG_DMA;
2432
2433                 if (io->io_hdr.flags & CTL_FLAG_ABORT)
2434                         entry->cmd_flags |= CTL_OOACMD_FLAG_ABORT;
2435
2436                 if (io->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR)
2437                         entry->cmd_flags |= CTL_OOACMD_FLAG_RTR;
2438
2439                 if (io->io_hdr.flags & CTL_FLAG_DMA_QUEUED)
2440                         entry->cmd_flags |= CTL_OOACMD_FLAG_DMA_QUEUED;
2441         }
2442         mtx_unlock(&lun->lun_lock);
2443 }
2444
2445 /*
2446  * Escape characters that are illegal or not recommended in XML.
2447  */
2448 int
2449 ctl_sbuf_printf_esc(struct sbuf *sb, char *str, int size)
2450 {
2451         char *end = str + size;
2452         int retval;
2453
2454         retval = 0;
2455
2456         for (; *str && str < end; str++) {
2457                 switch (*str) {
2458                 case '&':
2459                         retval = sbuf_printf(sb, "&amp;");
2460                         break;
2461                 case '>':
2462                         retval = sbuf_printf(sb, "&gt;");
2463                         break;
2464                 case '<':
2465                         retval = sbuf_printf(sb, "&lt;");
2466                         break;
2467                 default:
2468                         retval = sbuf_putc(sb, *str);
2469                         break;
2470                 }
2471
2472                 if (retval != 0)
2473                         break;
2474
2475         }
2476
2477         return (retval);
2478 }
2479
2480 static void
2481 ctl_id_sbuf(struct ctl_devid *id, struct sbuf *sb)
2482 {
2483         struct scsi_vpd_id_descriptor *desc;
2484         int i;
2485
2486         if (id == NULL || id->len < 4)
2487                 return;
2488         desc = (struct scsi_vpd_id_descriptor *)id->data;
2489         switch (desc->id_type & SVPD_ID_TYPE_MASK) {
2490         case SVPD_ID_TYPE_T10:
2491                 sbuf_printf(sb, "t10.");
2492                 break;
2493         case SVPD_ID_TYPE_EUI64:
2494                 sbuf_printf(sb, "eui.");
2495                 break;
2496         case SVPD_ID_TYPE_NAA:
2497                 sbuf_printf(sb, "naa.");
2498                 break;
2499         case SVPD_ID_TYPE_SCSI_NAME:
2500                 break;
2501         }
2502         switch (desc->proto_codeset & SVPD_ID_CODESET_MASK) {
2503         case SVPD_ID_CODESET_BINARY:
2504                 for (i = 0; i < desc->length; i++)
2505                         sbuf_printf(sb, "%02x", desc->identifier[i]);
2506                 break;
2507         case SVPD_ID_CODESET_ASCII:
2508                 sbuf_printf(sb, "%.*s", (int)desc->length,
2509                     (char *)desc->identifier);
2510                 break;
2511         case SVPD_ID_CODESET_UTF8:
2512                 sbuf_printf(sb, "%s", (char *)desc->identifier);
2513                 break;
2514         }
2515 }
2516
2517 static int
2518 ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
2519           struct thread *td)
2520 {
2521         struct ctl_softc *softc = dev->si_drv1;
2522         struct ctl_port *port;
2523         struct ctl_lun *lun;
2524         int retval;
2525
2526         retval = 0;
2527
2528         switch (cmd) {
2529         case CTL_IO:
2530                 retval = ctl_ioctl_io(dev, cmd, addr, flag, td);
2531                 break;
2532         case CTL_ENABLE_PORT:
2533         case CTL_DISABLE_PORT:
2534         case CTL_SET_PORT_WWNS: {
2535                 struct ctl_port *port;
2536                 struct ctl_port_entry *entry;
2537
2538                 entry = (struct ctl_port_entry *)addr;
2539
2540                 mtx_lock(&softc->ctl_lock);
2541                 STAILQ_FOREACH(port, &softc->port_list, links) {
2542                         int action, done;
2543
2544                         if (port->targ_port < softc->port_min ||
2545                             port->targ_port >= softc->port_max)
2546                                 continue;
2547
2548                         action = 0;
2549                         done = 0;
2550                         if ((entry->port_type == CTL_PORT_NONE)
2551                          && (entry->targ_port == port->targ_port)) {
2552                                 /*
2553                                  * If the user only wants to enable or
2554                                  * disable or set WWNs on a specific port,
2555                                  * do the operation and we're done.
2556                                  */
2557                                 action = 1;
2558                                 done = 1;
2559                         } else if (entry->port_type & port->port_type) {
2560                                 /*
2561                                  * Compare the user's type mask with the
2562                                  * particular frontend type to see if we
2563                                  * have a match.
2564                                  */
2565                                 action = 1;
2566                                 done = 0;
2567
2568                                 /*
2569                                  * Make sure the user isn't trying to set
2570                                  * WWNs on multiple ports at the same time.
2571                                  */
2572                                 if (cmd == CTL_SET_PORT_WWNS) {
2573                                         printf("%s: Can't set WWNs on "
2574                                                "multiple ports\n", __func__);
2575                                         retval = EINVAL;
2576                                         break;
2577                                 }
2578                         }
2579                         if (action == 0)
2580                                 continue;
2581
2582                         /*
2583                          * XXX KDM we have to drop the lock here, because
2584                          * the online/offline operations can potentially
2585                          * block.  We need to reference count the frontends
2586                          * so they can't go away,
2587                          */
2588                         if (cmd == CTL_ENABLE_PORT) {
2589                                 mtx_unlock(&softc->ctl_lock);
2590                                 ctl_port_online(port);
2591                                 mtx_lock(&softc->ctl_lock);
2592                         } else if (cmd == CTL_DISABLE_PORT) {
2593                                 mtx_unlock(&softc->ctl_lock);
2594                                 ctl_port_offline(port);
2595                                 mtx_lock(&softc->ctl_lock);
2596                         } else if (cmd == CTL_SET_PORT_WWNS) {
2597                                 ctl_port_set_wwns(port,
2598                                     (entry->flags & CTL_PORT_WWNN_VALID) ?
2599                                     1 : 0, entry->wwnn,
2600                                     (entry->flags & CTL_PORT_WWPN_VALID) ?
2601                                     1 : 0, entry->wwpn);
2602                         }
2603                         if (done != 0)
2604                                 break;
2605                 }
2606                 mtx_unlock(&softc->ctl_lock);
2607                 break;
2608         }
2609         case CTL_GET_OOA: {
2610                 struct ctl_ooa *ooa_hdr;
2611                 struct ctl_ooa_entry *entries;
2612                 uint32_t cur_fill_num;
2613
2614                 ooa_hdr = (struct ctl_ooa *)addr;
2615
2616                 if ((ooa_hdr->alloc_len == 0)
2617                  || (ooa_hdr->alloc_num == 0)) {
2618                         printf("%s: CTL_GET_OOA: alloc len %u and alloc num %u "
2619                                "must be non-zero\n", __func__,
2620                                ooa_hdr->alloc_len, ooa_hdr->alloc_num);
2621                         retval = EINVAL;
2622                         break;
2623                 }
2624
2625                 if (ooa_hdr->alloc_len != (ooa_hdr->alloc_num *
2626                     sizeof(struct ctl_ooa_entry))) {
2627                         printf("%s: CTL_GET_OOA: alloc len %u must be alloc "
2628                                "num %d * sizeof(struct ctl_ooa_entry) %zd\n",
2629                                __func__, ooa_hdr->alloc_len,
2630                                ooa_hdr->alloc_num,sizeof(struct ctl_ooa_entry));
2631                         retval = EINVAL;
2632                         break;
2633                 }
2634
2635                 entries = malloc(ooa_hdr->alloc_len, M_CTL, M_WAITOK | M_ZERO);
2636                 if (entries == NULL) {
2637                         printf("%s: could not allocate %d bytes for OOA "
2638                                "dump\n", __func__, ooa_hdr->alloc_len);
2639                         retval = ENOMEM;
2640                         break;
2641                 }
2642
2643                 mtx_lock(&softc->ctl_lock);
2644                 if ((ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) == 0 &&
2645                     (ooa_hdr->lun_num >= ctl_max_luns ||
2646                      softc->ctl_luns[ooa_hdr->lun_num] == NULL)) {
2647                         mtx_unlock(&softc->ctl_lock);
2648                         free(entries, M_CTL);
2649                         printf("%s: CTL_GET_OOA: invalid LUN %ju\n",
2650                                __func__, (uintmax_t)ooa_hdr->lun_num);
2651                         retval = EINVAL;
2652                         break;
2653                 }
2654
2655                 cur_fill_num = 0;
2656
2657                 if (ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) {
2658                         STAILQ_FOREACH(lun, &softc->lun_list, links) {
2659                                 ctl_ioctl_fill_ooa(lun, &cur_fill_num,
2660                                     ooa_hdr, entries);
2661                         }
2662                 } else {
2663                         lun = softc->ctl_luns[ooa_hdr->lun_num];
2664                         ctl_ioctl_fill_ooa(lun, &cur_fill_num, ooa_hdr,
2665                             entries);
2666                 }
2667                 mtx_unlock(&softc->ctl_lock);
2668
2669                 ooa_hdr->fill_num = min(cur_fill_num, ooa_hdr->alloc_num);
2670                 ooa_hdr->fill_len = ooa_hdr->fill_num *
2671                         sizeof(struct ctl_ooa_entry);
2672                 retval = copyout(entries, ooa_hdr->entries, ooa_hdr->fill_len);
2673                 if (retval != 0) {
2674                         printf("%s: error copying out %d bytes for OOA dump\n",
2675                                __func__, ooa_hdr->fill_len);
2676                 }
2677
2678                 getbinuptime(&ooa_hdr->cur_bt);
2679
2680                 if (cur_fill_num > ooa_hdr->alloc_num) {
2681                         ooa_hdr->dropped_num = cur_fill_num -ooa_hdr->alloc_num;
2682                         ooa_hdr->status = CTL_OOA_NEED_MORE_SPACE;
2683                 } else {
2684                         ooa_hdr->dropped_num = 0;
2685                         ooa_hdr->status = CTL_OOA_OK;
2686                 }
2687
2688                 free(entries, M_CTL);
2689                 break;
2690         }
2691         case CTL_DELAY_IO: {
2692                 struct ctl_io_delay_info *delay_info;
2693
2694                 delay_info = (struct ctl_io_delay_info *)addr;
2695
2696 #ifdef CTL_IO_DELAY
2697                 mtx_lock(&softc->ctl_lock);
2698                 if (delay_info->lun_id >= ctl_max_luns ||
2699                     (lun = softc->ctl_luns[delay_info->lun_id]) == NULL) {
2700                         mtx_unlock(&softc->ctl_lock);
2701                         delay_info->status = CTL_DELAY_STATUS_INVALID_LUN;
2702                         break;
2703                 }
2704                 mtx_lock(&lun->lun_lock);
2705                 mtx_unlock(&softc->ctl_lock);
2706                 delay_info->status = CTL_DELAY_STATUS_OK;
2707                 switch (delay_info->delay_type) {
2708                 case CTL_DELAY_TYPE_CONT:
2709                 case CTL_DELAY_TYPE_ONESHOT:
2710                         break;
2711                 default:
2712                         delay_info->status = CTL_DELAY_STATUS_INVALID_TYPE;
2713                         break;
2714                 }
2715                 switch (delay_info->delay_loc) {
2716                 case CTL_DELAY_LOC_DATAMOVE:
2717                         lun->delay_info.datamove_type = delay_info->delay_type;
2718                         lun->delay_info.datamove_delay = delay_info->delay_secs;
2719                         break;
2720                 case CTL_DELAY_LOC_DONE:
2721                         lun->delay_info.done_type = delay_info->delay_type;
2722                         lun->delay_info.done_delay = delay_info->delay_secs;
2723                         break;
2724                 default:
2725                         delay_info->status = CTL_DELAY_STATUS_INVALID_LOC;
2726                         break;
2727                 }
2728                 mtx_unlock(&lun->lun_lock);
2729 #else
2730                 delay_info->status = CTL_DELAY_STATUS_NOT_IMPLEMENTED;
2731 #endif /* CTL_IO_DELAY */
2732                 break;
2733         }
2734         case CTL_ERROR_INJECT: {
2735                 struct ctl_error_desc *err_desc, *new_err_desc;
2736
2737                 err_desc = (struct ctl_error_desc *)addr;
2738
2739                 new_err_desc = malloc(sizeof(*new_err_desc), M_CTL,
2740                                       M_WAITOK | M_ZERO);
2741                 bcopy(err_desc, new_err_desc, sizeof(*new_err_desc));
2742
2743                 mtx_lock(&softc->ctl_lock);
2744                 if (err_desc->lun_id >= ctl_max_luns ||
2745                     (lun = softc->ctl_luns[err_desc->lun_id]) == NULL) {
2746                         mtx_unlock(&softc->ctl_lock);
2747                         free(new_err_desc, M_CTL);
2748                         printf("%s: CTL_ERROR_INJECT: invalid LUN %ju\n",
2749                                __func__, (uintmax_t)err_desc->lun_id);
2750                         retval = EINVAL;
2751                         break;
2752                 }
2753                 mtx_lock(&lun->lun_lock);
2754                 mtx_unlock(&softc->ctl_lock);
2755
2756                 /*
2757                  * We could do some checking here to verify the validity
2758                  * of the request, but given the complexity of error
2759                  * injection requests, the checking logic would be fairly
2760                  * complex.
2761                  *
2762                  * For now, if the request is invalid, it just won't get
2763                  * executed and might get deleted.
2764                  */
2765                 STAILQ_INSERT_TAIL(&lun->error_list, new_err_desc, links);
2766
2767                 /*
2768                  * XXX KDM check to make sure the serial number is unique,
2769                  * in case we somehow manage to wrap.  That shouldn't
2770                  * happen for a very long time, but it's the right thing to
2771                  * do.
2772                  */
2773                 new_err_desc->serial = lun->error_serial;
2774                 err_desc->serial = lun->error_serial;
2775                 lun->error_serial++;
2776
2777                 mtx_unlock(&lun->lun_lock);
2778                 break;
2779         }
2780         case CTL_ERROR_INJECT_DELETE: {
2781                 struct ctl_error_desc *delete_desc, *desc, *desc2;
2782                 int delete_done;
2783
2784                 delete_desc = (struct ctl_error_desc *)addr;
2785                 delete_done = 0;
2786
2787                 mtx_lock(&softc->ctl_lock);
2788                 if (delete_desc->lun_id >= ctl_max_luns ||
2789                     (lun = softc->ctl_luns[delete_desc->lun_id]) == NULL) {
2790                         mtx_unlock(&softc->ctl_lock);
2791                         printf("%s: CTL_ERROR_INJECT_DELETE: invalid LUN %ju\n",
2792                                __func__, (uintmax_t)delete_desc->lun_id);
2793                         retval = EINVAL;
2794                         break;
2795                 }
2796                 mtx_lock(&lun->lun_lock);
2797                 mtx_unlock(&softc->ctl_lock);
2798                 STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
2799                         if (desc->serial != delete_desc->serial)
2800                                 continue;
2801
2802                         STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc,
2803                                       links);
2804                         free(desc, M_CTL);
2805                         delete_done = 1;
2806                 }
2807                 mtx_unlock(&lun->lun_lock);
2808                 if (delete_done == 0) {
2809                         printf("%s: CTL_ERROR_INJECT_DELETE: can't find "
2810                                "error serial %ju on LUN %u\n", __func__,
2811                                delete_desc->serial, delete_desc->lun_id);
2812                         retval = EINVAL;
2813                         break;
2814                 }
2815                 break;
2816         }
2817         case CTL_DUMP_STRUCTS: {
2818                 int j, k;
2819                 struct ctl_port *port;
2820                 struct ctl_frontend *fe;
2821
2822                 mtx_lock(&softc->ctl_lock);
2823                 printf("CTL Persistent Reservation information start:\n");
2824                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
2825                         mtx_lock(&lun->lun_lock);
2826                         if ((lun->flags & CTL_LUN_DISABLED) != 0) {
2827                                 mtx_unlock(&lun->lun_lock);
2828                                 continue;
2829                         }
2830
2831                         for (j = 0; j < ctl_max_ports; j++) {
2832                                 if (lun->pr_keys[j] == NULL)
2833                                         continue;
2834                                 for (k = 0; k < CTL_MAX_INIT_PER_PORT; k++){
2835                                         if (lun->pr_keys[j][k] == 0)
2836                                                 continue;
2837                                         printf("  LUN %ju port %d iid %d key "
2838                                                "%#jx\n", lun->lun, j, k,
2839                                                (uintmax_t)lun->pr_keys[j][k]);
2840                                 }
2841                         }
2842                         mtx_unlock(&lun->lun_lock);
2843                 }
2844                 printf("CTL Persistent Reservation information end\n");
2845                 printf("CTL Ports:\n");
2846                 STAILQ_FOREACH(port, &softc->port_list, links) {
2847                         printf("  Port %d '%s' Frontend '%s' Type %u pp %d vp %d WWNN "
2848                                "%#jx WWPN %#jx\n", port->targ_port, port->port_name,
2849                                port->frontend->name, port->port_type,
2850                                port->physical_port, port->virtual_port,
2851                                (uintmax_t)port->wwnn, (uintmax_t)port->wwpn);
2852                         for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
2853                                 if (port->wwpn_iid[j].in_use == 0 &&
2854                                     port->wwpn_iid[j].wwpn == 0 &&
2855                                     port->wwpn_iid[j].name == NULL)
2856                                         continue;
2857
2858                                 printf("    iid %u use %d WWPN %#jx '%s'\n",
2859                                     j, port->wwpn_iid[j].in_use,
2860                                     (uintmax_t)port->wwpn_iid[j].wwpn,
2861                                     port->wwpn_iid[j].name);
2862                         }
2863                 }
2864                 printf("CTL Port information end\n");
2865                 mtx_unlock(&softc->ctl_lock);
2866                 /*
2867                  * XXX KDM calling this without a lock.  We'd likely want
2868                  * to drop the lock before calling the frontend's dump
2869                  * routine anyway.
2870                  */
2871                 printf("CTL Frontends:\n");
2872                 STAILQ_FOREACH(fe, &softc->fe_list, links) {
2873                         printf("  Frontend '%s'\n", fe->name);
2874                         if (fe->fe_dump != NULL)
2875                                 fe->fe_dump();
2876                 }
2877                 printf("CTL Frontend information end\n");
2878                 break;
2879         }
2880         case CTL_LUN_REQ: {
2881                 struct ctl_lun_req *lun_req;
2882                 struct ctl_backend_driver *backend;
2883                 void *packed;
2884                 nvlist_t *tmp_args_nvl;
2885                 size_t packed_len;
2886
2887                 lun_req = (struct ctl_lun_req *)addr;
2888                 tmp_args_nvl = lun_req->args_nvl;
2889
2890                 backend = ctl_backend_find(lun_req->backend);
2891                 if (backend == NULL) {
2892                         lun_req->status = CTL_LUN_ERROR;
2893                         snprintf(lun_req->error_str,
2894                                  sizeof(lun_req->error_str),
2895                                  "Backend \"%s\" not found.",
2896                                  lun_req->backend);
2897                         break;
2898                 }
2899
2900                 if (lun_req->args != NULL) {
2901                         packed = malloc(lun_req->args_len, M_CTL, M_WAITOK);
2902                         if (copyin(lun_req->args, packed, lun_req->args_len) != 0) {
2903                                 free(packed, M_CTL);
2904                                 lun_req->status = CTL_LUN_ERROR;
2905                                 snprintf(lun_req->error_str, sizeof(lun_req->error_str),
2906                                     "Cannot copyin args.");
2907                                 break;
2908                         }
2909                         lun_req->args_nvl = nvlist_unpack(packed,
2910                             lun_req->args_len, 0);
2911                         free(packed, M_CTL);
2912
2913                         if (lun_req->args_nvl == NULL) {
2914                                 lun_req->status = CTL_LUN_ERROR;
2915                                 snprintf(lun_req->error_str, sizeof(lun_req->error_str),
2916                                     "Cannot unpack args nvlist.");
2917                                 break;
2918                         }
2919                 } else
2920                         lun_req->args_nvl = nvlist_create(0);
2921
2922                 retval = backend->ioctl(dev, cmd, addr, flag, td);
2923                 nvlist_destroy(lun_req->args_nvl);
2924                 lun_req->args_nvl = tmp_args_nvl;
2925
2926                 if (lun_req->result_nvl != NULL) {
2927                         if (lun_req->result != NULL) {
2928                                 packed = nvlist_pack(lun_req->result_nvl,
2929                                     &packed_len);
2930                                 if (packed == NULL) {
2931                                         lun_req->status = CTL_LUN_ERROR;
2932                                         snprintf(lun_req->error_str,
2933                                             sizeof(lun_req->error_str),
2934                                             "Cannot pack result nvlist.");
2935                                         break;
2936                                 }
2937
2938                                 if (packed_len > lun_req->result_len) {
2939                                         lun_req->status = CTL_LUN_ERROR;
2940                                         snprintf(lun_req->error_str,
2941                                             sizeof(lun_req->error_str),
2942                                             "Result nvlist too large.");
2943                                         free(packed, M_NVLIST);
2944                                         break;
2945                                 }
2946
2947                                 if (copyout(packed, lun_req->result, packed_len)) {
2948                                         lun_req->status = CTL_LUN_ERROR;
2949                                         snprintf(lun_req->error_str,
2950                                             sizeof(lun_req->error_str),
2951                                             "Cannot copyout() the result.");
2952                                         free(packed, M_NVLIST);
2953                                         break;
2954                                 }
2955
2956                                 lun_req->result_len = packed_len;
2957                                 free(packed, M_NVLIST);
2958                         }
2959
2960                         nvlist_destroy(lun_req->result_nvl);
2961                 }
2962                 break;
2963         }
2964         case CTL_LUN_LIST: {
2965                 struct sbuf *sb;
2966                 struct ctl_lun_list *list;
2967                 const char *name, *value;
2968                 void *cookie;
2969                 int type;
2970
2971                 list = (struct ctl_lun_list *)addr;
2972
2973                 /*
2974                  * Allocate a fixed length sbuf here, based on the length
2975                  * of the user's buffer.  We could allocate an auto-extending
2976                  * buffer, and then tell the user how much larger our
2977                  * amount of data is than his buffer, but that presents
2978                  * some problems:
2979                  *
2980                  * 1.  The sbuf(9) routines use a blocking malloc, and so
2981                  *     we can't hold a lock while calling them with an
2982                  *     auto-extending buffer.
2983                  *
2984                  * 2.  There is not currently a LUN reference counting
2985                  *     mechanism, outside of outstanding transactions on
2986                  *     the LUN's OOA queue.  So a LUN could go away on us
2987                  *     while we're getting the LUN number, backend-specific
2988                  *     information, etc.  Thus, given the way things
2989                  *     currently work, we need to hold the CTL lock while
2990                  *     grabbing LUN information.
2991                  *
2992                  * So, from the user's standpoint, the best thing to do is
2993                  * allocate what he thinks is a reasonable buffer length,
2994                  * and then if he gets a CTL_LUN_LIST_NEED_MORE_SPACE error,
2995                  * double the buffer length and try again.  (And repeat
2996                  * that until he succeeds.)
2997                  */
2998                 sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
2999                 if (sb == NULL) {
3000                         list->status = CTL_LUN_LIST_ERROR;
3001                         snprintf(list->error_str, sizeof(list->error_str),
3002                                  "Unable to allocate %d bytes for LUN list",
3003                                  list->alloc_len);
3004                         break;
3005                 }
3006
3007                 sbuf_printf(sb, "<ctllunlist>\n");
3008
3009                 mtx_lock(&softc->ctl_lock);
3010                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
3011                         mtx_lock(&lun->lun_lock);
3012                         retval = sbuf_printf(sb, "<lun id=\"%ju\">\n",
3013                                              (uintmax_t)lun->lun);
3014
3015                         /*
3016                          * Bail out as soon as we see that we've overfilled
3017                          * the buffer.
3018                          */
3019                         if (retval != 0)
3020                                 break;
3021
3022                         retval = sbuf_printf(sb, "\t<backend_type>%s"
3023                                              "</backend_type>\n",
3024                                              (lun->backend == NULL) ?  "none" :
3025                                              lun->backend->name);
3026
3027                         if (retval != 0)
3028                                 break;
3029
3030                         retval = sbuf_printf(sb, "\t<lun_type>%d</lun_type>\n",
3031                                              lun->be_lun->lun_type);
3032
3033                         if (retval != 0)
3034                                 break;
3035
3036                         if (lun->backend == NULL) {
3037                                 retval = sbuf_printf(sb, "</lun>\n");
3038                                 if (retval != 0)
3039                                         break;
3040                                 continue;
3041                         }
3042
3043                         retval = sbuf_printf(sb, "\t<size>%ju</size>\n",
3044                                              (lun->be_lun->maxlba > 0) ?
3045                                              lun->be_lun->maxlba + 1 : 0);
3046
3047                         if (retval != 0)
3048                                 break;
3049
3050                         retval = sbuf_printf(sb, "\t<blocksize>%u</blocksize>\n",
3051                                              lun->be_lun->blocksize);
3052
3053                         if (retval != 0)
3054                                 break;
3055
3056                         retval = sbuf_printf(sb, "\t<serial_number>");
3057
3058                         if (retval != 0)
3059                                 break;
3060
3061                         retval = ctl_sbuf_printf_esc(sb,
3062                             lun->be_lun->serial_num,
3063                             sizeof(lun->be_lun->serial_num));
3064
3065                         if (retval != 0)
3066                                 break;
3067
3068                         retval = sbuf_printf(sb, "</serial_number>\n");
3069
3070                         if (retval != 0)
3071                                 break;
3072
3073                         retval = sbuf_printf(sb, "\t<device_id>");
3074
3075                         if (retval != 0)
3076                                 break;
3077
3078                         retval = ctl_sbuf_printf_esc(sb,
3079                             lun->be_lun->device_id,
3080                             sizeof(lun->be_lun->device_id));
3081
3082                         if (retval != 0)
3083                                 break;
3084
3085                         retval = sbuf_printf(sb, "</device_id>\n");
3086
3087                         if (retval != 0)
3088                                 break;
3089
3090                         if (lun->backend->lun_info != NULL) {
3091                                 retval = lun->backend->lun_info(lun->be_lun->be_lun, sb);
3092                                 if (retval != 0)
3093                                         break;
3094                         }
3095
3096                         cookie = NULL;
3097                         while ((name = nvlist_next(lun->be_lun->options, &type,
3098                             &cookie)) != NULL) {
3099                                 sbuf_printf(sb, "\t<%s>", name);
3100
3101                                 if (type == NV_TYPE_STRING) {
3102                                         value = dnvlist_get_string(
3103                                             lun->be_lun->options, name, NULL);
3104                                         if (value != NULL)
3105                                                 sbuf_printf(sb, "%s", value);
3106                                 }
3107
3108                                 sbuf_printf(sb, "</%s>\n", name);
3109                         }
3110
3111                         retval = sbuf_printf(sb, "</lun>\n");
3112
3113                         if (retval != 0)
3114                                 break;
3115                         mtx_unlock(&lun->lun_lock);
3116                 }
3117                 if (lun != NULL)
3118                         mtx_unlock(&lun->lun_lock);
3119                 mtx_unlock(&softc->ctl_lock);
3120
3121                 if ((retval != 0)
3122                  || ((retval = sbuf_printf(sb, "</ctllunlist>\n")) != 0)) {
3123                         retval = 0;
3124                         sbuf_delete(sb);
3125                         list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3126                         snprintf(list->error_str, sizeof(list->error_str),
3127                                  "Out of space, %d bytes is too small",
3128                                  list->alloc_len);
3129                         break;
3130                 }
3131
3132                 sbuf_finish(sb);
3133
3134                 retval = copyout(sbuf_data(sb), list->lun_xml,
3135                                  sbuf_len(sb) + 1);
3136
3137                 list->fill_len = sbuf_len(sb) + 1;
3138                 list->status = CTL_LUN_LIST_OK;
3139                 sbuf_delete(sb);
3140                 break;
3141         }
3142         case CTL_ISCSI: {
3143                 struct ctl_iscsi *ci;
3144                 struct ctl_frontend *fe;
3145
3146                 ci = (struct ctl_iscsi *)addr;
3147
3148                 fe = ctl_frontend_find("iscsi");
3149                 if (fe == NULL) {
3150                         ci->status = CTL_ISCSI_ERROR;
3151                         snprintf(ci->error_str, sizeof(ci->error_str),
3152                             "Frontend \"iscsi\" not found.");
3153                         break;
3154                 }
3155
3156                 retval = fe->ioctl(dev, cmd, addr, flag, td);
3157                 break;
3158         }
3159         case CTL_PORT_REQ: {
3160                 struct ctl_req *req;
3161                 struct ctl_frontend *fe;
3162                 void *packed;
3163                 nvlist_t *tmp_args_nvl;
3164                 size_t packed_len;
3165
3166                 req = (struct ctl_req *)addr;
3167                 tmp_args_nvl = req->args_nvl;
3168
3169                 fe = ctl_frontend_find(req->driver);
3170                 if (fe == NULL) {
3171                         req->status = CTL_LUN_ERROR;
3172                         snprintf(req->error_str, sizeof(req->error_str),
3173                             "Frontend \"%s\" not found.", req->driver);
3174                         break;
3175                 }
3176
3177                 if (req->args != NULL) {
3178                         packed = malloc(req->args_len, M_CTL, M_WAITOK);
3179                         if (copyin(req->args, packed, req->args_len) != 0) {
3180                                 free(packed, M_CTL);
3181                                 req->status = CTL_LUN_ERROR;
3182                                 snprintf(req->error_str, sizeof(req->error_str),
3183                                     "Cannot copyin args.");
3184                                 break;
3185                         }
3186                         req->args_nvl = nvlist_unpack(packed,
3187                             req->args_len, 0);
3188                         free(packed, M_CTL);
3189
3190                         if (req->args_nvl == NULL) {
3191                                 req->status = CTL_LUN_ERROR;
3192                                 snprintf(req->error_str, sizeof(req->error_str),
3193                                     "Cannot unpack args nvlist.");
3194                                 break;
3195                         }
3196                 } else
3197                         req->args_nvl = nvlist_create(0);
3198
3199                 if (fe->ioctl)
3200                         retval = fe->ioctl(dev, cmd, addr, flag, td);
3201                 else
3202                         retval = ENODEV;
3203
3204                 nvlist_destroy(req->args_nvl);
3205                 req->args_nvl = tmp_args_nvl;
3206
3207                 if (req->result_nvl != NULL) {
3208                         if (req->result != NULL) {
3209                                 packed = nvlist_pack(req->result_nvl,
3210                                     &packed_len);
3211                                 if (packed == NULL) {
3212                                         req->status = CTL_LUN_ERROR;
3213                                         snprintf(req->error_str,
3214                                             sizeof(req->error_str),
3215                                             "Cannot pack result nvlist.");
3216                                         break;
3217                                 }
3218
3219                                 if (packed_len > req->result_len) {
3220                                         req->status = CTL_LUN_ERROR;
3221                                         snprintf(req->error_str,
3222                                             sizeof(req->error_str),
3223                                             "Result nvlist too large.");
3224                                         free(packed, M_NVLIST);
3225                                         break;
3226                                 }
3227
3228                                 if (copyout(packed, req->result, packed_len)) {
3229                                         req->status = CTL_LUN_ERROR;
3230                                         snprintf(req->error_str,
3231                                             sizeof(req->error_str),
3232                                             "Cannot copyout() the result.");
3233                                         free(packed, M_NVLIST);
3234                                         break;
3235                                 }
3236
3237                                 req->result_len = packed_len;
3238                                 free(packed, M_NVLIST);
3239                         }
3240
3241                         nvlist_destroy(req->result_nvl);
3242                 }
3243                 break;
3244         }
3245         case CTL_PORT_LIST: {
3246                 struct sbuf *sb;
3247                 struct ctl_port *port;
3248                 struct ctl_lun_list *list;
3249                 const char *name, *value;
3250                 void *cookie;
3251                 int j, type;
3252                 uint32_t plun;
3253
3254                 list = (struct ctl_lun_list *)addr;
3255
3256                 sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3257                 if (sb == NULL) {
3258                         list->status = CTL_LUN_LIST_ERROR;
3259                         snprintf(list->error_str, sizeof(list->error_str),
3260                                  "Unable to allocate %d bytes for LUN list",
3261                                  list->alloc_len);
3262                         break;
3263                 }
3264
3265                 sbuf_printf(sb, "<ctlportlist>\n");
3266
3267                 mtx_lock(&softc->ctl_lock);
3268                 STAILQ_FOREACH(port, &softc->port_list, links) {
3269                         retval = sbuf_printf(sb, "<targ_port id=\"%ju\">\n",
3270                                              (uintmax_t)port->targ_port);
3271
3272                         /*
3273                          * Bail out as soon as we see that we've overfilled
3274                          * the buffer.
3275                          */
3276                         if (retval != 0)
3277                                 break;
3278
3279                         retval = sbuf_printf(sb, "\t<frontend_type>%s"
3280                             "</frontend_type>\n", port->frontend->name);
3281                         if (retval != 0)
3282                                 break;
3283
3284                         retval = sbuf_printf(sb, "\t<port_type>%d</port_type>\n",
3285                                              port->port_type);
3286                         if (retval != 0)
3287                                 break;
3288
3289                         retval = sbuf_printf(sb, "\t<online>%s</online>\n",
3290                             (port->status & CTL_PORT_STATUS_ONLINE) ? "YES" : "NO");
3291                         if (retval != 0)
3292                                 break;
3293
3294                         retval = sbuf_printf(sb, "\t<port_name>%s</port_name>\n",
3295                             port->port_name);
3296                         if (retval != 0)
3297                                 break;
3298
3299                         retval = sbuf_printf(sb, "\t<physical_port>%d</physical_port>\n",
3300                             port->physical_port);
3301                         if (retval != 0)
3302                                 break;
3303
3304                         retval = sbuf_printf(sb, "\t<virtual_port>%d</virtual_port>\n",
3305                             port->virtual_port);
3306                         if (retval != 0)
3307                                 break;
3308
3309                         if (port->target_devid != NULL) {
3310                                 sbuf_printf(sb, "\t<target>");
3311                                 ctl_id_sbuf(port->target_devid, sb);
3312                                 sbuf_printf(sb, "</target>\n");
3313                         }
3314
3315                         if (port->port_devid != NULL) {
3316                                 sbuf_printf(sb, "\t<port>");
3317                                 ctl_id_sbuf(port->port_devid, sb);
3318                                 sbuf_printf(sb, "</port>\n");
3319                         }
3320
3321                         if (port->port_info != NULL) {
3322                                 retval = port->port_info(port->onoff_arg, sb);
3323                                 if (retval != 0)
3324                                         break;
3325                         }
3326
3327                         cookie = NULL;
3328                         while ((name = nvlist_next(port->options, &type,
3329                             &cookie)) != NULL) {
3330                                 sbuf_printf(sb, "\t<%s>", name);
3331
3332                                 if (type == NV_TYPE_STRING) {
3333                                         value = dnvlist_get_string(port->options,
3334                                             name, NULL);
3335                                         if (value != NULL)
3336                                                 sbuf_printf(sb, "%s", value);
3337                                 }
3338
3339                                 sbuf_printf(sb, "</%s>\n", name);
3340                         }
3341
3342                         if (port->lun_map != NULL) {
3343                                 sbuf_printf(sb, "\t<lun_map>on</lun_map>\n");
3344                                 for (j = 0; j < port->lun_map_size; j++) {
3345                                         plun = ctl_lun_map_from_port(port, j);
3346                                         if (plun == UINT32_MAX)
3347                                                 continue;
3348                                         sbuf_printf(sb,
3349                                             "\t<lun id=\"%u\">%u</lun>\n",
3350                                             j, plun);
3351                                 }
3352                         }
3353
3354                         for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
3355                                 if (port->wwpn_iid[j].in_use == 0 ||
3356                                     (port->wwpn_iid[j].wwpn == 0 &&
3357                                      port->wwpn_iid[j].name == NULL))
3358                                         continue;
3359
3360                                 if (port->wwpn_iid[j].name != NULL)
3361                                         retval = sbuf_printf(sb,
3362                                             "\t<initiator id=\"%u\">%s</initiator>\n",
3363                                             j, port->wwpn_iid[j].name);
3364                                 else
3365                                         retval = sbuf_printf(sb,
3366                                             "\t<initiator id=\"%u\">naa.%08jx</initiator>\n",
3367                                             j, port->wwpn_iid[j].wwpn);
3368                                 if (retval != 0)
3369                                         break;
3370                         }
3371                         if (retval != 0)
3372                                 break;
3373
3374                         retval = sbuf_printf(sb, "</targ_port>\n");
3375                         if (retval != 0)
3376                                 break;
3377                 }
3378                 mtx_unlock(&softc->ctl_lock);
3379
3380                 if ((retval != 0)
3381                  || ((retval = sbuf_printf(sb, "</ctlportlist>\n")) != 0)) {
3382                         retval = 0;
3383                         sbuf_delete(sb);
3384                         list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3385                         snprintf(list->error_str, sizeof(list->error_str),
3386                                  "Out of space, %d bytes is too small",
3387                                  list->alloc_len);
3388                         break;
3389                 }
3390
3391                 sbuf_finish(sb);
3392
3393                 retval = copyout(sbuf_data(sb), list->lun_xml,
3394                                  sbuf_len(sb) + 1);
3395
3396                 list->fill_len = sbuf_len(sb) + 1;
3397                 list->status = CTL_LUN_LIST_OK;
3398                 sbuf_delete(sb);
3399                 break;
3400         }
3401         case CTL_LUN_MAP: {
3402                 struct ctl_lun_map *lm  = (struct ctl_lun_map *)addr;
3403                 struct ctl_port *port;
3404
3405                 mtx_lock(&softc->ctl_lock);
3406                 if (lm->port < softc->port_min ||
3407                     lm->port >= softc->port_max ||
3408                     (port = softc->ctl_ports[lm->port]) == NULL) {
3409                         mtx_unlock(&softc->ctl_lock);
3410                         return (ENXIO);
3411                 }
3412                 if (port->status & CTL_PORT_STATUS_ONLINE) {
3413                         STAILQ_FOREACH(lun, &softc->lun_list, links) {
3414                                 if (ctl_lun_map_to_port(port, lun->lun) ==
3415                                     UINT32_MAX)
3416                                         continue;
3417                                 mtx_lock(&lun->lun_lock);
3418                                 ctl_est_ua_port(lun, lm->port, -1,
3419                                     CTL_UA_LUN_CHANGE);
3420                                 mtx_unlock(&lun->lun_lock);
3421                         }
3422                 }
3423                 mtx_unlock(&softc->ctl_lock); // XXX: port_enable sleeps
3424                 if (lm->plun != UINT32_MAX) {
3425                         if (lm->lun == UINT32_MAX)
3426                                 retval = ctl_lun_map_unset(port, lm->plun);
3427                         else if (lm->lun < ctl_max_luns &&
3428                             softc->ctl_luns[lm->lun] != NULL)
3429                                 retval = ctl_lun_map_set(port, lm->plun, lm->lun);
3430                         else
3431                                 return (ENXIO);
3432                 } else {
3433                         if (lm->lun == UINT32_MAX)
3434                                 retval = ctl_lun_map_deinit(port);
3435                         else
3436                                 retval = ctl_lun_map_init(port);
3437                 }
3438                 if (port->status & CTL_PORT_STATUS_ONLINE)
3439                         ctl_isc_announce_port(port);
3440                 break;
3441         }
3442         case CTL_GET_LUN_STATS: {
3443                 struct ctl_get_io_stats *stats = (struct ctl_get_io_stats *)addr;
3444                 int i;
3445
3446                 /*
3447                  * XXX KDM no locking here.  If the LUN list changes,
3448                  * things can blow up.
3449                  */
3450                 i = 0;
3451                 stats->status = CTL_SS_OK;
3452                 stats->fill_len = 0;
3453                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
3454                         if (lun->lun < stats->first_item)
3455                                 continue;
3456                         if (stats->fill_len + sizeof(lun->stats) >
3457                             stats->alloc_len) {
3458                                 stats->status = CTL_SS_NEED_MORE_SPACE;
3459                                 break;
3460                         }
3461                         retval = copyout(&lun->stats, &stats->stats[i++],
3462                                          sizeof(lun->stats));
3463                         if (retval != 0)
3464                                 break;
3465                         stats->fill_len += sizeof(lun->stats);
3466                 }
3467                 stats->num_items = softc->num_luns;
3468                 stats->flags = CTL_STATS_FLAG_NONE;
3469 #ifdef CTL_TIME_IO
3470                 stats->flags |= CTL_STATS_FLAG_TIME_VALID;
3471 #endif
3472                 getnanouptime(&stats->timestamp);
3473                 break;
3474         }
3475         case CTL_GET_PORT_STATS: {
3476                 struct ctl_get_io_stats *stats = (struct ctl_get_io_stats *)addr;
3477                 int i;
3478
3479                 /*
3480                  * XXX KDM no locking here.  If the LUN list changes,
3481                  * things can blow up.
3482                  */
3483                 i = 0;
3484                 stats->status = CTL_SS_OK;
3485                 stats->fill_len = 0;
3486                 STAILQ_FOREACH(port, &softc->port_list, links) {
3487                         if (port->targ_port < stats->first_item)
3488                                 continue;
3489                         if (stats->fill_len + sizeof(port->stats) >
3490                             stats->alloc_len) {
3491                                 stats->status = CTL_SS_NEED_MORE_SPACE;
3492                                 break;
3493                         }
3494                         retval = copyout(&port->stats, &stats->stats[i++],
3495                                          sizeof(port->stats));
3496                         if (retval != 0)
3497                                 break;
3498                         stats->fill_len += sizeof(port->stats);
3499                 }
3500                 stats->num_items = softc->num_ports;
3501                 stats->flags = CTL_STATS_FLAG_NONE;
3502 #ifdef CTL_TIME_IO
3503                 stats->flags |= CTL_STATS_FLAG_TIME_VALID;
3504 #endif
3505                 getnanouptime(&stats->timestamp);
3506                 break;
3507         }
3508         default: {
3509                 /* XXX KDM should we fix this? */
3510 #if 0
3511                 struct ctl_backend_driver *backend;
3512                 unsigned int type;
3513                 int found;
3514
3515                 found = 0;
3516
3517                 /*
3518                  * We encode the backend type as the ioctl type for backend
3519                  * ioctls.  So parse it out here, and then search for a
3520                  * backend of this type.
3521                  */
3522                 type = _IOC_TYPE(cmd);
3523
3524                 STAILQ_FOREACH(backend, &softc->be_list, links) {
3525                         if (backend->type == type) {
3526                                 found = 1;
3527                                 break;
3528                         }
3529                 }
3530                 if (found == 0) {
3531                         printf("ctl: unknown ioctl command %#lx or backend "
3532                                "%d\n", cmd, type);
3533                         retval = EINVAL;
3534                         break;
3535                 }
3536                 retval = backend->ioctl(dev, cmd, addr, flag, td);
3537 #endif
3538                 retval = ENOTTY;
3539                 break;
3540         }
3541         }
3542         return (retval);
3543 }
3544
3545 uint32_t
3546 ctl_get_initindex(struct ctl_nexus *nexus)
3547 {
3548         return (nexus->initid + (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3549 }
3550
3551 int
3552 ctl_lun_map_init(struct ctl_port *port)
3553 {
3554         struct ctl_softc *softc = port->ctl_softc;
3555         struct ctl_lun *lun;
3556         int size = ctl_lun_map_size;
3557         uint32_t i;
3558
3559         if (port->lun_map == NULL || port->lun_map_size < size) {
3560                 port->lun_map_size = 0;
3561                 free(port->lun_map, M_CTL);
3562                 port->lun_map = malloc(size * sizeof(uint32_t),
3563                     M_CTL, M_NOWAIT);
3564         }
3565         if (port->lun_map == NULL)
3566                 return (ENOMEM);
3567         for (i = 0; i < size; i++)
3568                 port->lun_map[i] = UINT32_MAX;
3569         port->lun_map_size = size;
3570         if (port->status & CTL_PORT_STATUS_ONLINE) {
3571                 if (port->lun_disable != NULL) {
3572                         STAILQ_FOREACH(lun, &softc->lun_list, links)
3573                                 port->lun_disable(port->targ_lun_arg, lun->lun);
3574                 }
3575                 ctl_isc_announce_port(port);
3576         }
3577         return (0);
3578 }
3579
3580 int
3581 ctl_lun_map_deinit(struct ctl_port *port)
3582 {
3583         struct ctl_softc *softc = port->ctl_softc;
3584         struct ctl_lun *lun;
3585
3586         if (port->lun_map == NULL)
3587                 return (0);
3588         port->lun_map_size = 0;
3589         free(port->lun_map, M_CTL);
3590         port->lun_map = NULL;
3591         if (port->status & CTL_PORT_STATUS_ONLINE) {
3592                 if (port->lun_enable != NULL) {
3593                         STAILQ_FOREACH(lun, &softc->lun_list, links)
3594                                 port->lun_enable(port->targ_lun_arg, lun->lun);
3595                 }
3596                 ctl_isc_announce_port(port);
3597         }
3598         return (0);
3599 }
3600
3601 int
3602 ctl_lun_map_set(struct ctl_port *port, uint32_t plun, uint32_t glun)
3603 {
3604         int status;
3605         uint32_t old;
3606
3607         if (port->lun_map == NULL) {
3608                 status = ctl_lun_map_init(port);
3609                 if (status != 0)
3610                         return (status);
3611         }
3612         if (plun >= port->lun_map_size)
3613                 return (EINVAL);
3614         old = port->lun_map[plun];
3615         port->lun_map[plun] = glun;
3616         if ((port->status & CTL_PORT_STATUS_ONLINE) && old == UINT32_MAX) {
3617                 if (port->lun_enable != NULL)
3618                         port->lun_enable(port->targ_lun_arg, plun);
3619                 ctl_isc_announce_port(port);
3620         }
3621         return (0);
3622 }
3623
3624 int
3625 ctl_lun_map_unset(struct ctl_port *port, uint32_t plun)
3626 {
3627         uint32_t old;
3628
3629         if (port->lun_map == NULL || plun >= port->lun_map_size)
3630                 return (0);
3631         old = port->lun_map[plun];
3632         port->lun_map[plun] = UINT32_MAX;
3633         if ((port->status & CTL_PORT_STATUS_ONLINE) && old != UINT32_MAX) {
3634                 if (port->lun_disable != NULL)
3635                         port->lun_disable(port->targ_lun_arg, plun);
3636                 ctl_isc_announce_port(port);
3637         }
3638         return (0);
3639 }
3640
3641 uint32_t
3642 ctl_lun_map_from_port(struct ctl_port *port, uint32_t lun_id)
3643 {
3644
3645         if (port == NULL)
3646                 return (UINT32_MAX);
3647         if (port->lun_map == NULL)
3648                 return (lun_id);
3649         if (lun_id > port->lun_map_size)
3650                 return (UINT32_MAX);
3651         return (port->lun_map[lun_id]);
3652 }
3653
3654 uint32_t
3655 ctl_lun_map_to_port(struct ctl_port *port, uint32_t lun_id)
3656 {
3657         uint32_t i;
3658
3659         if (port == NULL)
3660                 return (UINT32_MAX);
3661         if (port->lun_map == NULL)
3662                 return (lun_id);
3663         for (i = 0; i < port->lun_map_size; i++) {
3664                 if (port->lun_map[i] == lun_id)
3665                         return (i);
3666         }
3667         return (UINT32_MAX);
3668 }
3669
3670 uint32_t
3671 ctl_decode_lun(uint64_t encoded)
3672 {
3673         uint8_t lun[8];
3674         uint32_t result = 0xffffffff;
3675
3676         be64enc(lun, encoded);
3677         switch (lun[0] & RPL_LUNDATA_ATYP_MASK) {
3678         case RPL_LUNDATA_ATYP_PERIPH:
3679                 if ((lun[0] & 0x3f) == 0 && lun[2] == 0 && lun[3] == 0 &&
3680                     lun[4] == 0 && lun[5] == 0 && lun[6] == 0 && lun[7] == 0)
3681                         result = lun[1];
3682                 break;
3683         case RPL_LUNDATA_ATYP_FLAT:
3684                 if (lun[2] == 0 && lun[3] == 0 && lun[4] == 0 && lun[5] == 0 &&
3685                     lun[6] == 0 && lun[7] == 0)
3686                         result = ((lun[0] & 0x3f) << 8) + lun[1];
3687                 break;
3688         case RPL_LUNDATA_ATYP_EXTLUN:
3689                 switch (lun[0] & RPL_LUNDATA_EXT_EAM_MASK) {
3690                 case 0x02:
3691                         switch (lun[0] & RPL_LUNDATA_EXT_LEN_MASK) {
3692                         case 0x00:
3693                                 result = lun[1];
3694                                 break;
3695                         case 0x10:
3696                                 result = (lun[1] << 16) + (lun[2] << 8) +
3697                                     lun[3];
3698                                 break;
3699                         case 0x20:
3700                                 if (lun[1] == 0 && lun[6] == 0 && lun[7] == 0)
3701                                         result = (lun[2] << 24) +
3702                                             (lun[3] << 16) + (lun[4] << 8) +
3703                                             lun[5];
3704                                 break;
3705                         }
3706                         break;
3707                 case RPL_LUNDATA_EXT_EAM_NOT_SPEC:
3708                         result = 0xffffffff;
3709                         break;
3710                 }
3711                 break;
3712         }
3713         return (result);
3714 }
3715
3716 uint64_t
3717 ctl_encode_lun(uint32_t decoded)
3718 {
3719         uint64_t l = decoded;
3720
3721         if (l <= 0xff)
3722                 return (((uint64_t)RPL_LUNDATA_ATYP_PERIPH << 56) | (l << 48));
3723         if (l <= 0x3fff)
3724                 return (((uint64_t)RPL_LUNDATA_ATYP_FLAT << 56) | (l << 48));
3725         if (l <= 0xffffff)
3726                 return (((uint64_t)(RPL_LUNDATA_ATYP_EXTLUN | 0x12) << 56) |
3727                     (l << 32));
3728         return ((((uint64_t)RPL_LUNDATA_ATYP_EXTLUN | 0x22) << 56) | (l << 16));
3729 }
3730
3731 int
3732 ctl_ffz(uint32_t *mask, uint32_t first, uint32_t last)
3733 {
3734         int i;
3735
3736         for (i = first; i < last; i++) {
3737                 if ((mask[i / 32] & (1 << (i % 32))) == 0)
3738                         return (i);
3739         }
3740         return (-1);
3741 }
3742
3743 int
3744 ctl_set_mask(uint32_t *mask, uint32_t bit)
3745 {
3746         uint32_t chunk, piece;
3747
3748         chunk = bit >> 5;
3749         piece = bit % (sizeof(uint32_t) * 8);
3750
3751         if ((mask[chunk] & (1 << piece)) != 0)
3752                 return (-1);
3753         else
3754                 mask[chunk] |= (1 << piece);
3755
3756         return (0);
3757 }
3758
3759 int
3760 ctl_clear_mask(uint32_t *mask, uint32_t bit)
3761 {
3762         uint32_t chunk, piece;
3763
3764         chunk = bit >> 5;
3765         piece = bit % (sizeof(uint32_t) * 8);
3766
3767         if ((mask[chunk] & (1 << piece)) == 0)
3768                 return (-1);
3769         else
3770                 mask[chunk] &= ~(1 << piece);
3771
3772         return (0);
3773 }
3774
3775 int
3776 ctl_is_set(uint32_t *mask, uint32_t bit)
3777 {
3778         uint32_t chunk, piece;
3779
3780         chunk = bit >> 5;
3781         piece = bit % (sizeof(uint32_t) * 8);
3782
3783         if ((mask[chunk] & (1 << piece)) == 0)
3784                 return (0);
3785         else
3786                 return (1);
3787 }
3788
3789 static uint64_t
3790 ctl_get_prkey(struct ctl_lun *lun, uint32_t residx)
3791 {
3792         uint64_t *t;
3793
3794         t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3795         if (t == NULL)
3796                 return (0);
3797         return (t[residx % CTL_MAX_INIT_PER_PORT]);
3798 }
3799
3800 static void
3801 ctl_clr_prkey(struct ctl_lun *lun, uint32_t residx)
3802 {
3803         uint64_t *t;
3804
3805         t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3806         if (t == NULL)
3807                 return;
3808         t[residx % CTL_MAX_INIT_PER_PORT] = 0;
3809 }
3810
3811 static void
3812 ctl_alloc_prkey(struct ctl_lun *lun, uint32_t residx)
3813 {
3814         uint64_t *p;
3815         u_int i;
3816
3817         i = residx/CTL_MAX_INIT_PER_PORT;
3818         if (lun->pr_keys[i] != NULL)
3819                 return;
3820         mtx_unlock(&lun->lun_lock);
3821         p = malloc(sizeof(uint64_t) * CTL_MAX_INIT_PER_PORT, M_CTL,
3822             M_WAITOK | M_ZERO);
3823         mtx_lock(&lun->lun_lock);
3824         if (lun->pr_keys[i] == NULL)
3825                 lun->pr_keys[i] = p;
3826         else
3827                 free(p, M_CTL);
3828 }
3829
3830 static void
3831 ctl_set_prkey(struct ctl_lun *lun, uint32_t residx, uint64_t key)
3832 {
3833         uint64_t *t;
3834
3835         t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3836         KASSERT(t != NULL, ("prkey %d is not allocated", residx));
3837         t[residx % CTL_MAX_INIT_PER_PORT] = key;
3838 }
3839
3840 /*
3841  * ctl_softc, pool_name, total_ctl_io are passed in.
3842  * npool is passed out.
3843  */
3844 int
3845 ctl_pool_create(struct ctl_softc *ctl_softc, const char *pool_name,
3846                 uint32_t total_ctl_io, void **npool)
3847 {
3848         struct ctl_io_pool *pool;
3849
3850         pool = (struct ctl_io_pool *)malloc(sizeof(*pool), M_CTL,
3851                                             M_NOWAIT | M_ZERO);
3852         if (pool == NULL)
3853                 return (ENOMEM);
3854
3855         snprintf(pool->name, sizeof(pool->name), "CTL IO %s", pool_name);
3856         pool->ctl_softc = ctl_softc;
3857 #ifdef IO_POOLS
3858         pool->zone = uma_zsecond_create(pool->name, NULL,
3859             NULL, NULL, NULL, ctl_softc->io_zone);
3860         /* uma_prealloc(pool->zone, total_ctl_io); */
3861 #else
3862         pool->zone = ctl_softc->io_zone;
3863 #endif
3864
3865         *npool = pool;
3866         return (0);
3867 }
3868
3869 void
3870 ctl_pool_free(struct ctl_io_pool *pool)
3871 {
3872
3873         if (pool == NULL)
3874                 return;
3875
3876 #ifdef IO_POOLS
3877         uma_zdestroy(pool->zone);
3878 #endif
3879         free(pool, M_CTL);
3880 }
3881
3882 union ctl_io *
3883 ctl_alloc_io(void *pool_ref)
3884 {
3885         struct ctl_io_pool *pool = (struct ctl_io_pool *)pool_ref;
3886         union ctl_io *io;
3887
3888         io = uma_zalloc(pool->zone, M_WAITOK);
3889         if (io != NULL) {
3890                 io->io_hdr.pool = pool_ref;
3891                 CTL_SOFTC(io) = pool->ctl_softc;
3892         }
3893         return (io);
3894 }
3895
3896 union ctl_io *
3897 ctl_alloc_io_nowait(void *pool_ref)
3898 {
3899         struct ctl_io_pool *pool = (struct ctl_io_pool *)pool_ref;
3900         union ctl_io *io;
3901
3902         io = uma_zalloc(pool->zone, M_NOWAIT);
3903         if (io != NULL) {
3904                 io->io_hdr.pool = pool_ref;
3905                 CTL_SOFTC(io) = pool->ctl_softc;
3906         }
3907         return (io);
3908 }
3909
3910 void
3911 ctl_free_io(union ctl_io *io)
3912 {
3913         struct ctl_io_pool *pool;
3914
3915         if (io == NULL)
3916                 return;
3917
3918         pool = (struct ctl_io_pool *)io->io_hdr.pool;
3919         uma_zfree(pool->zone, io);
3920 }
3921
3922 void
3923 ctl_zero_io(union ctl_io *io)
3924 {
3925         struct ctl_io_pool *pool;
3926
3927         if (io == NULL)
3928                 return;
3929
3930         /*
3931          * May need to preserve linked list pointers at some point too.
3932          */
3933         pool = io->io_hdr.pool;
3934         memset(io, 0, sizeof(*io));
3935         io->io_hdr.pool = pool;
3936         CTL_SOFTC(io) = pool->ctl_softc;
3937 }
3938
3939 int
3940 ctl_expand_number(const char *buf, uint64_t *num)
3941 {
3942         char *endptr;
3943         uint64_t number;
3944         unsigned shift;
3945
3946         number = strtoq(buf, &endptr, 0);
3947
3948         switch (tolower((unsigned char)*endptr)) {
3949         case 'e':
3950                 shift = 60;
3951                 break;
3952         case 'p':
3953                 shift = 50;
3954                 break;
3955         case 't':
3956                 shift = 40;
3957                 break;
3958         case 'g':
3959                 shift = 30;
3960                 break;
3961         case 'm':
3962                 shift = 20;
3963                 break;
3964         case 'k':
3965                 shift = 10;
3966                 break;
3967         case 'b':
3968         case '\0': /* No unit. */
3969                 *num = number;
3970                 return (0);
3971         default:
3972                 /* Unrecognized unit. */
3973                 return (-1);
3974         }
3975
3976         if ((number << shift) >> shift != number) {
3977                 /* Overflow */
3978                 return (-1);
3979         }
3980         *num = number << shift;
3981         return (0);
3982 }
3983
3984
3985 /*
3986  * This routine could be used in the future to load default and/or saved
3987  * mode page parameters for a particuar lun.
3988  */
3989 static int
3990 ctl_init_page_index(struct ctl_lun *lun)
3991 {
3992         int i, page_code;
3993         struct ctl_page_index *page_index;
3994         const char *value;
3995         uint64_t ival;
3996
3997         memcpy(&lun->mode_pages.index, page_index_template,
3998                sizeof(page_index_template));
3999
4000         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
4001
4002                 page_index = &lun->mode_pages.index[i];
4003                 if (lun->be_lun->lun_type == T_DIRECT &&
4004                     (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
4005                         continue;
4006                 if (lun->be_lun->lun_type == T_PROCESSOR &&
4007                     (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
4008                         continue;
4009                 if (lun->be_lun->lun_type == T_CDROM &&
4010                     (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
4011                         continue;
4012
4013                 page_code = page_index->page_code & SMPH_PC_MASK;
4014                 switch (page_code) {
4015                 case SMS_RW_ERROR_RECOVERY_PAGE: {
4016                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4017                             ("subpage %#x for page %#x is incorrect!",
4018                             page_index->subpage, page_code));
4019                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_CURRENT],
4020                                &rw_er_page_default,
4021                                sizeof(rw_er_page_default));
4022                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_CHANGEABLE],
4023                                &rw_er_page_changeable,
4024                                sizeof(rw_er_page_changeable));
4025                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_DEFAULT],
4026                                &rw_er_page_default,
4027                                sizeof(rw_er_page_default));
4028                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_SAVED],
4029                                &rw_er_page_default,
4030                                sizeof(rw_er_page_default));
4031                         page_index->page_data =
4032                                 (uint8_t *)lun->mode_pages.rw_er_page;
4033                         break;
4034                 }
4035                 case SMS_FORMAT_DEVICE_PAGE: {
4036                         struct scsi_format_page *format_page;
4037
4038                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4039                             ("subpage %#x for page %#x is incorrect!",
4040                             page_index->subpage, page_code));
4041
4042                         /*
4043                          * Sectors per track are set above.  Bytes per
4044                          * sector need to be set here on a per-LUN basis.
4045                          */
4046                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_CURRENT],
4047                                &format_page_default,
4048                                sizeof(format_page_default));
4049                         memcpy(&lun->mode_pages.format_page[
4050                                CTL_PAGE_CHANGEABLE], &format_page_changeable,
4051                                sizeof(format_page_changeable));
4052                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_DEFAULT],
4053                                &format_page_default,
4054                                sizeof(format_page_default));
4055                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_SAVED],
4056                                &format_page_default,
4057                                sizeof(format_page_default));
4058
4059                         format_page = &lun->mode_pages.format_page[
4060                                 CTL_PAGE_CURRENT];
4061                         scsi_ulto2b(lun->be_lun->blocksize,
4062                                     format_page->bytes_per_sector);
4063
4064                         format_page = &lun->mode_pages.format_page[
4065                                 CTL_PAGE_DEFAULT];
4066                         scsi_ulto2b(lun->be_lun->blocksize,
4067                                     format_page->bytes_per_sector);
4068
4069                         format_page = &lun->mode_pages.format_page[
4070                                 CTL_PAGE_SAVED];
4071                         scsi_ulto2b(lun->be_lun->blocksize,
4072                                     format_page->bytes_per_sector);
4073
4074                         page_index->page_data =
4075                                 (uint8_t *)lun->mode_pages.format_page;
4076                         break;
4077                 }
4078                 case SMS_RIGID_DISK_PAGE: {
4079                         struct scsi_rigid_disk_page *rigid_disk_page;
4080                         uint32_t sectors_per_cylinder;
4081                         uint64_t cylinders;
4082 #ifndef __XSCALE__
4083                         int shift;
4084 #endif /* !__XSCALE__ */
4085
4086                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4087                             ("subpage %#x for page %#x is incorrect!",
4088                             page_index->subpage, page_code));
4089
4090                         /*
4091                          * Rotation rate and sectors per track are set
4092                          * above.  We calculate the cylinders here based on
4093                          * capacity.  Due to the number of heads and
4094                          * sectors per track we're using, smaller arrays
4095                          * may turn out to have 0 cylinders.  Linux and
4096                          * FreeBSD don't pay attention to these mode pages
4097                          * to figure out capacity, but Solaris does.  It
4098                          * seems to deal with 0 cylinders just fine, and
4099                          * works out a fake geometry based on the capacity.
4100                          */
4101                         memcpy(&lun->mode_pages.rigid_disk_page[
4102                                CTL_PAGE_DEFAULT], &rigid_disk_page_default,
4103                                sizeof(rigid_disk_page_default));
4104                         memcpy(&lun->mode_pages.rigid_disk_page[
4105                                CTL_PAGE_CHANGEABLE],&rigid_disk_page_changeable,
4106                                sizeof(rigid_disk_page_changeable));
4107
4108                         sectors_per_cylinder = CTL_DEFAULT_SECTORS_PER_TRACK *
4109                                 CTL_DEFAULT_HEADS;
4110
4111                         /*
4112                          * The divide method here will be more accurate,
4113                          * probably, but results in floating point being
4114                          * used in the kernel on i386 (__udivdi3()).  On the
4115                          * XScale, though, __udivdi3() is implemented in
4116                          * software.
4117                          *
4118                          * The shift method for cylinder calculation is
4119                          * accurate if sectors_per_cylinder is a power of
4120                          * 2.  Otherwise it might be slightly off -- you
4121                          * might have a bit of a truncation problem.
4122                          */
4123 #ifdef  __XSCALE__
4124                         cylinders = (lun->be_lun->maxlba + 1) /
4125                                 sectors_per_cylinder;
4126 #else
4127                         for (shift = 31; shift > 0; shift--) {
4128                                 if (sectors_per_cylinder & (1 << shift))
4129                                         break;
4130                         }
4131                         cylinders = (lun->be_lun->maxlba + 1) >> shift;
4132 #endif
4133
4134                         /*
4135                          * We've basically got 3 bytes, or 24 bits for the
4136                          * cylinder size in the mode page.  If we're over,
4137                          * just round down to 2^24.
4138                          */
4139                         if (cylinders > 0xffffff)
4140                                 cylinders = 0xffffff;
4141
4142                         rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4143                                 CTL_PAGE_DEFAULT];
4144                         scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4145
4146                         if ((value = dnvlist_get_string(lun->be_lun->options,
4147                             "rpm", NULL)) != NULL) {
4148                                 scsi_ulto2b(strtol(value, NULL, 0),
4149                                      rigid_disk_page->rotation_rate);
4150                         }
4151
4152                         memcpy(&lun->mode_pages.rigid_disk_page[CTL_PAGE_CURRENT],
4153                                &lun->mode_pages.rigid_disk_page[CTL_PAGE_DEFAULT],
4154                                sizeof(rigid_disk_page_default));
4155                         memcpy(&lun->mode_pages.rigid_disk_page[CTL_PAGE_SAVED],
4156                                &lun->mode_pages.rigid_disk_page[CTL_PAGE_DEFAULT],
4157                                sizeof(rigid_disk_page_default));
4158
4159                         page_index->page_data =
4160                                 (uint8_t *)lun->mode_pages.rigid_disk_page;
4161                         break;
4162                 }
4163                 case SMS_VERIFY_ERROR_RECOVERY_PAGE: {
4164                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4165                             ("subpage %#x for page %#x is incorrect!",
4166                             page_index->subpage, page_code));
4167                         memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_CURRENT],
4168                                &verify_er_page_default,
4169                                sizeof(verify_er_page_default));
4170                         memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_CHANGEABLE],
4171                                &verify_er_page_changeable,
4172                                sizeof(verify_er_page_changeable));
4173                         memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_DEFAULT],
4174                                &verify_er_page_default,
4175                                sizeof(verify_er_page_default));
4176                         memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_SAVED],
4177                                &verify_er_page_default,
4178                                sizeof(verify_er_page_default));
4179                         page_index->page_data =
4180                                 (uint8_t *)lun->mode_pages.verify_er_page;
4181                         break;
4182                 }
4183                 case SMS_CACHING_PAGE: {
4184                         struct scsi_caching_page *caching_page;
4185
4186                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4187                             ("subpage %#x for page %#x is incorrect!",
4188                             page_index->subpage, page_code));
4189                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_DEFAULT],
4190                                &caching_page_default,
4191                                sizeof(caching_page_default));
4192                         memcpy(&lun->mode_pages.caching_page[
4193                                CTL_PAGE_CHANGEABLE], &caching_page_changeable,
4194                                sizeof(caching_page_changeable));
4195                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4196                                &caching_page_default,
4197                                sizeof(caching_page_default));
4198                         caching_page = &lun->mode_pages.caching_page[
4199                             CTL_PAGE_SAVED];
4200                         value = dnvlist_get_string(lun->be_lun->options,
4201                             "writecache", NULL);
4202                         if (value != NULL && strcmp(value, "off") == 0)
4203                                 caching_page->flags1 &= ~SCP_WCE;
4204                         value = dnvlist_get_string(lun->be_lun->options,
4205                             "readcache", NULL);
4206                         if (value != NULL && strcmp(value, "off") == 0)
4207                                 caching_page->flags1 |= SCP_RCD;
4208                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_CURRENT],
4209                                &lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4210                                sizeof(caching_page_default));
4211                         page_index->page_data =
4212                                 (uint8_t *)lun->mode_pages.caching_page;
4213                         break;
4214                 }
4215                 case SMS_CONTROL_MODE_PAGE: {
4216                         switch (page_index->subpage) {
4217                         case SMS_SUBPAGE_PAGE_0: {
4218                                 struct scsi_control_page *control_page;
4219
4220                                 memcpy(&lun->mode_pages.control_page[
4221                                     CTL_PAGE_DEFAULT],
4222                                        &control_page_default,
4223                                        sizeof(control_page_default));
4224                                 memcpy(&lun->mode_pages.control_page[
4225                                     CTL_PAGE_CHANGEABLE],
4226                                        &control_page_changeable,
4227                                        sizeof(control_page_changeable));
4228                                 memcpy(&lun->mode_pages.control_page[
4229                                     CTL_PAGE_SAVED],
4230                                        &control_page_default,
4231                                        sizeof(control_page_default));
4232                                 control_page = &lun->mode_pages.control_page[
4233                                     CTL_PAGE_SAVED];
4234                                 value = dnvlist_get_string(lun->be_lun->options,
4235                                     "reordering", NULL);
4236                                 if (value != NULL &&
4237                                     strcmp(value, "unrestricted") == 0) {
4238                                         control_page->queue_flags &=
4239                                             ~SCP_QUEUE_ALG_MASK;
4240                                         control_page->queue_flags |=
4241                                             SCP_QUEUE_ALG_UNRESTRICTED;
4242                                 }
4243                                 memcpy(&lun->mode_pages.control_page[
4244                                     CTL_PAGE_CURRENT],
4245                                        &lun->mode_pages.control_page[
4246                                     CTL_PAGE_SAVED],
4247                                        sizeof(control_page_default));
4248                                 page_index->page_data =
4249                                     (uint8_t *)lun->mode_pages.control_page;
4250                                 break;
4251                         }
4252                         case 0x01:
4253                                 memcpy(&lun->mode_pages.control_ext_page[
4254                                     CTL_PAGE_DEFAULT],
4255                                        &control_ext_page_default,
4256                                        sizeof(control_ext_page_default));
4257                                 memcpy(&lun->mode_pages.control_ext_page[
4258                                     CTL_PAGE_CHANGEABLE],
4259                                        &control_ext_page_changeable,
4260                                        sizeof(control_ext_page_changeable));
4261                                 memcpy(&lun->mode_pages.control_ext_page[
4262                                     CTL_PAGE_SAVED],
4263                                        &control_ext_page_default,
4264                                        sizeof(control_ext_page_default));
4265                                 memcpy(&lun->mode_pages.control_ext_page[
4266                                     CTL_PAGE_CURRENT],
4267                                        &lun->mode_pages.control_ext_page[
4268                                     CTL_PAGE_SAVED],
4269                                        sizeof(control_ext_page_default));
4270                                 page_index->page_data =
4271                                     (uint8_t *)lun->mode_pages.control_ext_page;
4272                                 break;
4273                         default:
4274                                 panic("subpage %#x for page %#x is incorrect!",
4275                                       page_index->subpage, page_code);
4276                         }
4277                         break;
4278                 }
4279                 case SMS_INFO_EXCEPTIONS_PAGE: {
4280                         switch (page_index->subpage) {
4281                         case SMS_SUBPAGE_PAGE_0:
4282                                 memcpy(&lun->mode_pages.ie_page[CTL_PAGE_CURRENT],
4283                                        &ie_page_default,
4284                                        sizeof(ie_page_default));
4285                                 memcpy(&lun->mode_pages.ie_page[
4286                                        CTL_PAGE_CHANGEABLE], &ie_page_changeable,
4287                                        sizeof(ie_page_changeable));
4288                                 memcpy(&lun->mode_pages.ie_page[CTL_PAGE_DEFAULT],
4289                                        &ie_page_default,
4290                                        sizeof(ie_page_default));
4291                                 memcpy(&lun->mode_pages.ie_page[CTL_PAGE_SAVED],
4292                                        &ie_page_default,
4293                                        sizeof(ie_page_default));
4294                                 page_index->page_data =
4295                                         (uint8_t *)lun->mode_pages.ie_page;
4296                                 break;
4297                         case 0x02: {
4298                                 struct ctl_logical_block_provisioning_page *page;
4299
4300                                 memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_DEFAULT],
4301                                        &lbp_page_default,
4302                                        sizeof(lbp_page_default));
4303                                 memcpy(&lun->mode_pages.lbp_page[
4304                                        CTL_PAGE_CHANGEABLE], &lbp_page_changeable,
4305                                        sizeof(lbp_page_changeable));
4306                                 memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_SAVED],
4307                                        &lbp_page_default,
4308                                        sizeof(lbp_page_default));
4309                                 page = &lun->mode_pages.lbp_page[CTL_PAGE_SAVED];
4310                                 value = dnvlist_get_string(lun->be_lun->options,
4311                                     "avail-threshold", NULL);
4312                                 if (value != NULL &&
4313                                     ctl_expand_number(value, &ival) == 0) {
4314                                         page->descr[0].flags |= SLBPPD_ENABLED |
4315                                             SLBPPD_ARMING_DEC;
4316                                         if (lun->be_lun->blocksize)
4317                                                 ival /= lun->be_lun->blocksize;
4318                                         else
4319                                                 ival /= 512;
4320                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4321                                             page->descr[0].count);
4322                                 }
4323                                 value = dnvlist_get_string(lun->be_lun->options,
4324                                     "used-threshold", NULL);
4325                                 if (value != NULL &&
4326                                     ctl_expand_number(value, &ival) == 0) {
4327                                         page->descr[1].flags |= SLBPPD_ENABLED |
4328                                             SLBPPD_ARMING_INC;
4329                                         if (lun->be_lun->blocksize)
4330                                                 ival /= lun->be_lun->blocksize;
4331                                         else
4332                                                 ival /= 512;
4333                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4334                                             page->descr[1].count);
4335                                 }
4336                                 value = dnvlist_get_string(lun->be_lun->options,
4337                                     "pool-avail-threshold", NULL);
4338                                 if (value != NULL &&
4339                                     ctl_expand_number(value, &ival) == 0) {
4340                                         page->descr[2].flags |= SLBPPD_ENABLED |
4341                                             SLBPPD_ARMING_DEC;
4342                                         if (lun->be_lun->blocksize)
4343                                                 ival /= lun->be_lun->blocksize;
4344                                         else
4345                                                 ival /= 512;
4346                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4347                                             page->descr[2].count);
4348                                 }
4349                                 value = dnvlist_get_string(lun->be_lun->options,
4350                                     "pool-used-threshold", NULL);
4351                                 if (value != NULL &&
4352                                     ctl_expand_number(value, &ival) == 0) {
4353                                         page->descr[3].flags |= SLBPPD_ENABLED |
4354                                             SLBPPD_ARMING_INC;
4355                                         if (lun->be_lun->blocksize)
4356                                                 ival /= lun->be_lun->blocksize;
4357                                         else
4358                                                 ival /= 512;
4359                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4360                                             page->descr[3].count);
4361                                 }
4362                                 memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_CURRENT],
4363                                        &lun->mode_pages.lbp_page[CTL_PAGE_SAVED],
4364                                        sizeof(lbp_page_default));
4365                                 page_index->page_data =
4366                                         (uint8_t *)lun->mode_pages.lbp_page;
4367                                 break;
4368                         }
4369                         default:
4370                                 panic("subpage %#x for page %#x is incorrect!",
4371                                       page_index->subpage, page_code);
4372                         }
4373                         break;
4374                 }
4375                 case SMS_CDDVD_CAPS_PAGE:{
4376                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4377                             ("subpage %#x for page %#x is incorrect!",
4378                             page_index->subpage, page_code));
4379                         memcpy(&lun->mode_pages.cddvd_page[CTL_PAGE_DEFAULT],
4380                                &cddvd_page_default,
4381                                sizeof(cddvd_page_default));
4382                         memcpy(&lun->mode_pages.cddvd_page[
4383                                CTL_PAGE_CHANGEABLE], &cddvd_page_changeable,
4384                                sizeof(cddvd_page_changeable));
4385                         memcpy(&lun->mode_pages.cddvd_page[CTL_PAGE_SAVED],
4386                                &cddvd_page_default,
4387                                sizeof(cddvd_page_default));
4388                         memcpy(&lun->mode_pages.cddvd_page[CTL_PAGE_CURRENT],
4389                                &lun->mode_pages.cddvd_page[CTL_PAGE_SAVED],
4390                                sizeof(cddvd_page_default));
4391                         page_index->page_data =
4392                                 (uint8_t *)lun->mode_pages.cddvd_page;
4393                         break;
4394                 }
4395                 default:
4396                         panic("invalid page code value %#x", page_code);
4397                 }
4398         }
4399
4400         return (CTL_RETVAL_COMPLETE);
4401 }
4402
4403 static int
4404 ctl_init_log_page_index(struct ctl_lun *lun)
4405 {
4406         struct ctl_page_index *page_index;
4407         int i, j, k, prev;
4408
4409         memcpy(&lun->log_pages.index, log_page_index_template,
4410                sizeof(log_page_index_template));
4411
4412         prev = -1;
4413         for (i = 0, j = 0, k = 0; i < CTL_NUM_LOG_PAGES; i++) {
4414
4415                 page_index = &lun->log_pages.index[i];
4416                 if (lun->be_lun->lun_type == T_DIRECT &&
4417                     (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
4418                         continue;
4419                 if (lun->be_lun->lun_type == T_PROCESSOR &&
4420                     (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
4421                         continue;
4422                 if (lun->be_lun->lun_type == T_CDROM &&
4423                     (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
4424                         continue;
4425
4426                 if (page_index->page_code == SLS_LOGICAL_BLOCK_PROVISIONING &&
4427                     lun->backend->lun_attr == NULL)
4428                         continue;
4429
4430                 if (page_index->page_code != prev) {
4431                         lun->log_pages.pages_page[j] = page_index->page_code;
4432                         prev = page_index->page_code;
4433                         j++;
4434                 }
4435                 lun->log_pages.subpages_page[k*2] = page_index->page_code;
4436                 lun->log_pages.subpages_page[k*2+1] = page_index->subpage;
4437                 k++;
4438         }
4439         lun->log_pages.index[0].page_data = &lun->log_pages.pages_page[0];
4440         lun->log_pages.index[0].page_len = j;
4441         lun->log_pages.index[1].page_data = &lun->log_pages.subpages_page[0];
4442         lun->log_pages.index[1].page_len = k * 2;
4443         lun->log_pages.index[2].page_data = &lun->log_pages.lbp_page[0];
4444         lun->log_pages.index[2].page_len = 12*CTL_NUM_LBP_PARAMS;
4445         lun->log_pages.index[3].page_data = (uint8_t *)&lun->log_pages.stat_page;
4446         lun->log_pages.index[3].page_len = sizeof(lun->log_pages.stat_page);
4447         lun->log_pages.index[4].page_data = (uint8_t *)&lun->log_pages.ie_page;
4448         lun->log_pages.index[4].page_len = sizeof(lun->log_pages.ie_page);
4449
4450         return (CTL_RETVAL_COMPLETE);
4451 }
4452
4453 static int
4454 hex2bin(const char *str, uint8_t *buf, int buf_size)
4455 {
4456         int i;
4457         u_char c;
4458
4459         memset(buf, 0, buf_size);
4460         while (isspace(str[0]))
4461                 str++;
4462         if (str[0] == '0' && (str[1] == 'x' || str[1] == 'X'))
4463                 str += 2;
4464         buf_size *= 2;
4465         for (i = 0; str[i] != 0 && i < buf_size; i++) {
4466                 while (str[i] == '-')   /* Skip dashes in UUIDs. */
4467                         str++;
4468                 c = str[i];
4469                 if (isdigit(c))
4470                         c -= '0';
4471                 else if (isalpha(c))
4472                         c -= isupper(c) ? 'A' - 10 : 'a' - 10;
4473                 else
4474                         break;
4475                 if (c >= 16)
4476                         break;
4477                 if ((i & 1) == 0)
4478                         buf[i / 2] |= (c << 4);
4479                 else
4480                         buf[i / 2] |= c;
4481         }
4482         return ((i + 1) / 2);
4483 }
4484
4485 /*
4486  * LUN allocation.
4487  *
4488  * Requirements:
4489  * - caller allocates and zeros LUN storage, or passes in a NULL LUN if he
4490  *   wants us to allocate the LUN and he can block.
4491  * - ctl_softc is always set
4492  * - be_lun is set if the LUN has a backend (needed for disk LUNs)
4493  *
4494  * Returns 0 for success, non-zero (errno) for failure.
4495  */
4496 static int
4497 ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *ctl_lun,
4498               struct ctl_be_lun *const be_lun)
4499 {
4500         struct ctl_lun *nlun, *lun;
4501         struct scsi_vpd_id_descriptor *desc;
4502         struct scsi_vpd_id_t10 *t10id;
4503         const char *eui, *naa, *scsiname, *uuid, *vendor, *value;
4504         int lun_number, lun_malloced;
4505         int devidlen, idlen1, idlen2 = 0, len;
4506
4507         if (be_lun == NULL)
4508                 return (EINVAL);
4509
4510         /*
4511          * We currently only support Direct Access or Processor LUN types.
4512          */
4513         switch (be_lun->lun_type) {
4514         case T_DIRECT:
4515         case T_PROCESSOR:
4516         case T_CDROM:
4517                 break;
4518         case T_SEQUENTIAL:
4519         case T_CHANGER:
4520         default:
4521                 be_lun->lun_config_status(be_lun->be_lun,
4522                                           CTL_LUN_CONFIG_FAILURE);
4523                 break;
4524         }
4525         if (ctl_lun == NULL) {
4526                 lun = malloc(sizeof(*lun), M_CTL, M_WAITOK);
4527                 lun_malloced = 1;
4528         } else {
4529                 lun_malloced = 0;
4530                 lun = ctl_lun;
4531         }
4532
4533         memset(lun, 0, sizeof(*lun));
4534         if (lun_malloced)
4535                 lun->flags = CTL_LUN_MALLOCED;
4536
4537         lun->pending_sense = malloc(sizeof(struct scsi_sense_data *) *
4538             ctl_max_ports, M_DEVBUF, M_WAITOK | M_ZERO);
4539         lun->pending_ua = malloc(sizeof(ctl_ua_type *) * ctl_max_ports,
4540             M_DEVBUF, M_WAITOK | M_ZERO);
4541         lun->pr_keys = malloc(sizeof(uint64_t *) * ctl_max_ports,
4542             M_DEVBUF, M_WAITOK | M_ZERO);
4543
4544         /* Generate LUN ID. */
4545         devidlen = max(CTL_DEVID_MIN_LEN,
4546             strnlen(be_lun->device_id, CTL_DEVID_LEN));
4547         idlen1 = sizeof(*t10id) + devidlen;
4548         len = sizeof(struct scsi_vpd_id_descriptor) + idlen1;
4549         scsiname = dnvlist_get_string(be_lun->options, "scsiname", NULL);
4550         if (scsiname != NULL) {
4551                 idlen2 = roundup2(strlen(scsiname) + 1, 4);
4552                 len += sizeof(struct scsi_vpd_id_descriptor) + idlen2;
4553         }
4554         eui = dnvlist_get_string(be_lun->options, "eui", NULL);
4555         if (eui != NULL) {
4556                 len += sizeof(struct scsi_vpd_id_descriptor) + 16;
4557         }
4558         naa = dnvlist_get_string(be_lun->options, "naa", NULL);
4559         if (naa != NULL) {
4560                 len += sizeof(struct scsi_vpd_id_descriptor) + 16;
4561         }
4562         uuid = dnvlist_get_string(be_lun->options, "uuid", NULL);
4563         if (uuid != NULL) {
4564                 len += sizeof(struct scsi_vpd_id_descriptor) + 18;
4565         }
4566         lun->lun_devid = malloc(sizeof(struct ctl_devid) + len,
4567             M_CTL, M_WAITOK | M_ZERO);
4568         desc = (struct scsi_vpd_id_descriptor *)lun->lun_devid->data;
4569         desc->proto_codeset = SVPD_ID_CODESET_ASCII;
4570         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_T10;
4571         desc->length = idlen1;
4572         t10id = (struct scsi_vpd_id_t10 *)&desc->identifier[0];
4573         memset(t10id->vendor, ' ', sizeof(t10id->vendor));
4574         if ((vendor = dnvlist_get_string(be_lun->options, "vendor", NULL)) == NULL) {
4575                 strncpy((char *)t10id->vendor, CTL_VENDOR, sizeof(t10id->vendor));
4576         } else {
4577                 strncpy(t10id->vendor, vendor,
4578                     min(sizeof(t10id->vendor), strlen(vendor)));
4579         }
4580         strncpy((char *)t10id->vendor_spec_id,
4581             (char *)be_lun->device_id, devidlen);
4582         if (scsiname != NULL) {
4583                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4584                     desc->length);
4585                 desc->proto_codeset = SVPD_ID_CODESET_UTF8;
4586                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4587                     SVPD_ID_TYPE_SCSI_NAME;
4588                 desc->length = idlen2;
4589                 strlcpy(desc->identifier, scsiname, idlen2);
4590         }
4591         if (eui != NULL) {
4592                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4593                     desc->length);
4594                 desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4595                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4596                     SVPD_ID_TYPE_EUI64;
4597                 desc->length = hex2bin(eui, desc->identifier, 16);
4598                 desc->length = desc->length > 12 ? 16 :
4599                     (desc->length > 8 ? 12 : 8);
4600                 len -= 16 - desc->length;
4601         }
4602         if (naa != NULL) {
4603                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4604                     desc->length);
4605                 desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4606                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4607                     SVPD_ID_TYPE_NAA;
4608                 desc->length = hex2bin(naa, desc->identifier, 16);
4609                 desc->length = desc->length > 8 ? 16 : 8;
4610                 len -= 16 - desc->length;
4611         }
4612         if (uuid != NULL) {
4613                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4614                     desc->length);
4615                 desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4616                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4617                     SVPD_ID_TYPE_UUID;
4618                 desc->identifier[0] = 0x10;
4619                 hex2bin(uuid, &desc->identifier[2], 16);
4620                 desc->length = 18;
4621         }
4622         lun->lun_devid->len = len;
4623
4624         mtx_lock(&ctl_softc->ctl_lock);
4625         /*
4626          * See if the caller requested a particular LUN number.  If so, see
4627          * if it is available.  Otherwise, allocate the first available LUN.
4628          */
4629         if (be_lun->flags & CTL_LUN_FLAG_ID_REQ) {
4630                 if ((be_lun->req_lun_id > (ctl_max_luns - 1))
4631                  || (ctl_is_set(ctl_softc->ctl_lun_mask, be_lun->req_lun_id))) {
4632                         mtx_unlock(&ctl_softc->ctl_lock);
4633                         if (be_lun->req_lun_id > (ctl_max_luns - 1)) {
4634                                 printf("ctl: requested LUN ID %d is higher "
4635                                        "than ctl_max_luns - 1 (%d)\n",
4636                                        be_lun->req_lun_id, ctl_max_luns - 1);
4637                         } else {
4638                                 /*
4639                                  * XXX KDM return an error, or just assign
4640                                  * another LUN ID in this case??
4641                                  */
4642                                 printf("ctl: requested LUN ID %d is already "
4643                                        "in use\n", be_lun->req_lun_id);
4644                         }
4645 fail:
4646                         free(lun->lun_devid, M_CTL);
4647                         if (lun->flags & CTL_LUN_MALLOCED)
4648                                 free(lun, M_CTL);
4649                         be_lun->lun_config_status(be_lun->be_lun,
4650                                                   CTL_LUN_CONFIG_FAILURE);
4651                         return (ENOSPC);
4652                 }
4653                 lun_number = be_lun->req_lun_id;
4654         } else {
4655                 lun_number = ctl_ffz(ctl_softc->ctl_lun_mask, 0, ctl_max_luns);
4656                 if (lun_number == -1) {
4657                         mtx_unlock(&ctl_softc->ctl_lock);
4658                         printf("ctl: can't allocate LUN, out of LUNs\n");
4659                         goto fail;
4660                 }
4661         }
4662         ctl_set_mask(ctl_softc->ctl_lun_mask, lun_number);
4663         mtx_unlock(&ctl_softc->ctl_lock);
4664
4665         mtx_init(&lun->lun_lock, "CTL LUN", NULL, MTX_DEF);
4666         lun->lun = lun_number;
4667         lun->be_lun = be_lun;
4668         /*
4669          * The processor LUN is always enabled.  Disk LUNs come on line
4670          * disabled, and must be enabled by the backend.
4671          */
4672         lun->flags |= CTL_LUN_DISABLED;
4673         lun->backend = be_lun->be;
4674         be_lun->ctl_lun = lun;
4675         be_lun->lun_id = lun_number;
4676         atomic_add_int(&be_lun->be->num_luns, 1);
4677         if (be_lun->flags & CTL_LUN_FLAG_EJECTED)
4678                 lun->flags |= CTL_LUN_EJECTED;
4679         if (be_lun->flags & CTL_LUN_FLAG_NO_MEDIA)
4680                 lun->flags |= CTL_LUN_NO_MEDIA;
4681         if (be_lun->flags & CTL_LUN_FLAG_STOPPED)
4682                 lun->flags |= CTL_LUN_STOPPED;
4683
4684         if (be_lun->flags & CTL_LUN_FLAG_PRIMARY)
4685                 lun->flags |= CTL_LUN_PRIMARY_SC;
4686
4687         value = dnvlist_get_string(be_lun->options, "removable", NULL);
4688         if (value != NULL) {
4689                 if (strcmp(value, "on") == 0)
4690                         lun->flags |= CTL_LUN_REMOVABLE;
4691         } else if (be_lun->lun_type == T_CDROM)
4692                 lun->flags |= CTL_LUN_REMOVABLE;
4693
4694         lun->ctl_softc = ctl_softc;
4695 #ifdef CTL_TIME_IO
4696         lun->last_busy = getsbinuptime();
4697 #endif
4698         TAILQ_INIT(&lun->ooa_queue);
4699         TAILQ_INIT(&lun->blocked_queue);
4700         STAILQ_INIT(&lun->error_list);
4701         lun->ie_reported = 1;
4702         callout_init_mtx(&lun->ie_callout, &lun->lun_lock, 0);
4703         ctl_tpc_lun_init(lun);
4704         if (lun->flags & CTL_LUN_REMOVABLE) {
4705                 lun->prevent = malloc((CTL_MAX_INITIATORS + 31) / 32 * 4,
4706                     M_CTL, M_WAITOK);
4707         }
4708
4709         /*
4710          * Initialize the mode and log page index.
4711          */
4712         ctl_init_page_index(lun);
4713         ctl_init_log_page_index(lun);
4714
4715         /* Setup statistics gathering */
4716         lun->stats.item = lun_number;
4717
4718         /*
4719          * Now, before we insert this lun on the lun list, set the lun
4720          * inventory changed UA for all other luns.
4721          */
4722         mtx_lock(&ctl_softc->ctl_lock);
4723         STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) {
4724                 mtx_lock(&nlun->lun_lock);
4725                 ctl_est_ua_all(nlun, -1, CTL_UA_LUN_CHANGE);
4726                 mtx_unlock(&nlun->lun_lock);
4727         }
4728         STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links);
4729         ctl_softc->ctl_luns[lun_number] = lun;
4730         ctl_softc->num_luns++;
4731         mtx_unlock(&ctl_softc->ctl_lock);
4732
4733         lun->be_lun->lun_config_status(lun->be_lun->be_lun, CTL_LUN_CONFIG_OK);
4734         return (0);
4735 }
4736
4737 /*
4738  * Delete a LUN.
4739  * Assumptions:
4740  * - LUN has already been marked invalid and any pending I/O has been taken
4741  *   care of.
4742  */
4743 static int
4744 ctl_free_lun(struct ctl_lun *lun)
4745 {
4746         struct ctl_softc *softc = lun->ctl_softc;
4747         struct ctl_lun *nlun;
4748         int i;
4749
4750         KASSERT(TAILQ_EMPTY(&lun->ooa_queue),
4751             ("Freeing a LUN %p with outstanding I/O!\n", lun));
4752
4753         mtx_lock(&softc->ctl_lock);
4754         STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links);
4755         ctl_clear_mask(softc->ctl_lun_mask, lun->lun);
4756         softc->ctl_luns[lun->lun] = NULL;
4757         softc->num_luns--;
4758         STAILQ_FOREACH(nlun, &softc->lun_list, links) {
4759                 mtx_lock(&nlun->lun_lock);
4760                 ctl_est_ua_all(nlun, -1, CTL_UA_LUN_CHANGE);
4761                 mtx_unlock(&nlun->lun_lock);
4762         }
4763         mtx_unlock(&softc->ctl_lock);
4764
4765         /*
4766          * Tell the backend to free resources, if this LUN has a backend.
4767          */
4768         atomic_subtract_int(&lun->be_lun->be->num_luns, 1);
4769         lun->be_lun->lun_shutdown(lun->be_lun->be_lun);
4770
4771         lun->ie_reportcnt = UINT32_MAX;
4772         callout_drain(&lun->ie_callout);
4773         ctl_tpc_lun_shutdown(lun);
4774         mtx_destroy(&lun->lun_lock);
4775         free(lun->lun_devid, M_CTL);
4776         for (i = 0; i < ctl_max_ports; i++)
4777                 free(lun->pending_ua[i], M_CTL);
4778         free(lun->pending_ua, M_DEVBUF);
4779         for (i = 0; i < ctl_max_ports; i++)
4780                 free(lun->pr_keys[i], M_CTL);
4781         free(lun->pr_keys, M_DEVBUF);
4782         free(lun->write_buffer, M_CTL);
4783         free(lun->prevent, M_CTL);
4784         if (lun->flags & CTL_LUN_MALLOCED)
4785                 free(lun, M_CTL);
4786
4787         return (0);
4788 }
4789
4790 static void
4791 ctl_create_lun(struct ctl_be_lun *be_lun)
4792 {
4793
4794         /*
4795          * ctl_alloc_lun() should handle all potential failure cases.
4796          */
4797         ctl_alloc_lun(control_softc, NULL, be_lun);
4798 }
4799
4800 int
4801 ctl_add_lun(struct ctl_be_lun *be_lun)
4802 {
4803         struct ctl_softc *softc = control_softc;
4804
4805         mtx_lock(&softc->ctl_lock);
4806         STAILQ_INSERT_TAIL(&softc->pending_lun_queue, be_lun, links);
4807         mtx_unlock(&softc->ctl_lock);
4808         wakeup(&softc->pending_lun_queue);
4809
4810         return (0);
4811 }
4812
4813 int
4814 ctl_enable_lun(struct ctl_be_lun *be_lun)
4815 {
4816         struct ctl_softc *softc;
4817         struct ctl_port *port, *nport;
4818         struct ctl_lun *lun;
4819         int retval;
4820
4821         lun = (struct ctl_lun *)be_lun->ctl_lun;
4822         softc = lun->ctl_softc;
4823
4824         mtx_lock(&softc->ctl_lock);
4825         mtx_lock(&lun->lun_lock);
4826         if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4827                 /*
4828                  * eh?  Why did we get called if the LUN is already
4829                  * enabled?
4830                  */
4831                 mtx_unlock(&lun->lun_lock);
4832                 mtx_unlock(&softc->ctl_lock);
4833                 return (0);
4834         }
4835         lun->flags &= ~CTL_LUN_DISABLED;
4836         mtx_unlock(&lun->lun_lock);
4837
4838         STAILQ_FOREACH_SAFE(port, &softc->port_list, links, nport) {
4839                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0 ||
4840                     port->lun_map != NULL || port->lun_enable == NULL)
4841                         continue;
4842
4843                 /*
4844                  * Drop the lock while we call the FETD's enable routine.
4845                  * This can lead to a callback into CTL (at least in the
4846                  * case of the internal initiator frontend.
4847                  */
4848                 mtx_unlock(&softc->ctl_lock);
4849                 retval = port->lun_enable(port->targ_lun_arg, lun->lun);
4850                 mtx_lock(&softc->ctl_lock);
4851                 if (retval != 0) {
4852                         printf("%s: FETD %s port %d returned error "
4853                                "%d for lun_enable on lun %jd\n",
4854                                __func__, port->port_name, port->targ_port,
4855                                retval, (intmax_t)lun->lun);
4856                 }
4857         }
4858
4859         mtx_unlock(&softc->ctl_lock);
4860         ctl_isc_announce_lun(lun);
4861
4862         return (0);
4863 }
4864
4865 int
4866 ctl_disable_lun(struct ctl_be_lun *be_lun)
4867 {
4868         struct ctl_softc *softc;
4869         struct ctl_port *port;
4870         struct ctl_lun *lun;
4871         int retval;
4872
4873         lun = (struct ctl_lun *)be_lun->ctl_lun;
4874         softc = lun->ctl_softc;
4875
4876         mtx_lock(&softc->ctl_lock);
4877         mtx_lock(&lun->lun_lock);
4878         if (lun->flags & CTL_LUN_DISABLED) {
4879                 mtx_unlock(&lun->lun_lock);
4880                 mtx_unlock(&softc->ctl_lock);
4881                 return (0);
4882         }
4883         lun->flags |= CTL_LUN_DISABLED;
4884         mtx_unlock(&lun->lun_lock);
4885
4886         STAILQ_FOREACH(port, &softc->port_list, links) {
4887                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0 ||
4888                     port->lun_map != NULL || port->lun_disable == NULL)
4889                         continue;
4890
4891                 /*
4892                  * Drop the lock before we call the frontend's disable
4893                  * routine, to avoid lock order reversals.
4894                  *
4895                  * XXX KDM what happens if the frontend list changes while
4896                  * we're traversing it?  It's unlikely, but should be handled.
4897                  */
4898                 mtx_unlock(&softc->ctl_lock);
4899                 retval = port->lun_disable(port->targ_lun_arg, lun->lun);
4900                 mtx_lock(&softc->ctl_lock);
4901                 if (retval != 0) {
4902                         printf("%s: FETD %s port %d returned error "
4903                                "%d for lun_disable on lun %jd\n",
4904                                __func__, port->port_name, port->targ_port,
4905                                retval, (intmax_t)lun->lun);
4906                 }
4907         }
4908
4909         mtx_unlock(&softc->ctl_lock);
4910         ctl_isc_announce_lun(lun);
4911
4912         return (0);
4913 }
4914
4915 int
4916 ctl_start_lun(struct ctl_be_lun *be_lun)
4917 {
4918         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4919
4920         mtx_lock(&lun->lun_lock);
4921         lun->flags &= ~CTL_LUN_STOPPED;
4922         mtx_unlock(&lun->lun_lock);
4923         return (0);
4924 }
4925
4926 int
4927 ctl_stop_lun(struct ctl_be_lun *be_lun)
4928 {
4929         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4930
4931         mtx_lock(&lun->lun_lock);
4932         lun->flags |= CTL_LUN_STOPPED;
4933         mtx_unlock(&lun->lun_lock);
4934         return (0);
4935 }
4936
4937 int
4938 ctl_lun_no_media(struct ctl_be_lun *be_lun)
4939 {
4940         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4941
4942         mtx_lock(&lun->lun_lock);
4943         lun->flags |= CTL_LUN_NO_MEDIA;
4944         mtx_unlock(&lun->lun_lock);
4945         return (0);
4946 }
4947
4948 int
4949 ctl_lun_has_media(struct ctl_be_lun *be_lun)
4950 {
4951         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4952         union ctl_ha_msg msg;
4953
4954         mtx_lock(&lun->lun_lock);
4955         lun->flags &= ~(CTL_LUN_NO_MEDIA | CTL_LUN_EJECTED);
4956         if (lun->flags & CTL_LUN_REMOVABLE)
4957                 ctl_est_ua_all(lun, -1, CTL_UA_MEDIUM_CHANGE);
4958         mtx_unlock(&lun->lun_lock);
4959         if ((lun->flags & CTL_LUN_REMOVABLE) &&
4960             lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
4961                 bzero(&msg.ua, sizeof(msg.ua));
4962                 msg.hdr.msg_type = CTL_MSG_UA;
4963                 msg.hdr.nexus.initid = -1;
4964                 msg.hdr.nexus.targ_port = -1;
4965                 msg.hdr.nexus.targ_lun = lun->lun;
4966                 msg.hdr.nexus.targ_mapped_lun = lun->lun;
4967                 msg.ua.ua_all = 1;
4968                 msg.ua.ua_set = 1;
4969                 msg.ua.ua_type = CTL_UA_MEDIUM_CHANGE;
4970                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg.ua),
4971                     M_WAITOK);
4972         }
4973         return (0);
4974 }
4975
4976 int
4977 ctl_lun_ejected(struct ctl_be_lun *be_lun)
4978 {
4979         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4980
4981         mtx_lock(&lun->lun_lock);
4982         lun->flags |= CTL_LUN_EJECTED;
4983         mtx_unlock(&lun->lun_lock);
4984         return (0);
4985 }
4986
4987 int
4988 ctl_lun_primary(struct ctl_be_lun *be_lun)
4989 {
4990         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4991
4992         mtx_lock(&lun->lun_lock);
4993         lun->flags |= CTL_LUN_PRIMARY_SC;
4994         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
4995         mtx_unlock(&lun->lun_lock);
4996         ctl_isc_announce_lun(lun);
4997         return (0);
4998 }
4999
5000 int
5001 ctl_lun_secondary(struct ctl_be_lun *be_lun)
5002 {
5003         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
5004
5005         mtx_lock(&lun->lun_lock);
5006         lun->flags &= ~CTL_LUN_PRIMARY_SC;
5007         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
5008         mtx_unlock(&lun->lun_lock);
5009         ctl_isc_announce_lun(lun);
5010         return (0);
5011 }
5012
5013 int
5014 ctl_invalidate_lun(struct ctl_be_lun *be_lun)
5015 {
5016         struct ctl_lun *lun;
5017
5018         lun = (struct ctl_lun *)be_lun->ctl_lun;
5019
5020         mtx_lock(&lun->lun_lock);
5021
5022         /*
5023          * The LUN needs to be disabled before it can be marked invalid.
5024          */
5025         if ((lun->flags & CTL_LUN_DISABLED) == 0) {
5026                 mtx_unlock(&lun->lun_lock);
5027                 return (-1);
5028         }
5029         /*
5030          * Mark the LUN invalid.
5031          */
5032         lun->flags |= CTL_LUN_INVALID;
5033
5034         /*
5035          * If there is nothing in the OOA queue, go ahead and free the LUN.
5036          * If we have something in the OOA queue, we'll free it when the
5037          * last I/O completes.
5038          */
5039         if (TAILQ_EMPTY(&lun->ooa_queue)) {
5040                 mtx_unlock(&lun->lun_lock);
5041                 ctl_free_lun(lun);
5042         } else
5043                 mtx_unlock(&lun->lun_lock);
5044
5045         return (0);
5046 }
5047
5048 void
5049 ctl_lun_capacity_changed(struct ctl_be_lun *be_lun)
5050 {
5051         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
5052         union ctl_ha_msg msg;
5053
5054         mtx_lock(&lun->lun_lock);
5055         ctl_est_ua_all(lun, -1, CTL_UA_CAPACITY_CHANGE);
5056         mtx_unlock(&lun->lun_lock);
5057         if (lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
5058                 /* Send msg to other side. */
5059                 bzero(&msg.ua, sizeof(msg.ua));
5060                 msg.hdr.msg_type = CTL_MSG_UA;
5061                 msg.hdr.nexus.initid = -1;
5062                 msg.hdr.nexus.targ_port = -1;
5063                 msg.hdr.nexus.targ_lun = lun->lun;
5064                 msg.hdr.nexus.targ_mapped_lun = lun->lun;
5065                 msg.ua.ua_all = 1;
5066                 msg.ua.ua_set = 1;
5067                 msg.ua.ua_type = CTL_UA_CAPACITY_CHANGE;
5068                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg.ua),
5069                     M_WAITOK);
5070         }
5071 }
5072
5073 /*
5074  * Backend "memory move is complete" callback for requests that never
5075  * make it down to say RAIDCore's configuration code.
5076  */
5077 int
5078 ctl_config_move_done(union ctl_io *io)
5079 {
5080         int retval;
5081
5082         CTL_DEBUG_PRINT(("ctl_config_move_done\n"));
5083         KASSERT(io->io_hdr.io_type == CTL_IO_SCSI,
5084             ("Config I/O type isn't CTL_IO_SCSI (%d)!", io->io_hdr.io_type));
5085
5086         if ((io->io_hdr.port_status != 0) &&
5087             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5088              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5089                 ctl_set_internal_failure(&io->scsiio, /*sks_valid*/ 1,
5090                     /*retry_count*/ io->io_hdr.port_status);
5091         } else if (io->scsiio.kern_data_resid != 0 &&
5092             (io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT &&
5093             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5094              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5095                 ctl_set_invalid_field_ciu(&io->scsiio);
5096         }
5097
5098         if (ctl_debug & CTL_DEBUG_CDB_DATA)
5099                 ctl_data_print(io);
5100         if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN) ||
5101             ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
5102              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS) ||
5103             ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) {
5104                 /*
5105                  * XXX KDM just assuming a single pointer here, and not a
5106                  * S/G list.  If we start using S/G lists for config data,
5107                  * we'll need to know how to clean them up here as well.
5108                  */
5109                 if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5110                         free(io->scsiio.kern_data_ptr, M_CTL);
5111                 ctl_done(io);
5112                 retval = CTL_RETVAL_COMPLETE;
5113         } else {
5114                 /*
5115                  * XXX KDM now we need to continue data movement.  Some
5116                  * options:
5117                  * - call ctl_scsiio() again?  We don't do this for data
5118                  *   writes, because for those at least we know ahead of
5119                  *   time where the write will go and how long it is.  For
5120                  *   config writes, though, that information is largely
5121                  *   contained within the write itself, thus we need to
5122                  *   parse out the data again.
5123                  *
5124                  * - Call some other function once the data is in?
5125                  */
5126
5127                 /*
5128                  * XXX KDM call ctl_scsiio() again for now, and check flag
5129                  * bits to see whether we're allocated or not.
5130                  */
5131                 retval = ctl_scsiio(&io->scsiio);
5132         }
5133         return (retval);
5134 }
5135
5136 /*
5137  * This gets called by a backend driver when it is done with a
5138  * data_submit method.
5139  */
5140 void
5141 ctl_data_submit_done(union ctl_io *io)
5142 {
5143         /*
5144          * If the IO_CONT flag is set, we need to call the supplied
5145          * function to continue processing the I/O, instead of completing
5146          * the I/O just yet.
5147          *
5148          * If there is an error, though, we don't want to keep processing.
5149          * Instead, just send status back to the initiator.
5150          */
5151         if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5152             (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5153             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5154              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5155                 io->scsiio.io_cont(io);
5156                 return;
5157         }
5158         ctl_done(io);
5159 }
5160
5161 /*
5162  * This gets called by a backend driver when it is done with a
5163  * configuration write.
5164  */
5165 void
5166 ctl_config_write_done(union ctl_io *io)
5167 {
5168         uint8_t *buf;
5169
5170         /*
5171          * If the IO_CONT flag is set, we need to call the supplied
5172          * function to continue processing the I/O, instead of completing
5173          * the I/O just yet.
5174          *
5175          * If there is an error, though, we don't want to keep processing.
5176          * Instead, just send status back to the initiator.
5177          */
5178         if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5179             (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5180             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5181              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5182                 io->scsiio.io_cont(io);
5183                 return;
5184         }
5185         /*
5186          * Since a configuration write can be done for commands that actually
5187          * have data allocated, like write buffer, and commands that have
5188          * no data, like start/stop unit, we need to check here.
5189          */
5190         if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5191                 buf = io->scsiio.kern_data_ptr;
5192         else
5193                 buf = NULL;
5194         ctl_done(io);
5195         if (buf)
5196                 free(buf, M_CTL);
5197 }
5198
5199 void
5200 ctl_config_read_done(union ctl_io *io)
5201 {
5202         uint8_t *buf;
5203
5204         /*
5205          * If there is some error -- we are done, skip data transfer.
5206          */
5207         if ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0 ||
5208             ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
5209              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) {
5210                 if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5211                         buf = io->scsiio.kern_data_ptr;
5212                 else
5213                         buf = NULL;
5214                 ctl_done(io);
5215                 if (buf)
5216                         free(buf, M_CTL);
5217                 return;
5218         }
5219
5220         /*
5221          * If the IO_CONT flag is set, we need to call the supplied
5222          * function to continue processing the I/O, instead of completing
5223          * the I/O just yet.
5224          */
5225         if (io->io_hdr.flags & CTL_FLAG_IO_CONT) {
5226                 io->scsiio.io_cont(io);
5227                 return;
5228         }
5229
5230         ctl_datamove(io);
5231 }
5232
5233 /*
5234  * SCSI release command.
5235  */
5236 int
5237 ctl_scsi_release(struct ctl_scsiio *ctsio)
5238 {
5239         struct ctl_lun *lun = CTL_LUN(ctsio);
5240         uint32_t residx;
5241
5242         CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
5243
5244         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5245
5246         /*
5247          * XXX KDM right now, we only support LUN reservation.  We don't
5248          * support 3rd party reservations, or extent reservations, which
5249          * might actually need the parameter list.  If we've gotten this
5250          * far, we've got a LUN reservation.  Anything else got kicked out
5251          * above.  So, according to SPC, ignore the length.
5252          */
5253
5254         mtx_lock(&lun->lun_lock);
5255
5256         /*
5257          * According to SPC, it is not an error for an intiator to attempt
5258          * to release a reservation on a LUN that isn't reserved, or that
5259          * is reserved by another initiator.  The reservation can only be
5260          * released, though, by the initiator who made it or by one of
5261          * several reset type events.
5262          */
5263         if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == residx))
5264                         lun->flags &= ~CTL_LUN_RESERVED;
5265
5266         mtx_unlock(&lun->lun_lock);
5267
5268         ctl_set_success(ctsio);
5269         ctl_done((union ctl_io *)ctsio);
5270         return (CTL_RETVAL_COMPLETE);
5271 }
5272
5273 int
5274 ctl_scsi_reserve(struct ctl_scsiio *ctsio)
5275 {
5276         struct ctl_lun *lun = CTL_LUN(ctsio);
5277         uint32_t residx;
5278
5279         CTL_DEBUG_PRINT(("ctl_reserve\n"));
5280
5281         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5282
5283         /*
5284          * XXX KDM right now, we only support LUN reservation.  We don't
5285          * support 3rd party reservations, or extent reservations, which
5286          * might actually need the parameter list.  If we've gotten this
5287          * far, we've got a LUN reservation.  Anything else got kicked out
5288          * above.  So, according to SPC, ignore the length.
5289          */
5290
5291         mtx_lock(&lun->lun_lock);
5292         if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx != residx)) {
5293                 ctl_set_reservation_conflict(ctsio);
5294                 goto bailout;
5295         }
5296
5297         /* SPC-3 exceptions to SPC-2 RESERVE and RELEASE behavior. */
5298         if (lun->flags & CTL_LUN_PR_RESERVED) {
5299                 ctl_set_success(ctsio);
5300                 goto bailout;
5301         }
5302
5303         lun->flags |= CTL_LUN_RESERVED;
5304         lun->res_idx = residx;
5305         ctl_set_success(ctsio);
5306
5307 bailout:
5308         mtx_unlock(&lun->lun_lock);
5309         ctl_done((union ctl_io *)ctsio);
5310         return (CTL_RETVAL_COMPLETE);
5311 }
5312
5313 int
5314 ctl_start_stop(struct ctl_scsiio *ctsio)
5315 {
5316         struct ctl_lun *lun = CTL_LUN(ctsio);
5317         struct scsi_start_stop_unit *cdb;
5318         int retval;
5319
5320         CTL_DEBUG_PRINT(("ctl_start_stop\n"));
5321
5322         cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
5323
5324         if ((cdb->how & SSS_PC_MASK) == 0) {
5325                 if ((lun->flags & CTL_LUN_PR_RESERVED) &&
5326                     (cdb->how & SSS_START) == 0) {
5327                         uint32_t residx;
5328
5329                         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5330                         if (ctl_get_prkey(lun, residx) == 0 ||
5331                             (lun->pr_res_idx != residx && lun->pr_res_type < 4)) {
5332
5333                                 ctl_set_reservation_conflict(ctsio);
5334                                 ctl_done((union ctl_io *)ctsio);
5335                                 return (CTL_RETVAL_COMPLETE);
5336                         }
5337                 }
5338
5339                 if ((cdb->how & SSS_LOEJ) &&
5340                     (lun->flags & CTL_LUN_REMOVABLE) == 0) {
5341                         ctl_set_invalid_field(ctsio,
5342                                               /*sks_valid*/ 1,
5343                                               /*command*/ 1,
5344                                               /*field*/ 4,
5345                                               /*bit_valid*/ 1,
5346                                               /*bit*/ 1);
5347                         ctl_done((union ctl_io *)ctsio);
5348                         return (CTL_RETVAL_COMPLETE);
5349                 }
5350
5351                 if ((cdb->how & SSS_START) == 0 && (cdb->how & SSS_LOEJ) &&
5352                     lun->prevent_count > 0) {
5353                         /* "Medium removal prevented" */
5354                         ctl_set_sense(ctsio, /*current_error*/ 1,
5355                             /*sense_key*/(lun->flags & CTL_LUN_NO_MEDIA) ?
5356                              SSD_KEY_NOT_READY : SSD_KEY_ILLEGAL_REQUEST,
5357                             /*asc*/ 0x53, /*ascq*/ 0x02, SSD_ELEM_NONE);
5358                         ctl_done((union ctl_io *)ctsio);
5359                         return (CTL_RETVAL_COMPLETE);
5360                 }
5361         }
5362
5363         retval = lun->backend->config_write((union ctl_io *)ctsio);
5364         return (retval);
5365 }
5366
5367 int
5368 ctl_prevent_allow(struct ctl_scsiio *ctsio)
5369 {
5370         struct ctl_lun *lun = CTL_LUN(ctsio);
5371         struct scsi_prevent *cdb;
5372         int retval;
5373         uint32_t initidx;
5374
5375         CTL_DEBUG_PRINT(("ctl_prevent_allow\n"));
5376
5377         cdb = (struct scsi_prevent *)ctsio->cdb;
5378
5379         if ((lun->flags & CTL_LUN_REMOVABLE) == 0 || lun->prevent == NULL) {
5380                 ctl_set_invalid_opcode(ctsio);
5381                 ctl_done((union ctl_io *)ctsio);
5382                 return (CTL_RETVAL_COMPLETE);
5383         }
5384
5385         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5386         mtx_lock(&lun->lun_lock);
5387         if ((cdb->how & PR_PREVENT) &&
5388             ctl_is_set(lun->prevent, initidx) == 0) {
5389                 ctl_set_mask(lun->prevent, initidx);
5390                 lun->prevent_count++;
5391         } else if ((cdb->how & PR_PREVENT) == 0 &&
5392             ctl_is_set(lun->prevent, initidx)) {
5393                 ctl_clear_mask(lun->prevent, initidx);
5394                 lun->prevent_count--;
5395         }
5396         mtx_unlock(&lun->lun_lock);
5397         retval = lun->backend->config_write((union ctl_io *)ctsio);
5398         return (retval);
5399 }
5400
5401 /*
5402  * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
5403  * we don't really do anything with the LBA and length fields if the user
5404  * passes them in.  Instead we'll just flush out the cache for the entire
5405  * LUN.
5406  */
5407 int
5408 ctl_sync_cache(struct ctl_scsiio *ctsio)
5409 {
5410         struct ctl_lun *lun = CTL_LUN(ctsio);
5411         struct ctl_lba_len_flags *lbalen;
5412         uint64_t starting_lba;
5413         uint32_t block_count;
5414         int retval;
5415         uint8_t byte2;
5416
5417         CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
5418
5419         retval = 0;
5420
5421         switch (ctsio->cdb[0]) {
5422         case SYNCHRONIZE_CACHE: {
5423                 struct scsi_sync_cache *cdb;
5424                 cdb = (struct scsi_sync_cache *)ctsio->cdb;
5425
5426                 starting_lba = scsi_4btoul(cdb->begin_lba);
5427                 block_count = scsi_2btoul(cdb->lb_count);
5428                 byte2 = cdb->byte2;
5429                 break;
5430         }
5431         case SYNCHRONIZE_CACHE_16: {
5432                 struct scsi_sync_cache_16 *cdb;
5433                 cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5434
5435                 starting_lba = scsi_8btou64(cdb->begin_lba);
5436                 block_count = scsi_4btoul(cdb->lb_count);
5437                 byte2 = cdb->byte2;
5438                 break;
5439         }
5440         default:
5441                 ctl_set_invalid_opcode(ctsio);
5442                 ctl_done((union ctl_io *)ctsio);
5443                 goto bailout;
5444                 break; /* NOTREACHED */
5445         }
5446
5447         /*
5448          * We check the LBA and length, but don't do anything with them.
5449          * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5450          * get flushed.  This check will just help satisfy anyone who wants
5451          * to see an error for an out of range LBA.
5452          */
5453         if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5454                 ctl_set_lba_out_of_range(ctsio,
5455                     MAX(starting_lba, lun->be_lun->maxlba + 1));
5456                 ctl_done((union ctl_io *)ctsio);
5457                 goto bailout;
5458         }
5459
5460         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5461         lbalen->lba = starting_lba;
5462         lbalen->len = block_count;
5463         lbalen->flags = byte2;
5464         retval = lun->backend->config_write((union ctl_io *)ctsio);
5465
5466 bailout:
5467         return (retval);
5468 }
5469
5470 int
5471 ctl_format(struct ctl_scsiio *ctsio)
5472 {
5473         struct scsi_format *cdb;
5474         int length, defect_list_len;
5475
5476         CTL_DEBUG_PRINT(("ctl_format\n"));
5477
5478         cdb = (struct scsi_format *)ctsio->cdb;
5479
5480         length = 0;
5481         if (cdb->byte2 & SF_FMTDATA) {
5482                 if (cdb->byte2 & SF_LONGLIST)
5483                         length = sizeof(struct scsi_format_header_long);
5484                 else
5485                         length = sizeof(struct scsi_format_header_short);
5486         }
5487
5488         if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5489          && (length > 0)) {
5490                 ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5491                 ctsio->kern_data_len = length;
5492                 ctsio->kern_total_len = length;
5493                 ctsio->kern_rel_offset = 0;
5494                 ctsio->kern_sg_entries = 0;
5495                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5496                 ctsio->be_move_done = ctl_config_move_done;
5497                 ctl_datamove((union ctl_io *)ctsio);
5498
5499                 return (CTL_RETVAL_COMPLETE);
5500         }
5501
5502         defect_list_len = 0;
5503
5504         if (cdb->byte2 & SF_FMTDATA) {
5505                 if (cdb->byte2 & SF_LONGLIST) {
5506                         struct scsi_format_header_long *header;
5507
5508                         header = (struct scsi_format_header_long *)
5509                                 ctsio->kern_data_ptr;
5510
5511                         defect_list_len = scsi_4btoul(header->defect_list_len);
5512                         if (defect_list_len != 0) {
5513                                 ctl_set_invalid_field(ctsio,
5514                                                       /*sks_valid*/ 1,
5515                                                       /*command*/ 0,
5516                                                       /*field*/ 2,
5517                                                       /*bit_valid*/ 0,
5518                                                       /*bit*/ 0);
5519                                 goto bailout;
5520                         }
5521                 } else {
5522                         struct scsi_format_header_short *header;
5523
5524                         header = (struct scsi_format_header_short *)
5525                                 ctsio->kern_data_ptr;
5526
5527                         defect_list_len = scsi_2btoul(header->defect_list_len);
5528                         if (defect_list_len != 0) {
5529                                 ctl_set_invalid_field(ctsio,
5530                                                       /*sks_valid*/ 1,
5531                                                       /*command*/ 0,
5532                                                       /*field*/ 2,
5533                                                       /*bit_valid*/ 0,
5534                                                       /*bit*/ 0);
5535                                 goto bailout;
5536                         }
5537                 }
5538         }
5539
5540         ctl_set_success(ctsio);
5541 bailout:
5542
5543         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5544                 free(ctsio->kern_data_ptr, M_CTL);
5545                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5546         }
5547
5548         ctl_done((union ctl_io *)ctsio);
5549         return (CTL_RETVAL_COMPLETE);
5550 }
5551
5552 int
5553 ctl_read_buffer(struct ctl_scsiio *ctsio)
5554 {
5555         struct ctl_lun *lun = CTL_LUN(ctsio);
5556         uint64_t buffer_offset;
5557         uint32_t len;
5558         uint8_t byte2;
5559         static uint8_t descr[4];
5560         static uint8_t echo_descr[4] = { 0 };
5561
5562         CTL_DEBUG_PRINT(("ctl_read_buffer\n"));
5563
5564         switch (ctsio->cdb[0]) {
5565         case READ_BUFFER: {
5566                 struct scsi_read_buffer *cdb;
5567
5568                 cdb = (struct scsi_read_buffer *)ctsio->cdb;
5569                 buffer_offset = scsi_3btoul(cdb->offset);
5570                 len = scsi_3btoul(cdb->length);
5571                 byte2 = cdb->byte2;
5572                 break;
5573         }
5574         case READ_BUFFER_16: {
5575                 struct scsi_read_buffer_16 *cdb;
5576
5577                 cdb = (struct scsi_read_buffer_16 *)ctsio->cdb;
5578                 buffer_offset = scsi_8btou64(cdb->offset);
5579                 len = scsi_4btoul(cdb->length);
5580                 byte2 = cdb->byte2;
5581                 break;
5582         }
5583         default: /* This shouldn't happen. */
5584                 ctl_set_invalid_opcode(ctsio);
5585                 ctl_done((union ctl_io *)ctsio);
5586                 return (CTL_RETVAL_COMPLETE);
5587         }
5588
5589         if (buffer_offset > CTL_WRITE_BUFFER_SIZE ||
5590             buffer_offset + len > CTL_WRITE_BUFFER_SIZE) {
5591                 ctl_set_invalid_field(ctsio,
5592                                       /*sks_valid*/ 1,
5593                                       /*command*/ 1,
5594                                       /*field*/ 6,
5595                                       /*bit_valid*/ 0,
5596                                       /*bit*/ 0);
5597                 ctl_done((union ctl_io *)ctsio);
5598                 return (CTL_RETVAL_COMPLETE);
5599         }
5600
5601         if ((byte2 & RWB_MODE) == RWB_MODE_DESCR) {
5602                 descr[0] = 0;
5603                 scsi_ulto3b(CTL_WRITE_BUFFER_SIZE, &descr[1]);
5604                 ctsio->kern_data_ptr = descr;
5605                 len = min(len, sizeof(descr));
5606         } else if ((byte2 & RWB_MODE) == RWB_MODE_ECHO_DESCR) {
5607                 ctsio->kern_data_ptr = echo_descr;
5608                 len = min(len, sizeof(echo_descr));
5609         } else {
5610                 if (lun->write_buffer == NULL) {
5611                         lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE,
5612                             M_CTL, M_WAITOK);
5613                 }
5614                 ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5615         }
5616         ctsio->kern_data_len = len;
5617         ctsio->kern_total_len = len;
5618         ctsio->kern_rel_offset = 0;
5619         ctsio->kern_sg_entries = 0;
5620         ctl_set_success(ctsio);
5621         ctsio->be_move_done = ctl_config_move_done;
5622         ctl_datamove((union ctl_io *)ctsio);
5623         return (CTL_RETVAL_COMPLETE);
5624 }
5625
5626 int
5627 ctl_write_buffer(struct ctl_scsiio *ctsio)
5628 {
5629         struct ctl_lun *lun = CTL_LUN(ctsio);
5630         struct scsi_write_buffer *cdb;
5631         int buffer_offset, len;
5632
5633         CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5634
5635         cdb = (struct scsi_write_buffer *)ctsio->cdb;
5636
5637         len = scsi_3btoul(cdb->length);
5638         buffer_offset = scsi_3btoul(cdb->offset);
5639
5640         if (buffer_offset + len > CTL_WRITE_BUFFER_SIZE) {
5641                 ctl_set_invalid_field(ctsio,
5642                                       /*sks_valid*/ 1,
5643                                       /*command*/ 1,
5644                                       /*field*/ 6,
5645                                       /*bit_valid*/ 0,
5646                                       /*bit*/ 0);
5647                 ctl_done((union ctl_io *)ctsio);
5648                 return (CTL_RETVAL_COMPLETE);
5649         }
5650
5651         /*
5652          * If we've got a kernel request that hasn't been malloced yet,
5653          * malloc it and tell the caller the data buffer is here.
5654          */
5655         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5656                 if (lun->write_buffer == NULL) {
5657                         lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE,
5658                             M_CTL, M_WAITOK);
5659                 }
5660                 ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5661                 ctsio->kern_data_len = len;
5662                 ctsio->kern_total_len = len;
5663                 ctsio->kern_rel_offset = 0;
5664                 ctsio->kern_sg_entries = 0;
5665                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5666                 ctsio->be_move_done = ctl_config_move_done;
5667                 ctl_datamove((union ctl_io *)ctsio);
5668
5669                 return (CTL_RETVAL_COMPLETE);
5670         }
5671
5672         ctl_set_success(ctsio);
5673         ctl_done((union ctl_io *)ctsio);
5674         return (CTL_RETVAL_COMPLETE);
5675 }
5676
5677 int
5678 ctl_write_same(struct ctl_scsiio *ctsio)
5679 {
5680         struct ctl_lun *lun = CTL_LUN(ctsio);
5681         struct ctl_lba_len_flags *lbalen;
5682         uint64_t lba;
5683         uint32_t num_blocks;
5684         int len, retval;
5685         uint8_t byte2;
5686
5687         CTL_DEBUG_PRINT(("ctl_write_same\n"));
5688
5689         switch (ctsio->cdb[0]) {
5690         case WRITE_SAME_10: {
5691                 struct scsi_write_same_10 *cdb;
5692
5693                 cdb = (struct scsi_write_same_10 *)ctsio->cdb;
5694
5695                 lba = scsi_4btoul(cdb->addr);
5696                 num_blocks = scsi_2btoul(cdb->length);
5697                 byte2 = cdb->byte2;
5698                 break;
5699         }
5700         case WRITE_SAME_16: {
5701                 struct scsi_write_same_16 *cdb;
5702
5703                 cdb = (struct scsi_write_same_16 *)ctsio->cdb;
5704
5705                 lba = scsi_8btou64(cdb->addr);
5706                 num_blocks = scsi_4btoul(cdb->length);
5707                 byte2 = cdb->byte2;
5708                 break;
5709         }
5710         default:
5711                 /*
5712                  * We got a command we don't support.  This shouldn't
5713                  * happen, commands should be filtered out above us.
5714                  */
5715                 ctl_set_invalid_opcode(ctsio);
5716                 ctl_done((union ctl_io *)ctsio);
5717
5718                 return (CTL_RETVAL_COMPLETE);
5719                 break; /* NOTREACHED */
5720         }
5721
5722         /* ANCHOR flag can be used only together with UNMAP */
5723         if ((byte2 & SWS_UNMAP) == 0 && (byte2 & SWS_ANCHOR) != 0) {
5724                 ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
5725                     /*command*/ 1, /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 0);
5726                 ctl_done((union ctl_io *)ctsio);
5727                 return (CTL_RETVAL_COMPLETE);
5728         }
5729
5730         /*
5731          * The first check is to make sure we're in bounds, the second
5732          * check is to catch wrap-around problems.  If the lba + num blocks
5733          * is less than the lba, then we've wrapped around and the block
5734          * range is invalid anyway.
5735          */
5736         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5737          || ((lba + num_blocks) < lba)) {
5738                 ctl_set_lba_out_of_range(ctsio,
5739                     MAX(lba, lun->be_lun->maxlba + 1));
5740                 ctl_done((union ctl_io *)ctsio);
5741                 return (CTL_RETVAL_COMPLETE);
5742         }
5743
5744         /* Zero number of blocks means "to the last logical block" */
5745         if (num_blocks == 0) {
5746                 if ((lun->be_lun->maxlba + 1) - lba > UINT32_MAX) {
5747                         ctl_set_invalid_field(ctsio,
5748                                               /*sks_valid*/ 0,
5749                                               /*command*/ 1,
5750                                               /*field*/ 0,
5751                                               /*bit_valid*/ 0,
5752                                               /*bit*/ 0);
5753                         ctl_done((union ctl_io *)ctsio);
5754                         return (CTL_RETVAL_COMPLETE);
5755                 }
5756                 num_blocks = (lun->be_lun->maxlba + 1) - lba;
5757         }
5758
5759         len = lun->be_lun->blocksize;
5760
5761         /*
5762          * If we've got a kernel request that hasn't been malloced yet,
5763          * malloc it and tell the caller the data buffer is here.
5764          */
5765         if ((byte2 & SWS_NDOB) == 0 &&
5766             (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5767                 ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);
5768                 ctsio->kern_data_len = len;
5769                 ctsio->kern_total_len = len;
5770                 ctsio->kern_rel_offset = 0;
5771                 ctsio->kern_sg_entries = 0;
5772                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5773                 ctsio->be_move_done = ctl_config_move_done;
5774                 ctl_datamove((union ctl_io *)ctsio);
5775
5776                 return (CTL_RETVAL_COMPLETE);
5777         }
5778
5779         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5780         lbalen->lba = lba;
5781         lbalen->len = num_blocks;
5782         lbalen->flags = byte2;
5783         retval = lun->backend->config_write((union ctl_io *)ctsio);
5784
5785         return (retval);
5786 }
5787
5788 int
5789 ctl_unmap(struct ctl_scsiio *ctsio)
5790 {
5791         struct ctl_lun *lun = CTL_LUN(ctsio);
5792         struct scsi_unmap *cdb;
5793         struct ctl_ptr_len_flags *ptrlen;
5794         struct scsi_unmap_header *hdr;
5795         struct scsi_unmap_desc *buf, *end, *endnz, *range;
5796         uint64_t lba;
5797         uint32_t num_blocks;
5798         int len, retval;
5799         uint8_t byte2;
5800
5801         CTL_DEBUG_PRINT(("ctl_unmap\n"));
5802
5803         cdb = (struct scsi_unmap *)ctsio->cdb;
5804         len = scsi_2btoul(cdb->length);
5805         byte2 = cdb->byte2;
5806
5807         /*
5808          * If we've got a kernel request that hasn't been malloced yet,
5809          * malloc it and tell the caller the data buffer is here.
5810          */
5811         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5812                 ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);
5813                 ctsio->kern_data_len = len;
5814                 ctsio->kern_total_len = len;
5815                 ctsio->kern_rel_offset = 0;
5816                 ctsio->kern_sg_entries = 0;
5817                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5818                 ctsio->be_move_done = ctl_config_move_done;
5819                 ctl_datamove((union ctl_io *)ctsio);
5820
5821                 return (CTL_RETVAL_COMPLETE);
5822         }
5823
5824         len = ctsio->kern_total_len - ctsio->kern_data_resid;
5825         hdr = (struct scsi_unmap_header *)ctsio->kern_data_ptr;
5826         if (len < sizeof (*hdr) ||
5827             len < (scsi_2btoul(hdr->length) + sizeof(hdr->length)) ||
5828             len < (scsi_2btoul(hdr->desc_length) + sizeof (*hdr)) ||
5829             scsi_2btoul(hdr->desc_length) % sizeof(*buf) != 0) {
5830                 ctl_set_invalid_field(ctsio,
5831                                       /*sks_valid*/ 0,
5832                                       /*command*/ 0,
5833                                       /*field*/ 0,
5834                                       /*bit_valid*/ 0,
5835                                       /*bit*/ 0);
5836                 goto done;
5837         }
5838         len = scsi_2btoul(hdr->desc_length);
5839         buf = (struct scsi_unmap_desc *)(hdr + 1);
5840         end = buf + len / sizeof(*buf);
5841
5842         endnz = buf;
5843         for (range = buf; range < end; range++) {
5844                 lba = scsi_8btou64(range->lba);
5845                 num_blocks = scsi_4btoul(range->length);
5846                 if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5847                  || ((lba + num_blocks) < lba)) {
5848                         ctl_set_lba_out_of_range(ctsio,
5849                             MAX(lba, lun->be_lun->maxlba + 1));
5850                         ctl_done((union ctl_io *)ctsio);
5851                         return (CTL_RETVAL_COMPLETE);
5852                 }
5853                 if (num_blocks != 0)
5854                         endnz = range + 1;
5855         }
5856
5857         /*
5858          * Block backend can not handle zero last range.
5859          * Filter it out and return if there is nothing left.
5860          */
5861         len = (uint8_t *)endnz - (uint8_t *)buf;
5862         if (len == 0) {
5863                 ctl_set_success(ctsio);
5864                 goto done;
5865         }
5866
5867         mtx_lock(&lun->lun_lock);
5868         ptrlen = (struct ctl_ptr_len_flags *)
5869             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5870         ptrlen->ptr = (void *)buf;
5871         ptrlen->len = len;
5872         ptrlen->flags = byte2;
5873         ctl_check_blocked(lun);
5874         mtx_unlock(&lun->lun_lock);
5875
5876         retval = lun->backend->config_write((union ctl_io *)ctsio);
5877         return (retval);
5878
5879 done:
5880         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5881                 free(ctsio->kern_data_ptr, M_CTL);
5882                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5883         }
5884         ctl_done((union ctl_io *)ctsio);
5885         return (CTL_RETVAL_COMPLETE);
5886 }
5887
5888 int
5889 ctl_default_page_handler(struct ctl_scsiio *ctsio,
5890                          struct ctl_page_index *page_index, uint8_t *page_ptr)
5891 {
5892         struct ctl_lun *lun = CTL_LUN(ctsio);
5893         uint8_t *current_cp;
5894         int set_ua;
5895         uint32_t initidx;
5896
5897         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5898         set_ua = 0;
5899
5900         current_cp = (page_index->page_data + (page_index->page_len *
5901             CTL_PAGE_CURRENT));
5902
5903         mtx_lock(&lun->lun_lock);
5904         if (memcmp(current_cp, page_ptr, page_index->page_len)) {
5905                 memcpy(current_cp, page_ptr, page_index->page_len);
5906                 set_ua = 1;
5907         }
5908         if (set_ua != 0)
5909                 ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
5910         mtx_unlock(&lun->lun_lock);
5911         if (set_ua) {
5912                 ctl_isc_announce_mode(lun,
5913                     ctl_get_initindex(&ctsio->io_hdr.nexus),
5914                     page_index->page_code, page_index->subpage);
5915         }
5916         return (CTL_RETVAL_COMPLETE);
5917 }
5918
5919 static void
5920 ctl_ie_timer(void *arg)
5921 {
5922         struct ctl_lun *lun = arg;
5923         uint64_t t;
5924
5925         if (lun->ie_asc == 0)
5926                 return;
5927
5928         if (lun->MODE_IE.mrie == SIEP_MRIE_UA)
5929                 ctl_est_ua_all(lun, -1, CTL_UA_IE);
5930         else
5931                 lun->ie_reported = 0;
5932
5933         if (lun->ie_reportcnt < scsi_4btoul(lun->MODE_IE.report_count)) {
5934                 lun->ie_reportcnt++;
5935                 t = scsi_4btoul(lun->MODE_IE.interval_timer);
5936                 if (t == 0 || t == UINT32_MAX)
5937                         t = 3000;  /* 5 min */
5938                 callout_schedule(&lun->ie_callout, t * hz / 10);
5939         }
5940 }
5941
5942 int
5943 ctl_ie_page_handler(struct ctl_scsiio *ctsio,
5944                          struct ctl_page_index *page_index, uint8_t *page_ptr)
5945 {
5946         struct ctl_lun *lun = CTL_LUN(ctsio);
5947         struct scsi_info_exceptions_page *pg;
5948         uint64_t t;
5949
5950         (void)ctl_default_page_handler(ctsio, page_index, page_ptr);
5951
5952         pg = (struct scsi_info_exceptions_page *)page_ptr;
5953         mtx_lock(&lun->lun_lock);
5954         if (pg->info_flags & SIEP_FLAGS_TEST) {
5955                 lun->ie_asc = 0x5d;
5956                 lun->ie_ascq = 0xff;
5957                 if (pg->mrie == SIEP_MRIE_UA) {
5958                         ctl_est_ua_all(lun, -1, CTL_UA_IE);
5959                         lun->ie_reported = 1;
5960                 } else {
5961                         ctl_clr_ua_all(lun, -1, CTL_UA_IE);
5962                         lun->ie_reported = -1;
5963                 }
5964                 lun->ie_reportcnt = 1;
5965                 if (lun->ie_reportcnt < scsi_4btoul(pg->report_count)) {
5966                         lun->ie_reportcnt++;
5967                         t = scsi_4btoul(pg->interval_timer);
5968                         if (t == 0 || t == UINT32_MAX)
5969                                 t = 3000;  /* 5 min */
5970                         callout_reset(&lun->ie_callout, t * hz / 10,
5971                             ctl_ie_timer, lun);
5972                 }
5973         } else {
5974                 lun->ie_asc = 0;
5975                 lun->ie_ascq = 0;
5976                 lun->ie_reported = 1;
5977                 ctl_clr_ua_all(lun, -1, CTL_UA_IE);
5978                 lun->ie_reportcnt = UINT32_MAX;
5979                 callout_stop(&lun->ie_callout);
5980         }
5981         mtx_unlock(&lun->lun_lock);
5982         return (CTL_RETVAL_COMPLETE);
5983 }
5984
5985 static int
5986 ctl_do_mode_select(union ctl_io *io)
5987 {
5988         struct ctl_lun *lun = CTL_LUN(io);
5989         struct scsi_mode_page_header *page_header;
5990         struct ctl_page_index *page_index;
5991         struct ctl_scsiio *ctsio;
5992         int page_len, page_len_offset, page_len_size;
5993         union ctl_modepage_info *modepage_info;
5994         uint16_t *len_left, *len_used;
5995         int retval, i;
5996
5997         ctsio = &io->scsiio;
5998         page_index = NULL;
5999         page_len = 0;
6000
6001         modepage_info = (union ctl_modepage_info *)
6002                 ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6003         len_left = &modepage_info->header.len_left;
6004         len_used = &modepage_info->header.len_used;
6005
6006 do_next_page:
6007
6008         page_header = (struct scsi_mode_page_header *)
6009                 (ctsio->kern_data_ptr + *len_used);
6010
6011         if (*len_left == 0) {
6012                 free(ctsio->kern_data_ptr, M_CTL);
6013                 ctl_set_success(ctsio);
6014                 ctl_done((union ctl_io *)ctsio);
6015                 return (CTL_RETVAL_COMPLETE);
6016         } else if (*len_left < sizeof(struct scsi_mode_page_header)) {
6017
6018                 free(ctsio->kern_data_ptr, M_CTL);
6019                 ctl_set_param_len_error(ctsio);
6020                 ctl_done((union ctl_io *)ctsio);
6021                 return (CTL_RETVAL_COMPLETE);
6022
6023         } else if ((page_header->page_code & SMPH_SPF)
6024                 && (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
6025
6026                 free(ctsio->kern_data_ptr, M_CTL);
6027                 ctl_set_param_len_error(ctsio);
6028                 ctl_done((union ctl_io *)ctsio);
6029                 return (CTL_RETVAL_COMPLETE);
6030         }
6031
6032
6033         /*
6034          * XXX KDM should we do something with the block descriptor?
6035          */
6036         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6037                 page_index = &lun->mode_pages.index[i];
6038                 if (lun->be_lun->lun_type == T_DIRECT &&
6039                     (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6040                         continue;
6041                 if (lun->be_lun->lun_type == T_PROCESSOR &&
6042                     (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6043                         continue;
6044                 if (lun->be_lun->lun_type == T_CDROM &&
6045                     (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6046                         continue;
6047
6048                 if ((page_index->page_code & SMPH_PC_MASK) !=
6049                     (page_header->page_code & SMPH_PC_MASK))
6050                         continue;
6051
6052                 /*
6053                  * If neither page has a subpage code, then we've got a
6054                  * match.
6055                  */
6056                 if (((page_index->page_code & SMPH_SPF) == 0)
6057                  && ((page_header->page_code & SMPH_SPF) == 0)) {
6058                         page_len = page_header->page_length;
6059                         break;
6060                 }
6061
6062                 /*
6063                  * If both pages have subpages, then the subpage numbers
6064                  * have to match.
6065                  */
6066                 if ((page_index->page_code & SMPH_SPF)
6067                   && (page_header->page_code & SMPH_SPF)) {
6068                         struct scsi_mode_page_header_sp *sph;
6069
6070                         sph = (struct scsi_mode_page_header_sp *)page_header;
6071                         if (page_index->subpage == sph->subpage) {
6072                                 page_len = scsi_2btoul(sph->page_length);
6073                                 break;
6074                         }
6075                 }
6076         }
6077
6078         /*
6079          * If we couldn't find the page, or if we don't have a mode select
6080          * handler for it, send back an error to the user.
6081          */
6082         if ((i >= CTL_NUM_MODE_PAGES)
6083          || (page_index->select_handler == NULL)) {
6084                 ctl_set_invalid_field(ctsio,
6085                                       /*sks_valid*/ 1,
6086                                       /*command*/ 0,
6087                                       /*field*/ *len_used,
6088                                       /*bit_valid*/ 0,
6089                                       /*bit*/ 0);
6090                 free(ctsio->kern_data_ptr, M_CTL);
6091                 ctl_done((union ctl_io *)ctsio);
6092                 return (CTL_RETVAL_COMPLETE);
6093         }
6094
6095         if (page_index->page_code & SMPH_SPF) {
6096                 page_len_offset = 2;
6097                 page_len_size = 2;
6098         } else {
6099                 page_len_size = 1;
6100                 page_len_offset = 1;
6101         }
6102
6103         /*
6104          * If the length the initiator gives us isn't the one we specify in
6105          * the mode page header, or if they didn't specify enough data in
6106          * the CDB to avoid truncating this page, kick out the request.
6107          */
6108         if (page_len != page_index->page_len - page_len_offset - page_len_size) {
6109                 ctl_set_invalid_field(ctsio,
6110                                       /*sks_valid*/ 1,
6111                                       /*command*/ 0,
6112                                       /*field*/ *len_used + page_len_offset,
6113                                       /*bit_valid*/ 0,
6114                                       /*bit*/ 0);
6115                 free(ctsio->kern_data_ptr, M_CTL);
6116                 ctl_done((union ctl_io *)ctsio);
6117                 return (CTL_RETVAL_COMPLETE);
6118         }
6119         if (*len_left < page_index->page_len) {
6120                 free(ctsio->kern_data_ptr, M_CTL);
6121                 ctl_set_param_len_error(ctsio);
6122                 ctl_done((union ctl_io *)ctsio);
6123                 return (CTL_RETVAL_COMPLETE);
6124         }
6125
6126         /*
6127          * Run through the mode page, checking to make sure that the bits
6128          * the user changed are actually legal for him to change.
6129          */
6130         for (i = 0; i < page_index->page_len; i++) {
6131                 uint8_t *user_byte, *change_mask, *current_byte;
6132                 int bad_bit;
6133                 int j;
6134
6135                 user_byte = (uint8_t *)page_header + i;
6136                 change_mask = page_index->page_data +
6137                               (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
6138                 current_byte = page_index->page_data +
6139                                (page_index->page_len * CTL_PAGE_CURRENT) + i;
6140
6141                 /*
6142                  * Check to see whether the user set any bits in this byte
6143                  * that he is not allowed to set.
6144                  */
6145                 if ((*user_byte & ~(*change_mask)) ==
6146                     (*current_byte & ~(*change_mask)))
6147                         continue;
6148
6149                 /*
6150                  * Go through bit by bit to determine which one is illegal.
6151                  */
6152                 bad_bit = 0;
6153                 for (j = 7; j >= 0; j--) {
6154                         if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
6155                             (((1 << i) & ~(*change_mask)) & *current_byte)) {
6156                                 bad_bit = i;
6157                                 break;
6158                         }
6159                 }
6160                 ctl_set_invalid_field(ctsio,
6161                                       /*sks_valid*/ 1,
6162                                       /*command*/ 0,
6163                                       /*field*/ *len_used + i,
6164                                       /*bit_valid*/ 1,
6165                                       /*bit*/ bad_bit);
6166                 free(ctsio->kern_data_ptr, M_CTL);
6167                 ctl_done((union ctl_io *)ctsio);
6168                 return (CTL_RETVAL_COMPLETE);
6169         }
6170
6171         /*
6172          * Decrement these before we call the page handler, since we may
6173          * end up getting called back one way or another before the handler
6174          * returns to this context.
6175          */
6176         *len_left -= page_index->page_len;
6177         *len_used += page_index->page_len;
6178
6179         retval = page_index->select_handler(ctsio, page_index,
6180                                             (uint8_t *)page_header);
6181
6182         /*
6183          * If the page handler returns CTL_RETVAL_QUEUED, then we need to
6184          * wait until this queued command completes to finish processing
6185          * the mode page.  If it returns anything other than
6186          * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
6187          * already set the sense information, freed the data pointer, and
6188          * completed the io for us.
6189          */
6190         if (retval != CTL_RETVAL_COMPLETE)
6191                 goto bailout_no_done;
6192
6193         /*
6194          * If the initiator sent us more than one page, parse the next one.
6195          */
6196         if (*len_left > 0)
6197                 goto do_next_page;
6198
6199         ctl_set_success(ctsio);
6200         free(ctsio->kern_data_ptr, M_CTL);
6201         ctl_done((union ctl_io *)ctsio);
6202
6203 bailout_no_done:
6204
6205         return (CTL_RETVAL_COMPLETE);
6206
6207 }
6208
6209 int
6210 ctl_mode_select(struct ctl_scsiio *ctsio)
6211 {
6212         struct ctl_lun *lun = CTL_LUN(ctsio);
6213         union ctl_modepage_info *modepage_info;
6214         int bd_len, i, header_size, param_len, rtd;
6215         uint32_t initidx;
6216
6217         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
6218         switch (ctsio->cdb[0]) {
6219         case MODE_SELECT_6: {
6220                 struct scsi_mode_select_6 *cdb;
6221
6222                 cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
6223
6224                 rtd = (cdb->byte2 & SMS_RTD) ? 1 : 0;
6225                 param_len = cdb->length;
6226                 header_size = sizeof(struct scsi_mode_header_6);
6227                 break;
6228         }
6229         case MODE_SELECT_10: {
6230                 struct scsi_mode_select_10 *cdb;
6231
6232                 cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
6233
6234                 rtd = (cdb->byte2 & SMS_RTD) ? 1 : 0;
6235                 param_len = scsi_2btoul(cdb->length);
6236                 header_size = sizeof(struct scsi_mode_header_10);
6237                 break;
6238         }
6239         default:
6240                 ctl_set_invalid_opcode(ctsio);
6241                 ctl_done((union ctl_io *)ctsio);
6242                 return (CTL_RETVAL_COMPLETE);
6243         }
6244
6245         if (rtd) {
6246                 if (param_len != 0) {
6247                         ctl_set_invalid_field(ctsio, /*sks_valid*/ 0,
6248                             /*command*/ 1, /*field*/ 0,
6249                             /*bit_valid*/ 0, /*bit*/ 0);
6250                         ctl_done((union ctl_io *)ctsio);
6251                         return (CTL_RETVAL_COMPLETE);
6252                 }
6253
6254                 /* Revert to defaults. */
6255                 ctl_init_page_index(lun);
6256                 mtx_lock(&lun->lun_lock);
6257                 ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
6258                 mtx_unlock(&lun->lun_lock);
6259                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6260                         ctl_isc_announce_mode(lun, -1,
6261                             lun->mode_pages.index[i].page_code & SMPH_PC_MASK,
6262                             lun->mode_pages.index[i].subpage);
6263                 }
6264                 ctl_set_success(ctsio);
6265                 ctl_done((union ctl_io *)ctsio);
6266                 return (CTL_RETVAL_COMPLETE);
6267         }
6268
6269         /*
6270          * From SPC-3:
6271          * "A parameter list length of zero indicates that the Data-Out Buffer
6272          * shall be empty. This condition shall not be considered as an error."
6273          */
6274         if (param_len == 0) {
6275                 ctl_set_success(ctsio);
6276                 ctl_done((union ctl_io *)ctsio);
6277                 return (CTL_RETVAL_COMPLETE);
6278         }
6279
6280         /*
6281          * Since we'll hit this the first time through, prior to
6282          * allocation, we don't need to free a data buffer here.
6283          */
6284         if (param_len < header_size) {
6285                 ctl_set_param_len_error(ctsio);
6286                 ctl_done((union ctl_io *)ctsio);
6287                 return (CTL_RETVAL_COMPLETE);
6288         }
6289
6290         /*
6291          * Allocate the data buffer and grab the user's data.  In theory,
6292          * we shouldn't have to sanity check the parameter list length here
6293          * because the maximum size is 64K.  We should be able to malloc
6294          * that much without too many problems.
6295          */
6296         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6297                 ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6298                 ctsio->kern_data_len = param_len;
6299                 ctsio->kern_total_len = param_len;
6300                 ctsio->kern_rel_offset = 0;
6301                 ctsio->kern_sg_entries = 0;
6302                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6303                 ctsio->be_move_done = ctl_config_move_done;
6304                 ctl_datamove((union ctl_io *)ctsio);
6305
6306                 return (CTL_RETVAL_COMPLETE);
6307         }
6308
6309         switch (ctsio->cdb[0]) {
6310         case MODE_SELECT_6: {
6311                 struct scsi_mode_header_6 *mh6;
6312
6313                 mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6314                 bd_len = mh6->blk_desc_len;
6315                 break;
6316         }
6317         case MODE_SELECT_10: {
6318                 struct scsi_mode_header_10 *mh10;
6319
6320                 mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6321                 bd_len = scsi_2btoul(mh10->blk_desc_len);
6322                 break;
6323         }
6324         default:
6325                 panic("%s: Invalid CDB type %#x", __func__, ctsio->cdb[0]);
6326         }
6327
6328         if (param_len < (header_size + bd_len)) {
6329                 free(ctsio->kern_data_ptr, M_CTL);
6330                 ctl_set_param_len_error(ctsio);
6331                 ctl_done((union ctl_io *)ctsio);
6332                 return (CTL_RETVAL_COMPLETE);
6333         }
6334
6335         /*
6336          * Set the IO_CONT flag, so that if this I/O gets passed to
6337          * ctl_config_write_done(), it'll get passed back to
6338          * ctl_do_mode_select() for further processing, or completion if
6339          * we're all done.
6340          */
6341         ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6342         ctsio->io_cont = ctl_do_mode_select;
6343
6344         modepage_info = (union ctl_modepage_info *)
6345                 ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6346         memset(modepage_info, 0, sizeof(*modepage_info));
6347         modepage_info->header.len_left = param_len - header_size - bd_len;
6348         modepage_info->header.len_used = header_size + bd_len;
6349
6350         return (ctl_do_mode_select((union ctl_io *)ctsio));
6351 }
6352
6353 int
6354 ctl_mode_sense(struct ctl_scsiio *ctsio)
6355 {
6356         struct ctl_lun *lun = CTL_LUN(ctsio);
6357         int pc, page_code, dbd, subpage;
6358         int alloc_len, page_len, header_len, total_len;
6359         struct scsi_mode_block_descr *block_desc;
6360         struct ctl_page_index *page_index;
6361
6362         dbd = 0;
6363         block_desc = NULL;
6364
6365         CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6366
6367         switch (ctsio->cdb[0]) {
6368         case MODE_SENSE_6: {
6369                 struct scsi_mode_sense_6 *cdb;
6370
6371                 cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6372
6373                 header_len = sizeof(struct scsi_mode_hdr_6);
6374                 if (cdb->byte2 & SMS_DBD)
6375                         dbd = 1;
6376                 else
6377                         header_len += sizeof(struct scsi_mode_block_descr);
6378
6379                 pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6380                 page_code = cdb->page & SMS_PAGE_CODE;
6381                 subpage = cdb->subpage;
6382                 alloc_len = cdb->length;
6383                 break;
6384         }
6385         case MODE_SENSE_10: {
6386                 struct scsi_mode_sense_10 *cdb;
6387
6388                 cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6389
6390                 header_len = sizeof(struct scsi_mode_hdr_10);
6391
6392                 if (cdb->byte2 & SMS_DBD)
6393                         dbd = 1;
6394                 else
6395                         header_len += sizeof(struct scsi_mode_block_descr);
6396                 pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6397                 page_code = cdb->page & SMS_PAGE_CODE;
6398                 subpage = cdb->subpage;
6399                 alloc_len = scsi_2btoul(cdb->length);
6400                 break;
6401         }
6402         default:
6403                 ctl_set_invalid_opcode(ctsio);
6404                 ctl_done((union ctl_io *)ctsio);
6405                 return (CTL_RETVAL_COMPLETE);
6406                 break; /* NOTREACHED */
6407         }
6408
6409         /*
6410          * We have to make a first pass through to calculate the size of
6411          * the pages that match the user's query.  Then we allocate enough
6412          * memory to hold it, and actually copy the data into the buffer.
6413          */
6414         switch (page_code) {
6415         case SMS_ALL_PAGES_PAGE: {
6416                 u_int i;
6417
6418                 page_len = 0;
6419
6420                 /*
6421                  * At the moment, values other than 0 and 0xff here are
6422                  * reserved according to SPC-3.
6423                  */
6424                 if ((subpage != SMS_SUBPAGE_PAGE_0)
6425                  && (subpage != SMS_SUBPAGE_ALL)) {
6426                         ctl_set_invalid_field(ctsio,
6427                                               /*sks_valid*/ 1,
6428                                               /*command*/ 1,
6429                                               /*field*/ 3,
6430                                               /*bit_valid*/ 0,
6431                                               /*bit*/ 0);
6432                         ctl_done((union ctl_io *)ctsio);
6433                         return (CTL_RETVAL_COMPLETE);
6434                 }
6435
6436                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6437                         page_index = &lun->mode_pages.index[i];
6438
6439                         /* Make sure the page is supported for this dev type */
6440                         if (lun->be_lun->lun_type == T_DIRECT &&
6441                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6442                                 continue;
6443                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6444                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6445                                 continue;
6446                         if (lun->be_lun->lun_type == T_CDROM &&
6447                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6448                                 continue;
6449
6450                         /*
6451                          * We don't use this subpage if the user didn't
6452                          * request all subpages.
6453                          */
6454                         if ((page_index->subpage != 0)
6455                          && (subpage == SMS_SUBPAGE_PAGE_0))
6456                                 continue;
6457
6458                         page_len += page_index->page_len;
6459                 }
6460                 break;
6461         }
6462         default: {
6463                 u_int i;
6464
6465                 page_len = 0;
6466
6467                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6468                         page_index = &lun->mode_pages.index[i];
6469
6470                         /* Make sure the page is supported for this dev type */
6471                         if (lun->be_lun->lun_type == T_DIRECT &&
6472                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6473                                 continue;
6474                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6475                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6476                                 continue;
6477                         if (lun->be_lun->lun_type == T_CDROM &&
6478                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6479                                 continue;
6480
6481                         /* Look for the right page code */
6482                         if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6483                                 continue;
6484
6485                         /* Look for the right subpage or the subpage wildcard*/
6486                         if ((page_index->subpage != subpage)
6487                          && (subpage != SMS_SUBPAGE_ALL))
6488                                 continue;
6489
6490                         page_len += page_index->page_len;
6491                 }
6492
6493                 if (page_len == 0) {
6494                         ctl_set_invalid_field(ctsio,
6495                                               /*sks_valid*/ 1,
6496                                               /*command*/ 1,
6497                                               /*field*/ 2,
6498                                               /*bit_valid*/ 1,
6499                                               /*bit*/ 5);
6500                         ctl_done((union ctl_io *)ctsio);
6501                         return (CTL_RETVAL_COMPLETE);
6502                 }
6503                 break;
6504         }
6505         }
6506
6507         total_len = header_len + page_len;
6508
6509         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6510         ctsio->kern_sg_entries = 0;
6511         ctsio->kern_rel_offset = 0;
6512         ctsio->kern_data_len = min(total_len, alloc_len);
6513         ctsio->kern_total_len = ctsio->kern_data_len;
6514
6515         switch (ctsio->cdb[0]) {
6516         case MODE_SENSE_6: {
6517                 struct scsi_mode_hdr_6 *header;
6518
6519                 header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
6520
6521                 header->datalen = MIN(total_len - 1, 254);
6522                 if (lun->be_lun->lun_type == T_DIRECT) {
6523                         header->dev_specific = 0x10; /* DPOFUA */
6524                         if ((lun->be_lun->flags & CTL_LUN_FLAG_READONLY) ||
6525                             (lun->MODE_CTRL.eca_and_aen & SCP_SWP) != 0)
6526                                 header->dev_specific |= 0x80; /* WP */
6527                 }
6528                 if (dbd)
6529                         header->block_descr_len = 0;
6530                 else
6531                         header->block_descr_len =
6532                                 sizeof(struct scsi_mode_block_descr);
6533                 block_desc = (struct scsi_mode_block_descr *)&header[1];
6534                 break;
6535         }
6536         case MODE_SENSE_10: {
6537                 struct scsi_mode_hdr_10 *header;
6538                 int datalen;
6539
6540                 header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
6541
6542                 datalen = MIN(total_len - 2, 65533);
6543                 scsi_ulto2b(datalen, header->datalen);
6544                 if (lun->be_lun->lun_type == T_DIRECT) {
6545                         header->dev_specific = 0x10; /* DPOFUA */
6546                         if ((lun->be_lun->flags & CTL_LUN_FLAG_READONLY) ||
6547                             (lun->MODE_CTRL.eca_and_aen & SCP_SWP) != 0)
6548                                 header->dev_specific |= 0x80; /* WP */
6549                 }
6550                 if (dbd)
6551                         scsi_ulto2b(0, header->block_descr_len);
6552                 else
6553                         scsi_ulto2b(sizeof(struct scsi_mode_block_descr),
6554                                     header->block_descr_len);
6555                 block_desc = (struct scsi_mode_block_descr *)&header[1];
6556                 break;
6557         }
6558         default:
6559                 panic("%s: Invalid CDB type %#x", __func__, ctsio->cdb[0]);
6560         }
6561
6562         /*
6563          * If we've got a disk, use its blocksize in the block
6564          * descriptor.  Otherwise, just set it to 0.
6565          */
6566         if (dbd == 0) {
6567                 if (lun->be_lun->lun_type == T_DIRECT)
6568                         scsi_ulto3b(lun->be_lun->blocksize,
6569                                     block_desc->block_len);
6570                 else
6571                         scsi_ulto3b(0, block_desc->block_len);
6572         }
6573
6574         switch (page_code) {
6575         case SMS_ALL_PAGES_PAGE: {
6576                 int i, data_used;
6577
6578                 data_used = header_len;
6579                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6580                         struct ctl_page_index *page_index;
6581
6582                         page_index = &lun->mode_pages.index[i];
6583                         if (lun->be_lun->lun_type == T_DIRECT &&
6584                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6585                                 continue;
6586                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6587                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6588                                 continue;
6589                         if (lun->be_lun->lun_type == T_CDROM &&
6590                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6591                                 continue;
6592
6593                         /*
6594                          * We don't use this subpage if the user didn't
6595                          * request all subpages.  We already checked (above)
6596                          * to make sure the user only specified a subpage
6597                          * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case.
6598                          */
6599                         if ((page_index->subpage != 0)
6600                          && (subpage == SMS_SUBPAGE_PAGE_0))
6601                                 continue;
6602
6603                         /*
6604                          * Call the handler, if it exists, to update the
6605                          * page to the latest values.
6606                          */
6607                         if (page_index->sense_handler != NULL)
6608                                 page_index->sense_handler(ctsio, page_index,pc);
6609
6610                         memcpy(ctsio->kern_data_ptr + data_used,
6611                                page_index->page_data +
6612                                (page_index->page_len * pc),
6613                                page_index->page_len);
6614                         data_used += page_index->page_len;
6615                 }
6616                 break;
6617         }
6618         default: {
6619                 int i, data_used;
6620
6621                 data_used = header_len;
6622
6623                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6624                         struct ctl_page_index *page_index;
6625
6626                         page_index = &lun->mode_pages.index[i];
6627
6628                         /* Look for the right page code */
6629                         if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6630                                 continue;
6631
6632                         /* Look for the right subpage or the subpage wildcard*/
6633                         if ((page_index->subpage != subpage)
6634                          && (subpage != SMS_SUBPAGE_ALL))
6635                                 continue;
6636
6637                         /* Make sure the page is supported for this dev type */
6638                         if (lun->be_lun->lun_type == T_DIRECT &&
6639                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6640                                 continue;
6641                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6642                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6643                                 continue;
6644                         if (lun->be_lun->lun_type == T_CDROM &&
6645                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6646                                 continue;
6647
6648                         /*
6649                          * Call the handler, if it exists, to update the
6650                          * page to the latest values.
6651                          */
6652                         if (page_index->sense_handler != NULL)
6653                                 page_index->sense_handler(ctsio, page_index,pc);
6654
6655                         memcpy(ctsio->kern_data_ptr + data_used,
6656                                page_index->page_data +
6657                                (page_index->page_len * pc),
6658                                page_index->page_len);
6659                         data_used += page_index->page_len;
6660                 }
6661                 break;
6662         }
6663         }
6664
6665         ctl_set_success(ctsio);
6666         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6667         ctsio->be_move_done = ctl_config_move_done;
6668         ctl_datamove((union ctl_io *)ctsio);
6669         return (CTL_RETVAL_COMPLETE);
6670 }
6671
6672 int
6673 ctl_lbp_log_sense_handler(struct ctl_scsiio *ctsio,
6674                                struct ctl_page_index *page_index,
6675                                int pc)
6676 {
6677         struct ctl_lun *lun = CTL_LUN(ctsio);
6678         struct scsi_log_param_header *phdr;
6679         uint8_t *data;
6680         uint64_t val;
6681
6682         data = page_index->page_data;
6683
6684         if (lun->backend->lun_attr != NULL &&
6685             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "blocksavail"))
6686              != UINT64_MAX) {
6687                 phdr = (struct scsi_log_param_header *)data;
6688                 scsi_ulto2b(0x0001, phdr->param_code);
6689                 phdr->param_control = SLP_LBIN | SLP_LP;
6690                 phdr->param_len = 8;
6691                 data = (uint8_t *)(phdr + 1);
6692                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6693                 data[4] = 0x02; /* per-pool */
6694                 data += phdr->param_len;
6695         }
6696
6697         if (lun->backend->lun_attr != NULL &&
6698             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "blocksused"))
6699              != UINT64_MAX) {
6700                 phdr = (struct scsi_log_param_header *)data;
6701                 scsi_ulto2b(0x0002, phdr->param_code);
6702                 phdr->param_control = SLP_LBIN | SLP_LP;
6703                 phdr->param_len = 8;
6704                 data = (uint8_t *)(phdr + 1);
6705                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6706                 data[4] = 0x01; /* per-LUN */
6707                 data += phdr->param_len;
6708         }
6709
6710         if (lun->backend->lun_attr != NULL &&
6711             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "poolblocksavail"))
6712              != UINT64_MAX) {
6713                 phdr = (struct scsi_log_param_header *)data;
6714                 scsi_ulto2b(0x00f1, phdr->param_code);
6715                 phdr->param_control = SLP_LBIN | SLP_LP;
6716                 phdr->param_len = 8;
6717                 data = (uint8_t *)(phdr + 1);
6718                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6719                 data[4] = 0x02; /* per-pool */
6720                 data += phdr->param_len;
6721         }
6722
6723         if (lun->backend->lun_attr != NULL &&
6724             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "poolblocksused"))
6725              != UINT64_MAX) {
6726                 phdr = (struct scsi_log_param_header *)data;
6727                 scsi_ulto2b(0x00f2, phdr->param_code);
6728                 phdr->param_control = SLP_LBIN | SLP_LP;
6729                 phdr->param_len = 8;
6730                 data = (uint8_t *)(phdr + 1);
6731                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6732                 data[4] = 0x02; /* per-pool */
6733                 data += phdr->param_len;
6734         }
6735
6736         page_index->page_len = data - page_index->page_data;
6737         return (0);
6738 }
6739
6740 int
6741 ctl_sap_log_sense_handler(struct ctl_scsiio *ctsio,
6742                                struct ctl_page_index *page_index,
6743                                int pc)
6744 {
6745         struct ctl_lun *lun = CTL_LUN(ctsio);
6746         struct stat_page *data;
6747         struct bintime *t;
6748
6749         data = (struct stat_page *)page_index->page_data;
6750
6751         scsi_ulto2b(SLP_SAP, data->sap.hdr.param_code);
6752         data->sap.hdr.param_control = SLP_LBIN;
6753         data->sap.hdr.param_len = sizeof(struct scsi_log_stat_and_perf) -
6754             sizeof(struct scsi_log_param_header);
6755         scsi_u64to8b(lun->stats.operations[CTL_STATS_READ],
6756             data->sap.read_num);
6757         scsi_u64to8b(lun->stats.operations[CTL_STATS_WRITE],
6758             data->sap.write_num);
6759         if (lun->be_lun->blocksize > 0) {
6760                 scsi_u64to8b(lun->stats.bytes[CTL_STATS_WRITE] /
6761                     lun->be_lun->blocksize, data->sap.recvieved_lba);
6762                 scsi_u64to8b(lun->stats.bytes[CTL_STATS_READ] /
6763                     lun->be_lun->blocksize, data->sap.transmitted_lba);
6764         }
6765         t = &lun->stats.time[CTL_STATS_READ];
6766         scsi_u64to8b((uint64_t)t->sec * 1000 + t->frac / (UINT64_MAX / 1000),
6767             data->sap.read_int);
6768         t = &lun->stats.time[CTL_STATS_WRITE];
6769         scsi_u64to8b((uint64_t)t->sec * 1000 + t->frac / (UINT64_MAX / 1000),
6770             data->sap.write_int);
6771         scsi_u64to8b(0, data->sap.weighted_num);
6772         scsi_u64to8b(0, data->sap.weighted_int);
6773         scsi_ulto2b(SLP_IT, data->it.hdr.param_code);
6774         data->it.hdr.param_control = SLP_LBIN;
6775         data->it.hdr.param_len = sizeof(struct scsi_log_idle_time) -
6776             sizeof(struct scsi_log_param_header);
6777 #ifdef CTL_TIME_IO
6778         scsi_u64to8b(lun->idle_time / SBT_1MS, data->it.idle_int);
6779 #endif
6780         scsi_ulto2b(SLP_TI, data->ti.hdr.param_code);
6781         data->it.hdr.param_control = SLP_LBIN;
6782         data->ti.hdr.param_len = sizeof(struct scsi_log_time_interval) -
6783             sizeof(struct scsi_log_param_header);
6784         scsi_ulto4b(3, data->ti.exponent);
6785         scsi_ulto4b(1, data->ti.integer);
6786         return (0);
6787 }
6788
6789 int
6790 ctl_ie_log_sense_handler(struct ctl_scsiio *ctsio,
6791                                struct ctl_page_index *page_index,
6792                                int pc)
6793 {
6794         struct ctl_lun *lun = CTL_LUN(ctsio);
6795         struct scsi_log_informational_exceptions *data;
6796
6797         data = (struct scsi_log_informational_exceptions *)page_index->page_data;
6798
6799         scsi_ulto2b(SLP_IE_GEN, data->hdr.param_code);
6800         data->hdr.param_control = SLP_LBIN;
6801         data->hdr.param_len = sizeof(struct scsi_log_informational_exceptions) -
6802             sizeof(struct scsi_log_param_header);
6803         data->ie_asc = lun->ie_asc;
6804         data->ie_ascq = lun->ie_ascq;
6805         data->temperature = 0xff;
6806         return (0);
6807 }
6808
6809 int
6810 ctl_log_sense(struct ctl_scsiio *ctsio)
6811 {
6812         struct ctl_lun *lun = CTL_LUN(ctsio);
6813         int i, pc, page_code, subpage;
6814         int alloc_len, total_len;
6815         struct ctl_page_index *page_index;
6816         struct scsi_log_sense *cdb;
6817         struct scsi_log_header *header;
6818
6819         CTL_DEBUG_PRINT(("ctl_log_sense\n"));
6820
6821         cdb = (struct scsi_log_sense *)ctsio->cdb;
6822         pc = (cdb->page & SLS_PAGE_CTRL_MASK) >> 6;
6823         page_code = cdb->page & SLS_PAGE_CODE;
6824         subpage = cdb->subpage;
6825         alloc_len = scsi_2btoul(cdb->length);
6826
6827         page_index = NULL;
6828         for (i = 0; i < CTL_NUM_LOG_PAGES; i++) {
6829                 page_index = &lun->log_pages.index[i];
6830
6831                 /* Look for the right page code */
6832                 if ((page_index->page_code & SL_PAGE_CODE) != page_code)
6833                         continue;
6834
6835                 /* Look for the right subpage or the subpage wildcard*/
6836                 if (page_index->subpage != subpage)
6837                         continue;
6838
6839                 break;
6840         }
6841         if (i >= CTL_NUM_LOG_PAGES) {
6842                 ctl_set_invalid_field(ctsio,
6843                                       /*sks_valid*/ 1,
6844                                       /*command*/ 1,
6845                                       /*field*/ 2,
6846                                       /*bit_valid*/ 0,
6847                                       /*bit*/ 0);
6848                 ctl_done((union ctl_io *)ctsio);
6849                 return (CTL_RETVAL_COMPLETE);
6850         }
6851
6852         total_len = sizeof(struct scsi_log_header) + page_index->page_len;
6853
6854         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6855         ctsio->kern_sg_entries = 0;
6856         ctsio->kern_rel_offset = 0;
6857         ctsio->kern_data_len = min(total_len, alloc_len);
6858         ctsio->kern_total_len = ctsio->kern_data_len;
6859
6860         header = (struct scsi_log_header *)ctsio->kern_data_ptr;
6861         header->page = page_index->page_code;
6862         if (page_index->page_code == SLS_LOGICAL_BLOCK_PROVISIONING)
6863                 header->page |= SL_DS;
6864         if (page_index->subpage) {
6865                 header->page |= SL_SPF;
6866                 header->subpage = page_index->subpage;
6867         }
6868         scsi_ulto2b(page_index->page_len, header->datalen);
6869
6870         /*
6871          * Call the handler, if it exists, to update the
6872          * page to the latest values.
6873          */
6874         if (page_index->sense_handler != NULL)
6875                 page_index->sense_handler(ctsio, page_index, pc);
6876
6877         memcpy(header + 1, page_index->page_data, page_index->page_len);
6878
6879         ctl_set_success(ctsio);
6880         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6881         ctsio->be_move_done = ctl_config_move_done;
6882         ctl_datamove((union ctl_io *)ctsio);
6883         return (CTL_RETVAL_COMPLETE);
6884 }
6885
6886 int
6887 ctl_read_capacity(struct ctl_scsiio *ctsio)
6888 {
6889         struct ctl_lun *lun = CTL_LUN(ctsio);
6890         struct scsi_read_capacity *cdb;
6891         struct scsi_read_capacity_data *data;
6892         uint32_t lba;
6893
6894         CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
6895
6896         cdb = (struct scsi_read_capacity *)ctsio->cdb;
6897
6898         lba = scsi_4btoul(cdb->addr);
6899         if (((cdb->pmi & SRC_PMI) == 0)
6900          && (lba != 0)) {
6901                 ctl_set_invalid_field(/*ctsio*/ ctsio,
6902                                       /*sks_valid*/ 1,
6903                                       /*command*/ 1,
6904                                       /*field*/ 2,
6905                                       /*bit_valid*/ 0,
6906                                       /*bit*/ 0);
6907                 ctl_done((union ctl_io *)ctsio);
6908                 return (CTL_RETVAL_COMPLETE);
6909         }
6910
6911         ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
6912         data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
6913         ctsio->kern_data_len = sizeof(*data);
6914         ctsio->kern_total_len = sizeof(*data);
6915         ctsio->kern_rel_offset = 0;
6916         ctsio->kern_sg_entries = 0;
6917
6918         /*
6919          * If the maximum LBA is greater than 0xfffffffe, the user must
6920          * issue a SERVICE ACTION IN (16) command, with the read capacity
6921          * serivce action set.
6922          */
6923         if (lun->be_lun->maxlba > 0xfffffffe)
6924                 scsi_ulto4b(0xffffffff, data->addr);
6925         else
6926                 scsi_ulto4b(lun->be_lun->maxlba, data->addr);
6927
6928         /*
6929          * XXX KDM this may not be 512 bytes...
6930          */
6931         scsi_ulto4b(lun->be_lun->blocksize, data->length);
6932
6933         ctl_set_success(ctsio);
6934         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6935         ctsio->be_move_done = ctl_config_move_done;
6936         ctl_datamove((union ctl_io *)ctsio);
6937         return (CTL_RETVAL_COMPLETE);
6938 }
6939
6940 int
6941 ctl_read_capacity_16(struct ctl_scsiio *ctsio)
6942 {
6943         struct ctl_lun *lun = CTL_LUN(ctsio);
6944         struct scsi_read_capacity_16 *cdb;
6945         struct scsi_read_capacity_data_long *data;
6946         uint64_t lba;
6947         uint32_t alloc_len;
6948
6949         CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
6950
6951         cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
6952
6953         alloc_len = scsi_4btoul(cdb->alloc_len);
6954         lba = scsi_8btou64(cdb->addr);
6955
6956         if ((cdb->reladr & SRC16_PMI)
6957          && (lba != 0)) {
6958                 ctl_set_invalid_field(/*ctsio*/ ctsio,
6959                                       /*sks_valid*/ 1,
6960                                       /*command*/ 1,
6961                                       /*field*/ 2,
6962                                       /*bit_valid*/ 0,
6963                                       /*bit*/ 0);
6964                 ctl_done((union ctl_io *)ctsio);
6965                 return (CTL_RETVAL_COMPLETE);
6966         }
6967
6968         ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
6969         data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
6970         ctsio->kern_rel_offset = 0;
6971         ctsio->kern_sg_entries = 0;
6972         ctsio->kern_data_len = min(sizeof(*data), alloc_len);
6973         ctsio->kern_total_len = ctsio->kern_data_len;
6974
6975         scsi_u64to8b(lun->be_lun->maxlba, data->addr);
6976         /* XXX KDM this may not be 512 bytes... */
6977         scsi_ulto4b(lun->be_lun->blocksize, data->length);
6978         data->prot_lbppbe = lun->be_lun->pblockexp & SRC16_LBPPBE;
6979         scsi_ulto2b(lun->be_lun->pblockoff & SRC16_LALBA_A, data->lalba_lbp);
6980         if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
6981                 data->lalba_lbp[0] |= SRC16_LBPME | SRC16_LBPRZ;
6982
6983         ctl_set_success(ctsio);
6984         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6985         ctsio->be_move_done = ctl_config_move_done;
6986         ctl_datamove((union ctl_io *)ctsio);
6987         return (CTL_RETVAL_COMPLETE);
6988 }
6989
6990 int
6991 ctl_get_lba_status(struct ctl_scsiio *ctsio)
6992 {
6993         struct ctl_lun *lun = CTL_LUN(ctsio);
6994         struct scsi_get_lba_status *cdb;
6995         struct scsi_get_lba_status_data *data;
6996         struct ctl_lba_len_flags *lbalen;
6997         uint64_t lba;
6998         uint32_t alloc_len, total_len;
6999         int retval;
7000
7001         CTL_DEBUG_PRINT(("ctl_get_lba_status\n"));
7002
7003         cdb = (struct scsi_get_lba_status *)ctsio->cdb;
7004         lba = scsi_8btou64(cdb->addr);
7005         alloc_len = scsi_4btoul(cdb->alloc_len);
7006
7007         if (lba > lun->be_lun->maxlba) {
7008                 ctl_set_lba_out_of_range(ctsio, lba);
7009                 ctl_done((union ctl_io *)ctsio);
7010                 return (CTL_RETVAL_COMPLETE);
7011         }
7012
7013         total_len = sizeof(*data) + sizeof(data->descr[0]);
7014         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7015         data = (struct scsi_get_lba_status_data *)ctsio->kern_data_ptr;
7016         ctsio->kern_rel_offset = 0;
7017         ctsio->kern_sg_entries = 0;
7018         ctsio->kern_data_len = min(total_len, alloc_len);
7019         ctsio->kern_total_len = ctsio->kern_data_len;
7020
7021         /* Fill dummy data in case backend can't tell anything. */
7022         scsi_ulto4b(4 + sizeof(data->descr[0]), data->length);
7023         scsi_u64to8b(lba, data->descr[0].addr);
7024         scsi_ulto4b(MIN(UINT32_MAX, lun->be_lun->maxlba + 1 - lba),
7025             data->descr[0].length);
7026         data->descr[0].status = 0; /* Mapped or unknown. */
7027
7028         ctl_set_success(ctsio);
7029         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7030         ctsio->be_move_done = ctl_config_move_done;
7031
7032         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
7033         lbalen->lba = lba;
7034         lbalen->len = total_len;
7035         lbalen->flags = 0;
7036         retval = lun->backend->config_read((union ctl_io *)ctsio);
7037         return (retval);
7038 }
7039
7040 int
7041 ctl_read_defect(struct ctl_scsiio *ctsio)
7042 {
7043         struct scsi_read_defect_data_10 *ccb10;
7044         struct scsi_read_defect_data_12 *ccb12;
7045         struct scsi_read_defect_data_hdr_10 *data10;
7046         struct scsi_read_defect_data_hdr_12 *data12;
7047         uint32_t alloc_len, data_len;
7048         uint8_t format;
7049
7050         CTL_DEBUG_PRINT(("ctl_read_defect\n"));
7051
7052         if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
7053                 ccb10 = (struct scsi_read_defect_data_10 *)&ctsio->cdb;
7054                 format = ccb10->format;
7055                 alloc_len = scsi_2btoul(ccb10->alloc_length);
7056                 data_len = sizeof(*data10);
7057         } else {
7058                 ccb12 = (struct scsi_read_defect_data_12 *)&ctsio->cdb;
7059                 format = ccb12->format;
7060                 alloc_len = scsi_4btoul(ccb12->alloc_length);
7061                 data_len = sizeof(*data12);
7062         }
7063         if (alloc_len == 0) {
7064                 ctl_set_success(ctsio);
7065                 ctl_done((union ctl_io *)ctsio);
7066                 return (CTL_RETVAL_COMPLETE);
7067         }
7068
7069         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
7070         ctsio->kern_rel_offset = 0;
7071         ctsio->kern_sg_entries = 0;
7072         ctsio->kern_data_len = min(data_len, alloc_len);
7073         ctsio->kern_total_len = ctsio->kern_data_len;
7074
7075         if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
7076                 data10 = (struct scsi_read_defect_data_hdr_10 *)
7077                     ctsio->kern_data_ptr;
7078                 data10->format = format;
7079                 scsi_ulto2b(0, data10->length);
7080         } else {
7081                 data12 = (struct scsi_read_defect_data_hdr_12 *)
7082                     ctsio->kern_data_ptr;
7083                 data12->format = format;
7084                 scsi_ulto2b(0, data12->generation);
7085                 scsi_ulto4b(0, data12->length);
7086         }
7087
7088         ctl_set_success(ctsio);
7089         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7090         ctsio->be_move_done = ctl_config_move_done;
7091         ctl_datamove((union ctl_io *)ctsio);
7092         return (CTL_RETVAL_COMPLETE);
7093 }
7094
7095 int
7096 ctl_report_tagret_port_groups(struct ctl_scsiio *ctsio)
7097 {
7098         struct ctl_softc *softc = CTL_SOFTC(ctsio);
7099         struct ctl_lun *lun = CTL_LUN(ctsio);
7100         struct scsi_maintenance_in *cdb;
7101         int retval;
7102         int alloc_len, ext, total_len = 0, g, pc, pg, ts, os;
7103         int num_ha_groups, num_target_ports, shared_group;
7104         struct ctl_port *port;
7105         struct scsi_target_group_data *rtg_ptr;
7106         struct scsi_target_group_data_extended *rtg_ext_ptr;
7107         struct scsi_target_port_group_descriptor *tpg_desc;
7108
7109         CTL_DEBUG_PRINT(("ctl_report_tagret_port_groups\n"));
7110
7111         cdb = (struct scsi_maintenance_in *)ctsio->cdb;
7112         retval = CTL_RETVAL_COMPLETE;
7113
7114         switch (cdb->byte2 & STG_PDF_MASK) {
7115         case STG_PDF_LENGTH:
7116                 ext = 0;
7117                 break;
7118         case STG_PDF_EXTENDED:
7119                 ext = 1;
7120                 break;
7121         default:
7122                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7123                                       /*sks_valid*/ 1,
7124                                       /*command*/ 1,
7125                                       /*field*/ 2,
7126                                       /*bit_valid*/ 1,
7127                                       /*bit*/ 5);
7128                 ctl_done((union ctl_io *)ctsio);
7129                 return(retval);
7130         }
7131
7132         num_target_ports = 0;
7133         shared_group = (softc->is_single != 0);
7134         mtx_lock(&softc->ctl_lock);
7135         STAILQ_FOREACH(port, &softc->port_list, links) {
7136                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7137                         continue;
7138                 if (ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
7139                         continue;
7140                 num_target_ports++;
7141                 if (port->status & CTL_PORT_STATUS_HA_SHARED)
7142                         shared_group = 1;
7143         }
7144         mtx_unlock(&softc->ctl_lock);
7145         num_ha_groups = (softc->is_single) ? 0 : NUM_HA_SHELVES;
7146
7147         if (ext)
7148                 total_len = sizeof(struct scsi_target_group_data_extended);
7149         else
7150                 total_len = sizeof(struct scsi_target_group_data);
7151         total_len += sizeof(struct scsi_target_port_group_descriptor) *
7152                 (shared_group + num_ha_groups) +
7153             sizeof(struct scsi_target_port_descriptor) * num_target_ports;
7154
7155         alloc_len = scsi_4btoul(cdb->length);
7156
7157         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7158         ctsio->kern_sg_entries = 0;
7159         ctsio->kern_rel_offset = 0;
7160         ctsio->kern_data_len = min(total_len, alloc_len);
7161         ctsio->kern_total_len = ctsio->kern_data_len;
7162
7163         if (ext) {
7164                 rtg_ext_ptr = (struct scsi_target_group_data_extended *)
7165                     ctsio->kern_data_ptr;
7166                 scsi_ulto4b(total_len - 4, rtg_ext_ptr->length);
7167                 rtg_ext_ptr->format_type = 0x10;
7168                 rtg_ext_ptr->implicit_transition_time = 0;
7169                 tpg_desc = &rtg_ext_ptr->groups[0];
7170         } else {
7171                 rtg_ptr = (struct scsi_target_group_data *)
7172                     ctsio->kern_data_ptr;
7173                 scsi_ulto4b(total_len - 4, rtg_ptr->length);
7174                 tpg_desc = &rtg_ptr->groups[0];
7175         }
7176
7177         mtx_lock(&softc->ctl_lock);
7178         pg = softc->port_min / softc->port_cnt;
7179         if (lun->flags & (CTL_LUN_PRIMARY_SC | CTL_LUN_PEER_SC_PRIMARY)) {
7180                 /* Some shelf is known to be primary. */
7181                 if (softc->ha_link == CTL_HA_LINK_OFFLINE)
7182                         os = TPG_ASYMMETRIC_ACCESS_UNAVAILABLE;
7183                 else if (softc->ha_link == CTL_HA_LINK_UNKNOWN)
7184                         os = TPG_ASYMMETRIC_ACCESS_TRANSITIONING;
7185                 else if (softc->ha_mode == CTL_HA_MODE_ACT_STBY)
7186                         os = TPG_ASYMMETRIC_ACCESS_STANDBY;
7187                 else
7188                         os = TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7189                 if (lun->flags & CTL_LUN_PRIMARY_SC) {
7190                         ts = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7191                 } else {
7192                         ts = os;
7193                         os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7194                 }
7195         } else {
7196                 /* No known primary shelf. */
7197                 if (softc->ha_link == CTL_HA_LINK_OFFLINE) {
7198                         ts = TPG_ASYMMETRIC_ACCESS_UNAVAILABLE;
7199                         os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7200                 } else if (softc->ha_link == CTL_HA_LINK_UNKNOWN) {
7201                         ts = TPG_ASYMMETRIC_ACCESS_TRANSITIONING;
7202                         os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7203                 } else {
7204                         ts = os = TPG_ASYMMETRIC_ACCESS_TRANSITIONING;
7205                 }
7206         }
7207         if (shared_group) {
7208                 tpg_desc->pref_state = ts;
7209                 tpg_desc->support = TPG_AO_SUP | TPG_AN_SUP | TPG_S_SUP |
7210                     TPG_U_SUP | TPG_T_SUP;
7211                 scsi_ulto2b(1, tpg_desc->target_port_group);
7212                 tpg_desc->status = TPG_IMPLICIT;
7213                 pc = 0;
7214                 STAILQ_FOREACH(port, &softc->port_list, links) {
7215                         if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7216                                 continue;
7217                         if (!softc->is_single &&
7218                             (port->status & CTL_PORT_STATUS_HA_SHARED) == 0)
7219                                 continue;
7220                         if (ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
7221                                 continue;
7222                         scsi_ulto2b(port->targ_port, tpg_desc->descriptors[pc].
7223                             relative_target_port_identifier);
7224                         pc++;
7225                 }
7226                 tpg_desc->target_port_count = pc;
7227                 tpg_desc = (struct scsi_target_port_group_descriptor *)
7228                     &tpg_desc->descriptors[pc];
7229         }
7230         for (g = 0; g < num_ha_groups; g++) {
7231                 tpg_desc->pref_state = (g == pg) ? ts : os;
7232                 tpg_desc->support = TPG_AO_SUP | TPG_AN_SUP | TPG_S_SUP |
7233                     TPG_U_SUP | TPG_T_SUP;
7234                 scsi_ulto2b(2 + g, tpg_desc->target_port_group);
7235                 tpg_desc->status = TPG_IMPLICIT;
7236                 pc = 0;
7237                 STAILQ_FOREACH(port, &softc->port_list, links) {
7238                         if (port->targ_port < g * softc->port_cnt ||
7239                             port->targ_port >= (g + 1) * softc->port_cnt)
7240                                 continue;
7241                         if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7242                                 continue;
7243                         if (port->status & CTL_PORT_STATUS_HA_SHARED)
7244                                 continue;
7245                         if (ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
7246                                 continue;
7247                         scsi_ulto2b(port->targ_port, tpg_desc->descriptors[pc].
7248                             relative_target_port_identifier);
7249                         pc++;
7250                 }
7251                 tpg_desc->target_port_count = pc;
7252                 tpg_desc = (struct scsi_target_port_group_descriptor *)
7253                     &tpg_desc->descriptors[pc];
7254         }
7255         mtx_unlock(&softc->ctl_lock);
7256
7257         ctl_set_success(ctsio);
7258         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7259         ctsio->be_move_done = ctl_config_move_done;
7260         ctl_datamove((union ctl_io *)ctsio);
7261         return(retval);
7262 }
7263
7264 int
7265 ctl_report_supported_opcodes(struct ctl_scsiio *ctsio)
7266 {
7267         struct ctl_lun *lun = CTL_LUN(ctsio);
7268         struct scsi_report_supported_opcodes *cdb;
7269         const struct ctl_cmd_entry *entry, *sentry;
7270         struct scsi_report_supported_opcodes_all *all;
7271         struct scsi_report_supported_opcodes_descr *descr;
7272         struct scsi_report_supported_opcodes_one *one;
7273         int retval;
7274         int alloc_len, total_len;
7275         int opcode, service_action, i, j, num;
7276
7277         CTL_DEBUG_PRINT(("ctl_report_supported_opcodes\n"));
7278
7279         cdb = (struct scsi_report_supported_opcodes *)ctsio->cdb;
7280         retval = CTL_RETVAL_COMPLETE;
7281
7282         opcode = cdb->requested_opcode;
7283         service_action = scsi_2btoul(cdb->requested_service_action);
7284         switch (cdb->options & RSO_OPTIONS_MASK) {
7285         case RSO_OPTIONS_ALL:
7286                 num = 0;
7287                 for (i = 0; i < 256; i++) {
7288                         entry = &ctl_cmd_table[i];
7289                         if (entry->flags & CTL_CMD_FLAG_SA5) {
7290                                 for (j = 0; j < 32; j++) {
7291                                         sentry = &((const struct ctl_cmd_entry *)
7292                                             entry->execute)[j];
7293                                         if (ctl_cmd_applicable(
7294                                             lun->be_lun->lun_type, sentry))
7295                                                 num++;
7296                                 }
7297                         } else {
7298                                 if (ctl_cmd_applicable(lun->be_lun->lun_type,
7299                                     entry))
7300                                         num++;
7301                         }
7302                 }
7303                 total_len = sizeof(struct scsi_report_supported_opcodes_all) +
7304                     num * sizeof(struct scsi_report_supported_opcodes_descr);
7305                 break;
7306         case RSO_OPTIONS_OC:
7307                 if (ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) {
7308                         ctl_set_invalid_field(/*ctsio*/ ctsio,
7309                                               /*sks_valid*/ 1,
7310                                               /*command*/ 1,
7311                                               /*field*/ 2,
7312                                               /*bit_valid*/ 1,
7313                                               /*bit*/ 2);
7314                         ctl_done((union ctl_io *)ctsio);
7315                         return (CTL_RETVAL_COMPLETE);
7316                 }
7317                 total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7318                 break;
7319         case RSO_OPTIONS_OC_SA:
7320                 if ((ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) == 0 ||
7321                     service_action >= 32) {
7322                         ctl_set_invalid_field(/*ctsio*/ ctsio,
7323                                               /*sks_valid*/ 1,
7324                                               /*command*/ 1,
7325                                               /*field*/ 2,
7326                                               /*bit_valid*/ 1,
7327                                               /*bit*/ 2);
7328                         ctl_done((union ctl_io *)ctsio);
7329                         return (CTL_RETVAL_COMPLETE);
7330                 }
7331                 /* FALLTHROUGH */
7332         case RSO_OPTIONS_OC_ASA:
7333                 total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7334                 break;
7335         default:
7336                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7337                                       /*sks_valid*/ 1,
7338                                       /*command*/ 1,
7339                                       /*field*/ 2,
7340                                       /*bit_valid*/ 1,
7341                                       /*bit*/ 2);
7342                 ctl_done((union ctl_io *)ctsio);
7343                 return (CTL_RETVAL_COMPLETE);
7344         }
7345
7346         alloc_len = scsi_4btoul(cdb->length);
7347
7348         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7349         ctsio->kern_sg_entries = 0;
7350         ctsio->kern_rel_offset = 0;
7351         ctsio->kern_data_len = min(total_len, alloc_len);
7352         ctsio->kern_total_len = ctsio->kern_data_len;
7353
7354         switch (cdb->options & RSO_OPTIONS_MASK) {
7355         case RSO_OPTIONS_ALL:
7356                 all = (struct scsi_report_supported_opcodes_all *)
7357                     ctsio->kern_data_ptr;
7358                 num = 0;
7359                 for (i = 0; i < 256; i++) {
7360                         entry = &ctl_cmd_table[i];
7361                         if (entry->flags & CTL_CMD_FLAG_SA5) {
7362                                 for (j = 0; j < 32; j++) {
7363                                         sentry = &((const struct ctl_cmd_entry *)
7364                                             entry->execute)[j];
7365                                         if (!ctl_cmd_applicable(
7366                                             lun->be_lun->lun_type, sentry))
7367                                                 continue;
7368                                         descr = &all->descr[num++];
7369                                         descr->opcode = i;
7370                                         scsi_ulto2b(j, descr->service_action);
7371                                         descr->flags = RSO_SERVACTV;
7372                                         scsi_ulto2b(sentry->length,
7373                                             descr->cdb_length);
7374                                 }
7375                         } else {
7376                                 if (!ctl_cmd_applicable(lun->be_lun->lun_type,
7377                                     entry))
7378                                         continue;
7379                                 descr = &all->descr[num++];
7380                                 descr->opcode = i;
7381                                 scsi_ulto2b(0, descr->service_action);
7382                                 descr->flags = 0;
7383                                 scsi_ulto2b(entry->length, descr->cdb_length);
7384                         }
7385                 }
7386                 scsi_ulto4b(
7387                     num * sizeof(struct scsi_report_supported_opcodes_descr),
7388                     all->length);
7389                 break;
7390         case RSO_OPTIONS_OC:
7391                 one = (struct scsi_report_supported_opcodes_one *)
7392                     ctsio->kern_data_ptr;
7393                 entry = &ctl_cmd_table[opcode];
7394                 goto fill_one;
7395         case RSO_OPTIONS_OC_SA:
7396                 one = (struct scsi_report_supported_opcodes_one *)
7397                     ctsio->kern_data_ptr;
7398                 entry = &ctl_cmd_table[opcode];
7399                 entry = &((const struct ctl_cmd_entry *)
7400                     entry->execute)[service_action];
7401 fill_one:
7402                 if (ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
7403                         one->support = 3;
7404                         scsi_ulto2b(entry->length, one->cdb_length);
7405                         one->cdb_usage[0] = opcode;
7406                         memcpy(&one->cdb_usage[1], entry->usage,
7407                             entry->length - 1);
7408                 } else
7409                         one->support = 1;
7410                 break;
7411         case RSO_OPTIONS_OC_ASA:
7412                 one = (struct scsi_report_supported_opcodes_one *)
7413                     ctsio->kern_data_ptr;
7414                 entry = &ctl_cmd_table[opcode];
7415                 if (entry->flags & CTL_CMD_FLAG_SA5) {
7416                         entry = &((const struct ctl_cmd_entry *)
7417                             entry->execute)[service_action];
7418                 } else if (service_action != 0) {
7419                         one->support = 1;
7420                         break;
7421                 }
7422                 goto fill_one;
7423         }
7424
7425         ctl_set_success(ctsio);
7426         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7427         ctsio->be_move_done = ctl_config_move_done;
7428         ctl_datamove((union ctl_io *)ctsio);
7429         return(retval);
7430 }
7431
7432 int
7433 ctl_report_supported_tmf(struct ctl_scsiio *ctsio)
7434 {
7435         struct scsi_report_supported_tmf *cdb;
7436         struct scsi_report_supported_tmf_ext_data *data;
7437         int retval;
7438         int alloc_len, total_len;
7439
7440         CTL_DEBUG_PRINT(("ctl_report_supported_tmf\n"));
7441
7442         cdb = (struct scsi_report_supported_tmf *)ctsio->cdb;
7443
7444         retval = CTL_RETVAL_COMPLETE;
7445
7446         if (cdb->options & RST_REPD)
7447                 total_len = sizeof(struct scsi_report_supported_tmf_ext_data);
7448         else
7449                 total_len = sizeof(struct scsi_report_supported_tmf_data);
7450         alloc_len = scsi_4btoul(cdb->length);
7451
7452         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7453         ctsio->kern_sg_entries = 0;
7454         ctsio->kern_rel_offset = 0;
7455         ctsio->kern_data_len = min(total_len, alloc_len);
7456         ctsio->kern_total_len = ctsio->kern_data_len;
7457
7458         data = (struct scsi_report_supported_tmf_ext_data *)ctsio->kern_data_ptr;
7459         data->byte1 |= RST_ATS | RST_ATSS | RST_CTSS | RST_LURS | RST_QTS |
7460             RST_TRS;
7461         data->byte2 |= RST_QAES | RST_QTSS | RST_ITNRS;
7462         data->length = total_len - 4;
7463
7464         ctl_set_success(ctsio);
7465         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7466         ctsio->be_move_done = ctl_config_move_done;
7467         ctl_datamove((union ctl_io *)ctsio);
7468         return (retval);
7469 }
7470
7471 int
7472 ctl_report_timestamp(struct ctl_scsiio *ctsio)
7473 {
7474         struct scsi_report_timestamp *cdb;
7475         struct scsi_report_timestamp_data *data;
7476         struct timeval tv;
7477         int64_t timestamp;
7478         int retval;
7479         int alloc_len, total_len;
7480
7481         CTL_DEBUG_PRINT(("ctl_report_timestamp\n"));
7482
7483         cdb = (struct scsi_report_timestamp *)ctsio->cdb;
7484
7485         retval = CTL_RETVAL_COMPLETE;
7486
7487         total_len = sizeof(struct scsi_report_timestamp_data);
7488         alloc_len = scsi_4btoul(cdb->length);
7489
7490         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7491         ctsio->kern_sg_entries = 0;
7492         ctsio->kern_rel_offset = 0;
7493         ctsio->kern_data_len = min(total_len, alloc_len);
7494         ctsio->kern_total_len = ctsio->kern_data_len;
7495
7496         data = (struct scsi_report_timestamp_data *)ctsio->kern_data_ptr;
7497         scsi_ulto2b(sizeof(*data) - 2, data->length);
7498         data->origin = RTS_ORIG_OUTSIDE;
7499         getmicrotime(&tv);
7500         timestamp = (int64_t)tv.tv_sec * 1000 + tv.tv_usec / 1000;
7501         scsi_ulto4b(timestamp >> 16, data->timestamp);
7502         scsi_ulto2b(timestamp & 0xffff, &data->timestamp[4]);
7503
7504         ctl_set_success(ctsio);
7505         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7506         ctsio->be_move_done = ctl_config_move_done;
7507         ctl_datamove((union ctl_io *)ctsio);
7508         return (retval);
7509 }
7510
7511 int
7512 ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7513 {
7514         struct ctl_softc *softc = CTL_SOFTC(ctsio);
7515         struct ctl_lun *lun = CTL_LUN(ctsio);
7516         struct scsi_per_res_in *cdb;
7517         int alloc_len, total_len = 0;
7518         /* struct scsi_per_res_in_rsrv in_data; */
7519         uint64_t key;
7520
7521         CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7522
7523         cdb = (struct scsi_per_res_in *)ctsio->cdb;
7524
7525         alloc_len = scsi_2btoul(cdb->length);
7526
7527 retry:
7528         mtx_lock(&lun->lun_lock);
7529         switch (cdb->action) {
7530         case SPRI_RK: /* read keys */
7531                 total_len = sizeof(struct scsi_per_res_in_keys) +
7532                         lun->pr_key_count *
7533                         sizeof(struct scsi_per_res_key);
7534                 break;
7535         case SPRI_RR: /* read reservation */
7536                 if (lun->flags & CTL_LUN_PR_RESERVED)
7537                         total_len = sizeof(struct scsi_per_res_in_rsrv);
7538                 else
7539                         total_len = sizeof(struct scsi_per_res_in_header);
7540                 break;
7541         case SPRI_RC: /* report capabilities */
7542                 total_len = sizeof(struct scsi_per_res_cap);
7543                 break;
7544         case SPRI_RS: /* read full status */
7545                 total_len = sizeof(struct scsi_per_res_in_header) +
7546                     (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7547                     lun->pr_key_count;
7548                 break;
7549         default:
7550                 panic("%s: Invalid PR type %#x", __func__, cdb->action);
7551         }
7552         mtx_unlock(&lun->lun_lock);
7553
7554         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7555         ctsio->kern_rel_offset = 0;
7556         ctsio->kern_sg_entries = 0;
7557         ctsio->kern_data_len = min(total_len, alloc_len);
7558         ctsio->kern_total_len = ctsio->kern_data_len;
7559
7560         mtx_lock(&lun->lun_lock);
7561         switch (cdb->action) {
7562         case SPRI_RK: { // read keys
7563         struct scsi_per_res_in_keys *res_keys;
7564                 int i, key_count;
7565
7566                 res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7567
7568                 /*
7569                  * We had to drop the lock to allocate our buffer, which
7570                  * leaves time for someone to come in with another
7571                  * persistent reservation.  (That is unlikely, though,
7572                  * since this should be the only persistent reservation
7573                  * command active right now.)
7574                  */
7575                 if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7576                     (lun->pr_key_count *
7577                      sizeof(struct scsi_per_res_key)))){
7578                         mtx_unlock(&lun->lun_lock);
7579                         free(ctsio->kern_data_ptr, M_CTL);
7580                         printf("%s: reservation length changed, retrying\n",
7581                                __func__);
7582                         goto retry;
7583                 }
7584
7585                 scsi_ulto4b(lun->pr_generation, res_keys->header.generation);
7586
7587                 scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7588                              lun->pr_key_count, res_keys->header.length);
7589
7590                 for (i = 0, key_count = 0; i < CTL_MAX_INITIATORS; i++) {
7591                         if ((key = ctl_get_prkey(lun, i)) == 0)
7592                                 continue;
7593
7594                         /*
7595                          * We used lun->pr_key_count to calculate the
7596                          * size to allocate.  If it turns out the number of
7597                          * initiators with the registered flag set is
7598                          * larger than that (i.e. they haven't been kept in
7599                          * sync), we've got a problem.
7600                          */
7601                         if (key_count >= lun->pr_key_count) {
7602                                 key_count++;
7603                                 continue;
7604                         }
7605                         scsi_u64to8b(key, res_keys->keys[key_count].key);
7606                         key_count++;
7607                 }
7608                 break;
7609         }
7610         case SPRI_RR: { // read reservation
7611                 struct scsi_per_res_in_rsrv *res;
7612                 int tmp_len, header_only;
7613
7614                 res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7615
7616                 scsi_ulto4b(lun->pr_generation, res->header.generation);
7617
7618                 if (lun->flags & CTL_LUN_PR_RESERVED)
7619                 {
7620                         tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7621                         scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7622                                     res->header.length);
7623                         header_only = 0;
7624                 } else {
7625                         tmp_len = sizeof(struct scsi_per_res_in_header);
7626                         scsi_ulto4b(0, res->header.length);
7627                         header_only = 1;
7628                 }
7629
7630                 /*
7631                  * We had to drop the lock to allocate our buffer, which
7632                  * leaves time for someone to come in with another
7633                  * persistent reservation.  (That is unlikely, though,
7634                  * since this should be the only persistent reservation
7635                  * command active right now.)
7636                  */
7637                 if (tmp_len != total_len) {
7638                         mtx_unlock(&lun->lun_lock);
7639                         free(ctsio->kern_data_ptr, M_CTL);
7640                         printf("%s: reservation status changed, retrying\n",
7641                                __func__);
7642                         goto retry;
7643                 }
7644
7645                 /*
7646                  * No reservation held, so we're done.
7647                  */
7648                 if (header_only != 0)
7649                         break;
7650
7651                 /*
7652                  * If the registration is an All Registrants type, the key
7653                  * is 0, since it doesn't really matter.
7654                  */
7655                 if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7656                         scsi_u64to8b(ctl_get_prkey(lun, lun->pr_res_idx),
7657                             res->data.reservation);
7658                 }
7659                 res->data.scopetype = lun->pr_res_type;
7660                 break;
7661         }
7662         case SPRI_RC:     //report capabilities
7663         {
7664                 struct scsi_per_res_cap *res_cap;
7665                 uint16_t type_mask;
7666
7667                 res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7668                 scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7669                 res_cap->flags1 = SPRI_CRH;
7670                 res_cap->flags2 = SPRI_TMV | SPRI_ALLOW_5;
7671                 type_mask = SPRI_TM_WR_EX_AR |
7672                             SPRI_TM_EX_AC_RO |
7673                             SPRI_TM_WR_EX_RO |
7674                             SPRI_TM_EX_AC |
7675                             SPRI_TM_WR_EX |
7676                             SPRI_TM_EX_AC_AR;
7677                 scsi_ulto2b(type_mask, res_cap->type_mask);
7678                 break;
7679         }
7680         case SPRI_RS: { // read full status
7681                 struct scsi_per_res_in_full *res_status;
7682                 struct scsi_per_res_in_full_desc *res_desc;
7683                 struct ctl_port *port;
7684                 int i, len;
7685
7686                 res_status = (struct scsi_per_res_in_full*)ctsio->kern_data_ptr;
7687
7688                 /*
7689                  * We had to drop the lock to allocate our buffer, which
7690                  * leaves time for someone to come in with another
7691                  * persistent reservation.  (That is unlikely, though,
7692                  * since this should be the only persistent reservation
7693                  * command active right now.)
7694                  */
7695                 if (total_len < (sizeof(struct scsi_per_res_in_header) +
7696                     (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7697                      lun->pr_key_count)){
7698                         mtx_unlock(&lun->lun_lock);
7699                         free(ctsio->kern_data_ptr, M_CTL);
7700                         printf("%s: reservation length changed, retrying\n",
7701                                __func__);
7702                         goto retry;
7703                 }
7704
7705                 scsi_ulto4b(lun->pr_generation, res_status->header.generation);
7706
7707                 res_desc = &res_status->desc[0];
7708                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
7709                         if ((key = ctl_get_prkey(lun, i)) == 0)
7710                                 continue;
7711
7712                         scsi_u64to8b(key, res_desc->res_key.key);
7713                         if ((lun->flags & CTL_LUN_PR_RESERVED) &&
7714                             (lun->pr_res_idx == i ||
7715                              lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS)) {
7716                                 res_desc->flags = SPRI_FULL_R_HOLDER;
7717                                 res_desc->scopetype = lun->pr_res_type;
7718                         }
7719                         scsi_ulto2b(i / CTL_MAX_INIT_PER_PORT,
7720                             res_desc->rel_trgt_port_id);
7721                         len = 0;
7722                         port = softc->ctl_ports[i / CTL_MAX_INIT_PER_PORT];
7723                         if (port != NULL)
7724                                 len = ctl_create_iid(port,
7725                                     i % CTL_MAX_INIT_PER_PORT,
7726                                     res_desc->transport_id);
7727                         scsi_ulto4b(len, res_desc->additional_length);
7728                         res_desc = (struct scsi_per_res_in_full_desc *)
7729                             &res_desc->transport_id[len];
7730                 }
7731                 scsi_ulto4b((uint8_t *)res_desc - (uint8_t *)&res_status->desc[0],
7732                     res_status->header.length);
7733                 break;
7734         }
7735         default:
7736                 panic("%s: Invalid PR type %#x", __func__, cdb->action);
7737         }
7738         mtx_unlock(&lun->lun_lock);
7739
7740         ctl_set_success(ctsio);
7741         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7742         ctsio->be_move_done = ctl_config_move_done;
7743         ctl_datamove((union ctl_io *)ctsio);
7744         return (CTL_RETVAL_COMPLETE);
7745 }
7746
7747 /*
7748  * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
7749  * it should return.
7750  */
7751 static int
7752 ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
7753                 uint64_t sa_res_key, uint8_t type, uint32_t residx,
7754                 struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
7755                 struct scsi_per_res_out_parms* param)
7756 {
7757         union ctl_ha_msg persis_io;
7758         int i;
7759
7760         mtx_lock(&lun->lun_lock);
7761         if (sa_res_key == 0) {
7762                 if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
7763                         /* validate scope and type */
7764                         if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7765                              SPR_LU_SCOPE) {
7766                                 mtx_unlock(&lun->lun_lock);
7767                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7768                                                       /*sks_valid*/ 1,
7769                                                       /*command*/ 1,
7770                                                       /*field*/ 2,
7771                                                       /*bit_valid*/ 1,
7772                                                       /*bit*/ 4);
7773                                 ctl_done((union ctl_io *)ctsio);
7774                                 return (1);
7775                         }
7776
7777                         if (type>8 || type==2 || type==4 || type==0) {
7778                                 mtx_unlock(&lun->lun_lock);
7779                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7780                                                       /*sks_valid*/ 1,
7781                                                       /*command*/ 1,
7782                                                       /*field*/ 2,
7783                                                       /*bit_valid*/ 1,
7784                                                       /*bit*/ 0);
7785                                 ctl_done((union ctl_io *)ctsio);
7786                                 return (1);
7787                         }
7788
7789                         /*
7790                          * Unregister everybody else and build UA for
7791                          * them
7792                          */
7793                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
7794                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
7795                                         continue;
7796
7797                                 ctl_clr_prkey(lun, i);
7798                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7799                         }
7800                         lun->pr_key_count = 1;
7801                         lun->pr_res_type = type;
7802                         if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
7803                             lun->pr_res_type != SPR_TYPE_EX_AC_AR)
7804                                 lun->pr_res_idx = residx;
7805                         lun->pr_generation++;
7806                         mtx_unlock(&lun->lun_lock);
7807
7808                         /* send msg to other side */
7809                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7810                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7811                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7812                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
7813                         persis_io.pr.pr_info.res_type = type;
7814                         memcpy(persis_io.pr.pr_info.sa_res_key,
7815                                param->serv_act_res_key,
7816                                sizeof(param->serv_act_res_key));
7817                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
7818                             sizeof(persis_io.pr), M_WAITOK);
7819                 } else {
7820                         /* not all registrants */
7821                         mtx_unlock(&lun->lun_lock);
7822                         free(ctsio->kern_data_ptr, M_CTL);
7823                         ctl_set_invalid_field(ctsio,
7824                                               /*sks_valid*/ 1,
7825                                               /*command*/ 0,
7826                                               /*field*/ 8,
7827                                               /*bit_valid*/ 0,
7828                                               /*bit*/ 0);
7829                         ctl_done((union ctl_io *)ctsio);
7830                         return (1);
7831                 }
7832         } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
7833                 || !(lun->flags & CTL_LUN_PR_RESERVED)) {
7834                 int found = 0;
7835
7836                 if (res_key == sa_res_key) {
7837                         /* special case */
7838                         /*
7839                          * The spec implies this is not good but doesn't
7840                          * say what to do. There are two choices either
7841                          * generate a res conflict or check condition
7842                          * with illegal field in parameter data. Since
7843                          * that is what is done when the sa_res_key is
7844                          * zero I'll take that approach since this has
7845                          * to do with the sa_res_key.
7846                          */
7847                         mtx_unlock(&lun->lun_lock);
7848                         free(ctsio->kern_data_ptr, M_CTL);
7849                         ctl_set_invalid_field(ctsio,
7850                                               /*sks_valid*/ 1,
7851                                               /*command*/ 0,
7852                                               /*field*/ 8,
7853                                               /*bit_valid*/ 0,
7854                                               /*bit*/ 0);
7855                         ctl_done((union ctl_io *)ctsio);
7856                         return (1);
7857                 }
7858
7859                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
7860                         if (ctl_get_prkey(lun, i) != sa_res_key)
7861                                 continue;
7862
7863                         found = 1;
7864                         ctl_clr_prkey(lun, i);
7865                         lun->pr_key_count--;
7866                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7867                 }
7868                 if (!found) {
7869                         mtx_unlock(&lun->lun_lock);
7870                         free(ctsio->kern_data_ptr, M_CTL);
7871                         ctl_set_reservation_conflict(ctsio);
7872                         ctl_done((union ctl_io *)ctsio);
7873                         return (CTL_RETVAL_COMPLETE);
7874                 }
7875                 lun->pr_generation++;
7876                 mtx_unlock(&lun->lun_lock);
7877
7878                 /* send msg to other side */
7879                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7880                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7881                 persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7882                 persis_io.pr.pr_info.residx = lun->pr_res_idx;
7883                 persis_io.pr.pr_info.res_type = type;
7884                 memcpy(persis_io.pr.pr_info.sa_res_key,
7885                        param->serv_act_res_key,
7886                        sizeof(param->serv_act_res_key));
7887                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
7888                     sizeof(persis_io.pr), M_WAITOK);
7889         } else {
7890                 /* Reserved but not all registrants */
7891                 /* sa_res_key is res holder */
7892                 if (sa_res_key == ctl_get_prkey(lun, lun->pr_res_idx)) {
7893                         /* validate scope and type */
7894                         if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7895                              SPR_LU_SCOPE) {
7896                                 mtx_unlock(&lun->lun_lock);
7897                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7898                                                       /*sks_valid*/ 1,
7899                                                       /*command*/ 1,
7900                                                       /*field*/ 2,
7901                                                       /*bit_valid*/ 1,
7902                                                       /*bit*/ 4);
7903                                 ctl_done((union ctl_io *)ctsio);
7904                                 return (1);
7905                         }
7906
7907                         if (type>8 || type==2 || type==4 || type==0) {
7908                                 mtx_unlock(&lun->lun_lock);
7909                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7910                                                       /*sks_valid*/ 1,
7911                                                       /*command*/ 1,
7912                                                       /*field*/ 2,
7913                                                       /*bit_valid*/ 1,
7914                                                       /*bit*/ 0);
7915                                 ctl_done((union ctl_io *)ctsio);
7916                                 return (1);
7917                         }
7918
7919                         /*
7920                          * Do the following:
7921                          * if sa_res_key != res_key remove all
7922                          * registrants w/sa_res_key and generate UA
7923                          * for these registrants(Registrations
7924                          * Preempted) if it wasn't an exclusive
7925                          * reservation generate UA(Reservations
7926                          * Preempted) for all other registered nexuses
7927                          * if the type has changed. Establish the new
7928                          * reservation and holder. If res_key and
7929                          * sa_res_key are the same do the above
7930                          * except don't unregister the res holder.
7931                          */
7932
7933                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
7934                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
7935                                         continue;
7936
7937                                 if (sa_res_key == ctl_get_prkey(lun, i)) {
7938                                         ctl_clr_prkey(lun, i);
7939                                         lun->pr_key_count--;
7940                                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7941                                 } else if (type != lun->pr_res_type &&
7942                                     (lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
7943                                      lun->pr_res_type == SPR_TYPE_EX_AC_RO)) {
7944                                         ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
7945                                 }
7946                         }
7947                         lun->pr_res_type = type;
7948                         if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
7949                             lun->pr_res_type != SPR_TYPE_EX_AC_AR)
7950                                 lun->pr_res_idx = residx;
7951                         else
7952                                 lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
7953                         lun->pr_generation++;
7954                         mtx_unlock(&lun->lun_lock);
7955
7956                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7957                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7958                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7959                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
7960                         persis_io.pr.pr_info.res_type = type;
7961                         memcpy(persis_io.pr.pr_info.sa_res_key,
7962                                param->serv_act_res_key,
7963                                sizeof(param->serv_act_res_key));
7964                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
7965                             sizeof(persis_io.pr), M_WAITOK);
7966                 } else {
7967                         /*
7968                          * sa_res_key is not the res holder just
7969                          * remove registrants
7970                          */
7971                         int found=0;
7972
7973                         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
7974                                 if (sa_res_key != ctl_get_prkey(lun, i))
7975                                         continue;
7976
7977                                 found = 1;
7978                                 ctl_clr_prkey(lun, i);
7979                                 lun->pr_key_count--;
7980                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7981                         }
7982
7983                         if (!found) {
7984                                 mtx_unlock(&lun->lun_lock);
7985                                 free(ctsio->kern_data_ptr, M_CTL);
7986                                 ctl_set_reservation_conflict(ctsio);
7987                                 ctl_done((union ctl_io *)ctsio);
7988                                 return (1);
7989                         }
7990                         lun->pr_generation++;
7991                         mtx_unlock(&lun->lun_lock);
7992
7993                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7994                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7995                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7996                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
7997                         persis_io.pr.pr_info.res_type = type;
7998                         memcpy(persis_io.pr.pr_info.sa_res_key,
7999                                param->serv_act_res_key,
8000                                sizeof(param->serv_act_res_key));
8001                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8002                             sizeof(persis_io.pr), M_WAITOK);
8003                 }
8004         }
8005         return (0);
8006 }
8007
8008 static void
8009 ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
8010 {
8011         uint64_t sa_res_key;
8012         int i;
8013
8014         sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
8015
8016         if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8017          || lun->pr_res_idx == CTL_PR_NO_RESERVATION
8018          || sa_res_key != ctl_get_prkey(lun, lun->pr_res_idx)) {
8019                 if (sa_res_key == 0) {
8020                         /*
8021                          * Unregister everybody else and build UA for
8022                          * them
8023                          */
8024                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
8025                                 if (i == msg->pr.pr_info.residx ||
8026                                     ctl_get_prkey(lun, i) == 0)
8027                                         continue;
8028
8029                                 ctl_clr_prkey(lun, i);
8030                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8031                         }
8032
8033                         lun->pr_key_count = 1;
8034                         lun->pr_res_type = msg->pr.pr_info.res_type;
8035                         if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
8036                             lun->pr_res_type != SPR_TYPE_EX_AC_AR)
8037                                 lun->pr_res_idx = msg->pr.pr_info.residx;
8038                 } else {
8039                         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8040                                 if (sa_res_key == ctl_get_prkey(lun, i))
8041                                         continue;
8042
8043                                 ctl_clr_prkey(lun, i);
8044                                 lun->pr_key_count--;
8045                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8046                         }
8047                 }
8048         } else {
8049                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8050                         if (i == msg->pr.pr_info.residx ||
8051                             ctl_get_prkey(lun, i) == 0)
8052                                 continue;
8053
8054                         if (sa_res_key == ctl_get_prkey(lun, i)) {
8055                                 ctl_clr_prkey(lun, i);
8056                                 lun->pr_key_count--;
8057                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8058                         } else if (msg->pr.pr_info.res_type != lun->pr_res_type
8059                             && (lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8060                              lun->pr_res_type == SPR_TYPE_EX_AC_RO)) {
8061                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8062                         }
8063                 }
8064                 lun->pr_res_type = msg->pr.pr_info.res_type;
8065                 if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
8066                     lun->pr_res_type != SPR_TYPE_EX_AC_AR)
8067                         lun->pr_res_idx = msg->pr.pr_info.residx;
8068                 else
8069                         lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8070         }
8071         lun->pr_generation++;
8072
8073 }
8074
8075
8076 int
8077 ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
8078 {
8079         struct ctl_softc *softc = CTL_SOFTC(ctsio);
8080         struct ctl_lun *lun = CTL_LUN(ctsio);
8081         int retval;
8082         u_int32_t param_len;
8083         struct scsi_per_res_out *cdb;
8084         struct scsi_per_res_out_parms* param;
8085         uint32_t residx;
8086         uint64_t res_key, sa_res_key, key;
8087         uint8_t type;
8088         union ctl_ha_msg persis_io;
8089         int    i;
8090
8091         CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n"));
8092
8093         cdb = (struct scsi_per_res_out *)ctsio->cdb;
8094         retval = CTL_RETVAL_COMPLETE;
8095
8096         /*
8097          * We only support whole-LUN scope.  The scope & type are ignored for
8098          * register, register and ignore existing key and clear.
8099          * We sometimes ignore scope and type on preempts too!!
8100          * Verify reservation type here as well.
8101          */
8102         type = cdb->scope_type & SPR_TYPE_MASK;
8103         if ((cdb->action == SPRO_RESERVE)
8104          || (cdb->action == SPRO_RELEASE)) {
8105                 if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
8106                         ctl_set_invalid_field(/*ctsio*/ ctsio,
8107                                               /*sks_valid*/ 1,
8108                                               /*command*/ 1,
8109                                               /*field*/ 2,
8110                                               /*bit_valid*/ 1,
8111                                               /*bit*/ 4);
8112                         ctl_done((union ctl_io *)ctsio);
8113                         return (CTL_RETVAL_COMPLETE);
8114                 }
8115
8116                 if (type>8 || type==2 || type==4 || type==0) {
8117                         ctl_set_invalid_field(/*ctsio*/ ctsio,
8118                                               /*sks_valid*/ 1,
8119                                               /*command*/ 1,
8120                                               /*field*/ 2,
8121                                               /*bit_valid*/ 1,
8122                                               /*bit*/ 0);
8123                         ctl_done((union ctl_io *)ctsio);
8124                         return (CTL_RETVAL_COMPLETE);
8125                 }
8126         }
8127
8128         param_len = scsi_4btoul(cdb->length);
8129
8130         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
8131                 ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
8132                 ctsio->kern_data_len = param_len;
8133                 ctsio->kern_total_len = param_len;
8134                 ctsio->kern_rel_offset = 0;
8135                 ctsio->kern_sg_entries = 0;
8136                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
8137                 ctsio->be_move_done = ctl_config_move_done;
8138                 ctl_datamove((union ctl_io *)ctsio);
8139
8140                 return (CTL_RETVAL_COMPLETE);
8141         }
8142
8143         param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
8144
8145         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
8146         res_key = scsi_8btou64(param->res_key.key);
8147         sa_res_key = scsi_8btou64(param->serv_act_res_key);
8148
8149         /*
8150          * Validate the reservation key here except for SPRO_REG_IGNO
8151          * This must be done for all other service actions
8152          */
8153         if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
8154                 mtx_lock(&lun->lun_lock);
8155                 if ((key = ctl_get_prkey(lun, residx)) != 0) {
8156                         if (res_key != key) {
8157                                 /*
8158                                  * The current key passed in doesn't match
8159                                  * the one the initiator previously
8160                                  * registered.
8161                                  */
8162                                 mtx_unlock(&lun->lun_lock);
8163                                 free(ctsio->kern_data_ptr, M_CTL);
8164                                 ctl_set_reservation_conflict(ctsio);
8165                                 ctl_done((union ctl_io *)ctsio);
8166                                 return (CTL_RETVAL_COMPLETE);
8167                         }
8168                 } else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
8169                         /*
8170                          * We are not registered
8171                          */
8172                         mtx_unlock(&lun->lun_lock);
8173                         free(ctsio->kern_data_ptr, M_CTL);
8174                         ctl_set_reservation_conflict(ctsio);
8175                         ctl_done((union ctl_io *)ctsio);
8176                         return (CTL_RETVAL_COMPLETE);
8177                 } else if (res_key != 0) {
8178                         /*
8179                          * We are not registered and trying to register but
8180                          * the register key isn't zero.
8181                          */
8182                         mtx_unlock(&lun->lun_lock);
8183                         free(ctsio->kern_data_ptr, M_CTL);
8184                         ctl_set_reservation_conflict(ctsio);
8185                         ctl_done((union ctl_io *)ctsio);
8186                         return (CTL_RETVAL_COMPLETE);
8187                 }
8188                 mtx_unlock(&lun->lun_lock);
8189         }
8190
8191         switch (cdb->action & SPRO_ACTION_MASK) {
8192         case SPRO_REGISTER:
8193         case SPRO_REG_IGNO: {
8194
8195                 /*
8196                  * We don't support any of these options, as we report in
8197                  * the read capabilities request (see
8198                  * ctl_persistent_reserve_in(), above).
8199                  */
8200                 if ((param->flags & SPR_SPEC_I_PT)
8201                  || (param->flags & SPR_ALL_TG_PT)
8202                  || (param->flags & SPR_APTPL)) {
8203                         int bit_ptr;
8204
8205                         if (param->flags & SPR_APTPL)
8206                                 bit_ptr = 0;
8207                         else if (param->flags & SPR_ALL_TG_PT)
8208                                 bit_ptr = 2;
8209                         else /* SPR_SPEC_I_PT */
8210                                 bit_ptr = 3;
8211
8212                         free(ctsio->kern_data_ptr, M_CTL);
8213                         ctl_set_invalid_field(ctsio,
8214                                               /*sks_valid*/ 1,
8215                                               /*command*/ 0,
8216                                               /*field*/ 20,
8217                                               /*bit_valid*/ 1,
8218                                               /*bit*/ bit_ptr);
8219                         ctl_done((union ctl_io *)ctsio);
8220                         return (CTL_RETVAL_COMPLETE);
8221                 }
8222
8223                 mtx_lock(&lun->lun_lock);
8224
8225                 /*
8226                  * The initiator wants to clear the
8227                  * key/unregister.
8228                  */
8229                 if (sa_res_key == 0) {
8230                         if ((res_key == 0
8231                           && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
8232                          || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
8233                           && ctl_get_prkey(lun, residx) == 0)) {
8234                                 mtx_unlock(&lun->lun_lock);
8235                                 goto done;
8236                         }
8237
8238                         ctl_clr_prkey(lun, residx);
8239                         lun->pr_key_count--;
8240
8241                         if (residx == lun->pr_res_idx) {
8242                                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8243                                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8244
8245                                 if ((lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8246                                      lun->pr_res_type == SPR_TYPE_EX_AC_RO) &&
8247                                     lun->pr_key_count) {
8248                                         /*
8249                                          * If the reservation is a registrants
8250                                          * only type we need to generate a UA
8251                                          * for other registered inits.  The
8252                                          * sense code should be RESERVATIONS
8253                                          * RELEASED
8254                                          */
8255
8256                                         for (i = softc->init_min; i < softc->init_max; i++){
8257                                                 if (ctl_get_prkey(lun, i) == 0)
8258                                                         continue;
8259                                                 ctl_est_ua(lun, i,
8260                                                     CTL_UA_RES_RELEASE);
8261                                         }
8262                                 }
8263                                 lun->pr_res_type = 0;
8264                         } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8265                                 if (lun->pr_key_count==0) {
8266                                         lun->flags &= ~CTL_LUN_PR_RESERVED;
8267                                         lun->pr_res_type = 0;
8268                                         lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8269                                 }
8270                         }
8271                         lun->pr_generation++;
8272                         mtx_unlock(&lun->lun_lock);
8273
8274                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8275                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8276                         persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
8277                         persis_io.pr.pr_info.residx = residx;
8278                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8279                             sizeof(persis_io.pr), M_WAITOK);
8280                 } else /* sa_res_key != 0 */ {
8281
8282                         /*
8283                          * If we aren't registered currently then increment
8284                          * the key count and set the registered flag.
8285                          */
8286                         ctl_alloc_prkey(lun, residx);
8287                         if (ctl_get_prkey(lun, residx) == 0)
8288                                 lun->pr_key_count++;
8289                         ctl_set_prkey(lun, residx, sa_res_key);
8290                         lun->pr_generation++;
8291                         mtx_unlock(&lun->lun_lock);
8292
8293                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8294                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8295                         persis_io.pr.pr_info.action = CTL_PR_REG_KEY;
8296                         persis_io.pr.pr_info.residx = residx;
8297                         memcpy(persis_io.pr.pr_info.sa_res_key,
8298                                param->serv_act_res_key,
8299                                sizeof(param->serv_act_res_key));
8300                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8301                             sizeof(persis_io.pr), M_WAITOK);
8302                 }
8303
8304                 break;
8305         }
8306         case SPRO_RESERVE:
8307                 mtx_lock(&lun->lun_lock);
8308                 if (lun->flags & CTL_LUN_PR_RESERVED) {
8309                         /*
8310                          * if this isn't the reservation holder and it's
8311                          * not a "all registrants" type or if the type is
8312                          * different then we have a conflict
8313                          */
8314                         if ((lun->pr_res_idx != residx
8315                           && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8316                          || lun->pr_res_type != type) {
8317                                 mtx_unlock(&lun->lun_lock);
8318                                 free(ctsio->kern_data_ptr, M_CTL);
8319                                 ctl_set_reservation_conflict(ctsio);
8320                                 ctl_done((union ctl_io *)ctsio);
8321                                 return (CTL_RETVAL_COMPLETE);
8322                         }
8323                         mtx_unlock(&lun->lun_lock);
8324                 } else /* create a reservation */ {
8325                         /*
8326                          * If it's not an "all registrants" type record
8327                          * reservation holder
8328                          */
8329                         if (type != SPR_TYPE_WR_EX_AR
8330                          && type != SPR_TYPE_EX_AC_AR)
8331                                 lun->pr_res_idx = residx; /* Res holder */
8332                         else
8333                                 lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8334
8335                         lun->flags |= CTL_LUN_PR_RESERVED;
8336                         lun->pr_res_type = type;
8337
8338                         mtx_unlock(&lun->lun_lock);
8339
8340                         /* send msg to other side */
8341                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8342                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8343                         persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8344                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8345                         persis_io.pr.pr_info.res_type = type;
8346                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8347                             sizeof(persis_io.pr), M_WAITOK);
8348                 }
8349                 break;
8350
8351         case SPRO_RELEASE:
8352                 mtx_lock(&lun->lun_lock);
8353                 if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8354                         /* No reservation exists return good status */
8355                         mtx_unlock(&lun->lun_lock);
8356                         goto done;
8357                 }
8358                 /*
8359                  * Is this nexus a reservation holder?
8360                  */
8361                 if (lun->pr_res_idx != residx
8362                  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8363                         /*
8364                          * not a res holder return good status but
8365                          * do nothing
8366                          */
8367                         mtx_unlock(&lun->lun_lock);
8368                         goto done;
8369                 }
8370
8371                 if (lun->pr_res_type != type) {
8372                         mtx_unlock(&lun->lun_lock);
8373                         free(ctsio->kern_data_ptr, M_CTL);
8374                         ctl_set_illegal_pr_release(ctsio);
8375                         ctl_done((union ctl_io *)ctsio);
8376                         return (CTL_RETVAL_COMPLETE);
8377                 }
8378
8379                 /* okay to release */
8380                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8381                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8382                 lun->pr_res_type = 0;
8383
8384                 /*
8385                  * If this isn't an exclusive access reservation and NUAR
8386                  * is not set, generate UA for all other registrants.
8387                  */
8388                 if (type != SPR_TYPE_EX_AC && type != SPR_TYPE_WR_EX &&
8389                     (lun->MODE_CTRL.queue_flags & SCP_NUAR) == 0) {
8390                         for (i = softc->init_min; i < softc->init_max; i++) {
8391                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
8392                                         continue;
8393                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8394                         }
8395                 }
8396                 mtx_unlock(&lun->lun_lock);
8397
8398                 /* Send msg to other side */
8399                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8400                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8401                 persis_io.pr.pr_info.action = CTL_PR_RELEASE;
8402                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8403                      sizeof(persis_io.pr), M_WAITOK);
8404                 break;
8405
8406         case SPRO_CLEAR:
8407                 /* send msg to other side */
8408
8409                 mtx_lock(&lun->lun_lock);
8410                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8411                 lun->pr_res_type = 0;
8412                 lun->pr_key_count = 0;
8413                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8414
8415                 ctl_clr_prkey(lun, residx);
8416                 for (i = 0; i < CTL_MAX_INITIATORS; i++)
8417                         if (ctl_get_prkey(lun, i) != 0) {
8418                                 ctl_clr_prkey(lun, i);
8419                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8420                         }
8421                 lun->pr_generation++;
8422                 mtx_unlock(&lun->lun_lock);
8423
8424                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8425                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8426                 persis_io.pr.pr_info.action = CTL_PR_CLEAR;
8427                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8428                      sizeof(persis_io.pr), M_WAITOK);
8429                 break;
8430
8431         case SPRO_PREEMPT:
8432         case SPRO_PRE_ABO: {
8433                 int nretval;
8434
8435                 nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8436                                           residx, ctsio, cdb, param);
8437                 if (nretval != 0)
8438                         return (CTL_RETVAL_COMPLETE);
8439                 break;
8440         }
8441         default:
8442                 panic("%s: Invalid PR type %#x", __func__, cdb->action);
8443         }
8444
8445 done:
8446         free(ctsio->kern_data_ptr, M_CTL);
8447         ctl_set_success(ctsio);
8448         ctl_done((union ctl_io *)ctsio);
8449
8450         return (retval);
8451 }
8452
8453 /*
8454  * This routine is for handling a message from the other SC pertaining to
8455  * persistent reserve out. All the error checking will have been done
8456  * so only perorming the action need be done here to keep the two
8457  * in sync.
8458  */
8459 static void
8460 ctl_hndl_per_res_out_on_other_sc(union ctl_io *io)
8461 {
8462         struct ctl_softc *softc = CTL_SOFTC(io);
8463         union ctl_ha_msg *msg = (union ctl_ha_msg *)&io->presio.pr_msg;
8464         struct ctl_lun *lun;
8465         int i;
8466         uint32_t residx, targ_lun;
8467
8468         targ_lun = msg->hdr.nexus.targ_mapped_lun;
8469         mtx_lock(&softc->ctl_lock);
8470         if (targ_lun >= ctl_max_luns ||
8471             (lun = softc->ctl_luns[targ_lun]) == NULL) {
8472                 mtx_unlock(&softc->ctl_lock);
8473                 return;
8474         }
8475         mtx_lock(&lun->lun_lock);
8476         mtx_unlock(&softc->ctl_lock);
8477         if (lun->flags & CTL_LUN_DISABLED) {
8478                 mtx_unlock(&lun->lun_lock);
8479                 return;
8480         }
8481         residx = ctl_get_initindex(&msg->hdr.nexus);
8482         switch(msg->pr.pr_info.action) {
8483         case CTL_PR_REG_KEY:
8484                 ctl_alloc_prkey(lun, msg->pr.pr_info.residx);
8485                 if (ctl_get_prkey(lun, msg->pr.pr_info.residx) == 0)
8486                         lun->pr_key_count++;
8487                 ctl_set_prkey(lun, msg->pr.pr_info.residx,
8488                     scsi_8btou64(msg->pr.pr_info.sa_res_key));
8489                 lun->pr_generation++;
8490                 break;
8491
8492         case CTL_PR_UNREG_KEY:
8493                 ctl_clr_prkey(lun, msg->pr.pr_info.residx);
8494                 lun->pr_key_count--;
8495
8496                 /* XXX Need to see if the reservation has been released */
8497                 /* if so do we need to generate UA? */
8498                 if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8499                         lun->flags &= ~CTL_LUN_PR_RESERVED;
8500                         lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8501
8502                         if ((lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8503                              lun->pr_res_type == SPR_TYPE_EX_AC_RO) &&
8504                             lun->pr_key_count) {
8505                                 /*
8506                                  * If the reservation is a registrants
8507                                  * only type we need to generate a UA
8508                                  * for other registered inits.  The
8509                                  * sense code should be RESERVATIONS
8510                                  * RELEASED
8511                                  */
8512
8513                                 for (i = softc->init_min; i < softc->init_max; i++) {
8514                                         if (ctl_get_prkey(lun, i) == 0)
8515                                                 continue;
8516
8517                                         ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8518                                 }
8519                         }
8520                         lun->pr_res_type = 0;
8521                 } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8522                         if (lun->pr_key_count==0) {
8523                                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8524                                 lun->pr_res_type = 0;
8525                                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8526                         }
8527                 }
8528                 lun->pr_generation++;
8529                 break;
8530
8531         case CTL_PR_RESERVE:
8532                 lun->flags |= CTL_LUN_PR_RESERVED;
8533                 lun->pr_res_type = msg->pr.pr_info.res_type;
8534                 lun->pr_res_idx = msg->pr.pr_info.residx;
8535
8536                 break;
8537
8538         case CTL_PR_RELEASE:
8539                 /*
8540                  * If this isn't an exclusive access reservation and NUAR
8541                  * is not set, generate UA for all other registrants.
8542                  */
8543                 if (lun->pr_res_type != SPR_TYPE_EX_AC &&
8544                     lun->pr_res_type != SPR_TYPE_WR_EX &&
8545                     (lun->MODE_CTRL.queue_flags & SCP_NUAR) == 0) {
8546                         for (i = softc->init_min; i < softc->init_max; i++) {
8547                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
8548                                         continue;
8549                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8550                         }
8551                 }
8552
8553                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8554                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8555                 lun->pr_res_type = 0;
8556                 break;
8557
8558         case CTL_PR_PREEMPT:
8559                 ctl_pro_preempt_other(lun, msg);
8560                 break;
8561         case CTL_PR_CLEAR:
8562                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8563                 lun->pr_res_type = 0;
8564                 lun->pr_key_count = 0;
8565                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8566
8567                 for (i=0; i < CTL_MAX_INITIATORS; i++) {
8568                         if (ctl_get_prkey(lun, i) == 0)
8569                                 continue;
8570                         ctl_clr_prkey(lun, i);
8571                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8572                 }
8573                 lun->pr_generation++;
8574                 break;
8575         }
8576
8577         mtx_unlock(&lun->lun_lock);
8578 }
8579
8580 int
8581 ctl_read_write(struct ctl_scsiio *ctsio)
8582 {
8583         struct ctl_lun *lun = CTL_LUN(ctsio);
8584         struct ctl_lba_len_flags *lbalen;
8585         uint64_t lba;
8586         uint32_t num_blocks;
8587         int flags, retval;
8588         int isread;
8589
8590         CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
8591
8592         flags = 0;
8593         isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
8594               || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
8595         switch (ctsio->cdb[0]) {
8596         case READ_6:
8597         case WRITE_6: {
8598                 struct scsi_rw_6 *cdb;
8599
8600                 cdb = (struct scsi_rw_6 *)ctsio->cdb;
8601
8602                 lba = scsi_3btoul(cdb->addr);
8603                 /* only 5 bits are valid in the most significant address byte */
8604                 lba &= 0x1fffff;
8605                 num_blocks = cdb->length;
8606                 /*
8607                  * This is correct according to SBC-2.
8608                  */
8609                 if (num_blocks == 0)
8610                         num_blocks = 256;
8611                 break;
8612         }
8613         case READ_10:
8614         case WRITE_10: {
8615                 struct scsi_rw_10 *cdb;
8616
8617                 cdb = (struct scsi_rw_10 *)ctsio->cdb;
8618                 if (cdb->byte2 & SRW10_FUA)
8619                         flags |= CTL_LLF_FUA;
8620                 if (cdb->byte2 & SRW10_DPO)
8621                         flags |= CTL_LLF_DPO;
8622                 lba = scsi_4btoul(cdb->addr);
8623                 num_blocks = scsi_2btoul(cdb->length);
8624                 break;
8625         }
8626         case WRITE_VERIFY_10: {
8627                 struct scsi_write_verify_10 *cdb;
8628
8629                 cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
8630                 flags |= CTL_LLF_FUA;
8631                 if (cdb->byte2 & SWV_DPO)
8632                         flags |= CTL_LLF_DPO;
8633                 lba = scsi_4btoul(cdb->addr);
8634                 num_blocks = scsi_2btoul(cdb->length);
8635                 break;
8636         }
8637         case READ_12:
8638         case WRITE_12: {
8639                 struct scsi_rw_12 *cdb;
8640
8641                 cdb = (struct scsi_rw_12 *)ctsio->cdb;
8642                 if (cdb->byte2 & SRW12_FUA)
8643                         flags |= CTL_LLF_FUA;
8644                 if (cdb->byte2 & SRW12_DPO)
8645                         flags |= CTL_LLF_DPO;
8646                 lba = scsi_4btoul(cdb->addr);
8647                 num_blocks = scsi_4btoul(cdb->length);
8648                 break;
8649         }
8650         case WRITE_VERIFY_12: {
8651                 struct scsi_write_verify_12 *cdb;
8652
8653                 cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
8654                 flags |= CTL_LLF_FUA;
8655                 if (cdb->byte2 & SWV_DPO)
8656                         flags |= CTL_LLF_DPO;
8657                 lba = scsi_4btoul(cdb->addr);
8658                 num_blocks = scsi_4btoul(cdb->length);
8659                 break;
8660         }
8661         case READ_16:
8662         case WRITE_16: {
8663                 struct scsi_rw_16 *cdb;
8664
8665                 cdb = (struct scsi_rw_16 *)ctsio->cdb;
8666                 if (cdb->byte2 & SRW12_FUA)
8667                         flags |= CTL_LLF_FUA;
8668                 if (cdb->byte2 & SRW12_DPO)
8669                         flags |= CTL_LLF_DPO;
8670                 lba = scsi_8btou64(cdb->addr);
8671                 num_blocks = scsi_4btoul(cdb->length);
8672                 break;
8673         }
8674         case WRITE_ATOMIC_16: {
8675                 struct scsi_write_atomic_16 *cdb;
8676
8677                 if (lun->be_lun->atomicblock == 0) {
8678                         ctl_set_invalid_opcode(ctsio);
8679                         ctl_done((union ctl_io *)ctsio);
8680                         return (CTL_RETVAL_COMPLETE);
8681                 }
8682
8683                 cdb = (struct scsi_write_atomic_16 *)ctsio->cdb;
8684                 if (cdb->byte2 & SRW12_FUA)
8685                         flags |= CTL_LLF_FUA;
8686                 if (cdb->byte2 & SRW12_DPO)
8687                         flags |= CTL_LLF_DPO;
8688                 lba = scsi_8btou64(cdb->addr);
8689                 num_blocks = scsi_2btoul(cdb->length);
8690                 if (num_blocks > lun->be_lun->atomicblock) {
8691                         ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
8692                             /*command*/ 1, /*field*/ 12, /*bit_valid*/ 0,
8693                             /*bit*/ 0);
8694                         ctl_done((union ctl_io *)ctsio);
8695                         return (CTL_RETVAL_COMPLETE);
8696                 }
8697                 break;
8698         }
8699         case WRITE_VERIFY_16: {
8700                 struct scsi_write_verify_16 *cdb;
8701
8702                 cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
8703                 flags |= CTL_LLF_FUA;
8704                 if (cdb->byte2 & SWV_DPO)
8705                         flags |= CTL_LLF_DPO;
8706                 lba = scsi_8btou64(cdb->addr);
8707                 num_blocks = scsi_4btoul(cdb->length);
8708                 break;
8709         }
8710         default:
8711                 /*
8712                  * We got a command we don't support.  This shouldn't
8713                  * happen, commands should be filtered out above us.
8714                  */
8715                 ctl_set_invalid_opcode(ctsio);
8716                 ctl_done((union ctl_io *)ctsio);
8717
8718                 return (CTL_RETVAL_COMPLETE);
8719                 break; /* NOTREACHED */
8720         }
8721
8722         /*
8723          * The first check is to make sure we're in bounds, the second
8724          * check is to catch wrap-around problems.  If the lba + num blocks
8725          * is less than the lba, then we've wrapped around and the block
8726          * range is invalid anyway.
8727          */
8728         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8729          || ((lba + num_blocks) < lba)) {
8730                 ctl_set_lba_out_of_range(ctsio,
8731                     MAX(lba, lun->be_lun->maxlba + 1));
8732                 ctl_done((union ctl_io *)ctsio);
8733                 return (CTL_RETVAL_COMPLETE);
8734         }
8735
8736         /*
8737          * According to SBC-3, a transfer length of 0 is not an error.
8738          * Note that this cannot happen with WRITE(6) or READ(6), since 0
8739          * translates to 256 blocks for those commands.
8740          */
8741         if (num_blocks == 0) {
8742                 ctl_set_success(ctsio);
8743                 ctl_done((union ctl_io *)ctsio);
8744                 return (CTL_RETVAL_COMPLETE);
8745         }
8746
8747         /* Set FUA and/or DPO if caches are disabled. */
8748         if (isread) {
8749                 if ((lun->MODE_CACHING.flags1 & SCP_RCD) != 0)
8750                         flags |= CTL_LLF_FUA | CTL_LLF_DPO;
8751         } else {
8752                 if ((lun->MODE_CACHING.flags1 & SCP_WCE) == 0)
8753                         flags |= CTL_LLF_FUA;
8754         }
8755
8756         lbalen = (struct ctl_lba_len_flags *)
8757             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8758         lbalen->lba = lba;
8759         lbalen->len = num_blocks;
8760         lbalen->flags = (isread ? CTL_LLF_READ : CTL_LLF_WRITE) | flags;
8761
8762         ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
8763         ctsio->kern_rel_offset = 0;
8764
8765         CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
8766
8767         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8768         return (retval);
8769 }
8770
8771 static int
8772 ctl_cnw_cont(union ctl_io *io)
8773 {
8774         struct ctl_lun *lun = CTL_LUN(io);
8775         struct ctl_scsiio *ctsio;
8776         struct ctl_lba_len_flags *lbalen;
8777         int retval;
8778
8779         ctsio = &io->scsiio;
8780         ctsio->io_hdr.status = CTL_STATUS_NONE;
8781         ctsio->io_hdr.flags &= ~CTL_FLAG_IO_CONT;
8782         lbalen = (struct ctl_lba_len_flags *)
8783             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8784         lbalen->flags &= ~CTL_LLF_COMPARE;
8785         lbalen->flags |= CTL_LLF_WRITE;
8786
8787         CTL_DEBUG_PRINT(("ctl_cnw_cont: calling data_submit()\n"));
8788         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8789         return (retval);
8790 }
8791
8792 int
8793 ctl_cnw(struct ctl_scsiio *ctsio)
8794 {
8795         struct ctl_lun *lun = CTL_LUN(ctsio);
8796         struct ctl_lba_len_flags *lbalen;
8797         uint64_t lba;
8798         uint32_t num_blocks;
8799         int flags, retval;
8800
8801         CTL_DEBUG_PRINT(("ctl_cnw: command: %#x\n", ctsio->cdb[0]));
8802
8803         flags = 0;
8804         switch (ctsio->cdb[0]) {
8805         case COMPARE_AND_WRITE: {
8806                 struct scsi_compare_and_write *cdb;
8807
8808                 cdb = (struct scsi_compare_and_write *)ctsio->cdb;
8809                 if (cdb->byte2 & SRW10_FUA)
8810                         flags |= CTL_LLF_FUA;
8811                 if (cdb->byte2 & SRW10_DPO)
8812                         flags |= CTL_LLF_DPO;
8813                 lba = scsi_8btou64(cdb->addr);
8814                 num_blocks = cdb->length;
8815                 break;
8816         }
8817         default:
8818                 /*
8819                  * We got a command we don't support.  This shouldn't
8820                  * happen, commands should be filtered out above us.
8821                  */
8822                 ctl_set_invalid_opcode(ctsio);
8823                 ctl_done((union ctl_io *)ctsio);
8824
8825                 return (CTL_RETVAL_COMPLETE);
8826                 break; /* NOTREACHED */
8827         }
8828
8829         /*
8830          * The first check is to make sure we're in bounds, the second
8831          * check is to catch wrap-around problems.  If the lba + num blocks
8832          * is less than the lba, then we've wrapped around and the block
8833          * range is invalid anyway.
8834          */
8835         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8836          || ((lba + num_blocks) < lba)) {
8837                 ctl_set_lba_out_of_range(ctsio,
8838                     MAX(lba, lun->be_lun->maxlba + 1));
8839                 ctl_done((union ctl_io *)ctsio);
8840                 return (CTL_RETVAL_COMPLETE);
8841         }
8842
8843         /*
8844          * According to SBC-3, a transfer length of 0 is not an error.
8845          */
8846         if (num_blocks == 0) {
8847                 ctl_set_success(ctsio);
8848                 ctl_done((union ctl_io *)ctsio);
8849                 return (CTL_RETVAL_COMPLETE);
8850         }
8851
8852         /* Set FUA if write cache is disabled. */
8853         if ((lun->MODE_CACHING.flags1 & SCP_WCE) == 0)
8854                 flags |= CTL_LLF_FUA;
8855
8856         ctsio->kern_total_len = 2 * num_blocks * lun->be_lun->blocksize;
8857         ctsio->kern_rel_offset = 0;
8858
8859         /*
8860          * Set the IO_CONT flag, so that if this I/O gets passed to
8861          * ctl_data_submit_done(), it'll get passed back to
8862          * ctl_ctl_cnw_cont() for further processing.
8863          */
8864         ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
8865         ctsio->io_cont = ctl_cnw_cont;
8866
8867         lbalen = (struct ctl_lba_len_flags *)
8868             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8869         lbalen->lba = lba;
8870         lbalen->len = num_blocks;
8871         lbalen->flags = CTL_LLF_COMPARE | flags;
8872
8873         CTL_DEBUG_PRINT(("ctl_cnw: calling data_submit()\n"));
8874         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8875         return (retval);
8876 }
8877
8878 int
8879 ctl_verify(struct ctl_scsiio *ctsio)
8880 {
8881         struct ctl_lun *lun = CTL_LUN(ctsio);
8882         struct ctl_lba_len_flags *lbalen;
8883         uint64_t lba;
8884         uint32_t num_blocks;
8885         int bytchk, flags;
8886         int retval;
8887
8888         CTL_DEBUG_PRINT(("ctl_verify: command: %#x\n", ctsio->cdb[0]));
8889
8890         bytchk = 0;
8891         flags = CTL_LLF_FUA;
8892         switch (ctsio->cdb[0]) {
8893         case VERIFY_10: {
8894                 struct scsi_verify_10 *cdb;
8895
8896                 cdb = (struct scsi_verify_10 *)ctsio->cdb;
8897                 if (cdb->byte2 & SVFY_BYTCHK)
8898                         bytchk = 1;
8899                 if (cdb->byte2 & SVFY_DPO)
8900                         flags |= CTL_LLF_DPO;
8901                 lba = scsi_4btoul(cdb->addr);
8902                 num_blocks = scsi_2btoul(cdb->length);
8903                 break;
8904         }
8905         case VERIFY_12: {
8906                 struct scsi_verify_12 *cdb;
8907
8908                 cdb = (struct scsi_verify_12 *)ctsio->cdb;
8909                 if (cdb->byte2 & SVFY_BYTCHK)
8910                         bytchk = 1;
8911                 if (cdb->byte2 & SVFY_DPO)
8912                         flags |= CTL_LLF_DPO;
8913                 lba = scsi_4btoul(cdb->addr);
8914                 num_blocks = scsi_4btoul(cdb->length);
8915                 break;
8916         }
8917         case VERIFY_16: {
8918                 struct scsi_rw_16 *cdb;
8919
8920                 cdb = (struct scsi_rw_16 *)ctsio->cdb;
8921                 if (cdb->byte2 & SVFY_BYTCHK)
8922                         bytchk = 1;
8923                 if (cdb->byte2 & SVFY_DPO)
8924                         flags |= CTL_LLF_DPO;
8925                 lba = scsi_8btou64(cdb->addr);
8926                 num_blocks = scsi_4btoul(cdb->length);
8927                 break;
8928         }
8929         default:
8930                 /*
8931                  * We got a command we don't support.  This shouldn't
8932                  * happen, commands should be filtered out above us.
8933                  */
8934                 ctl_set_invalid_opcode(ctsio);
8935                 ctl_done((union ctl_io *)ctsio);
8936                 return (CTL_RETVAL_COMPLETE);
8937         }
8938
8939         /*
8940          * The first check is to make sure we're in bounds, the second
8941          * check is to catch wrap-around problems.  If the lba + num blocks
8942          * is less than the lba, then we've wrapped around and the block
8943          * range is invalid anyway.
8944          */
8945         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8946          || ((lba + num_blocks) < lba)) {
8947                 ctl_set_lba_out_of_range(ctsio,
8948                     MAX(lba, lun->be_lun->maxlba + 1));
8949                 ctl_done((union ctl_io *)ctsio);
8950                 return (CTL_RETVAL_COMPLETE);
8951         }
8952
8953         /*
8954          * According to SBC-3, a transfer length of 0 is not an error.
8955          */
8956         if (num_blocks == 0) {
8957                 ctl_set_success(ctsio);
8958                 ctl_done((union ctl_io *)ctsio);
8959                 return (CTL_RETVAL_COMPLETE);
8960         }
8961
8962         lbalen = (struct ctl_lba_len_flags *)
8963             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8964         lbalen->lba = lba;
8965         lbalen->len = num_blocks;
8966         if (bytchk) {
8967                 lbalen->flags = CTL_LLF_COMPARE | flags;
8968                 ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
8969         } else {
8970                 lbalen->flags = CTL_LLF_VERIFY | flags;
8971                 ctsio->kern_total_len = 0;
8972         }
8973         ctsio->kern_rel_offset = 0;
8974
8975         CTL_DEBUG_PRINT(("ctl_verify: calling data_submit()\n"));
8976         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8977         return (retval);
8978 }
8979
8980 int
8981 ctl_report_luns(struct ctl_scsiio *ctsio)
8982 {
8983         struct ctl_softc *softc = CTL_SOFTC(ctsio);
8984         struct ctl_port *port = CTL_PORT(ctsio);
8985         struct ctl_lun *lun, *request_lun = CTL_LUN(ctsio);
8986         struct scsi_report_luns *cdb;
8987         struct scsi_report_luns_data *lun_data;
8988         int num_filled, num_luns, num_port_luns, retval;
8989         uint32_t alloc_len, lun_datalen;
8990         uint32_t initidx, targ_lun_id, lun_id;
8991
8992         retval = CTL_RETVAL_COMPLETE;
8993         cdb = (struct scsi_report_luns *)ctsio->cdb;
8994
8995         CTL_DEBUG_PRINT(("ctl_report_luns\n"));
8996
8997         num_luns = 0;
8998         num_port_luns = port->lun_map ? port->lun_map_size : ctl_max_luns;
8999         mtx_lock(&softc->ctl_lock);
9000         for (targ_lun_id = 0; targ_lun_id < num_port_luns; targ_lun_id++) {
9001                 if (ctl_lun_map_from_port(port, targ_lun_id) != UINT32_MAX)
9002                         num_luns++;
9003         }
9004         mtx_unlock(&softc->ctl_lock);
9005
9006         switch (cdb->select_report) {
9007         case RPL_REPORT_DEFAULT:
9008         case RPL_REPORT_ALL:
9009         case RPL_REPORT_NONSUBSID:
9010                 break;
9011         case RPL_REPORT_WELLKNOWN:
9012         case RPL_REPORT_ADMIN:
9013         case RPL_REPORT_CONGLOM:
9014                 num_luns = 0;
9015                 break;
9016         default:
9017                 ctl_set_invalid_field(ctsio,
9018                                       /*sks_valid*/ 1,
9019                                       /*command*/ 1,
9020                                       /*field*/ 2,
9021                                       /*bit_valid*/ 0,
9022                                       /*bit*/ 0);
9023                 ctl_done((union ctl_io *)ctsio);
9024                 return (retval);
9025                 break; /* NOTREACHED */
9026         }
9027
9028         alloc_len = scsi_4btoul(cdb->length);
9029         /*
9030          * The initiator has to allocate at least 16 bytes for this request,
9031          * so he can at least get the header and the first LUN.  Otherwise
9032          * we reject the request (per SPC-3 rev 14, section 6.21).
9033          */
9034         if (alloc_len < (sizeof(struct scsi_report_luns_data) +
9035             sizeof(struct scsi_report_luns_lundata))) {
9036                 ctl_set_invalid_field(ctsio,
9037                                       /*sks_valid*/ 1,
9038                                       /*command*/ 1,
9039                                       /*field*/ 6,
9040                                       /*bit_valid*/ 0,
9041                                       /*bit*/ 0);
9042                 ctl_done((union ctl_io *)ctsio);
9043                 return (retval);
9044         }
9045
9046         lun_datalen = sizeof(*lun_data) +
9047                 (num_luns * sizeof(struct scsi_report_luns_lundata));
9048
9049         ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO);
9050         lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
9051         ctsio->kern_sg_entries = 0;
9052
9053         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9054
9055         mtx_lock(&softc->ctl_lock);
9056         for (targ_lun_id = 0, num_filled = 0;
9057             targ_lun_id < num_port_luns && num_filled < num_luns;
9058             targ_lun_id++) {
9059                 lun_id = ctl_lun_map_from_port(port, targ_lun_id);
9060                 if (lun_id == UINT32_MAX)
9061                         continue;
9062                 lun = softc->ctl_luns[lun_id];
9063                 if (lun == NULL)
9064                         continue;
9065
9066                 be64enc(lun_data->luns[num_filled++].lundata,
9067                     ctl_encode_lun(targ_lun_id));
9068
9069                 /*
9070                  * According to SPC-3, rev 14 section 6.21:
9071                  *
9072                  * "The execution of a REPORT LUNS command to any valid and
9073                  * installed logical unit shall clear the REPORTED LUNS DATA
9074                  * HAS CHANGED unit attention condition for all logical
9075                  * units of that target with respect to the requesting
9076                  * initiator. A valid and installed logical unit is one
9077                  * having a PERIPHERAL QUALIFIER of 000b in the standard
9078                  * INQUIRY data (see 6.4.2)."
9079                  *
9080                  * If request_lun is NULL, the LUN this report luns command
9081                  * was issued to is either disabled or doesn't exist. In that
9082                  * case, we shouldn't clear any pending lun change unit
9083                  * attention.
9084                  */
9085                 if (request_lun != NULL) {
9086                         mtx_lock(&lun->lun_lock);
9087                         ctl_clr_ua(lun, initidx, CTL_UA_LUN_CHANGE);
9088                         mtx_unlock(&lun->lun_lock);
9089                 }
9090         }
9091         mtx_unlock(&softc->ctl_lock);
9092
9093         /*
9094          * It's quite possible that we've returned fewer LUNs than we allocated
9095          * space for.  Trim it.
9096          */
9097         lun_datalen = sizeof(*lun_data) +
9098                 (num_filled * sizeof(struct scsi_report_luns_lundata));
9099         ctsio->kern_rel_offset = 0;
9100         ctsio->kern_sg_entries = 0;
9101         ctsio->kern_data_len = min(lun_datalen, alloc_len);
9102         ctsio->kern_total_len = ctsio->kern_data_len;
9103
9104         /*
9105          * We set this to the actual data length, regardless of how much
9106          * space we actually have to return results.  If the user looks at
9107          * this value, he'll know whether or not he allocated enough space
9108          * and reissue the command if necessary.  We don't support well
9109          * known logical units, so if the user asks for that, return none.
9110          */
9111         scsi_ulto4b(lun_datalen - 8, lun_data->length);
9112
9113         /*
9114          * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
9115          * this request.
9116          */
9117         ctl_set_success(ctsio);
9118         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9119         ctsio->be_move_done = ctl_config_move_done;
9120         ctl_datamove((union ctl_io *)ctsio);
9121         return (retval);
9122 }
9123
9124 int
9125 ctl_request_sense(struct ctl_scsiio *ctsio)
9126 {
9127         struct ctl_softc *softc = CTL_SOFTC(ctsio);
9128         struct ctl_lun *lun = CTL_LUN(ctsio);
9129         struct scsi_request_sense *cdb;
9130         struct scsi_sense_data *sense_ptr, *ps;
9131         uint32_t initidx;
9132         int have_error;
9133         u_int sense_len = SSD_FULL_SIZE;
9134         scsi_sense_data_type sense_format;
9135         ctl_ua_type ua_type;
9136         uint8_t asc = 0, ascq = 0;
9137
9138         cdb = (struct scsi_request_sense *)ctsio->cdb;
9139
9140         CTL_DEBUG_PRINT(("ctl_request_sense\n"));
9141
9142         /*
9143          * Determine which sense format the user wants.
9144          */
9145         if (cdb->byte2 & SRS_DESC)
9146                 sense_format = SSD_TYPE_DESC;
9147         else
9148                 sense_format = SSD_TYPE_FIXED;
9149
9150         ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
9151         sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
9152         ctsio->kern_sg_entries = 0;
9153         ctsio->kern_rel_offset = 0;
9154
9155         /*
9156          * struct scsi_sense_data, which is currently set to 256 bytes, is
9157          * larger than the largest allowed value for the length field in the
9158          * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
9159          */
9160         ctsio->kern_data_len = cdb->length;
9161         ctsio->kern_total_len = cdb->length;
9162
9163         /*
9164          * If we don't have a LUN, we don't have any pending sense.
9165          */
9166         if (lun == NULL ||
9167             ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
9168              softc->ha_link < CTL_HA_LINK_UNKNOWN)) {
9169                 /* "Logical unit not supported" */
9170                 ctl_set_sense_data(sense_ptr, &sense_len, NULL, sense_format,
9171                     /*current_error*/ 1,
9172                     /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
9173                     /*asc*/ 0x25,
9174                     /*ascq*/ 0x00,
9175                     SSD_ELEM_NONE);
9176                 goto send;
9177         }
9178
9179         have_error = 0;
9180         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9181         /*
9182          * Check for pending sense, and then for pending unit attentions.
9183          * Pending sense gets returned first, then pending unit attentions.
9184          */
9185         mtx_lock(&lun->lun_lock);
9186         ps = lun->pending_sense[initidx / CTL_MAX_INIT_PER_PORT];
9187         if (ps != NULL)
9188                 ps += initidx % CTL_MAX_INIT_PER_PORT;
9189         if (ps != NULL && ps->error_code != 0) {
9190                 scsi_sense_data_type stored_format;
9191
9192                 /*
9193                  * Check to see which sense format was used for the stored
9194                  * sense data.
9195                  */
9196                 stored_format = scsi_sense_type(ps);
9197
9198                 /*
9199                  * If the user requested a different sense format than the
9200                  * one we stored, then we need to convert it to the other
9201                  * format.  If we're going from descriptor to fixed format
9202                  * sense data, we may lose things in translation, depending
9203                  * on what options were used.
9204                  *
9205                  * If the stored format is SSD_TYPE_NONE (i.e. invalid),
9206                  * for some reason we'll just copy it out as-is.
9207                  */
9208                 if ((stored_format == SSD_TYPE_FIXED)
9209                  && (sense_format == SSD_TYPE_DESC))
9210                         ctl_sense_to_desc((struct scsi_sense_data_fixed *)
9211                             ps, (struct scsi_sense_data_desc *)sense_ptr);
9212                 else if ((stored_format == SSD_TYPE_DESC)
9213                       && (sense_format == SSD_TYPE_FIXED))
9214                         ctl_sense_to_fixed((struct scsi_sense_data_desc *)
9215                             ps, (struct scsi_sense_data_fixed *)sense_ptr);
9216                 else
9217                         memcpy(sense_ptr, ps, sizeof(*sense_ptr));
9218
9219                 ps->error_code = 0;
9220                 have_error = 1;
9221         } else {
9222                 ua_type = ctl_build_ua(lun, initidx, sense_ptr, &sense_len,
9223                     sense_format);
9224                 if (ua_type != CTL_UA_NONE)
9225                         have_error = 1;
9226         }
9227         if (have_error == 0) {
9228                 /*
9229                  * Report informational exception if have one and allowed.
9230                  */
9231                 if (lun->MODE_IE.mrie != SIEP_MRIE_NO) {
9232                         asc = lun->ie_asc;
9233                         ascq = lun->ie_ascq;
9234                 }
9235                 ctl_set_sense_data(sense_ptr, &sense_len, lun, sense_format,
9236                     /*current_error*/ 1,
9237                     /*sense_key*/ SSD_KEY_NO_SENSE,
9238                     /*asc*/ asc,
9239                     /*ascq*/ ascq,
9240                     SSD_ELEM_NONE);
9241         }
9242         mtx_unlock(&lun->lun_lock);
9243
9244 send:
9245         /*
9246          * We report the SCSI status as OK, since the status of the command
9247          * itself is OK.  We're reporting sense as parameter data.
9248          */
9249         ctl_set_success(ctsio);
9250         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9251         ctsio->be_move_done = ctl_config_move_done;
9252         ctl_datamove((union ctl_io *)ctsio);
9253         return (CTL_RETVAL_COMPLETE);
9254 }
9255
9256 int
9257 ctl_tur(struct ctl_scsiio *ctsio)
9258 {
9259
9260         CTL_DEBUG_PRINT(("ctl_tur\n"));
9261
9262         ctl_set_success(ctsio);
9263         ctl_done((union ctl_io *)ctsio);
9264
9265         return (CTL_RETVAL_COMPLETE);
9266 }
9267
9268 /*
9269  * SCSI VPD page 0x00, the Supported VPD Pages page.
9270  */
9271 static int
9272 ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
9273 {
9274         struct ctl_lun *lun = CTL_LUN(ctsio);
9275         struct scsi_vpd_supported_pages *pages;
9276         int sup_page_size;
9277         int p;
9278
9279         sup_page_size = sizeof(struct scsi_vpd_supported_pages) *
9280             SCSI_EVPD_NUM_SUPPORTED_PAGES;
9281         ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO);
9282         pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
9283         ctsio->kern_rel_offset = 0;
9284         ctsio->kern_sg_entries = 0;
9285         ctsio->kern_data_len = min(sup_page_size, alloc_len);
9286         ctsio->kern_total_len = ctsio->kern_data_len;
9287
9288         /*
9289          * The control device is always connected.  The disk device, on the
9290          * other hand, may not be online all the time.  Need to change this
9291          * to figure out whether the disk device is actually online or not.
9292          */
9293         if (lun != NULL)
9294                 pages->device = (SID_QUAL_LU_CONNECTED << 5) |
9295                                 lun->be_lun->lun_type;
9296         else
9297                 pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9298
9299         p = 0;
9300         /* Supported VPD pages */
9301         pages->page_list[p++] = SVPD_SUPPORTED_PAGES;
9302         /* Serial Number */
9303         pages->page_list[p++] = SVPD_UNIT_SERIAL_NUMBER;
9304         /* Device Identification */
9305         pages->page_list[p++] = SVPD_DEVICE_ID;
9306         /* Extended INQUIRY Data */
9307         pages->page_list[p++] = SVPD_EXTENDED_INQUIRY_DATA;
9308         /* Mode Page Policy */
9309         pages->page_list[p++] = SVPD_MODE_PAGE_POLICY;
9310         /* SCSI Ports */
9311         pages->page_list[p++] = SVPD_SCSI_PORTS;
9312         /* Third-party Copy */
9313         pages->page_list[p++] = SVPD_SCSI_TPC;
9314         if (lun != NULL && lun->be_lun->lun_type == T_DIRECT) {
9315                 /* Block limits */
9316                 pages->page_list[p++] = SVPD_BLOCK_LIMITS;
9317                 /* Block Device Characteristics */
9318                 pages->page_list[p++] = SVPD_BDC;
9319                 /* Logical Block Provisioning */
9320                 pages->page_list[p++] = SVPD_LBP;
9321         }
9322         pages->length = p;
9323
9324         ctl_set_success(ctsio);
9325         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9326         ctsio->be_move_done = ctl_config_move_done;
9327         ctl_datamove((union ctl_io *)ctsio);
9328         return (CTL_RETVAL_COMPLETE);
9329 }
9330
9331 /*
9332  * SCSI VPD page 0x80, the Unit Serial Number page.
9333  */
9334 static int
9335 ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
9336 {
9337         struct ctl_lun *lun = CTL_LUN(ctsio);
9338         struct scsi_vpd_unit_serial_number *sn_ptr;
9339         int data_len;
9340
9341         data_len = 4 + CTL_SN_LEN;
9342         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9343         sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
9344         ctsio->kern_rel_offset = 0;
9345         ctsio->kern_sg_entries = 0;
9346         ctsio->kern_data_len = min(data_len, alloc_len);
9347         ctsio->kern_total_len = ctsio->kern_data_len;
9348
9349         /*
9350          * The control device is always connected.  The disk device, on the
9351          * other hand, may not be online all the time.  Need to change this
9352          * to figure out whether the disk device is actually online or not.
9353          */
9354         if (lun != NULL)
9355                 sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9356                                   lun->be_lun->lun_type;
9357         else
9358                 sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9359
9360         sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9361         sn_ptr->length = CTL_SN_LEN;
9362         /*
9363          * If we don't have a LUN, we just leave the serial number as
9364          * all spaces.
9365          */
9366         if (lun != NULL) {
9367                 strncpy((char *)sn_ptr->serial_num,
9368                         (char *)lun->be_lun->serial_num, CTL_SN_LEN);
9369         } else
9370                 memset(sn_ptr->serial_num, 0x20, CTL_SN_LEN);
9371
9372         ctl_set_success(ctsio);
9373         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9374         ctsio->be_move_done = ctl_config_move_done;
9375         ctl_datamove((union ctl_io *)ctsio);
9376         return (CTL_RETVAL_COMPLETE);
9377 }
9378
9379
9380 /*
9381  * SCSI VPD page 0x86, the Extended INQUIRY Data page.
9382  */
9383 static int
9384 ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len)
9385 {
9386         struct ctl_lun *lun = CTL_LUN(ctsio);
9387         struct scsi_vpd_extended_inquiry_data *eid_ptr;
9388         int data_len;
9389
9390         data_len = sizeof(struct scsi_vpd_extended_inquiry_data);
9391         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9392         eid_ptr = (struct scsi_vpd_extended_inquiry_data *)ctsio->kern_data_ptr;
9393         ctsio->kern_sg_entries = 0;
9394         ctsio->kern_rel_offset = 0;
9395         ctsio->kern_data_len = min(data_len, alloc_len);
9396         ctsio->kern_total_len = ctsio->kern_data_len;
9397
9398         /*
9399          * The control device is always connected.  The disk device, on the
9400          * other hand, may not be online all the time.
9401          */
9402         if (lun != NULL)
9403                 eid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9404                                      lun->be_lun->lun_type;
9405         else
9406                 eid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9407         eid_ptr->page_code = SVPD_EXTENDED_INQUIRY_DATA;
9408         scsi_ulto2b(data_len - 4, eid_ptr->page_length);
9409         /*
9410          * We support head of queue, ordered and simple tags.
9411          */
9412         eid_ptr->flags2 = SVPD_EID_HEADSUP | SVPD_EID_ORDSUP | SVPD_EID_SIMPSUP;
9413         /*
9414          * Volatile cache supported.
9415          */
9416         eid_ptr->flags3 = SVPD_EID_V_SUP;
9417
9418         /*
9419          * This means that we clear the REPORTED LUNS DATA HAS CHANGED unit
9420          * attention for a particular IT nexus on all LUNs once we report
9421          * it to that nexus once.  This bit is required as of SPC-4.
9422          */
9423         eid_ptr->flags4 = SVPD_EID_LUICLR;
9424
9425         /*
9426          * We support revert to defaults (RTD) bit in MODE SELECT.
9427          */
9428         eid_ptr->flags5 = SVPD_EID_RTD_SUP;
9429
9430         /*
9431          * XXX KDM in order to correctly answer this, we would need
9432          * information from the SIM to determine how much sense data it
9433          * can send.  So this would really be a path inquiry field, most
9434          * likely.  This can be set to a maximum of 252 according to SPC-4,
9435          * but the hardware may or may not be able to support that much.
9436          * 0 just means that the maximum sense data length is not reported.
9437          */
9438         eid_ptr->max_sense_length = 0;
9439
9440         ctl_set_success(ctsio);
9441         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9442         ctsio->be_move_done = ctl_config_move_done;
9443         ctl_datamove((union ctl_io *)ctsio);
9444         return (CTL_RETVAL_COMPLETE);
9445 }
9446
9447 static int
9448 ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len)
9449 {
9450         struct ctl_lun *lun = CTL_LUN(ctsio);
9451         struct scsi_vpd_mode_page_policy *mpp_ptr;
9452         int data_len;
9453
9454         data_len = sizeof(struct scsi_vpd_mode_page_policy) +
9455             sizeof(struct scsi_vpd_mode_page_policy_descr);
9456
9457         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9458         mpp_ptr = (struct scsi_vpd_mode_page_policy *)ctsio->kern_data_ptr;
9459         ctsio->kern_rel_offset = 0;
9460         ctsio->kern_sg_entries = 0;
9461         ctsio->kern_data_len = min(data_len, alloc_len);
9462         ctsio->kern_total_len = ctsio->kern_data_len;
9463
9464         /*
9465          * The control device is always connected.  The disk device, on the
9466          * other hand, may not be online all the time.
9467          */
9468         if (lun != NULL)
9469                 mpp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9470                                      lun->be_lun->lun_type;
9471         else
9472                 mpp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9473         mpp_ptr->page_code = SVPD_MODE_PAGE_POLICY;
9474         scsi_ulto2b(data_len - 4, mpp_ptr->page_length);
9475         mpp_ptr->descr[0].page_code = 0x3f;
9476         mpp_ptr->descr[0].subpage_code = 0xff;
9477         mpp_ptr->descr[0].policy = SVPD_MPP_SHARED;
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  * SCSI VPD page 0x83, the Device Identification page.
9488  */
9489 static int
9490 ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
9491 {
9492         struct ctl_softc *softc = CTL_SOFTC(ctsio);
9493         struct ctl_port *port = CTL_PORT(ctsio);
9494         struct ctl_lun *lun = CTL_LUN(ctsio);
9495         struct scsi_vpd_device_id *devid_ptr;
9496         struct scsi_vpd_id_descriptor *desc;
9497         int data_len, g;
9498         uint8_t proto;
9499
9500         data_len = sizeof(struct scsi_vpd_device_id) +
9501             sizeof(struct scsi_vpd_id_descriptor) +
9502                 sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
9503             sizeof(struct scsi_vpd_id_descriptor) +
9504                 sizeof(struct scsi_vpd_id_trgt_port_grp_id);
9505         if (lun && lun->lun_devid)
9506                 data_len += lun->lun_devid->len;
9507         if (port && port->port_devid)
9508                 data_len += port->port_devid->len;
9509         if (port && port->target_devid)
9510                 data_len += port->target_devid->len;
9511
9512         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9513         devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
9514         ctsio->kern_sg_entries = 0;
9515         ctsio->kern_rel_offset = 0;
9516         ctsio->kern_sg_entries = 0;
9517         ctsio->kern_data_len = min(data_len, alloc_len);
9518         ctsio->kern_total_len = ctsio->kern_data_len;
9519
9520         /*
9521          * The control device is always connected.  The disk device, on the
9522          * other hand, may not be online all the time.
9523          */
9524         if (lun != NULL)
9525                 devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9526                                      lun->be_lun->lun_type;
9527         else
9528                 devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9529         devid_ptr->page_code = SVPD_DEVICE_ID;
9530         scsi_ulto2b(data_len - 4, devid_ptr->length);
9531
9532         if (port && port->port_type == CTL_PORT_FC)
9533                 proto = SCSI_PROTO_FC << 4;
9534         else if (port && port->port_type == CTL_PORT_SAS)
9535                 proto = SCSI_PROTO_SAS << 4;
9536         else if (port && port->port_type == CTL_PORT_ISCSI)
9537                 proto = SCSI_PROTO_ISCSI << 4;
9538         else
9539                 proto = SCSI_PROTO_SPI << 4;
9540         desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
9541
9542         /*
9543          * We're using a LUN association here.  i.e., this device ID is a
9544          * per-LUN identifier.
9545          */
9546         if (lun && lun->lun_devid) {
9547                 memcpy(desc, lun->lun_devid->data, lun->lun_devid->len);
9548                 desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9549                     lun->lun_devid->len);
9550         }
9551
9552         /*
9553          * This is for the WWPN which is a port association.
9554          */
9555         if (port && port->port_devid) {
9556                 memcpy(desc, port->port_devid->data, port->port_devid->len);
9557                 desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9558                     port->port_devid->len);
9559         }
9560
9561         /*
9562          * This is for the Relative Target Port(type 4h) identifier
9563          */
9564         desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9565         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9566             SVPD_ID_TYPE_RELTARG;
9567         desc->length = 4;
9568         scsi_ulto2b(ctsio->io_hdr.nexus.targ_port, &desc->identifier[2]);
9569         desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9570             sizeof(struct scsi_vpd_id_rel_trgt_port_id));
9571
9572         /*
9573          * This is for the Target Port Group(type 5h) identifier
9574          */
9575         desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9576         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9577             SVPD_ID_TYPE_TPORTGRP;
9578         desc->length = 4;
9579         if (softc->is_single ||
9580             (port && port->status & CTL_PORT_STATUS_HA_SHARED))
9581                 g = 1;
9582         else
9583                 g = 2 + ctsio->io_hdr.nexus.targ_port / softc->port_cnt;
9584         scsi_ulto2b(g, &desc->identifier[2]);
9585         desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9586             sizeof(struct scsi_vpd_id_trgt_port_grp_id));
9587
9588         /*
9589          * This is for the Target identifier
9590          */
9591         if (port && port->target_devid) {
9592                 memcpy(desc, port->target_devid->data, port->target_devid->len);
9593         }
9594
9595         ctl_set_success(ctsio);
9596         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9597         ctsio->be_move_done = ctl_config_move_done;
9598         ctl_datamove((union ctl_io *)ctsio);
9599         return (CTL_RETVAL_COMPLETE);
9600 }
9601
9602 static int
9603 ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio, int alloc_len)
9604 {
9605         struct ctl_softc *softc = CTL_SOFTC(ctsio);
9606         struct ctl_lun *lun = CTL_LUN(ctsio);
9607         struct scsi_vpd_scsi_ports *sp;
9608         struct scsi_vpd_port_designation *pd;
9609         struct scsi_vpd_port_designation_cont *pdc;
9610         struct ctl_port *port;
9611         int data_len, num_target_ports, iid_len, id_len;
9612
9613         num_target_ports = 0;
9614         iid_len = 0;
9615         id_len = 0;
9616         mtx_lock(&softc->ctl_lock);
9617         STAILQ_FOREACH(port, &softc->port_list, links) {
9618                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
9619                         continue;
9620                 if (lun != NULL &&
9621                     ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
9622                         continue;
9623                 num_target_ports++;
9624                 if (port->init_devid)
9625                         iid_len += port->init_devid->len;
9626                 if (port->port_devid)
9627                         id_len += port->port_devid->len;
9628         }
9629         mtx_unlock(&softc->ctl_lock);
9630
9631         data_len = sizeof(struct scsi_vpd_scsi_ports) +
9632             num_target_ports * (sizeof(struct scsi_vpd_port_designation) +
9633              sizeof(struct scsi_vpd_port_designation_cont)) + iid_len + id_len;
9634         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9635         sp = (struct scsi_vpd_scsi_ports *)ctsio->kern_data_ptr;
9636         ctsio->kern_sg_entries = 0;
9637         ctsio->kern_rel_offset = 0;
9638         ctsio->kern_sg_entries = 0;
9639         ctsio->kern_data_len = min(data_len, alloc_len);
9640         ctsio->kern_total_len = ctsio->kern_data_len;
9641
9642         /*
9643          * The control device is always connected.  The disk device, on the
9644          * other hand, may not be online all the time.  Need to change this
9645          * to figure out whether the disk device is actually online or not.
9646          */
9647         if (lun != NULL)
9648                 sp->device = (SID_QUAL_LU_CONNECTED << 5) |
9649                                   lun->be_lun->lun_type;
9650         else
9651                 sp->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9652
9653         sp->page_code = SVPD_SCSI_PORTS;
9654         scsi_ulto2b(data_len - sizeof(struct scsi_vpd_scsi_ports),
9655             sp->page_length);
9656         pd = &sp->design[0];
9657
9658         mtx_lock(&softc->ctl_lock);
9659         STAILQ_FOREACH(port, &softc->port_list, links) {
9660                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
9661                         continue;
9662                 if (lun != NULL &&
9663                     ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
9664                         continue;
9665                 scsi_ulto2b(port->targ_port, pd->relative_port_id);
9666                 if (port->init_devid) {
9667                         iid_len = port->init_devid->len;
9668                         memcpy(pd->initiator_transportid,
9669                             port->init_devid->data, port->init_devid->len);
9670                 } else
9671                         iid_len = 0;
9672                 scsi_ulto2b(iid_len, pd->initiator_transportid_length);
9673                 pdc = (struct scsi_vpd_port_designation_cont *)
9674                     (&pd->initiator_transportid[iid_len]);
9675                 if (port->port_devid) {
9676                         id_len = port->port_devid->len;
9677                         memcpy(pdc->target_port_descriptors,
9678                             port->port_devid->data, port->port_devid->len);
9679                 } else
9680                         id_len = 0;
9681                 scsi_ulto2b(id_len, pdc->target_port_descriptors_length);
9682                 pd = (struct scsi_vpd_port_designation *)
9683                     ((uint8_t *)pdc->target_port_descriptors + id_len);
9684         }
9685         mtx_unlock(&softc->ctl_lock);
9686
9687         ctl_set_success(ctsio);
9688         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9689         ctsio->be_move_done = ctl_config_move_done;
9690         ctl_datamove((union ctl_io *)ctsio);
9691         return (CTL_RETVAL_COMPLETE);
9692 }
9693
9694 static int
9695 ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio, int alloc_len)
9696 {
9697         struct ctl_lun *lun = CTL_LUN(ctsio);
9698         struct scsi_vpd_block_limits *bl_ptr;
9699         const char *val;
9700         uint64_t ival;
9701
9702         ctsio->kern_data_ptr = malloc(sizeof(*bl_ptr), M_CTL, M_WAITOK | M_ZERO);
9703         bl_ptr = (struct scsi_vpd_block_limits *)ctsio->kern_data_ptr;
9704         ctsio->kern_sg_entries = 0;
9705         ctsio->kern_rel_offset = 0;
9706         ctsio->kern_sg_entries = 0;
9707         ctsio->kern_data_len = min(sizeof(*bl_ptr), alloc_len);
9708         ctsio->kern_total_len = ctsio->kern_data_len;
9709
9710         /*
9711          * The control device is always connected.  The disk device, on the
9712          * other hand, may not be online all the time.  Need to change this
9713          * to figure out whether the disk device is actually online or not.
9714          */
9715         if (lun != NULL)
9716                 bl_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9717                                   lun->be_lun->lun_type;
9718         else
9719                 bl_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9720
9721         bl_ptr->page_code = SVPD_BLOCK_LIMITS;
9722         scsi_ulto2b(sizeof(*bl_ptr) - 4, bl_ptr->page_length);
9723         bl_ptr->max_cmp_write_len = 0xff;
9724         scsi_ulto4b(0xffffffff, bl_ptr->max_txfer_len);
9725         if (lun != NULL) {
9726                 scsi_ulto4b(lun->be_lun->opttxferlen, bl_ptr->opt_txfer_len);
9727                 if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
9728                         ival = 0xffffffff;
9729                         val = dnvlist_get_string(lun->be_lun->options,
9730                             "unmap_max_lba", NULL);
9731                         if (val != NULL)
9732                                 ctl_expand_number(val, &ival);
9733                         scsi_ulto4b(ival, bl_ptr->max_unmap_lba_cnt);
9734                         ival = 0xffffffff;
9735                         val = dnvlist_get_string(lun->be_lun->options,
9736                             "unmap_max_descr", NULL);
9737                         if (val != NULL)
9738                                 ctl_expand_number(val, &ival);
9739                         scsi_ulto4b(ival, bl_ptr->max_unmap_blk_cnt);
9740                         if (lun->be_lun->ublockexp != 0) {
9741                                 scsi_ulto4b((1 << lun->be_lun->ublockexp),
9742                                     bl_ptr->opt_unmap_grain);
9743                                 scsi_ulto4b(0x80000000 | lun->be_lun->ublockoff,
9744                                     bl_ptr->unmap_grain_align);
9745                         }
9746                 }
9747                 scsi_ulto4b(lun->be_lun->atomicblock,
9748                     bl_ptr->max_atomic_transfer_length);
9749                 scsi_ulto4b(0, bl_ptr->atomic_alignment);
9750                 scsi_ulto4b(0, bl_ptr->atomic_transfer_length_granularity);
9751                 scsi_ulto4b(0, bl_ptr->max_atomic_transfer_length_with_atomic_boundary);
9752                 scsi_ulto4b(0, bl_ptr->max_atomic_boundary_size);
9753                 ival = UINT64_MAX;
9754                 val = dnvlist_get_string(lun->be_lun->options,
9755                     "write_same_max_lba", NULL);
9756                 if (val != NULL)
9757                         ctl_expand_number(val, &ival);
9758                 scsi_u64to8b(ival, bl_ptr->max_write_same_length);
9759         }
9760
9761         ctl_set_success(ctsio);
9762         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9763         ctsio->be_move_done = ctl_config_move_done;
9764         ctl_datamove((union ctl_io *)ctsio);
9765         return (CTL_RETVAL_COMPLETE);
9766 }
9767
9768 static int
9769 ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len)
9770 {
9771         struct ctl_lun *lun = CTL_LUN(ctsio);
9772         struct scsi_vpd_block_device_characteristics *bdc_ptr;
9773         const char *value;
9774         u_int i;
9775
9776         ctsio->kern_data_ptr = malloc(sizeof(*bdc_ptr), M_CTL, M_WAITOK | M_ZERO);
9777         bdc_ptr = (struct scsi_vpd_block_device_characteristics *)ctsio->kern_data_ptr;
9778         ctsio->kern_sg_entries = 0;
9779         ctsio->kern_rel_offset = 0;
9780         ctsio->kern_data_len = min(sizeof(*bdc_ptr), alloc_len);
9781         ctsio->kern_total_len = ctsio->kern_data_len;
9782
9783         /*
9784          * The control device is always connected.  The disk device, on the
9785          * other hand, may not be online all the time.  Need to change this
9786          * to figure out whether the disk device is actually online or not.
9787          */
9788         if (lun != NULL)
9789                 bdc_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9790                                   lun->be_lun->lun_type;
9791         else
9792                 bdc_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9793         bdc_ptr->page_code = SVPD_BDC;
9794         scsi_ulto2b(sizeof(*bdc_ptr) - 4, bdc_ptr->page_length);
9795         if (lun != NULL &&
9796             (value = dnvlist_get_string(lun->be_lun->options, "rpm", NULL)) != NULL)
9797                 i = strtol(value, NULL, 0);
9798         else
9799                 i = CTL_DEFAULT_ROTATION_RATE;
9800         scsi_ulto2b(i, bdc_ptr->medium_rotation_rate);
9801         if (lun != NULL &&
9802             (value = dnvlist_get_string(lun->be_lun->options, "formfactor", NULL)) != NULL)
9803                 i = strtol(value, NULL, 0);
9804         else
9805                 i = 0;
9806         bdc_ptr->wab_wac_ff = (i & 0x0f);
9807         bdc_ptr->flags = SVPD_FUAB | SVPD_VBULS;
9808
9809         ctl_set_success(ctsio);
9810         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9811         ctsio->be_move_done = ctl_config_move_done;
9812         ctl_datamove((union ctl_io *)ctsio);
9813         return (CTL_RETVAL_COMPLETE);
9814 }
9815
9816 static int
9817 ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len)
9818 {
9819         struct ctl_lun *lun = CTL_LUN(ctsio);
9820         struct scsi_vpd_logical_block_prov *lbp_ptr;
9821         const char *value;
9822
9823         ctsio->kern_data_ptr = malloc(sizeof(*lbp_ptr), M_CTL, M_WAITOK | M_ZERO);
9824         lbp_ptr = (struct scsi_vpd_logical_block_prov *)ctsio->kern_data_ptr;
9825         ctsio->kern_sg_entries = 0;
9826         ctsio->kern_rel_offset = 0;
9827         ctsio->kern_data_len = min(sizeof(*lbp_ptr), alloc_len);
9828         ctsio->kern_total_len = ctsio->kern_data_len;
9829
9830         /*
9831          * The control device is always connected.  The disk device, on the
9832          * other hand, may not be online all the time.  Need to change this
9833          * to figure out whether the disk device is actually online or not.
9834          */
9835         if (lun != NULL)
9836                 lbp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9837                                   lun->be_lun->lun_type;
9838         else
9839                 lbp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9840
9841         lbp_ptr->page_code = SVPD_LBP;
9842         scsi_ulto2b(sizeof(*lbp_ptr) - 4, lbp_ptr->page_length);
9843         lbp_ptr->threshold_exponent = CTL_LBP_EXPONENT;
9844         if (lun != NULL && lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
9845                 lbp_ptr->flags = SVPD_LBP_UNMAP | SVPD_LBP_WS16 |
9846                     SVPD_LBP_WS10 | SVPD_LBP_RZ | SVPD_LBP_ANC_SUP;
9847                 value = dnvlist_get_string(lun->be_lun->options,
9848                     "provisioning_type", NULL);
9849                 if (value != NULL) {
9850                         if (strcmp(value, "resource") == 0)
9851                                 lbp_ptr->prov_type = SVPD_LBP_RESOURCE;
9852                         else if (strcmp(value, "thin") == 0)
9853                                 lbp_ptr->prov_type = SVPD_LBP_THIN;
9854                 } else
9855                         lbp_ptr->prov_type = SVPD_LBP_THIN;
9856         }
9857
9858         ctl_set_success(ctsio);
9859         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9860         ctsio->be_move_done = ctl_config_move_done;
9861         ctl_datamove((union ctl_io *)ctsio);
9862         return (CTL_RETVAL_COMPLETE);
9863 }
9864
9865 /*
9866  * INQUIRY with the EVPD bit set.
9867  */
9868 static int
9869 ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
9870 {
9871         struct ctl_lun *lun = CTL_LUN(ctsio);
9872         struct scsi_inquiry *cdb;
9873         int alloc_len, retval;
9874
9875         cdb = (struct scsi_inquiry *)ctsio->cdb;
9876         alloc_len = scsi_2btoul(cdb->length);
9877
9878         switch (cdb->page_code) {
9879         case SVPD_SUPPORTED_PAGES:
9880                 retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
9881                 break;
9882         case SVPD_UNIT_SERIAL_NUMBER:
9883                 retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
9884                 break;
9885         case SVPD_DEVICE_ID:
9886                 retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
9887                 break;
9888         case SVPD_EXTENDED_INQUIRY_DATA:
9889                 retval = ctl_inquiry_evpd_eid(ctsio, alloc_len);
9890                 break;
9891         case SVPD_MODE_PAGE_POLICY:
9892                 retval = ctl_inquiry_evpd_mpp(ctsio, alloc_len);
9893                 break;
9894         case SVPD_SCSI_PORTS:
9895                 retval = ctl_inquiry_evpd_scsi_ports(ctsio, alloc_len);
9896                 break;
9897         case SVPD_SCSI_TPC:
9898                 retval = ctl_inquiry_evpd_tpc(ctsio, alloc_len);
9899                 break;
9900         case SVPD_BLOCK_LIMITS:
9901                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
9902                         goto err;
9903                 retval = ctl_inquiry_evpd_block_limits(ctsio, alloc_len);
9904                 break;
9905         case SVPD_BDC:
9906                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
9907                         goto err;
9908                 retval = ctl_inquiry_evpd_bdc(ctsio, alloc_len);
9909                 break;
9910         case SVPD_LBP:
9911                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
9912                         goto err;
9913                 retval = ctl_inquiry_evpd_lbp(ctsio, alloc_len);
9914                 break;
9915         default:
9916 err:
9917                 ctl_set_invalid_field(ctsio,
9918                                       /*sks_valid*/ 1,
9919                                       /*command*/ 1,
9920                                       /*field*/ 2,
9921                                       /*bit_valid*/ 0,
9922                                       /*bit*/ 0);
9923                 ctl_done((union ctl_io *)ctsio);
9924                 retval = CTL_RETVAL_COMPLETE;
9925                 break;
9926         }
9927
9928         return (retval);
9929 }
9930
9931 /*
9932  * Standard INQUIRY data.
9933  */
9934 static int
9935 ctl_inquiry_std(struct ctl_scsiio *ctsio)
9936 {
9937         struct ctl_softc *softc = CTL_SOFTC(ctsio);
9938         struct ctl_port *port = CTL_PORT(ctsio);
9939         struct ctl_lun *lun = CTL_LUN(ctsio);
9940         struct scsi_inquiry_data *inq_ptr;
9941         struct scsi_inquiry *cdb;
9942         const char *val;
9943         uint32_t alloc_len, data_len;
9944         ctl_port_type port_type;
9945
9946         port_type = port->port_type;
9947         if (port_type == CTL_PORT_IOCTL || port_type == CTL_PORT_INTERNAL)
9948                 port_type = CTL_PORT_SCSI;
9949
9950         cdb = (struct scsi_inquiry *)ctsio->cdb;
9951         alloc_len = scsi_2btoul(cdb->length);
9952
9953         /*
9954          * We malloc the full inquiry data size here and fill it
9955          * in.  If the user only asks for less, we'll give him
9956          * that much.
9957          */
9958         data_len = offsetof(struct scsi_inquiry_data, vendor_specific1);
9959         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9960         inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
9961         ctsio->kern_sg_entries = 0;
9962         ctsio->kern_rel_offset = 0;
9963         ctsio->kern_data_len = min(data_len, alloc_len);
9964         ctsio->kern_total_len = ctsio->kern_data_len;
9965
9966         if (lun != NULL) {
9967                 if ((lun->flags & CTL_LUN_PRIMARY_SC) ||
9968                     softc->ha_link >= CTL_HA_LINK_UNKNOWN) {
9969                         inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9970                             lun->be_lun->lun_type;
9971                 } else {
9972                         inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) |
9973                             lun->be_lun->lun_type;
9974                 }
9975                 if (lun->flags & CTL_LUN_REMOVABLE)
9976                         inq_ptr->dev_qual2 |= SID_RMB;
9977         } else
9978                 inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE;
9979
9980         /* RMB in byte 2 is 0 */
9981         inq_ptr->version = SCSI_REV_SPC5;
9982
9983         /*
9984          * According to SAM-3, even if a device only supports a single
9985          * level of LUN addressing, it should still set the HISUP bit:
9986          *
9987          * 4.9.1 Logical unit numbers overview
9988          *
9989          * All logical unit number formats described in this standard are
9990          * hierarchical in structure even when only a single level in that
9991          * hierarchy is used. The HISUP bit shall be set to one in the
9992          * standard INQUIRY data (see SPC-2) when any logical unit number
9993          * format described in this standard is used.  Non-hierarchical
9994          * formats are outside the scope of this standard.
9995          *
9996          * Therefore we set the HiSup bit here.
9997          *
9998          * The response format is 2, per SPC-3.
9999          */
10000         inq_ptr->response_format = SID_HiSup | 2;
10001
10002         inq_ptr->additional_length = data_len -
10003             (offsetof(struct scsi_inquiry_data, additional_length) + 1);
10004         CTL_DEBUG_PRINT(("additional_length = %d\n",
10005                          inq_ptr->additional_length));
10006
10007         inq_ptr->spc3_flags = SPC3_SID_3PC | SPC3_SID_TPGS_IMPLICIT;
10008         if (port_type == CTL_PORT_SCSI)
10009                 inq_ptr->spc2_flags = SPC2_SID_ADDR16;
10010         inq_ptr->spc2_flags |= SPC2_SID_MultiP;
10011         inq_ptr->flags = SID_CmdQue;
10012         if (port_type == CTL_PORT_SCSI)
10013                 inq_ptr->flags |= SID_WBus16 | SID_Sync;
10014
10015         /*
10016          * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
10017          * We have 8 bytes for the vendor name, and 16 bytes for the device
10018          * name and 4 bytes for the revision.
10019          */
10020         if (lun == NULL || (val = dnvlist_get_string(lun->be_lun->options,
10021             "vendor", NULL)) == NULL) {
10022                 strncpy(inq_ptr->vendor, CTL_VENDOR, sizeof(inq_ptr->vendor));
10023         } else {
10024                 memset(inq_ptr->vendor, ' ', sizeof(inq_ptr->vendor));
10025                 strncpy(inq_ptr->vendor, val,
10026                     min(sizeof(inq_ptr->vendor), strlen(val)));
10027         }
10028         if (lun == NULL) {
10029                 strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10030                     sizeof(inq_ptr->product));
10031         } else if ((val = dnvlist_get_string(lun->be_lun->options, "product",
10032             NULL)) == NULL) {
10033                 switch (lun->be_lun->lun_type) {
10034                 case T_DIRECT:
10035                         strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10036                             sizeof(inq_ptr->product));
10037                         break;
10038                 case T_PROCESSOR:
10039                         strncpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT,
10040                             sizeof(inq_ptr->product));
10041                         break;
10042                 case T_CDROM:
10043                         strncpy(inq_ptr->product, CTL_CDROM_PRODUCT,
10044                             sizeof(inq_ptr->product));
10045                         break;
10046                 default:
10047                         strncpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT,
10048                             sizeof(inq_ptr->product));
10049                         break;
10050                 }
10051         } else {
10052                 memset(inq_ptr->product, ' ', sizeof(inq_ptr->product));
10053                 strncpy(inq_ptr->product, val,
10054                     min(sizeof(inq_ptr->product), strlen(val)));
10055         }
10056
10057         /*
10058          * XXX make this a macro somewhere so it automatically gets
10059          * incremented when we make changes.
10060          */
10061         if (lun == NULL || (val = dnvlist_get_string(lun->be_lun->options,
10062             "revision", NULL)) == NULL) {
10063                 strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
10064         } else {
10065                 memset(inq_ptr->revision, ' ', sizeof(inq_ptr->revision));
10066                 strncpy(inq_ptr->revision, val,
10067                     min(sizeof(inq_ptr->revision), strlen(val)));
10068         }
10069
10070         /*
10071          * For parallel SCSI, we support double transition and single
10072          * transition clocking.  We also support QAS (Quick Arbitration
10073          * and Selection) and Information Unit transfers on both the
10074          * control and array devices.
10075          */
10076         if (port_type == CTL_PORT_SCSI)
10077                 inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
10078                                     SID_SPI_IUS;
10079
10080         /* SAM-6 (no version claimed) */
10081         scsi_ulto2b(0x00C0, inq_ptr->version1);
10082         /* SPC-5 (no version claimed) */
10083         scsi_ulto2b(0x05C0, inq_ptr->version2);
10084         if (port_type == CTL_PORT_FC) {
10085                 /* FCP-2 ANSI INCITS.350:2003 */
10086                 scsi_ulto2b(0x0917, inq_ptr->version3);
10087         } else if (port_type == CTL_PORT_SCSI) {
10088                 /* SPI-4 ANSI INCITS.362:200x */
10089                 scsi_ulto2b(0x0B56, inq_ptr->version3);
10090         } else if (port_type == CTL_PORT_ISCSI) {
10091                 /* iSCSI (no version claimed) */
10092                 scsi_ulto2b(0x0960, inq_ptr->version3);
10093         } else if (port_type == CTL_PORT_SAS) {
10094                 /* SAS (no version claimed) */
10095                 scsi_ulto2b(0x0BE0, inq_ptr->version3);
10096         } else if (port_type == CTL_PORT_UMASS) {
10097                 /* USB Mass Storage Class Bulk-Only Transport, Revision 1.0 */
10098                 scsi_ulto2b(0x1730, inq_ptr->version3);
10099         }
10100
10101         if (lun == NULL) {
10102                 /* SBC-4 (no version claimed) */
10103                 scsi_ulto2b(0x0600, inq_ptr->version4);
10104         } else {
10105                 switch (lun->be_lun->lun_type) {
10106                 case T_DIRECT:
10107                         /* SBC-4 (no version claimed) */
10108                         scsi_ulto2b(0x0600, inq_ptr->version4);
10109                         break;
10110                 case T_PROCESSOR:
10111                         break;
10112                 case T_CDROM:
10113                         /* MMC-6 (no version claimed) */
10114                         scsi_ulto2b(0x04E0, inq_ptr->version4);
10115                         break;
10116                 default:
10117                         break;
10118                 }
10119         }
10120
10121         ctl_set_success(ctsio);
10122         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10123         ctsio->be_move_done = ctl_config_move_done;
10124         ctl_datamove((union ctl_io *)ctsio);
10125         return (CTL_RETVAL_COMPLETE);
10126 }
10127
10128 int
10129 ctl_inquiry(struct ctl_scsiio *ctsio)
10130 {
10131         struct scsi_inquiry *cdb;
10132         int retval;
10133
10134         CTL_DEBUG_PRINT(("ctl_inquiry\n"));
10135
10136         cdb = (struct scsi_inquiry *)ctsio->cdb;
10137         if (cdb->byte2 & SI_EVPD)
10138                 retval = ctl_inquiry_evpd(ctsio);
10139         else if (cdb->page_code == 0)
10140                 retval = ctl_inquiry_std(ctsio);
10141         else {
10142                 ctl_set_invalid_field(ctsio,
10143                                       /*sks_valid*/ 1,
10144                                       /*command*/ 1,
10145                                       /*field*/ 2,
10146                                       /*bit_valid*/ 0,
10147                                       /*bit*/ 0);
10148                 ctl_done((union ctl_io *)ctsio);
10149                 return (CTL_RETVAL_COMPLETE);
10150         }
10151
10152         return (retval);
10153 }
10154
10155 int
10156 ctl_get_config(struct ctl_scsiio *ctsio)
10157 {
10158         struct ctl_lun *lun = CTL_LUN(ctsio);
10159         struct scsi_get_config_header *hdr;
10160         struct scsi_get_config_feature *feature;
10161         struct scsi_get_config *cdb;
10162         uint32_t alloc_len, data_len;
10163         int rt, starting;
10164
10165         cdb = (struct scsi_get_config *)ctsio->cdb;
10166         rt = (cdb->rt & SGC_RT_MASK);
10167         starting = scsi_2btoul(cdb->starting_feature);
10168         alloc_len = scsi_2btoul(cdb->length);
10169
10170         data_len = sizeof(struct scsi_get_config_header) +
10171             sizeof(struct scsi_get_config_feature) + 8 +
10172             sizeof(struct scsi_get_config_feature) + 8 +
10173             sizeof(struct scsi_get_config_feature) + 4 +
10174             sizeof(struct scsi_get_config_feature) + 4 +
10175             sizeof(struct scsi_get_config_feature) + 8 +
10176             sizeof(struct scsi_get_config_feature) +
10177             sizeof(struct scsi_get_config_feature) + 4 +
10178             sizeof(struct scsi_get_config_feature) + 4 +
10179             sizeof(struct scsi_get_config_feature) + 4 +
10180             sizeof(struct scsi_get_config_feature) + 4 +
10181             sizeof(struct scsi_get_config_feature) + 4 +
10182             sizeof(struct scsi_get_config_feature) + 4;
10183         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10184         ctsio->kern_sg_entries = 0;
10185         ctsio->kern_rel_offset = 0;
10186
10187         hdr = (struct scsi_get_config_header *)ctsio->kern_data_ptr;
10188         if (lun->flags & CTL_LUN_NO_MEDIA)
10189                 scsi_ulto2b(0x0000, hdr->current_profile);
10190         else
10191                 scsi_ulto2b(0x0010, hdr->current_profile);
10192         feature = (struct scsi_get_config_feature *)(hdr + 1);
10193
10194         if (starting > 0x003b)
10195                 goto done;
10196         if (starting > 0x003a)
10197                 goto f3b;
10198         if (starting > 0x002b)
10199                 goto f3a;
10200         if (starting > 0x002a)
10201                 goto f2b;
10202         if (starting > 0x001f)
10203                 goto f2a;
10204         if (starting > 0x001e)
10205                 goto f1f;
10206         if (starting > 0x001d)
10207                 goto f1e;
10208         if (starting > 0x0010)
10209                 goto f1d;
10210         if (starting > 0x0003)
10211                 goto f10;
10212         if (starting > 0x0002)
10213                 goto f3;
10214         if (starting > 0x0001)
10215                 goto f2;
10216         if (starting > 0x0000)
10217                 goto f1;
10218
10219         /* Profile List */
10220         scsi_ulto2b(0x0000, feature->feature_code);
10221         feature->flags = SGC_F_PERSISTENT | SGC_F_CURRENT;
10222         feature->add_length = 8;
10223         scsi_ulto2b(0x0008, &feature->feature_data[0]); /* CD-ROM */
10224         feature->feature_data[2] = 0x00;
10225         scsi_ulto2b(0x0010, &feature->feature_data[4]); /* DVD-ROM */
10226         feature->feature_data[6] = 0x01;
10227         feature = (struct scsi_get_config_feature *)
10228             &feature->feature_data[feature->add_length];
10229
10230 f1:     /* Core */
10231         scsi_ulto2b(0x0001, feature->feature_code);
10232         feature->flags = 0x08 | SGC_F_PERSISTENT | SGC_F_CURRENT;
10233         feature->add_length = 8;
10234         scsi_ulto4b(0x00000000, &feature->feature_data[0]);
10235         feature->feature_data[4] = 0x03;
10236         feature = (struct scsi_get_config_feature *)
10237             &feature->feature_data[feature->add_length];
10238
10239 f2:     /* Morphing */
10240         scsi_ulto2b(0x0002, feature->feature_code);
10241         feature->flags = 0x04 | SGC_F_PERSISTENT | SGC_F_CURRENT;
10242         feature->add_length = 4;
10243         feature->feature_data[0] = 0x02;
10244         feature = (struct scsi_get_config_feature *)
10245             &feature->feature_data[feature->add_length];
10246
10247 f3:     /* Removable Medium */
10248         scsi_ulto2b(0x0003, feature->feature_code);
10249         feature->flags = 0x04 | SGC_F_PERSISTENT | SGC_F_CURRENT;
10250         feature->add_length = 4;
10251         feature->feature_data[0] = 0x39;
10252         feature = (struct scsi_get_config_feature *)
10253             &feature->feature_data[feature->add_length];
10254
10255         if (rt == SGC_RT_CURRENT && (lun->flags & CTL_LUN_NO_MEDIA))
10256                 goto done;
10257
10258 f10:    /* Random Read */
10259         scsi_ulto2b(0x0010, feature->feature_code);
10260         feature->flags = 0x00;
10261         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10262                 feature->flags |= SGC_F_CURRENT;
10263         feature->add_length = 8;
10264         scsi_ulto4b(lun->be_lun->blocksize, &feature->feature_data[0]);
10265         scsi_ulto2b(1, &feature->feature_data[4]);
10266         feature->feature_data[6] = 0x00;
10267         feature = (struct scsi_get_config_feature *)
10268             &feature->feature_data[feature->add_length];
10269
10270 f1d:    /* Multi-Read */
10271         scsi_ulto2b(0x001D, feature->feature_code);
10272         feature->flags = 0x00;
10273         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10274                 feature->flags |= SGC_F_CURRENT;
10275         feature->add_length = 0;
10276         feature = (struct scsi_get_config_feature *)
10277             &feature->feature_data[feature->add_length];
10278
10279 f1e:    /* CD Read */
10280         scsi_ulto2b(0x001E, feature->feature_code);
10281         feature->flags = 0x00;
10282         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10283                 feature->flags |= SGC_F_CURRENT;
10284         feature->add_length = 4;
10285         feature->feature_data[0] = 0x00;
10286         feature = (struct scsi_get_config_feature *)
10287             &feature->feature_data[feature->add_length];
10288
10289 f1f:    /* DVD Read */
10290         scsi_ulto2b(0x001F, feature->feature_code);
10291         feature->flags = 0x08;
10292         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10293                 feature->flags |= SGC_F_CURRENT;
10294         feature->add_length = 4;
10295         feature->feature_data[0] = 0x01;
10296         feature->feature_data[2] = 0x03;
10297         feature = (struct scsi_get_config_feature *)
10298             &feature->feature_data[feature->add_length];
10299
10300 f2a:    /* DVD+RW */
10301         scsi_ulto2b(0x002A, feature->feature_code);
10302         feature->flags = 0x04;
10303         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10304                 feature->flags |= SGC_F_CURRENT;
10305         feature->add_length = 4;
10306         feature->feature_data[0] = 0x00;
10307         feature->feature_data[1] = 0x00;
10308         feature = (struct scsi_get_config_feature *)
10309             &feature->feature_data[feature->add_length];
10310
10311 f2b:    /* DVD+R */
10312         scsi_ulto2b(0x002B, feature->feature_code);
10313         feature->flags = 0x00;
10314         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10315                 feature->flags |= SGC_F_CURRENT;
10316         feature->add_length = 4;
10317         feature->feature_data[0] = 0x00;
10318         feature = (struct scsi_get_config_feature *)
10319             &feature->feature_data[feature->add_length];
10320
10321 f3a:    /* DVD+RW Dual Layer */
10322         scsi_ulto2b(0x003A, feature->feature_code);
10323         feature->flags = 0x00;
10324         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10325                 feature->flags |= SGC_F_CURRENT;
10326         feature->add_length = 4;
10327         feature->feature_data[0] = 0x00;
10328         feature->feature_data[1] = 0x00;
10329         feature = (struct scsi_get_config_feature *)
10330             &feature->feature_data[feature->add_length];
10331
10332 f3b:    /* DVD+R Dual Layer */
10333         scsi_ulto2b(0x003B, feature->feature_code);
10334         feature->flags = 0x00;
10335         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10336                 feature->flags |= SGC_F_CURRENT;
10337         feature->add_length = 4;
10338         feature->feature_data[0] = 0x00;
10339         feature = (struct scsi_get_config_feature *)
10340             &feature->feature_data[feature->add_length];
10341
10342 done:
10343         data_len = (uint8_t *)feature - (uint8_t *)hdr;
10344         if (rt == SGC_RT_SPECIFIC && data_len > 4) {
10345                 feature = (struct scsi_get_config_feature *)(hdr + 1);
10346                 if (scsi_2btoul(feature->feature_code) == starting)
10347                         feature = (struct scsi_get_config_feature *)
10348                             &feature->feature_data[feature->add_length];
10349                 data_len = (uint8_t *)feature - (uint8_t *)hdr;
10350         }
10351         scsi_ulto4b(data_len - 4, hdr->data_length);
10352         ctsio->kern_data_len = min(data_len, alloc_len);
10353         ctsio->kern_total_len = ctsio->kern_data_len;
10354
10355         ctl_set_success(ctsio);
10356         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10357         ctsio->be_move_done = ctl_config_move_done;
10358         ctl_datamove((union ctl_io *)ctsio);
10359         return (CTL_RETVAL_COMPLETE);
10360 }
10361
10362 int
10363 ctl_get_event_status(struct ctl_scsiio *ctsio)
10364 {
10365         struct scsi_get_event_status_header *hdr;
10366         struct scsi_get_event_status *cdb;
10367         uint32_t alloc_len, data_len;
10368
10369         cdb = (struct scsi_get_event_status *)ctsio->cdb;
10370         if ((cdb->byte2 & SGESN_POLLED) == 0) {
10371                 ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 1,
10372                     /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 0);
10373                 ctl_done((union ctl_io *)ctsio);
10374                 return (CTL_RETVAL_COMPLETE);
10375         }
10376         alloc_len = scsi_2btoul(cdb->length);
10377
10378         data_len = sizeof(struct scsi_get_event_status_header);
10379         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10380         ctsio->kern_sg_entries = 0;
10381         ctsio->kern_rel_offset = 0;
10382         ctsio->kern_data_len = min(data_len, alloc_len);
10383         ctsio->kern_total_len = ctsio->kern_data_len;
10384
10385         hdr = (struct scsi_get_event_status_header *)ctsio->kern_data_ptr;
10386         scsi_ulto2b(0, hdr->descr_length);
10387         hdr->nea_class = SGESN_NEA;
10388         hdr->supported_class = 0;
10389
10390         ctl_set_success(ctsio);
10391         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10392         ctsio->be_move_done = ctl_config_move_done;
10393         ctl_datamove((union ctl_io *)ctsio);
10394         return (CTL_RETVAL_COMPLETE);
10395 }
10396
10397 int
10398 ctl_mechanism_status(struct ctl_scsiio *ctsio)
10399 {
10400         struct scsi_mechanism_status_header *hdr;
10401         struct scsi_mechanism_status *cdb;
10402         uint32_t alloc_len, data_len;
10403
10404         cdb = (struct scsi_mechanism_status *)ctsio->cdb;
10405         alloc_len = scsi_2btoul(cdb->length);
10406
10407         data_len = sizeof(struct scsi_mechanism_status_header);
10408         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10409         ctsio->kern_sg_entries = 0;
10410         ctsio->kern_rel_offset = 0;
10411         ctsio->kern_data_len = min(data_len, alloc_len);
10412         ctsio->kern_total_len = ctsio->kern_data_len;
10413
10414         hdr = (struct scsi_mechanism_status_header *)ctsio->kern_data_ptr;
10415         hdr->state1 = 0x00;
10416         hdr->state2 = 0xe0;
10417         scsi_ulto3b(0, hdr->lba);
10418         hdr->slots_num = 0;
10419         scsi_ulto2b(0, hdr->slots_length);
10420
10421         ctl_set_success(ctsio);
10422         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10423         ctsio->be_move_done = ctl_config_move_done;
10424         ctl_datamove((union ctl_io *)ctsio);
10425         return (CTL_RETVAL_COMPLETE);
10426 }
10427
10428 static void
10429 ctl_ultomsf(uint32_t lba, uint8_t *buf)
10430 {
10431
10432         lba += 150;
10433         buf[0] = 0;
10434         buf[1] = bin2bcd((lba / 75) / 60);
10435         buf[2] = bin2bcd((lba / 75) % 60);
10436         buf[3] = bin2bcd(lba % 75);
10437 }
10438
10439 int
10440 ctl_read_toc(struct ctl_scsiio *ctsio)
10441 {
10442         struct ctl_lun *lun = CTL_LUN(ctsio);
10443         struct scsi_read_toc_hdr *hdr;
10444         struct scsi_read_toc_type01_descr *descr;
10445         struct scsi_read_toc *cdb;
10446         uint32_t alloc_len, data_len;
10447         int format, msf;
10448
10449         cdb = (struct scsi_read_toc *)ctsio->cdb;
10450         msf = (cdb->byte2 & CD_MSF) != 0;
10451         format = cdb->format;
10452         alloc_len = scsi_2btoul(cdb->data_len);
10453
10454         data_len = sizeof(struct scsi_read_toc_hdr);
10455         if (format == 0)
10456                 data_len += 2 * sizeof(struct scsi_read_toc_type01_descr);
10457         else
10458                 data_len += sizeof(struct scsi_read_toc_type01_descr);
10459         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10460         ctsio->kern_sg_entries = 0;
10461         ctsio->kern_rel_offset = 0;
10462         ctsio->kern_data_len = min(data_len, alloc_len);
10463         ctsio->kern_total_len = ctsio->kern_data_len;
10464
10465         hdr = (struct scsi_read_toc_hdr *)ctsio->kern_data_ptr;
10466         if (format == 0) {
10467                 scsi_ulto2b(0x12, hdr->data_length);
10468                 hdr->first = 1;
10469                 hdr->last = 1;
10470                 descr = (struct scsi_read_toc_type01_descr *)(hdr + 1);
10471                 descr->addr_ctl = 0x14;
10472                 descr->track_number = 1;
10473                 if (msf)
10474                         ctl_ultomsf(0, descr->track_start);
10475                 else
10476                         scsi_ulto4b(0, descr->track_start);
10477                 descr++;
10478                 descr->addr_ctl = 0x14;
10479                 descr->track_number = 0xaa;
10480                 if (msf)
10481                         ctl_ultomsf(lun->be_lun->maxlba+1, descr->track_start);
10482                 else
10483                         scsi_ulto4b(lun->be_lun->maxlba+1, descr->track_start);
10484         } else {
10485                 scsi_ulto2b(0x0a, hdr->data_length);
10486                 hdr->first = 1;
10487                 hdr->last = 1;
10488                 descr = (struct scsi_read_toc_type01_descr *)(hdr + 1);
10489                 descr->addr_ctl = 0x14;
10490                 descr->track_number = 1;
10491                 if (msf)
10492                         ctl_ultomsf(0, descr->track_start);
10493                 else
10494                         scsi_ulto4b(0, descr->track_start);
10495         }
10496
10497         ctl_set_success(ctsio);
10498         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10499         ctsio->be_move_done = ctl_config_move_done;
10500         ctl_datamove((union ctl_io *)ctsio);
10501         return (CTL_RETVAL_COMPLETE);
10502 }
10503
10504 /*
10505  * For known CDB types, parse the LBA and length.
10506  */
10507 static int
10508 ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len)
10509 {
10510         if (io->io_hdr.io_type != CTL_IO_SCSI)
10511                 return (1);
10512
10513         switch (io->scsiio.cdb[0]) {
10514         case COMPARE_AND_WRITE: {
10515                 struct scsi_compare_and_write *cdb;
10516
10517                 cdb = (struct scsi_compare_and_write *)io->scsiio.cdb;
10518
10519                 *lba = scsi_8btou64(cdb->addr);
10520                 *len = cdb->length;
10521                 break;
10522         }
10523         case READ_6:
10524         case WRITE_6: {
10525                 struct scsi_rw_6 *cdb;
10526
10527                 cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
10528
10529                 *lba = scsi_3btoul(cdb->addr);
10530                 /* only 5 bits are valid in the most significant address byte */
10531                 *lba &= 0x1fffff;
10532                 *len = cdb->length;
10533                 break;
10534         }
10535         case READ_10:
10536         case WRITE_10: {
10537                 struct scsi_rw_10 *cdb;
10538
10539                 cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
10540
10541                 *lba = scsi_4btoul(cdb->addr);
10542                 *len = scsi_2btoul(cdb->length);
10543                 break;
10544         }
10545         case WRITE_VERIFY_10: {
10546                 struct scsi_write_verify_10 *cdb;
10547
10548                 cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
10549
10550                 *lba = scsi_4btoul(cdb->addr);
10551                 *len = scsi_2btoul(cdb->length);
10552                 break;
10553         }
10554         case READ_12:
10555         case WRITE_12: {
10556                 struct scsi_rw_12 *cdb;
10557
10558                 cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
10559
10560                 *lba = scsi_4btoul(cdb->addr);
10561                 *len = scsi_4btoul(cdb->length);
10562                 break;
10563         }
10564         case WRITE_VERIFY_12: {
10565                 struct scsi_write_verify_12 *cdb;
10566
10567                 cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
10568
10569                 *lba = scsi_4btoul(cdb->addr);
10570                 *len = scsi_4btoul(cdb->length);
10571                 break;
10572         }
10573         case READ_16:
10574         case WRITE_16: {
10575                 struct scsi_rw_16 *cdb;
10576
10577                 cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
10578
10579                 *lba = scsi_8btou64(cdb->addr);
10580                 *len = scsi_4btoul(cdb->length);
10581                 break;
10582         }
10583         case WRITE_ATOMIC_16: {
10584                 struct scsi_write_atomic_16 *cdb;
10585
10586                 cdb = (struct scsi_write_atomic_16 *)io->scsiio.cdb;
10587
10588                 *lba = scsi_8btou64(cdb->addr);
10589                 *len = scsi_2btoul(cdb->length);
10590                 break;
10591         }
10592         case WRITE_VERIFY_16: {
10593                 struct scsi_write_verify_16 *cdb;
10594
10595                 cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
10596
10597                 *lba = scsi_8btou64(cdb->addr);
10598                 *len = scsi_4btoul(cdb->length);
10599                 break;
10600         }
10601         case WRITE_SAME_10: {
10602                 struct scsi_write_same_10 *cdb;
10603
10604                 cdb = (struct scsi_write_same_10 *)io->scsiio.cdb;
10605
10606                 *lba = scsi_4btoul(cdb->addr);
10607                 *len = scsi_2btoul(cdb->length);
10608                 break;
10609         }
10610         case WRITE_SAME_16: {
10611                 struct scsi_write_same_16 *cdb;
10612
10613                 cdb = (struct scsi_write_same_16 *)io->scsiio.cdb;
10614
10615                 *lba = scsi_8btou64(cdb->addr);
10616                 *len = scsi_4btoul(cdb->length);
10617                 break;
10618         }
10619         case VERIFY_10: {
10620                 struct scsi_verify_10 *cdb;
10621
10622                 cdb = (struct scsi_verify_10 *)io->scsiio.cdb;
10623
10624                 *lba = scsi_4btoul(cdb->addr);
10625                 *len = scsi_2btoul(cdb->length);
10626                 break;
10627         }
10628         case VERIFY_12: {
10629                 struct scsi_verify_12 *cdb;
10630
10631                 cdb = (struct scsi_verify_12 *)io->scsiio.cdb;
10632
10633                 *lba = scsi_4btoul(cdb->addr);
10634                 *len = scsi_4btoul(cdb->length);
10635                 break;
10636         }
10637         case VERIFY_16: {
10638                 struct scsi_verify_16 *cdb;
10639
10640                 cdb = (struct scsi_verify_16 *)io->scsiio.cdb;
10641
10642                 *lba = scsi_8btou64(cdb->addr);
10643                 *len = scsi_4btoul(cdb->length);
10644                 break;
10645         }
10646         case UNMAP: {
10647                 *lba = 0;
10648                 *len = UINT64_MAX;
10649                 break;
10650         }
10651         case SERVICE_ACTION_IN: {       /* GET LBA STATUS */
10652                 struct scsi_get_lba_status *cdb;
10653
10654                 cdb = (struct scsi_get_lba_status *)io->scsiio.cdb;
10655                 *lba = scsi_8btou64(cdb->addr);
10656                 *len = UINT32_MAX;
10657                 break;
10658         }
10659         default:
10660                 return (1);
10661                 break; /* NOTREACHED */
10662         }
10663
10664         return (0);
10665 }
10666
10667 static ctl_action
10668 ctl_extent_check_lba(uint64_t lba1, uint64_t len1, uint64_t lba2, uint64_t len2,
10669     bool seq)
10670 {
10671         uint64_t endlba1, endlba2;
10672
10673         endlba1 = lba1 + len1 - (seq ? 0 : 1);
10674         endlba2 = lba2 + len2 - 1;
10675
10676         if ((endlba1 < lba2) || (endlba2 < lba1))
10677                 return (CTL_ACTION_PASS);
10678         else
10679                 return (CTL_ACTION_BLOCK);
10680 }
10681
10682 static int
10683 ctl_extent_check_unmap(union ctl_io *io, uint64_t lba2, uint64_t len2)
10684 {
10685         struct ctl_ptr_len_flags *ptrlen;
10686         struct scsi_unmap_desc *buf, *end, *range;
10687         uint64_t lba;
10688         uint32_t len;
10689
10690         /* If not UNMAP -- go other way. */
10691         if (io->io_hdr.io_type != CTL_IO_SCSI ||
10692             io->scsiio.cdb[0] != UNMAP)
10693                 return (CTL_ACTION_ERROR);
10694
10695         /* If UNMAP without data -- block and wait for data. */
10696         ptrlen = (struct ctl_ptr_len_flags *)
10697             &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
10698         if ((io->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0 ||
10699             ptrlen->ptr == NULL)
10700                 return (CTL_ACTION_BLOCK);
10701
10702         /* UNMAP with data -- check for collision. */
10703         buf = (struct scsi_unmap_desc *)ptrlen->ptr;
10704         end = buf + ptrlen->len / sizeof(*buf);
10705         for (range = buf; range < end; range++) {
10706                 lba = scsi_8btou64(range->lba);
10707                 len = scsi_4btoul(range->length);
10708                 if ((lba < lba2 + len2) && (lba + len > lba2))
10709                         return (CTL_ACTION_BLOCK);
10710         }
10711         return (CTL_ACTION_PASS);
10712 }
10713
10714 static ctl_action
10715 ctl_extent_check(union ctl_io *io1, union ctl_io *io2, bool seq)
10716 {
10717         uint64_t lba1, lba2;
10718         uint64_t len1, len2;
10719         int retval;
10720
10721         if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
10722                 return (CTL_ACTION_ERROR);
10723
10724         retval = ctl_extent_check_unmap(io1, lba2, len2);
10725         if (retval != CTL_ACTION_ERROR)
10726                 return (retval);
10727
10728         if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
10729                 return (CTL_ACTION_ERROR);
10730
10731         if (io1->io_hdr.flags & CTL_FLAG_SERSEQ_DONE)
10732                 seq = FALSE;
10733         return (ctl_extent_check_lba(lba1, len1, lba2, len2, seq));
10734 }
10735
10736 static ctl_action
10737 ctl_extent_check_seq(union ctl_io *io1, union ctl_io *io2)
10738 {
10739         uint64_t lba1, lba2;
10740         uint64_t len1, len2;
10741
10742         if (io1->io_hdr.flags & CTL_FLAG_SERSEQ_DONE)
10743                 return (CTL_ACTION_PASS);
10744         if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
10745                 return (CTL_ACTION_ERROR);
10746         if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
10747                 return (CTL_ACTION_ERROR);
10748
10749         if (lba1 + len1 == lba2)
10750                 return (CTL_ACTION_BLOCK);
10751         return (CTL_ACTION_PASS);
10752 }
10753
10754 static ctl_action
10755 ctl_check_for_blockage(struct ctl_lun *lun, union ctl_io *pending_io,
10756     union ctl_io *ooa_io)
10757 {
10758         const struct ctl_cmd_entry *pending_entry, *ooa_entry;
10759         const ctl_serialize_action *serialize_row;
10760
10761         /*
10762          * The initiator attempted multiple untagged commands at the same
10763          * time.  Can't do that.
10764          */
10765         if ((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10766          && (ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10767          && ((pending_io->io_hdr.nexus.targ_port ==
10768               ooa_io->io_hdr.nexus.targ_port)
10769           && (pending_io->io_hdr.nexus.initid ==
10770               ooa_io->io_hdr.nexus.initid))
10771          && ((ooa_io->io_hdr.flags & (CTL_FLAG_ABORT |
10772               CTL_FLAG_STATUS_SENT)) == 0))
10773                 return (CTL_ACTION_OVERLAP);
10774
10775         /*
10776          * The initiator attempted to send multiple tagged commands with
10777          * the same ID.  (It's fine if different initiators have the same
10778          * tag ID.)
10779          *
10780          * Even if all of those conditions are true, we don't kill the I/O
10781          * if the command ahead of us has been aborted.  We won't end up
10782          * sending it to the FETD, and it's perfectly legal to resend a
10783          * command with the same tag number as long as the previous
10784          * instance of this tag number has been aborted somehow.
10785          */
10786         if ((pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10787          && (ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10788          && (pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
10789          && ((pending_io->io_hdr.nexus.targ_port ==
10790               ooa_io->io_hdr.nexus.targ_port)
10791           && (pending_io->io_hdr.nexus.initid ==
10792               ooa_io->io_hdr.nexus.initid))
10793          && ((ooa_io->io_hdr.flags & (CTL_FLAG_ABORT |
10794               CTL_FLAG_STATUS_SENT)) == 0))
10795                 return (CTL_ACTION_OVERLAP_TAG);
10796
10797         /*
10798          * If we get a head of queue tag, SAM-3 says that we should
10799          * immediately execute it.
10800          *
10801          * What happens if this command would normally block for some other
10802          * reason?  e.g. a request sense with a head of queue tag
10803          * immediately after a write.  Normally that would block, but this
10804          * will result in its getting executed immediately...
10805          *
10806          * We currently return "pass" instead of "skip", so we'll end up
10807          * going through the rest of the queue to check for overlapped tags.
10808          *
10809          * XXX KDM check for other types of blockage first??
10810          */
10811         if (pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10812                 return (CTL_ACTION_PASS);
10813
10814         /*
10815          * Ordered tags have to block until all items ahead of them
10816          * have completed.  If we get called with an ordered tag, we always
10817          * block, if something else is ahead of us in the queue.
10818          */
10819         if (pending_io->scsiio.tag_type == CTL_TAG_ORDERED)
10820                 return (CTL_ACTION_BLOCK);
10821
10822         /*
10823          * Simple tags get blocked until all head of queue and ordered tags
10824          * ahead of them have completed.  I'm lumping untagged commands in
10825          * with simple tags here.  XXX KDM is that the right thing to do?
10826          */
10827         if (((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10828           || (pending_io->scsiio.tag_type == CTL_TAG_SIMPLE))
10829          && ((ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10830           || (ooa_io->scsiio.tag_type == CTL_TAG_ORDERED)))
10831                 return (CTL_ACTION_BLOCK);
10832
10833         pending_entry = ctl_get_cmd_entry(&pending_io->scsiio, NULL);
10834         KASSERT(pending_entry->seridx < CTL_SERIDX_COUNT,
10835             ("%s: Invalid seridx %d for pending CDB %02x %02x @ %p",
10836              __func__, pending_entry->seridx, pending_io->scsiio.cdb[0],
10837              pending_io->scsiio.cdb[1], pending_io));
10838         ooa_entry = ctl_get_cmd_entry(&ooa_io->scsiio, NULL);
10839         if (ooa_entry->seridx == CTL_SERIDX_INVLD)
10840                 return (CTL_ACTION_PASS); /* Unsupported command in OOA queue */
10841         KASSERT(ooa_entry->seridx < CTL_SERIDX_COUNT,
10842             ("%s: Invalid seridx %d for ooa CDB %02x %02x @ %p",
10843              __func__, ooa_entry->seridx, ooa_io->scsiio.cdb[0],
10844              ooa_io->scsiio.cdb[1], ooa_io));
10845
10846         serialize_row = ctl_serialize_table[ooa_entry->seridx];
10847
10848         switch (serialize_row[pending_entry->seridx]) {
10849         case CTL_SER_BLOCK:
10850                 return (CTL_ACTION_BLOCK);
10851         case CTL_SER_EXTENT:
10852                 return (ctl_extent_check(ooa_io, pending_io,
10853                     (lun->be_lun && lun->be_lun->serseq == CTL_LUN_SERSEQ_ON)));
10854         case CTL_SER_EXTENTOPT:
10855                 if ((lun->MODE_CTRL.queue_flags & SCP_QUEUE_ALG_MASK) !=
10856                     SCP_QUEUE_ALG_UNRESTRICTED)
10857                         return (ctl_extent_check(ooa_io, pending_io,
10858                             (lun->be_lun &&
10859                              lun->be_lun->serseq == CTL_LUN_SERSEQ_ON)));
10860                 return (CTL_ACTION_PASS);
10861         case CTL_SER_EXTENTSEQ:
10862                 if (lun->be_lun && lun->be_lun->serseq != CTL_LUN_SERSEQ_OFF)
10863                         return (ctl_extent_check_seq(ooa_io, pending_io));
10864                 return (CTL_ACTION_PASS);
10865         case CTL_SER_PASS:
10866                 return (CTL_ACTION_PASS);
10867         case CTL_SER_BLOCKOPT:
10868                 if ((lun->MODE_CTRL.queue_flags & SCP_QUEUE_ALG_MASK) !=
10869                     SCP_QUEUE_ALG_UNRESTRICTED)
10870                         return (CTL_ACTION_BLOCK);
10871                 return (CTL_ACTION_PASS);
10872         case CTL_SER_SKIP:
10873                 return (CTL_ACTION_SKIP);
10874         default:
10875                 panic("%s: Invalid serialization value %d for %d => %d",
10876                     __func__, serialize_row[pending_entry->seridx],
10877                     pending_entry->seridx, ooa_entry->seridx);
10878         }
10879
10880         return (CTL_ACTION_ERROR);
10881 }
10882
10883 /*
10884  * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
10885  * Assumptions:
10886  * - pending_io is generally either incoming, or on the blocked queue
10887  * - starting I/O is the I/O we want to start the check with.
10888  */
10889 static ctl_action
10890 ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
10891               union ctl_io *starting_io)
10892 {
10893         union ctl_io *ooa_io;
10894         ctl_action action;
10895
10896         mtx_assert(&lun->lun_lock, MA_OWNED);
10897
10898         /*
10899          * Run back along the OOA queue, starting with the current
10900          * blocked I/O and going through every I/O before it on the
10901          * queue.  If starting_io is NULL, we'll just end up returning
10902          * CTL_ACTION_PASS.
10903          */
10904         for (ooa_io = starting_io; ooa_io != NULL;
10905              ooa_io = (union ctl_io *)TAILQ_PREV(&ooa_io->io_hdr, ctl_ooaq,
10906              ooa_links)){
10907
10908                 /*
10909                  * This routine just checks to see whether
10910                  * cur_blocked is blocked by ooa_io, which is ahead
10911                  * of it in the queue.  It doesn't queue/dequeue
10912                  * cur_blocked.
10913                  */
10914                 action = ctl_check_for_blockage(lun, pending_io, ooa_io);
10915                 switch (action) {
10916                 case CTL_ACTION_BLOCK:
10917                 case CTL_ACTION_OVERLAP:
10918                 case CTL_ACTION_OVERLAP_TAG:
10919                 case CTL_ACTION_SKIP:
10920                 case CTL_ACTION_ERROR:
10921                         return (action);
10922                         break; /* NOTREACHED */
10923                 case CTL_ACTION_PASS:
10924                         break;
10925                 default:
10926                         panic("%s: Invalid action %d\n", __func__, action);
10927                 }
10928         }
10929
10930         return (CTL_ACTION_PASS);
10931 }
10932
10933 /*
10934  * Assumptions:
10935  * - An I/O has just completed, and has been removed from the per-LUN OOA
10936  *   queue, so some items on the blocked queue may now be unblocked.
10937  */
10938 static int
10939 ctl_check_blocked(struct ctl_lun *lun)
10940 {
10941         struct ctl_softc *softc = lun->ctl_softc;
10942         union ctl_io *cur_blocked, *next_blocked;
10943
10944         mtx_assert(&lun->lun_lock, MA_OWNED);
10945
10946         /*
10947          * Run forward from the head of the blocked queue, checking each
10948          * entry against the I/Os prior to it on the OOA queue to see if
10949          * there is still any blockage.
10950          *
10951          * We cannot use the TAILQ_FOREACH() macro, because it can't deal
10952          * with our removing a variable on it while it is traversing the
10953          * list.
10954          */
10955         for (cur_blocked = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue);
10956              cur_blocked != NULL; cur_blocked = next_blocked) {
10957                 union ctl_io *prev_ooa;
10958                 ctl_action action;
10959
10960                 next_blocked = (union ctl_io *)TAILQ_NEXT(&cur_blocked->io_hdr,
10961                                                           blocked_links);
10962
10963                 prev_ooa = (union ctl_io *)TAILQ_PREV(&cur_blocked->io_hdr,
10964                                                       ctl_ooaq, ooa_links);
10965
10966                 /*
10967                  * If cur_blocked happens to be the first item in the OOA
10968                  * queue now, prev_ooa will be NULL, and the action
10969                  * returned will just be CTL_ACTION_PASS.
10970                  */
10971                 action = ctl_check_ooa(lun, cur_blocked, prev_ooa);
10972
10973                 switch (action) {
10974                 case CTL_ACTION_BLOCK:
10975                         /* Nothing to do here, still blocked */
10976                         break;
10977                 case CTL_ACTION_OVERLAP:
10978                 case CTL_ACTION_OVERLAP_TAG:
10979                         /*
10980                          * This shouldn't happen!  In theory we've already
10981                          * checked this command for overlap...
10982                          */
10983                         break;
10984                 case CTL_ACTION_PASS:
10985                 case CTL_ACTION_SKIP: {
10986                         const struct ctl_cmd_entry *entry;
10987
10988                         /*
10989                          * The skip case shouldn't happen, this transaction
10990                          * should have never made it onto the blocked queue.
10991                          */
10992                         /*
10993                          * This I/O is no longer blocked, we can remove it
10994                          * from the blocked queue.  Since this is a TAILQ
10995                          * (doubly linked list), we can do O(1) removals
10996                          * from any place on the list.
10997                          */
10998                         TAILQ_REMOVE(&lun->blocked_queue, &cur_blocked->io_hdr,
10999                                      blocked_links);
11000                         cur_blocked->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
11001
11002                         if ((softc->ha_mode != CTL_HA_MODE_XFER) &&
11003                             (cur_blocked->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)){
11004                                 /*
11005                                  * Need to send IO back to original side to
11006                                  * run
11007                                  */
11008                                 union ctl_ha_msg msg_info;
11009
11010                                 cur_blocked->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11011                                 msg_info.hdr.original_sc =
11012                                         cur_blocked->io_hdr.remote_io;
11013                                 msg_info.hdr.serializing_sc = cur_blocked;
11014                                 msg_info.hdr.msg_type = CTL_MSG_R2R;
11015                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11016                                     sizeof(msg_info.hdr), M_NOWAIT);
11017                                 break;
11018                         }
11019                         entry = ctl_get_cmd_entry(&cur_blocked->scsiio, NULL);
11020
11021                         /*
11022                          * Check this I/O for LUN state changes that may
11023                          * have happened while this command was blocked.
11024                          * The LUN state may have been changed by a command
11025                          * ahead of us in the queue, so we need to re-check
11026                          * for any states that can be caused by SCSI
11027                          * commands.
11028                          */
11029                         if (ctl_scsiio_lun_check(lun, entry,
11030                                                  &cur_blocked->scsiio) == 0) {
11031                                 cur_blocked->io_hdr.flags |=
11032                                                       CTL_FLAG_IS_WAS_ON_RTR;
11033                                 ctl_enqueue_rtr(cur_blocked);
11034                         } else
11035                                 ctl_done(cur_blocked);
11036                         break;
11037                 }
11038                 default:
11039                         /*
11040                          * This probably shouldn't happen -- we shouldn't
11041                          * get CTL_ACTION_ERROR, or anything else.
11042                          */
11043                         break;
11044                 }
11045         }
11046
11047         return (CTL_RETVAL_COMPLETE);
11048 }
11049
11050 /*
11051  * This routine (with one exception) checks LUN flags that can be set by
11052  * commands ahead of us in the OOA queue.  These flags have to be checked
11053  * when a command initially comes in, and when we pull a command off the
11054  * blocked queue and are preparing to execute it.  The reason we have to
11055  * check these flags for commands on the blocked queue is that the LUN
11056  * state may have been changed by a command ahead of us while we're on the
11057  * blocked queue.
11058  *
11059  * Ordering is somewhat important with these checks, so please pay
11060  * careful attention to the placement of any new checks.
11061  */
11062 static int
11063 ctl_scsiio_lun_check(struct ctl_lun *lun,
11064     const struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
11065 {
11066         struct ctl_softc *softc = lun->ctl_softc;
11067         int retval;
11068         uint32_t residx;
11069
11070         retval = 0;
11071
11072         mtx_assert(&lun->lun_lock, MA_OWNED);
11073
11074         /*
11075          * If this shelf is a secondary shelf controller, we may have to
11076          * reject some commands disallowed by HA mode and link state.
11077          */
11078         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
11079                 if (softc->ha_link == CTL_HA_LINK_OFFLINE &&
11080                     (entry->flags & CTL_CMD_FLAG_OK_ON_UNAVAIL) == 0) {
11081                         ctl_set_lun_unavail(ctsio);
11082                         retval = 1;
11083                         goto bailout;
11084                 }
11085                 if ((lun->flags & CTL_LUN_PEER_SC_PRIMARY) == 0 &&
11086                     (entry->flags & CTL_CMD_FLAG_OK_ON_UNAVAIL) == 0) {
11087                         ctl_set_lun_transit(ctsio);
11088                         retval = 1;
11089                         goto bailout;
11090                 }
11091                 if (softc->ha_mode == CTL_HA_MODE_ACT_STBY &&
11092                     (entry->flags & CTL_CMD_FLAG_OK_ON_STANDBY) == 0) {
11093                         ctl_set_lun_standby(ctsio);
11094                         retval = 1;
11095                         goto bailout;
11096                 }
11097
11098                 /* The rest of checks are only done on executing side */
11099                 if (softc->ha_mode == CTL_HA_MODE_XFER)
11100                         goto bailout;
11101         }
11102
11103         if (entry->pattern & CTL_LUN_PAT_WRITE) {
11104                 if (lun->be_lun &&
11105                     lun->be_lun->flags & CTL_LUN_FLAG_READONLY) {
11106                         ctl_set_hw_write_protected(ctsio);
11107                         retval = 1;
11108                         goto bailout;
11109                 }
11110                 if ((lun->MODE_CTRL.eca_and_aen & SCP_SWP) != 0) {
11111                         ctl_set_sense(ctsio, /*current_error*/ 1,
11112                             /*sense_key*/ SSD_KEY_DATA_PROTECT,
11113                             /*asc*/ 0x27, /*ascq*/ 0x02, SSD_ELEM_NONE);
11114                         retval = 1;
11115                         goto bailout;
11116                 }
11117         }
11118
11119         /*
11120          * Check for a reservation conflict.  If this command isn't allowed
11121          * even on reserved LUNs, and if this initiator isn't the one who
11122          * reserved us, reject the command with a reservation conflict.
11123          */
11124         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11125         if ((lun->flags & CTL_LUN_RESERVED)
11126          && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
11127                 if (lun->res_idx != residx) {
11128                         ctl_set_reservation_conflict(ctsio);
11129                         retval = 1;
11130                         goto bailout;
11131                 }
11132         }
11133
11134         if ((lun->flags & CTL_LUN_PR_RESERVED) == 0 ||
11135             (entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV)) {
11136                 /* No reservation or command is allowed. */;
11137         } else if ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_WRESV) &&
11138             (lun->pr_res_type == SPR_TYPE_WR_EX ||
11139              lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
11140              lun->pr_res_type == SPR_TYPE_WR_EX_AR)) {
11141                 /* The command is allowed for Write Exclusive resv. */;
11142         } else {
11143                 /*
11144                  * if we aren't registered or it's a res holder type
11145                  * reservation and this isn't the res holder then set a
11146                  * conflict.
11147                  */
11148                 if (ctl_get_prkey(lun, residx) == 0 ||
11149                     (residx != lun->pr_res_idx && lun->pr_res_type < 4)) {
11150                         ctl_set_reservation_conflict(ctsio);
11151                         retval = 1;
11152                         goto bailout;
11153                 }
11154         }
11155
11156         if ((entry->flags & CTL_CMD_FLAG_OK_ON_NO_MEDIA) == 0) {
11157                 if (lun->flags & CTL_LUN_EJECTED)
11158                         ctl_set_lun_ejected(ctsio);
11159                 else if (lun->flags & CTL_LUN_NO_MEDIA) {
11160                         if (lun->flags & CTL_LUN_REMOVABLE)
11161                                 ctl_set_lun_no_media(ctsio);
11162                         else
11163                                 ctl_set_lun_int_reqd(ctsio);
11164                 } else if (lun->flags & CTL_LUN_STOPPED)
11165                         ctl_set_lun_stopped(ctsio);
11166                 else
11167                         goto bailout;
11168                 retval = 1;
11169                 goto bailout;
11170         }
11171
11172 bailout:
11173         return (retval);
11174 }
11175
11176 static void
11177 ctl_failover_io(union ctl_io *io, int have_lock)
11178 {
11179         ctl_set_busy(&io->scsiio);
11180         ctl_done(io);
11181 }
11182
11183 static void
11184 ctl_failover_lun(union ctl_io *rio)
11185 {
11186         struct ctl_softc *softc = CTL_SOFTC(rio);
11187         struct ctl_lun *lun;
11188         struct ctl_io_hdr *io, *next_io;
11189         uint32_t targ_lun;
11190
11191         targ_lun = rio->io_hdr.nexus.targ_mapped_lun;
11192         CTL_DEBUG_PRINT(("FAILOVER for lun %ju\n", targ_lun));
11193
11194         /* Find and lock the LUN. */
11195         mtx_lock(&softc->ctl_lock);
11196         if (targ_lun > ctl_max_luns ||
11197             (lun = softc->ctl_luns[targ_lun]) == NULL) {
11198                 mtx_unlock(&softc->ctl_lock);
11199                 return;
11200         }
11201         mtx_lock(&lun->lun_lock);
11202         mtx_unlock(&softc->ctl_lock);
11203         if (lun->flags & CTL_LUN_DISABLED) {
11204                 mtx_unlock(&lun->lun_lock);
11205                 return;
11206         }
11207
11208         if (softc->ha_mode == CTL_HA_MODE_XFER) {
11209                 TAILQ_FOREACH_SAFE(io, &lun->ooa_queue, ooa_links, next_io) {
11210                         /* We are master */
11211                         if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
11212                                 if (io->flags & CTL_FLAG_IO_ACTIVE) {
11213                                         io->flags |= CTL_FLAG_ABORT;
11214                                         io->flags |= CTL_FLAG_FAILOVER;
11215                                 } else { /* This can be only due to DATAMOVE */
11216                                         io->msg_type = CTL_MSG_DATAMOVE_DONE;
11217                                         io->flags &= ~CTL_FLAG_DMA_INPROG;
11218                                         io->flags |= CTL_FLAG_IO_ACTIVE;
11219                                         io->port_status = 31340;
11220                                         ctl_enqueue_isc((union ctl_io *)io);
11221                                 }
11222                         }
11223                         /* We are slave */
11224                         if (io->flags & CTL_FLAG_SENT_2OTHER_SC) {
11225                                 io->flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11226                                 if (io->flags & CTL_FLAG_IO_ACTIVE) {
11227                                         io->flags |= CTL_FLAG_FAILOVER;
11228                                 } else {
11229                                         ctl_set_busy(&((union ctl_io *)io)->
11230                                             scsiio);
11231                                         ctl_done((union ctl_io *)io);
11232                                 }
11233                         }
11234                 }
11235         } else { /* SERIALIZE modes */
11236                 TAILQ_FOREACH_SAFE(io, &lun->blocked_queue, blocked_links,
11237                     next_io) {
11238                         /* We are master */
11239                         if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
11240                                 TAILQ_REMOVE(&lun->blocked_queue, io,
11241                                     blocked_links);
11242                                 io->flags &= ~CTL_FLAG_BLOCKED;
11243                                 TAILQ_REMOVE(&lun->ooa_queue, io, ooa_links);
11244                                 ctl_free_io((union ctl_io *)io);
11245                         }
11246                 }
11247                 TAILQ_FOREACH_SAFE(io, &lun->ooa_queue, ooa_links, next_io) {
11248                         /* We are master */
11249                         if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
11250                                 TAILQ_REMOVE(&lun->ooa_queue, io, ooa_links);
11251                                 ctl_free_io((union ctl_io *)io);
11252                         }
11253                         /* We are slave */
11254                         if (io->flags & CTL_FLAG_SENT_2OTHER_SC) {
11255                                 io->flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11256                                 if (!(io->flags & CTL_FLAG_IO_ACTIVE)) {
11257                                         ctl_set_busy(&((union ctl_io *)io)->
11258                                             scsiio);
11259                                         ctl_done((union ctl_io *)io);
11260                                 }
11261                         }
11262                 }
11263                 ctl_check_blocked(lun);
11264         }
11265         mtx_unlock(&lun->lun_lock);
11266 }
11267
11268 static int
11269 ctl_scsiio_precheck(struct ctl_softc *softc, struct ctl_scsiio *ctsio)
11270 {
11271         struct ctl_lun *lun;
11272         const struct ctl_cmd_entry *entry;
11273         uint32_t initidx, targ_lun;
11274         int retval = 0;
11275
11276         lun = NULL;
11277         targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
11278         if (targ_lun < ctl_max_luns)
11279                 lun = softc->ctl_luns[targ_lun];
11280         if (lun) {
11281                 /*
11282                  * If the LUN is invalid, pretend that it doesn't exist.
11283                  * It will go away as soon as all pending I/O has been
11284                  * completed.
11285                  */
11286                 mtx_lock(&lun->lun_lock);
11287                 if (lun->flags & CTL_LUN_DISABLED) {
11288                         mtx_unlock(&lun->lun_lock);
11289                         lun = NULL;
11290                 }
11291         }
11292         CTL_LUN(ctsio) = lun;
11293         if (lun) {
11294                 CTL_BACKEND_LUN(ctsio) = lun->be_lun;
11295
11296                 /*
11297                  * Every I/O goes into the OOA queue for a particular LUN,
11298                  * and stays there until completion.
11299                  */
11300 #ifdef CTL_TIME_IO
11301                 if (TAILQ_EMPTY(&lun->ooa_queue))
11302                         lun->idle_time += getsbinuptime() - lun->last_busy;
11303 #endif
11304                 TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
11305         }
11306
11307         /* Get command entry and return error if it is unsuppotyed. */
11308         entry = ctl_validate_command(ctsio);
11309         if (entry == NULL) {
11310                 if (lun)
11311                         mtx_unlock(&lun->lun_lock);
11312                 return (retval);
11313         }
11314
11315         ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
11316         ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
11317
11318         /*
11319          * Check to see whether we can send this command to LUNs that don't
11320          * exist.  This should pretty much only be the case for inquiry
11321          * and request sense.  Further checks, below, really require having
11322          * a LUN, so we can't really check the command anymore.  Just put
11323          * it on the rtr queue.
11324          */
11325         if (lun == NULL) {
11326                 if (entry->flags & CTL_CMD_FLAG_OK_ON_NO_LUN) {
11327                         ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11328                         ctl_enqueue_rtr((union ctl_io *)ctsio);
11329                         return (retval);
11330                 }
11331
11332                 ctl_set_unsupported_lun(ctsio);
11333                 ctl_done((union ctl_io *)ctsio);
11334                 CTL_DEBUG_PRINT(("ctl_scsiio_precheck: bailing out due to invalid LUN\n"));
11335                 return (retval);
11336         } else {
11337                 /*
11338                  * Make sure we support this particular command on this LUN.
11339                  * e.g., we don't support writes to the control LUN.
11340                  */
11341                 if (!ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
11342                         mtx_unlock(&lun->lun_lock);
11343                         ctl_set_invalid_opcode(ctsio);
11344                         ctl_done((union ctl_io *)ctsio);
11345                         return (retval);
11346                 }
11347         }
11348
11349         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11350
11351         /*
11352          * If we've got a request sense, it'll clear the contingent
11353          * allegiance condition.  Otherwise, if we have a CA condition for
11354          * this initiator, clear it, because it sent down a command other
11355          * than request sense.
11356          */
11357         if (ctsio->cdb[0] != REQUEST_SENSE) {
11358                 struct scsi_sense_data *ps;
11359
11360                 ps = lun->pending_sense[initidx / CTL_MAX_INIT_PER_PORT];
11361                 if (ps != NULL)
11362                         ps[initidx % CTL_MAX_INIT_PER_PORT].error_code = 0;
11363         }
11364
11365         /*
11366          * If the command has this flag set, it handles its own unit
11367          * attention reporting, we shouldn't do anything.  Otherwise we
11368          * check for any pending unit attentions, and send them back to the
11369          * initiator.  We only do this when a command initially comes in,
11370          * not when we pull it off the blocked queue.
11371          *
11372          * According to SAM-3, section 5.3.2, the order that things get
11373          * presented back to the host is basically unit attentions caused
11374          * by some sort of reset event, busy status, reservation conflicts
11375          * or task set full, and finally any other status.
11376          *
11377          * One issue here is that some of the unit attentions we report
11378          * don't fall into the "reset" category (e.g. "reported luns data
11379          * has changed").  So reporting it here, before the reservation
11380          * check, may be technically wrong.  I guess the only thing to do
11381          * would be to check for and report the reset events here, and then
11382          * check for the other unit attention types after we check for a
11383          * reservation conflict.
11384          *
11385          * XXX KDM need to fix this
11386          */
11387         if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
11388                 ctl_ua_type ua_type;
11389                 u_int sense_len = 0;
11390
11391                 ua_type = ctl_build_ua(lun, initidx, &ctsio->sense_data,
11392                     &sense_len, SSD_TYPE_NONE);
11393                 if (ua_type != CTL_UA_NONE) {
11394                         mtx_unlock(&lun->lun_lock);
11395                         ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
11396                         ctsio->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
11397                         ctsio->sense_len = sense_len;
11398                         ctl_done((union ctl_io *)ctsio);
11399                         return (retval);
11400                 }
11401         }
11402
11403
11404         if (ctl_scsiio_lun_check(lun, entry, ctsio) != 0) {
11405                 mtx_unlock(&lun->lun_lock);
11406                 ctl_done((union ctl_io *)ctsio);
11407                 return (retval);
11408         }
11409
11410         /*
11411          * XXX CHD this is where we want to send IO to other side if
11412          * this LUN is secondary on this SC. We will need to make a copy
11413          * of the IO and flag the IO on this side as SENT_2OTHER and the flag
11414          * the copy we send as FROM_OTHER.
11415          * We also need to stuff the address of the original IO so we can
11416          * find it easily. Something similar will need be done on the other
11417          * side so when we are done we can find the copy.
11418          */
11419         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
11420             (lun->flags & CTL_LUN_PEER_SC_PRIMARY) != 0 &&
11421             (entry->flags & CTL_CMD_FLAG_RUN_HERE) == 0) {
11422                 union ctl_ha_msg msg_info;
11423                 int isc_retval;
11424
11425                 ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11426                 ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11427                 mtx_unlock(&lun->lun_lock);
11428
11429                 msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
11430                 msg_info.hdr.original_sc = (union ctl_io *)ctsio;
11431                 msg_info.hdr.serializing_sc = NULL;
11432                 msg_info.hdr.nexus = ctsio->io_hdr.nexus;
11433                 msg_info.scsi.tag_num = ctsio->tag_num;
11434                 msg_info.scsi.tag_type = ctsio->tag_type;
11435                 msg_info.scsi.cdb_len = ctsio->cdb_len;
11436                 memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
11437
11438                 if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11439                     sizeof(msg_info.scsi) - sizeof(msg_info.scsi.sense_data),
11440                     M_WAITOK)) > CTL_HA_STATUS_SUCCESS) {
11441                         ctl_set_busy(ctsio);
11442                         ctl_done((union ctl_io *)ctsio);
11443                         return (retval);
11444                 }
11445                 return (retval);
11446         }
11447
11448         switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
11449                               (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr,
11450                               ctl_ooaq, ooa_links))) {
11451         case CTL_ACTION_BLOCK:
11452                 ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
11453                 TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
11454                                   blocked_links);
11455                 mtx_unlock(&lun->lun_lock);
11456                 return (retval);
11457         case CTL_ACTION_PASS:
11458         case CTL_ACTION_SKIP:
11459                 ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11460                 mtx_unlock(&lun->lun_lock);
11461                 ctl_enqueue_rtr((union ctl_io *)ctsio);
11462                 break;
11463         case CTL_ACTION_OVERLAP:
11464                 mtx_unlock(&lun->lun_lock);
11465                 ctl_set_overlapped_cmd(ctsio);
11466                 ctl_done((union ctl_io *)ctsio);
11467                 break;
11468         case CTL_ACTION_OVERLAP_TAG:
11469                 mtx_unlock(&lun->lun_lock);
11470                 ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
11471                 ctl_done((union ctl_io *)ctsio);
11472                 break;
11473         case CTL_ACTION_ERROR:
11474         default:
11475                 mtx_unlock(&lun->lun_lock);
11476                 ctl_set_internal_failure(ctsio,
11477                                          /*sks_valid*/ 0,
11478                                          /*retry_count*/ 0);
11479                 ctl_done((union ctl_io *)ctsio);
11480                 break;
11481         }
11482         return (retval);
11483 }
11484
11485 const struct ctl_cmd_entry *
11486 ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa)
11487 {
11488         const struct ctl_cmd_entry *entry;
11489         int service_action;
11490
11491         entry = &ctl_cmd_table[ctsio->cdb[0]];
11492         if (sa)
11493                 *sa = ((entry->flags & CTL_CMD_FLAG_SA5) != 0);
11494         if (entry->flags & CTL_CMD_FLAG_SA5) {
11495                 service_action = ctsio->cdb[1] & SERVICE_ACTION_MASK;
11496                 entry = &((const struct ctl_cmd_entry *)
11497                     entry->execute)[service_action];
11498         }
11499         return (entry);
11500 }
11501
11502 const struct ctl_cmd_entry *
11503 ctl_validate_command(struct ctl_scsiio *ctsio)
11504 {
11505         const struct ctl_cmd_entry *entry;
11506         int i, sa;
11507         uint8_t diff;
11508
11509         entry = ctl_get_cmd_entry(ctsio, &sa);
11510         if (entry->execute == NULL) {
11511                 if (sa)
11512                         ctl_set_invalid_field(ctsio,
11513                                               /*sks_valid*/ 1,
11514                                               /*command*/ 1,
11515                                               /*field*/ 1,
11516                                               /*bit_valid*/ 1,
11517                                               /*bit*/ 4);
11518                 else
11519                         ctl_set_invalid_opcode(ctsio);
11520                 ctl_done((union ctl_io *)ctsio);
11521                 return (NULL);
11522         }
11523         KASSERT(entry->length > 0,
11524             ("Not defined length for command 0x%02x/0x%02x",
11525              ctsio->cdb[0], ctsio->cdb[1]));
11526         for (i = 1; i < entry->length; i++) {
11527                 diff = ctsio->cdb[i] & ~entry->usage[i - 1];
11528                 if (diff == 0)
11529                         continue;
11530                 ctl_set_invalid_field(ctsio,
11531                                       /*sks_valid*/ 1,
11532                                       /*command*/ 1,
11533                                       /*field*/ i,
11534                                       /*bit_valid*/ 1,
11535                                       /*bit*/ fls(diff) - 1);
11536                 ctl_done((union ctl_io *)ctsio);
11537                 return (NULL);
11538         }
11539         return (entry);
11540 }
11541
11542 static int
11543 ctl_cmd_applicable(uint8_t lun_type, const struct ctl_cmd_entry *entry)
11544 {
11545
11546         switch (lun_type) {
11547         case T_DIRECT:
11548                 if ((entry->flags & CTL_CMD_FLAG_OK_ON_DIRECT) == 0)
11549                         return (0);
11550                 break;
11551         case T_PROCESSOR:
11552                 if ((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0)
11553                         return (0);
11554                 break;
11555         case T_CDROM:
11556                 if ((entry->flags & CTL_CMD_FLAG_OK_ON_CDROM) == 0)
11557                         return (0);
11558                 break;
11559         default:
11560                 return (0);
11561         }
11562         return (1);
11563 }
11564
11565 static int
11566 ctl_scsiio(struct ctl_scsiio *ctsio)
11567 {
11568         int retval;
11569         const struct ctl_cmd_entry *entry;
11570
11571         retval = CTL_RETVAL_COMPLETE;
11572
11573         CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
11574
11575         entry = ctl_get_cmd_entry(ctsio, NULL);
11576
11577         /*
11578          * If this I/O has been aborted, just send it straight to
11579          * ctl_done() without executing it.
11580          */
11581         if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
11582                 ctl_done((union ctl_io *)ctsio);
11583                 goto bailout;
11584         }
11585
11586         /*
11587          * All the checks should have been handled by ctl_scsiio_precheck().
11588          * We should be clear now to just execute the I/O.
11589          */
11590         retval = entry->execute(ctsio);
11591
11592 bailout:
11593         return (retval);
11594 }
11595
11596 static int
11597 ctl_target_reset(union ctl_io *io)
11598 {
11599         struct ctl_softc *softc = CTL_SOFTC(io);
11600         struct ctl_port *port = CTL_PORT(io);
11601         struct ctl_lun *lun;
11602         uint32_t initidx;
11603         ctl_ua_type ua_type;
11604
11605         if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11606                 union ctl_ha_msg msg_info;
11607
11608                 msg_info.hdr.nexus = io->io_hdr.nexus;
11609                 msg_info.task.task_action = io->taskio.task_action;
11610                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11611                 msg_info.hdr.original_sc = NULL;
11612                 msg_info.hdr.serializing_sc = NULL;
11613                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11614                     sizeof(msg_info.task), M_WAITOK);
11615         }
11616
11617         initidx = ctl_get_initindex(&io->io_hdr.nexus);
11618         if (io->taskio.task_action == CTL_TASK_TARGET_RESET)
11619                 ua_type = CTL_UA_TARG_RESET;
11620         else
11621                 ua_type = CTL_UA_BUS_RESET;
11622         mtx_lock(&softc->ctl_lock);
11623         STAILQ_FOREACH(lun, &softc->lun_list, links) {
11624                 if (port != NULL &&
11625                     ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
11626                         continue;
11627                 ctl_do_lun_reset(lun, initidx, ua_type);
11628         }
11629         mtx_unlock(&softc->ctl_lock);
11630         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11631         return (0);
11632 }
11633
11634 /*
11635  * The LUN should always be set.  The I/O is optional, and is used to
11636  * distinguish between I/Os sent by this initiator, and by other
11637  * initiators.  We set unit attention for initiators other than this one.
11638  * SAM-3 is vague on this point.  It does say that a unit attention should
11639  * be established for other initiators when a LUN is reset (see section
11640  * 5.7.3), but it doesn't specifically say that the unit attention should
11641  * be established for this particular initiator when a LUN is reset.  Here
11642  * is the relevant text, from SAM-3 rev 8:
11643  *
11644  * 5.7.2 When a SCSI initiator port aborts its own tasks
11645  *
11646  * When a SCSI initiator port causes its own task(s) to be aborted, no
11647  * notification that the task(s) have been aborted shall be returned to
11648  * the SCSI initiator port other than the completion response for the
11649  * command or task management function action that caused the task(s) to
11650  * be aborted and notification(s) associated with related effects of the
11651  * action (e.g., a reset unit attention condition).
11652  *
11653  * XXX KDM for now, we're setting unit attention for all initiators.
11654  */
11655 static void
11656 ctl_do_lun_reset(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua_type)
11657 {
11658         union ctl_io *xio;
11659         int i;
11660
11661         mtx_lock(&lun->lun_lock);
11662         /* Abort tasks. */
11663         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11664              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11665                 xio->io_hdr.flags |= CTL_FLAG_ABORT | CTL_FLAG_ABORT_STATUS;
11666         }
11667         /* Clear CA. */
11668         for (i = 0; i < ctl_max_ports; i++) {
11669                 free(lun->pending_sense[i], M_CTL);
11670                 lun->pending_sense[i] = NULL;
11671         }
11672         /* Clear reservation. */
11673         lun->flags &= ~CTL_LUN_RESERVED;
11674         /* Clear prevent media removal. */
11675         if (lun->prevent) {
11676                 for (i = 0; i < CTL_MAX_INITIATORS; i++)
11677                         ctl_clear_mask(lun->prevent, i);
11678                 lun->prevent_count = 0;
11679         }
11680         /* Clear TPC status */
11681         ctl_tpc_lun_clear(lun, -1);
11682         /* Establish UA. */
11683 #if 0
11684         ctl_est_ua_all(lun, initidx, ua_type);
11685 #else
11686         ctl_est_ua_all(lun, -1, ua_type);
11687 #endif
11688         mtx_unlock(&lun->lun_lock);
11689 }
11690
11691 static int
11692 ctl_lun_reset(union ctl_io *io)
11693 {
11694         struct ctl_softc *softc = CTL_SOFTC(io);
11695         struct ctl_lun *lun;
11696         uint32_t targ_lun, initidx;
11697
11698         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11699         initidx = ctl_get_initindex(&io->io_hdr.nexus);
11700         mtx_lock(&softc->ctl_lock);
11701         if (targ_lun >= ctl_max_luns ||
11702             (lun = softc->ctl_luns[targ_lun]) == NULL) {
11703                 mtx_unlock(&softc->ctl_lock);
11704                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
11705                 return (1);
11706         }
11707         ctl_do_lun_reset(lun, initidx, CTL_UA_LUN_RESET);
11708         mtx_unlock(&softc->ctl_lock);
11709         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11710
11711         if ((io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) == 0) {
11712                 union ctl_ha_msg msg_info;
11713
11714                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11715                 msg_info.hdr.nexus = io->io_hdr.nexus;
11716                 msg_info.task.task_action = CTL_TASK_LUN_RESET;
11717                 msg_info.hdr.original_sc = NULL;
11718                 msg_info.hdr.serializing_sc = NULL;
11719                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11720                     sizeof(msg_info.task), M_WAITOK);
11721         }
11722         return (0);
11723 }
11724
11725 static void
11726 ctl_abort_tasks_lun(struct ctl_lun *lun, uint32_t targ_port, uint32_t init_id,
11727     int other_sc)
11728 {
11729         union ctl_io *xio;
11730
11731         mtx_assert(&lun->lun_lock, MA_OWNED);
11732
11733         /*
11734          * Run through the OOA queue and attempt to find the given I/O.
11735          * The target port, initiator ID, tag type and tag number have to
11736          * match the values that we got from the initiator.  If we have an
11737          * untagged command to abort, simply abort the first untagged command
11738          * we come to.  We only allow one untagged command at a time of course.
11739          */
11740         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11741              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11742
11743                 if ((targ_port == UINT32_MAX ||
11744                      targ_port == xio->io_hdr.nexus.targ_port) &&
11745                     (init_id == UINT32_MAX ||
11746                      init_id == xio->io_hdr.nexus.initid)) {
11747                         if (targ_port != xio->io_hdr.nexus.targ_port ||
11748                             init_id != xio->io_hdr.nexus.initid)
11749                                 xio->io_hdr.flags |= CTL_FLAG_ABORT_STATUS;
11750                         xio->io_hdr.flags |= CTL_FLAG_ABORT;
11751                         if (!other_sc && !(lun->flags & CTL_LUN_PRIMARY_SC)) {
11752                                 union ctl_ha_msg msg_info;
11753
11754                                 msg_info.hdr.nexus = xio->io_hdr.nexus;
11755                                 msg_info.task.task_action = CTL_TASK_ABORT_TASK;
11756                                 msg_info.task.tag_num = xio->scsiio.tag_num;
11757                                 msg_info.task.tag_type = xio->scsiio.tag_type;
11758                                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11759                                 msg_info.hdr.original_sc = NULL;
11760                                 msg_info.hdr.serializing_sc = NULL;
11761                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11762                                     sizeof(msg_info.task), M_NOWAIT);
11763                         }
11764                 }
11765         }
11766 }
11767
11768 static int
11769 ctl_abort_task_set(union ctl_io *io)
11770 {
11771         struct ctl_softc *softc = CTL_SOFTC(io);
11772         struct ctl_lun *lun;
11773         uint32_t targ_lun;
11774
11775         /*
11776          * Look up the LUN.
11777          */
11778         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11779         mtx_lock(&softc->ctl_lock);
11780         if (targ_lun >= ctl_max_luns ||
11781             (lun = softc->ctl_luns[targ_lun]) == NULL) {
11782                 mtx_unlock(&softc->ctl_lock);
11783                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
11784                 return (1);
11785         }
11786
11787         mtx_lock(&lun->lun_lock);
11788         mtx_unlock(&softc->ctl_lock);
11789         if (io->taskio.task_action == CTL_TASK_ABORT_TASK_SET) {
11790                 ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
11791                     io->io_hdr.nexus.initid,
11792                     (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
11793         } else { /* CTL_TASK_CLEAR_TASK_SET */
11794                 ctl_abort_tasks_lun(lun, UINT32_MAX, UINT32_MAX,
11795                     (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
11796         }
11797         mtx_unlock(&lun->lun_lock);
11798         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11799         return (0);
11800 }
11801
11802 static void
11803 ctl_i_t_nexus_loss(struct ctl_softc *softc, uint32_t initidx,
11804     ctl_ua_type ua_type)
11805 {
11806         struct ctl_lun *lun;
11807         struct scsi_sense_data *ps;
11808         uint32_t p, i;
11809
11810         p = initidx / CTL_MAX_INIT_PER_PORT;
11811         i = initidx % CTL_MAX_INIT_PER_PORT;
11812         mtx_lock(&softc->ctl_lock);
11813         STAILQ_FOREACH(lun, &softc->lun_list, links) {
11814                 mtx_lock(&lun->lun_lock);
11815                 /* Abort tasks. */
11816                 ctl_abort_tasks_lun(lun, p, i, 1);
11817                 /* Clear CA. */
11818                 ps = lun->pending_sense[p];
11819                 if (ps != NULL)
11820                         ps[i].error_code = 0;
11821                 /* Clear reservation. */
11822                 if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == initidx))
11823                         lun->flags &= ~CTL_LUN_RESERVED;
11824                 /* Clear prevent media removal. */
11825                 if (lun->prevent && ctl_is_set(lun->prevent, initidx)) {
11826                         ctl_clear_mask(lun->prevent, initidx);
11827                         lun->prevent_count--;
11828                 }
11829                 /* Clear TPC status */
11830                 ctl_tpc_lun_clear(lun, initidx);
11831                 /* Establish UA. */
11832                 ctl_est_ua(lun, initidx, ua_type);
11833                 mtx_unlock(&lun->lun_lock);
11834         }
11835         mtx_unlock(&softc->ctl_lock);
11836 }
11837
11838 static int
11839 ctl_i_t_nexus_reset(union ctl_io *io)
11840 {
11841         struct ctl_softc *softc = CTL_SOFTC(io);
11842         uint32_t initidx;
11843
11844         if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11845                 union ctl_ha_msg msg_info;
11846
11847                 msg_info.hdr.nexus = io->io_hdr.nexus;
11848                 msg_info.task.task_action = CTL_TASK_I_T_NEXUS_RESET;
11849                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11850                 msg_info.hdr.original_sc = NULL;
11851                 msg_info.hdr.serializing_sc = NULL;
11852                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11853                     sizeof(msg_info.task), M_WAITOK);
11854         }
11855
11856         initidx = ctl_get_initindex(&io->io_hdr.nexus);
11857         ctl_i_t_nexus_loss(softc, initidx, CTL_UA_I_T_NEXUS_LOSS);
11858         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11859         return (0);
11860 }
11861
11862 static int
11863 ctl_abort_task(union ctl_io *io)
11864 {
11865         struct ctl_softc *softc = CTL_SOFTC(io);
11866         union ctl_io *xio;
11867         struct ctl_lun *lun;
11868         uint32_t targ_lun;
11869
11870         /*
11871          * Look up the LUN.
11872          */
11873         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11874         mtx_lock(&softc->ctl_lock);
11875         if (targ_lun >= ctl_max_luns ||
11876             (lun = softc->ctl_luns[targ_lun]) == NULL) {
11877                 mtx_unlock(&softc->ctl_lock);
11878                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
11879                 return (1);
11880         }
11881
11882         mtx_lock(&lun->lun_lock);
11883         mtx_unlock(&softc->ctl_lock);
11884         /*
11885          * Run through the OOA queue and attempt to find the given I/O.
11886          * The target port, initiator ID, tag type and tag number have to
11887          * match the values that we got from the initiator.  If we have an
11888          * untagged command to abort, simply abort the first untagged command
11889          * we come to.  We only allow one untagged command at a time of course.
11890          */
11891         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11892              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11893
11894                 if ((xio->io_hdr.nexus.targ_port != io->io_hdr.nexus.targ_port)
11895                  || (xio->io_hdr.nexus.initid != io->io_hdr.nexus.initid)
11896                  || (xio->io_hdr.flags & CTL_FLAG_ABORT))
11897                         continue;
11898
11899                 /*
11900                  * If the abort says that the task is untagged, the
11901                  * task in the queue must be untagged.  Otherwise,
11902                  * we just check to see whether the tag numbers
11903                  * match.  This is because the QLogic firmware
11904                  * doesn't pass back the tag type in an abort
11905                  * request.
11906                  */
11907 #if 0
11908                 if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
11909                   && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
11910                  || (xio->scsiio.tag_num == io->taskio.tag_num)) {
11911 #else
11912                 /*
11913                  * XXX KDM we've got problems with FC, because it
11914                  * doesn't send down a tag type with aborts.  So we
11915                  * can only really go by the tag number...
11916                  * This may cause problems with parallel SCSI.
11917                  * Need to figure that out!!
11918                  */
11919                 if (xio->scsiio.tag_num == io->taskio.tag_num) {
11920 #endif
11921                         xio->io_hdr.flags |= CTL_FLAG_ABORT;
11922                         if ((io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) == 0 &&
11923                             !(lun->flags & CTL_LUN_PRIMARY_SC)) {
11924                                 union ctl_ha_msg msg_info;
11925
11926                                 msg_info.hdr.nexus = io->io_hdr.nexus;
11927                                 msg_info.task.task_action = CTL_TASK_ABORT_TASK;
11928                                 msg_info.task.tag_num = io->taskio.tag_num;
11929                                 msg_info.task.tag_type = io->taskio.tag_type;
11930                                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11931                                 msg_info.hdr.original_sc = NULL;
11932                                 msg_info.hdr.serializing_sc = NULL;
11933                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11934                                     sizeof(msg_info.task), M_NOWAIT);
11935                         }
11936                 }
11937         }
11938         mtx_unlock(&lun->lun_lock);
11939         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11940         return (0);
11941 }
11942
11943 static int
11944 ctl_query_task(union ctl_io *io, int task_set)
11945 {
11946         struct ctl_softc *softc = CTL_SOFTC(io);
11947         union ctl_io *xio;
11948         struct ctl_lun *lun;
11949         int found = 0;
11950         uint32_t targ_lun;
11951
11952         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11953         mtx_lock(&softc->ctl_lock);
11954         if (targ_lun >= ctl_max_luns ||
11955             (lun = softc->ctl_luns[targ_lun]) == NULL) {
11956                 mtx_unlock(&softc->ctl_lock);
11957                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
11958                 return (1);
11959         }
11960         mtx_lock(&lun->lun_lock);
11961         mtx_unlock(&softc->ctl_lock);
11962         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11963              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11964
11965                 if ((xio->io_hdr.nexus.targ_port != io->io_hdr.nexus.targ_port)
11966                  || (xio->io_hdr.nexus.initid != io->io_hdr.nexus.initid)
11967                  || (xio->io_hdr.flags & CTL_FLAG_ABORT))
11968                         continue;
11969
11970                 if (task_set || xio->scsiio.tag_num == io->taskio.tag_num) {
11971                         found = 1;
11972                         break;
11973                 }
11974         }
11975         mtx_unlock(&lun->lun_lock);
11976         if (found)
11977                 io->taskio.task_status = CTL_TASK_FUNCTION_SUCCEEDED;
11978         else
11979                 io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11980         return (0);
11981 }
11982
11983 static int
11984 ctl_query_async_event(union ctl_io *io)
11985 {
11986         struct ctl_softc *softc = CTL_SOFTC(io);
11987         struct ctl_lun *lun;
11988         ctl_ua_type ua;
11989         uint32_t targ_lun, initidx;
11990
11991         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11992         mtx_lock(&softc->ctl_lock);
11993         if (targ_lun >= ctl_max_luns ||
11994             (lun = softc->ctl_luns[targ_lun]) == NULL) {
11995                 mtx_unlock(&softc->ctl_lock);
11996                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
11997                 return (1);
11998         }
11999         mtx_lock(&lun->lun_lock);
12000         mtx_unlock(&softc->ctl_lock);
12001         initidx = ctl_get_initindex(&io->io_hdr.nexus);
12002         ua = ctl_build_qae(lun, initidx, io->taskio.task_resp);
12003         mtx_unlock(&lun->lun_lock);
12004         if (ua != CTL_UA_NONE)
12005                 io->taskio.task_status = CTL_TASK_FUNCTION_SUCCEEDED;
12006         else
12007                 io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12008         return (0);
12009 }
12010
12011 static void
12012 ctl_run_task(union ctl_io *io)
12013 {
12014         int retval = 1;
12015
12016         CTL_DEBUG_PRINT(("ctl_run_task\n"));
12017         KASSERT(io->io_hdr.io_type == CTL_IO_TASK,
12018             ("ctl_run_task: Unextected io_type %d\n", io->io_hdr.io_type));
12019         io->taskio.task_status = CTL_TASK_FUNCTION_NOT_SUPPORTED;
12020         bzero(io->taskio.task_resp, sizeof(io->taskio.task_resp));
12021         switch (io->taskio.task_action) {
12022         case CTL_TASK_ABORT_TASK:
12023                 retval = ctl_abort_task(io);
12024                 break;
12025         case CTL_TASK_ABORT_TASK_SET:
12026         case CTL_TASK_CLEAR_TASK_SET:
12027                 retval = ctl_abort_task_set(io);
12028                 break;
12029         case CTL_TASK_CLEAR_ACA:
12030                 break;
12031         case CTL_TASK_I_T_NEXUS_RESET:
12032                 retval = ctl_i_t_nexus_reset(io);
12033                 break;
12034         case CTL_TASK_LUN_RESET:
12035                 retval = ctl_lun_reset(io);
12036                 break;
12037         case CTL_TASK_TARGET_RESET:
12038         case CTL_TASK_BUS_RESET:
12039                 retval = ctl_target_reset(io);
12040                 break;
12041         case CTL_TASK_PORT_LOGIN:
12042                 break;
12043         case CTL_TASK_PORT_LOGOUT:
12044                 break;
12045         case CTL_TASK_QUERY_TASK:
12046                 retval = ctl_query_task(io, 0);
12047                 break;
12048         case CTL_TASK_QUERY_TASK_SET:
12049                 retval = ctl_query_task(io, 1);
12050                 break;
12051         case CTL_TASK_QUERY_ASYNC_EVENT:
12052                 retval = ctl_query_async_event(io);
12053                 break;
12054         default:
12055                 printf("%s: got unknown task management event %d\n",
12056                        __func__, io->taskio.task_action);
12057                 break;
12058         }
12059         if (retval == 0)
12060                 io->io_hdr.status = CTL_SUCCESS;
12061         else
12062                 io->io_hdr.status = CTL_ERROR;
12063         ctl_done(io);
12064 }
12065
12066 /*
12067  * For HA operation.  Handle commands that come in from the other
12068  * controller.
12069  */
12070 static void
12071 ctl_handle_isc(union ctl_io *io)
12072 {
12073         struct ctl_softc *softc = CTL_SOFTC(io);
12074         struct ctl_lun *lun;
12075         const struct ctl_cmd_entry *entry;
12076         uint32_t targ_lun;
12077
12078         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12079         switch (io->io_hdr.msg_type) {
12080         case CTL_MSG_SERIALIZE:
12081                 ctl_serialize_other_sc_cmd(&io->scsiio);
12082                 break;
12083         case CTL_MSG_R2R:               /* Only used in SER_ONLY mode. */
12084                 entry = ctl_get_cmd_entry(&io->scsiio, NULL);
12085                 if (targ_lun >= ctl_max_luns ||
12086                     (lun = softc->ctl_luns[targ_lun]) == NULL) {
12087                         ctl_done(io);
12088                         break;
12089                 }
12090                 mtx_lock(&lun->lun_lock);
12091                 if (ctl_scsiio_lun_check(lun, entry, &io->scsiio) != 0) {
12092                         mtx_unlock(&lun->lun_lock);
12093                         ctl_done(io);
12094                         break;
12095                 }
12096                 io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
12097                 mtx_unlock(&lun->lun_lock);
12098                 ctl_enqueue_rtr(io);
12099                 break;
12100         case CTL_MSG_FINISH_IO:
12101                 if (softc->ha_mode == CTL_HA_MODE_XFER) {
12102                         ctl_done(io);
12103                         break;
12104                 }
12105                 if (targ_lun >= ctl_max_luns ||
12106                     (lun = softc->ctl_luns[targ_lun]) == NULL) {
12107                         ctl_free_io(io);
12108                         break;
12109                 }
12110                 mtx_lock(&lun->lun_lock);
12111                 TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
12112                 ctl_check_blocked(lun);
12113                 mtx_unlock(&lun->lun_lock);
12114                 ctl_free_io(io);
12115                 break;
12116         case CTL_MSG_PERS_ACTION:
12117                 ctl_hndl_per_res_out_on_other_sc(io);
12118                 ctl_free_io(io);
12119                 break;
12120         case CTL_MSG_BAD_JUJU:
12121                 ctl_done(io);
12122                 break;
12123         case CTL_MSG_DATAMOVE:          /* Only used in XFER mode */
12124                 ctl_datamove_remote(io);
12125                 break;
12126         case CTL_MSG_DATAMOVE_DONE:     /* Only used in XFER mode */
12127                 io->scsiio.be_move_done(io);
12128                 break;
12129         case CTL_MSG_FAILOVER:
12130                 ctl_failover_lun(io);
12131                 ctl_free_io(io);
12132                 break;
12133         default:
12134                 printf("%s: Invalid message type %d\n",
12135                        __func__, io->io_hdr.msg_type);
12136                 ctl_free_io(io);
12137                 break;
12138         }
12139
12140 }
12141
12142
12143 /*
12144  * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
12145  * there is no match.
12146  */
12147 static ctl_lun_error_pattern
12148 ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
12149 {
12150         const struct ctl_cmd_entry *entry;
12151         ctl_lun_error_pattern filtered_pattern, pattern;
12152
12153         pattern = desc->error_pattern;
12154
12155         /*
12156          * XXX KDM we need more data passed into this function to match a
12157          * custom pattern, and we actually need to implement custom pattern
12158          * matching.
12159          */
12160         if (pattern & CTL_LUN_PAT_CMD)
12161                 return (CTL_LUN_PAT_CMD);
12162
12163         if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
12164                 return (CTL_LUN_PAT_ANY);
12165
12166         entry = ctl_get_cmd_entry(ctsio, NULL);
12167
12168         filtered_pattern = entry->pattern & pattern;
12169
12170         /*
12171          * If the user requested specific flags in the pattern (e.g.
12172          * CTL_LUN_PAT_RANGE), make sure the command supports all of those
12173          * flags.
12174          *
12175          * If the user did not specify any flags, it doesn't matter whether
12176          * or not the command supports the flags.
12177          */
12178         if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
12179              (pattern & ~CTL_LUN_PAT_MASK))
12180                 return (CTL_LUN_PAT_NONE);
12181
12182         /*
12183          * If the user asked for a range check, see if the requested LBA
12184          * range overlaps with this command's LBA range.
12185          */
12186         if (filtered_pattern & CTL_LUN_PAT_RANGE) {
12187                 uint64_t lba1;
12188                 uint64_t len1;
12189                 ctl_action action;
12190                 int retval;
12191
12192                 retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
12193                 if (retval != 0)
12194                         return (CTL_LUN_PAT_NONE);
12195
12196                 action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
12197                                               desc->lba_range.len, FALSE);
12198                 /*
12199                  * A "pass" means that the LBA ranges don't overlap, so
12200                  * this doesn't match the user's range criteria.
12201                  */
12202                 if (action == CTL_ACTION_PASS)
12203                         return (CTL_LUN_PAT_NONE);
12204         }
12205
12206         return (filtered_pattern);
12207 }
12208
12209 static void
12210 ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
12211 {
12212         struct ctl_error_desc *desc, *desc2;
12213
12214         mtx_assert(&lun->lun_lock, MA_OWNED);
12215
12216         STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
12217                 ctl_lun_error_pattern pattern;
12218                 /*
12219                  * Check to see whether this particular command matches
12220                  * the pattern in the descriptor.
12221                  */
12222                 pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
12223                 if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
12224                         continue;
12225
12226                 switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
12227                 case CTL_LUN_INJ_ABORTED:
12228                         ctl_set_aborted(&io->scsiio);
12229                         break;
12230                 case CTL_LUN_INJ_MEDIUM_ERR:
12231                         ctl_set_medium_error(&io->scsiio,
12232                             (io->io_hdr.flags & CTL_FLAG_DATA_MASK) !=
12233                              CTL_FLAG_DATA_OUT);
12234                         break;
12235                 case CTL_LUN_INJ_UA:
12236                         /* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
12237                          * OCCURRED */
12238                         ctl_set_ua(&io->scsiio, 0x29, 0x00);
12239                         break;
12240                 case CTL_LUN_INJ_CUSTOM:
12241                         /*
12242                          * We're assuming the user knows what he is doing.
12243                          * Just copy the sense information without doing
12244                          * checks.
12245                          */
12246                         bcopy(&desc->custom_sense, &io->scsiio.sense_data,
12247                               MIN(sizeof(desc->custom_sense),
12248                                   sizeof(io->scsiio.sense_data)));
12249                         io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
12250                         io->scsiio.sense_len = SSD_FULL_SIZE;
12251                         io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
12252                         break;
12253                 case CTL_LUN_INJ_NONE:
12254                 default:
12255                         /*
12256                          * If this is an error injection type we don't know
12257                          * about, clear the continuous flag (if it is set)
12258                          * so it will get deleted below.
12259                          */
12260                         desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
12261                         break;
12262                 }
12263                 /*
12264                  * By default, each error injection action is a one-shot
12265                  */
12266                 if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
12267                         continue;
12268
12269                 STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
12270
12271                 free(desc, M_CTL);
12272         }
12273 }
12274
12275 #ifdef CTL_IO_DELAY
12276 static void
12277 ctl_datamove_timer_wakeup(void *arg)
12278 {
12279         union ctl_io *io;
12280
12281         io = (union ctl_io *)arg;
12282
12283         ctl_datamove(io);
12284 }
12285 #endif /* CTL_IO_DELAY */
12286
12287 void
12288 ctl_datamove(union ctl_io *io)
12289 {
12290         void (*fe_datamove)(union ctl_io *io);
12291
12292         mtx_assert(&((struct ctl_softc *)CTL_SOFTC(io))->ctl_lock, MA_NOTOWNED);
12293
12294         CTL_DEBUG_PRINT(("ctl_datamove\n"));
12295
12296         /* No data transferred yet.  Frontend must update this when done. */
12297         io->scsiio.kern_data_resid = io->scsiio.kern_data_len;
12298
12299 #ifdef CTL_TIME_IO
12300         if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12301                 char str[256];
12302                 char path_str[64];
12303                 struct sbuf sb;
12304
12305                 ctl_scsi_path_string(io, path_str, sizeof(path_str));
12306                 sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12307
12308                 sbuf_cat(&sb, path_str);
12309                 switch (io->io_hdr.io_type) {
12310                 case CTL_IO_SCSI:
12311                         ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12312                         sbuf_printf(&sb, "\n");
12313                         sbuf_cat(&sb, path_str);
12314                         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12315                                     io->scsiio.tag_num, io->scsiio.tag_type);
12316                         break;
12317                 case CTL_IO_TASK:
12318                         sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12319                                     "Tag Type: %d\n", io->taskio.task_action,
12320                                     io->taskio.tag_num, io->taskio.tag_type);
12321                         break;
12322                 default:
12323                         panic("%s: Invalid CTL I/O type %d\n",
12324                             __func__, io->io_hdr.io_type);
12325                 }
12326                 sbuf_cat(&sb, path_str);
12327                 sbuf_printf(&sb, "ctl_datamove: %jd seconds\n",
12328                             (intmax_t)time_uptime - io->io_hdr.start_time);
12329                 sbuf_finish(&sb);
12330                 printf("%s", sbuf_data(&sb));
12331         }
12332 #endif /* CTL_TIME_IO */
12333
12334 #ifdef CTL_IO_DELAY
12335         if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
12336                 io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
12337         } else {
12338                 struct ctl_lun *lun;
12339
12340                 lun = CTL_LUN(io);
12341                 if ((lun != NULL)
12342                  && (lun->delay_info.datamove_delay > 0)) {
12343
12344                         callout_init(&io->io_hdr.delay_callout, /*mpsafe*/ 1);
12345                         io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
12346                         callout_reset(&io->io_hdr.delay_callout,
12347                                       lun->delay_info.datamove_delay * hz,
12348                                       ctl_datamove_timer_wakeup, io);
12349                         if (lun->delay_info.datamove_type ==
12350                             CTL_DELAY_TYPE_ONESHOT)
12351                                 lun->delay_info.datamove_delay = 0;
12352                         return;
12353                 }
12354         }
12355 #endif
12356
12357         /*
12358          * This command has been aborted.  Set the port status, so we fail
12359          * the data move.
12360          */
12361         if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12362                 printf("ctl_datamove: tag 0x%04x on (%u:%u:%u) aborted\n",
12363                        io->scsiio.tag_num, io->io_hdr.nexus.initid,
12364                        io->io_hdr.nexus.targ_port,
12365                        io->io_hdr.nexus.targ_lun);
12366                 io->io_hdr.port_status = 31337;
12367                 /*
12368                  * Note that the backend, in this case, will get the
12369                  * callback in its context.  In other cases it may get
12370                  * called in the frontend's interrupt thread context.
12371                  */
12372                 io->scsiio.be_move_done(io);
12373                 return;
12374         }
12375
12376         /* Don't confuse frontend with zero length data move. */
12377         if (io->scsiio.kern_data_len == 0) {
12378                 io->scsiio.be_move_done(io);
12379                 return;
12380         }
12381
12382         fe_datamove = CTL_PORT(io)->fe_datamove;
12383         fe_datamove(io);
12384 }
12385
12386 static void
12387 ctl_send_datamove_done(union ctl_io *io, int have_lock)
12388 {
12389         union ctl_ha_msg msg;
12390 #ifdef CTL_TIME_IO
12391         struct bintime cur_bt;
12392 #endif
12393
12394         memset(&msg, 0, sizeof(msg));
12395         msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
12396         msg.hdr.original_sc = io;
12397         msg.hdr.serializing_sc = io->io_hdr.remote_io;
12398         msg.hdr.nexus = io->io_hdr.nexus;
12399         msg.hdr.status = io->io_hdr.status;
12400         msg.scsi.kern_data_resid = io->scsiio.kern_data_resid;
12401         msg.scsi.tag_num = io->scsiio.tag_num;
12402         msg.scsi.tag_type = io->scsiio.tag_type;
12403         msg.scsi.scsi_status = io->scsiio.scsi_status;
12404         memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
12405                io->scsiio.sense_len);
12406         msg.scsi.sense_len = io->scsiio.sense_len;
12407         msg.scsi.port_status = io->io_hdr.port_status;
12408         io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12409         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12410                 ctl_failover_io(io, /*have_lock*/ have_lock);
12411                 return;
12412         }
12413         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12414             sizeof(msg.scsi) - sizeof(msg.scsi.sense_data) +
12415             msg.scsi.sense_len, M_WAITOK);
12416
12417 #ifdef CTL_TIME_IO
12418         getbinuptime(&cur_bt);
12419         bintime_sub(&cur_bt, &io->io_hdr.dma_start_bt);
12420         bintime_add(&io->io_hdr.dma_bt, &cur_bt);
12421 #endif
12422         io->io_hdr.num_dmas++;
12423 }
12424
12425 /*
12426  * The DMA to the remote side is done, now we need to tell the other side
12427  * we're done so it can continue with its data movement.
12428  */
12429 static void
12430 ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
12431 {
12432         union ctl_io *io;
12433         uint32_t i;
12434
12435         io = rq->context;
12436
12437         if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12438                 printf("%s: ISC DMA write failed with error %d", __func__,
12439                        rq->ret);
12440                 ctl_set_internal_failure(&io->scsiio,
12441                                          /*sks_valid*/ 1,
12442                                          /*retry_count*/ rq->ret);
12443         }
12444
12445         ctl_dt_req_free(rq);
12446
12447         for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12448                 free(CTL_LSGLT(io)[i].addr, M_CTL);
12449         free(CTL_RSGL(io), M_CTL);
12450         CTL_RSGL(io) = NULL;
12451         CTL_LSGL(io) = NULL;
12452
12453         /*
12454          * The data is in local and remote memory, so now we need to send
12455          * status (good or back) back to the other side.
12456          */
12457         ctl_send_datamove_done(io, /*have_lock*/ 0);
12458 }
12459
12460 /*
12461  * We've moved the data from the host/controller into local memory.  Now we
12462  * need to push it over to the remote controller's memory.
12463  */
12464 static int
12465 ctl_datamove_remote_dm_write_cb(union ctl_io *io)
12466 {
12467         int retval;
12468
12469         retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
12470                                           ctl_datamove_remote_write_cb);
12471         return (retval);
12472 }
12473
12474 static void
12475 ctl_datamove_remote_write(union ctl_io *io)
12476 {
12477         int retval;
12478         void (*fe_datamove)(union ctl_io *io);
12479
12480         /*
12481          * - Get the data from the host/HBA into local memory.
12482          * - DMA memory from the local controller to the remote controller.
12483          * - Send status back to the remote controller.
12484          */
12485
12486         retval = ctl_datamove_remote_sgl_setup(io);
12487         if (retval != 0)
12488                 return;
12489
12490         /* Switch the pointer over so the FETD knows what to do */
12491         io->scsiio.kern_data_ptr = (uint8_t *)CTL_LSGL(io);
12492
12493         /*
12494          * Use a custom move done callback, since we need to send completion
12495          * back to the other controller, not to the backend on this side.
12496          */
12497         io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
12498
12499         fe_datamove = CTL_PORT(io)->fe_datamove;
12500         fe_datamove(io);
12501 }
12502
12503 static int
12504 ctl_datamove_remote_dm_read_cb(union ctl_io *io)
12505 {
12506         uint32_t i;
12507
12508         for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12509                 free(CTL_LSGLT(io)[i].addr, M_CTL);
12510         free(CTL_RSGL(io), M_CTL);
12511         CTL_RSGL(io) = NULL;
12512         CTL_LSGL(io) = NULL;
12513
12514         /*
12515          * The read is done, now we need to send status (good or bad) back
12516          * to the other side.
12517          */
12518         ctl_send_datamove_done(io, /*have_lock*/ 0);
12519
12520         return (0);
12521 }
12522
12523 static void
12524 ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
12525 {
12526         union ctl_io *io;
12527         void (*fe_datamove)(union ctl_io *io);
12528
12529         io = rq->context;
12530
12531         if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12532                 printf("%s: ISC DMA read failed with error %d\n", __func__,
12533                        rq->ret);
12534                 ctl_set_internal_failure(&io->scsiio,
12535                                          /*sks_valid*/ 1,
12536                                          /*retry_count*/ rq->ret);
12537         }
12538
12539         ctl_dt_req_free(rq);
12540
12541         /* Switch the pointer over so the FETD knows what to do */
12542         io->scsiio.kern_data_ptr = (uint8_t *)CTL_LSGL(io);
12543
12544         /*
12545          * Use a custom move done callback, since we need to send completion
12546          * back to the other controller, not to the backend on this side.
12547          */
12548         io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
12549
12550         /* XXX KDM add checks like the ones in ctl_datamove? */
12551
12552         fe_datamove = CTL_PORT(io)->fe_datamove;
12553         fe_datamove(io);
12554 }
12555
12556 static int
12557 ctl_datamove_remote_sgl_setup(union ctl_io *io)
12558 {
12559         struct ctl_sg_entry *local_sglist;
12560         uint32_t len_to_go;
12561         int retval;
12562         int i;
12563
12564         retval = 0;
12565         local_sglist = CTL_LSGL(io);
12566         len_to_go = io->scsiio.kern_data_len;
12567
12568         /*
12569          * The difficult thing here is that the size of the various
12570          * S/G segments may be different than the size from the
12571          * remote controller.  That'll make it harder when DMAing
12572          * the data back to the other side.
12573          */
12574         for (i = 0; len_to_go > 0; i++) {
12575                 local_sglist[i].len = MIN(len_to_go, CTL_HA_DATAMOVE_SEGMENT);
12576                 local_sglist[i].addr =
12577                     malloc(local_sglist[i].len, M_CTL, M_WAITOK);
12578
12579                 len_to_go -= local_sglist[i].len;
12580         }
12581         /*
12582          * Reset the number of S/G entries accordingly.  The original
12583          * number of S/G entries is available in rem_sg_entries.
12584          */
12585         io->scsiio.kern_sg_entries = i;
12586
12587         return (retval);
12588 }
12589
12590 static int
12591 ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
12592                          ctl_ha_dt_cb callback)
12593 {
12594         struct ctl_ha_dt_req *rq;
12595         struct ctl_sg_entry *remote_sglist, *local_sglist;
12596         uint32_t local_used, remote_used, total_used;
12597         int i, j, isc_ret;
12598
12599         rq = ctl_dt_req_alloc();
12600
12601         /*
12602          * If we failed to allocate the request, and if the DMA didn't fail
12603          * anyway, set busy status.  This is just a resource allocation
12604          * failure.
12605          */
12606         if ((rq == NULL)
12607          && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
12608              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS))
12609                 ctl_set_busy(&io->scsiio);
12610
12611         if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
12612             (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS) {
12613
12614                 if (rq != NULL)
12615                         ctl_dt_req_free(rq);
12616
12617                 /*
12618                  * The data move failed.  We need to return status back
12619                  * to the other controller.  No point in trying to DMA
12620                  * data to the remote controller.
12621                  */
12622
12623                 ctl_send_datamove_done(io, /*have_lock*/ 0);
12624
12625                 return (1);
12626         }
12627
12628         local_sglist = CTL_LSGL(io);
12629         remote_sglist = CTL_RSGL(io);
12630         local_used = 0;
12631         remote_used = 0;
12632         total_used = 0;
12633
12634         /*
12635          * Pull/push the data over the wire from/to the other controller.
12636          * This takes into account the possibility that the local and
12637          * remote sglists may not be identical in terms of the size of
12638          * the elements and the number of elements.
12639          *
12640          * One fundamental assumption here is that the length allocated for
12641          * both the local and remote sglists is identical.  Otherwise, we've
12642          * essentially got a coding error of some sort.
12643          */
12644         isc_ret = CTL_HA_STATUS_SUCCESS;
12645         for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
12646                 uint32_t cur_len;
12647                 uint8_t *tmp_ptr;
12648
12649                 rq->command = command;
12650                 rq->context = io;
12651
12652                 /*
12653                  * Both pointers should be aligned.  But it is possible
12654                  * that the allocation length is not.  They should both
12655                  * also have enough slack left over at the end, though,
12656                  * to round up to the next 8 byte boundary.
12657                  */
12658                 cur_len = MIN(local_sglist[i].len - local_used,
12659                               remote_sglist[j].len - remote_used);
12660                 rq->size = cur_len;
12661
12662                 tmp_ptr = (uint8_t *)local_sglist[i].addr;
12663                 tmp_ptr += local_used;
12664
12665 #if 0
12666                 /* Use physical addresses when talking to ISC hardware */
12667                 if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
12668                         /* XXX KDM use busdma */
12669                         rq->local = vtophys(tmp_ptr);
12670                 } else
12671                         rq->local = tmp_ptr;
12672 #else
12673                 KASSERT((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0,
12674                     ("HA does not support BUS_ADDR"));
12675                 rq->local = tmp_ptr;
12676 #endif
12677
12678                 tmp_ptr = (uint8_t *)remote_sglist[j].addr;
12679                 tmp_ptr += remote_used;
12680                 rq->remote = tmp_ptr;
12681
12682                 rq->callback = NULL;
12683
12684                 local_used += cur_len;
12685                 if (local_used >= local_sglist[i].len) {
12686                         i++;
12687                         local_used = 0;
12688                 }
12689
12690                 remote_used += cur_len;
12691                 if (remote_used >= remote_sglist[j].len) {
12692                         j++;
12693                         remote_used = 0;
12694                 }
12695                 total_used += cur_len;
12696
12697                 if (total_used >= io->scsiio.kern_data_len)
12698                         rq->callback = callback;
12699
12700                 isc_ret = ctl_dt_single(rq);
12701                 if (isc_ret > CTL_HA_STATUS_SUCCESS)
12702                         break;
12703         }
12704         if (isc_ret != CTL_HA_STATUS_WAIT) {
12705                 rq->ret = isc_ret;
12706                 callback(rq);
12707         }
12708
12709         return (0);
12710 }
12711
12712 static void
12713 ctl_datamove_remote_read(union ctl_io *io)
12714 {
12715         int retval;
12716         uint32_t i;
12717
12718         /*
12719          * This will send an error to the other controller in the case of a
12720          * failure.
12721          */
12722         retval = ctl_datamove_remote_sgl_setup(io);
12723         if (retval != 0)
12724                 return;
12725
12726         retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
12727                                           ctl_datamove_remote_read_cb);
12728         if (retval != 0) {
12729                 /*
12730                  * Make sure we free memory if there was an error..  The
12731                  * ctl_datamove_remote_xfer() function will send the
12732                  * datamove done message, or call the callback with an
12733                  * error if there is a problem.
12734                  */
12735                 for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12736                         free(CTL_LSGLT(io)[i].addr, M_CTL);
12737                 free(CTL_RSGL(io), M_CTL);
12738                 CTL_RSGL(io) = NULL;
12739                 CTL_LSGL(io) = NULL;
12740         }
12741 }
12742
12743 /*
12744  * Process a datamove request from the other controller.  This is used for
12745  * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
12746  * first.  Once that is complete, the data gets DMAed into the remote
12747  * controller's memory.  For reads, we DMA from the remote controller's
12748  * memory into our memory first, and then move it out to the FETD.
12749  */
12750 static void
12751 ctl_datamove_remote(union ctl_io *io)
12752 {
12753
12754         mtx_assert(&((struct ctl_softc *)CTL_SOFTC(io))->ctl_lock, MA_NOTOWNED);
12755
12756         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12757                 ctl_failover_io(io, /*have_lock*/ 0);
12758                 return;
12759         }
12760
12761         /*
12762          * Note that we look for an aborted I/O here, but don't do some of
12763          * the other checks that ctl_datamove() normally does.
12764          * We don't need to run the datamove delay code, since that should
12765          * have been done if need be on the other controller.
12766          */
12767         if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12768                 printf("%s: tag 0x%04x on (%u:%u:%u) aborted\n", __func__,
12769                        io->scsiio.tag_num, io->io_hdr.nexus.initid,
12770                        io->io_hdr.nexus.targ_port,
12771                        io->io_hdr.nexus.targ_lun);
12772                 io->io_hdr.port_status = 31338;
12773                 ctl_send_datamove_done(io, /*have_lock*/ 0);
12774                 return;
12775         }
12776
12777         if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT)
12778                 ctl_datamove_remote_write(io);
12779         else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN)
12780                 ctl_datamove_remote_read(io);
12781         else {
12782                 io->io_hdr.port_status = 31339;
12783                 ctl_send_datamove_done(io, /*have_lock*/ 0);
12784         }
12785 }
12786
12787 static void
12788 ctl_process_done(union ctl_io *io)
12789 {
12790         struct ctl_softc *softc = CTL_SOFTC(io);
12791         struct ctl_port *port = CTL_PORT(io);
12792         struct ctl_lun *lun = CTL_LUN(io);
12793         void (*fe_done)(union ctl_io *io);
12794         union ctl_ha_msg msg;
12795
12796         CTL_DEBUG_PRINT(("ctl_process_done\n"));
12797         fe_done = port->fe_done;
12798
12799 #ifdef CTL_TIME_IO
12800         if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12801                 char str[256];
12802                 char path_str[64];
12803                 struct sbuf sb;
12804
12805                 ctl_scsi_path_string(io, path_str, sizeof(path_str));
12806                 sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12807
12808                 sbuf_cat(&sb, path_str);
12809                 switch (io->io_hdr.io_type) {
12810                 case CTL_IO_SCSI:
12811                         ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12812                         sbuf_printf(&sb, "\n");
12813                         sbuf_cat(&sb, path_str);
12814                         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12815                                     io->scsiio.tag_num, io->scsiio.tag_type);
12816                         break;
12817                 case CTL_IO_TASK:
12818                         sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12819                                     "Tag Type: %d\n", io->taskio.task_action,
12820                                     io->taskio.tag_num, io->taskio.tag_type);
12821                         break;
12822                 default:
12823                         panic("%s: Invalid CTL I/O type %d\n",
12824                             __func__, io->io_hdr.io_type);
12825                 }
12826                 sbuf_cat(&sb, path_str);
12827                 sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
12828                             (intmax_t)time_uptime - io->io_hdr.start_time);
12829                 sbuf_finish(&sb);
12830                 printf("%s", sbuf_data(&sb));
12831         }
12832 #endif /* CTL_TIME_IO */
12833
12834         switch (io->io_hdr.io_type) {
12835         case CTL_IO_SCSI:
12836                 break;
12837         case CTL_IO_TASK:
12838                 if (ctl_debug & CTL_DEBUG_INFO)
12839                         ctl_io_error_print(io, NULL);
12840                 fe_done(io);
12841                 return;
12842         default:
12843                 panic("%s: Invalid CTL I/O type %d\n",
12844                     __func__, io->io_hdr.io_type);
12845         }
12846
12847         if (lun == NULL) {
12848                 CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
12849                                  io->io_hdr.nexus.targ_mapped_lun));
12850                 goto bailout;
12851         }
12852
12853         mtx_lock(&lun->lun_lock);
12854
12855         /*
12856          * Check to see if we have any informational exception and status
12857          * of this command can be modified to report it in form of either
12858          * RECOVERED ERROR or NO SENSE, depending on MRIE mode page field.
12859          */
12860         if (lun->ie_reported == 0 && lun->ie_asc != 0 &&
12861             io->io_hdr.status == CTL_SUCCESS &&
12862             (io->io_hdr.flags & CTL_FLAG_STATUS_SENT) == 0) {
12863                 uint8_t mrie = lun->MODE_IE.mrie;
12864                 uint8_t per = ((lun->MODE_RWER.byte3 & SMS_RWER_PER) ||
12865                     (lun->MODE_VER.byte3 & SMS_VER_PER));
12866                 if (((mrie == SIEP_MRIE_REC_COND && per) ||
12867                      mrie == SIEP_MRIE_REC_UNCOND ||
12868                      mrie == SIEP_MRIE_NO_SENSE) &&
12869                     (ctl_get_cmd_entry(&io->scsiio, NULL)->flags &
12870                      CTL_CMD_FLAG_NO_SENSE) == 0) {
12871                         ctl_set_sense(&io->scsiio,
12872                               /*current_error*/ 1,
12873                               /*sense_key*/ (mrie == SIEP_MRIE_NO_SENSE) ?
12874                                 SSD_KEY_NO_SENSE : SSD_KEY_RECOVERED_ERROR,
12875                               /*asc*/ lun->ie_asc,
12876                               /*ascq*/ lun->ie_ascq,
12877                               SSD_ELEM_NONE);
12878                         lun->ie_reported = 1;
12879                 }
12880         } else if (lun->ie_reported < 0)
12881                 lun->ie_reported = 0;
12882
12883         /*
12884          * Check to see if we have any errors to inject here.  We only
12885          * inject errors for commands that don't already have errors set.
12886          */
12887         if (!STAILQ_EMPTY(&lun->error_list) &&
12888             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS) &&
12889             ((io->io_hdr.flags & CTL_FLAG_STATUS_SENT) == 0))
12890                 ctl_inject_error(lun, io);
12891
12892         /*
12893          * XXX KDM how do we treat commands that aren't completed
12894          * successfully?
12895          *
12896          * XXX KDM should we also track I/O latency?
12897          */
12898         if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS &&
12899             io->io_hdr.io_type == CTL_IO_SCSI) {
12900                 int type;
12901 #ifdef CTL_TIME_IO
12902                 struct bintime bt;
12903
12904                 getbinuptime(&bt);
12905                 bintime_sub(&bt, &io->io_hdr.start_bt);
12906 #endif
12907                 if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
12908                     CTL_FLAG_DATA_IN)
12909                         type = CTL_STATS_READ;
12910                 else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
12911                     CTL_FLAG_DATA_OUT)
12912                         type = CTL_STATS_WRITE;
12913                 else
12914                         type = CTL_STATS_NO_IO;
12915
12916                 lun->stats.bytes[type] += io->scsiio.kern_total_len;
12917                 lun->stats.operations[type] ++;
12918                 lun->stats.dmas[type] += io->io_hdr.num_dmas;
12919 #ifdef CTL_TIME_IO
12920                 bintime_add(&lun->stats.dma_time[type], &io->io_hdr.dma_bt);
12921                 bintime_add(&lun->stats.time[type], &bt);
12922 #endif
12923
12924                 mtx_lock(&port->port_lock);
12925                 port->stats.bytes[type] += io->scsiio.kern_total_len;
12926                 port->stats.operations[type] ++;
12927                 port->stats.dmas[type] += io->io_hdr.num_dmas;
12928 #ifdef CTL_TIME_IO
12929                 bintime_add(&port->stats.dma_time[type], &io->io_hdr.dma_bt);
12930                 bintime_add(&port->stats.time[type], &bt);
12931 #endif
12932                 mtx_unlock(&port->port_lock);
12933         }
12934
12935         /*
12936          * Remove this from the OOA queue.
12937          */
12938         TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
12939 #ifdef CTL_TIME_IO
12940         if (TAILQ_EMPTY(&lun->ooa_queue))
12941                 lun->last_busy = getsbinuptime();
12942 #endif
12943
12944         /*
12945          * Run through the blocked queue on this LUN and see if anything
12946          * has become unblocked, now that this transaction is done.
12947          */
12948         ctl_check_blocked(lun);
12949
12950         /*
12951          * If the LUN has been invalidated, free it if there is nothing
12952          * left on its OOA queue.
12953          */
12954         if ((lun->flags & CTL_LUN_INVALID)
12955          && TAILQ_EMPTY(&lun->ooa_queue)) {
12956                 mtx_unlock(&lun->lun_lock);
12957                 ctl_free_lun(lun);
12958         } else
12959                 mtx_unlock(&lun->lun_lock);
12960
12961 bailout:
12962
12963         /*
12964          * If this command has been aborted, make sure we set the status
12965          * properly.  The FETD is responsible for freeing the I/O and doing
12966          * whatever it needs to do to clean up its state.
12967          */
12968         if (io->io_hdr.flags & CTL_FLAG_ABORT)
12969                 ctl_set_task_aborted(&io->scsiio);
12970
12971         /*
12972          * If enabled, print command error status.
12973          */
12974         if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS &&
12975             (ctl_debug & CTL_DEBUG_INFO) != 0)
12976                 ctl_io_error_print(io, NULL);
12977
12978         /*
12979          * Tell the FETD or the other shelf controller we're done with this
12980          * command.  Note that only SCSI commands get to this point.  Task
12981          * management commands are completed above.
12982          */
12983         if ((softc->ha_mode != CTL_HA_MODE_XFER) &&
12984             (io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)) {
12985                 memset(&msg, 0, sizeof(msg));
12986                 msg.hdr.msg_type = CTL_MSG_FINISH_IO;
12987                 msg.hdr.serializing_sc = io->io_hdr.remote_io;
12988                 msg.hdr.nexus = io->io_hdr.nexus;
12989                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12990                     sizeof(msg.scsi) - sizeof(msg.scsi.sense_data),
12991                     M_WAITOK);
12992         }
12993
12994         fe_done(io);
12995 }
12996
12997 /*
12998  * Front end should call this if it doesn't do autosense.  When the request
12999  * sense comes back in from the initiator, we'll dequeue this and send it.
13000  */
13001 int
13002 ctl_queue_sense(union ctl_io *io)
13003 {
13004         struct ctl_softc *softc = CTL_SOFTC(io);
13005         struct ctl_port *port = CTL_PORT(io);
13006         struct ctl_lun *lun;
13007         struct scsi_sense_data *ps;
13008         uint32_t initidx, p, targ_lun;
13009
13010         CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
13011
13012         targ_lun = ctl_lun_map_from_port(port, io->io_hdr.nexus.targ_lun);
13013
13014         /*
13015          * LUN lookup will likely move to the ctl_work_thread() once we
13016          * have our new queueing infrastructure (that doesn't put things on
13017          * a per-LUN queue initially).  That is so that we can handle
13018          * things like an INQUIRY to a LUN that we don't have enabled.  We
13019          * can't deal with that right now.
13020          * If we don't have a LUN for this, just toss the sense information.
13021          */
13022         mtx_lock(&softc->ctl_lock);
13023         if (targ_lun >= ctl_max_luns ||
13024             (lun = softc->ctl_luns[targ_lun]) == NULL) {
13025                 mtx_unlock(&softc->ctl_lock);
13026                 goto bailout;
13027         }
13028         mtx_lock(&lun->lun_lock);
13029         mtx_unlock(&softc->ctl_lock);
13030
13031         initidx = ctl_get_initindex(&io->io_hdr.nexus);
13032         p = initidx / CTL_MAX_INIT_PER_PORT;
13033         if (lun->pending_sense[p] == NULL) {
13034                 lun->pending_sense[p] = malloc(sizeof(*ps) * CTL_MAX_INIT_PER_PORT,
13035                     M_CTL, M_NOWAIT | M_ZERO);
13036         }
13037         if ((ps = lun->pending_sense[p]) != NULL) {
13038                 ps += initidx % CTL_MAX_INIT_PER_PORT;
13039                 memset(ps, 0, sizeof(*ps));
13040                 memcpy(ps, &io->scsiio.sense_data, io->scsiio.sense_len);
13041         }
13042         mtx_unlock(&lun->lun_lock);
13043
13044 bailout:
13045         ctl_free_io(io);
13046         return (CTL_RETVAL_COMPLETE);
13047 }
13048
13049 /*
13050  * Primary command inlet from frontend ports.  All SCSI and task I/O
13051  * requests must go through this function.
13052  */
13053 int
13054 ctl_queue(union ctl_io *io)
13055 {
13056         struct ctl_port *port = CTL_PORT(io);
13057
13058         CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
13059
13060 #ifdef CTL_TIME_IO
13061         io->io_hdr.start_time = time_uptime;
13062         getbinuptime(&io->io_hdr.start_bt);
13063 #endif /* CTL_TIME_IO */
13064
13065         /* Map FE-specific LUN ID into global one. */
13066         io->io_hdr.nexus.targ_mapped_lun =
13067             ctl_lun_map_from_port(port, io->io_hdr.nexus.targ_lun);
13068
13069         switch (io->io_hdr.io_type) {
13070         case CTL_IO_SCSI:
13071         case CTL_IO_TASK:
13072                 if (ctl_debug & CTL_DEBUG_CDB)
13073                         ctl_io_print(io);
13074                 ctl_enqueue_incoming(io);
13075                 break;
13076         default:
13077                 printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
13078                 return (EINVAL);
13079         }
13080
13081         return (CTL_RETVAL_COMPLETE);
13082 }
13083
13084 #ifdef CTL_IO_DELAY
13085 static void
13086 ctl_done_timer_wakeup(void *arg)
13087 {
13088         union ctl_io *io;
13089
13090         io = (union ctl_io *)arg;
13091         ctl_done(io);
13092 }
13093 #endif /* CTL_IO_DELAY */
13094
13095 void
13096 ctl_serseq_done(union ctl_io *io)
13097 {
13098         struct ctl_lun *lun = CTL_LUN(io);;
13099
13100         if (lun->be_lun == NULL ||
13101             lun->be_lun->serseq == CTL_LUN_SERSEQ_OFF)
13102                 return;
13103         mtx_lock(&lun->lun_lock);
13104         io->io_hdr.flags |= CTL_FLAG_SERSEQ_DONE;
13105         ctl_check_blocked(lun);
13106         mtx_unlock(&lun->lun_lock);
13107 }
13108
13109 void
13110 ctl_done(union ctl_io *io)
13111 {
13112
13113         /*
13114          * Enable this to catch duplicate completion issues.
13115          */
13116 #if 0
13117         if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
13118                 printf("%s: type %d msg %d cdb %x iptl: "
13119                        "%u:%u:%u tag 0x%04x "
13120                        "flag %#x status %x\n",
13121                         __func__,
13122                         io->io_hdr.io_type,
13123                         io->io_hdr.msg_type,
13124                         io->scsiio.cdb[0],
13125                         io->io_hdr.nexus.initid,
13126                         io->io_hdr.nexus.targ_port,
13127                         io->io_hdr.nexus.targ_lun,
13128                         (io->io_hdr.io_type ==
13129                         CTL_IO_TASK) ?
13130                         io->taskio.tag_num :
13131                         io->scsiio.tag_num,
13132                         io->io_hdr.flags,
13133                         io->io_hdr.status);
13134         } else
13135                 io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
13136 #endif
13137
13138         /*
13139          * This is an internal copy of an I/O, and should not go through
13140          * the normal done processing logic.
13141          */
13142         if (io->io_hdr.flags & CTL_FLAG_INT_COPY)
13143                 return;
13144
13145 #ifdef CTL_IO_DELAY
13146         if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
13147                 io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
13148         } else {
13149                 struct ctl_lun *lun = CTL_LUN(io);
13150
13151                 if ((lun != NULL)
13152                  && (lun->delay_info.done_delay > 0)) {
13153
13154                         callout_init(&io->io_hdr.delay_callout, /*mpsafe*/ 1);
13155                         io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
13156                         callout_reset(&io->io_hdr.delay_callout,
13157                                       lun->delay_info.done_delay * hz,
13158                                       ctl_done_timer_wakeup, io);
13159                         if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
13160                                 lun->delay_info.done_delay = 0;
13161                         return;
13162                 }
13163         }
13164 #endif /* CTL_IO_DELAY */
13165
13166         ctl_enqueue_done(io);
13167 }
13168
13169 static void
13170 ctl_work_thread(void *arg)
13171 {
13172         struct ctl_thread *thr = (struct ctl_thread *)arg;
13173         struct ctl_softc *softc = thr->ctl_softc;
13174         union ctl_io *io;
13175         int retval;
13176
13177         CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
13178
13179         while (!softc->shutdown) {
13180                 /*
13181                  * We handle the queues in this order:
13182                  * - ISC
13183                  * - done queue (to free up resources, unblock other commands)
13184                  * - incoming queue
13185                  * - RtR queue
13186                  *
13187                  * If those queues are empty, we break out of the loop and
13188                  * go to sleep.
13189                  */
13190                 mtx_lock(&thr->queue_lock);
13191                 io = (union ctl_io *)STAILQ_FIRST(&thr->isc_queue);
13192                 if (io != NULL) {
13193                         STAILQ_REMOVE_HEAD(&thr->isc_queue, links);
13194                         mtx_unlock(&thr->queue_lock);
13195                         ctl_handle_isc(io);
13196                         continue;
13197                 }
13198                 io = (union ctl_io *)STAILQ_FIRST(&thr->done_queue);
13199                 if (io != NULL) {
13200                         STAILQ_REMOVE_HEAD(&thr->done_queue, links);
13201                         /* clear any blocked commands, call fe_done */
13202                         mtx_unlock(&thr->queue_lock);
13203                         ctl_process_done(io);
13204                         continue;
13205                 }
13206                 io = (union ctl_io *)STAILQ_FIRST(&thr->incoming_queue);
13207                 if (io != NULL) {
13208                         STAILQ_REMOVE_HEAD(&thr->incoming_queue, links);
13209                         mtx_unlock(&thr->queue_lock);
13210                         if (io->io_hdr.io_type == CTL_IO_TASK)
13211                                 ctl_run_task(io);
13212                         else
13213                                 ctl_scsiio_precheck(softc, &io->scsiio);
13214                         continue;
13215                 }
13216                 io = (union ctl_io *)STAILQ_FIRST(&thr->rtr_queue);
13217                 if (io != NULL) {
13218                         STAILQ_REMOVE_HEAD(&thr->rtr_queue, links);
13219                         mtx_unlock(&thr->queue_lock);
13220                         retval = ctl_scsiio(&io->scsiio);
13221                         if (retval != CTL_RETVAL_COMPLETE)
13222                                 CTL_DEBUG_PRINT(("ctl_scsiio failed\n"));
13223                         continue;
13224                 }
13225
13226                 /* Sleep until we have something to do. */
13227                 mtx_sleep(thr, &thr->queue_lock, PDROP | PRIBIO, "-", 0);
13228         }
13229         thr->thread = NULL;
13230         kthread_exit();
13231 }
13232
13233 static void
13234 ctl_lun_thread(void *arg)
13235 {
13236         struct ctl_softc *softc = (struct ctl_softc *)arg;
13237         struct ctl_be_lun *be_lun;
13238
13239         CTL_DEBUG_PRINT(("ctl_lun_thread starting\n"));
13240
13241         while (!softc->shutdown) {
13242                 mtx_lock(&softc->ctl_lock);
13243                 be_lun = STAILQ_FIRST(&softc->pending_lun_queue);
13244                 if (be_lun != NULL) {
13245                         STAILQ_REMOVE_HEAD(&softc->pending_lun_queue, links);
13246                         mtx_unlock(&softc->ctl_lock);
13247                         ctl_create_lun(be_lun);
13248                         continue;
13249                 }
13250
13251                 /* Sleep until we have something to do. */
13252                 mtx_sleep(&softc->pending_lun_queue, &softc->ctl_lock,
13253                     PDROP | PRIBIO, "-", 0);
13254         }
13255         softc->lun_thread = NULL;
13256         kthread_exit();
13257 }
13258
13259 static void
13260 ctl_thresh_thread(void *arg)
13261 {
13262         struct ctl_softc *softc = (struct ctl_softc *)arg;
13263         struct ctl_lun *lun;
13264         struct ctl_logical_block_provisioning_page *page;
13265         const char *attr;
13266         union ctl_ha_msg msg;
13267         uint64_t thres, val;
13268         int i, e, set;
13269
13270         CTL_DEBUG_PRINT(("ctl_thresh_thread starting\n"));
13271
13272         while (!softc->shutdown) {
13273                 mtx_lock(&softc->ctl_lock);
13274                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
13275                         if ((lun->flags & CTL_LUN_DISABLED) ||
13276                             (lun->flags & CTL_LUN_NO_MEDIA) ||
13277                             lun->backend->lun_attr == NULL)
13278                                 continue;
13279                         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
13280                             softc->ha_mode == CTL_HA_MODE_XFER)
13281                                 continue;
13282                         if ((lun->MODE_RWER.byte8 & SMS_RWER_LBPERE) == 0)
13283                                 continue;
13284                         e = 0;
13285                         page = &lun->MODE_LBP;
13286                         for (i = 0; i < CTL_NUM_LBP_THRESH; i++) {
13287                                 if ((page->descr[i].flags & SLBPPD_ENABLED) == 0)
13288                                         continue;
13289                                 thres = scsi_4btoul(page->descr[i].count);
13290                                 thres <<= CTL_LBP_EXPONENT;
13291                                 switch (page->descr[i].resource) {
13292                                 case 0x01:
13293                                         attr = "blocksavail";
13294                                         break;
13295                                 case 0x02:
13296                                         attr = "blocksused";
13297                                         break;
13298                                 case 0xf1:
13299                                         attr = "poolblocksavail";
13300                                         break;
13301                                 case 0xf2:
13302                                         attr = "poolblocksused";
13303                                         break;
13304                                 default:
13305                                         continue;
13306                                 }
13307                                 mtx_unlock(&softc->ctl_lock); // XXX
13308                                 val = lun->backend->lun_attr(
13309                                     lun->be_lun->be_lun, attr);
13310                                 mtx_lock(&softc->ctl_lock);
13311                                 if (val == UINT64_MAX)
13312                                         continue;
13313                                 if ((page->descr[i].flags & SLBPPD_ARMING_MASK)
13314                                     == SLBPPD_ARMING_INC)
13315                                         e = (val >= thres);
13316                                 else
13317                                         e = (val <= thres);
13318                                 if (e)
13319                                         break;
13320                         }
13321                         mtx_lock(&lun->lun_lock);
13322                         if (e) {
13323                                 scsi_u64to8b((uint8_t *)&page->descr[i] -
13324                                     (uint8_t *)page, lun->ua_tpt_info);
13325                                 if (lun->lasttpt == 0 ||
13326                                     time_uptime - lun->lasttpt >= CTL_LBP_UA_PERIOD) {
13327                                         lun->lasttpt = time_uptime;
13328                                         ctl_est_ua_all(lun, -1, CTL_UA_THIN_PROV_THRES);
13329                                         set = 1;
13330                                 } else
13331                                         set = 0;
13332                         } else {
13333                                 lun->lasttpt = 0;
13334                                 ctl_clr_ua_all(lun, -1, CTL_UA_THIN_PROV_THRES);
13335                                 set = -1;
13336                         }
13337                         mtx_unlock(&lun->lun_lock);
13338                         if (set != 0 &&
13339                             lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
13340                                 /* Send msg to other side. */
13341                                 bzero(&msg.ua, sizeof(msg.ua));
13342                                 msg.hdr.msg_type = CTL_MSG_UA;
13343                                 msg.hdr.nexus.initid = -1;
13344                                 msg.hdr.nexus.targ_port = -1;
13345                                 msg.hdr.nexus.targ_lun = lun->lun;
13346                                 msg.hdr.nexus.targ_mapped_lun = lun->lun;
13347                                 msg.ua.ua_all = 1;
13348                                 msg.ua.ua_set = (set > 0);
13349                                 msg.ua.ua_type = CTL_UA_THIN_PROV_THRES;
13350                                 memcpy(msg.ua.ua_info, lun->ua_tpt_info, 8);
13351                                 mtx_unlock(&softc->ctl_lock); // XXX
13352                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13353                                     sizeof(msg.ua), M_WAITOK);
13354                                 mtx_lock(&softc->ctl_lock);
13355                         }
13356                 }
13357                 mtx_sleep(&softc->thresh_thread, &softc->ctl_lock,
13358                     PDROP | PRIBIO, "-", CTL_LBP_PERIOD * hz);
13359         }
13360         softc->thresh_thread = NULL;
13361         kthread_exit();
13362 }
13363
13364 static void
13365 ctl_enqueue_incoming(union ctl_io *io)
13366 {
13367         struct ctl_softc *softc = CTL_SOFTC(io);
13368         struct ctl_thread *thr;
13369         u_int idx;
13370
13371         idx = (io->io_hdr.nexus.targ_port * 127 +
13372                io->io_hdr.nexus.initid) % worker_threads;
13373         thr = &softc->threads[idx];
13374         mtx_lock(&thr->queue_lock);
13375         STAILQ_INSERT_TAIL(&thr->incoming_queue, &io->io_hdr, links);
13376         mtx_unlock(&thr->queue_lock);
13377         wakeup(thr);
13378 }
13379
13380 static void
13381 ctl_enqueue_rtr(union ctl_io *io)
13382 {
13383         struct ctl_softc *softc = CTL_SOFTC(io);
13384         struct ctl_thread *thr;
13385
13386         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13387         mtx_lock(&thr->queue_lock);
13388         STAILQ_INSERT_TAIL(&thr->rtr_queue, &io->io_hdr, links);
13389         mtx_unlock(&thr->queue_lock);
13390         wakeup(thr);
13391 }
13392
13393 static void
13394 ctl_enqueue_done(union ctl_io *io)
13395 {
13396         struct ctl_softc *softc = CTL_SOFTC(io);
13397         struct ctl_thread *thr;
13398
13399         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13400         mtx_lock(&thr->queue_lock);
13401         STAILQ_INSERT_TAIL(&thr->done_queue, &io->io_hdr, links);
13402         mtx_unlock(&thr->queue_lock);
13403         wakeup(thr);
13404 }
13405
13406 static void
13407 ctl_enqueue_isc(union ctl_io *io)
13408 {
13409         struct ctl_softc *softc = CTL_SOFTC(io);
13410         struct ctl_thread *thr;
13411
13412         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13413         mtx_lock(&thr->queue_lock);
13414         STAILQ_INSERT_TAIL(&thr->isc_queue, &io->io_hdr, links);
13415         mtx_unlock(&thr->queue_lock);
13416         wakeup(thr);
13417 }
13418
13419 /*
13420  *  vim: ts=8
13421  */