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 #if 0
1456                         printf("port %u, iid %u, lun %u\n",
1457                                io->io_hdr.nexus.targ_port,
1458                                io->io_hdr.nexus.initid,
1459                                io->io_hdr.nexus.targ_lun);
1460 #endif
1461                         io->scsiio.tag_num = msg->scsi.tag_num;
1462                         io->scsiio.tag_type = msg->scsi.tag_type;
1463 #ifdef CTL_TIME_IO
1464                         io->io_hdr.start_time = time_uptime;
1465                         getbinuptime(&io->io_hdr.start_bt);
1466 #endif /* CTL_TIME_IO */
1467                         io->scsiio.cdb_len = msg->scsi.cdb_len;
1468                         memcpy(io->scsiio.cdb, msg->scsi.cdb,
1469                                CTL_MAX_CDBLEN);
1470                         if (softc->ha_mode == CTL_HA_MODE_XFER) {
1471                                 const struct ctl_cmd_entry *entry;
1472
1473                                 entry = ctl_get_cmd_entry(&io->scsiio, NULL);
1474                                 io->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
1475                                 io->io_hdr.flags |=
1476                                         entry->flags & CTL_FLAG_DATA_MASK;
1477                         }
1478                         ctl_enqueue_isc(io);
1479                         break;
1480
1481                 /* Performed on the Originating SC, XFER mode only */
1482                 case CTL_MSG_DATAMOVE: {
1483                         struct ctl_sg_entry *sgl;
1484                         int i, j;
1485
1486                         io = msg->hdr.original_sc;
1487                         if (io == NULL) {
1488                                 printf("%s: original_sc == NULL!\n", __func__);
1489                                 /* XXX KDM do something here */
1490                                 break;
1491                         }
1492                         io->io_hdr.msg_type = CTL_MSG_DATAMOVE;
1493                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1494                         /*
1495                          * Keep track of this, we need to send it back over
1496                          * when the datamove is complete.
1497                          */
1498                         io->io_hdr.remote_io = msg->hdr.serializing_sc;
1499                         if (msg->hdr.status == CTL_SUCCESS)
1500                                 io->io_hdr.status = msg->hdr.status;
1501
1502                         if (msg->dt.sg_sequence == 0) {
1503 #ifdef CTL_TIME_IO
1504                                 getbinuptime(&io->io_hdr.dma_start_bt);
1505 #endif
1506                                 i = msg->dt.kern_sg_entries +
1507                                     msg->dt.kern_data_len /
1508                                     CTL_HA_DATAMOVE_SEGMENT + 1;
1509                                 sgl = malloc(sizeof(*sgl) * i, M_CTL,
1510                                     M_WAITOK | M_ZERO);
1511                                 CTL_RSGL(io) = sgl;
1512                                 CTL_LSGL(io) = &sgl[msg->dt.kern_sg_entries];
1513
1514                                 io->scsiio.kern_data_ptr = (uint8_t *)sgl;
1515
1516                                 io->scsiio.kern_sg_entries =
1517                                         msg->dt.kern_sg_entries;
1518                                 io->scsiio.rem_sg_entries =
1519                                         msg->dt.kern_sg_entries;
1520                                 io->scsiio.kern_data_len =
1521                                         msg->dt.kern_data_len;
1522                                 io->scsiio.kern_total_len =
1523                                         msg->dt.kern_total_len;
1524                                 io->scsiio.kern_data_resid =
1525                                         msg->dt.kern_data_resid;
1526                                 io->scsiio.kern_rel_offset =
1527                                         msg->dt.kern_rel_offset;
1528                                 io->io_hdr.flags &= ~CTL_FLAG_BUS_ADDR;
1529                                 io->io_hdr.flags |= msg->dt.flags &
1530                                     CTL_FLAG_BUS_ADDR;
1531                         } else
1532                                 sgl = (struct ctl_sg_entry *)
1533                                         io->scsiio.kern_data_ptr;
1534
1535                         for (i = msg->dt.sent_sg_entries, j = 0;
1536                              i < (msg->dt.sent_sg_entries +
1537                              msg->dt.cur_sg_entries); i++, j++) {
1538                                 sgl[i].addr = msg->dt.sg_list[j].addr;
1539                                 sgl[i].len = msg->dt.sg_list[j].len;
1540
1541 #if 0
1542                                 printf("%s: DATAMOVE: %p,%lu j=%d, i=%d\n",
1543                                     __func__, sgl[i].addr, sgl[i].len, j, i);
1544 #endif
1545                         }
1546
1547                         /*
1548                          * If this is the last piece of the I/O, we've got
1549                          * the full S/G list.  Queue processing in the thread.
1550                          * Otherwise wait for the next piece.
1551                          */
1552                         if (msg->dt.sg_last != 0)
1553                                 ctl_enqueue_isc(io);
1554                         break;
1555                 }
1556                 /* Performed on the Serializing (primary) SC, XFER mode only */
1557                 case CTL_MSG_DATAMOVE_DONE: {
1558                         if (msg->hdr.serializing_sc == NULL) {
1559                                 printf("%s: serializing_sc == NULL!\n",
1560                                        __func__);
1561                                 /* XXX KDM now what? */
1562                                 break;
1563                         }
1564                         /*
1565                          * We grab the sense information here in case
1566                          * there was a failure, so we can return status
1567                          * back to the initiator.
1568                          */
1569                         io = msg->hdr.serializing_sc;
1570                         io->io_hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
1571                         io->io_hdr.flags &= ~CTL_FLAG_DMA_INPROG;
1572                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1573                         io->io_hdr.port_status = msg->scsi.port_status;
1574                         io->scsiio.kern_data_resid = msg->scsi.kern_data_resid;
1575                         if (msg->hdr.status != CTL_STATUS_NONE) {
1576                                 io->io_hdr.status = msg->hdr.status;
1577                                 io->scsiio.scsi_status = msg->scsi.scsi_status;
1578                                 io->scsiio.sense_len = msg->scsi.sense_len;
1579                                 memcpy(&io->scsiio.sense_data,
1580                                     &msg->scsi.sense_data,
1581                                     msg->scsi.sense_len);
1582                                 if (msg->hdr.status == CTL_SUCCESS)
1583                                         io->io_hdr.flags |= CTL_FLAG_STATUS_SENT;
1584                         }
1585                         ctl_enqueue_isc(io);
1586                         break;
1587                 }
1588
1589                 /* Preformed on Originating SC, SER_ONLY mode */
1590                 case CTL_MSG_R2R:
1591                         io = msg->hdr.original_sc;
1592                         if (io == NULL) {
1593                                 printf("%s: original_sc == NULL!\n",
1594                                     __func__);
1595                                 break;
1596                         }
1597                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1598                         io->io_hdr.msg_type = CTL_MSG_R2R;
1599                         io->io_hdr.remote_io = msg->hdr.serializing_sc;
1600                         ctl_enqueue_isc(io);
1601                         break;
1602
1603                 /*
1604                  * Performed on Serializing(i.e. primary SC) SC in SER_ONLY
1605                  * mode.
1606                  * Performed on the Originating (i.e. secondary) SC in XFER
1607                  * mode
1608                  */
1609                 case CTL_MSG_FINISH_IO:
1610                         if (softc->ha_mode == CTL_HA_MODE_XFER)
1611                                 ctl_isc_handler_finish_xfer(softc, msg);
1612                         else
1613                                 ctl_isc_handler_finish_ser_only(softc, msg);
1614                         break;
1615
1616                 /* Preformed on Originating SC */
1617                 case CTL_MSG_BAD_JUJU:
1618                         io = msg->hdr.original_sc;
1619                         if (io == NULL) {
1620                                 printf("%s: Bad JUJU!, original_sc is NULL!\n",
1621                                        __func__);
1622                                 break;
1623                         }
1624                         ctl_copy_sense_data(msg, io);
1625                         /*
1626                          * IO should have already been cleaned up on other
1627                          * SC so clear this flag so we won't send a message
1628                          * back to finish the IO there.
1629                          */
1630                         io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
1631                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1632
1633                         /* io = msg->hdr.serializing_sc; */
1634                         io->io_hdr.msg_type = CTL_MSG_BAD_JUJU;
1635                         ctl_enqueue_isc(io);
1636                         break;
1637
1638                 /* Handle resets sent from the other side */
1639                 case CTL_MSG_MANAGE_TASKS: {
1640                         struct ctl_taskio *taskio;
1641                         taskio = (struct ctl_taskio *)ctl_alloc_io(
1642                             softc->othersc_pool);
1643                         ctl_zero_io((union ctl_io *)taskio);
1644                         taskio->io_hdr.io_type = CTL_IO_TASK;
1645                         taskio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
1646                         taskio->io_hdr.nexus = msg->hdr.nexus;
1647                         taskio->task_action = msg->task.task_action;
1648                         taskio->tag_num = msg->task.tag_num;
1649                         taskio->tag_type = msg->task.tag_type;
1650 #ifdef CTL_TIME_IO
1651                         taskio->io_hdr.start_time = time_uptime;
1652                         getbinuptime(&taskio->io_hdr.start_bt);
1653 #endif /* CTL_TIME_IO */
1654                         ctl_run_task((union ctl_io *)taskio);
1655                         break;
1656                 }
1657                 /* Persistent Reserve action which needs attention */
1658                 case CTL_MSG_PERS_ACTION:
1659                         presio = (struct ctl_prio *)ctl_alloc_io(
1660                             softc->othersc_pool);
1661                         ctl_zero_io((union ctl_io *)presio);
1662                         presio->io_hdr.msg_type = CTL_MSG_PERS_ACTION;
1663                         presio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
1664                         presio->io_hdr.nexus = msg->hdr.nexus;
1665                         presio->pr_msg = msg->pr;
1666                         ctl_enqueue_isc((union ctl_io *)presio);
1667                         break;
1668                 case CTL_MSG_UA:
1669                         ctl_isc_ua(softc, msg, param);
1670                         break;
1671                 case CTL_MSG_PORT_SYNC:
1672                         ctl_isc_port_sync(softc, msg, param);
1673                         break;
1674                 case CTL_MSG_LUN_SYNC:
1675                         ctl_isc_lun_sync(softc, msg, param);
1676                         break;
1677                 case CTL_MSG_IID_SYNC:
1678                         ctl_isc_iid_sync(softc, msg, param);
1679                         break;
1680                 case CTL_MSG_LOGIN:
1681                         ctl_isc_login(softc, msg, param);
1682                         break;
1683                 case CTL_MSG_MODE_SYNC:
1684                         ctl_isc_mode_sync(softc, msg, param);
1685                         break;
1686                 default:
1687                         printf("Received HA message of unknown type %d\n",
1688                             msg->hdr.msg_type);
1689                         ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1690                         break;
1691                 }
1692                 if (msg != &msgbuf)
1693                         free(msg, M_CTL);
1694         } else if (event == CTL_HA_EVT_LINK_CHANGE) {
1695                 printf("CTL: HA link status changed from %d to %d\n",
1696                     softc->ha_link, param);
1697                 if (param == softc->ha_link)
1698                         return;
1699                 if (softc->ha_link == CTL_HA_LINK_ONLINE) {
1700                         softc->ha_link = param;
1701                         ctl_isc_ha_link_down(softc);
1702                 } else {
1703                         softc->ha_link = param;
1704                         if (softc->ha_link == CTL_HA_LINK_ONLINE)
1705                                 ctl_isc_ha_link_up(softc);
1706                 }
1707                 return;
1708         } else {
1709                 printf("ctl_isc_event_handler: Unknown event %d\n", event);
1710                 return;
1711         }
1712 }
1713
1714 static void
1715 ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest)
1716 {
1717
1718         memcpy(&dest->scsiio.sense_data, &src->scsi.sense_data,
1719             src->scsi.sense_len);
1720         dest->scsiio.scsi_status = src->scsi.scsi_status;
1721         dest->scsiio.sense_len = src->scsi.sense_len;
1722         dest->io_hdr.status = src->hdr.status;
1723 }
1724
1725 static void
1726 ctl_copy_sense_data_back(union ctl_io *src, union ctl_ha_msg *dest)
1727 {
1728
1729         memcpy(&dest->scsi.sense_data, &src->scsiio.sense_data,
1730             src->scsiio.sense_len);
1731         dest->scsi.scsi_status = src->scsiio.scsi_status;
1732         dest->scsi.sense_len = src->scsiio.sense_len;
1733         dest->hdr.status = src->io_hdr.status;
1734 }
1735
1736 void
1737 ctl_est_ua(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua)
1738 {
1739         struct ctl_softc *softc = lun->ctl_softc;
1740         ctl_ua_type *pu;
1741
1742         if (initidx < softc->init_min || initidx >= softc->init_max)
1743                 return;
1744         mtx_assert(&lun->lun_lock, MA_OWNED);
1745         pu = lun->pending_ua[initidx / CTL_MAX_INIT_PER_PORT];
1746         if (pu == NULL)
1747                 return;
1748         pu[initidx % CTL_MAX_INIT_PER_PORT] |= ua;
1749 }
1750
1751 void
1752 ctl_est_ua_port(struct ctl_lun *lun, int port, uint32_t except, ctl_ua_type ua)
1753 {
1754         int i;
1755
1756         mtx_assert(&lun->lun_lock, MA_OWNED);
1757         if (lun->pending_ua[port] == NULL)
1758                 return;
1759         for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1760                 if (port * CTL_MAX_INIT_PER_PORT + i == except)
1761                         continue;
1762                 lun->pending_ua[port][i] |= ua;
1763         }
1764 }
1765
1766 void
1767 ctl_est_ua_all(struct ctl_lun *lun, uint32_t except, ctl_ua_type ua)
1768 {
1769         struct ctl_softc *softc = lun->ctl_softc;
1770         int i;
1771
1772         mtx_assert(&lun->lun_lock, MA_OWNED);
1773         for (i = softc->port_min; i < softc->port_max; i++)
1774                 ctl_est_ua_port(lun, i, except, ua);
1775 }
1776
1777 void
1778 ctl_clr_ua(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua)
1779 {
1780         struct ctl_softc *softc = lun->ctl_softc;
1781         ctl_ua_type *pu;
1782
1783         if (initidx < softc->init_min || initidx >= softc->init_max)
1784                 return;
1785         mtx_assert(&lun->lun_lock, MA_OWNED);
1786         pu = lun->pending_ua[initidx / CTL_MAX_INIT_PER_PORT];
1787         if (pu == NULL)
1788                 return;
1789         pu[initidx % CTL_MAX_INIT_PER_PORT] &= ~ua;
1790 }
1791
1792 void
1793 ctl_clr_ua_all(struct ctl_lun *lun, uint32_t except, ctl_ua_type ua)
1794 {
1795         struct ctl_softc *softc = lun->ctl_softc;
1796         int i, j;
1797
1798         mtx_assert(&lun->lun_lock, MA_OWNED);
1799         for (i = softc->port_min; i < softc->port_max; i++) {
1800                 if (lun->pending_ua[i] == NULL)
1801                         continue;
1802                 for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
1803                         if (i * CTL_MAX_INIT_PER_PORT + j == except)
1804                                 continue;
1805                         lun->pending_ua[i][j] &= ~ua;
1806                 }
1807         }
1808 }
1809
1810 void
1811 ctl_clr_ua_allluns(struct ctl_softc *ctl_softc, uint32_t initidx,
1812     ctl_ua_type ua_type)
1813 {
1814         struct ctl_lun *lun;
1815
1816         mtx_assert(&ctl_softc->ctl_lock, MA_OWNED);
1817         STAILQ_FOREACH(lun, &ctl_softc->lun_list, links) {
1818                 mtx_lock(&lun->lun_lock);
1819                 ctl_clr_ua(lun, initidx, ua_type);
1820                 mtx_unlock(&lun->lun_lock);
1821         }
1822 }
1823
1824 static int
1825 ctl_ha_role_sysctl(SYSCTL_HANDLER_ARGS)
1826 {
1827         struct ctl_softc *softc = (struct ctl_softc *)arg1;
1828         struct ctl_lun *lun;
1829         struct ctl_lun_req ireq;
1830         int error, value;
1831
1832         value = (softc->flags & CTL_FLAG_ACTIVE_SHELF) ? 0 : 1;
1833         error = sysctl_handle_int(oidp, &value, 0, req);
1834         if ((error != 0) || (req->newptr == NULL))
1835                 return (error);
1836
1837         mtx_lock(&softc->ctl_lock);
1838         if (value == 0)
1839                 softc->flags |= CTL_FLAG_ACTIVE_SHELF;
1840         else
1841                 softc->flags &= ~CTL_FLAG_ACTIVE_SHELF;
1842         STAILQ_FOREACH(lun, &softc->lun_list, links) {
1843                 mtx_unlock(&softc->ctl_lock);
1844                 bzero(&ireq, sizeof(ireq));
1845                 ireq.reqtype = CTL_LUNREQ_MODIFY;
1846                 ireq.reqdata.modify.lun_id = lun->lun;
1847                 lun->backend->ioctl(NULL, CTL_LUN_REQ, (caddr_t)&ireq, 0,
1848                     curthread);
1849                 if (ireq.status != CTL_LUN_OK) {
1850                         printf("%s: CTL_LUNREQ_MODIFY returned %d '%s'\n",
1851                             __func__, ireq.status, ireq.error_str);
1852                 }
1853                 mtx_lock(&softc->ctl_lock);
1854         }
1855         mtx_unlock(&softc->ctl_lock);
1856         return (0);
1857 }
1858
1859 static int
1860 ctl_init(void)
1861 {
1862         struct make_dev_args args;
1863         struct ctl_softc *softc;
1864         int i, error;
1865
1866         softc = control_softc = malloc(sizeof(*control_softc), M_DEVBUF,
1867                                M_WAITOK | M_ZERO);
1868
1869         make_dev_args_init(&args);
1870         args.mda_devsw = &ctl_cdevsw;
1871         args.mda_uid = UID_ROOT;
1872         args.mda_gid = GID_OPERATOR;
1873         args.mda_mode = 0600;
1874         args.mda_si_drv1 = softc;
1875         args.mda_si_drv2 = NULL;
1876         error = make_dev_s(&args, &softc->dev, "cam/ctl");
1877         if (error != 0) {
1878                 free(softc, M_DEVBUF);
1879                 control_softc = NULL;
1880                 return (error);
1881         }
1882
1883         sysctl_ctx_init(&softc->sysctl_ctx);
1884         softc->sysctl_tree = SYSCTL_ADD_NODE(&softc->sysctl_ctx,
1885                 SYSCTL_STATIC_CHILDREN(_kern_cam), OID_AUTO, "ctl",
1886                 CTLFLAG_RD, 0, "CAM Target Layer");
1887
1888         if (softc->sysctl_tree == NULL) {
1889                 printf("%s: unable to allocate sysctl tree\n", __func__);
1890                 destroy_dev(softc->dev);
1891                 free(softc, M_DEVBUF);
1892                 control_softc = NULL;
1893                 return (ENOMEM);
1894         }
1895
1896         mtx_init(&softc->ctl_lock, "CTL mutex", NULL, MTX_DEF);
1897         softc->io_zone = uma_zcreate("CTL IO", sizeof(union ctl_io),
1898             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1899         softc->flags = 0;
1900
1901         SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1902             OID_AUTO, "ha_mode", CTLFLAG_RDTUN, (int *)&softc->ha_mode, 0,
1903             "HA mode (0 - act/stby, 1 - serialize only, 2 - xfer)");
1904
1905         if (ctl_max_luns <= 0 || powerof2(ctl_max_luns) == 0) {
1906                 printf("Bad value %d for kern.cam.ctl.max_luns, must be a power of two, using %d\n",
1907                     ctl_max_luns, CTL_DEFAULT_MAX_LUNS);
1908                 ctl_max_luns = CTL_DEFAULT_MAX_LUNS;
1909         }
1910         softc->ctl_luns = malloc(sizeof(struct ctl_lun *) * ctl_max_luns,
1911             M_DEVBUF, M_WAITOK | M_ZERO);
1912         softc->ctl_lun_mask = malloc(sizeof(uint32_t) *
1913             ((ctl_max_luns + 31) / 32), M_DEVBUF, M_WAITOK | M_ZERO);
1914         if (ctl_max_ports <= 0 || powerof2(ctl_max_ports) == 0) {
1915                 printf("Bad value %d for kern.cam.ctl.max_ports, must be a power of two, using %d\n",
1916                     ctl_max_ports, CTL_DEFAULT_MAX_PORTS);
1917                 ctl_max_ports = CTL_DEFAULT_MAX_PORTS;
1918         }
1919         softc->ctl_port_mask = malloc(sizeof(uint32_t) *
1920           ((ctl_max_ports + 31) / 32), M_DEVBUF, M_WAITOK | M_ZERO);
1921         softc->ctl_ports = malloc(sizeof(struct ctl_port *) * ctl_max_ports,
1922              M_DEVBUF, M_WAITOK | M_ZERO);
1923
1924
1925         /*
1926          * In Copan's HA scheme, the "master" and "slave" roles are
1927          * figured out through the slot the controller is in.  Although it
1928          * is an active/active system, someone has to be in charge.
1929          */
1930         SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1931             OID_AUTO, "ha_id", CTLFLAG_RDTUN, &softc->ha_id, 0,
1932             "HA head ID (0 - no HA)");
1933         if (softc->ha_id == 0 || softc->ha_id > NUM_HA_SHELVES) {
1934                 softc->flags |= CTL_FLAG_ACTIVE_SHELF;
1935                 softc->is_single = 1;
1936                 softc->port_cnt = ctl_max_ports;
1937                 softc->port_min = 0;
1938         } else {
1939                 softc->port_cnt = ctl_max_ports / NUM_HA_SHELVES;
1940                 softc->port_min = (softc->ha_id - 1) * softc->port_cnt;
1941         }
1942         softc->port_max = softc->port_min + softc->port_cnt;
1943         softc->init_min = softc->port_min * CTL_MAX_INIT_PER_PORT;
1944         softc->init_max = softc->port_max * CTL_MAX_INIT_PER_PORT;
1945
1946         SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1947             OID_AUTO, "ha_link", CTLFLAG_RD, (int *)&softc->ha_link, 0,
1948             "HA link state (0 - offline, 1 - unknown, 2 - online)");
1949
1950         STAILQ_INIT(&softc->lun_list);
1951         STAILQ_INIT(&softc->pending_lun_queue);
1952         STAILQ_INIT(&softc->fe_list);
1953         STAILQ_INIT(&softc->port_list);
1954         STAILQ_INIT(&softc->be_list);
1955         ctl_tpc_init(softc);
1956
1957         if (worker_threads <= 0)
1958                 worker_threads = max(1, mp_ncpus / 4);
1959         if (worker_threads > CTL_MAX_THREADS)
1960                 worker_threads = CTL_MAX_THREADS;
1961
1962         for (i = 0; i < worker_threads; i++) {
1963                 struct ctl_thread *thr = &softc->threads[i];
1964
1965                 mtx_init(&thr->queue_lock, "CTL queue mutex", NULL, MTX_DEF);
1966                 thr->ctl_softc = softc;
1967                 STAILQ_INIT(&thr->incoming_queue);
1968                 STAILQ_INIT(&thr->rtr_queue);
1969                 STAILQ_INIT(&thr->done_queue);
1970                 STAILQ_INIT(&thr->isc_queue);
1971
1972                 error = kproc_kthread_add(ctl_work_thread, thr,
1973                     &softc->ctl_proc, &thr->thread, 0, 0, "ctl", "work%d", i);
1974                 if (error != 0) {
1975                         printf("error creating CTL work thread!\n");
1976                         return (error);
1977                 }
1978         }
1979         error = kproc_kthread_add(ctl_lun_thread, softc,
1980             &softc->ctl_proc, &softc->lun_thread, 0, 0, "ctl", "lun");
1981         if (error != 0) {
1982                 printf("error creating CTL lun thread!\n");
1983                 return (error);
1984         }
1985         error = kproc_kthread_add(ctl_thresh_thread, softc,
1986             &softc->ctl_proc, &softc->thresh_thread, 0, 0, "ctl", "thresh");
1987         if (error != 0) {
1988                 printf("error creating CTL threshold thread!\n");
1989                 return (error);
1990         }
1991
1992         SYSCTL_ADD_PROC(&softc->sysctl_ctx,SYSCTL_CHILDREN(softc->sysctl_tree),
1993             OID_AUTO, "ha_role", CTLTYPE_INT | CTLFLAG_RWTUN,
1994             softc, 0, ctl_ha_role_sysctl, "I", "HA role for this head");
1995
1996         if (softc->is_single == 0) {
1997                 if (ctl_frontend_register(&ha_frontend) != 0)
1998                         softc->is_single = 1;
1999         }
2000         return (0);
2001 }
2002
2003 static int
2004 ctl_shutdown(void)
2005 {
2006         struct ctl_softc *softc = control_softc;
2007         int i;
2008
2009         if (softc->is_single == 0)
2010                 ctl_frontend_deregister(&ha_frontend);
2011
2012         destroy_dev(softc->dev);
2013
2014         /* Shutdown CTL threads. */
2015         softc->shutdown = 1;
2016         for (i = 0; i < worker_threads; i++) {
2017                 struct ctl_thread *thr = &softc->threads[i];
2018                 while (thr->thread != NULL) {
2019                         wakeup(thr);
2020                         if (thr->thread != NULL)
2021                                 pause("CTL thr shutdown", 1);
2022                 }
2023                 mtx_destroy(&thr->queue_lock);
2024         }
2025         while (softc->lun_thread != NULL) {
2026                 wakeup(&softc->pending_lun_queue);
2027                 if (softc->lun_thread != NULL)
2028                         pause("CTL thr shutdown", 1);
2029         }
2030         while (softc->thresh_thread != NULL) {
2031                 wakeup(softc->thresh_thread);
2032                 if (softc->thresh_thread != NULL)
2033                         pause("CTL thr shutdown", 1);
2034         }
2035
2036         ctl_tpc_shutdown(softc);
2037         uma_zdestroy(softc->io_zone);
2038         mtx_destroy(&softc->ctl_lock);
2039
2040         free(softc->ctl_luns, M_DEVBUF);
2041         free(softc->ctl_lun_mask, M_DEVBUF);
2042         free(softc->ctl_port_mask, M_DEVBUF);
2043         free(softc->ctl_ports, M_DEVBUF);
2044
2045         sysctl_ctx_free(&softc->sysctl_ctx);
2046
2047         free(softc, M_DEVBUF);
2048         control_softc = NULL;
2049         return (0);
2050 }
2051
2052 static int
2053 ctl_module_event_handler(module_t mod, int what, void *arg)
2054 {
2055
2056         switch (what) {
2057         case MOD_LOAD:
2058                 return (ctl_init());
2059         case MOD_UNLOAD:
2060                 return (ctl_shutdown());
2061         default:
2062                 return (EOPNOTSUPP);
2063         }
2064 }
2065
2066 /*
2067  * XXX KDM should we do some access checks here?  Bump a reference count to
2068  * prevent a CTL module from being unloaded while someone has it open?
2069  */
2070 static int
2071 ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td)
2072 {
2073         return (0);
2074 }
2075
2076 static int
2077 ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td)
2078 {
2079         return (0);
2080 }
2081
2082 /*
2083  * Remove an initiator by port number and initiator ID.
2084  * Returns 0 for success, -1 for failure.
2085  */
2086 int
2087 ctl_remove_initiator(struct ctl_port *port, int iid)
2088 {
2089         struct ctl_softc *softc = port->ctl_softc;
2090         int last;
2091
2092         mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
2093
2094         if (iid > CTL_MAX_INIT_PER_PORT) {
2095                 printf("%s: initiator ID %u > maximun %u!\n",
2096                        __func__, iid, CTL_MAX_INIT_PER_PORT);
2097                 return (-1);
2098         }
2099
2100         mtx_lock(&softc->ctl_lock);
2101         last = (--port->wwpn_iid[iid].in_use == 0);
2102         port->wwpn_iid[iid].last_use = time_uptime;
2103         mtx_unlock(&softc->ctl_lock);
2104         if (last)
2105                 ctl_i_t_nexus_loss(softc, iid, CTL_UA_POWERON);
2106         ctl_isc_announce_iid(port, iid);
2107
2108         return (0);
2109 }
2110
2111 /*
2112  * Add an initiator to the initiator map.
2113  * Returns iid for success, < 0 for failure.
2114  */
2115 int
2116 ctl_add_initiator(struct ctl_port *port, int iid, uint64_t wwpn, char *name)
2117 {
2118         struct ctl_softc *softc = port->ctl_softc;
2119         time_t best_time;
2120         int i, best;
2121
2122         mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
2123
2124         if (iid >= CTL_MAX_INIT_PER_PORT) {
2125                 printf("%s: WWPN %#jx initiator ID %u > maximum %u!\n",
2126                        __func__, wwpn, iid, CTL_MAX_INIT_PER_PORT);
2127                 free(name, M_CTL);
2128                 return (-1);
2129         }
2130
2131         mtx_lock(&softc->ctl_lock);
2132
2133         if (iid < 0 && (wwpn != 0 || name != NULL)) {
2134                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
2135                         if (wwpn != 0 && wwpn == port->wwpn_iid[i].wwpn) {
2136                                 iid = i;
2137                                 break;
2138                         }
2139                         if (name != NULL && port->wwpn_iid[i].name != NULL &&
2140                             strcmp(name, port->wwpn_iid[i].name) == 0) {
2141                                 iid = i;
2142                                 break;
2143                         }
2144                 }
2145         }
2146
2147         if (iid < 0) {
2148                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
2149                         if (port->wwpn_iid[i].in_use == 0 &&
2150                             port->wwpn_iid[i].wwpn == 0 &&
2151                             port->wwpn_iid[i].name == NULL) {
2152                                 iid = i;
2153                                 break;
2154                         }
2155                 }
2156         }
2157
2158         if (iid < 0) {
2159                 best = -1;
2160                 best_time = INT32_MAX;
2161                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
2162                         if (port->wwpn_iid[i].in_use == 0) {
2163                                 if (port->wwpn_iid[i].last_use < best_time) {
2164                                         best = i;
2165                                         best_time = port->wwpn_iid[i].last_use;
2166                                 }
2167                         }
2168                 }
2169                 iid = best;
2170         }
2171
2172         if (iid < 0) {
2173                 mtx_unlock(&softc->ctl_lock);
2174                 free(name, M_CTL);
2175                 return (-2);
2176         }
2177
2178         if (port->wwpn_iid[iid].in_use > 0 && (wwpn != 0 || name != NULL)) {
2179                 /*
2180                  * This is not an error yet.
2181                  */
2182                 if (wwpn != 0 && wwpn == port->wwpn_iid[iid].wwpn) {
2183 #if 0
2184                         printf("%s: port %d iid %u WWPN %#jx arrived"
2185                             " again\n", __func__, port->targ_port,
2186                             iid, (uintmax_t)wwpn);
2187 #endif
2188                         goto take;
2189                 }
2190                 if (name != NULL && port->wwpn_iid[iid].name != NULL &&
2191                     strcmp(name, port->wwpn_iid[iid].name) == 0) {
2192 #if 0
2193                         printf("%s: port %d iid %u name '%s' arrived"
2194                             " again\n", __func__, port->targ_port,
2195                             iid, name);
2196 #endif
2197                         goto take;
2198                 }
2199
2200                 /*
2201                  * This is an error, but what do we do about it?  The
2202                  * driver is telling us we have a new WWPN for this
2203                  * initiator ID, so we pretty much need to use it.
2204                  */
2205                 printf("%s: port %d iid %u WWPN %#jx '%s' arrived,"
2206                     " but WWPN %#jx '%s' is still at that address\n",
2207                     __func__, port->targ_port, iid, wwpn, name,
2208                     (uintmax_t)port->wwpn_iid[iid].wwpn,
2209                     port->wwpn_iid[iid].name);
2210         }
2211 take:
2212         free(port->wwpn_iid[iid].name, M_CTL);
2213         port->wwpn_iid[iid].name = name;
2214         port->wwpn_iid[iid].wwpn = wwpn;
2215         port->wwpn_iid[iid].in_use++;
2216         mtx_unlock(&softc->ctl_lock);
2217         ctl_isc_announce_iid(port, iid);
2218
2219         return (iid);
2220 }
2221
2222 static int
2223 ctl_create_iid(struct ctl_port *port, int iid, uint8_t *buf)
2224 {
2225         int len;
2226
2227         switch (port->port_type) {
2228         case CTL_PORT_FC:
2229         {
2230                 struct scsi_transportid_fcp *id =
2231                     (struct scsi_transportid_fcp *)buf;
2232                 if (port->wwpn_iid[iid].wwpn == 0)
2233                         return (0);
2234                 memset(id, 0, sizeof(*id));
2235                 id->format_protocol = SCSI_PROTO_FC;
2236                 scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->n_port_name);
2237                 return (sizeof(*id));
2238         }
2239         case CTL_PORT_ISCSI:
2240         {
2241                 struct scsi_transportid_iscsi_port *id =
2242                     (struct scsi_transportid_iscsi_port *)buf;
2243                 if (port->wwpn_iid[iid].name == NULL)
2244                         return (0);
2245                 memset(id, 0, 256);
2246                 id->format_protocol = SCSI_TRN_ISCSI_FORMAT_PORT |
2247                     SCSI_PROTO_ISCSI;
2248                 len = strlcpy(id->iscsi_name, port->wwpn_iid[iid].name, 252) + 1;
2249                 len = roundup2(min(len, 252), 4);
2250                 scsi_ulto2b(len, id->additional_length);
2251                 return (sizeof(*id) + len);
2252         }
2253         case CTL_PORT_SAS:
2254         {
2255                 struct scsi_transportid_sas *id =
2256                     (struct scsi_transportid_sas *)buf;
2257                 if (port->wwpn_iid[iid].wwpn == 0)
2258                         return (0);
2259                 memset(id, 0, sizeof(*id));
2260                 id->format_protocol = SCSI_PROTO_SAS;
2261                 scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->sas_address);
2262                 return (sizeof(*id));
2263         }
2264         default:
2265         {
2266                 struct scsi_transportid_spi *id =
2267                     (struct scsi_transportid_spi *)buf;
2268                 memset(id, 0, sizeof(*id));
2269                 id->format_protocol = SCSI_PROTO_SPI;
2270                 scsi_ulto2b(iid, id->scsi_addr);
2271                 scsi_ulto2b(port->targ_port, id->rel_trgt_port_id);
2272                 return (sizeof(*id));
2273         }
2274         }
2275 }
2276
2277 /*
2278  * Serialize a command that went down the "wrong" side, and so was sent to
2279  * this controller for execution.  The logic is a little different than the
2280  * standard case in ctl_scsiio_precheck().  Errors in this case need to get
2281  * sent back to the other side, but in the success case, we execute the
2282  * command on this side (XFER mode) or tell the other side to execute it
2283  * (SER_ONLY mode).
2284  */
2285 static void
2286 ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio)
2287 {
2288         struct ctl_softc *softc = CTL_SOFTC(ctsio);
2289         struct ctl_port *port = CTL_PORT(ctsio);
2290         union ctl_ha_msg msg_info;
2291         struct ctl_lun *lun;
2292         const struct ctl_cmd_entry *entry;
2293         uint32_t targ_lun;
2294
2295         targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
2296
2297         /* Make sure that we know about this port. */
2298         if (port == NULL || (port->status & CTL_PORT_STATUS_ONLINE) == 0) {
2299                 ctl_set_internal_failure(ctsio, /*sks_valid*/ 0,
2300                                          /*retry_count*/ 1);
2301                 goto badjuju;
2302         }
2303
2304         /* Make sure that we know about this LUN. */
2305         mtx_lock(&softc->ctl_lock);
2306         if (targ_lun >= ctl_max_luns ||
2307             (lun = softc->ctl_luns[targ_lun]) == NULL) {
2308                 mtx_unlock(&softc->ctl_lock);
2309
2310                 /*
2311                  * The other node would not send this request to us unless
2312                  * received announce that we are primary node for this LUN.
2313                  * If this LUN does not exist now, it is probably result of
2314                  * a race, so respond to initiator in the most opaque way.
2315                  */
2316                 ctl_set_busy(ctsio);
2317                 goto badjuju;
2318         }
2319         mtx_lock(&lun->lun_lock);
2320         mtx_unlock(&softc->ctl_lock);
2321
2322         /*
2323          * If the LUN is invalid, pretend that it doesn't exist.
2324          * It will go away as soon as all pending I/Os completed.
2325          */
2326         if (lun->flags & CTL_LUN_DISABLED) {
2327                 mtx_unlock(&lun->lun_lock);
2328                 ctl_set_busy(ctsio);
2329                 goto badjuju;
2330         }
2331
2332         entry = ctl_get_cmd_entry(ctsio, NULL);
2333         if (ctl_scsiio_lun_check(lun, entry, ctsio) != 0) {
2334                 mtx_unlock(&lun->lun_lock);
2335                 goto badjuju;
2336         }
2337
2338         CTL_LUN(ctsio) = lun;
2339         CTL_BACKEND_LUN(ctsio) = lun->be_lun;
2340
2341         /*
2342          * Every I/O goes into the OOA queue for a
2343          * particular LUN, and stays there until completion.
2344          */
2345 #ifdef CTL_TIME_IO
2346         if (TAILQ_EMPTY(&lun->ooa_queue))
2347                 lun->idle_time += getsbinuptime() - lun->last_busy;
2348 #endif
2349         TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2350
2351         switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
2352                 (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr, ctl_ooaq,
2353                  ooa_links))) {
2354         case CTL_ACTION_BLOCK:
2355                 ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
2356                 TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
2357                                   blocked_links);
2358                 mtx_unlock(&lun->lun_lock);
2359                 break;
2360         case CTL_ACTION_PASS:
2361         case CTL_ACTION_SKIP:
2362                 if (softc->ha_mode == CTL_HA_MODE_XFER) {
2363                         ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
2364                         ctl_enqueue_rtr((union ctl_io *)ctsio);
2365                         mtx_unlock(&lun->lun_lock);
2366                 } else {
2367                         ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
2368                         mtx_unlock(&lun->lun_lock);
2369
2370                         /* send msg back to other side */
2371                         msg_info.hdr.original_sc = ctsio->io_hdr.remote_io;
2372                         msg_info.hdr.serializing_sc = (union ctl_io *)ctsio;
2373                         msg_info.hdr.msg_type = CTL_MSG_R2R;
2374                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
2375                             sizeof(msg_info.hdr), M_WAITOK);
2376                 }
2377                 break;
2378         case CTL_ACTION_OVERLAP:
2379                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2380                 mtx_unlock(&lun->lun_lock);
2381                 ctl_set_overlapped_cmd(ctsio);
2382                 goto badjuju;
2383         case CTL_ACTION_OVERLAP_TAG:
2384                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2385                 mtx_unlock(&lun->lun_lock);
2386                 ctl_set_overlapped_tag(ctsio, ctsio->tag_num);
2387                 goto badjuju;
2388         case CTL_ACTION_ERROR:
2389         default:
2390                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2391                 mtx_unlock(&lun->lun_lock);
2392
2393                 ctl_set_internal_failure(ctsio, /*sks_valid*/ 0,
2394                                          /*retry_count*/ 0);
2395 badjuju:
2396                 ctl_copy_sense_data_back((union ctl_io *)ctsio, &msg_info);
2397                 msg_info.hdr.original_sc = ctsio->io_hdr.remote_io;
2398                 msg_info.hdr.serializing_sc = NULL;
2399                 msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
2400                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
2401                     sizeof(msg_info.scsi), M_WAITOK);
2402                 ctl_free_io((union ctl_io *)ctsio);
2403                 break;
2404         }
2405 }
2406
2407 /*
2408  * Returns 0 for success, errno for failure.
2409  */
2410 static void
2411 ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
2412                    struct ctl_ooa *ooa_hdr, struct ctl_ooa_entry *kern_entries)
2413 {
2414         union ctl_io *io;
2415
2416         mtx_lock(&lun->lun_lock);
2417         for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); (io != NULL);
2418              (*cur_fill_num)++, io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2419              ooa_links)) {
2420                 struct ctl_ooa_entry *entry;
2421
2422                 /*
2423                  * If we've got more than we can fit, just count the
2424                  * remaining entries.
2425                  */
2426                 if (*cur_fill_num >= ooa_hdr->alloc_num)
2427                         continue;
2428
2429                 entry = &kern_entries[*cur_fill_num];
2430
2431                 entry->tag_num = io->scsiio.tag_num;
2432                 entry->lun_num = lun->lun;
2433 #ifdef CTL_TIME_IO
2434                 entry->start_bt = io->io_hdr.start_bt;
2435 #endif
2436                 bcopy(io->scsiio.cdb, entry->cdb, io->scsiio.cdb_len);
2437                 entry->cdb_len = io->scsiio.cdb_len;
2438                 if (io->io_hdr.flags & CTL_FLAG_BLOCKED)
2439                         entry->cmd_flags |= CTL_OOACMD_FLAG_BLOCKED;
2440
2441                 if (io->io_hdr.flags & CTL_FLAG_DMA_INPROG)
2442                         entry->cmd_flags |= CTL_OOACMD_FLAG_DMA;
2443
2444                 if (io->io_hdr.flags & CTL_FLAG_ABORT)
2445                         entry->cmd_flags |= CTL_OOACMD_FLAG_ABORT;
2446
2447                 if (io->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR)
2448                         entry->cmd_flags |= CTL_OOACMD_FLAG_RTR;
2449
2450                 if (io->io_hdr.flags & CTL_FLAG_DMA_QUEUED)
2451                         entry->cmd_flags |= CTL_OOACMD_FLAG_DMA_QUEUED;
2452         }
2453         mtx_unlock(&lun->lun_lock);
2454 }
2455
2456 /*
2457  * Escape characters that are illegal or not recommended in XML.
2458  */
2459 int
2460 ctl_sbuf_printf_esc(struct sbuf *sb, char *str, int size)
2461 {
2462         char *end = str + size;
2463         int retval;
2464
2465         retval = 0;
2466
2467         for (; *str && str < end; str++) {
2468                 switch (*str) {
2469                 case '&':
2470                         retval = sbuf_printf(sb, "&amp;");
2471                         break;
2472                 case '>':
2473                         retval = sbuf_printf(sb, "&gt;");
2474                         break;
2475                 case '<':
2476                         retval = sbuf_printf(sb, "&lt;");
2477                         break;
2478                 default:
2479                         retval = sbuf_putc(sb, *str);
2480                         break;
2481                 }
2482
2483                 if (retval != 0)
2484                         break;
2485
2486         }
2487
2488         return (retval);
2489 }
2490
2491 static void
2492 ctl_id_sbuf(struct ctl_devid *id, struct sbuf *sb)
2493 {
2494         struct scsi_vpd_id_descriptor *desc;
2495         int i;
2496
2497         if (id == NULL || id->len < 4)
2498                 return;
2499         desc = (struct scsi_vpd_id_descriptor *)id->data;
2500         switch (desc->id_type & SVPD_ID_TYPE_MASK) {
2501         case SVPD_ID_TYPE_T10:
2502                 sbuf_printf(sb, "t10.");
2503                 break;
2504         case SVPD_ID_TYPE_EUI64:
2505                 sbuf_printf(sb, "eui.");
2506                 break;
2507         case SVPD_ID_TYPE_NAA:
2508                 sbuf_printf(sb, "naa.");
2509                 break;
2510         case SVPD_ID_TYPE_SCSI_NAME:
2511                 break;
2512         }
2513         switch (desc->proto_codeset & SVPD_ID_CODESET_MASK) {
2514         case SVPD_ID_CODESET_BINARY:
2515                 for (i = 0; i < desc->length; i++)
2516                         sbuf_printf(sb, "%02x", desc->identifier[i]);
2517                 break;
2518         case SVPD_ID_CODESET_ASCII:
2519                 sbuf_printf(sb, "%.*s", (int)desc->length,
2520                     (char *)desc->identifier);
2521                 break;
2522         case SVPD_ID_CODESET_UTF8:
2523                 sbuf_printf(sb, "%s", (char *)desc->identifier);
2524                 break;
2525         }
2526 }
2527
2528 static int
2529 ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
2530           struct thread *td)
2531 {
2532         struct ctl_softc *softc = dev->si_drv1;
2533         struct ctl_port *port;
2534         struct ctl_lun *lun;
2535         int retval;
2536
2537         retval = 0;
2538
2539         switch (cmd) {
2540         case CTL_IO:
2541                 retval = ctl_ioctl_io(dev, cmd, addr, flag, td);
2542                 break;
2543         case CTL_ENABLE_PORT:
2544         case CTL_DISABLE_PORT:
2545         case CTL_SET_PORT_WWNS: {
2546                 struct ctl_port *port;
2547                 struct ctl_port_entry *entry;
2548
2549                 entry = (struct ctl_port_entry *)addr;
2550
2551                 mtx_lock(&softc->ctl_lock);
2552                 STAILQ_FOREACH(port, &softc->port_list, links) {
2553                         int action, done;
2554
2555                         if (port->targ_port < softc->port_min ||
2556                             port->targ_port >= softc->port_max)
2557                                 continue;
2558
2559                         action = 0;
2560                         done = 0;
2561                         if ((entry->port_type == CTL_PORT_NONE)
2562                          && (entry->targ_port == port->targ_port)) {
2563                                 /*
2564                                  * If the user only wants to enable or
2565                                  * disable or set WWNs on a specific port,
2566                                  * do the operation and we're done.
2567                                  */
2568                                 action = 1;
2569                                 done = 1;
2570                         } else if (entry->port_type & port->port_type) {
2571                                 /*
2572                                  * Compare the user's type mask with the
2573                                  * particular frontend type to see if we
2574                                  * have a match.
2575                                  */
2576                                 action = 1;
2577                                 done = 0;
2578
2579                                 /*
2580                                  * Make sure the user isn't trying to set
2581                                  * WWNs on multiple ports at the same time.
2582                                  */
2583                                 if (cmd == CTL_SET_PORT_WWNS) {
2584                                         printf("%s: Can't set WWNs on "
2585                                                "multiple ports\n", __func__);
2586                                         retval = EINVAL;
2587                                         break;
2588                                 }
2589                         }
2590                         if (action == 0)
2591                                 continue;
2592
2593                         /*
2594                          * XXX KDM we have to drop the lock here, because
2595                          * the online/offline operations can potentially
2596                          * block.  We need to reference count the frontends
2597                          * so they can't go away,
2598                          */
2599                         if (cmd == CTL_ENABLE_PORT) {
2600                                 mtx_unlock(&softc->ctl_lock);
2601                                 ctl_port_online(port);
2602                                 mtx_lock(&softc->ctl_lock);
2603                         } else if (cmd == CTL_DISABLE_PORT) {
2604                                 mtx_unlock(&softc->ctl_lock);
2605                                 ctl_port_offline(port);
2606                                 mtx_lock(&softc->ctl_lock);
2607                         } else if (cmd == CTL_SET_PORT_WWNS) {
2608                                 ctl_port_set_wwns(port,
2609                                     (entry->flags & CTL_PORT_WWNN_VALID) ?
2610                                     1 : 0, entry->wwnn,
2611                                     (entry->flags & CTL_PORT_WWPN_VALID) ?
2612                                     1 : 0, entry->wwpn);
2613                         }
2614                         if (done != 0)
2615                                 break;
2616                 }
2617                 mtx_unlock(&softc->ctl_lock);
2618                 break;
2619         }
2620         case CTL_GET_OOA: {
2621                 struct ctl_ooa *ooa_hdr;
2622                 struct ctl_ooa_entry *entries;
2623                 uint32_t cur_fill_num;
2624
2625                 ooa_hdr = (struct ctl_ooa *)addr;
2626
2627                 if ((ooa_hdr->alloc_len == 0)
2628                  || (ooa_hdr->alloc_num == 0)) {
2629                         printf("%s: CTL_GET_OOA: alloc len %u and alloc num %u "
2630                                "must be non-zero\n", __func__,
2631                                ooa_hdr->alloc_len, ooa_hdr->alloc_num);
2632                         retval = EINVAL;
2633                         break;
2634                 }
2635
2636                 if (ooa_hdr->alloc_len != (ooa_hdr->alloc_num *
2637                     sizeof(struct ctl_ooa_entry))) {
2638                         printf("%s: CTL_GET_OOA: alloc len %u must be alloc "
2639                                "num %d * sizeof(struct ctl_ooa_entry) %zd\n",
2640                                __func__, ooa_hdr->alloc_len,
2641                                ooa_hdr->alloc_num,sizeof(struct ctl_ooa_entry));
2642                         retval = EINVAL;
2643                         break;
2644                 }
2645
2646                 entries = malloc(ooa_hdr->alloc_len, M_CTL, M_WAITOK | M_ZERO);
2647                 if (entries == NULL) {
2648                         printf("%s: could not allocate %d bytes for OOA "
2649                                "dump\n", __func__, ooa_hdr->alloc_len);
2650                         retval = ENOMEM;
2651                         break;
2652                 }
2653
2654                 mtx_lock(&softc->ctl_lock);
2655                 if ((ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) == 0 &&
2656                     (ooa_hdr->lun_num >= ctl_max_luns ||
2657                      softc->ctl_luns[ooa_hdr->lun_num] == NULL)) {
2658                         mtx_unlock(&softc->ctl_lock);
2659                         free(entries, M_CTL);
2660                         printf("%s: CTL_GET_OOA: invalid LUN %ju\n",
2661                                __func__, (uintmax_t)ooa_hdr->lun_num);
2662                         retval = EINVAL;
2663                         break;
2664                 }
2665
2666                 cur_fill_num = 0;
2667
2668                 if (ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) {
2669                         STAILQ_FOREACH(lun, &softc->lun_list, links) {
2670                                 ctl_ioctl_fill_ooa(lun, &cur_fill_num,
2671                                     ooa_hdr, entries);
2672                         }
2673                 } else {
2674                         lun = softc->ctl_luns[ooa_hdr->lun_num];
2675                         ctl_ioctl_fill_ooa(lun, &cur_fill_num, ooa_hdr,
2676                             entries);
2677                 }
2678                 mtx_unlock(&softc->ctl_lock);
2679
2680                 ooa_hdr->fill_num = min(cur_fill_num, ooa_hdr->alloc_num);
2681                 ooa_hdr->fill_len = ooa_hdr->fill_num *
2682                         sizeof(struct ctl_ooa_entry);
2683                 retval = copyout(entries, ooa_hdr->entries, ooa_hdr->fill_len);
2684                 if (retval != 0) {
2685                         printf("%s: error copying out %d bytes for OOA dump\n",
2686                                __func__, ooa_hdr->fill_len);
2687                 }
2688
2689                 getbinuptime(&ooa_hdr->cur_bt);
2690
2691                 if (cur_fill_num > ooa_hdr->alloc_num) {
2692                         ooa_hdr->dropped_num = cur_fill_num -ooa_hdr->alloc_num;
2693                         ooa_hdr->status = CTL_OOA_NEED_MORE_SPACE;
2694                 } else {
2695                         ooa_hdr->dropped_num = 0;
2696                         ooa_hdr->status = CTL_OOA_OK;
2697                 }
2698
2699                 free(entries, M_CTL);
2700                 break;
2701         }
2702         case CTL_DELAY_IO: {
2703                 struct ctl_io_delay_info *delay_info;
2704
2705                 delay_info = (struct ctl_io_delay_info *)addr;
2706
2707 #ifdef CTL_IO_DELAY
2708                 mtx_lock(&softc->ctl_lock);
2709                 if (delay_info->lun_id >= ctl_max_luns ||
2710                     (lun = softc->ctl_luns[delay_info->lun_id]) == NULL) {
2711                         mtx_unlock(&softc->ctl_lock);
2712                         delay_info->status = CTL_DELAY_STATUS_INVALID_LUN;
2713                         break;
2714                 }
2715                 mtx_lock(&lun->lun_lock);
2716                 mtx_unlock(&softc->ctl_lock);
2717                 delay_info->status = CTL_DELAY_STATUS_OK;
2718                 switch (delay_info->delay_type) {
2719                 case CTL_DELAY_TYPE_CONT:
2720                 case CTL_DELAY_TYPE_ONESHOT:
2721                         break;
2722                 default:
2723                         delay_info->status = CTL_DELAY_STATUS_INVALID_TYPE;
2724                         break;
2725                 }
2726                 switch (delay_info->delay_loc) {
2727                 case CTL_DELAY_LOC_DATAMOVE:
2728                         lun->delay_info.datamove_type = delay_info->delay_type;
2729                         lun->delay_info.datamove_delay = delay_info->delay_secs;
2730                         break;
2731                 case CTL_DELAY_LOC_DONE:
2732                         lun->delay_info.done_type = delay_info->delay_type;
2733                         lun->delay_info.done_delay = delay_info->delay_secs;
2734                         break;
2735                 default:
2736                         delay_info->status = CTL_DELAY_STATUS_INVALID_LOC;
2737                         break;
2738                 }
2739                 mtx_unlock(&lun->lun_lock);
2740 #else
2741                 delay_info->status = CTL_DELAY_STATUS_NOT_IMPLEMENTED;
2742 #endif /* CTL_IO_DELAY */
2743                 break;
2744         }
2745         case CTL_ERROR_INJECT: {
2746                 struct ctl_error_desc *err_desc, *new_err_desc;
2747
2748                 err_desc = (struct ctl_error_desc *)addr;
2749
2750                 new_err_desc = malloc(sizeof(*new_err_desc), M_CTL,
2751                                       M_WAITOK | M_ZERO);
2752                 bcopy(err_desc, new_err_desc, sizeof(*new_err_desc));
2753
2754                 mtx_lock(&softc->ctl_lock);
2755                 if (err_desc->lun_id >= ctl_max_luns ||
2756                     (lun = softc->ctl_luns[err_desc->lun_id]) == NULL) {
2757                         mtx_unlock(&softc->ctl_lock);
2758                         free(new_err_desc, M_CTL);
2759                         printf("%s: CTL_ERROR_INJECT: invalid LUN %ju\n",
2760                                __func__, (uintmax_t)err_desc->lun_id);
2761                         retval = EINVAL;
2762                         break;
2763                 }
2764                 mtx_lock(&lun->lun_lock);
2765                 mtx_unlock(&softc->ctl_lock);
2766
2767                 /*
2768                  * We could do some checking here to verify the validity
2769                  * of the request, but given the complexity of error
2770                  * injection requests, the checking logic would be fairly
2771                  * complex.
2772                  *
2773                  * For now, if the request is invalid, it just won't get
2774                  * executed and might get deleted.
2775                  */
2776                 STAILQ_INSERT_TAIL(&lun->error_list, new_err_desc, links);
2777
2778                 /*
2779                  * XXX KDM check to make sure the serial number is unique,
2780                  * in case we somehow manage to wrap.  That shouldn't
2781                  * happen for a very long time, but it's the right thing to
2782                  * do.
2783                  */
2784                 new_err_desc->serial = lun->error_serial;
2785                 err_desc->serial = lun->error_serial;
2786                 lun->error_serial++;
2787
2788                 mtx_unlock(&lun->lun_lock);
2789                 break;
2790         }
2791         case CTL_ERROR_INJECT_DELETE: {
2792                 struct ctl_error_desc *delete_desc, *desc, *desc2;
2793                 int delete_done;
2794
2795                 delete_desc = (struct ctl_error_desc *)addr;
2796                 delete_done = 0;
2797
2798                 mtx_lock(&softc->ctl_lock);
2799                 if (delete_desc->lun_id >= ctl_max_luns ||
2800                     (lun = softc->ctl_luns[delete_desc->lun_id]) == NULL) {
2801                         mtx_unlock(&softc->ctl_lock);
2802                         printf("%s: CTL_ERROR_INJECT_DELETE: invalid LUN %ju\n",
2803                                __func__, (uintmax_t)delete_desc->lun_id);
2804                         retval = EINVAL;
2805                         break;
2806                 }
2807                 mtx_lock(&lun->lun_lock);
2808                 mtx_unlock(&softc->ctl_lock);
2809                 STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
2810                         if (desc->serial != delete_desc->serial)
2811                                 continue;
2812
2813                         STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc,
2814                                       links);
2815                         free(desc, M_CTL);
2816                         delete_done = 1;
2817                 }
2818                 mtx_unlock(&lun->lun_lock);
2819                 if (delete_done == 0) {
2820                         printf("%s: CTL_ERROR_INJECT_DELETE: can't find "
2821                                "error serial %ju on LUN %u\n", __func__,
2822                                delete_desc->serial, delete_desc->lun_id);
2823                         retval = EINVAL;
2824                         break;
2825                 }
2826                 break;
2827         }
2828         case CTL_DUMP_STRUCTS: {
2829                 int j, k;
2830                 struct ctl_port *port;
2831                 struct ctl_frontend *fe;
2832
2833                 mtx_lock(&softc->ctl_lock);
2834                 printf("CTL Persistent Reservation information start:\n");
2835                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
2836                         mtx_lock(&lun->lun_lock);
2837                         if ((lun->flags & CTL_LUN_DISABLED) != 0) {
2838                                 mtx_unlock(&lun->lun_lock);
2839                                 continue;
2840                         }
2841
2842                         for (j = 0; j < ctl_max_ports; j++) {
2843                                 if (lun->pr_keys[j] == NULL)
2844                                         continue;
2845                                 for (k = 0; k < CTL_MAX_INIT_PER_PORT; k++){
2846                                         if (lun->pr_keys[j][k] == 0)
2847                                                 continue;
2848                                         printf("  LUN %ju port %d iid %d key "
2849                                                "%#jx\n", lun->lun, j, k,
2850                                                (uintmax_t)lun->pr_keys[j][k]);
2851                                 }
2852                         }
2853                         mtx_unlock(&lun->lun_lock);
2854                 }
2855                 printf("CTL Persistent Reservation information end\n");
2856                 printf("CTL Ports:\n");
2857                 STAILQ_FOREACH(port, &softc->port_list, links) {
2858                         printf("  Port %d '%s' Frontend '%s' Type %u pp %d vp %d WWNN "
2859                                "%#jx WWPN %#jx\n", port->targ_port, port->port_name,
2860                                port->frontend->name, port->port_type,
2861                                port->physical_port, port->virtual_port,
2862                                (uintmax_t)port->wwnn, (uintmax_t)port->wwpn);
2863                         for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
2864                                 if (port->wwpn_iid[j].in_use == 0 &&
2865                                     port->wwpn_iid[j].wwpn == 0 &&
2866                                     port->wwpn_iid[j].name == NULL)
2867                                         continue;
2868
2869                                 printf("    iid %u use %d WWPN %#jx '%s'\n",
2870                                     j, port->wwpn_iid[j].in_use,
2871                                     (uintmax_t)port->wwpn_iid[j].wwpn,
2872                                     port->wwpn_iid[j].name);
2873                         }
2874                 }
2875                 printf("CTL Port information end\n");
2876                 mtx_unlock(&softc->ctl_lock);
2877                 /*
2878                  * XXX KDM calling this without a lock.  We'd likely want
2879                  * to drop the lock before calling the frontend's dump
2880                  * routine anyway.
2881                  */
2882                 printf("CTL Frontends:\n");
2883                 STAILQ_FOREACH(fe, &softc->fe_list, links) {
2884                         printf("  Frontend '%s'\n", fe->name);
2885                         if (fe->fe_dump != NULL)
2886                                 fe->fe_dump();
2887                 }
2888                 printf("CTL Frontend information end\n");
2889                 break;
2890         }
2891         case CTL_LUN_REQ: {
2892                 struct ctl_lun_req *lun_req;
2893                 struct ctl_backend_driver *backend;
2894                 void *packed;
2895                 nvlist_t *tmp_args_nvl;
2896                 size_t packed_len;
2897
2898                 lun_req = (struct ctl_lun_req *)addr;
2899                 tmp_args_nvl = lun_req->args_nvl;
2900
2901                 backend = ctl_backend_find(lun_req->backend);
2902                 if (backend == NULL) {
2903                         lun_req->status = CTL_LUN_ERROR;
2904                         snprintf(lun_req->error_str,
2905                                  sizeof(lun_req->error_str),
2906                                  "Backend \"%s\" not found.",
2907                                  lun_req->backend);
2908                         break;
2909                 }
2910
2911                 if (lun_req->args != NULL) {
2912                         packed = malloc(lun_req->args_len, M_CTL, M_WAITOK);
2913                         if (copyin(lun_req->args, packed, lun_req->args_len) != 0) {
2914                                 free(packed, M_CTL);
2915                                 lun_req->status = CTL_LUN_ERROR;
2916                                 snprintf(lun_req->error_str, sizeof(lun_req->error_str),
2917                                     "Cannot copyin args.");
2918                                 break;
2919                         }
2920                         lun_req->args_nvl = nvlist_unpack(packed,
2921                             lun_req->args_len, 0);
2922                         free(packed, M_CTL);
2923
2924                         if (lun_req->args_nvl == NULL) {
2925                                 lun_req->status = CTL_LUN_ERROR;
2926                                 snprintf(lun_req->error_str, sizeof(lun_req->error_str),
2927                                     "Cannot unpack args nvlist.");
2928                                 break;
2929                         }
2930                 } else
2931                         lun_req->args_nvl = nvlist_create(0);
2932
2933                 retval = backend->ioctl(dev, cmd, addr, flag, td);
2934                 nvlist_destroy(lun_req->args_nvl);
2935                 lun_req->args_nvl = tmp_args_nvl;
2936
2937                 if (lun_req->result_nvl != NULL) {
2938                         if (lun_req->result != NULL) {
2939                                 packed = nvlist_pack(lun_req->result_nvl,
2940                                     &packed_len);
2941                                 if (packed == NULL) {
2942                                         lun_req->status = CTL_LUN_ERROR;
2943                                         snprintf(lun_req->error_str,
2944                                             sizeof(lun_req->error_str),
2945                                             "Cannot pack result nvlist.");
2946                                         break;
2947                                 }
2948
2949                                 if (packed_len > lun_req->result_len) {
2950                                         lun_req->status = CTL_LUN_ERROR;
2951                                         snprintf(lun_req->error_str,
2952                                             sizeof(lun_req->error_str),
2953                                             "Result nvlist too large.");
2954                                         free(packed, M_NVLIST);
2955                                         break;
2956                                 }
2957
2958                                 if (copyout(packed, lun_req->result, packed_len)) {
2959                                         lun_req->status = CTL_LUN_ERROR;
2960                                         snprintf(lun_req->error_str,
2961                                             sizeof(lun_req->error_str),
2962                                             "Cannot copyout() the result.");
2963                                         free(packed, M_NVLIST);
2964                                         break;
2965                                 }
2966
2967                                 lun_req->result_len = packed_len;
2968                                 free(packed, M_NVLIST);
2969                         }
2970
2971                         nvlist_destroy(lun_req->result_nvl);
2972                 }
2973                 break;
2974         }
2975         case CTL_LUN_LIST: {
2976                 struct sbuf *sb;
2977                 struct ctl_lun_list *list;
2978                 const char *name, *value;
2979                 void *cookie;
2980                 int type;
2981
2982                 list = (struct ctl_lun_list *)addr;
2983
2984                 /*
2985                  * Allocate a fixed length sbuf here, based on the length
2986                  * of the user's buffer.  We could allocate an auto-extending
2987                  * buffer, and then tell the user how much larger our
2988                  * amount of data is than his buffer, but that presents
2989                  * some problems:
2990                  *
2991                  * 1.  The sbuf(9) routines use a blocking malloc, and so
2992                  *     we can't hold a lock while calling them with an
2993                  *     auto-extending buffer.
2994                  *
2995                  * 2.  There is not currently a LUN reference counting
2996                  *     mechanism, outside of outstanding transactions on
2997                  *     the LUN's OOA queue.  So a LUN could go away on us
2998                  *     while we're getting the LUN number, backend-specific
2999                  *     information, etc.  Thus, given the way things
3000                  *     currently work, we need to hold the CTL lock while
3001                  *     grabbing LUN information.
3002                  *
3003                  * So, from the user's standpoint, the best thing to do is
3004                  * allocate what he thinks is a reasonable buffer length,
3005                  * and then if he gets a CTL_LUN_LIST_NEED_MORE_SPACE error,
3006                  * double the buffer length and try again.  (And repeat
3007                  * that until he succeeds.)
3008                  */
3009                 sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3010                 if (sb == NULL) {
3011                         list->status = CTL_LUN_LIST_ERROR;
3012                         snprintf(list->error_str, sizeof(list->error_str),
3013                                  "Unable to allocate %d bytes for LUN list",
3014                                  list->alloc_len);
3015                         break;
3016                 }
3017
3018                 sbuf_printf(sb, "<ctllunlist>\n");
3019
3020                 mtx_lock(&softc->ctl_lock);
3021                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
3022                         mtx_lock(&lun->lun_lock);
3023                         retval = sbuf_printf(sb, "<lun id=\"%ju\">\n",
3024                                              (uintmax_t)lun->lun);
3025
3026                         /*
3027                          * Bail out as soon as we see that we've overfilled
3028                          * the buffer.
3029                          */
3030                         if (retval != 0)
3031                                 break;
3032
3033                         retval = sbuf_printf(sb, "\t<backend_type>%s"
3034                                              "</backend_type>\n",
3035                                              (lun->backend == NULL) ?  "none" :
3036                                              lun->backend->name);
3037
3038                         if (retval != 0)
3039                                 break;
3040
3041                         retval = sbuf_printf(sb, "\t<lun_type>%d</lun_type>\n",
3042                                              lun->be_lun->lun_type);
3043
3044                         if (retval != 0)
3045                                 break;
3046
3047                         if (lun->backend == NULL) {
3048                                 retval = sbuf_printf(sb, "</lun>\n");
3049                                 if (retval != 0)
3050                                         break;
3051                                 continue;
3052                         }
3053
3054                         retval = sbuf_printf(sb, "\t<size>%ju</size>\n",
3055                                              (lun->be_lun->maxlba > 0) ?
3056                                              lun->be_lun->maxlba + 1 : 0);
3057
3058                         if (retval != 0)
3059                                 break;
3060
3061                         retval = sbuf_printf(sb, "\t<blocksize>%u</blocksize>\n",
3062                                              lun->be_lun->blocksize);
3063
3064                         if (retval != 0)
3065                                 break;
3066
3067                         retval = sbuf_printf(sb, "\t<serial_number>");
3068
3069                         if (retval != 0)
3070                                 break;
3071
3072                         retval = ctl_sbuf_printf_esc(sb,
3073                             lun->be_lun->serial_num,
3074                             sizeof(lun->be_lun->serial_num));
3075
3076                         if (retval != 0)
3077                                 break;
3078
3079                         retval = sbuf_printf(sb, "</serial_number>\n");
3080
3081                         if (retval != 0)
3082                                 break;
3083
3084                         retval = sbuf_printf(sb, "\t<device_id>");
3085
3086                         if (retval != 0)
3087                                 break;
3088
3089                         retval = ctl_sbuf_printf_esc(sb,
3090                             lun->be_lun->device_id,
3091                             sizeof(lun->be_lun->device_id));
3092
3093                         if (retval != 0)
3094                                 break;
3095
3096                         retval = sbuf_printf(sb, "</device_id>\n");
3097
3098                         if (retval != 0)
3099                                 break;
3100
3101                         if (lun->backend->lun_info != NULL) {
3102                                 retval = lun->backend->lun_info(lun->be_lun->be_lun, sb);
3103                                 if (retval != 0)
3104                                         break;
3105                         }
3106
3107                         cookie = NULL;
3108                         while ((name = nvlist_next(lun->be_lun->options, &type,
3109                             &cookie)) != NULL) {
3110                                 sbuf_printf(sb, "\t<%s>", name);
3111
3112                                 if (type == NV_TYPE_STRING) {
3113                                         value = dnvlist_get_string(
3114                                             lun->be_lun->options, name, NULL);
3115                                         if (value != NULL)
3116                                                 sbuf_printf(sb, "%s", value);
3117                                 }
3118
3119                                 sbuf_printf(sb, "</%s>\n", name);
3120                         }
3121
3122                         retval = sbuf_printf(sb, "</lun>\n");
3123
3124                         if (retval != 0)
3125                                 break;
3126                         mtx_unlock(&lun->lun_lock);
3127                 }
3128                 if (lun != NULL)
3129                         mtx_unlock(&lun->lun_lock);
3130                 mtx_unlock(&softc->ctl_lock);
3131
3132                 if ((retval != 0)
3133                  || ((retval = sbuf_printf(sb, "</ctllunlist>\n")) != 0)) {
3134                         retval = 0;
3135                         sbuf_delete(sb);
3136                         list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3137                         snprintf(list->error_str, sizeof(list->error_str),
3138                                  "Out of space, %d bytes is too small",
3139                                  list->alloc_len);
3140                         break;
3141                 }
3142
3143                 sbuf_finish(sb);
3144
3145                 retval = copyout(sbuf_data(sb), list->lun_xml,
3146                                  sbuf_len(sb) + 1);
3147
3148                 list->fill_len = sbuf_len(sb) + 1;
3149                 list->status = CTL_LUN_LIST_OK;
3150                 sbuf_delete(sb);
3151                 break;
3152         }
3153         case CTL_ISCSI: {
3154                 struct ctl_iscsi *ci;
3155                 struct ctl_frontend *fe;
3156
3157                 ci = (struct ctl_iscsi *)addr;
3158
3159                 fe = ctl_frontend_find("iscsi");
3160                 if (fe == NULL) {
3161                         ci->status = CTL_ISCSI_ERROR;
3162                         snprintf(ci->error_str, sizeof(ci->error_str),
3163                             "Frontend \"iscsi\" not found.");
3164                         break;
3165                 }
3166
3167                 retval = fe->ioctl(dev, cmd, addr, flag, td);
3168                 break;
3169         }
3170         case CTL_PORT_REQ: {
3171                 struct ctl_req *req;
3172                 struct ctl_frontend *fe;
3173                 void *packed;
3174                 nvlist_t *tmp_args_nvl;
3175                 size_t packed_len;
3176
3177                 req = (struct ctl_req *)addr;
3178                 tmp_args_nvl = req->args_nvl;
3179
3180                 fe = ctl_frontend_find(req->driver);
3181                 if (fe == NULL) {
3182                         req->status = CTL_LUN_ERROR;
3183                         snprintf(req->error_str, sizeof(req->error_str),
3184                             "Frontend \"%s\" not found.", req->driver);
3185                         break;
3186                 }
3187
3188                 if (req->args != NULL) {
3189                         packed = malloc(req->args_len, M_CTL, M_WAITOK);
3190                         if (copyin(req->args, packed, req->args_len) != 0) {
3191                                 free(packed, M_CTL);
3192                                 req->status = CTL_LUN_ERROR;
3193                                 snprintf(req->error_str, sizeof(req->error_str),
3194                                     "Cannot copyin args.");
3195                                 break;
3196                         }
3197                         req->args_nvl = nvlist_unpack(packed,
3198                             req->args_len, 0);
3199                         free(packed, M_CTL);
3200
3201                         if (req->args_nvl == NULL) {
3202                                 req->status = CTL_LUN_ERROR;
3203                                 snprintf(req->error_str, sizeof(req->error_str),
3204                                     "Cannot unpack args nvlist.");
3205                                 break;
3206                         }
3207                 } else
3208                         req->args_nvl = nvlist_create(0);
3209
3210                 if (fe->ioctl)
3211                         retval = fe->ioctl(dev, cmd, addr, flag, td);
3212                 else
3213                         retval = ENODEV;
3214
3215                 nvlist_destroy(req->args_nvl);
3216                 req->args_nvl = tmp_args_nvl;
3217
3218                 if (req->result_nvl != NULL) {
3219                         if (req->result != NULL) {
3220                                 packed = nvlist_pack(req->result_nvl,
3221                                     &packed_len);
3222                                 if (packed == NULL) {
3223                                         req->status = CTL_LUN_ERROR;
3224                                         snprintf(req->error_str,
3225                                             sizeof(req->error_str),
3226                                             "Cannot pack result nvlist.");
3227                                         break;
3228                                 }
3229
3230                                 if (packed_len > req->result_len) {
3231                                         req->status = CTL_LUN_ERROR;
3232                                         snprintf(req->error_str,
3233                                             sizeof(req->error_str),
3234                                             "Result nvlist too large.");
3235                                         free(packed, M_NVLIST);
3236                                         break;
3237                                 }
3238
3239                                 if (copyout(packed, req->result, packed_len)) {
3240                                         req->status = CTL_LUN_ERROR;
3241                                         snprintf(req->error_str,
3242                                             sizeof(req->error_str),
3243                                             "Cannot copyout() the result.");
3244                                         free(packed, M_NVLIST);
3245                                         break;
3246                                 }
3247
3248                                 req->result_len = packed_len;
3249                                 free(packed, M_NVLIST);
3250                         }
3251
3252                         nvlist_destroy(req->result_nvl);
3253                 }
3254                 break;
3255         }
3256         case CTL_PORT_LIST: {
3257                 struct sbuf *sb;
3258                 struct ctl_port *port;
3259                 struct ctl_lun_list *list;
3260                 const char *name, *value;
3261                 void *cookie;
3262                 int j, type;
3263                 uint32_t plun;
3264
3265                 list = (struct ctl_lun_list *)addr;
3266
3267                 sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3268                 if (sb == NULL) {
3269                         list->status = CTL_LUN_LIST_ERROR;
3270                         snprintf(list->error_str, sizeof(list->error_str),
3271                                  "Unable to allocate %d bytes for LUN list",
3272                                  list->alloc_len);
3273                         break;
3274                 }
3275
3276                 sbuf_printf(sb, "<ctlportlist>\n");
3277
3278                 mtx_lock(&softc->ctl_lock);
3279                 STAILQ_FOREACH(port, &softc->port_list, links) {
3280                         retval = sbuf_printf(sb, "<targ_port id=\"%ju\">\n",
3281                                              (uintmax_t)port->targ_port);
3282
3283                         /*
3284                          * Bail out as soon as we see that we've overfilled
3285                          * the buffer.
3286                          */
3287                         if (retval != 0)
3288                                 break;
3289
3290                         retval = sbuf_printf(sb, "\t<frontend_type>%s"
3291                             "</frontend_type>\n", port->frontend->name);
3292                         if (retval != 0)
3293                                 break;
3294
3295                         retval = sbuf_printf(sb, "\t<port_type>%d</port_type>\n",
3296                                              port->port_type);
3297                         if (retval != 0)
3298                                 break;
3299
3300                         retval = sbuf_printf(sb, "\t<online>%s</online>\n",
3301                             (port->status & CTL_PORT_STATUS_ONLINE) ? "YES" : "NO");
3302                         if (retval != 0)
3303                                 break;
3304
3305                         retval = sbuf_printf(sb, "\t<port_name>%s</port_name>\n",
3306                             port->port_name);
3307                         if (retval != 0)
3308                                 break;
3309
3310                         retval = sbuf_printf(sb, "\t<physical_port>%d</physical_port>\n",
3311                             port->physical_port);
3312                         if (retval != 0)
3313                                 break;
3314
3315                         retval = sbuf_printf(sb, "\t<virtual_port>%d</virtual_port>\n",
3316                             port->virtual_port);
3317                         if (retval != 0)
3318                                 break;
3319
3320                         if (port->target_devid != NULL) {
3321                                 sbuf_printf(sb, "\t<target>");
3322                                 ctl_id_sbuf(port->target_devid, sb);
3323                                 sbuf_printf(sb, "</target>\n");
3324                         }
3325
3326                         if (port->port_devid != NULL) {
3327                                 sbuf_printf(sb, "\t<port>");
3328                                 ctl_id_sbuf(port->port_devid, sb);
3329                                 sbuf_printf(sb, "</port>\n");
3330                         }
3331
3332                         if (port->port_info != NULL) {
3333                                 retval = port->port_info(port->onoff_arg, sb);
3334                                 if (retval != 0)
3335                                         break;
3336                         }
3337
3338                         cookie = NULL;
3339                         while ((name = nvlist_next(port->options, &type,
3340                             &cookie)) != NULL) {
3341                                 sbuf_printf(sb, "\t<%s>", name);
3342
3343                                 if (type == NV_TYPE_STRING) {
3344                                         value = dnvlist_get_string(port->options,
3345                                             name, NULL);
3346                                         if (value != NULL)
3347                                                 sbuf_printf(sb, "%s", value);
3348                                 }
3349
3350                                 sbuf_printf(sb, "</%s>\n", name);
3351                         }
3352
3353                         if (port->lun_map != NULL) {
3354                                 sbuf_printf(sb, "\t<lun_map>on</lun_map>\n");
3355                                 for (j = 0; j < port->lun_map_size; j++) {
3356                                         plun = ctl_lun_map_from_port(port, j);
3357                                         if (plun == UINT32_MAX)
3358                                                 continue;
3359                                         sbuf_printf(sb,
3360                                             "\t<lun id=\"%u\">%u</lun>\n",
3361                                             j, plun);
3362                                 }
3363                         }
3364
3365                         for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
3366                                 if (port->wwpn_iid[j].in_use == 0 ||
3367                                     (port->wwpn_iid[j].wwpn == 0 &&
3368                                      port->wwpn_iid[j].name == NULL))
3369                                         continue;
3370
3371                                 if (port->wwpn_iid[j].name != NULL)
3372                                         retval = sbuf_printf(sb,
3373                                             "\t<initiator id=\"%u\">%s</initiator>\n",
3374                                             j, port->wwpn_iid[j].name);
3375                                 else
3376                                         retval = sbuf_printf(sb,
3377                                             "\t<initiator id=\"%u\">naa.%08jx</initiator>\n",
3378                                             j, port->wwpn_iid[j].wwpn);
3379                                 if (retval != 0)
3380                                         break;
3381                         }
3382                         if (retval != 0)
3383                                 break;
3384
3385                         retval = sbuf_printf(sb, "</targ_port>\n");
3386                         if (retval != 0)
3387                                 break;
3388                 }
3389                 mtx_unlock(&softc->ctl_lock);
3390
3391                 if ((retval != 0)
3392                  || ((retval = sbuf_printf(sb, "</ctlportlist>\n")) != 0)) {
3393                         retval = 0;
3394                         sbuf_delete(sb);
3395                         list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3396                         snprintf(list->error_str, sizeof(list->error_str),
3397                                  "Out of space, %d bytes is too small",
3398                                  list->alloc_len);
3399                         break;
3400                 }
3401
3402                 sbuf_finish(sb);
3403
3404                 retval = copyout(sbuf_data(sb), list->lun_xml,
3405                                  sbuf_len(sb) + 1);
3406
3407                 list->fill_len = sbuf_len(sb) + 1;
3408                 list->status = CTL_LUN_LIST_OK;
3409                 sbuf_delete(sb);
3410                 break;
3411         }
3412         case CTL_LUN_MAP: {
3413                 struct ctl_lun_map *lm  = (struct ctl_lun_map *)addr;
3414                 struct ctl_port *port;
3415
3416                 mtx_lock(&softc->ctl_lock);
3417                 if (lm->port < softc->port_min ||
3418                     lm->port >= softc->port_max ||
3419                     (port = softc->ctl_ports[lm->port]) == NULL) {
3420                         mtx_unlock(&softc->ctl_lock);
3421                         return (ENXIO);
3422                 }
3423                 if (port->status & CTL_PORT_STATUS_ONLINE) {
3424                         STAILQ_FOREACH(lun, &softc->lun_list, links) {
3425                                 if (ctl_lun_map_to_port(port, lun->lun) ==
3426                                     UINT32_MAX)
3427                                         continue;
3428                                 mtx_lock(&lun->lun_lock);
3429                                 ctl_est_ua_port(lun, lm->port, -1,
3430                                     CTL_UA_LUN_CHANGE);
3431                                 mtx_unlock(&lun->lun_lock);
3432                         }
3433                 }
3434                 mtx_unlock(&softc->ctl_lock); // XXX: port_enable sleeps
3435                 if (lm->plun != UINT32_MAX) {
3436                         if (lm->lun == UINT32_MAX)
3437                                 retval = ctl_lun_map_unset(port, lm->plun);
3438                         else if (lm->lun < ctl_max_luns &&
3439                             softc->ctl_luns[lm->lun] != NULL)
3440                                 retval = ctl_lun_map_set(port, lm->plun, lm->lun);
3441                         else
3442                                 return (ENXIO);
3443                 } else {
3444                         if (lm->lun == UINT32_MAX)
3445                                 retval = ctl_lun_map_deinit(port);
3446                         else
3447                                 retval = ctl_lun_map_init(port);
3448                 }
3449                 if (port->status & CTL_PORT_STATUS_ONLINE)
3450                         ctl_isc_announce_port(port);
3451                 break;
3452         }
3453         case CTL_GET_LUN_STATS: {
3454                 struct ctl_get_io_stats *stats = (struct ctl_get_io_stats *)addr;
3455                 int i;
3456
3457                 /*
3458                  * XXX KDM no locking here.  If the LUN list changes,
3459                  * things can blow up.
3460                  */
3461                 i = 0;
3462                 stats->status = CTL_SS_OK;
3463                 stats->fill_len = 0;
3464                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
3465                         if (lun->lun < stats->first_item)
3466                                 continue;
3467                         if (stats->fill_len + sizeof(lun->stats) >
3468                             stats->alloc_len) {
3469                                 stats->status = CTL_SS_NEED_MORE_SPACE;
3470                                 break;
3471                         }
3472                         retval = copyout(&lun->stats, &stats->stats[i++],
3473                                          sizeof(lun->stats));
3474                         if (retval != 0)
3475                                 break;
3476                         stats->fill_len += sizeof(lun->stats);
3477                 }
3478                 stats->num_items = softc->num_luns;
3479                 stats->flags = CTL_STATS_FLAG_NONE;
3480 #ifdef CTL_TIME_IO
3481                 stats->flags |= CTL_STATS_FLAG_TIME_VALID;
3482 #endif
3483                 getnanouptime(&stats->timestamp);
3484                 break;
3485         }
3486         case CTL_GET_PORT_STATS: {
3487                 struct ctl_get_io_stats *stats = (struct ctl_get_io_stats *)addr;
3488                 int i;
3489
3490                 /*
3491                  * XXX KDM no locking here.  If the LUN list changes,
3492                  * things can blow up.
3493                  */
3494                 i = 0;
3495                 stats->status = CTL_SS_OK;
3496                 stats->fill_len = 0;
3497                 STAILQ_FOREACH(port, &softc->port_list, links) {
3498                         if (port->targ_port < stats->first_item)
3499                                 continue;
3500                         if (stats->fill_len + sizeof(port->stats) >
3501                             stats->alloc_len) {
3502                                 stats->status = CTL_SS_NEED_MORE_SPACE;
3503                                 break;
3504                         }
3505                         retval = copyout(&port->stats, &stats->stats[i++],
3506                                          sizeof(port->stats));
3507                         if (retval != 0)
3508                                 break;
3509                         stats->fill_len += sizeof(port->stats);
3510                 }
3511                 stats->num_items = softc->num_ports;
3512                 stats->flags = CTL_STATS_FLAG_NONE;
3513 #ifdef CTL_TIME_IO
3514                 stats->flags |= CTL_STATS_FLAG_TIME_VALID;
3515 #endif
3516                 getnanouptime(&stats->timestamp);
3517                 break;
3518         }
3519         default: {
3520                 /* XXX KDM should we fix this? */
3521 #if 0
3522                 struct ctl_backend_driver *backend;
3523                 unsigned int type;
3524                 int found;
3525
3526                 found = 0;
3527
3528                 /*
3529                  * We encode the backend type as the ioctl type for backend
3530                  * ioctls.  So parse it out here, and then search for a
3531                  * backend of this type.
3532                  */
3533                 type = _IOC_TYPE(cmd);
3534
3535                 STAILQ_FOREACH(backend, &softc->be_list, links) {
3536                         if (backend->type == type) {
3537                                 found = 1;
3538                                 break;
3539                         }
3540                 }
3541                 if (found == 0) {
3542                         printf("ctl: unknown ioctl command %#lx or backend "
3543                                "%d\n", cmd, type);
3544                         retval = EINVAL;
3545                         break;
3546                 }
3547                 retval = backend->ioctl(dev, cmd, addr, flag, td);
3548 #endif
3549                 retval = ENOTTY;
3550                 break;
3551         }
3552         }
3553         return (retval);
3554 }
3555
3556 uint32_t
3557 ctl_get_initindex(struct ctl_nexus *nexus)
3558 {
3559         return (nexus->initid + (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3560 }
3561
3562 int
3563 ctl_lun_map_init(struct ctl_port *port)
3564 {
3565         struct ctl_softc *softc = port->ctl_softc;
3566         struct ctl_lun *lun;
3567         int size = ctl_lun_map_size;
3568         uint32_t i;
3569
3570         if (port->lun_map == NULL || port->lun_map_size < size) {
3571                 port->lun_map_size = 0;
3572                 free(port->lun_map, M_CTL);
3573                 port->lun_map = malloc(size * sizeof(uint32_t),
3574                     M_CTL, M_NOWAIT);
3575         }
3576         if (port->lun_map == NULL)
3577                 return (ENOMEM);
3578         for (i = 0; i < size; i++)
3579                 port->lun_map[i] = UINT32_MAX;
3580         port->lun_map_size = size;
3581         if (port->status & CTL_PORT_STATUS_ONLINE) {
3582                 if (port->lun_disable != NULL) {
3583                         STAILQ_FOREACH(lun, &softc->lun_list, links)
3584                                 port->lun_disable(port->targ_lun_arg, lun->lun);
3585                 }
3586                 ctl_isc_announce_port(port);
3587         }
3588         return (0);
3589 }
3590
3591 int
3592 ctl_lun_map_deinit(struct ctl_port *port)
3593 {
3594         struct ctl_softc *softc = port->ctl_softc;
3595         struct ctl_lun *lun;
3596
3597         if (port->lun_map == NULL)
3598                 return (0);
3599         port->lun_map_size = 0;
3600         free(port->lun_map, M_CTL);
3601         port->lun_map = NULL;
3602         if (port->status & CTL_PORT_STATUS_ONLINE) {
3603                 if (port->lun_enable != NULL) {
3604                         STAILQ_FOREACH(lun, &softc->lun_list, links)
3605                                 port->lun_enable(port->targ_lun_arg, lun->lun);
3606                 }
3607                 ctl_isc_announce_port(port);
3608         }
3609         return (0);
3610 }
3611
3612 int
3613 ctl_lun_map_set(struct ctl_port *port, uint32_t plun, uint32_t glun)
3614 {
3615         int status;
3616         uint32_t old;
3617
3618         if (port->lun_map == NULL) {
3619                 status = ctl_lun_map_init(port);
3620                 if (status != 0)
3621                         return (status);
3622         }
3623         if (plun >= port->lun_map_size)
3624                 return (EINVAL);
3625         old = port->lun_map[plun];
3626         port->lun_map[plun] = glun;
3627         if ((port->status & CTL_PORT_STATUS_ONLINE) && old == UINT32_MAX) {
3628                 if (port->lun_enable != NULL)
3629                         port->lun_enable(port->targ_lun_arg, plun);
3630                 ctl_isc_announce_port(port);
3631         }
3632         return (0);
3633 }
3634
3635 int
3636 ctl_lun_map_unset(struct ctl_port *port, uint32_t plun)
3637 {
3638         uint32_t old;
3639
3640         if (port->lun_map == NULL || plun >= port->lun_map_size)
3641                 return (0);
3642         old = port->lun_map[plun];
3643         port->lun_map[plun] = UINT32_MAX;
3644         if ((port->status & CTL_PORT_STATUS_ONLINE) && old != UINT32_MAX) {
3645                 if (port->lun_disable != NULL)
3646                         port->lun_disable(port->targ_lun_arg, plun);
3647                 ctl_isc_announce_port(port);
3648         }
3649         return (0);
3650 }
3651
3652 uint32_t
3653 ctl_lun_map_from_port(struct ctl_port *port, uint32_t lun_id)
3654 {
3655
3656         if (port == NULL)
3657                 return (UINT32_MAX);
3658         if (port->lun_map == NULL)
3659                 return (lun_id);
3660         if (lun_id > port->lun_map_size)
3661                 return (UINT32_MAX);
3662         return (port->lun_map[lun_id]);
3663 }
3664
3665 uint32_t
3666 ctl_lun_map_to_port(struct ctl_port *port, uint32_t lun_id)
3667 {
3668         uint32_t i;
3669
3670         if (port == NULL)
3671                 return (UINT32_MAX);
3672         if (port->lun_map == NULL)
3673                 return (lun_id);
3674         for (i = 0; i < port->lun_map_size; i++) {
3675                 if (port->lun_map[i] == lun_id)
3676                         return (i);
3677         }
3678         return (UINT32_MAX);
3679 }
3680
3681 uint32_t
3682 ctl_decode_lun(uint64_t encoded)
3683 {
3684         uint8_t lun[8];
3685         uint32_t result = 0xffffffff;
3686
3687         be64enc(lun, encoded);
3688         switch (lun[0] & RPL_LUNDATA_ATYP_MASK) {
3689         case RPL_LUNDATA_ATYP_PERIPH:
3690                 if ((lun[0] & 0x3f) == 0 && lun[2] == 0 && lun[3] == 0 &&
3691                     lun[4] == 0 && lun[5] == 0 && lun[6] == 0 && lun[7] == 0)
3692                         result = lun[1];
3693                 break;
3694         case RPL_LUNDATA_ATYP_FLAT:
3695                 if (lun[2] == 0 && lun[3] == 0 && lun[4] == 0 && lun[5] == 0 &&
3696                     lun[6] == 0 && lun[7] == 0)
3697                         result = ((lun[0] & 0x3f) << 8) + lun[1];
3698                 break;
3699         case RPL_LUNDATA_ATYP_EXTLUN:
3700                 switch (lun[0] & RPL_LUNDATA_EXT_EAM_MASK) {
3701                 case 0x02:
3702                         switch (lun[0] & RPL_LUNDATA_EXT_LEN_MASK) {
3703                         case 0x00:
3704                                 result = lun[1];
3705                                 break;
3706                         case 0x10:
3707                                 result = (lun[1] << 16) + (lun[2] << 8) +
3708                                     lun[3];
3709                                 break;
3710                         case 0x20:
3711                                 if (lun[1] == 0 && lun[6] == 0 && lun[7] == 0)
3712                                         result = (lun[2] << 24) +
3713                                             (lun[3] << 16) + (lun[4] << 8) +
3714                                             lun[5];
3715                                 break;
3716                         }
3717                         break;
3718                 case RPL_LUNDATA_EXT_EAM_NOT_SPEC:
3719                         result = 0xffffffff;
3720                         break;
3721                 }
3722                 break;
3723         }
3724         return (result);
3725 }
3726
3727 uint64_t
3728 ctl_encode_lun(uint32_t decoded)
3729 {
3730         uint64_t l = decoded;
3731
3732         if (l <= 0xff)
3733                 return (((uint64_t)RPL_LUNDATA_ATYP_PERIPH << 56) | (l << 48));
3734         if (l <= 0x3fff)
3735                 return (((uint64_t)RPL_LUNDATA_ATYP_FLAT << 56) | (l << 48));
3736         if (l <= 0xffffff)
3737                 return (((uint64_t)(RPL_LUNDATA_ATYP_EXTLUN | 0x12) << 56) |
3738                     (l << 32));
3739         return ((((uint64_t)RPL_LUNDATA_ATYP_EXTLUN | 0x22) << 56) | (l << 16));
3740 }
3741
3742 int
3743 ctl_ffz(uint32_t *mask, uint32_t first, uint32_t last)
3744 {
3745         int i;
3746
3747         for (i = first; i < last; i++) {
3748                 if ((mask[i / 32] & (1 << (i % 32))) == 0)
3749                         return (i);
3750         }
3751         return (-1);
3752 }
3753
3754 int
3755 ctl_set_mask(uint32_t *mask, uint32_t bit)
3756 {
3757         uint32_t chunk, piece;
3758
3759         chunk = bit >> 5;
3760         piece = bit % (sizeof(uint32_t) * 8);
3761
3762         if ((mask[chunk] & (1 << piece)) != 0)
3763                 return (-1);
3764         else
3765                 mask[chunk] |= (1 << piece);
3766
3767         return (0);
3768 }
3769
3770 int
3771 ctl_clear_mask(uint32_t *mask, uint32_t bit)
3772 {
3773         uint32_t chunk, piece;
3774
3775         chunk = bit >> 5;
3776         piece = bit % (sizeof(uint32_t) * 8);
3777
3778         if ((mask[chunk] & (1 << piece)) == 0)
3779                 return (-1);
3780         else
3781                 mask[chunk] &= ~(1 << piece);
3782
3783         return (0);
3784 }
3785
3786 int
3787 ctl_is_set(uint32_t *mask, uint32_t bit)
3788 {
3789         uint32_t chunk, piece;
3790
3791         chunk = bit >> 5;
3792         piece = bit % (sizeof(uint32_t) * 8);
3793
3794         if ((mask[chunk] & (1 << piece)) == 0)
3795                 return (0);
3796         else
3797                 return (1);
3798 }
3799
3800 static uint64_t
3801 ctl_get_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 (0);
3808         return (t[residx % CTL_MAX_INIT_PER_PORT]);
3809 }
3810
3811 static void
3812 ctl_clr_prkey(struct ctl_lun *lun, uint32_t residx)
3813 {
3814         uint64_t *t;
3815
3816         t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3817         if (t == NULL)
3818                 return;
3819         t[residx % CTL_MAX_INIT_PER_PORT] = 0;
3820 }
3821
3822 static void
3823 ctl_alloc_prkey(struct ctl_lun *lun, uint32_t residx)
3824 {
3825         uint64_t *p;
3826         u_int i;
3827
3828         i = residx/CTL_MAX_INIT_PER_PORT;
3829         if (lun->pr_keys[i] != NULL)
3830                 return;
3831         mtx_unlock(&lun->lun_lock);
3832         p = malloc(sizeof(uint64_t) * CTL_MAX_INIT_PER_PORT, M_CTL,
3833             M_WAITOK | M_ZERO);
3834         mtx_lock(&lun->lun_lock);
3835         if (lun->pr_keys[i] == NULL)
3836                 lun->pr_keys[i] = p;
3837         else
3838                 free(p, M_CTL);
3839 }
3840
3841 static void
3842 ctl_set_prkey(struct ctl_lun *lun, uint32_t residx, uint64_t key)
3843 {
3844         uint64_t *t;
3845
3846         t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3847         KASSERT(t != NULL, ("prkey %d is not allocated", residx));
3848         t[residx % CTL_MAX_INIT_PER_PORT] = key;
3849 }
3850
3851 /*
3852  * ctl_softc, pool_name, total_ctl_io are passed in.
3853  * npool is passed out.
3854  */
3855 int
3856 ctl_pool_create(struct ctl_softc *ctl_softc, const char *pool_name,
3857                 uint32_t total_ctl_io, void **npool)
3858 {
3859         struct ctl_io_pool *pool;
3860
3861         pool = (struct ctl_io_pool *)malloc(sizeof(*pool), M_CTL,
3862                                             M_NOWAIT | M_ZERO);
3863         if (pool == NULL)
3864                 return (ENOMEM);
3865
3866         snprintf(pool->name, sizeof(pool->name), "CTL IO %s", pool_name);
3867         pool->ctl_softc = ctl_softc;
3868 #ifdef IO_POOLS
3869         pool->zone = uma_zsecond_create(pool->name, NULL,
3870             NULL, NULL, NULL, ctl_softc->io_zone);
3871         /* uma_prealloc(pool->zone, total_ctl_io); */
3872 #else
3873         pool->zone = ctl_softc->io_zone;
3874 #endif
3875
3876         *npool = pool;
3877         return (0);
3878 }
3879
3880 void
3881 ctl_pool_free(struct ctl_io_pool *pool)
3882 {
3883
3884         if (pool == NULL)
3885                 return;
3886
3887 #ifdef IO_POOLS
3888         uma_zdestroy(pool->zone);
3889 #endif
3890         free(pool, M_CTL);
3891 }
3892
3893 union ctl_io *
3894 ctl_alloc_io(void *pool_ref)
3895 {
3896         struct ctl_io_pool *pool = (struct ctl_io_pool *)pool_ref;
3897         union ctl_io *io;
3898
3899         io = uma_zalloc(pool->zone, M_WAITOK);
3900         if (io != NULL) {
3901                 io->io_hdr.pool = pool_ref;
3902                 CTL_SOFTC(io) = pool->ctl_softc;
3903         }
3904         return (io);
3905 }
3906
3907 union ctl_io *
3908 ctl_alloc_io_nowait(void *pool_ref)
3909 {
3910         struct ctl_io_pool *pool = (struct ctl_io_pool *)pool_ref;
3911         union ctl_io *io;
3912
3913         io = uma_zalloc(pool->zone, M_NOWAIT);
3914         if (io != NULL) {
3915                 io->io_hdr.pool = pool_ref;
3916                 CTL_SOFTC(io) = pool->ctl_softc;
3917         }
3918         return (io);
3919 }
3920
3921 void
3922 ctl_free_io(union ctl_io *io)
3923 {
3924         struct ctl_io_pool *pool;
3925
3926         if (io == NULL)
3927                 return;
3928
3929         pool = (struct ctl_io_pool *)io->io_hdr.pool;
3930         uma_zfree(pool->zone, io);
3931 }
3932
3933 void
3934 ctl_zero_io(union ctl_io *io)
3935 {
3936         struct ctl_io_pool *pool;
3937
3938         if (io == NULL)
3939                 return;
3940
3941         /*
3942          * May need to preserve linked list pointers at some point too.
3943          */
3944         pool = io->io_hdr.pool;
3945         memset(io, 0, sizeof(*io));
3946         io->io_hdr.pool = pool;
3947         CTL_SOFTC(io) = pool->ctl_softc;
3948 }
3949
3950 int
3951 ctl_expand_number(const char *buf, uint64_t *num)
3952 {
3953         char *endptr;
3954         uint64_t number;
3955         unsigned shift;
3956
3957         number = strtoq(buf, &endptr, 0);
3958
3959         switch (tolower((unsigned char)*endptr)) {
3960         case 'e':
3961                 shift = 60;
3962                 break;
3963         case 'p':
3964                 shift = 50;
3965                 break;
3966         case 't':
3967                 shift = 40;
3968                 break;
3969         case 'g':
3970                 shift = 30;
3971                 break;
3972         case 'm':
3973                 shift = 20;
3974                 break;
3975         case 'k':
3976                 shift = 10;
3977                 break;
3978         case 'b':
3979         case '\0': /* No unit. */
3980                 *num = number;
3981                 return (0);
3982         default:
3983                 /* Unrecognized unit. */
3984                 return (-1);
3985         }
3986
3987         if ((number << shift) >> shift != number) {
3988                 /* Overflow */
3989                 return (-1);
3990         }
3991         *num = number << shift;
3992         return (0);
3993 }
3994
3995
3996 /*
3997  * This routine could be used in the future to load default and/or saved
3998  * mode page parameters for a particuar lun.
3999  */
4000 static int
4001 ctl_init_page_index(struct ctl_lun *lun)
4002 {
4003         int i, page_code;
4004         struct ctl_page_index *page_index;
4005         const char *value;
4006         uint64_t ival;
4007
4008         memcpy(&lun->mode_pages.index, page_index_template,
4009                sizeof(page_index_template));
4010
4011         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
4012
4013                 page_index = &lun->mode_pages.index[i];
4014                 if (lun->be_lun->lun_type == T_DIRECT &&
4015                     (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
4016                         continue;
4017                 if (lun->be_lun->lun_type == T_PROCESSOR &&
4018                     (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
4019                         continue;
4020                 if (lun->be_lun->lun_type == T_CDROM &&
4021                     (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
4022                         continue;
4023
4024                 page_code = page_index->page_code & SMPH_PC_MASK;
4025                 switch (page_code) {
4026                 case SMS_RW_ERROR_RECOVERY_PAGE: {
4027                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4028                             ("subpage %#x for page %#x is incorrect!",
4029                             page_index->subpage, page_code));
4030                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_CURRENT],
4031                                &rw_er_page_default,
4032                                sizeof(rw_er_page_default));
4033                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_CHANGEABLE],
4034                                &rw_er_page_changeable,
4035                                sizeof(rw_er_page_changeable));
4036                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_DEFAULT],
4037                                &rw_er_page_default,
4038                                sizeof(rw_er_page_default));
4039                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_SAVED],
4040                                &rw_er_page_default,
4041                                sizeof(rw_er_page_default));
4042                         page_index->page_data =
4043                                 (uint8_t *)lun->mode_pages.rw_er_page;
4044                         break;
4045                 }
4046                 case SMS_FORMAT_DEVICE_PAGE: {
4047                         struct scsi_format_page *format_page;
4048
4049                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4050                             ("subpage %#x for page %#x is incorrect!",
4051                             page_index->subpage, page_code));
4052
4053                         /*
4054                          * Sectors per track are set above.  Bytes per
4055                          * sector need to be set here on a per-LUN basis.
4056                          */
4057                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_CURRENT],
4058                                &format_page_default,
4059                                sizeof(format_page_default));
4060                         memcpy(&lun->mode_pages.format_page[
4061                                CTL_PAGE_CHANGEABLE], &format_page_changeable,
4062                                sizeof(format_page_changeable));
4063                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_DEFAULT],
4064                                &format_page_default,
4065                                sizeof(format_page_default));
4066                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_SAVED],
4067                                &format_page_default,
4068                                sizeof(format_page_default));
4069
4070                         format_page = &lun->mode_pages.format_page[
4071                                 CTL_PAGE_CURRENT];
4072                         scsi_ulto2b(lun->be_lun->blocksize,
4073                                     format_page->bytes_per_sector);
4074
4075                         format_page = &lun->mode_pages.format_page[
4076                                 CTL_PAGE_DEFAULT];
4077                         scsi_ulto2b(lun->be_lun->blocksize,
4078                                     format_page->bytes_per_sector);
4079
4080                         format_page = &lun->mode_pages.format_page[
4081                                 CTL_PAGE_SAVED];
4082                         scsi_ulto2b(lun->be_lun->blocksize,
4083                                     format_page->bytes_per_sector);
4084
4085                         page_index->page_data =
4086                                 (uint8_t *)lun->mode_pages.format_page;
4087                         break;
4088                 }
4089                 case SMS_RIGID_DISK_PAGE: {
4090                         struct scsi_rigid_disk_page *rigid_disk_page;
4091                         uint32_t sectors_per_cylinder;
4092                         uint64_t cylinders;
4093 #ifndef __XSCALE__
4094                         int shift;
4095 #endif /* !__XSCALE__ */
4096
4097                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4098                             ("subpage %#x for page %#x is incorrect!",
4099                             page_index->subpage, page_code));
4100
4101                         /*
4102                          * Rotation rate and sectors per track are set
4103                          * above.  We calculate the cylinders here based on
4104                          * capacity.  Due to the number of heads and
4105                          * sectors per track we're using, smaller arrays
4106                          * may turn out to have 0 cylinders.  Linux and
4107                          * FreeBSD don't pay attention to these mode pages
4108                          * to figure out capacity, but Solaris does.  It
4109                          * seems to deal with 0 cylinders just fine, and
4110                          * works out a fake geometry based on the capacity.
4111                          */
4112                         memcpy(&lun->mode_pages.rigid_disk_page[
4113                                CTL_PAGE_DEFAULT], &rigid_disk_page_default,
4114                                sizeof(rigid_disk_page_default));
4115                         memcpy(&lun->mode_pages.rigid_disk_page[
4116                                CTL_PAGE_CHANGEABLE],&rigid_disk_page_changeable,
4117                                sizeof(rigid_disk_page_changeable));
4118
4119                         sectors_per_cylinder = CTL_DEFAULT_SECTORS_PER_TRACK *
4120                                 CTL_DEFAULT_HEADS;
4121
4122                         /*
4123                          * The divide method here will be more accurate,
4124                          * probably, but results in floating point being
4125                          * used in the kernel on i386 (__udivdi3()).  On the
4126                          * XScale, though, __udivdi3() is implemented in
4127                          * software.
4128                          *
4129                          * The shift method for cylinder calculation is
4130                          * accurate if sectors_per_cylinder is a power of
4131                          * 2.  Otherwise it might be slightly off -- you
4132                          * might have a bit of a truncation problem.
4133                          */
4134 #ifdef  __XSCALE__
4135                         cylinders = (lun->be_lun->maxlba + 1) /
4136                                 sectors_per_cylinder;
4137 #else
4138                         for (shift = 31; shift > 0; shift--) {
4139                                 if (sectors_per_cylinder & (1 << shift))
4140                                         break;
4141                         }
4142                         cylinders = (lun->be_lun->maxlba + 1) >> shift;
4143 #endif
4144
4145                         /*
4146                          * We've basically got 3 bytes, or 24 bits for the
4147                          * cylinder size in the mode page.  If we're over,
4148                          * just round down to 2^24.
4149                          */
4150                         if (cylinders > 0xffffff)
4151                                 cylinders = 0xffffff;
4152
4153                         rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4154                                 CTL_PAGE_DEFAULT];
4155                         scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4156
4157                         if ((value = dnvlist_get_string(lun->be_lun->options,
4158                             "rpm", NULL)) != NULL) {
4159                                 scsi_ulto2b(strtol(value, NULL, 0),
4160                                      rigid_disk_page->rotation_rate);
4161                         }
4162
4163                         memcpy(&lun->mode_pages.rigid_disk_page[CTL_PAGE_CURRENT],
4164                                &lun->mode_pages.rigid_disk_page[CTL_PAGE_DEFAULT],
4165                                sizeof(rigid_disk_page_default));
4166                         memcpy(&lun->mode_pages.rigid_disk_page[CTL_PAGE_SAVED],
4167                                &lun->mode_pages.rigid_disk_page[CTL_PAGE_DEFAULT],
4168                                sizeof(rigid_disk_page_default));
4169
4170                         page_index->page_data =
4171                                 (uint8_t *)lun->mode_pages.rigid_disk_page;
4172                         break;
4173                 }
4174                 case SMS_VERIFY_ERROR_RECOVERY_PAGE: {
4175                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4176                             ("subpage %#x for page %#x is incorrect!",
4177                             page_index->subpage, page_code));
4178                         memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_CURRENT],
4179                                &verify_er_page_default,
4180                                sizeof(verify_er_page_default));
4181                         memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_CHANGEABLE],
4182                                &verify_er_page_changeable,
4183                                sizeof(verify_er_page_changeable));
4184                         memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_DEFAULT],
4185                                &verify_er_page_default,
4186                                sizeof(verify_er_page_default));
4187                         memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_SAVED],
4188                                &verify_er_page_default,
4189                                sizeof(verify_er_page_default));
4190                         page_index->page_data =
4191                                 (uint8_t *)lun->mode_pages.verify_er_page;
4192                         break;
4193                 }
4194                 case SMS_CACHING_PAGE: {
4195                         struct scsi_caching_page *caching_page;
4196
4197                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4198                             ("subpage %#x for page %#x is incorrect!",
4199                             page_index->subpage, page_code));
4200                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_DEFAULT],
4201                                &caching_page_default,
4202                                sizeof(caching_page_default));
4203                         memcpy(&lun->mode_pages.caching_page[
4204                                CTL_PAGE_CHANGEABLE], &caching_page_changeable,
4205                                sizeof(caching_page_changeable));
4206                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4207                                &caching_page_default,
4208                                sizeof(caching_page_default));
4209                         caching_page = &lun->mode_pages.caching_page[
4210                             CTL_PAGE_SAVED];
4211                         value = dnvlist_get_string(lun->be_lun->options,
4212                             "writecache", NULL);
4213                         if (value != NULL && strcmp(value, "off") == 0)
4214                                 caching_page->flags1 &= ~SCP_WCE;
4215                         value = dnvlist_get_string(lun->be_lun->options,
4216                             "readcache", NULL);
4217                         if (value != NULL && strcmp(value, "off") == 0)
4218                                 caching_page->flags1 |= SCP_RCD;
4219                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_CURRENT],
4220                                &lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4221                                sizeof(caching_page_default));
4222                         page_index->page_data =
4223                                 (uint8_t *)lun->mode_pages.caching_page;
4224                         break;
4225                 }
4226                 case SMS_CONTROL_MODE_PAGE: {
4227                         switch (page_index->subpage) {
4228                         case SMS_SUBPAGE_PAGE_0: {
4229                                 struct scsi_control_page *control_page;
4230
4231                                 memcpy(&lun->mode_pages.control_page[
4232                                     CTL_PAGE_DEFAULT],
4233                                        &control_page_default,
4234                                        sizeof(control_page_default));
4235                                 memcpy(&lun->mode_pages.control_page[
4236                                     CTL_PAGE_CHANGEABLE],
4237                                        &control_page_changeable,
4238                                        sizeof(control_page_changeable));
4239                                 memcpy(&lun->mode_pages.control_page[
4240                                     CTL_PAGE_SAVED],
4241                                        &control_page_default,
4242                                        sizeof(control_page_default));
4243                                 control_page = &lun->mode_pages.control_page[
4244                                     CTL_PAGE_SAVED];
4245                                 value = dnvlist_get_string(lun->be_lun->options,
4246                                     "reordering", NULL);
4247                                 if (value != NULL &&
4248                                     strcmp(value, "unrestricted") == 0) {
4249                                         control_page->queue_flags &=
4250                                             ~SCP_QUEUE_ALG_MASK;
4251                                         control_page->queue_flags |=
4252                                             SCP_QUEUE_ALG_UNRESTRICTED;
4253                                 }
4254                                 memcpy(&lun->mode_pages.control_page[
4255                                     CTL_PAGE_CURRENT],
4256                                        &lun->mode_pages.control_page[
4257                                     CTL_PAGE_SAVED],
4258                                        sizeof(control_page_default));
4259                                 page_index->page_data =
4260                                     (uint8_t *)lun->mode_pages.control_page;
4261                                 break;
4262                         }
4263                         case 0x01:
4264                                 memcpy(&lun->mode_pages.control_ext_page[
4265                                     CTL_PAGE_DEFAULT],
4266                                        &control_ext_page_default,
4267                                        sizeof(control_ext_page_default));
4268                                 memcpy(&lun->mode_pages.control_ext_page[
4269                                     CTL_PAGE_CHANGEABLE],
4270                                        &control_ext_page_changeable,
4271                                        sizeof(control_ext_page_changeable));
4272                                 memcpy(&lun->mode_pages.control_ext_page[
4273                                     CTL_PAGE_SAVED],
4274                                        &control_ext_page_default,
4275                                        sizeof(control_ext_page_default));
4276                                 memcpy(&lun->mode_pages.control_ext_page[
4277                                     CTL_PAGE_CURRENT],
4278                                        &lun->mode_pages.control_ext_page[
4279                                     CTL_PAGE_SAVED],
4280                                        sizeof(control_ext_page_default));
4281                                 page_index->page_data =
4282                                     (uint8_t *)lun->mode_pages.control_ext_page;
4283                                 break;
4284                         default:
4285                                 panic("subpage %#x for page %#x is incorrect!",
4286                                       page_index->subpage, page_code);
4287                         }
4288                         break;
4289                 }
4290                 case SMS_INFO_EXCEPTIONS_PAGE: {
4291                         switch (page_index->subpage) {
4292                         case SMS_SUBPAGE_PAGE_0:
4293                                 memcpy(&lun->mode_pages.ie_page[CTL_PAGE_CURRENT],
4294                                        &ie_page_default,
4295                                        sizeof(ie_page_default));
4296                                 memcpy(&lun->mode_pages.ie_page[
4297                                        CTL_PAGE_CHANGEABLE], &ie_page_changeable,
4298                                        sizeof(ie_page_changeable));
4299                                 memcpy(&lun->mode_pages.ie_page[CTL_PAGE_DEFAULT],
4300                                        &ie_page_default,
4301                                        sizeof(ie_page_default));
4302                                 memcpy(&lun->mode_pages.ie_page[CTL_PAGE_SAVED],
4303                                        &ie_page_default,
4304                                        sizeof(ie_page_default));
4305                                 page_index->page_data =
4306                                         (uint8_t *)lun->mode_pages.ie_page;
4307                                 break;
4308                         case 0x02: {
4309                                 struct ctl_logical_block_provisioning_page *page;
4310
4311                                 memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_DEFAULT],
4312                                        &lbp_page_default,
4313                                        sizeof(lbp_page_default));
4314                                 memcpy(&lun->mode_pages.lbp_page[
4315                                        CTL_PAGE_CHANGEABLE], &lbp_page_changeable,
4316                                        sizeof(lbp_page_changeable));
4317                                 memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_SAVED],
4318                                        &lbp_page_default,
4319                                        sizeof(lbp_page_default));
4320                                 page = &lun->mode_pages.lbp_page[CTL_PAGE_SAVED];
4321                                 value = dnvlist_get_string(lun->be_lun->options,
4322                                     "avail-threshold", NULL);
4323                                 if (value != NULL &&
4324                                     ctl_expand_number(value, &ival) == 0) {
4325                                         page->descr[0].flags |= SLBPPD_ENABLED |
4326                                             SLBPPD_ARMING_DEC;
4327                                         if (lun->be_lun->blocksize)
4328                                                 ival /= lun->be_lun->blocksize;
4329                                         else
4330                                                 ival /= 512;
4331                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4332                                             page->descr[0].count);
4333                                 }
4334                                 value = dnvlist_get_string(lun->be_lun->options,
4335                                     "used-threshold", NULL);
4336                                 if (value != NULL &&
4337                                     ctl_expand_number(value, &ival) == 0) {
4338                                         page->descr[1].flags |= SLBPPD_ENABLED |
4339                                             SLBPPD_ARMING_INC;
4340                                         if (lun->be_lun->blocksize)
4341                                                 ival /= lun->be_lun->blocksize;
4342                                         else
4343                                                 ival /= 512;
4344                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4345                                             page->descr[1].count);
4346                                 }
4347                                 value = dnvlist_get_string(lun->be_lun->options,
4348                                     "pool-avail-threshold", NULL);
4349                                 if (value != NULL &&
4350                                     ctl_expand_number(value, &ival) == 0) {
4351                                         page->descr[2].flags |= SLBPPD_ENABLED |
4352                                             SLBPPD_ARMING_DEC;
4353                                         if (lun->be_lun->blocksize)
4354                                                 ival /= lun->be_lun->blocksize;
4355                                         else
4356                                                 ival /= 512;
4357                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4358                                             page->descr[2].count);
4359                                 }
4360                                 value = dnvlist_get_string(lun->be_lun->options,
4361                                     "pool-used-threshold", NULL);
4362                                 if (value != NULL &&
4363                                     ctl_expand_number(value, &ival) == 0) {
4364                                         page->descr[3].flags |= SLBPPD_ENABLED |
4365                                             SLBPPD_ARMING_INC;
4366                                         if (lun->be_lun->blocksize)
4367                                                 ival /= lun->be_lun->blocksize;
4368                                         else
4369                                                 ival /= 512;
4370                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4371                                             page->descr[3].count);
4372                                 }
4373                                 memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_CURRENT],
4374                                        &lun->mode_pages.lbp_page[CTL_PAGE_SAVED],
4375                                        sizeof(lbp_page_default));
4376                                 page_index->page_data =
4377                                         (uint8_t *)lun->mode_pages.lbp_page;
4378                                 break;
4379                         }
4380                         default:
4381                                 panic("subpage %#x for page %#x is incorrect!",
4382                                       page_index->subpage, page_code);
4383                         }
4384                         break;
4385                 }
4386                 case SMS_CDDVD_CAPS_PAGE:{
4387                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4388                             ("subpage %#x for page %#x is incorrect!",
4389                             page_index->subpage, page_code));
4390                         memcpy(&lun->mode_pages.cddvd_page[CTL_PAGE_DEFAULT],
4391                                &cddvd_page_default,
4392                                sizeof(cddvd_page_default));
4393                         memcpy(&lun->mode_pages.cddvd_page[
4394                                CTL_PAGE_CHANGEABLE], &cddvd_page_changeable,
4395                                sizeof(cddvd_page_changeable));
4396                         memcpy(&lun->mode_pages.cddvd_page[CTL_PAGE_SAVED],
4397                                &cddvd_page_default,
4398                                sizeof(cddvd_page_default));
4399                         memcpy(&lun->mode_pages.cddvd_page[CTL_PAGE_CURRENT],
4400                                &lun->mode_pages.cddvd_page[CTL_PAGE_SAVED],
4401                                sizeof(cddvd_page_default));
4402                         page_index->page_data =
4403                                 (uint8_t *)lun->mode_pages.cddvd_page;
4404                         break;
4405                 }
4406                 default:
4407                         panic("invalid page code value %#x", page_code);
4408                 }
4409         }
4410
4411         return (CTL_RETVAL_COMPLETE);
4412 }
4413
4414 static int
4415 ctl_init_log_page_index(struct ctl_lun *lun)
4416 {
4417         struct ctl_page_index *page_index;
4418         int i, j, k, prev;
4419
4420         memcpy(&lun->log_pages.index, log_page_index_template,
4421                sizeof(log_page_index_template));
4422
4423         prev = -1;
4424         for (i = 0, j = 0, k = 0; i < CTL_NUM_LOG_PAGES; i++) {
4425
4426                 page_index = &lun->log_pages.index[i];
4427                 if (lun->be_lun->lun_type == T_DIRECT &&
4428                     (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
4429                         continue;
4430                 if (lun->be_lun->lun_type == T_PROCESSOR &&
4431                     (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
4432                         continue;
4433                 if (lun->be_lun->lun_type == T_CDROM &&
4434                     (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
4435                         continue;
4436
4437                 if (page_index->page_code == SLS_LOGICAL_BLOCK_PROVISIONING &&
4438                     lun->backend->lun_attr == NULL)
4439                         continue;
4440
4441                 if (page_index->page_code != prev) {
4442                         lun->log_pages.pages_page[j] = page_index->page_code;
4443                         prev = page_index->page_code;
4444                         j++;
4445                 }
4446                 lun->log_pages.subpages_page[k*2] = page_index->page_code;
4447                 lun->log_pages.subpages_page[k*2+1] = page_index->subpage;
4448                 k++;
4449         }
4450         lun->log_pages.index[0].page_data = &lun->log_pages.pages_page[0];
4451         lun->log_pages.index[0].page_len = j;
4452         lun->log_pages.index[1].page_data = &lun->log_pages.subpages_page[0];
4453         lun->log_pages.index[1].page_len = k * 2;
4454         lun->log_pages.index[2].page_data = &lun->log_pages.lbp_page[0];
4455         lun->log_pages.index[2].page_len = 12*CTL_NUM_LBP_PARAMS;
4456         lun->log_pages.index[3].page_data = (uint8_t *)&lun->log_pages.stat_page;
4457         lun->log_pages.index[3].page_len = sizeof(lun->log_pages.stat_page);
4458         lun->log_pages.index[4].page_data = (uint8_t *)&lun->log_pages.ie_page;
4459         lun->log_pages.index[4].page_len = sizeof(lun->log_pages.ie_page);
4460
4461         return (CTL_RETVAL_COMPLETE);
4462 }
4463
4464 static int
4465 hex2bin(const char *str, uint8_t *buf, int buf_size)
4466 {
4467         int i;
4468         u_char c;
4469
4470         memset(buf, 0, buf_size);
4471         while (isspace(str[0]))
4472                 str++;
4473         if (str[0] == '0' && (str[1] == 'x' || str[1] == 'X'))
4474                 str += 2;
4475         buf_size *= 2;
4476         for (i = 0; str[i] != 0 && i < buf_size; i++) {
4477                 while (str[i] == '-')   /* Skip dashes in UUIDs. */
4478                         str++;
4479                 c = str[i];
4480                 if (isdigit(c))
4481                         c -= '0';
4482                 else if (isalpha(c))
4483                         c -= isupper(c) ? 'A' - 10 : 'a' - 10;
4484                 else
4485                         break;
4486                 if (c >= 16)
4487                         break;
4488                 if ((i & 1) == 0)
4489                         buf[i / 2] |= (c << 4);
4490                 else
4491                         buf[i / 2] |= c;
4492         }
4493         return ((i + 1) / 2);
4494 }
4495
4496 /*
4497  * LUN allocation.
4498  *
4499  * Requirements:
4500  * - caller allocates and zeros LUN storage, or passes in a NULL LUN if he
4501  *   wants us to allocate the LUN and he can block.
4502  * - ctl_softc is always set
4503  * - be_lun is set if the LUN has a backend (needed for disk LUNs)
4504  *
4505  * Returns 0 for success, non-zero (errno) for failure.
4506  */
4507 static int
4508 ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *ctl_lun,
4509               struct ctl_be_lun *const be_lun)
4510 {
4511         struct ctl_lun *nlun, *lun;
4512         struct scsi_vpd_id_descriptor *desc;
4513         struct scsi_vpd_id_t10 *t10id;
4514         const char *eui, *naa, *scsiname, *uuid, *vendor, *value;
4515         int lun_number, lun_malloced;
4516         int devidlen, idlen1, idlen2 = 0, len;
4517
4518         if (be_lun == NULL)
4519                 return (EINVAL);
4520
4521         /*
4522          * We currently only support Direct Access or Processor LUN types.
4523          */
4524         switch (be_lun->lun_type) {
4525         case T_DIRECT:
4526         case T_PROCESSOR:
4527         case T_CDROM:
4528                 break;
4529         case T_SEQUENTIAL:
4530         case T_CHANGER:
4531         default:
4532                 be_lun->lun_config_status(be_lun->be_lun,
4533                                           CTL_LUN_CONFIG_FAILURE);
4534                 break;
4535         }
4536         if (ctl_lun == NULL) {
4537                 lun = malloc(sizeof(*lun), M_CTL, M_WAITOK);
4538                 lun_malloced = 1;
4539         } else {
4540                 lun_malloced = 0;
4541                 lun = ctl_lun;
4542         }
4543
4544         memset(lun, 0, sizeof(*lun));
4545         if (lun_malloced)
4546                 lun->flags = CTL_LUN_MALLOCED;
4547
4548         lun->pending_sense = malloc(sizeof(struct scsi_sense_data *) *
4549             ctl_max_ports, M_DEVBUF, M_WAITOK | M_ZERO);
4550         lun->pending_ua = malloc(sizeof(ctl_ua_type *) * ctl_max_ports,
4551             M_DEVBUF, M_WAITOK | M_ZERO);
4552         lun->pr_keys = malloc(sizeof(uint64_t *) * ctl_max_ports,
4553             M_DEVBUF, M_WAITOK | M_ZERO);
4554
4555         /* Generate LUN ID. */
4556         devidlen = max(CTL_DEVID_MIN_LEN,
4557             strnlen(be_lun->device_id, CTL_DEVID_LEN));
4558         idlen1 = sizeof(*t10id) + devidlen;
4559         len = sizeof(struct scsi_vpd_id_descriptor) + idlen1;
4560         scsiname = dnvlist_get_string(be_lun->options, "scsiname", NULL);
4561         if (scsiname != NULL) {
4562                 idlen2 = roundup2(strlen(scsiname) + 1, 4);
4563                 len += sizeof(struct scsi_vpd_id_descriptor) + idlen2;
4564         }
4565         eui = dnvlist_get_string(be_lun->options, "eui", NULL);
4566         if (eui != NULL) {
4567                 len += sizeof(struct scsi_vpd_id_descriptor) + 16;
4568         }
4569         naa = dnvlist_get_string(be_lun->options, "naa", NULL);
4570         if (naa != NULL) {
4571                 len += sizeof(struct scsi_vpd_id_descriptor) + 16;
4572         }
4573         uuid = dnvlist_get_string(be_lun->options, "uuid", NULL);
4574         if (uuid != NULL) {
4575                 len += sizeof(struct scsi_vpd_id_descriptor) + 18;
4576         }
4577         lun->lun_devid = malloc(sizeof(struct ctl_devid) + len,
4578             M_CTL, M_WAITOK | M_ZERO);
4579         desc = (struct scsi_vpd_id_descriptor *)lun->lun_devid->data;
4580         desc->proto_codeset = SVPD_ID_CODESET_ASCII;
4581         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_T10;
4582         desc->length = idlen1;
4583         t10id = (struct scsi_vpd_id_t10 *)&desc->identifier[0];
4584         memset(t10id->vendor, ' ', sizeof(t10id->vendor));
4585         if ((vendor = dnvlist_get_string(be_lun->options, "vendor", NULL)) == NULL) {
4586                 strncpy((char *)t10id->vendor, CTL_VENDOR, sizeof(t10id->vendor));
4587         } else {
4588                 strncpy(t10id->vendor, vendor,
4589                     min(sizeof(t10id->vendor), strlen(vendor)));
4590         }
4591         strncpy((char *)t10id->vendor_spec_id,
4592             (char *)be_lun->device_id, devidlen);
4593         if (scsiname != NULL) {
4594                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4595                     desc->length);
4596                 desc->proto_codeset = SVPD_ID_CODESET_UTF8;
4597                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4598                     SVPD_ID_TYPE_SCSI_NAME;
4599                 desc->length = idlen2;
4600                 strlcpy(desc->identifier, scsiname, idlen2);
4601         }
4602         if (eui != 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_EUI64;
4608                 desc->length = hex2bin(eui, desc->identifier, 16);
4609                 desc->length = desc->length > 12 ? 16 :
4610                     (desc->length > 8 ? 12 : 8);
4611                 len -= 16 - desc->length;
4612         }
4613         if (naa != NULL) {
4614                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4615                     desc->length);
4616                 desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4617                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4618                     SVPD_ID_TYPE_NAA;
4619                 desc->length = hex2bin(naa, desc->identifier, 16);
4620                 desc->length = desc->length > 8 ? 16 : 8;
4621                 len -= 16 - desc->length;
4622         }
4623         if (uuid != NULL) {
4624                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4625                     desc->length);
4626                 desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4627                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4628                     SVPD_ID_TYPE_UUID;
4629                 desc->identifier[0] = 0x10;
4630                 hex2bin(uuid, &desc->identifier[2], 16);
4631                 desc->length = 18;
4632         }
4633         lun->lun_devid->len = len;
4634
4635         mtx_lock(&ctl_softc->ctl_lock);
4636         /*
4637          * See if the caller requested a particular LUN number.  If so, see
4638          * if it is available.  Otherwise, allocate the first available LUN.
4639          */
4640         if (be_lun->flags & CTL_LUN_FLAG_ID_REQ) {
4641                 if ((be_lun->req_lun_id > (ctl_max_luns - 1))
4642                  || (ctl_is_set(ctl_softc->ctl_lun_mask, be_lun->req_lun_id))) {
4643                         mtx_unlock(&ctl_softc->ctl_lock);
4644                         if (be_lun->req_lun_id > (ctl_max_luns - 1)) {
4645                                 printf("ctl: requested LUN ID %d is higher "
4646                                        "than ctl_max_luns - 1 (%d)\n",
4647                                        be_lun->req_lun_id, ctl_max_luns - 1);
4648                         } else {
4649                                 /*
4650                                  * XXX KDM return an error, or just assign
4651                                  * another LUN ID in this case??
4652                                  */
4653                                 printf("ctl: requested LUN ID %d is already "
4654                                        "in use\n", be_lun->req_lun_id);
4655                         }
4656 fail:
4657                         free(lun->lun_devid, M_CTL);
4658                         if (lun->flags & CTL_LUN_MALLOCED)
4659                                 free(lun, M_CTL);
4660                         be_lun->lun_config_status(be_lun->be_lun,
4661                                                   CTL_LUN_CONFIG_FAILURE);
4662                         return (ENOSPC);
4663                 }
4664                 lun_number = be_lun->req_lun_id;
4665         } else {
4666                 lun_number = ctl_ffz(ctl_softc->ctl_lun_mask, 0, ctl_max_luns);
4667                 if (lun_number == -1) {
4668                         mtx_unlock(&ctl_softc->ctl_lock);
4669                         printf("ctl: can't allocate LUN, out of LUNs\n");
4670                         goto fail;
4671                 }
4672         }
4673         ctl_set_mask(ctl_softc->ctl_lun_mask, lun_number);
4674         mtx_unlock(&ctl_softc->ctl_lock);
4675
4676         mtx_init(&lun->lun_lock, "CTL LUN", NULL, MTX_DEF);
4677         lun->lun = lun_number;
4678         lun->be_lun = be_lun;
4679         /*
4680          * The processor LUN is always enabled.  Disk LUNs come on line
4681          * disabled, and must be enabled by the backend.
4682          */
4683         lun->flags |= CTL_LUN_DISABLED;
4684         lun->backend = be_lun->be;
4685         be_lun->ctl_lun = lun;
4686         be_lun->lun_id = lun_number;
4687         atomic_add_int(&be_lun->be->num_luns, 1);
4688         if (be_lun->flags & CTL_LUN_FLAG_EJECTED)
4689                 lun->flags |= CTL_LUN_EJECTED;
4690         if (be_lun->flags & CTL_LUN_FLAG_NO_MEDIA)
4691                 lun->flags |= CTL_LUN_NO_MEDIA;
4692         if (be_lun->flags & CTL_LUN_FLAG_STOPPED)
4693                 lun->flags |= CTL_LUN_STOPPED;
4694
4695         if (be_lun->flags & CTL_LUN_FLAG_PRIMARY)
4696                 lun->flags |= CTL_LUN_PRIMARY_SC;
4697
4698         value = dnvlist_get_string(be_lun->options, "removable", NULL);
4699         if (value != NULL) {
4700                 if (strcmp(value, "on") == 0)
4701                         lun->flags |= CTL_LUN_REMOVABLE;
4702         } else if (be_lun->lun_type == T_CDROM)
4703                 lun->flags |= CTL_LUN_REMOVABLE;
4704
4705         lun->ctl_softc = ctl_softc;
4706 #ifdef CTL_TIME_IO
4707         lun->last_busy = getsbinuptime();
4708 #endif
4709         TAILQ_INIT(&lun->ooa_queue);
4710         TAILQ_INIT(&lun->blocked_queue);
4711         STAILQ_INIT(&lun->error_list);
4712         lun->ie_reported = 1;
4713         callout_init_mtx(&lun->ie_callout, &lun->lun_lock, 0);
4714         ctl_tpc_lun_init(lun);
4715         if (lun->flags & CTL_LUN_REMOVABLE) {
4716                 lun->prevent = malloc((CTL_MAX_INITIATORS + 31) / 32 * 4,
4717                     M_CTL, M_WAITOK);
4718         }
4719
4720         /*
4721          * Initialize the mode and log page index.
4722          */
4723         ctl_init_page_index(lun);
4724         ctl_init_log_page_index(lun);
4725
4726         /* Setup statistics gathering */
4727         lun->stats.item = lun_number;
4728
4729         /*
4730          * Now, before we insert this lun on the lun list, set the lun
4731          * inventory changed UA for all other luns.
4732          */
4733         mtx_lock(&ctl_softc->ctl_lock);
4734         STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) {
4735                 mtx_lock(&nlun->lun_lock);
4736                 ctl_est_ua_all(nlun, -1, CTL_UA_LUN_CHANGE);
4737                 mtx_unlock(&nlun->lun_lock);
4738         }
4739         STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links);
4740         ctl_softc->ctl_luns[lun_number] = lun;
4741         ctl_softc->num_luns++;
4742         mtx_unlock(&ctl_softc->ctl_lock);
4743
4744         lun->be_lun->lun_config_status(lun->be_lun->be_lun, CTL_LUN_CONFIG_OK);
4745         return (0);
4746 }
4747
4748 /*
4749  * Delete a LUN.
4750  * Assumptions:
4751  * - LUN has already been marked invalid and any pending I/O has been taken
4752  *   care of.
4753  */
4754 static int
4755 ctl_free_lun(struct ctl_lun *lun)
4756 {
4757         struct ctl_softc *softc = lun->ctl_softc;
4758         struct ctl_lun *nlun;
4759         int i;
4760
4761         KASSERT(TAILQ_EMPTY(&lun->ooa_queue),
4762             ("Freeing a LUN %p with outstanding I/O!\n", lun));
4763
4764         mtx_lock(&softc->ctl_lock);
4765         STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links);
4766         ctl_clear_mask(softc->ctl_lun_mask, lun->lun);
4767         softc->ctl_luns[lun->lun] = NULL;
4768         softc->num_luns--;
4769         STAILQ_FOREACH(nlun, &softc->lun_list, links) {
4770                 mtx_lock(&nlun->lun_lock);
4771                 ctl_est_ua_all(nlun, -1, CTL_UA_LUN_CHANGE);
4772                 mtx_unlock(&nlun->lun_lock);
4773         }
4774         mtx_unlock(&softc->ctl_lock);
4775
4776         /*
4777          * Tell the backend to free resources, if this LUN has a backend.
4778          */
4779         atomic_subtract_int(&lun->be_lun->be->num_luns, 1);
4780         lun->be_lun->lun_shutdown(lun->be_lun->be_lun);
4781
4782         lun->ie_reportcnt = UINT32_MAX;
4783         callout_drain(&lun->ie_callout);
4784         ctl_tpc_lun_shutdown(lun);
4785         mtx_destroy(&lun->lun_lock);
4786         free(lun->lun_devid, M_CTL);
4787         for (i = 0; i < ctl_max_ports; i++)
4788                 free(lun->pending_ua[i], M_CTL);
4789         free(lun->pending_ua, M_DEVBUF);
4790         for (i = 0; i < ctl_max_ports; i++)
4791                 free(lun->pr_keys[i], M_CTL);
4792         free(lun->pr_keys, M_DEVBUF);
4793         free(lun->write_buffer, M_CTL);
4794         free(lun->prevent, M_CTL);
4795         if (lun->flags & CTL_LUN_MALLOCED)
4796                 free(lun, M_CTL);
4797
4798         return (0);
4799 }
4800
4801 static void
4802 ctl_create_lun(struct ctl_be_lun *be_lun)
4803 {
4804
4805         /*
4806          * ctl_alloc_lun() should handle all potential failure cases.
4807          */
4808         ctl_alloc_lun(control_softc, NULL, be_lun);
4809 }
4810
4811 int
4812 ctl_add_lun(struct ctl_be_lun *be_lun)
4813 {
4814         struct ctl_softc *softc = control_softc;
4815
4816         mtx_lock(&softc->ctl_lock);
4817         STAILQ_INSERT_TAIL(&softc->pending_lun_queue, be_lun, links);
4818         mtx_unlock(&softc->ctl_lock);
4819         wakeup(&softc->pending_lun_queue);
4820
4821         return (0);
4822 }
4823
4824 int
4825 ctl_enable_lun(struct ctl_be_lun *be_lun)
4826 {
4827         struct ctl_softc *softc;
4828         struct ctl_port *port, *nport;
4829         struct ctl_lun *lun;
4830         int retval;
4831
4832         lun = (struct ctl_lun *)be_lun->ctl_lun;
4833         softc = lun->ctl_softc;
4834
4835         mtx_lock(&softc->ctl_lock);
4836         mtx_lock(&lun->lun_lock);
4837         if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4838                 /*
4839                  * eh?  Why did we get called if the LUN is already
4840                  * enabled?
4841                  */
4842                 mtx_unlock(&lun->lun_lock);
4843                 mtx_unlock(&softc->ctl_lock);
4844                 return (0);
4845         }
4846         lun->flags &= ~CTL_LUN_DISABLED;
4847         mtx_unlock(&lun->lun_lock);
4848
4849         STAILQ_FOREACH_SAFE(port, &softc->port_list, links, nport) {
4850                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0 ||
4851                     port->lun_map != NULL || port->lun_enable == NULL)
4852                         continue;
4853
4854                 /*
4855                  * Drop the lock while we call the FETD's enable routine.
4856                  * This can lead to a callback into CTL (at least in the
4857                  * case of the internal initiator frontend.
4858                  */
4859                 mtx_unlock(&softc->ctl_lock);
4860                 retval = port->lun_enable(port->targ_lun_arg, lun->lun);
4861                 mtx_lock(&softc->ctl_lock);
4862                 if (retval != 0) {
4863                         printf("%s: FETD %s port %d returned error "
4864                                "%d for lun_enable on lun %jd\n",
4865                                __func__, port->port_name, port->targ_port,
4866                                retval, (intmax_t)lun->lun);
4867                 }
4868         }
4869
4870         mtx_unlock(&softc->ctl_lock);
4871         ctl_isc_announce_lun(lun);
4872
4873         return (0);
4874 }
4875
4876 int
4877 ctl_disable_lun(struct ctl_be_lun *be_lun)
4878 {
4879         struct ctl_softc *softc;
4880         struct ctl_port *port;
4881         struct ctl_lun *lun;
4882         int retval;
4883
4884         lun = (struct ctl_lun *)be_lun->ctl_lun;
4885         softc = lun->ctl_softc;
4886
4887         mtx_lock(&softc->ctl_lock);
4888         mtx_lock(&lun->lun_lock);
4889         if (lun->flags & CTL_LUN_DISABLED) {
4890                 mtx_unlock(&lun->lun_lock);
4891                 mtx_unlock(&softc->ctl_lock);
4892                 return (0);
4893         }
4894         lun->flags |= CTL_LUN_DISABLED;
4895         mtx_unlock(&lun->lun_lock);
4896
4897         STAILQ_FOREACH(port, &softc->port_list, links) {
4898                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0 ||
4899                     port->lun_map != NULL || port->lun_disable == NULL)
4900                         continue;
4901
4902                 /*
4903                  * Drop the lock before we call the frontend's disable
4904                  * routine, to avoid lock order reversals.
4905                  *
4906                  * XXX KDM what happens if the frontend list changes while
4907                  * we're traversing it?  It's unlikely, but should be handled.
4908                  */
4909                 mtx_unlock(&softc->ctl_lock);
4910                 retval = port->lun_disable(port->targ_lun_arg, lun->lun);
4911                 mtx_lock(&softc->ctl_lock);
4912                 if (retval != 0) {
4913                         printf("%s: FETD %s port %d returned error "
4914                                "%d for lun_disable on lun %jd\n",
4915                                __func__, port->port_name, port->targ_port,
4916                                retval, (intmax_t)lun->lun);
4917                 }
4918         }
4919
4920         mtx_unlock(&softc->ctl_lock);
4921         ctl_isc_announce_lun(lun);
4922
4923         return (0);
4924 }
4925
4926 int
4927 ctl_start_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_stop_lun(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_STOPPED;
4944         mtx_unlock(&lun->lun_lock);
4945         return (0);
4946 }
4947
4948 int
4949 ctl_lun_no_media(struct ctl_be_lun *be_lun)
4950 {
4951         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4952
4953         mtx_lock(&lun->lun_lock);
4954         lun->flags |= CTL_LUN_NO_MEDIA;
4955         mtx_unlock(&lun->lun_lock);
4956         return (0);
4957 }
4958
4959 int
4960 ctl_lun_has_media(struct ctl_be_lun *be_lun)
4961 {
4962         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4963         union ctl_ha_msg msg;
4964
4965         mtx_lock(&lun->lun_lock);
4966         lun->flags &= ~(CTL_LUN_NO_MEDIA | CTL_LUN_EJECTED);
4967         if (lun->flags & CTL_LUN_REMOVABLE)
4968                 ctl_est_ua_all(lun, -1, CTL_UA_MEDIUM_CHANGE);
4969         mtx_unlock(&lun->lun_lock);
4970         if ((lun->flags & CTL_LUN_REMOVABLE) &&
4971             lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
4972                 bzero(&msg.ua, sizeof(msg.ua));
4973                 msg.hdr.msg_type = CTL_MSG_UA;
4974                 msg.hdr.nexus.initid = -1;
4975                 msg.hdr.nexus.targ_port = -1;
4976                 msg.hdr.nexus.targ_lun = lun->lun;
4977                 msg.hdr.nexus.targ_mapped_lun = lun->lun;
4978                 msg.ua.ua_all = 1;
4979                 msg.ua.ua_set = 1;
4980                 msg.ua.ua_type = CTL_UA_MEDIUM_CHANGE;
4981                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg.ua),
4982                     M_WAITOK);
4983         }
4984         return (0);
4985 }
4986
4987 int
4988 ctl_lun_ejected(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_EJECTED;
4994         mtx_unlock(&lun->lun_lock);
4995         return (0);
4996 }
4997
4998 int
4999 ctl_lun_primary(struct ctl_be_lun *be_lun)
5000 {
5001         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
5002
5003         mtx_lock(&lun->lun_lock);
5004         lun->flags |= CTL_LUN_PRIMARY_SC;
5005         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
5006         mtx_unlock(&lun->lun_lock);
5007         ctl_isc_announce_lun(lun);
5008         return (0);
5009 }
5010
5011 int
5012 ctl_lun_secondary(struct ctl_be_lun *be_lun)
5013 {
5014         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
5015
5016         mtx_lock(&lun->lun_lock);
5017         lun->flags &= ~CTL_LUN_PRIMARY_SC;
5018         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
5019         mtx_unlock(&lun->lun_lock);
5020         ctl_isc_announce_lun(lun);
5021         return (0);
5022 }
5023
5024 int
5025 ctl_invalidate_lun(struct ctl_be_lun *be_lun)
5026 {
5027         struct ctl_lun *lun;
5028
5029         lun = (struct ctl_lun *)be_lun->ctl_lun;
5030
5031         mtx_lock(&lun->lun_lock);
5032
5033         /*
5034          * The LUN needs to be disabled before it can be marked invalid.
5035          */
5036         if ((lun->flags & CTL_LUN_DISABLED) == 0) {
5037                 mtx_unlock(&lun->lun_lock);
5038                 return (-1);
5039         }
5040         /*
5041          * Mark the LUN invalid.
5042          */
5043         lun->flags |= CTL_LUN_INVALID;
5044
5045         /*
5046          * If there is nothing in the OOA queue, go ahead and free the LUN.
5047          * If we have something in the OOA queue, we'll free it when the
5048          * last I/O completes.
5049          */
5050         if (TAILQ_EMPTY(&lun->ooa_queue)) {
5051                 mtx_unlock(&lun->lun_lock);
5052                 ctl_free_lun(lun);
5053         } else
5054                 mtx_unlock(&lun->lun_lock);
5055
5056         return (0);
5057 }
5058
5059 void
5060 ctl_lun_capacity_changed(struct ctl_be_lun *be_lun)
5061 {
5062         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
5063         union ctl_ha_msg msg;
5064
5065         mtx_lock(&lun->lun_lock);
5066         ctl_est_ua_all(lun, -1, CTL_UA_CAPACITY_CHANGE);
5067         mtx_unlock(&lun->lun_lock);
5068         if (lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
5069                 /* Send msg to other side. */
5070                 bzero(&msg.ua, sizeof(msg.ua));
5071                 msg.hdr.msg_type = CTL_MSG_UA;
5072                 msg.hdr.nexus.initid = -1;
5073                 msg.hdr.nexus.targ_port = -1;
5074                 msg.hdr.nexus.targ_lun = lun->lun;
5075                 msg.hdr.nexus.targ_mapped_lun = lun->lun;
5076                 msg.ua.ua_all = 1;
5077                 msg.ua.ua_set = 1;
5078                 msg.ua.ua_type = CTL_UA_CAPACITY_CHANGE;
5079                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg.ua),
5080                     M_WAITOK);
5081         }
5082 }
5083
5084 /*
5085  * Backend "memory move is complete" callback for requests that never
5086  * make it down to say RAIDCore's configuration code.
5087  */
5088 int
5089 ctl_config_move_done(union ctl_io *io)
5090 {
5091         int retval;
5092
5093         CTL_DEBUG_PRINT(("ctl_config_move_done\n"));
5094         KASSERT(io->io_hdr.io_type == CTL_IO_SCSI,
5095             ("Config I/O type isn't CTL_IO_SCSI (%d)!", io->io_hdr.io_type));
5096
5097         if ((io->io_hdr.port_status != 0) &&
5098             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5099              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5100                 ctl_set_internal_failure(&io->scsiio, /*sks_valid*/ 1,
5101                     /*retry_count*/ io->io_hdr.port_status);
5102         } else if (io->scsiio.kern_data_resid != 0 &&
5103             (io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT &&
5104             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5105              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5106                 ctl_set_invalid_field_ciu(&io->scsiio);
5107         }
5108
5109         if (ctl_debug & CTL_DEBUG_CDB_DATA)
5110                 ctl_data_print(io);
5111         if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN) ||
5112             ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
5113              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS) ||
5114             ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) {
5115                 /*
5116                  * XXX KDM just assuming a single pointer here, and not a
5117                  * S/G list.  If we start using S/G lists for config data,
5118                  * we'll need to know how to clean them up here as well.
5119                  */
5120                 if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5121                         free(io->scsiio.kern_data_ptr, M_CTL);
5122                 ctl_done(io);
5123                 retval = CTL_RETVAL_COMPLETE;
5124         } else {
5125                 /*
5126                  * XXX KDM now we need to continue data movement.  Some
5127                  * options:
5128                  * - call ctl_scsiio() again?  We don't do this for data
5129                  *   writes, because for those at least we know ahead of
5130                  *   time where the write will go and how long it is.  For
5131                  *   config writes, though, that information is largely
5132                  *   contained within the write itself, thus we need to
5133                  *   parse out the data again.
5134                  *
5135                  * - Call some other function once the data is in?
5136                  */
5137
5138                 /*
5139                  * XXX KDM call ctl_scsiio() again for now, and check flag
5140                  * bits to see whether we're allocated or not.
5141                  */
5142                 retval = ctl_scsiio(&io->scsiio);
5143         }
5144         return (retval);
5145 }
5146
5147 /*
5148  * This gets called by a backend driver when it is done with a
5149  * data_submit method.
5150  */
5151 void
5152 ctl_data_submit_done(union ctl_io *io)
5153 {
5154         /*
5155          * If the IO_CONT flag is set, we need to call the supplied
5156          * function to continue processing the I/O, instead of completing
5157          * the I/O just yet.
5158          *
5159          * If there is an error, though, we don't want to keep processing.
5160          * Instead, just send status back to the initiator.
5161          */
5162         if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5163             (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5164             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5165              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5166                 io->scsiio.io_cont(io);
5167                 return;
5168         }
5169         ctl_done(io);
5170 }
5171
5172 /*
5173  * This gets called by a backend driver when it is done with a
5174  * configuration write.
5175  */
5176 void
5177 ctl_config_write_done(union ctl_io *io)
5178 {
5179         uint8_t *buf;
5180
5181         /*
5182          * If the IO_CONT flag is set, we need to call the supplied
5183          * function to continue processing the I/O, instead of completing
5184          * the I/O just yet.
5185          *
5186          * If there is an error, though, we don't want to keep processing.
5187          * Instead, just send status back to the initiator.
5188          */
5189         if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5190             (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5191             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5192              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5193                 io->scsiio.io_cont(io);
5194                 return;
5195         }
5196         /*
5197          * Since a configuration write can be done for commands that actually
5198          * have data allocated, like write buffer, and commands that have
5199          * no data, like start/stop unit, we need to check here.
5200          */
5201         if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5202                 buf = io->scsiio.kern_data_ptr;
5203         else
5204                 buf = NULL;
5205         ctl_done(io);
5206         if (buf)
5207                 free(buf, M_CTL);
5208 }
5209
5210 void
5211 ctl_config_read_done(union ctl_io *io)
5212 {
5213         uint8_t *buf;
5214
5215         /*
5216          * If there is some error -- we are done, skip data transfer.
5217          */
5218         if ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0 ||
5219             ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
5220              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) {
5221                 if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5222                         buf = io->scsiio.kern_data_ptr;
5223                 else
5224                         buf = NULL;
5225                 ctl_done(io);
5226                 if (buf)
5227                         free(buf, M_CTL);
5228                 return;
5229         }
5230
5231         /*
5232          * If the IO_CONT flag is set, we need to call the supplied
5233          * function to continue processing the I/O, instead of completing
5234          * the I/O just yet.
5235          */
5236         if (io->io_hdr.flags & CTL_FLAG_IO_CONT) {
5237                 io->scsiio.io_cont(io);
5238                 return;
5239         }
5240
5241         ctl_datamove(io);
5242 }
5243
5244 /*
5245  * SCSI release command.
5246  */
5247 int
5248 ctl_scsi_release(struct ctl_scsiio *ctsio)
5249 {
5250         struct ctl_lun *lun = CTL_LUN(ctsio);
5251         uint32_t residx;
5252
5253         CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
5254
5255         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5256
5257         /*
5258          * XXX KDM right now, we only support LUN reservation.  We don't
5259          * support 3rd party reservations, or extent reservations, which
5260          * might actually need the parameter list.  If we've gotten this
5261          * far, we've got a LUN reservation.  Anything else got kicked out
5262          * above.  So, according to SPC, ignore the length.
5263          */
5264
5265         mtx_lock(&lun->lun_lock);
5266
5267         /*
5268          * According to SPC, it is not an error for an intiator to attempt
5269          * to release a reservation on a LUN that isn't reserved, or that
5270          * is reserved by another initiator.  The reservation can only be
5271          * released, though, by the initiator who made it or by one of
5272          * several reset type events.
5273          */
5274         if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == residx))
5275                         lun->flags &= ~CTL_LUN_RESERVED;
5276
5277         mtx_unlock(&lun->lun_lock);
5278
5279         ctl_set_success(ctsio);
5280         ctl_done((union ctl_io *)ctsio);
5281         return (CTL_RETVAL_COMPLETE);
5282 }
5283
5284 int
5285 ctl_scsi_reserve(struct ctl_scsiio *ctsio)
5286 {
5287         struct ctl_lun *lun = CTL_LUN(ctsio);
5288         uint32_t residx;
5289
5290         CTL_DEBUG_PRINT(("ctl_reserve\n"));
5291
5292         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5293
5294         /*
5295          * XXX KDM right now, we only support LUN reservation.  We don't
5296          * support 3rd party reservations, or extent reservations, which
5297          * might actually need the parameter list.  If we've gotten this
5298          * far, we've got a LUN reservation.  Anything else got kicked out
5299          * above.  So, according to SPC, ignore the length.
5300          */
5301
5302         mtx_lock(&lun->lun_lock);
5303         if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx != residx)) {
5304                 ctl_set_reservation_conflict(ctsio);
5305                 goto bailout;
5306         }
5307
5308         /* SPC-3 exceptions to SPC-2 RESERVE and RELEASE behavior. */
5309         if (lun->flags & CTL_LUN_PR_RESERVED) {
5310                 ctl_set_success(ctsio);
5311                 goto bailout;
5312         }
5313
5314         lun->flags |= CTL_LUN_RESERVED;
5315         lun->res_idx = residx;
5316         ctl_set_success(ctsio);
5317
5318 bailout:
5319         mtx_unlock(&lun->lun_lock);
5320         ctl_done((union ctl_io *)ctsio);
5321         return (CTL_RETVAL_COMPLETE);
5322 }
5323
5324 int
5325 ctl_start_stop(struct ctl_scsiio *ctsio)
5326 {
5327         struct ctl_lun *lun = CTL_LUN(ctsio);
5328         struct scsi_start_stop_unit *cdb;
5329         int retval;
5330
5331         CTL_DEBUG_PRINT(("ctl_start_stop\n"));
5332
5333         cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
5334
5335         if ((cdb->how & SSS_PC_MASK) == 0) {
5336                 if ((lun->flags & CTL_LUN_PR_RESERVED) &&
5337                     (cdb->how & SSS_START) == 0) {
5338                         uint32_t residx;
5339
5340                         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5341                         if (ctl_get_prkey(lun, residx) == 0 ||
5342                             (lun->pr_res_idx != residx && lun->pr_res_type < 4)) {
5343
5344                                 ctl_set_reservation_conflict(ctsio);
5345                                 ctl_done((union ctl_io *)ctsio);
5346                                 return (CTL_RETVAL_COMPLETE);
5347                         }
5348                 }
5349
5350                 if ((cdb->how & SSS_LOEJ) &&
5351                     (lun->flags & CTL_LUN_REMOVABLE) == 0) {
5352                         ctl_set_invalid_field(ctsio,
5353                                               /*sks_valid*/ 1,
5354                                               /*command*/ 1,
5355                                               /*field*/ 4,
5356                                               /*bit_valid*/ 1,
5357                                               /*bit*/ 1);
5358                         ctl_done((union ctl_io *)ctsio);
5359                         return (CTL_RETVAL_COMPLETE);
5360                 }
5361
5362                 if ((cdb->how & SSS_START) == 0 && (cdb->how & SSS_LOEJ) &&
5363                     lun->prevent_count > 0) {
5364                         /* "Medium removal prevented" */
5365                         ctl_set_sense(ctsio, /*current_error*/ 1,
5366                             /*sense_key*/(lun->flags & CTL_LUN_NO_MEDIA) ?
5367                              SSD_KEY_NOT_READY : SSD_KEY_ILLEGAL_REQUEST,
5368                             /*asc*/ 0x53, /*ascq*/ 0x02, SSD_ELEM_NONE);
5369                         ctl_done((union ctl_io *)ctsio);
5370                         return (CTL_RETVAL_COMPLETE);
5371                 }
5372         }
5373
5374         retval = lun->backend->config_write((union ctl_io *)ctsio);
5375         return (retval);
5376 }
5377
5378 int
5379 ctl_prevent_allow(struct ctl_scsiio *ctsio)
5380 {
5381         struct ctl_lun *lun = CTL_LUN(ctsio);
5382         struct scsi_prevent *cdb;
5383         int retval;
5384         uint32_t initidx;
5385
5386         CTL_DEBUG_PRINT(("ctl_prevent_allow\n"));
5387
5388         cdb = (struct scsi_prevent *)ctsio->cdb;
5389
5390         if ((lun->flags & CTL_LUN_REMOVABLE) == 0 || lun->prevent == NULL) {
5391                 ctl_set_invalid_opcode(ctsio);
5392                 ctl_done((union ctl_io *)ctsio);
5393                 return (CTL_RETVAL_COMPLETE);
5394         }
5395
5396         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5397         mtx_lock(&lun->lun_lock);
5398         if ((cdb->how & PR_PREVENT) &&
5399             ctl_is_set(lun->prevent, initidx) == 0) {
5400                 ctl_set_mask(lun->prevent, initidx);
5401                 lun->prevent_count++;
5402         } else if ((cdb->how & PR_PREVENT) == 0 &&
5403             ctl_is_set(lun->prevent, initidx)) {
5404                 ctl_clear_mask(lun->prevent, initidx);
5405                 lun->prevent_count--;
5406         }
5407         mtx_unlock(&lun->lun_lock);
5408         retval = lun->backend->config_write((union ctl_io *)ctsio);
5409         return (retval);
5410 }
5411
5412 /*
5413  * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
5414  * we don't really do anything with the LBA and length fields if the user
5415  * passes them in.  Instead we'll just flush out the cache for the entire
5416  * LUN.
5417  */
5418 int
5419 ctl_sync_cache(struct ctl_scsiio *ctsio)
5420 {
5421         struct ctl_lun *lun = CTL_LUN(ctsio);
5422         struct ctl_lba_len_flags *lbalen;
5423         uint64_t starting_lba;
5424         uint32_t block_count;
5425         int retval;
5426         uint8_t byte2;
5427
5428         CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
5429
5430         retval = 0;
5431
5432         switch (ctsio->cdb[0]) {
5433         case SYNCHRONIZE_CACHE: {
5434                 struct scsi_sync_cache *cdb;
5435                 cdb = (struct scsi_sync_cache *)ctsio->cdb;
5436
5437                 starting_lba = scsi_4btoul(cdb->begin_lba);
5438                 block_count = scsi_2btoul(cdb->lb_count);
5439                 byte2 = cdb->byte2;
5440                 break;
5441         }
5442         case SYNCHRONIZE_CACHE_16: {
5443                 struct scsi_sync_cache_16 *cdb;
5444                 cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5445
5446                 starting_lba = scsi_8btou64(cdb->begin_lba);
5447                 block_count = scsi_4btoul(cdb->lb_count);
5448                 byte2 = cdb->byte2;
5449                 break;
5450         }
5451         default:
5452                 ctl_set_invalid_opcode(ctsio);
5453                 ctl_done((union ctl_io *)ctsio);
5454                 goto bailout;
5455                 break; /* NOTREACHED */
5456         }
5457
5458         /*
5459          * We check the LBA and length, but don't do anything with them.
5460          * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5461          * get flushed.  This check will just help satisfy anyone who wants
5462          * to see an error for an out of range LBA.
5463          */
5464         if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5465                 ctl_set_lba_out_of_range(ctsio,
5466                     MAX(starting_lba, lun->be_lun->maxlba + 1));
5467                 ctl_done((union ctl_io *)ctsio);
5468                 goto bailout;
5469         }
5470
5471         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5472         lbalen->lba = starting_lba;
5473         lbalen->len = block_count;
5474         lbalen->flags = byte2;
5475         retval = lun->backend->config_write((union ctl_io *)ctsio);
5476
5477 bailout:
5478         return (retval);
5479 }
5480
5481 int
5482 ctl_format(struct ctl_scsiio *ctsio)
5483 {
5484         struct scsi_format *cdb;
5485         int length, defect_list_len;
5486
5487         CTL_DEBUG_PRINT(("ctl_format\n"));
5488
5489         cdb = (struct scsi_format *)ctsio->cdb;
5490
5491         length = 0;
5492         if (cdb->byte2 & SF_FMTDATA) {
5493                 if (cdb->byte2 & SF_LONGLIST)
5494                         length = sizeof(struct scsi_format_header_long);
5495                 else
5496                         length = sizeof(struct scsi_format_header_short);
5497         }
5498
5499         if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5500          && (length > 0)) {
5501                 ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5502                 ctsio->kern_data_len = length;
5503                 ctsio->kern_total_len = length;
5504                 ctsio->kern_rel_offset = 0;
5505                 ctsio->kern_sg_entries = 0;
5506                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5507                 ctsio->be_move_done = ctl_config_move_done;
5508                 ctl_datamove((union ctl_io *)ctsio);
5509
5510                 return (CTL_RETVAL_COMPLETE);
5511         }
5512
5513         defect_list_len = 0;
5514
5515         if (cdb->byte2 & SF_FMTDATA) {
5516                 if (cdb->byte2 & SF_LONGLIST) {
5517                         struct scsi_format_header_long *header;
5518
5519                         header = (struct scsi_format_header_long *)
5520                                 ctsio->kern_data_ptr;
5521
5522                         defect_list_len = scsi_4btoul(header->defect_list_len);
5523                         if (defect_list_len != 0) {
5524                                 ctl_set_invalid_field(ctsio,
5525                                                       /*sks_valid*/ 1,
5526                                                       /*command*/ 0,
5527                                                       /*field*/ 2,
5528                                                       /*bit_valid*/ 0,
5529                                                       /*bit*/ 0);
5530                                 goto bailout;
5531                         }
5532                 } else {
5533                         struct scsi_format_header_short *header;
5534
5535                         header = (struct scsi_format_header_short *)
5536                                 ctsio->kern_data_ptr;
5537
5538                         defect_list_len = scsi_2btoul(header->defect_list_len);
5539                         if (defect_list_len != 0) {
5540                                 ctl_set_invalid_field(ctsio,
5541                                                       /*sks_valid*/ 1,
5542                                                       /*command*/ 0,
5543                                                       /*field*/ 2,
5544                                                       /*bit_valid*/ 0,
5545                                                       /*bit*/ 0);
5546                                 goto bailout;
5547                         }
5548                 }
5549         }
5550
5551         ctl_set_success(ctsio);
5552 bailout:
5553
5554         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5555                 free(ctsio->kern_data_ptr, M_CTL);
5556                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5557         }
5558
5559         ctl_done((union ctl_io *)ctsio);
5560         return (CTL_RETVAL_COMPLETE);
5561 }
5562
5563 int
5564 ctl_read_buffer(struct ctl_scsiio *ctsio)
5565 {
5566         struct ctl_lun *lun = CTL_LUN(ctsio);
5567         uint64_t buffer_offset;
5568         uint32_t len;
5569         uint8_t byte2;
5570         static uint8_t descr[4];
5571         static uint8_t echo_descr[4] = { 0 };
5572
5573         CTL_DEBUG_PRINT(("ctl_read_buffer\n"));
5574
5575         switch (ctsio->cdb[0]) {
5576         case READ_BUFFER: {
5577                 struct scsi_read_buffer *cdb;
5578
5579                 cdb = (struct scsi_read_buffer *)ctsio->cdb;
5580                 buffer_offset = scsi_3btoul(cdb->offset);
5581                 len = scsi_3btoul(cdb->length);
5582                 byte2 = cdb->byte2;
5583                 break;
5584         }
5585         case READ_BUFFER_16: {
5586                 struct scsi_read_buffer_16 *cdb;
5587
5588                 cdb = (struct scsi_read_buffer_16 *)ctsio->cdb;
5589                 buffer_offset = scsi_8btou64(cdb->offset);
5590                 len = scsi_4btoul(cdb->length);
5591                 byte2 = cdb->byte2;
5592                 break;
5593         }
5594         default: /* This shouldn't happen. */
5595                 ctl_set_invalid_opcode(ctsio);
5596                 ctl_done((union ctl_io *)ctsio);
5597                 return (CTL_RETVAL_COMPLETE);
5598         }
5599
5600         if (buffer_offset > CTL_WRITE_BUFFER_SIZE ||
5601             buffer_offset + len > CTL_WRITE_BUFFER_SIZE) {
5602                 ctl_set_invalid_field(ctsio,
5603                                       /*sks_valid*/ 1,
5604                                       /*command*/ 1,
5605                                       /*field*/ 6,
5606                                       /*bit_valid*/ 0,
5607                                       /*bit*/ 0);
5608                 ctl_done((union ctl_io *)ctsio);
5609                 return (CTL_RETVAL_COMPLETE);
5610         }
5611
5612         if ((byte2 & RWB_MODE) == RWB_MODE_DESCR) {
5613                 descr[0] = 0;
5614                 scsi_ulto3b(CTL_WRITE_BUFFER_SIZE, &descr[1]);
5615                 ctsio->kern_data_ptr = descr;
5616                 len = min(len, sizeof(descr));
5617         } else if ((byte2 & RWB_MODE) == RWB_MODE_ECHO_DESCR) {
5618                 ctsio->kern_data_ptr = echo_descr;
5619                 len = min(len, sizeof(echo_descr));
5620         } else {
5621                 if (lun->write_buffer == NULL) {
5622                         lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE,
5623                             M_CTL, M_WAITOK);
5624                 }
5625                 ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5626         }
5627         ctsio->kern_data_len = len;
5628         ctsio->kern_total_len = len;
5629         ctsio->kern_rel_offset = 0;
5630         ctsio->kern_sg_entries = 0;
5631         ctl_set_success(ctsio);
5632         ctsio->be_move_done = ctl_config_move_done;
5633         ctl_datamove((union ctl_io *)ctsio);
5634         return (CTL_RETVAL_COMPLETE);
5635 }
5636
5637 int
5638 ctl_write_buffer(struct ctl_scsiio *ctsio)
5639 {
5640         struct ctl_lun *lun = CTL_LUN(ctsio);
5641         struct scsi_write_buffer *cdb;
5642         int buffer_offset, len;
5643
5644         CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5645
5646         cdb = (struct scsi_write_buffer *)ctsio->cdb;
5647
5648         len = scsi_3btoul(cdb->length);
5649         buffer_offset = scsi_3btoul(cdb->offset);
5650
5651         if (buffer_offset + len > CTL_WRITE_BUFFER_SIZE) {
5652                 ctl_set_invalid_field(ctsio,
5653                                       /*sks_valid*/ 1,
5654                                       /*command*/ 1,
5655                                       /*field*/ 6,
5656                                       /*bit_valid*/ 0,
5657                                       /*bit*/ 0);
5658                 ctl_done((union ctl_io *)ctsio);
5659                 return (CTL_RETVAL_COMPLETE);
5660         }
5661
5662         /*
5663          * If we've got a kernel request that hasn't been malloced yet,
5664          * malloc it and tell the caller the data buffer is here.
5665          */
5666         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5667                 if (lun->write_buffer == NULL) {
5668                         lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE,
5669                             M_CTL, M_WAITOK);
5670                 }
5671                 ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5672                 ctsio->kern_data_len = len;
5673                 ctsio->kern_total_len = len;
5674                 ctsio->kern_rel_offset = 0;
5675                 ctsio->kern_sg_entries = 0;
5676                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5677                 ctsio->be_move_done = ctl_config_move_done;
5678                 ctl_datamove((union ctl_io *)ctsio);
5679
5680                 return (CTL_RETVAL_COMPLETE);
5681         }
5682
5683         ctl_set_success(ctsio);
5684         ctl_done((union ctl_io *)ctsio);
5685         return (CTL_RETVAL_COMPLETE);
5686 }
5687
5688 int
5689 ctl_write_same(struct ctl_scsiio *ctsio)
5690 {
5691         struct ctl_lun *lun = CTL_LUN(ctsio);
5692         struct ctl_lba_len_flags *lbalen;
5693         uint64_t lba;
5694         uint32_t num_blocks;
5695         int len, retval;
5696         uint8_t byte2;
5697
5698         CTL_DEBUG_PRINT(("ctl_write_same\n"));
5699
5700         switch (ctsio->cdb[0]) {
5701         case WRITE_SAME_10: {
5702                 struct scsi_write_same_10 *cdb;
5703
5704                 cdb = (struct scsi_write_same_10 *)ctsio->cdb;
5705
5706                 lba = scsi_4btoul(cdb->addr);
5707                 num_blocks = scsi_2btoul(cdb->length);
5708                 byte2 = cdb->byte2;
5709                 break;
5710         }
5711         case WRITE_SAME_16: {
5712                 struct scsi_write_same_16 *cdb;
5713
5714                 cdb = (struct scsi_write_same_16 *)ctsio->cdb;
5715
5716                 lba = scsi_8btou64(cdb->addr);
5717                 num_blocks = scsi_4btoul(cdb->length);
5718                 byte2 = cdb->byte2;
5719                 break;
5720         }
5721         default:
5722                 /*
5723                  * We got a command we don't support.  This shouldn't
5724                  * happen, commands should be filtered out above us.
5725                  */
5726                 ctl_set_invalid_opcode(ctsio);
5727                 ctl_done((union ctl_io *)ctsio);
5728
5729                 return (CTL_RETVAL_COMPLETE);
5730                 break; /* NOTREACHED */
5731         }
5732
5733         /* ANCHOR flag can be used only together with UNMAP */
5734         if ((byte2 & SWS_UNMAP) == 0 && (byte2 & SWS_ANCHOR) != 0) {
5735                 ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
5736                     /*command*/ 1, /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 0);
5737                 ctl_done((union ctl_io *)ctsio);
5738                 return (CTL_RETVAL_COMPLETE);
5739         }
5740
5741         /*
5742          * The first check is to make sure we're in bounds, the second
5743          * check is to catch wrap-around problems.  If the lba + num blocks
5744          * is less than the lba, then we've wrapped around and the block
5745          * range is invalid anyway.
5746          */
5747         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5748          || ((lba + num_blocks) < lba)) {
5749                 ctl_set_lba_out_of_range(ctsio,
5750                     MAX(lba, lun->be_lun->maxlba + 1));
5751                 ctl_done((union ctl_io *)ctsio);
5752                 return (CTL_RETVAL_COMPLETE);
5753         }
5754
5755         /* Zero number of blocks means "to the last logical block" */
5756         if (num_blocks == 0) {
5757                 if ((lun->be_lun->maxlba + 1) - lba > UINT32_MAX) {
5758                         ctl_set_invalid_field(ctsio,
5759                                               /*sks_valid*/ 0,
5760                                               /*command*/ 1,
5761                                               /*field*/ 0,
5762                                               /*bit_valid*/ 0,
5763                                               /*bit*/ 0);
5764                         ctl_done((union ctl_io *)ctsio);
5765                         return (CTL_RETVAL_COMPLETE);
5766                 }
5767                 num_blocks = (lun->be_lun->maxlba + 1) - lba;
5768         }
5769
5770         len = lun->be_lun->blocksize;
5771
5772         /*
5773          * If we've got a kernel request that hasn't been malloced yet,
5774          * malloc it and tell the caller the data buffer is here.
5775          */
5776         if ((byte2 & SWS_NDOB) == 0 &&
5777             (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5778                 ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);
5779                 ctsio->kern_data_len = len;
5780                 ctsio->kern_total_len = len;
5781                 ctsio->kern_rel_offset = 0;
5782                 ctsio->kern_sg_entries = 0;
5783                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5784                 ctsio->be_move_done = ctl_config_move_done;
5785                 ctl_datamove((union ctl_io *)ctsio);
5786
5787                 return (CTL_RETVAL_COMPLETE);
5788         }
5789
5790         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5791         lbalen->lba = lba;
5792         lbalen->len = num_blocks;
5793         lbalen->flags = byte2;
5794         retval = lun->backend->config_write((union ctl_io *)ctsio);
5795
5796         return (retval);
5797 }
5798
5799 int
5800 ctl_unmap(struct ctl_scsiio *ctsio)
5801 {
5802         struct ctl_lun *lun = CTL_LUN(ctsio);
5803         struct scsi_unmap *cdb;
5804         struct ctl_ptr_len_flags *ptrlen;
5805         struct scsi_unmap_header *hdr;
5806         struct scsi_unmap_desc *buf, *end, *endnz, *range;
5807         uint64_t lba;
5808         uint32_t num_blocks;
5809         int len, retval;
5810         uint8_t byte2;
5811
5812         CTL_DEBUG_PRINT(("ctl_unmap\n"));
5813
5814         cdb = (struct scsi_unmap *)ctsio->cdb;
5815         len = scsi_2btoul(cdb->length);
5816         byte2 = cdb->byte2;
5817
5818         /*
5819          * If we've got a kernel request that hasn't been malloced yet,
5820          * malloc it and tell the caller the data buffer is here.
5821          */
5822         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5823                 ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);
5824                 ctsio->kern_data_len = len;
5825                 ctsio->kern_total_len = len;
5826                 ctsio->kern_rel_offset = 0;
5827                 ctsio->kern_sg_entries = 0;
5828                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5829                 ctsio->be_move_done = ctl_config_move_done;
5830                 ctl_datamove((union ctl_io *)ctsio);
5831
5832                 return (CTL_RETVAL_COMPLETE);
5833         }
5834
5835         len = ctsio->kern_total_len - ctsio->kern_data_resid;
5836         hdr = (struct scsi_unmap_header *)ctsio->kern_data_ptr;
5837         if (len < sizeof (*hdr) ||
5838             len < (scsi_2btoul(hdr->length) + sizeof(hdr->length)) ||
5839             len < (scsi_2btoul(hdr->desc_length) + sizeof (*hdr)) ||
5840             scsi_2btoul(hdr->desc_length) % sizeof(*buf) != 0) {
5841                 ctl_set_invalid_field(ctsio,
5842                                       /*sks_valid*/ 0,
5843                                       /*command*/ 0,
5844                                       /*field*/ 0,
5845                                       /*bit_valid*/ 0,
5846                                       /*bit*/ 0);
5847                 goto done;
5848         }
5849         len = scsi_2btoul(hdr->desc_length);
5850         buf = (struct scsi_unmap_desc *)(hdr + 1);
5851         end = buf + len / sizeof(*buf);
5852
5853         endnz = buf;
5854         for (range = buf; range < end; range++) {
5855                 lba = scsi_8btou64(range->lba);
5856                 num_blocks = scsi_4btoul(range->length);
5857                 if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5858                  || ((lba + num_blocks) < lba)) {
5859                         ctl_set_lba_out_of_range(ctsio,
5860                             MAX(lba, lun->be_lun->maxlba + 1));
5861                         ctl_done((union ctl_io *)ctsio);
5862                         return (CTL_RETVAL_COMPLETE);
5863                 }
5864                 if (num_blocks != 0)
5865                         endnz = range + 1;
5866         }
5867
5868         /*
5869          * Block backend can not handle zero last range.
5870          * Filter it out and return if there is nothing left.
5871          */
5872         len = (uint8_t *)endnz - (uint8_t *)buf;
5873         if (len == 0) {
5874                 ctl_set_success(ctsio);
5875                 goto done;
5876         }
5877
5878         mtx_lock(&lun->lun_lock);
5879         ptrlen = (struct ctl_ptr_len_flags *)
5880             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5881         ptrlen->ptr = (void *)buf;
5882         ptrlen->len = len;
5883         ptrlen->flags = byte2;
5884         ctl_check_blocked(lun);
5885         mtx_unlock(&lun->lun_lock);
5886
5887         retval = lun->backend->config_write((union ctl_io *)ctsio);
5888         return (retval);
5889
5890 done:
5891         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5892                 free(ctsio->kern_data_ptr, M_CTL);
5893                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5894         }
5895         ctl_done((union ctl_io *)ctsio);
5896         return (CTL_RETVAL_COMPLETE);
5897 }
5898
5899 int
5900 ctl_default_page_handler(struct ctl_scsiio *ctsio,
5901                          struct ctl_page_index *page_index, uint8_t *page_ptr)
5902 {
5903         struct ctl_lun *lun = CTL_LUN(ctsio);
5904         uint8_t *current_cp;
5905         int set_ua;
5906         uint32_t initidx;
5907
5908         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5909         set_ua = 0;
5910
5911         current_cp = (page_index->page_data + (page_index->page_len *
5912             CTL_PAGE_CURRENT));
5913
5914         mtx_lock(&lun->lun_lock);
5915         if (memcmp(current_cp, page_ptr, page_index->page_len)) {
5916                 memcpy(current_cp, page_ptr, page_index->page_len);
5917                 set_ua = 1;
5918         }
5919         if (set_ua != 0)
5920                 ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
5921         mtx_unlock(&lun->lun_lock);
5922         if (set_ua) {
5923                 ctl_isc_announce_mode(lun,
5924                     ctl_get_initindex(&ctsio->io_hdr.nexus),
5925                     page_index->page_code, page_index->subpage);
5926         }
5927         return (CTL_RETVAL_COMPLETE);
5928 }
5929
5930 static void
5931 ctl_ie_timer(void *arg)
5932 {
5933         struct ctl_lun *lun = arg;
5934         uint64_t t;
5935
5936         if (lun->ie_asc == 0)
5937                 return;
5938
5939         if (lun->MODE_IE.mrie == SIEP_MRIE_UA)
5940                 ctl_est_ua_all(lun, -1, CTL_UA_IE);
5941         else
5942                 lun->ie_reported = 0;
5943
5944         if (lun->ie_reportcnt < scsi_4btoul(lun->MODE_IE.report_count)) {
5945                 lun->ie_reportcnt++;
5946                 t = scsi_4btoul(lun->MODE_IE.interval_timer);
5947                 if (t == 0 || t == UINT32_MAX)
5948                         t = 3000;  /* 5 min */
5949                 callout_schedule(&lun->ie_callout, t * hz / 10);
5950         }
5951 }
5952
5953 int
5954 ctl_ie_page_handler(struct ctl_scsiio *ctsio,
5955                          struct ctl_page_index *page_index, uint8_t *page_ptr)
5956 {
5957         struct ctl_lun *lun = CTL_LUN(ctsio);
5958         struct scsi_info_exceptions_page *pg;
5959         uint64_t t;
5960
5961         (void)ctl_default_page_handler(ctsio, page_index, page_ptr);
5962
5963         pg = (struct scsi_info_exceptions_page *)page_ptr;
5964         mtx_lock(&lun->lun_lock);
5965         if (pg->info_flags & SIEP_FLAGS_TEST) {
5966                 lun->ie_asc = 0x5d;
5967                 lun->ie_ascq = 0xff;
5968                 if (pg->mrie == SIEP_MRIE_UA) {
5969                         ctl_est_ua_all(lun, -1, CTL_UA_IE);
5970                         lun->ie_reported = 1;
5971                 } else {
5972                         ctl_clr_ua_all(lun, -1, CTL_UA_IE);
5973                         lun->ie_reported = -1;
5974                 }
5975                 lun->ie_reportcnt = 1;
5976                 if (lun->ie_reportcnt < scsi_4btoul(pg->report_count)) {
5977                         lun->ie_reportcnt++;
5978                         t = scsi_4btoul(pg->interval_timer);
5979                         if (t == 0 || t == UINT32_MAX)
5980                                 t = 3000;  /* 5 min */
5981                         callout_reset(&lun->ie_callout, t * hz / 10,
5982                             ctl_ie_timer, lun);
5983                 }
5984         } else {
5985                 lun->ie_asc = 0;
5986                 lun->ie_ascq = 0;
5987                 lun->ie_reported = 1;
5988                 ctl_clr_ua_all(lun, -1, CTL_UA_IE);
5989                 lun->ie_reportcnt = UINT32_MAX;
5990                 callout_stop(&lun->ie_callout);
5991         }
5992         mtx_unlock(&lun->lun_lock);
5993         return (CTL_RETVAL_COMPLETE);
5994 }
5995
5996 static int
5997 ctl_do_mode_select(union ctl_io *io)
5998 {
5999         struct ctl_lun *lun = CTL_LUN(io);
6000         struct scsi_mode_page_header *page_header;
6001         struct ctl_page_index *page_index;
6002         struct ctl_scsiio *ctsio;
6003         int page_len, page_len_offset, page_len_size;
6004         union ctl_modepage_info *modepage_info;
6005         uint16_t *len_left, *len_used;
6006         int retval, i;
6007
6008         ctsio = &io->scsiio;
6009         page_index = NULL;
6010         page_len = 0;
6011
6012         modepage_info = (union ctl_modepage_info *)
6013                 ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6014         len_left = &modepage_info->header.len_left;
6015         len_used = &modepage_info->header.len_used;
6016
6017 do_next_page:
6018
6019         page_header = (struct scsi_mode_page_header *)
6020                 (ctsio->kern_data_ptr + *len_used);
6021
6022         if (*len_left == 0) {
6023                 free(ctsio->kern_data_ptr, M_CTL);
6024                 ctl_set_success(ctsio);
6025                 ctl_done((union ctl_io *)ctsio);
6026                 return (CTL_RETVAL_COMPLETE);
6027         } else if (*len_left < sizeof(struct scsi_mode_page_header)) {
6028
6029                 free(ctsio->kern_data_ptr, M_CTL);
6030                 ctl_set_param_len_error(ctsio);
6031                 ctl_done((union ctl_io *)ctsio);
6032                 return (CTL_RETVAL_COMPLETE);
6033
6034         } else if ((page_header->page_code & SMPH_SPF)
6035                 && (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
6036
6037                 free(ctsio->kern_data_ptr, M_CTL);
6038                 ctl_set_param_len_error(ctsio);
6039                 ctl_done((union ctl_io *)ctsio);
6040                 return (CTL_RETVAL_COMPLETE);
6041         }
6042
6043
6044         /*
6045          * XXX KDM should we do something with the block descriptor?
6046          */
6047         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6048                 page_index = &lun->mode_pages.index[i];
6049                 if (lun->be_lun->lun_type == T_DIRECT &&
6050                     (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6051                         continue;
6052                 if (lun->be_lun->lun_type == T_PROCESSOR &&
6053                     (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6054                         continue;
6055                 if (lun->be_lun->lun_type == T_CDROM &&
6056                     (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6057                         continue;
6058
6059                 if ((page_index->page_code & SMPH_PC_MASK) !=
6060                     (page_header->page_code & SMPH_PC_MASK))
6061                         continue;
6062
6063                 /*
6064                  * If neither page has a subpage code, then we've got a
6065                  * match.
6066                  */
6067                 if (((page_index->page_code & SMPH_SPF) == 0)
6068                  && ((page_header->page_code & SMPH_SPF) == 0)) {
6069                         page_len = page_header->page_length;
6070                         break;
6071                 }
6072
6073                 /*
6074                  * If both pages have subpages, then the subpage numbers
6075                  * have to match.
6076                  */
6077                 if ((page_index->page_code & SMPH_SPF)
6078                   && (page_header->page_code & SMPH_SPF)) {
6079                         struct scsi_mode_page_header_sp *sph;
6080
6081                         sph = (struct scsi_mode_page_header_sp *)page_header;
6082                         if (page_index->subpage == sph->subpage) {
6083                                 page_len = scsi_2btoul(sph->page_length);
6084                                 break;
6085                         }
6086                 }
6087         }
6088
6089         /*
6090          * If we couldn't find the page, or if we don't have a mode select
6091          * handler for it, send back an error to the user.
6092          */
6093         if ((i >= CTL_NUM_MODE_PAGES)
6094          || (page_index->select_handler == NULL)) {
6095                 ctl_set_invalid_field(ctsio,
6096                                       /*sks_valid*/ 1,
6097                                       /*command*/ 0,
6098                                       /*field*/ *len_used,
6099                                       /*bit_valid*/ 0,
6100                                       /*bit*/ 0);
6101                 free(ctsio->kern_data_ptr, M_CTL);
6102                 ctl_done((union ctl_io *)ctsio);
6103                 return (CTL_RETVAL_COMPLETE);
6104         }
6105
6106         if (page_index->page_code & SMPH_SPF) {
6107                 page_len_offset = 2;
6108                 page_len_size = 2;
6109         } else {
6110                 page_len_size = 1;
6111                 page_len_offset = 1;
6112         }
6113
6114         /*
6115          * If the length the initiator gives us isn't the one we specify in
6116          * the mode page header, or if they didn't specify enough data in
6117          * the CDB to avoid truncating this page, kick out the request.
6118          */
6119         if (page_len != page_index->page_len - page_len_offset - page_len_size) {
6120                 ctl_set_invalid_field(ctsio,
6121                                       /*sks_valid*/ 1,
6122                                       /*command*/ 0,
6123                                       /*field*/ *len_used + page_len_offset,
6124                                       /*bit_valid*/ 0,
6125                                       /*bit*/ 0);
6126                 free(ctsio->kern_data_ptr, M_CTL);
6127                 ctl_done((union ctl_io *)ctsio);
6128                 return (CTL_RETVAL_COMPLETE);
6129         }
6130         if (*len_left < page_index->page_len) {
6131                 free(ctsio->kern_data_ptr, M_CTL);
6132                 ctl_set_param_len_error(ctsio);
6133                 ctl_done((union ctl_io *)ctsio);
6134                 return (CTL_RETVAL_COMPLETE);
6135         }
6136
6137         /*
6138          * Run through the mode page, checking to make sure that the bits
6139          * the user changed are actually legal for him to change.
6140          */
6141         for (i = 0; i < page_index->page_len; i++) {
6142                 uint8_t *user_byte, *change_mask, *current_byte;
6143                 int bad_bit;
6144                 int j;
6145
6146                 user_byte = (uint8_t *)page_header + i;
6147                 change_mask = page_index->page_data +
6148                               (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
6149                 current_byte = page_index->page_data +
6150                                (page_index->page_len * CTL_PAGE_CURRENT) + i;
6151
6152                 /*
6153                  * Check to see whether the user set any bits in this byte
6154                  * that he is not allowed to set.
6155                  */
6156                 if ((*user_byte & ~(*change_mask)) ==
6157                     (*current_byte & ~(*change_mask)))
6158                         continue;
6159
6160                 /*
6161                  * Go through bit by bit to determine which one is illegal.
6162                  */
6163                 bad_bit = 0;
6164                 for (j = 7; j >= 0; j--) {
6165                         if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
6166                             (((1 << i) & ~(*change_mask)) & *current_byte)) {
6167                                 bad_bit = i;
6168                                 break;
6169                         }
6170                 }
6171                 ctl_set_invalid_field(ctsio,
6172                                       /*sks_valid*/ 1,
6173                                       /*command*/ 0,
6174                                       /*field*/ *len_used + i,
6175                                       /*bit_valid*/ 1,
6176                                       /*bit*/ bad_bit);
6177                 free(ctsio->kern_data_ptr, M_CTL);
6178                 ctl_done((union ctl_io *)ctsio);
6179                 return (CTL_RETVAL_COMPLETE);
6180         }
6181
6182         /*
6183          * Decrement these before we call the page handler, since we may
6184          * end up getting called back one way or another before the handler
6185          * returns to this context.
6186          */
6187         *len_left -= page_index->page_len;
6188         *len_used += page_index->page_len;
6189
6190         retval = page_index->select_handler(ctsio, page_index,
6191                                             (uint8_t *)page_header);
6192
6193         /*
6194          * If the page handler returns CTL_RETVAL_QUEUED, then we need to
6195          * wait until this queued command completes to finish processing
6196          * the mode page.  If it returns anything other than
6197          * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
6198          * already set the sense information, freed the data pointer, and
6199          * completed the io for us.
6200          */
6201         if (retval != CTL_RETVAL_COMPLETE)
6202                 goto bailout_no_done;
6203
6204         /*
6205          * If the initiator sent us more than one page, parse the next one.
6206          */
6207         if (*len_left > 0)
6208                 goto do_next_page;
6209
6210         ctl_set_success(ctsio);
6211         free(ctsio->kern_data_ptr, M_CTL);
6212         ctl_done((union ctl_io *)ctsio);
6213
6214 bailout_no_done:
6215
6216         return (CTL_RETVAL_COMPLETE);
6217
6218 }
6219
6220 int
6221 ctl_mode_select(struct ctl_scsiio *ctsio)
6222 {
6223         struct ctl_lun *lun = CTL_LUN(ctsio);
6224         union ctl_modepage_info *modepage_info;
6225         int bd_len, i, header_size, param_len, rtd;
6226         uint32_t initidx;
6227
6228         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
6229         switch (ctsio->cdb[0]) {
6230         case MODE_SELECT_6: {
6231                 struct scsi_mode_select_6 *cdb;
6232
6233                 cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
6234
6235                 rtd = (cdb->byte2 & SMS_RTD) ? 1 : 0;
6236                 param_len = cdb->length;
6237                 header_size = sizeof(struct scsi_mode_header_6);
6238                 break;
6239         }
6240         case MODE_SELECT_10: {
6241                 struct scsi_mode_select_10 *cdb;
6242
6243                 cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
6244
6245                 rtd = (cdb->byte2 & SMS_RTD) ? 1 : 0;
6246                 param_len = scsi_2btoul(cdb->length);
6247                 header_size = sizeof(struct scsi_mode_header_10);
6248                 break;
6249         }
6250         default:
6251                 ctl_set_invalid_opcode(ctsio);
6252                 ctl_done((union ctl_io *)ctsio);
6253                 return (CTL_RETVAL_COMPLETE);
6254         }
6255
6256         if (rtd) {
6257                 if (param_len != 0) {
6258                         ctl_set_invalid_field(ctsio, /*sks_valid*/ 0,
6259                             /*command*/ 1, /*field*/ 0,
6260                             /*bit_valid*/ 0, /*bit*/ 0);
6261                         ctl_done((union ctl_io *)ctsio);
6262                         return (CTL_RETVAL_COMPLETE);
6263                 }
6264
6265                 /* Revert to defaults. */
6266                 ctl_init_page_index(lun);
6267                 mtx_lock(&lun->lun_lock);
6268                 ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
6269                 mtx_unlock(&lun->lun_lock);
6270                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6271                         ctl_isc_announce_mode(lun, -1,
6272                             lun->mode_pages.index[i].page_code & SMPH_PC_MASK,
6273                             lun->mode_pages.index[i].subpage);
6274                 }
6275                 ctl_set_success(ctsio);
6276                 ctl_done((union ctl_io *)ctsio);
6277                 return (CTL_RETVAL_COMPLETE);
6278         }
6279
6280         /*
6281          * From SPC-3:
6282          * "A parameter list length of zero indicates that the Data-Out Buffer
6283          * shall be empty. This condition shall not be considered as an error."
6284          */
6285         if (param_len == 0) {
6286                 ctl_set_success(ctsio);
6287                 ctl_done((union ctl_io *)ctsio);
6288                 return (CTL_RETVAL_COMPLETE);
6289         }
6290
6291         /*
6292          * Since we'll hit this the first time through, prior to
6293          * allocation, we don't need to free a data buffer here.
6294          */
6295         if (param_len < header_size) {
6296                 ctl_set_param_len_error(ctsio);
6297                 ctl_done((union ctl_io *)ctsio);
6298                 return (CTL_RETVAL_COMPLETE);
6299         }
6300
6301         /*
6302          * Allocate the data buffer and grab the user's data.  In theory,
6303          * we shouldn't have to sanity check the parameter list length here
6304          * because the maximum size is 64K.  We should be able to malloc
6305          * that much without too many problems.
6306          */
6307         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6308                 ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6309                 ctsio->kern_data_len = param_len;
6310                 ctsio->kern_total_len = param_len;
6311                 ctsio->kern_rel_offset = 0;
6312                 ctsio->kern_sg_entries = 0;
6313                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6314                 ctsio->be_move_done = ctl_config_move_done;
6315                 ctl_datamove((union ctl_io *)ctsio);
6316
6317                 return (CTL_RETVAL_COMPLETE);
6318         }
6319
6320         switch (ctsio->cdb[0]) {
6321         case MODE_SELECT_6: {
6322                 struct scsi_mode_header_6 *mh6;
6323
6324                 mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6325                 bd_len = mh6->blk_desc_len;
6326                 break;
6327         }
6328         case MODE_SELECT_10: {
6329                 struct scsi_mode_header_10 *mh10;
6330
6331                 mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6332                 bd_len = scsi_2btoul(mh10->blk_desc_len);
6333                 break;
6334         }
6335         default:
6336                 panic("%s: Invalid CDB type %#x", __func__, ctsio->cdb[0]);
6337         }
6338
6339         if (param_len < (header_size + bd_len)) {
6340                 free(ctsio->kern_data_ptr, M_CTL);
6341                 ctl_set_param_len_error(ctsio);
6342                 ctl_done((union ctl_io *)ctsio);
6343                 return (CTL_RETVAL_COMPLETE);
6344         }
6345
6346         /*
6347          * Set the IO_CONT flag, so that if this I/O gets passed to
6348          * ctl_config_write_done(), it'll get passed back to
6349          * ctl_do_mode_select() for further processing, or completion if
6350          * we're all done.
6351          */
6352         ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6353         ctsio->io_cont = ctl_do_mode_select;
6354
6355         modepage_info = (union ctl_modepage_info *)
6356                 ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6357         memset(modepage_info, 0, sizeof(*modepage_info));
6358         modepage_info->header.len_left = param_len - header_size - bd_len;
6359         modepage_info->header.len_used = header_size + bd_len;
6360
6361         return (ctl_do_mode_select((union ctl_io *)ctsio));
6362 }
6363
6364 int
6365 ctl_mode_sense(struct ctl_scsiio *ctsio)
6366 {
6367         struct ctl_lun *lun = CTL_LUN(ctsio);
6368         int pc, page_code, dbd, subpage;
6369         int alloc_len, page_len, header_len, total_len;
6370         struct scsi_mode_block_descr *block_desc;
6371         struct ctl_page_index *page_index;
6372
6373         dbd = 0;
6374         block_desc = NULL;
6375
6376         CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6377
6378         switch (ctsio->cdb[0]) {
6379         case MODE_SENSE_6: {
6380                 struct scsi_mode_sense_6 *cdb;
6381
6382                 cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6383
6384                 header_len = sizeof(struct scsi_mode_hdr_6);
6385                 if (cdb->byte2 & SMS_DBD)
6386                         dbd = 1;
6387                 else
6388                         header_len += sizeof(struct scsi_mode_block_descr);
6389
6390                 pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6391                 page_code = cdb->page & SMS_PAGE_CODE;
6392                 subpage = cdb->subpage;
6393                 alloc_len = cdb->length;
6394                 break;
6395         }
6396         case MODE_SENSE_10: {
6397                 struct scsi_mode_sense_10 *cdb;
6398
6399                 cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6400
6401                 header_len = sizeof(struct scsi_mode_hdr_10);
6402
6403                 if (cdb->byte2 & SMS_DBD)
6404                         dbd = 1;
6405                 else
6406                         header_len += sizeof(struct scsi_mode_block_descr);
6407                 pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6408                 page_code = cdb->page & SMS_PAGE_CODE;
6409                 subpage = cdb->subpage;
6410                 alloc_len = scsi_2btoul(cdb->length);
6411                 break;
6412         }
6413         default:
6414                 ctl_set_invalid_opcode(ctsio);
6415                 ctl_done((union ctl_io *)ctsio);
6416                 return (CTL_RETVAL_COMPLETE);
6417                 break; /* NOTREACHED */
6418         }
6419
6420         /*
6421          * We have to make a first pass through to calculate the size of
6422          * the pages that match the user's query.  Then we allocate enough
6423          * memory to hold it, and actually copy the data into the buffer.
6424          */
6425         switch (page_code) {
6426         case SMS_ALL_PAGES_PAGE: {
6427                 u_int i;
6428
6429                 page_len = 0;
6430
6431                 /*
6432                  * At the moment, values other than 0 and 0xff here are
6433                  * reserved according to SPC-3.
6434                  */
6435                 if ((subpage != SMS_SUBPAGE_PAGE_0)
6436                  && (subpage != SMS_SUBPAGE_ALL)) {
6437                         ctl_set_invalid_field(ctsio,
6438                                               /*sks_valid*/ 1,
6439                                               /*command*/ 1,
6440                                               /*field*/ 3,
6441                                               /*bit_valid*/ 0,
6442                                               /*bit*/ 0);
6443                         ctl_done((union ctl_io *)ctsio);
6444                         return (CTL_RETVAL_COMPLETE);
6445                 }
6446
6447                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6448                         page_index = &lun->mode_pages.index[i];
6449
6450                         /* Make sure the page is supported for this dev type */
6451                         if (lun->be_lun->lun_type == T_DIRECT &&
6452                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6453                                 continue;
6454                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6455                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6456                                 continue;
6457                         if (lun->be_lun->lun_type == T_CDROM &&
6458                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6459                                 continue;
6460
6461                         /*
6462                          * We don't use this subpage if the user didn't
6463                          * request all subpages.
6464                          */
6465                         if ((page_index->subpage != 0)
6466                          && (subpage == SMS_SUBPAGE_PAGE_0))
6467                                 continue;
6468
6469 #if 0
6470                         printf("found page %#x len %d\n",
6471                                page_index->page_code & SMPH_PC_MASK,
6472                                page_index->page_len);
6473 #endif
6474                         page_len += page_index->page_len;
6475                 }
6476                 break;
6477         }
6478         default: {
6479                 u_int i;
6480
6481                 page_len = 0;
6482
6483                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6484                         page_index = &lun->mode_pages.index[i];
6485
6486                         /* Make sure the page is supported for this dev type */
6487                         if (lun->be_lun->lun_type == T_DIRECT &&
6488                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6489                                 continue;
6490                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6491                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6492                                 continue;
6493                         if (lun->be_lun->lun_type == T_CDROM &&
6494                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6495                                 continue;
6496
6497                         /* Look for the right page code */
6498                         if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6499                                 continue;
6500
6501                         /* Look for the right subpage or the subpage wildcard*/
6502                         if ((page_index->subpage != subpage)
6503                          && (subpage != SMS_SUBPAGE_ALL))
6504                                 continue;
6505
6506 #if 0
6507                         printf("found page %#x len %d\n",
6508                                page_index->page_code & SMPH_PC_MASK,
6509                                page_index->page_len);
6510 #endif
6511
6512                         page_len += page_index->page_len;
6513                 }
6514
6515                 if (page_len == 0) {
6516                         ctl_set_invalid_field(ctsio,
6517                                               /*sks_valid*/ 1,
6518                                               /*command*/ 1,
6519                                               /*field*/ 2,
6520                                               /*bit_valid*/ 1,
6521                                               /*bit*/ 5);
6522                         ctl_done((union ctl_io *)ctsio);
6523                         return (CTL_RETVAL_COMPLETE);
6524                 }
6525                 break;
6526         }
6527         }
6528
6529         total_len = header_len + page_len;
6530 #if 0
6531         printf("header_len = %d, page_len = %d, total_len = %d\n",
6532                header_len, page_len, total_len);
6533 #endif
6534
6535         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6536         ctsio->kern_sg_entries = 0;
6537         ctsio->kern_rel_offset = 0;
6538         ctsio->kern_data_len = min(total_len, alloc_len);
6539         ctsio->kern_total_len = ctsio->kern_data_len;
6540
6541         switch (ctsio->cdb[0]) {
6542         case MODE_SENSE_6: {
6543                 struct scsi_mode_hdr_6 *header;
6544
6545                 header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
6546
6547                 header->datalen = MIN(total_len - 1, 254);
6548                 if (lun->be_lun->lun_type == T_DIRECT) {
6549                         header->dev_specific = 0x10; /* DPOFUA */
6550                         if ((lun->be_lun->flags & CTL_LUN_FLAG_READONLY) ||
6551                             (lun->MODE_CTRL.eca_and_aen & SCP_SWP) != 0)
6552                                 header->dev_specific |= 0x80; /* WP */
6553                 }
6554                 if (dbd)
6555                         header->block_descr_len = 0;
6556                 else
6557                         header->block_descr_len =
6558                                 sizeof(struct scsi_mode_block_descr);
6559                 block_desc = (struct scsi_mode_block_descr *)&header[1];
6560                 break;
6561         }
6562         case MODE_SENSE_10: {
6563                 struct scsi_mode_hdr_10 *header;
6564                 int datalen;
6565
6566                 header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
6567
6568                 datalen = MIN(total_len - 2, 65533);
6569                 scsi_ulto2b(datalen, header->datalen);
6570                 if (lun->be_lun->lun_type == T_DIRECT) {
6571                         header->dev_specific = 0x10; /* DPOFUA */
6572                         if ((lun->be_lun->flags & CTL_LUN_FLAG_READONLY) ||
6573                             (lun->MODE_CTRL.eca_and_aen & SCP_SWP) != 0)
6574                                 header->dev_specific |= 0x80; /* WP */
6575                 }
6576                 if (dbd)
6577                         scsi_ulto2b(0, header->block_descr_len);
6578                 else
6579                         scsi_ulto2b(sizeof(struct scsi_mode_block_descr),
6580                                     header->block_descr_len);
6581                 block_desc = (struct scsi_mode_block_descr *)&header[1];
6582                 break;
6583         }
6584         default:
6585                 panic("%s: Invalid CDB type %#x", __func__, ctsio->cdb[0]);
6586         }
6587
6588         /*
6589          * If we've got a disk, use its blocksize in the block
6590          * descriptor.  Otherwise, just set it to 0.
6591          */
6592         if (dbd == 0) {
6593                 if (lun->be_lun->lun_type == T_DIRECT)
6594                         scsi_ulto3b(lun->be_lun->blocksize,
6595                                     block_desc->block_len);
6596                 else
6597                         scsi_ulto3b(0, block_desc->block_len);
6598         }
6599
6600         switch (page_code) {
6601         case SMS_ALL_PAGES_PAGE: {
6602                 int i, data_used;
6603
6604                 data_used = header_len;
6605                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6606                         struct ctl_page_index *page_index;
6607
6608                         page_index = &lun->mode_pages.index[i];
6609                         if (lun->be_lun->lun_type == T_DIRECT &&
6610                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6611                                 continue;
6612                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6613                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6614                                 continue;
6615                         if (lun->be_lun->lun_type == T_CDROM &&
6616                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6617                                 continue;
6618
6619                         /*
6620                          * We don't use this subpage if the user didn't
6621                          * request all subpages.  We already checked (above)
6622                          * to make sure the user only specified a subpage
6623                          * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case.
6624                          */
6625                         if ((page_index->subpage != 0)
6626                          && (subpage == SMS_SUBPAGE_PAGE_0))
6627                                 continue;
6628
6629                         /*
6630                          * Call the handler, if it exists, to update the
6631                          * page to the latest values.
6632                          */
6633                         if (page_index->sense_handler != NULL)
6634                                 page_index->sense_handler(ctsio, page_index,pc);
6635
6636                         memcpy(ctsio->kern_data_ptr + data_used,
6637                                page_index->page_data +
6638                                (page_index->page_len * pc),
6639                                page_index->page_len);
6640                         data_used += page_index->page_len;
6641                 }
6642                 break;
6643         }
6644         default: {
6645                 int i, data_used;
6646
6647                 data_used = header_len;
6648
6649                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6650                         struct ctl_page_index *page_index;
6651
6652                         page_index = &lun->mode_pages.index[i];
6653
6654                         /* Look for the right page code */
6655                         if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6656                                 continue;
6657
6658                         /* Look for the right subpage or the subpage wildcard*/
6659                         if ((page_index->subpage != subpage)
6660                          && (subpage != SMS_SUBPAGE_ALL))
6661                                 continue;
6662
6663                         /* Make sure the page is supported for this dev type */
6664                         if (lun->be_lun->lun_type == T_DIRECT &&
6665                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6666                                 continue;
6667                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6668                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6669                                 continue;
6670                         if (lun->be_lun->lun_type == T_CDROM &&
6671                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6672                                 continue;
6673
6674                         /*
6675                          * Call the handler, if it exists, to update the
6676                          * page to the latest values.
6677                          */
6678                         if (page_index->sense_handler != NULL)
6679                                 page_index->sense_handler(ctsio, page_index,pc);
6680
6681                         memcpy(ctsio->kern_data_ptr + data_used,
6682                                page_index->page_data +
6683                                (page_index->page_len * pc),
6684                                page_index->page_len);
6685                         data_used += page_index->page_len;
6686                 }
6687                 break;
6688         }
6689         }
6690
6691         ctl_set_success(ctsio);
6692         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6693         ctsio->be_move_done = ctl_config_move_done;
6694         ctl_datamove((union ctl_io *)ctsio);
6695         return (CTL_RETVAL_COMPLETE);
6696 }
6697
6698 int
6699 ctl_lbp_log_sense_handler(struct ctl_scsiio *ctsio,
6700                                struct ctl_page_index *page_index,
6701                                int pc)
6702 {
6703         struct ctl_lun *lun = CTL_LUN(ctsio);
6704         struct scsi_log_param_header *phdr;
6705         uint8_t *data;
6706         uint64_t val;
6707
6708         data = page_index->page_data;
6709
6710         if (lun->backend->lun_attr != NULL &&
6711             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "blocksavail"))
6712              != UINT64_MAX) {
6713                 phdr = (struct scsi_log_param_header *)data;
6714                 scsi_ulto2b(0x0001, 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, "blocksused"))
6725              != UINT64_MAX) {
6726                 phdr = (struct scsi_log_param_header *)data;
6727                 scsi_ulto2b(0x0002, 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] = 0x01; /* per-LUN */
6733                 data += phdr->param_len;
6734         }
6735
6736         if (lun->backend->lun_attr != NULL &&
6737             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "poolblocksavail"))
6738              != UINT64_MAX) {
6739                 phdr = (struct scsi_log_param_header *)data;
6740                 scsi_ulto2b(0x00f1, phdr->param_code);
6741                 phdr->param_control = SLP_LBIN | SLP_LP;
6742                 phdr->param_len = 8;
6743                 data = (uint8_t *)(phdr + 1);
6744                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6745                 data[4] = 0x02; /* per-pool */
6746                 data += phdr->param_len;
6747         }
6748
6749         if (lun->backend->lun_attr != NULL &&
6750             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "poolblocksused"))
6751              != UINT64_MAX) {
6752                 phdr = (struct scsi_log_param_header *)data;
6753                 scsi_ulto2b(0x00f2, phdr->param_code);
6754                 phdr->param_control = SLP_LBIN | SLP_LP;
6755                 phdr->param_len = 8;
6756                 data = (uint8_t *)(phdr + 1);
6757                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6758                 data[4] = 0x02; /* per-pool */
6759                 data += phdr->param_len;
6760         }
6761
6762         page_index->page_len = data - page_index->page_data;
6763         return (0);
6764 }
6765
6766 int
6767 ctl_sap_log_sense_handler(struct ctl_scsiio *ctsio,
6768                                struct ctl_page_index *page_index,
6769                                int pc)
6770 {
6771         struct ctl_lun *lun = CTL_LUN(ctsio);
6772         struct stat_page *data;
6773         struct bintime *t;
6774
6775         data = (struct stat_page *)page_index->page_data;
6776
6777         scsi_ulto2b(SLP_SAP, data->sap.hdr.param_code);
6778         data->sap.hdr.param_control = SLP_LBIN;
6779         data->sap.hdr.param_len = sizeof(struct scsi_log_stat_and_perf) -
6780             sizeof(struct scsi_log_param_header);
6781         scsi_u64to8b(lun->stats.operations[CTL_STATS_READ],
6782             data->sap.read_num);
6783         scsi_u64to8b(lun->stats.operations[CTL_STATS_WRITE],
6784             data->sap.write_num);
6785         if (lun->be_lun->blocksize > 0) {
6786                 scsi_u64to8b(lun->stats.bytes[CTL_STATS_WRITE] /
6787                     lun->be_lun->blocksize, data->sap.recvieved_lba);
6788                 scsi_u64to8b(lun->stats.bytes[CTL_STATS_READ] /
6789                     lun->be_lun->blocksize, data->sap.transmitted_lba);
6790         }
6791         t = &lun->stats.time[CTL_STATS_READ];
6792         scsi_u64to8b((uint64_t)t->sec * 1000 + t->frac / (UINT64_MAX / 1000),
6793             data->sap.read_int);
6794         t = &lun->stats.time[CTL_STATS_WRITE];
6795         scsi_u64to8b((uint64_t)t->sec * 1000 + t->frac / (UINT64_MAX / 1000),
6796             data->sap.write_int);
6797         scsi_u64to8b(0, data->sap.weighted_num);
6798         scsi_u64to8b(0, data->sap.weighted_int);
6799         scsi_ulto2b(SLP_IT, data->it.hdr.param_code);
6800         data->it.hdr.param_control = SLP_LBIN;
6801         data->it.hdr.param_len = sizeof(struct scsi_log_idle_time) -
6802             sizeof(struct scsi_log_param_header);
6803 #ifdef CTL_TIME_IO
6804         scsi_u64to8b(lun->idle_time / SBT_1MS, data->it.idle_int);
6805 #endif
6806         scsi_ulto2b(SLP_TI, data->ti.hdr.param_code);
6807         data->it.hdr.param_control = SLP_LBIN;
6808         data->ti.hdr.param_len = sizeof(struct scsi_log_time_interval) -
6809             sizeof(struct scsi_log_param_header);
6810         scsi_ulto4b(3, data->ti.exponent);
6811         scsi_ulto4b(1, data->ti.integer);
6812         return (0);
6813 }
6814
6815 int
6816 ctl_ie_log_sense_handler(struct ctl_scsiio *ctsio,
6817                                struct ctl_page_index *page_index,
6818                                int pc)
6819 {
6820         struct ctl_lun *lun = CTL_LUN(ctsio);
6821         struct scsi_log_informational_exceptions *data;
6822
6823         data = (struct scsi_log_informational_exceptions *)page_index->page_data;
6824
6825         scsi_ulto2b(SLP_IE_GEN, data->hdr.param_code);
6826         data->hdr.param_control = SLP_LBIN;
6827         data->hdr.param_len = sizeof(struct scsi_log_informational_exceptions) -
6828             sizeof(struct scsi_log_param_header);
6829         data->ie_asc = lun->ie_asc;
6830         data->ie_ascq = lun->ie_ascq;
6831         data->temperature = 0xff;
6832         return (0);
6833 }
6834
6835 int
6836 ctl_log_sense(struct ctl_scsiio *ctsio)
6837 {
6838         struct ctl_lun *lun = CTL_LUN(ctsio);
6839         int i, pc, page_code, subpage;
6840         int alloc_len, total_len;
6841         struct ctl_page_index *page_index;
6842         struct scsi_log_sense *cdb;
6843         struct scsi_log_header *header;
6844
6845         CTL_DEBUG_PRINT(("ctl_log_sense\n"));
6846
6847         cdb = (struct scsi_log_sense *)ctsio->cdb;
6848         pc = (cdb->page & SLS_PAGE_CTRL_MASK) >> 6;
6849         page_code = cdb->page & SLS_PAGE_CODE;
6850         subpage = cdb->subpage;
6851         alloc_len = scsi_2btoul(cdb->length);
6852
6853         page_index = NULL;
6854         for (i = 0; i < CTL_NUM_LOG_PAGES; i++) {
6855                 page_index = &lun->log_pages.index[i];
6856
6857                 /* Look for the right page code */
6858                 if ((page_index->page_code & SL_PAGE_CODE) != page_code)
6859                         continue;
6860
6861                 /* Look for the right subpage or the subpage wildcard*/
6862                 if (page_index->subpage != subpage)
6863                         continue;
6864
6865                 break;
6866         }
6867         if (i >= CTL_NUM_LOG_PAGES) {
6868                 ctl_set_invalid_field(ctsio,
6869                                       /*sks_valid*/ 1,
6870                                       /*command*/ 1,
6871                                       /*field*/ 2,
6872                                       /*bit_valid*/ 0,
6873                                       /*bit*/ 0);
6874                 ctl_done((union ctl_io *)ctsio);
6875                 return (CTL_RETVAL_COMPLETE);
6876         }
6877
6878         total_len = sizeof(struct scsi_log_header) + page_index->page_len;
6879
6880         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6881         ctsio->kern_sg_entries = 0;
6882         ctsio->kern_rel_offset = 0;
6883         ctsio->kern_data_len = min(total_len, alloc_len);
6884         ctsio->kern_total_len = ctsio->kern_data_len;
6885
6886         header = (struct scsi_log_header *)ctsio->kern_data_ptr;
6887         header->page = page_index->page_code;
6888         if (page_index->page_code == SLS_LOGICAL_BLOCK_PROVISIONING)
6889                 header->page |= SL_DS;
6890         if (page_index->subpage) {
6891                 header->page |= SL_SPF;
6892                 header->subpage = page_index->subpage;
6893         }
6894         scsi_ulto2b(page_index->page_len, header->datalen);
6895
6896         /*
6897          * Call the handler, if it exists, to update the
6898          * page to the latest values.
6899          */
6900         if (page_index->sense_handler != NULL)
6901                 page_index->sense_handler(ctsio, page_index, pc);
6902
6903         memcpy(header + 1, page_index->page_data, page_index->page_len);
6904
6905         ctl_set_success(ctsio);
6906         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6907         ctsio->be_move_done = ctl_config_move_done;
6908         ctl_datamove((union ctl_io *)ctsio);
6909         return (CTL_RETVAL_COMPLETE);
6910 }
6911
6912 int
6913 ctl_read_capacity(struct ctl_scsiio *ctsio)
6914 {
6915         struct ctl_lun *lun = CTL_LUN(ctsio);
6916         struct scsi_read_capacity *cdb;
6917         struct scsi_read_capacity_data *data;
6918         uint32_t lba;
6919
6920         CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
6921
6922         cdb = (struct scsi_read_capacity *)ctsio->cdb;
6923
6924         lba = scsi_4btoul(cdb->addr);
6925         if (((cdb->pmi & SRC_PMI) == 0)
6926          && (lba != 0)) {
6927                 ctl_set_invalid_field(/*ctsio*/ ctsio,
6928                                       /*sks_valid*/ 1,
6929                                       /*command*/ 1,
6930                                       /*field*/ 2,
6931                                       /*bit_valid*/ 0,
6932                                       /*bit*/ 0);
6933                 ctl_done((union ctl_io *)ctsio);
6934                 return (CTL_RETVAL_COMPLETE);
6935         }
6936
6937         ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
6938         data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
6939         ctsio->kern_data_len = sizeof(*data);
6940         ctsio->kern_total_len = sizeof(*data);
6941         ctsio->kern_rel_offset = 0;
6942         ctsio->kern_sg_entries = 0;
6943
6944         /*
6945          * If the maximum LBA is greater than 0xfffffffe, the user must
6946          * issue a SERVICE ACTION IN (16) command, with the read capacity
6947          * serivce action set.
6948          */
6949         if (lun->be_lun->maxlba > 0xfffffffe)
6950                 scsi_ulto4b(0xffffffff, data->addr);
6951         else
6952                 scsi_ulto4b(lun->be_lun->maxlba, data->addr);
6953
6954         /*
6955          * XXX KDM this may not be 512 bytes...
6956          */
6957         scsi_ulto4b(lun->be_lun->blocksize, data->length);
6958
6959         ctl_set_success(ctsio);
6960         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6961         ctsio->be_move_done = ctl_config_move_done;
6962         ctl_datamove((union ctl_io *)ctsio);
6963         return (CTL_RETVAL_COMPLETE);
6964 }
6965
6966 int
6967 ctl_read_capacity_16(struct ctl_scsiio *ctsio)
6968 {
6969         struct ctl_lun *lun = CTL_LUN(ctsio);
6970         struct scsi_read_capacity_16 *cdb;
6971         struct scsi_read_capacity_data_long *data;
6972         uint64_t lba;
6973         uint32_t alloc_len;
6974
6975         CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
6976
6977         cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
6978
6979         alloc_len = scsi_4btoul(cdb->alloc_len);
6980         lba = scsi_8btou64(cdb->addr);
6981
6982         if ((cdb->reladr & SRC16_PMI)
6983          && (lba != 0)) {
6984                 ctl_set_invalid_field(/*ctsio*/ ctsio,
6985                                       /*sks_valid*/ 1,
6986                                       /*command*/ 1,
6987                                       /*field*/ 2,
6988                                       /*bit_valid*/ 0,
6989                                       /*bit*/ 0);
6990                 ctl_done((union ctl_io *)ctsio);
6991                 return (CTL_RETVAL_COMPLETE);
6992         }
6993
6994         ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
6995         data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
6996         ctsio->kern_rel_offset = 0;
6997         ctsio->kern_sg_entries = 0;
6998         ctsio->kern_data_len = min(sizeof(*data), alloc_len);
6999         ctsio->kern_total_len = ctsio->kern_data_len;
7000
7001         scsi_u64to8b(lun->be_lun->maxlba, data->addr);
7002         /* XXX KDM this may not be 512 bytes... */
7003         scsi_ulto4b(lun->be_lun->blocksize, data->length);
7004         data->prot_lbppbe = lun->be_lun->pblockexp & SRC16_LBPPBE;
7005         scsi_ulto2b(lun->be_lun->pblockoff & SRC16_LALBA_A, data->lalba_lbp);
7006         if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
7007                 data->lalba_lbp[0] |= SRC16_LBPME | SRC16_LBPRZ;
7008
7009         ctl_set_success(ctsio);
7010         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7011         ctsio->be_move_done = ctl_config_move_done;
7012         ctl_datamove((union ctl_io *)ctsio);
7013         return (CTL_RETVAL_COMPLETE);
7014 }
7015
7016 int
7017 ctl_get_lba_status(struct ctl_scsiio *ctsio)
7018 {
7019         struct ctl_lun *lun = CTL_LUN(ctsio);
7020         struct scsi_get_lba_status *cdb;
7021         struct scsi_get_lba_status_data *data;
7022         struct ctl_lba_len_flags *lbalen;
7023         uint64_t lba;
7024         uint32_t alloc_len, total_len;
7025         int retval;
7026
7027         CTL_DEBUG_PRINT(("ctl_get_lba_status\n"));
7028
7029         cdb = (struct scsi_get_lba_status *)ctsio->cdb;
7030         lba = scsi_8btou64(cdb->addr);
7031         alloc_len = scsi_4btoul(cdb->alloc_len);
7032
7033         if (lba > lun->be_lun->maxlba) {
7034                 ctl_set_lba_out_of_range(ctsio, lba);
7035                 ctl_done((union ctl_io *)ctsio);
7036                 return (CTL_RETVAL_COMPLETE);
7037         }
7038
7039         total_len = sizeof(*data) + sizeof(data->descr[0]);
7040         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7041         data = (struct scsi_get_lba_status_data *)ctsio->kern_data_ptr;
7042         ctsio->kern_rel_offset = 0;
7043         ctsio->kern_sg_entries = 0;
7044         ctsio->kern_data_len = min(total_len, alloc_len);
7045         ctsio->kern_total_len = ctsio->kern_data_len;
7046
7047         /* Fill dummy data in case backend can't tell anything. */
7048         scsi_ulto4b(4 + sizeof(data->descr[0]), data->length);
7049         scsi_u64to8b(lba, data->descr[0].addr);
7050         scsi_ulto4b(MIN(UINT32_MAX, lun->be_lun->maxlba + 1 - lba),
7051             data->descr[0].length);
7052         data->descr[0].status = 0; /* Mapped or unknown. */
7053
7054         ctl_set_success(ctsio);
7055         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7056         ctsio->be_move_done = ctl_config_move_done;
7057
7058         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
7059         lbalen->lba = lba;
7060         lbalen->len = total_len;
7061         lbalen->flags = 0;
7062         retval = lun->backend->config_read((union ctl_io *)ctsio);
7063         return (retval);
7064 }
7065
7066 int
7067 ctl_read_defect(struct ctl_scsiio *ctsio)
7068 {
7069         struct scsi_read_defect_data_10 *ccb10;
7070         struct scsi_read_defect_data_12 *ccb12;
7071         struct scsi_read_defect_data_hdr_10 *data10;
7072         struct scsi_read_defect_data_hdr_12 *data12;
7073         uint32_t alloc_len, data_len;
7074         uint8_t format;
7075
7076         CTL_DEBUG_PRINT(("ctl_read_defect\n"));
7077
7078         if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
7079                 ccb10 = (struct scsi_read_defect_data_10 *)&ctsio->cdb;
7080                 format = ccb10->format;
7081                 alloc_len = scsi_2btoul(ccb10->alloc_length);
7082                 data_len = sizeof(*data10);
7083         } else {
7084                 ccb12 = (struct scsi_read_defect_data_12 *)&ctsio->cdb;
7085                 format = ccb12->format;
7086                 alloc_len = scsi_4btoul(ccb12->alloc_length);
7087                 data_len = sizeof(*data12);
7088         }
7089         if (alloc_len == 0) {
7090                 ctl_set_success(ctsio);
7091                 ctl_done((union ctl_io *)ctsio);
7092                 return (CTL_RETVAL_COMPLETE);
7093         }
7094
7095         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
7096         ctsio->kern_rel_offset = 0;
7097         ctsio->kern_sg_entries = 0;
7098         ctsio->kern_data_len = min(data_len, alloc_len);
7099         ctsio->kern_total_len = ctsio->kern_data_len;
7100
7101         if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
7102                 data10 = (struct scsi_read_defect_data_hdr_10 *)
7103                     ctsio->kern_data_ptr;
7104                 data10->format = format;
7105                 scsi_ulto2b(0, data10->length);
7106         } else {
7107                 data12 = (struct scsi_read_defect_data_hdr_12 *)
7108                     ctsio->kern_data_ptr;
7109                 data12->format = format;
7110                 scsi_ulto2b(0, data12->generation);
7111                 scsi_ulto4b(0, data12->length);
7112         }
7113
7114         ctl_set_success(ctsio);
7115         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7116         ctsio->be_move_done = ctl_config_move_done;
7117         ctl_datamove((union ctl_io *)ctsio);
7118         return (CTL_RETVAL_COMPLETE);
7119 }
7120
7121 int
7122 ctl_report_tagret_port_groups(struct ctl_scsiio *ctsio)
7123 {
7124         struct ctl_softc *softc = CTL_SOFTC(ctsio);
7125         struct ctl_lun *lun = CTL_LUN(ctsio);
7126         struct scsi_maintenance_in *cdb;
7127         int retval;
7128         int alloc_len, ext, total_len = 0, g, pc, pg, ts, os;
7129         int num_ha_groups, num_target_ports, shared_group;
7130         struct ctl_port *port;
7131         struct scsi_target_group_data *rtg_ptr;
7132         struct scsi_target_group_data_extended *rtg_ext_ptr;
7133         struct scsi_target_port_group_descriptor *tpg_desc;
7134
7135         CTL_DEBUG_PRINT(("ctl_report_tagret_port_groups\n"));
7136
7137         cdb = (struct scsi_maintenance_in *)ctsio->cdb;
7138         retval = CTL_RETVAL_COMPLETE;
7139
7140         switch (cdb->byte2 & STG_PDF_MASK) {
7141         case STG_PDF_LENGTH:
7142                 ext = 0;
7143                 break;
7144         case STG_PDF_EXTENDED:
7145                 ext = 1;
7146                 break;
7147         default:
7148                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7149                                       /*sks_valid*/ 1,
7150                                       /*command*/ 1,
7151                                       /*field*/ 2,
7152                                       /*bit_valid*/ 1,
7153                                       /*bit*/ 5);
7154                 ctl_done((union ctl_io *)ctsio);
7155                 return(retval);
7156         }
7157
7158         num_target_ports = 0;
7159         shared_group = (softc->is_single != 0);
7160         mtx_lock(&softc->ctl_lock);
7161         STAILQ_FOREACH(port, &softc->port_list, links) {
7162                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7163                         continue;
7164                 if (ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
7165                         continue;
7166                 num_target_ports++;
7167                 if (port->status & CTL_PORT_STATUS_HA_SHARED)
7168                         shared_group = 1;
7169         }
7170         mtx_unlock(&softc->ctl_lock);
7171         num_ha_groups = (softc->is_single) ? 0 : NUM_HA_SHELVES;
7172
7173         if (ext)
7174                 total_len = sizeof(struct scsi_target_group_data_extended);
7175         else
7176                 total_len = sizeof(struct scsi_target_group_data);
7177         total_len += sizeof(struct scsi_target_port_group_descriptor) *
7178                 (shared_group + num_ha_groups) +
7179             sizeof(struct scsi_target_port_descriptor) * num_target_ports;
7180
7181         alloc_len = scsi_4btoul(cdb->length);
7182
7183         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7184         ctsio->kern_sg_entries = 0;
7185         ctsio->kern_rel_offset = 0;
7186         ctsio->kern_data_len = min(total_len, alloc_len);
7187         ctsio->kern_total_len = ctsio->kern_data_len;
7188
7189         if (ext) {
7190                 rtg_ext_ptr = (struct scsi_target_group_data_extended *)
7191                     ctsio->kern_data_ptr;
7192                 scsi_ulto4b(total_len - 4, rtg_ext_ptr->length);
7193                 rtg_ext_ptr->format_type = 0x10;
7194                 rtg_ext_ptr->implicit_transition_time = 0;
7195                 tpg_desc = &rtg_ext_ptr->groups[0];
7196         } else {
7197                 rtg_ptr = (struct scsi_target_group_data *)
7198                     ctsio->kern_data_ptr;
7199                 scsi_ulto4b(total_len - 4, rtg_ptr->length);
7200                 tpg_desc = &rtg_ptr->groups[0];
7201         }
7202
7203         mtx_lock(&softc->ctl_lock);
7204         pg = softc->port_min / softc->port_cnt;
7205         if (lun->flags & (CTL_LUN_PRIMARY_SC | CTL_LUN_PEER_SC_PRIMARY)) {
7206                 /* Some shelf is known to be primary. */
7207                 if (softc->ha_link == CTL_HA_LINK_OFFLINE)
7208                         os = TPG_ASYMMETRIC_ACCESS_UNAVAILABLE;
7209                 else if (softc->ha_link == CTL_HA_LINK_UNKNOWN)
7210                         os = TPG_ASYMMETRIC_ACCESS_TRANSITIONING;
7211                 else if (softc->ha_mode == CTL_HA_MODE_ACT_STBY)
7212                         os = TPG_ASYMMETRIC_ACCESS_STANDBY;
7213                 else
7214                         os = TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7215                 if (lun->flags & CTL_LUN_PRIMARY_SC) {
7216                         ts = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7217                 } else {
7218                         ts = os;
7219                         os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7220                 }
7221         } else {
7222                 /* No known primary shelf. */
7223                 if (softc->ha_link == CTL_HA_LINK_OFFLINE) {
7224                         ts = TPG_ASYMMETRIC_ACCESS_UNAVAILABLE;
7225                         os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7226                 } else if (softc->ha_link == CTL_HA_LINK_UNKNOWN) {
7227                         ts = TPG_ASYMMETRIC_ACCESS_TRANSITIONING;
7228                         os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7229                 } else {
7230                         ts = os = TPG_ASYMMETRIC_ACCESS_TRANSITIONING;
7231                 }
7232         }
7233         if (shared_group) {
7234                 tpg_desc->pref_state = ts;
7235                 tpg_desc->support = TPG_AO_SUP | TPG_AN_SUP | TPG_S_SUP |
7236                     TPG_U_SUP | TPG_T_SUP;
7237                 scsi_ulto2b(1, tpg_desc->target_port_group);
7238                 tpg_desc->status = TPG_IMPLICIT;
7239                 pc = 0;
7240                 STAILQ_FOREACH(port, &softc->port_list, links) {
7241                         if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7242                                 continue;
7243                         if (!softc->is_single &&
7244                             (port->status & CTL_PORT_STATUS_HA_SHARED) == 0)
7245                                 continue;
7246                         if (ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
7247                                 continue;
7248                         scsi_ulto2b(port->targ_port, tpg_desc->descriptors[pc].
7249                             relative_target_port_identifier);
7250                         pc++;
7251                 }
7252                 tpg_desc->target_port_count = pc;
7253                 tpg_desc = (struct scsi_target_port_group_descriptor *)
7254                     &tpg_desc->descriptors[pc];
7255         }
7256         for (g = 0; g < num_ha_groups; g++) {
7257                 tpg_desc->pref_state = (g == pg) ? ts : os;
7258                 tpg_desc->support = TPG_AO_SUP | TPG_AN_SUP | TPG_S_SUP |
7259                     TPG_U_SUP | TPG_T_SUP;
7260                 scsi_ulto2b(2 + g, tpg_desc->target_port_group);
7261                 tpg_desc->status = TPG_IMPLICIT;
7262                 pc = 0;
7263                 STAILQ_FOREACH(port, &softc->port_list, links) {
7264                         if (port->targ_port < g * softc->port_cnt ||
7265                             port->targ_port >= (g + 1) * softc->port_cnt)
7266                                 continue;
7267                         if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7268                                 continue;
7269                         if (port->status & CTL_PORT_STATUS_HA_SHARED)
7270                                 continue;
7271                         if (ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
7272                                 continue;
7273                         scsi_ulto2b(port->targ_port, tpg_desc->descriptors[pc].
7274                             relative_target_port_identifier);
7275                         pc++;
7276                 }
7277                 tpg_desc->target_port_count = pc;
7278                 tpg_desc = (struct scsi_target_port_group_descriptor *)
7279                     &tpg_desc->descriptors[pc];
7280         }
7281         mtx_unlock(&softc->ctl_lock);
7282
7283         ctl_set_success(ctsio);
7284         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7285         ctsio->be_move_done = ctl_config_move_done;
7286         ctl_datamove((union ctl_io *)ctsio);
7287         return(retval);
7288 }
7289
7290 int
7291 ctl_report_supported_opcodes(struct ctl_scsiio *ctsio)
7292 {
7293         struct ctl_lun *lun = CTL_LUN(ctsio);
7294         struct scsi_report_supported_opcodes *cdb;
7295         const struct ctl_cmd_entry *entry, *sentry;
7296         struct scsi_report_supported_opcodes_all *all;
7297         struct scsi_report_supported_opcodes_descr *descr;
7298         struct scsi_report_supported_opcodes_one *one;
7299         int retval;
7300         int alloc_len, total_len;
7301         int opcode, service_action, i, j, num;
7302
7303         CTL_DEBUG_PRINT(("ctl_report_supported_opcodes\n"));
7304
7305         cdb = (struct scsi_report_supported_opcodes *)ctsio->cdb;
7306         retval = CTL_RETVAL_COMPLETE;
7307
7308         opcode = cdb->requested_opcode;
7309         service_action = scsi_2btoul(cdb->requested_service_action);
7310         switch (cdb->options & RSO_OPTIONS_MASK) {
7311         case RSO_OPTIONS_ALL:
7312                 num = 0;
7313                 for (i = 0; i < 256; i++) {
7314                         entry = &ctl_cmd_table[i];
7315                         if (entry->flags & CTL_CMD_FLAG_SA5) {
7316                                 for (j = 0; j < 32; j++) {
7317                                         sentry = &((const struct ctl_cmd_entry *)
7318                                             entry->execute)[j];
7319                                         if (ctl_cmd_applicable(
7320                                             lun->be_lun->lun_type, sentry))
7321                                                 num++;
7322                                 }
7323                         } else {
7324                                 if (ctl_cmd_applicable(lun->be_lun->lun_type,
7325                                     entry))
7326                                         num++;
7327                         }
7328                 }
7329                 total_len = sizeof(struct scsi_report_supported_opcodes_all) +
7330                     num * sizeof(struct scsi_report_supported_opcodes_descr);
7331                 break;
7332         case RSO_OPTIONS_OC:
7333                 if (ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) {
7334                         ctl_set_invalid_field(/*ctsio*/ ctsio,
7335                                               /*sks_valid*/ 1,
7336                                               /*command*/ 1,
7337                                               /*field*/ 2,
7338                                               /*bit_valid*/ 1,
7339                                               /*bit*/ 2);
7340                         ctl_done((union ctl_io *)ctsio);
7341                         return (CTL_RETVAL_COMPLETE);
7342                 }
7343                 total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7344                 break;
7345         case RSO_OPTIONS_OC_SA:
7346                 if ((ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) == 0 ||
7347                     service_action >= 32) {
7348                         ctl_set_invalid_field(/*ctsio*/ ctsio,
7349                                               /*sks_valid*/ 1,
7350                                               /*command*/ 1,
7351                                               /*field*/ 2,
7352                                               /*bit_valid*/ 1,
7353                                               /*bit*/ 2);
7354                         ctl_done((union ctl_io *)ctsio);
7355                         return (CTL_RETVAL_COMPLETE);
7356                 }
7357                 /* FALLTHROUGH */
7358         case RSO_OPTIONS_OC_ASA:
7359                 total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7360                 break;
7361         default:
7362                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7363                                       /*sks_valid*/ 1,
7364                                       /*command*/ 1,
7365                                       /*field*/ 2,
7366                                       /*bit_valid*/ 1,
7367                                       /*bit*/ 2);
7368                 ctl_done((union ctl_io *)ctsio);
7369                 return (CTL_RETVAL_COMPLETE);
7370         }
7371
7372         alloc_len = scsi_4btoul(cdb->length);
7373
7374         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7375         ctsio->kern_sg_entries = 0;
7376         ctsio->kern_rel_offset = 0;
7377         ctsio->kern_data_len = min(total_len, alloc_len);
7378         ctsio->kern_total_len = ctsio->kern_data_len;
7379
7380         switch (cdb->options & RSO_OPTIONS_MASK) {
7381         case RSO_OPTIONS_ALL:
7382                 all = (struct scsi_report_supported_opcodes_all *)
7383                     ctsio->kern_data_ptr;
7384                 num = 0;
7385                 for (i = 0; i < 256; i++) {
7386                         entry = &ctl_cmd_table[i];
7387                         if (entry->flags & CTL_CMD_FLAG_SA5) {
7388                                 for (j = 0; j < 32; j++) {
7389                                         sentry = &((const struct ctl_cmd_entry *)
7390                                             entry->execute)[j];
7391                                         if (!ctl_cmd_applicable(
7392                                             lun->be_lun->lun_type, sentry))
7393                                                 continue;
7394                                         descr = &all->descr[num++];
7395                                         descr->opcode = i;
7396                                         scsi_ulto2b(j, descr->service_action);
7397                                         descr->flags = RSO_SERVACTV;
7398                                         scsi_ulto2b(sentry->length,
7399                                             descr->cdb_length);
7400                                 }
7401                         } else {
7402                                 if (!ctl_cmd_applicable(lun->be_lun->lun_type,
7403                                     entry))
7404                                         continue;
7405                                 descr = &all->descr[num++];
7406                                 descr->opcode = i;
7407                                 scsi_ulto2b(0, descr->service_action);
7408                                 descr->flags = 0;
7409                                 scsi_ulto2b(entry->length, descr->cdb_length);
7410                         }
7411                 }
7412                 scsi_ulto4b(
7413                     num * sizeof(struct scsi_report_supported_opcodes_descr),
7414                     all->length);
7415                 break;
7416         case RSO_OPTIONS_OC:
7417                 one = (struct scsi_report_supported_opcodes_one *)
7418                     ctsio->kern_data_ptr;
7419                 entry = &ctl_cmd_table[opcode];
7420                 goto fill_one;
7421         case RSO_OPTIONS_OC_SA:
7422                 one = (struct scsi_report_supported_opcodes_one *)
7423                     ctsio->kern_data_ptr;
7424                 entry = &ctl_cmd_table[opcode];
7425                 entry = &((const struct ctl_cmd_entry *)
7426                     entry->execute)[service_action];
7427 fill_one:
7428                 if (ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
7429                         one->support = 3;
7430                         scsi_ulto2b(entry->length, one->cdb_length);
7431                         one->cdb_usage[0] = opcode;
7432                         memcpy(&one->cdb_usage[1], entry->usage,
7433                             entry->length - 1);
7434                 } else
7435                         one->support = 1;
7436                 break;
7437         case RSO_OPTIONS_OC_ASA:
7438                 one = (struct scsi_report_supported_opcodes_one *)
7439                     ctsio->kern_data_ptr;
7440                 entry = &ctl_cmd_table[opcode];
7441                 if (entry->flags & CTL_CMD_FLAG_SA5) {
7442                         entry = &((const struct ctl_cmd_entry *)
7443                             entry->execute)[service_action];
7444                 } else if (service_action != 0) {
7445                         one->support = 1;
7446                         break;
7447                 }
7448                 goto fill_one;
7449         }
7450
7451         ctl_set_success(ctsio);
7452         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7453         ctsio->be_move_done = ctl_config_move_done;
7454         ctl_datamove((union ctl_io *)ctsio);
7455         return(retval);
7456 }
7457
7458 int
7459 ctl_report_supported_tmf(struct ctl_scsiio *ctsio)
7460 {
7461         struct scsi_report_supported_tmf *cdb;
7462         struct scsi_report_supported_tmf_ext_data *data;
7463         int retval;
7464         int alloc_len, total_len;
7465
7466         CTL_DEBUG_PRINT(("ctl_report_supported_tmf\n"));
7467
7468         cdb = (struct scsi_report_supported_tmf *)ctsio->cdb;
7469
7470         retval = CTL_RETVAL_COMPLETE;
7471
7472         if (cdb->options & RST_REPD)
7473                 total_len = sizeof(struct scsi_report_supported_tmf_ext_data);
7474         else
7475                 total_len = sizeof(struct scsi_report_supported_tmf_data);
7476         alloc_len = scsi_4btoul(cdb->length);
7477
7478         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7479         ctsio->kern_sg_entries = 0;
7480         ctsio->kern_rel_offset = 0;
7481         ctsio->kern_data_len = min(total_len, alloc_len);
7482         ctsio->kern_total_len = ctsio->kern_data_len;
7483
7484         data = (struct scsi_report_supported_tmf_ext_data *)ctsio->kern_data_ptr;
7485         data->byte1 |= RST_ATS | RST_ATSS | RST_CTSS | RST_LURS | RST_QTS |
7486             RST_TRS;
7487         data->byte2 |= RST_QAES | RST_QTSS | RST_ITNRS;
7488         data->length = total_len - 4;
7489
7490         ctl_set_success(ctsio);
7491         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7492         ctsio->be_move_done = ctl_config_move_done;
7493         ctl_datamove((union ctl_io *)ctsio);
7494         return (retval);
7495 }
7496
7497 int
7498 ctl_report_timestamp(struct ctl_scsiio *ctsio)
7499 {
7500         struct scsi_report_timestamp *cdb;
7501         struct scsi_report_timestamp_data *data;
7502         struct timeval tv;
7503         int64_t timestamp;
7504         int retval;
7505         int alloc_len, total_len;
7506
7507         CTL_DEBUG_PRINT(("ctl_report_timestamp\n"));
7508
7509         cdb = (struct scsi_report_timestamp *)ctsio->cdb;
7510
7511         retval = CTL_RETVAL_COMPLETE;
7512
7513         total_len = sizeof(struct scsi_report_timestamp_data);
7514         alloc_len = scsi_4btoul(cdb->length);
7515
7516         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7517         ctsio->kern_sg_entries = 0;
7518         ctsio->kern_rel_offset = 0;
7519         ctsio->kern_data_len = min(total_len, alloc_len);
7520         ctsio->kern_total_len = ctsio->kern_data_len;
7521
7522         data = (struct scsi_report_timestamp_data *)ctsio->kern_data_ptr;
7523         scsi_ulto2b(sizeof(*data) - 2, data->length);
7524         data->origin = RTS_ORIG_OUTSIDE;
7525         getmicrotime(&tv);
7526         timestamp = (int64_t)tv.tv_sec * 1000 + tv.tv_usec / 1000;
7527         scsi_ulto4b(timestamp >> 16, data->timestamp);
7528         scsi_ulto2b(timestamp & 0xffff, &data->timestamp[4]);
7529
7530         ctl_set_success(ctsio);
7531         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7532         ctsio->be_move_done = ctl_config_move_done;
7533         ctl_datamove((union ctl_io *)ctsio);
7534         return (retval);
7535 }
7536
7537 int
7538 ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7539 {
7540         struct ctl_softc *softc = CTL_SOFTC(ctsio);
7541         struct ctl_lun *lun = CTL_LUN(ctsio);
7542         struct scsi_per_res_in *cdb;
7543         int alloc_len, total_len = 0;
7544         /* struct scsi_per_res_in_rsrv in_data; */
7545         uint64_t key;
7546
7547         CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7548
7549         cdb = (struct scsi_per_res_in *)ctsio->cdb;
7550
7551         alloc_len = scsi_2btoul(cdb->length);
7552
7553 retry:
7554         mtx_lock(&lun->lun_lock);
7555         switch (cdb->action) {
7556         case SPRI_RK: /* read keys */
7557                 total_len = sizeof(struct scsi_per_res_in_keys) +
7558                         lun->pr_key_count *
7559                         sizeof(struct scsi_per_res_key);
7560                 break;
7561         case SPRI_RR: /* read reservation */
7562                 if (lun->flags & CTL_LUN_PR_RESERVED)
7563                         total_len = sizeof(struct scsi_per_res_in_rsrv);
7564                 else
7565                         total_len = sizeof(struct scsi_per_res_in_header);
7566                 break;
7567         case SPRI_RC: /* report capabilities */
7568                 total_len = sizeof(struct scsi_per_res_cap);
7569                 break;
7570         case SPRI_RS: /* read full status */
7571                 total_len = sizeof(struct scsi_per_res_in_header) +
7572                     (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7573                     lun->pr_key_count;
7574                 break;
7575         default:
7576                 panic("%s: Invalid PR type %#x", __func__, cdb->action);
7577         }
7578         mtx_unlock(&lun->lun_lock);
7579
7580         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7581         ctsio->kern_rel_offset = 0;
7582         ctsio->kern_sg_entries = 0;
7583         ctsio->kern_data_len = min(total_len, alloc_len);
7584         ctsio->kern_total_len = ctsio->kern_data_len;
7585
7586         mtx_lock(&lun->lun_lock);
7587         switch (cdb->action) {
7588         case SPRI_RK: { // read keys
7589         struct scsi_per_res_in_keys *res_keys;
7590                 int i, key_count;
7591
7592                 res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7593
7594                 /*
7595                  * We had to drop the lock to allocate our buffer, which
7596                  * leaves time for someone to come in with another
7597                  * persistent reservation.  (That is unlikely, though,
7598                  * since this should be the only persistent reservation
7599                  * command active right now.)
7600                  */
7601                 if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7602                     (lun->pr_key_count *
7603                      sizeof(struct scsi_per_res_key)))){
7604                         mtx_unlock(&lun->lun_lock);
7605                         free(ctsio->kern_data_ptr, M_CTL);
7606                         printf("%s: reservation length changed, retrying\n",
7607                                __func__);
7608                         goto retry;
7609                 }
7610
7611                 scsi_ulto4b(lun->pr_generation, res_keys->header.generation);
7612
7613                 scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7614                              lun->pr_key_count, res_keys->header.length);
7615
7616                 for (i = 0, key_count = 0; i < CTL_MAX_INITIATORS; i++) {
7617                         if ((key = ctl_get_prkey(lun, i)) == 0)
7618                                 continue;
7619
7620                         /*
7621                          * We used lun->pr_key_count to calculate the
7622                          * size to allocate.  If it turns out the number of
7623                          * initiators with the registered flag set is
7624                          * larger than that (i.e. they haven't been kept in
7625                          * sync), we've got a problem.
7626                          */
7627                         if (key_count >= lun->pr_key_count) {
7628                                 key_count++;
7629                                 continue;
7630                         }
7631                         scsi_u64to8b(key, res_keys->keys[key_count].key);
7632                         key_count++;
7633                 }
7634                 break;
7635         }
7636         case SPRI_RR: { // read reservation
7637                 struct scsi_per_res_in_rsrv *res;
7638                 int tmp_len, header_only;
7639
7640                 res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7641
7642                 scsi_ulto4b(lun->pr_generation, res->header.generation);
7643
7644                 if (lun->flags & CTL_LUN_PR_RESERVED)
7645                 {
7646                         tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7647                         scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7648                                     res->header.length);
7649                         header_only = 0;
7650                 } else {
7651                         tmp_len = sizeof(struct scsi_per_res_in_header);
7652                         scsi_ulto4b(0, res->header.length);
7653                         header_only = 1;
7654                 }
7655
7656                 /*
7657                  * We had to drop the lock to allocate our buffer, which
7658                  * leaves time for someone to come in with another
7659                  * persistent reservation.  (That is unlikely, though,
7660                  * since this should be the only persistent reservation
7661                  * command active right now.)
7662                  */
7663                 if (tmp_len != total_len) {
7664                         mtx_unlock(&lun->lun_lock);
7665                         free(ctsio->kern_data_ptr, M_CTL);
7666                         printf("%s: reservation status changed, retrying\n",
7667                                __func__);
7668                         goto retry;
7669                 }
7670
7671                 /*
7672                  * No reservation held, so we're done.
7673                  */
7674                 if (header_only != 0)
7675                         break;
7676
7677                 /*
7678                  * If the registration is an All Registrants type, the key
7679                  * is 0, since it doesn't really matter.
7680                  */
7681                 if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7682                         scsi_u64to8b(ctl_get_prkey(lun, lun->pr_res_idx),
7683                             res->data.reservation);
7684                 }
7685                 res->data.scopetype = lun->pr_res_type;
7686                 break;
7687         }
7688         case SPRI_RC:     //report capabilities
7689         {
7690                 struct scsi_per_res_cap *res_cap;
7691                 uint16_t type_mask;
7692
7693                 res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7694                 scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7695                 res_cap->flags1 = SPRI_CRH;
7696                 res_cap->flags2 = SPRI_TMV | SPRI_ALLOW_5;
7697                 type_mask = SPRI_TM_WR_EX_AR |
7698                             SPRI_TM_EX_AC_RO |
7699                             SPRI_TM_WR_EX_RO |
7700                             SPRI_TM_EX_AC |
7701                             SPRI_TM_WR_EX |
7702                             SPRI_TM_EX_AC_AR;
7703                 scsi_ulto2b(type_mask, res_cap->type_mask);
7704                 break;
7705         }
7706         case SPRI_RS: { // read full status
7707                 struct scsi_per_res_in_full *res_status;
7708                 struct scsi_per_res_in_full_desc *res_desc;
7709                 struct ctl_port *port;
7710                 int i, len;
7711
7712                 res_status = (struct scsi_per_res_in_full*)ctsio->kern_data_ptr;
7713
7714                 /*
7715                  * We had to drop the lock to allocate our buffer, which
7716                  * leaves time for someone to come in with another
7717                  * persistent reservation.  (That is unlikely, though,
7718                  * since this should be the only persistent reservation
7719                  * command active right now.)
7720                  */
7721                 if (total_len < (sizeof(struct scsi_per_res_in_header) +
7722                     (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7723                      lun->pr_key_count)){
7724                         mtx_unlock(&lun->lun_lock);
7725                         free(ctsio->kern_data_ptr, M_CTL);
7726                         printf("%s: reservation length changed, retrying\n",
7727                                __func__);
7728                         goto retry;
7729                 }
7730
7731                 scsi_ulto4b(lun->pr_generation, res_status->header.generation);
7732
7733                 res_desc = &res_status->desc[0];
7734                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
7735                         if ((key = ctl_get_prkey(lun, i)) == 0)
7736                                 continue;
7737
7738                         scsi_u64to8b(key, res_desc->res_key.key);
7739                         if ((lun->flags & CTL_LUN_PR_RESERVED) &&
7740                             (lun->pr_res_idx == i ||
7741                              lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS)) {
7742                                 res_desc->flags = SPRI_FULL_R_HOLDER;
7743                                 res_desc->scopetype = lun->pr_res_type;
7744                         }
7745                         scsi_ulto2b(i / CTL_MAX_INIT_PER_PORT,
7746                             res_desc->rel_trgt_port_id);
7747                         len = 0;
7748                         port = softc->ctl_ports[i / CTL_MAX_INIT_PER_PORT];
7749                         if (port != NULL)
7750                                 len = ctl_create_iid(port,
7751                                     i % CTL_MAX_INIT_PER_PORT,
7752                                     res_desc->transport_id);
7753                         scsi_ulto4b(len, res_desc->additional_length);
7754                         res_desc = (struct scsi_per_res_in_full_desc *)
7755                             &res_desc->transport_id[len];
7756                 }
7757                 scsi_ulto4b((uint8_t *)res_desc - (uint8_t *)&res_status->desc[0],
7758                     res_status->header.length);
7759                 break;
7760         }
7761         default:
7762                 panic("%s: Invalid PR type %#x", __func__, cdb->action);
7763         }
7764         mtx_unlock(&lun->lun_lock);
7765
7766         ctl_set_success(ctsio);
7767         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7768         ctsio->be_move_done = ctl_config_move_done;
7769         ctl_datamove((union ctl_io *)ctsio);
7770         return (CTL_RETVAL_COMPLETE);
7771 }
7772
7773 /*
7774  * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
7775  * it should return.
7776  */
7777 static int
7778 ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
7779                 uint64_t sa_res_key, uint8_t type, uint32_t residx,
7780                 struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
7781                 struct scsi_per_res_out_parms* param)
7782 {
7783         union ctl_ha_msg persis_io;
7784         int i;
7785
7786         mtx_lock(&lun->lun_lock);
7787         if (sa_res_key == 0) {
7788                 if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
7789                         /* validate scope and type */
7790                         if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7791                              SPR_LU_SCOPE) {
7792                                 mtx_unlock(&lun->lun_lock);
7793                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7794                                                       /*sks_valid*/ 1,
7795                                                       /*command*/ 1,
7796                                                       /*field*/ 2,
7797                                                       /*bit_valid*/ 1,
7798                                                       /*bit*/ 4);
7799                                 ctl_done((union ctl_io *)ctsio);
7800                                 return (1);
7801                         }
7802
7803                         if (type>8 || type==2 || type==4 || type==0) {
7804                                 mtx_unlock(&lun->lun_lock);
7805                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7806                                                       /*sks_valid*/ 1,
7807                                                       /*command*/ 1,
7808                                                       /*field*/ 2,
7809                                                       /*bit_valid*/ 1,
7810                                                       /*bit*/ 0);
7811                                 ctl_done((union ctl_io *)ctsio);
7812                                 return (1);
7813                         }
7814
7815                         /*
7816                          * Unregister everybody else and build UA for
7817                          * them
7818                          */
7819                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
7820                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
7821                                         continue;
7822
7823                                 ctl_clr_prkey(lun, i);
7824                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7825                         }
7826                         lun->pr_key_count = 1;
7827                         lun->pr_res_type = type;
7828                         if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
7829                             lun->pr_res_type != SPR_TYPE_EX_AC_AR)
7830                                 lun->pr_res_idx = residx;
7831                         lun->pr_generation++;
7832                         mtx_unlock(&lun->lun_lock);
7833
7834                         /* send msg to other side */
7835                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7836                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7837                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7838                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
7839                         persis_io.pr.pr_info.res_type = type;
7840                         memcpy(persis_io.pr.pr_info.sa_res_key,
7841                                param->serv_act_res_key,
7842                                sizeof(param->serv_act_res_key));
7843                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
7844                             sizeof(persis_io.pr), M_WAITOK);
7845                 } else {
7846                         /* not all registrants */
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         } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
7859                 || !(lun->flags & CTL_LUN_PR_RESERVED)) {
7860                 int found = 0;
7861
7862                 if (res_key == sa_res_key) {
7863                         /* special case */
7864                         /*
7865                          * The spec implies this is not good but doesn't
7866                          * say what to do. There are two choices either
7867                          * generate a res conflict or check condition
7868                          * with illegal field in parameter data. Since
7869                          * that is what is done when the sa_res_key is
7870                          * zero I'll take that approach since this has
7871                          * to do with the sa_res_key.
7872                          */
7873                         mtx_unlock(&lun->lun_lock);
7874                         free(ctsio->kern_data_ptr, M_CTL);
7875                         ctl_set_invalid_field(ctsio,
7876                                               /*sks_valid*/ 1,
7877                                               /*command*/ 0,
7878                                               /*field*/ 8,
7879                                               /*bit_valid*/ 0,
7880                                               /*bit*/ 0);
7881                         ctl_done((union ctl_io *)ctsio);
7882                         return (1);
7883                 }
7884
7885                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
7886                         if (ctl_get_prkey(lun, i) != sa_res_key)
7887                                 continue;
7888
7889                         found = 1;
7890                         ctl_clr_prkey(lun, i);
7891                         lun->pr_key_count--;
7892                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7893                 }
7894                 if (!found) {
7895                         mtx_unlock(&lun->lun_lock);
7896                         free(ctsio->kern_data_ptr, M_CTL);
7897                         ctl_set_reservation_conflict(ctsio);
7898                         ctl_done((union ctl_io *)ctsio);
7899                         return (CTL_RETVAL_COMPLETE);
7900                 }
7901                 lun->pr_generation++;
7902                 mtx_unlock(&lun->lun_lock);
7903
7904                 /* send msg to other side */
7905                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7906                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7907                 persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7908                 persis_io.pr.pr_info.residx = lun->pr_res_idx;
7909                 persis_io.pr.pr_info.res_type = type;
7910                 memcpy(persis_io.pr.pr_info.sa_res_key,
7911                        param->serv_act_res_key,
7912                        sizeof(param->serv_act_res_key));
7913                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
7914                     sizeof(persis_io.pr), M_WAITOK);
7915         } else {
7916                 /* Reserved but not all registrants */
7917                 /* sa_res_key is res holder */
7918                 if (sa_res_key == ctl_get_prkey(lun, lun->pr_res_idx)) {
7919                         /* validate scope and type */
7920                         if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7921                              SPR_LU_SCOPE) {
7922                                 mtx_unlock(&lun->lun_lock);
7923                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7924                                                       /*sks_valid*/ 1,
7925                                                       /*command*/ 1,
7926                                                       /*field*/ 2,
7927                                                       /*bit_valid*/ 1,
7928                                                       /*bit*/ 4);
7929                                 ctl_done((union ctl_io *)ctsio);
7930                                 return (1);
7931                         }
7932
7933                         if (type>8 || type==2 || type==4 || type==0) {
7934                                 mtx_unlock(&lun->lun_lock);
7935                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7936                                                       /*sks_valid*/ 1,
7937                                                       /*command*/ 1,
7938                                                       /*field*/ 2,
7939                                                       /*bit_valid*/ 1,
7940                                                       /*bit*/ 0);
7941                                 ctl_done((union ctl_io *)ctsio);
7942                                 return (1);
7943                         }
7944
7945                         /*
7946                          * Do the following:
7947                          * if sa_res_key != res_key remove all
7948                          * registrants w/sa_res_key and generate UA
7949                          * for these registrants(Registrations
7950                          * Preempted) if it wasn't an exclusive
7951                          * reservation generate UA(Reservations
7952                          * Preempted) for all other registered nexuses
7953                          * if the type has changed. Establish the new
7954                          * reservation and holder. If res_key and
7955                          * sa_res_key are the same do the above
7956                          * except don't unregister the res holder.
7957                          */
7958
7959                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
7960                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
7961                                         continue;
7962
7963                                 if (sa_res_key == ctl_get_prkey(lun, i)) {
7964                                         ctl_clr_prkey(lun, i);
7965                                         lun->pr_key_count--;
7966                                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7967                                 } else if (type != lun->pr_res_type &&
7968                                     (lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
7969                                      lun->pr_res_type == SPR_TYPE_EX_AC_RO)) {
7970                                         ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
7971                                 }
7972                         }
7973                         lun->pr_res_type = type;
7974                         if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
7975                             lun->pr_res_type != SPR_TYPE_EX_AC_AR)
7976                                 lun->pr_res_idx = residx;
7977                         else
7978                                 lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
7979                         lun->pr_generation++;
7980                         mtx_unlock(&lun->lun_lock);
7981
7982                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7983                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7984                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7985                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
7986                         persis_io.pr.pr_info.res_type = type;
7987                         memcpy(persis_io.pr.pr_info.sa_res_key,
7988                                param->serv_act_res_key,
7989                                sizeof(param->serv_act_res_key));
7990                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
7991                             sizeof(persis_io.pr), M_WAITOK);
7992                 } else {
7993                         /*
7994                          * sa_res_key is not the res holder just
7995                          * remove registrants
7996                          */
7997                         int found=0;
7998
7999                         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8000                                 if (sa_res_key != ctl_get_prkey(lun, i))
8001                                         continue;
8002
8003                                 found = 1;
8004                                 ctl_clr_prkey(lun, i);
8005                                 lun->pr_key_count--;
8006                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8007                         }
8008
8009                         if (!found) {
8010                                 mtx_unlock(&lun->lun_lock);
8011                                 free(ctsio->kern_data_ptr, M_CTL);
8012                                 ctl_set_reservation_conflict(ctsio);
8013                                 ctl_done((union ctl_io *)ctsio);
8014                                 return (1);
8015                         }
8016                         lun->pr_generation++;
8017                         mtx_unlock(&lun->lun_lock);
8018
8019                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8020                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8021                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8022                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8023                         persis_io.pr.pr_info.res_type = type;
8024                         memcpy(persis_io.pr.pr_info.sa_res_key,
8025                                param->serv_act_res_key,
8026                                sizeof(param->serv_act_res_key));
8027                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8028                             sizeof(persis_io.pr), M_WAITOK);
8029                 }
8030         }
8031         return (0);
8032 }
8033
8034 static void
8035 ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
8036 {
8037         uint64_t sa_res_key;
8038         int i;
8039
8040         sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
8041
8042         if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8043          || lun->pr_res_idx == CTL_PR_NO_RESERVATION
8044          || sa_res_key != ctl_get_prkey(lun, lun->pr_res_idx)) {
8045                 if (sa_res_key == 0) {
8046                         /*
8047                          * Unregister everybody else and build UA for
8048                          * them
8049                          */
8050                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
8051                                 if (i == msg->pr.pr_info.residx ||
8052                                     ctl_get_prkey(lun, i) == 0)
8053                                         continue;
8054
8055                                 ctl_clr_prkey(lun, i);
8056                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8057                         }
8058
8059                         lun->pr_key_count = 1;
8060                         lun->pr_res_type = msg->pr.pr_info.res_type;
8061                         if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
8062                             lun->pr_res_type != SPR_TYPE_EX_AC_AR)
8063                                 lun->pr_res_idx = msg->pr.pr_info.residx;
8064                 } else {
8065                         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8066                                 if (sa_res_key == ctl_get_prkey(lun, i))
8067                                         continue;
8068
8069                                 ctl_clr_prkey(lun, i);
8070                                 lun->pr_key_count--;
8071                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8072                         }
8073                 }
8074         } else {
8075                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8076                         if (i == msg->pr.pr_info.residx ||
8077                             ctl_get_prkey(lun, i) == 0)
8078                                 continue;
8079
8080                         if (sa_res_key == ctl_get_prkey(lun, i)) {
8081                                 ctl_clr_prkey(lun, i);
8082                                 lun->pr_key_count--;
8083                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8084                         } else if (msg->pr.pr_info.res_type != lun->pr_res_type
8085                             && (lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8086                              lun->pr_res_type == SPR_TYPE_EX_AC_RO)) {
8087                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8088                         }
8089                 }
8090                 lun->pr_res_type = msg->pr.pr_info.res_type;
8091                 if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
8092                     lun->pr_res_type != SPR_TYPE_EX_AC_AR)
8093                         lun->pr_res_idx = msg->pr.pr_info.residx;
8094                 else
8095                         lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8096         }
8097         lun->pr_generation++;
8098
8099 }
8100
8101
8102 int
8103 ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
8104 {
8105         struct ctl_softc *softc = CTL_SOFTC(ctsio);
8106         struct ctl_lun *lun = CTL_LUN(ctsio);
8107         int retval;
8108         u_int32_t param_len;
8109         struct scsi_per_res_out *cdb;
8110         struct scsi_per_res_out_parms* param;
8111         uint32_t residx;
8112         uint64_t res_key, sa_res_key, key;
8113         uint8_t type;
8114         union ctl_ha_msg persis_io;
8115         int    i;
8116
8117         CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n"));
8118
8119         cdb = (struct scsi_per_res_out *)ctsio->cdb;
8120         retval = CTL_RETVAL_COMPLETE;
8121
8122         /*
8123          * We only support whole-LUN scope.  The scope & type are ignored for
8124          * register, register and ignore existing key and clear.
8125          * We sometimes ignore scope and type on preempts too!!
8126          * Verify reservation type here as well.
8127          */
8128         type = cdb->scope_type & SPR_TYPE_MASK;
8129         if ((cdb->action == SPRO_RESERVE)
8130          || (cdb->action == SPRO_RELEASE)) {
8131                 if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
8132                         ctl_set_invalid_field(/*ctsio*/ ctsio,
8133                                               /*sks_valid*/ 1,
8134                                               /*command*/ 1,
8135                                               /*field*/ 2,
8136                                               /*bit_valid*/ 1,
8137                                               /*bit*/ 4);
8138                         ctl_done((union ctl_io *)ctsio);
8139                         return (CTL_RETVAL_COMPLETE);
8140                 }
8141
8142                 if (type>8 || type==2 || type==4 || type==0) {
8143                         ctl_set_invalid_field(/*ctsio*/ ctsio,
8144                                               /*sks_valid*/ 1,
8145                                               /*command*/ 1,
8146                                               /*field*/ 2,
8147                                               /*bit_valid*/ 1,
8148                                               /*bit*/ 0);
8149                         ctl_done((union ctl_io *)ctsio);
8150                         return (CTL_RETVAL_COMPLETE);
8151                 }
8152         }
8153
8154         param_len = scsi_4btoul(cdb->length);
8155
8156         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
8157                 ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
8158                 ctsio->kern_data_len = param_len;
8159                 ctsio->kern_total_len = param_len;
8160                 ctsio->kern_rel_offset = 0;
8161                 ctsio->kern_sg_entries = 0;
8162                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
8163                 ctsio->be_move_done = ctl_config_move_done;
8164                 ctl_datamove((union ctl_io *)ctsio);
8165
8166                 return (CTL_RETVAL_COMPLETE);
8167         }
8168
8169         param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
8170
8171         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
8172         res_key = scsi_8btou64(param->res_key.key);
8173         sa_res_key = scsi_8btou64(param->serv_act_res_key);
8174
8175         /*
8176          * Validate the reservation key here except for SPRO_REG_IGNO
8177          * This must be done for all other service actions
8178          */
8179         if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
8180                 mtx_lock(&lun->lun_lock);
8181                 if ((key = ctl_get_prkey(lun, residx)) != 0) {
8182                         if (res_key != key) {
8183                                 /*
8184                                  * The current key passed in doesn't match
8185                                  * the one the initiator previously
8186                                  * registered.
8187                                  */
8188                                 mtx_unlock(&lun->lun_lock);
8189                                 free(ctsio->kern_data_ptr, M_CTL);
8190                                 ctl_set_reservation_conflict(ctsio);
8191                                 ctl_done((union ctl_io *)ctsio);
8192                                 return (CTL_RETVAL_COMPLETE);
8193                         }
8194                 } else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
8195                         /*
8196                          * We are not registered
8197                          */
8198                         mtx_unlock(&lun->lun_lock);
8199                         free(ctsio->kern_data_ptr, M_CTL);
8200                         ctl_set_reservation_conflict(ctsio);
8201                         ctl_done((union ctl_io *)ctsio);
8202                         return (CTL_RETVAL_COMPLETE);
8203                 } else if (res_key != 0) {
8204                         /*
8205                          * We are not registered and trying to register but
8206                          * the register key isn't zero.
8207                          */
8208                         mtx_unlock(&lun->lun_lock);
8209                         free(ctsio->kern_data_ptr, M_CTL);
8210                         ctl_set_reservation_conflict(ctsio);
8211                         ctl_done((union ctl_io *)ctsio);
8212                         return (CTL_RETVAL_COMPLETE);
8213                 }
8214                 mtx_unlock(&lun->lun_lock);
8215         }
8216
8217         switch (cdb->action & SPRO_ACTION_MASK) {
8218         case SPRO_REGISTER:
8219         case SPRO_REG_IGNO: {
8220
8221 #if 0
8222                 printf("Registration received\n");
8223 #endif
8224
8225                 /*
8226                  * We don't support any of these options, as we report in
8227                  * the read capabilities request (see
8228                  * ctl_persistent_reserve_in(), above).
8229                  */
8230                 if ((param->flags & SPR_SPEC_I_PT)
8231                  || (param->flags & SPR_ALL_TG_PT)
8232                  || (param->flags & SPR_APTPL)) {
8233                         int bit_ptr;
8234
8235                         if (param->flags & SPR_APTPL)
8236                                 bit_ptr = 0;
8237                         else if (param->flags & SPR_ALL_TG_PT)
8238                                 bit_ptr = 2;
8239                         else /* SPR_SPEC_I_PT */
8240                                 bit_ptr = 3;
8241
8242                         free(ctsio->kern_data_ptr, M_CTL);
8243                         ctl_set_invalid_field(ctsio,
8244                                               /*sks_valid*/ 1,
8245                                               /*command*/ 0,
8246                                               /*field*/ 20,
8247                                               /*bit_valid*/ 1,
8248                                               /*bit*/ bit_ptr);
8249                         ctl_done((union ctl_io *)ctsio);
8250                         return (CTL_RETVAL_COMPLETE);
8251                 }
8252
8253                 mtx_lock(&lun->lun_lock);
8254
8255                 /*
8256                  * The initiator wants to clear the
8257                  * key/unregister.
8258                  */
8259                 if (sa_res_key == 0) {
8260                         if ((res_key == 0
8261                           && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
8262                          || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
8263                           && ctl_get_prkey(lun, residx) == 0)) {
8264                                 mtx_unlock(&lun->lun_lock);
8265                                 goto done;
8266                         }
8267
8268                         ctl_clr_prkey(lun, residx);
8269                         lun->pr_key_count--;
8270
8271                         if (residx == lun->pr_res_idx) {
8272                                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8273                                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8274
8275                                 if ((lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8276                                      lun->pr_res_type == SPR_TYPE_EX_AC_RO) &&
8277                                     lun->pr_key_count) {
8278                                         /*
8279                                          * If the reservation is a registrants
8280                                          * only type we need to generate a UA
8281                                          * for other registered inits.  The
8282                                          * sense code should be RESERVATIONS
8283                                          * RELEASED
8284                                          */
8285
8286                                         for (i = softc->init_min; i < softc->init_max; i++){
8287                                                 if (ctl_get_prkey(lun, i) == 0)
8288                                                         continue;
8289                                                 ctl_est_ua(lun, i,
8290                                                     CTL_UA_RES_RELEASE);
8291                                         }
8292                                 }
8293                                 lun->pr_res_type = 0;
8294                         } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8295                                 if (lun->pr_key_count==0) {
8296                                         lun->flags &= ~CTL_LUN_PR_RESERVED;
8297                                         lun->pr_res_type = 0;
8298                                         lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8299                                 }
8300                         }
8301                         lun->pr_generation++;
8302                         mtx_unlock(&lun->lun_lock);
8303
8304                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8305                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8306                         persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
8307                         persis_io.pr.pr_info.residx = residx;
8308                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8309                             sizeof(persis_io.pr), M_WAITOK);
8310                 } else /* sa_res_key != 0 */ {
8311
8312                         /*
8313                          * If we aren't registered currently then increment
8314                          * the key count and set the registered flag.
8315                          */
8316                         ctl_alloc_prkey(lun, residx);
8317                         if (ctl_get_prkey(lun, residx) == 0)
8318                                 lun->pr_key_count++;
8319                         ctl_set_prkey(lun, residx, sa_res_key);
8320                         lun->pr_generation++;
8321                         mtx_unlock(&lun->lun_lock);
8322
8323                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8324                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8325                         persis_io.pr.pr_info.action = CTL_PR_REG_KEY;
8326                         persis_io.pr.pr_info.residx = residx;
8327                         memcpy(persis_io.pr.pr_info.sa_res_key,
8328                                param->serv_act_res_key,
8329                                sizeof(param->serv_act_res_key));
8330                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8331                             sizeof(persis_io.pr), M_WAITOK);
8332                 }
8333
8334                 break;
8335         }
8336         case SPRO_RESERVE:
8337 #if 0
8338                 printf("Reserve executed type %d\n", type);
8339 #endif
8340                 mtx_lock(&lun->lun_lock);
8341                 if (lun->flags & CTL_LUN_PR_RESERVED) {
8342                         /*
8343                          * if this isn't the reservation holder and it's
8344                          * not a "all registrants" type or if the type is
8345                          * different then we have a conflict
8346                          */
8347                         if ((lun->pr_res_idx != residx
8348                           && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8349                          || lun->pr_res_type != type) {
8350                                 mtx_unlock(&lun->lun_lock);
8351                                 free(ctsio->kern_data_ptr, M_CTL);
8352                                 ctl_set_reservation_conflict(ctsio);
8353                                 ctl_done((union ctl_io *)ctsio);
8354                                 return (CTL_RETVAL_COMPLETE);
8355                         }
8356                         mtx_unlock(&lun->lun_lock);
8357                 } else /* create a reservation */ {
8358                         /*
8359                          * If it's not an "all registrants" type record
8360                          * reservation holder
8361                          */
8362                         if (type != SPR_TYPE_WR_EX_AR
8363                          && type != SPR_TYPE_EX_AC_AR)
8364                                 lun->pr_res_idx = residx; /* Res holder */
8365                         else
8366                                 lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8367
8368                         lun->flags |= CTL_LUN_PR_RESERVED;
8369                         lun->pr_res_type = type;
8370
8371                         mtx_unlock(&lun->lun_lock);
8372
8373                         /* send msg to other side */
8374                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8375                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8376                         persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8377                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8378                         persis_io.pr.pr_info.res_type = type;
8379                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8380                             sizeof(persis_io.pr), M_WAITOK);
8381                 }
8382                 break;
8383
8384         case SPRO_RELEASE:
8385                 mtx_lock(&lun->lun_lock);
8386                 if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8387                         /* No reservation exists return good status */
8388                         mtx_unlock(&lun->lun_lock);
8389                         goto done;
8390                 }
8391                 /*
8392                  * Is this nexus a reservation holder?
8393                  */
8394                 if (lun->pr_res_idx != residx
8395                  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8396                         /*
8397                          * not a res holder return good status but
8398                          * do nothing
8399                          */
8400                         mtx_unlock(&lun->lun_lock);
8401                         goto done;
8402                 }
8403
8404                 if (lun->pr_res_type != type) {
8405                         mtx_unlock(&lun->lun_lock);
8406                         free(ctsio->kern_data_ptr, M_CTL);
8407                         ctl_set_illegal_pr_release(ctsio);
8408                         ctl_done((union ctl_io *)ctsio);
8409                         return (CTL_RETVAL_COMPLETE);
8410                 }
8411
8412                 /* okay to release */
8413                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8414                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8415                 lun->pr_res_type = 0;
8416
8417                 /*
8418                  * If this isn't an exclusive access reservation and NUAR
8419                  * is not set, generate UA for all other registrants.
8420                  */
8421                 if (type != SPR_TYPE_EX_AC && type != SPR_TYPE_WR_EX &&
8422                     (lun->MODE_CTRL.queue_flags & SCP_NUAR) == 0) {
8423                         for (i = softc->init_min; i < softc->init_max; i++) {
8424                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
8425                                         continue;
8426                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8427                         }
8428                 }
8429                 mtx_unlock(&lun->lun_lock);
8430
8431                 /* Send msg to other side */
8432                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8433                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8434                 persis_io.pr.pr_info.action = CTL_PR_RELEASE;
8435                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8436                      sizeof(persis_io.pr), M_WAITOK);
8437                 break;
8438
8439         case SPRO_CLEAR:
8440                 /* send msg to other side */
8441
8442                 mtx_lock(&lun->lun_lock);
8443                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8444                 lun->pr_res_type = 0;
8445                 lun->pr_key_count = 0;
8446                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8447
8448                 ctl_clr_prkey(lun, residx);
8449                 for (i = 0; i < CTL_MAX_INITIATORS; i++)
8450                         if (ctl_get_prkey(lun, i) != 0) {
8451                                 ctl_clr_prkey(lun, i);
8452                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8453                         }
8454                 lun->pr_generation++;
8455                 mtx_unlock(&lun->lun_lock);
8456
8457                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8458                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8459                 persis_io.pr.pr_info.action = CTL_PR_CLEAR;
8460                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8461                      sizeof(persis_io.pr), M_WAITOK);
8462                 break;
8463
8464         case SPRO_PREEMPT:
8465         case SPRO_PRE_ABO: {
8466                 int nretval;
8467
8468                 nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8469                                           residx, ctsio, cdb, param);
8470                 if (nretval != 0)
8471                         return (CTL_RETVAL_COMPLETE);
8472                 break;
8473         }
8474         default:
8475                 panic("%s: Invalid PR type %#x", __func__, cdb->action);
8476         }
8477
8478 done:
8479         free(ctsio->kern_data_ptr, M_CTL);
8480         ctl_set_success(ctsio);
8481         ctl_done((union ctl_io *)ctsio);
8482
8483         return (retval);
8484 }
8485
8486 /*
8487  * This routine is for handling a message from the other SC pertaining to
8488  * persistent reserve out. All the error checking will have been done
8489  * so only perorming the action need be done here to keep the two
8490  * in sync.
8491  */
8492 static void
8493 ctl_hndl_per_res_out_on_other_sc(union ctl_io *io)
8494 {
8495         struct ctl_softc *softc = CTL_SOFTC(io);
8496         union ctl_ha_msg *msg = (union ctl_ha_msg *)&io->presio.pr_msg;
8497         struct ctl_lun *lun;
8498         int i;
8499         uint32_t residx, targ_lun;
8500
8501         targ_lun = msg->hdr.nexus.targ_mapped_lun;
8502         mtx_lock(&softc->ctl_lock);
8503         if (targ_lun >= ctl_max_luns ||
8504             (lun = softc->ctl_luns[targ_lun]) == NULL) {
8505                 mtx_unlock(&softc->ctl_lock);
8506                 return;
8507         }
8508         mtx_lock(&lun->lun_lock);
8509         mtx_unlock(&softc->ctl_lock);
8510         if (lun->flags & CTL_LUN_DISABLED) {
8511                 mtx_unlock(&lun->lun_lock);
8512                 return;
8513         }
8514         residx = ctl_get_initindex(&msg->hdr.nexus);
8515         switch(msg->pr.pr_info.action) {
8516         case CTL_PR_REG_KEY:
8517                 ctl_alloc_prkey(lun, msg->pr.pr_info.residx);
8518                 if (ctl_get_prkey(lun, msg->pr.pr_info.residx) == 0)
8519                         lun->pr_key_count++;
8520                 ctl_set_prkey(lun, msg->pr.pr_info.residx,
8521                     scsi_8btou64(msg->pr.pr_info.sa_res_key));
8522                 lun->pr_generation++;
8523                 break;
8524
8525         case CTL_PR_UNREG_KEY:
8526                 ctl_clr_prkey(lun, msg->pr.pr_info.residx);
8527                 lun->pr_key_count--;
8528
8529                 /* XXX Need to see if the reservation has been released */
8530                 /* if so do we need to generate UA? */
8531                 if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8532                         lun->flags &= ~CTL_LUN_PR_RESERVED;
8533                         lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8534
8535                         if ((lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8536                              lun->pr_res_type == SPR_TYPE_EX_AC_RO) &&
8537                             lun->pr_key_count) {
8538                                 /*
8539                                  * If the reservation is a registrants
8540                                  * only type we need to generate a UA
8541                                  * for other registered inits.  The
8542                                  * sense code should be RESERVATIONS
8543                                  * RELEASED
8544                                  */
8545
8546                                 for (i = softc->init_min; i < softc->init_max; i++) {
8547                                         if (ctl_get_prkey(lun, i) == 0)
8548                                                 continue;
8549
8550                                         ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8551                                 }
8552                         }
8553                         lun->pr_res_type = 0;
8554                 } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8555                         if (lun->pr_key_count==0) {
8556                                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8557                                 lun->pr_res_type = 0;
8558                                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8559                         }
8560                 }
8561                 lun->pr_generation++;
8562                 break;
8563
8564         case CTL_PR_RESERVE:
8565                 lun->flags |= CTL_LUN_PR_RESERVED;
8566                 lun->pr_res_type = msg->pr.pr_info.res_type;
8567                 lun->pr_res_idx = msg->pr.pr_info.residx;
8568
8569                 break;
8570
8571         case CTL_PR_RELEASE:
8572                 /*
8573                  * If this isn't an exclusive access reservation and NUAR
8574                  * is not set, generate UA for all other registrants.
8575                  */
8576                 if (lun->pr_res_type != SPR_TYPE_EX_AC &&
8577                     lun->pr_res_type != SPR_TYPE_WR_EX &&
8578                     (lun->MODE_CTRL.queue_flags & SCP_NUAR) == 0) {
8579                         for (i = softc->init_min; i < softc->init_max; i++) {
8580                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
8581                                         continue;
8582                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8583                         }
8584                 }
8585
8586                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8587                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8588                 lun->pr_res_type = 0;
8589                 break;
8590
8591         case CTL_PR_PREEMPT:
8592                 ctl_pro_preempt_other(lun, msg);
8593                 break;
8594         case CTL_PR_CLEAR:
8595                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8596                 lun->pr_res_type = 0;
8597                 lun->pr_key_count = 0;
8598                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8599
8600                 for (i=0; i < CTL_MAX_INITIATORS; i++) {
8601                         if (ctl_get_prkey(lun, i) == 0)
8602                                 continue;
8603                         ctl_clr_prkey(lun, i);
8604                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8605                 }
8606                 lun->pr_generation++;
8607                 break;
8608         }
8609
8610         mtx_unlock(&lun->lun_lock);
8611 }
8612
8613 int
8614 ctl_read_write(struct ctl_scsiio *ctsio)
8615 {
8616         struct ctl_lun *lun = CTL_LUN(ctsio);
8617         struct ctl_lba_len_flags *lbalen;
8618         uint64_t lba;
8619         uint32_t num_blocks;
8620         int flags, retval;
8621         int isread;
8622
8623         CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
8624
8625         flags = 0;
8626         isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
8627               || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
8628         switch (ctsio->cdb[0]) {
8629         case READ_6:
8630         case WRITE_6: {
8631                 struct scsi_rw_6 *cdb;
8632
8633                 cdb = (struct scsi_rw_6 *)ctsio->cdb;
8634
8635                 lba = scsi_3btoul(cdb->addr);
8636                 /* only 5 bits are valid in the most significant address byte */
8637                 lba &= 0x1fffff;
8638                 num_blocks = cdb->length;
8639                 /*
8640                  * This is correct according to SBC-2.
8641                  */
8642                 if (num_blocks == 0)
8643                         num_blocks = 256;
8644                 break;
8645         }
8646         case READ_10:
8647         case WRITE_10: {
8648                 struct scsi_rw_10 *cdb;
8649
8650                 cdb = (struct scsi_rw_10 *)ctsio->cdb;
8651                 if (cdb->byte2 & SRW10_FUA)
8652                         flags |= CTL_LLF_FUA;
8653                 if (cdb->byte2 & SRW10_DPO)
8654                         flags |= CTL_LLF_DPO;
8655                 lba = scsi_4btoul(cdb->addr);
8656                 num_blocks = scsi_2btoul(cdb->length);
8657                 break;
8658         }
8659         case WRITE_VERIFY_10: {
8660                 struct scsi_write_verify_10 *cdb;
8661
8662                 cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
8663                 flags |= CTL_LLF_FUA;
8664                 if (cdb->byte2 & SWV_DPO)
8665                         flags |= CTL_LLF_DPO;
8666                 lba = scsi_4btoul(cdb->addr);
8667                 num_blocks = scsi_2btoul(cdb->length);
8668                 break;
8669         }
8670         case READ_12:
8671         case WRITE_12: {
8672                 struct scsi_rw_12 *cdb;
8673
8674                 cdb = (struct scsi_rw_12 *)ctsio->cdb;
8675                 if (cdb->byte2 & SRW12_FUA)
8676                         flags |= CTL_LLF_FUA;
8677                 if (cdb->byte2 & SRW12_DPO)
8678                         flags |= CTL_LLF_DPO;
8679                 lba = scsi_4btoul(cdb->addr);
8680                 num_blocks = scsi_4btoul(cdb->length);
8681                 break;
8682         }
8683         case WRITE_VERIFY_12: {
8684                 struct scsi_write_verify_12 *cdb;
8685
8686                 cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
8687                 flags |= CTL_LLF_FUA;
8688                 if (cdb->byte2 & SWV_DPO)
8689                         flags |= CTL_LLF_DPO;
8690                 lba = scsi_4btoul(cdb->addr);
8691                 num_blocks = scsi_4btoul(cdb->length);
8692                 break;
8693         }
8694         case READ_16:
8695         case WRITE_16: {
8696                 struct scsi_rw_16 *cdb;
8697
8698                 cdb = (struct scsi_rw_16 *)ctsio->cdb;
8699                 if (cdb->byte2 & SRW12_FUA)
8700                         flags |= CTL_LLF_FUA;
8701                 if (cdb->byte2 & SRW12_DPO)
8702                         flags |= CTL_LLF_DPO;
8703                 lba = scsi_8btou64(cdb->addr);
8704                 num_blocks = scsi_4btoul(cdb->length);
8705                 break;
8706         }
8707         case WRITE_ATOMIC_16: {
8708                 struct scsi_write_atomic_16 *cdb;
8709
8710                 if (lun->be_lun->atomicblock == 0) {
8711                         ctl_set_invalid_opcode(ctsio);
8712                         ctl_done((union ctl_io *)ctsio);
8713                         return (CTL_RETVAL_COMPLETE);
8714                 }
8715
8716                 cdb = (struct scsi_write_atomic_16 *)ctsio->cdb;
8717                 if (cdb->byte2 & SRW12_FUA)
8718                         flags |= CTL_LLF_FUA;
8719                 if (cdb->byte2 & SRW12_DPO)
8720                         flags |= CTL_LLF_DPO;
8721                 lba = scsi_8btou64(cdb->addr);
8722                 num_blocks = scsi_2btoul(cdb->length);
8723                 if (num_blocks > lun->be_lun->atomicblock) {
8724                         ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
8725                             /*command*/ 1, /*field*/ 12, /*bit_valid*/ 0,
8726                             /*bit*/ 0);
8727                         ctl_done((union ctl_io *)ctsio);
8728                         return (CTL_RETVAL_COMPLETE);
8729                 }
8730                 break;
8731         }
8732         case WRITE_VERIFY_16: {
8733                 struct scsi_write_verify_16 *cdb;
8734
8735                 cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
8736                 flags |= CTL_LLF_FUA;
8737                 if (cdb->byte2 & SWV_DPO)
8738                         flags |= CTL_LLF_DPO;
8739                 lba = scsi_8btou64(cdb->addr);
8740                 num_blocks = scsi_4btoul(cdb->length);
8741                 break;
8742         }
8743         default:
8744                 /*
8745                  * We got a command we don't support.  This shouldn't
8746                  * happen, commands should be filtered out above us.
8747                  */
8748                 ctl_set_invalid_opcode(ctsio);
8749                 ctl_done((union ctl_io *)ctsio);
8750
8751                 return (CTL_RETVAL_COMPLETE);
8752                 break; /* NOTREACHED */
8753         }
8754
8755         /*
8756          * The first check is to make sure we're in bounds, the second
8757          * check is to catch wrap-around problems.  If the lba + num blocks
8758          * is less than the lba, then we've wrapped around and the block
8759          * range is invalid anyway.
8760          */
8761         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8762          || ((lba + num_blocks) < lba)) {
8763                 ctl_set_lba_out_of_range(ctsio,
8764                     MAX(lba, lun->be_lun->maxlba + 1));
8765                 ctl_done((union ctl_io *)ctsio);
8766                 return (CTL_RETVAL_COMPLETE);
8767         }
8768
8769         /*
8770          * According to SBC-3, a transfer length of 0 is not an error.
8771          * Note that this cannot happen with WRITE(6) or READ(6), since 0
8772          * translates to 256 blocks for those commands.
8773          */
8774         if (num_blocks == 0) {
8775                 ctl_set_success(ctsio);
8776                 ctl_done((union ctl_io *)ctsio);
8777                 return (CTL_RETVAL_COMPLETE);
8778         }
8779
8780         /* Set FUA and/or DPO if caches are disabled. */
8781         if (isread) {
8782                 if ((lun->MODE_CACHING.flags1 & SCP_RCD) != 0)
8783                         flags |= CTL_LLF_FUA | CTL_LLF_DPO;
8784         } else {
8785                 if ((lun->MODE_CACHING.flags1 & SCP_WCE) == 0)
8786                         flags |= CTL_LLF_FUA;
8787         }
8788
8789         lbalen = (struct ctl_lba_len_flags *)
8790             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8791         lbalen->lba = lba;
8792         lbalen->len = num_blocks;
8793         lbalen->flags = (isread ? CTL_LLF_READ : CTL_LLF_WRITE) | flags;
8794
8795         ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
8796         ctsio->kern_rel_offset = 0;
8797
8798         CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
8799
8800         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8801         return (retval);
8802 }
8803
8804 static int
8805 ctl_cnw_cont(union ctl_io *io)
8806 {
8807         struct ctl_lun *lun = CTL_LUN(io);
8808         struct ctl_scsiio *ctsio;
8809         struct ctl_lba_len_flags *lbalen;
8810         int retval;
8811
8812         ctsio = &io->scsiio;
8813         ctsio->io_hdr.status = CTL_STATUS_NONE;
8814         ctsio->io_hdr.flags &= ~CTL_FLAG_IO_CONT;
8815         lbalen = (struct ctl_lba_len_flags *)
8816             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8817         lbalen->flags &= ~CTL_LLF_COMPARE;
8818         lbalen->flags |= CTL_LLF_WRITE;
8819
8820         CTL_DEBUG_PRINT(("ctl_cnw_cont: calling data_submit()\n"));
8821         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8822         return (retval);
8823 }
8824
8825 int
8826 ctl_cnw(struct ctl_scsiio *ctsio)
8827 {
8828         struct ctl_lun *lun = CTL_LUN(ctsio);
8829         struct ctl_lba_len_flags *lbalen;
8830         uint64_t lba;
8831         uint32_t num_blocks;
8832         int flags, retval;
8833
8834         CTL_DEBUG_PRINT(("ctl_cnw: command: %#x\n", ctsio->cdb[0]));
8835
8836         flags = 0;
8837         switch (ctsio->cdb[0]) {
8838         case COMPARE_AND_WRITE: {
8839                 struct scsi_compare_and_write *cdb;
8840
8841                 cdb = (struct scsi_compare_and_write *)ctsio->cdb;
8842                 if (cdb->byte2 & SRW10_FUA)
8843                         flags |= CTL_LLF_FUA;
8844                 if (cdb->byte2 & SRW10_DPO)
8845                         flags |= CTL_LLF_DPO;
8846                 lba = scsi_8btou64(cdb->addr);
8847                 num_blocks = cdb->length;
8848                 break;
8849         }
8850         default:
8851                 /*
8852                  * We got a command we don't support.  This shouldn't
8853                  * happen, commands should be filtered out above us.
8854                  */
8855                 ctl_set_invalid_opcode(ctsio);
8856                 ctl_done((union ctl_io *)ctsio);
8857
8858                 return (CTL_RETVAL_COMPLETE);
8859                 break; /* NOTREACHED */
8860         }
8861
8862         /*
8863          * The first check is to make sure we're in bounds, the second
8864          * check is to catch wrap-around problems.  If the lba + num blocks
8865          * is less than the lba, then we've wrapped around and the block
8866          * range is invalid anyway.
8867          */
8868         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8869          || ((lba + num_blocks) < lba)) {
8870                 ctl_set_lba_out_of_range(ctsio,
8871                     MAX(lba, lun->be_lun->maxlba + 1));
8872                 ctl_done((union ctl_io *)ctsio);
8873                 return (CTL_RETVAL_COMPLETE);
8874         }
8875
8876         /*
8877          * According to SBC-3, a transfer length of 0 is not an error.
8878          */
8879         if (num_blocks == 0) {
8880                 ctl_set_success(ctsio);
8881                 ctl_done((union ctl_io *)ctsio);
8882                 return (CTL_RETVAL_COMPLETE);
8883         }
8884
8885         /* Set FUA if write cache is disabled. */
8886         if ((lun->MODE_CACHING.flags1 & SCP_WCE) == 0)
8887                 flags |= CTL_LLF_FUA;
8888
8889         ctsio->kern_total_len = 2 * num_blocks * lun->be_lun->blocksize;
8890         ctsio->kern_rel_offset = 0;
8891
8892         /*
8893          * Set the IO_CONT flag, so that if this I/O gets passed to
8894          * ctl_data_submit_done(), it'll get passed back to
8895          * ctl_ctl_cnw_cont() for further processing.
8896          */
8897         ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
8898         ctsio->io_cont = ctl_cnw_cont;
8899
8900         lbalen = (struct ctl_lba_len_flags *)
8901             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8902         lbalen->lba = lba;
8903         lbalen->len = num_blocks;
8904         lbalen->flags = CTL_LLF_COMPARE | flags;
8905
8906         CTL_DEBUG_PRINT(("ctl_cnw: calling data_submit()\n"));
8907         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8908         return (retval);
8909 }
8910
8911 int
8912 ctl_verify(struct ctl_scsiio *ctsio)
8913 {
8914         struct ctl_lun *lun = CTL_LUN(ctsio);
8915         struct ctl_lba_len_flags *lbalen;
8916         uint64_t lba;
8917         uint32_t num_blocks;
8918         int bytchk, flags;
8919         int retval;
8920
8921         CTL_DEBUG_PRINT(("ctl_verify: command: %#x\n", ctsio->cdb[0]));
8922
8923         bytchk = 0;
8924         flags = CTL_LLF_FUA;
8925         switch (ctsio->cdb[0]) {
8926         case VERIFY_10: {
8927                 struct scsi_verify_10 *cdb;
8928
8929                 cdb = (struct scsi_verify_10 *)ctsio->cdb;
8930                 if (cdb->byte2 & SVFY_BYTCHK)
8931                         bytchk = 1;
8932                 if (cdb->byte2 & SVFY_DPO)
8933                         flags |= CTL_LLF_DPO;
8934                 lba = scsi_4btoul(cdb->addr);
8935                 num_blocks = scsi_2btoul(cdb->length);
8936                 break;
8937         }
8938         case VERIFY_12: {
8939                 struct scsi_verify_12 *cdb;
8940
8941                 cdb = (struct scsi_verify_12 *)ctsio->cdb;
8942                 if (cdb->byte2 & SVFY_BYTCHK)
8943                         bytchk = 1;
8944                 if (cdb->byte2 & SVFY_DPO)
8945                         flags |= CTL_LLF_DPO;
8946                 lba = scsi_4btoul(cdb->addr);
8947                 num_blocks = scsi_4btoul(cdb->length);
8948                 break;
8949         }
8950         case VERIFY_16: {
8951                 struct scsi_rw_16 *cdb;
8952
8953                 cdb = (struct scsi_rw_16 *)ctsio->cdb;
8954                 if (cdb->byte2 & SVFY_BYTCHK)
8955                         bytchk = 1;
8956                 if (cdb->byte2 & SVFY_DPO)
8957                         flags |= CTL_LLF_DPO;
8958                 lba = scsi_8btou64(cdb->addr);
8959                 num_blocks = scsi_4btoul(cdb->length);
8960                 break;
8961         }
8962         default:
8963                 /*
8964                  * We got a command we don't support.  This shouldn't
8965                  * happen, commands should be filtered out above us.
8966                  */
8967                 ctl_set_invalid_opcode(ctsio);
8968                 ctl_done((union ctl_io *)ctsio);
8969                 return (CTL_RETVAL_COMPLETE);
8970         }
8971
8972         /*
8973          * The first check is to make sure we're in bounds, the second
8974          * check is to catch wrap-around problems.  If the lba + num blocks
8975          * is less than the lba, then we've wrapped around and the block
8976          * range is invalid anyway.
8977          */
8978         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8979          || ((lba + num_blocks) < lba)) {
8980                 ctl_set_lba_out_of_range(ctsio,
8981                     MAX(lba, lun->be_lun->maxlba + 1));
8982                 ctl_done((union ctl_io *)ctsio);
8983                 return (CTL_RETVAL_COMPLETE);
8984         }
8985
8986         /*
8987          * According to SBC-3, a transfer length of 0 is not an error.
8988          */
8989         if (num_blocks == 0) {
8990                 ctl_set_success(ctsio);
8991                 ctl_done((union ctl_io *)ctsio);
8992                 return (CTL_RETVAL_COMPLETE);
8993         }
8994
8995         lbalen = (struct ctl_lba_len_flags *)
8996             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8997         lbalen->lba = lba;
8998         lbalen->len = num_blocks;
8999         if (bytchk) {
9000                 lbalen->flags = CTL_LLF_COMPARE | flags;
9001                 ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9002         } else {
9003                 lbalen->flags = CTL_LLF_VERIFY | flags;
9004                 ctsio->kern_total_len = 0;
9005         }
9006         ctsio->kern_rel_offset = 0;
9007
9008         CTL_DEBUG_PRINT(("ctl_verify: calling data_submit()\n"));
9009         retval = lun->backend->data_submit((union ctl_io *)ctsio);
9010         return (retval);
9011 }
9012
9013 int
9014 ctl_report_luns(struct ctl_scsiio *ctsio)
9015 {
9016         struct ctl_softc *softc = CTL_SOFTC(ctsio);
9017         struct ctl_port *port = CTL_PORT(ctsio);
9018         struct ctl_lun *lun, *request_lun = CTL_LUN(ctsio);
9019         struct scsi_report_luns *cdb;
9020         struct scsi_report_luns_data *lun_data;
9021         int num_filled, num_luns, num_port_luns, retval;
9022         uint32_t alloc_len, lun_datalen;
9023         uint32_t initidx, targ_lun_id, lun_id;
9024
9025         retval = CTL_RETVAL_COMPLETE;
9026         cdb = (struct scsi_report_luns *)ctsio->cdb;
9027
9028         CTL_DEBUG_PRINT(("ctl_report_luns\n"));
9029
9030         num_luns = 0;
9031         num_port_luns = port->lun_map ? port->lun_map_size : ctl_max_luns;
9032         mtx_lock(&softc->ctl_lock);
9033         for (targ_lun_id = 0; targ_lun_id < num_port_luns; targ_lun_id++) {
9034                 if (ctl_lun_map_from_port(port, targ_lun_id) != UINT32_MAX)
9035                         num_luns++;
9036         }
9037         mtx_unlock(&softc->ctl_lock);
9038
9039         switch (cdb->select_report) {
9040         case RPL_REPORT_DEFAULT:
9041         case RPL_REPORT_ALL:
9042         case RPL_REPORT_NONSUBSID:
9043                 break;
9044         case RPL_REPORT_WELLKNOWN:
9045         case RPL_REPORT_ADMIN:
9046         case RPL_REPORT_CONGLOM:
9047                 num_luns = 0;
9048                 break;
9049         default:
9050                 ctl_set_invalid_field(ctsio,
9051                                       /*sks_valid*/ 1,
9052                                       /*command*/ 1,
9053                                       /*field*/ 2,
9054                                       /*bit_valid*/ 0,
9055                                       /*bit*/ 0);
9056                 ctl_done((union ctl_io *)ctsio);
9057                 return (retval);
9058                 break; /* NOTREACHED */
9059         }
9060
9061         alloc_len = scsi_4btoul(cdb->length);
9062         /*
9063          * The initiator has to allocate at least 16 bytes for this request,
9064          * so he can at least get the header and the first LUN.  Otherwise
9065          * we reject the request (per SPC-3 rev 14, section 6.21).
9066          */
9067         if (alloc_len < (sizeof(struct scsi_report_luns_data) +
9068             sizeof(struct scsi_report_luns_lundata))) {
9069                 ctl_set_invalid_field(ctsio,
9070                                       /*sks_valid*/ 1,
9071                                       /*command*/ 1,
9072                                       /*field*/ 6,
9073                                       /*bit_valid*/ 0,
9074                                       /*bit*/ 0);
9075                 ctl_done((union ctl_io *)ctsio);
9076                 return (retval);
9077         }
9078
9079         lun_datalen = sizeof(*lun_data) +
9080                 (num_luns * sizeof(struct scsi_report_luns_lundata));
9081
9082         ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO);
9083         lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
9084         ctsio->kern_sg_entries = 0;
9085
9086         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9087
9088         mtx_lock(&softc->ctl_lock);
9089         for (targ_lun_id = 0, num_filled = 0;
9090             targ_lun_id < num_port_luns && num_filled < num_luns;
9091             targ_lun_id++) {
9092                 lun_id = ctl_lun_map_from_port(port, targ_lun_id);
9093                 if (lun_id == UINT32_MAX)
9094                         continue;
9095                 lun = softc->ctl_luns[lun_id];
9096                 if (lun == NULL)
9097                         continue;
9098
9099                 be64enc(lun_data->luns[num_filled++].lundata,
9100                     ctl_encode_lun(targ_lun_id));
9101
9102                 /*
9103                  * According to SPC-3, rev 14 section 6.21:
9104                  *
9105                  * "The execution of a REPORT LUNS command to any valid and
9106                  * installed logical unit shall clear the REPORTED LUNS DATA
9107                  * HAS CHANGED unit attention condition for all logical
9108                  * units of that target with respect to the requesting
9109                  * initiator. A valid and installed logical unit is one
9110                  * having a PERIPHERAL QUALIFIER of 000b in the standard
9111                  * INQUIRY data (see 6.4.2)."
9112                  *
9113                  * If request_lun is NULL, the LUN this report luns command
9114                  * was issued to is either disabled or doesn't exist. In that
9115                  * case, we shouldn't clear any pending lun change unit
9116                  * attention.
9117                  */
9118                 if (request_lun != NULL) {
9119                         mtx_lock(&lun->lun_lock);
9120                         ctl_clr_ua(lun, initidx, CTL_UA_LUN_CHANGE);
9121                         mtx_unlock(&lun->lun_lock);
9122                 }
9123         }
9124         mtx_unlock(&softc->ctl_lock);
9125
9126         /*
9127          * It's quite possible that we've returned fewer LUNs than we allocated
9128          * space for.  Trim it.
9129          */
9130         lun_datalen = sizeof(*lun_data) +
9131                 (num_filled * sizeof(struct scsi_report_luns_lundata));
9132         ctsio->kern_rel_offset = 0;
9133         ctsio->kern_sg_entries = 0;
9134         ctsio->kern_data_len = min(lun_datalen, alloc_len);
9135         ctsio->kern_total_len = ctsio->kern_data_len;
9136
9137         /*
9138          * We set this to the actual data length, regardless of how much
9139          * space we actually have to return results.  If the user looks at
9140          * this value, he'll know whether or not he allocated enough space
9141          * and reissue the command if necessary.  We don't support well
9142          * known logical units, so if the user asks for that, return none.
9143          */
9144         scsi_ulto4b(lun_datalen - 8, lun_data->length);
9145
9146         /*
9147          * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
9148          * this request.
9149          */
9150         ctl_set_success(ctsio);
9151         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9152         ctsio->be_move_done = ctl_config_move_done;
9153         ctl_datamove((union ctl_io *)ctsio);
9154         return (retval);
9155 }
9156
9157 int
9158 ctl_request_sense(struct ctl_scsiio *ctsio)
9159 {
9160         struct ctl_softc *softc = CTL_SOFTC(ctsio);
9161         struct ctl_lun *lun = CTL_LUN(ctsio);
9162         struct scsi_request_sense *cdb;
9163         struct scsi_sense_data *sense_ptr, *ps;
9164         uint32_t initidx;
9165         int have_error;
9166         u_int sense_len = SSD_FULL_SIZE;
9167         scsi_sense_data_type sense_format;
9168         ctl_ua_type ua_type;
9169         uint8_t asc = 0, ascq = 0;
9170
9171         cdb = (struct scsi_request_sense *)ctsio->cdb;
9172
9173         CTL_DEBUG_PRINT(("ctl_request_sense\n"));
9174
9175         /*
9176          * Determine which sense format the user wants.
9177          */
9178         if (cdb->byte2 & SRS_DESC)
9179                 sense_format = SSD_TYPE_DESC;
9180         else
9181                 sense_format = SSD_TYPE_FIXED;
9182
9183         ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
9184         sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
9185         ctsio->kern_sg_entries = 0;
9186         ctsio->kern_rel_offset = 0;
9187
9188         /*
9189          * struct scsi_sense_data, which is currently set to 256 bytes, is
9190          * larger than the largest allowed value for the length field in the
9191          * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
9192          */
9193         ctsio->kern_data_len = cdb->length;
9194         ctsio->kern_total_len = cdb->length;
9195
9196         /*
9197          * If we don't have a LUN, we don't have any pending sense.
9198          */
9199         if (lun == NULL ||
9200             ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
9201              softc->ha_link < CTL_HA_LINK_UNKNOWN)) {
9202                 /* "Logical unit not supported" */
9203                 ctl_set_sense_data(sense_ptr, &sense_len, NULL, sense_format,
9204                     /*current_error*/ 1,
9205                     /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
9206                     /*asc*/ 0x25,
9207                     /*ascq*/ 0x00,
9208                     SSD_ELEM_NONE);
9209                 goto send;
9210         }
9211
9212         have_error = 0;
9213         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9214         /*
9215          * Check for pending sense, and then for pending unit attentions.
9216          * Pending sense gets returned first, then pending unit attentions.
9217          */
9218         mtx_lock(&lun->lun_lock);
9219         ps = lun->pending_sense[initidx / CTL_MAX_INIT_PER_PORT];
9220         if (ps != NULL)
9221                 ps += initidx % CTL_MAX_INIT_PER_PORT;
9222         if (ps != NULL && ps->error_code != 0) {
9223                 scsi_sense_data_type stored_format;
9224
9225                 /*
9226                  * Check to see which sense format was used for the stored
9227                  * sense data.
9228                  */
9229                 stored_format = scsi_sense_type(ps);
9230
9231                 /*
9232                  * If the user requested a different sense format than the
9233                  * one we stored, then we need to convert it to the other
9234                  * format.  If we're going from descriptor to fixed format
9235                  * sense data, we may lose things in translation, depending
9236                  * on what options were used.
9237                  *
9238                  * If the stored format is SSD_TYPE_NONE (i.e. invalid),
9239                  * for some reason we'll just copy it out as-is.
9240                  */
9241                 if ((stored_format == SSD_TYPE_FIXED)
9242                  && (sense_format == SSD_TYPE_DESC))
9243                         ctl_sense_to_desc((struct scsi_sense_data_fixed *)
9244                             ps, (struct scsi_sense_data_desc *)sense_ptr);
9245                 else if ((stored_format == SSD_TYPE_DESC)
9246                       && (sense_format == SSD_TYPE_FIXED))
9247                         ctl_sense_to_fixed((struct scsi_sense_data_desc *)
9248                             ps, (struct scsi_sense_data_fixed *)sense_ptr);
9249                 else
9250                         memcpy(sense_ptr, ps, sizeof(*sense_ptr));
9251
9252                 ps->error_code = 0;
9253                 have_error = 1;
9254         } else {
9255                 ua_type = ctl_build_ua(lun, initidx, sense_ptr, &sense_len,
9256                     sense_format);
9257                 if (ua_type != CTL_UA_NONE)
9258                         have_error = 1;
9259         }
9260         if (have_error == 0) {
9261                 /*
9262                  * Report informational exception if have one and allowed.
9263                  */
9264                 if (lun->MODE_IE.mrie != SIEP_MRIE_NO) {
9265                         asc = lun->ie_asc;
9266                         ascq = lun->ie_ascq;
9267                 }
9268                 ctl_set_sense_data(sense_ptr, &sense_len, lun, sense_format,
9269                     /*current_error*/ 1,
9270                     /*sense_key*/ SSD_KEY_NO_SENSE,
9271                     /*asc*/ asc,
9272                     /*ascq*/ ascq,
9273                     SSD_ELEM_NONE);
9274         }
9275         mtx_unlock(&lun->lun_lock);
9276
9277 send:
9278         /*
9279          * We report the SCSI status as OK, since the status of the command
9280          * itself is OK.  We're reporting sense as parameter data.
9281          */
9282         ctl_set_success(ctsio);
9283         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9284         ctsio->be_move_done = ctl_config_move_done;
9285         ctl_datamove((union ctl_io *)ctsio);
9286         return (CTL_RETVAL_COMPLETE);
9287 }
9288
9289 int
9290 ctl_tur(struct ctl_scsiio *ctsio)
9291 {
9292
9293         CTL_DEBUG_PRINT(("ctl_tur\n"));
9294
9295         ctl_set_success(ctsio);
9296         ctl_done((union ctl_io *)ctsio);
9297
9298         return (CTL_RETVAL_COMPLETE);
9299 }
9300
9301 /*
9302  * SCSI VPD page 0x00, the Supported VPD Pages page.
9303  */
9304 static int
9305 ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
9306 {
9307         struct ctl_lun *lun = CTL_LUN(ctsio);
9308         struct scsi_vpd_supported_pages *pages;
9309         int sup_page_size;
9310         int p;
9311
9312         sup_page_size = sizeof(struct scsi_vpd_supported_pages) *
9313             SCSI_EVPD_NUM_SUPPORTED_PAGES;
9314         ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO);
9315         pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
9316         ctsio->kern_rel_offset = 0;
9317         ctsio->kern_sg_entries = 0;
9318         ctsio->kern_data_len = min(sup_page_size, alloc_len);
9319         ctsio->kern_total_len = ctsio->kern_data_len;
9320
9321         /*
9322          * The control device is always connected.  The disk device, on the
9323          * other hand, may not be online all the time.  Need to change this
9324          * to figure out whether the disk device is actually online or not.
9325          */
9326         if (lun != NULL)
9327                 pages->device = (SID_QUAL_LU_CONNECTED << 5) |
9328                                 lun->be_lun->lun_type;
9329         else
9330                 pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9331
9332         p = 0;
9333         /* Supported VPD pages */
9334         pages->page_list[p++] = SVPD_SUPPORTED_PAGES;
9335         /* Serial Number */
9336         pages->page_list[p++] = SVPD_UNIT_SERIAL_NUMBER;
9337         /* Device Identification */
9338         pages->page_list[p++] = SVPD_DEVICE_ID;
9339         /* Extended INQUIRY Data */
9340         pages->page_list[p++] = SVPD_EXTENDED_INQUIRY_DATA;
9341         /* Mode Page Policy */
9342         pages->page_list[p++] = SVPD_MODE_PAGE_POLICY;
9343         /* SCSI Ports */
9344         pages->page_list[p++] = SVPD_SCSI_PORTS;
9345         /* Third-party Copy */
9346         pages->page_list[p++] = SVPD_SCSI_TPC;
9347         if (lun != NULL && lun->be_lun->lun_type == T_DIRECT) {
9348                 /* Block limits */
9349                 pages->page_list[p++] = SVPD_BLOCK_LIMITS;
9350                 /* Block Device Characteristics */
9351                 pages->page_list[p++] = SVPD_BDC;
9352                 /* Logical Block Provisioning */
9353                 pages->page_list[p++] = SVPD_LBP;
9354         }
9355         pages->length = p;
9356
9357         ctl_set_success(ctsio);
9358         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9359         ctsio->be_move_done = ctl_config_move_done;
9360         ctl_datamove((union ctl_io *)ctsio);
9361         return (CTL_RETVAL_COMPLETE);
9362 }
9363
9364 /*
9365  * SCSI VPD page 0x80, the Unit Serial Number page.
9366  */
9367 static int
9368 ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
9369 {
9370         struct ctl_lun *lun = CTL_LUN(ctsio);
9371         struct scsi_vpd_unit_serial_number *sn_ptr;
9372         int data_len;
9373
9374         data_len = 4 + CTL_SN_LEN;
9375         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9376         sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
9377         ctsio->kern_rel_offset = 0;
9378         ctsio->kern_sg_entries = 0;
9379         ctsio->kern_data_len = min(data_len, alloc_len);
9380         ctsio->kern_total_len = ctsio->kern_data_len;
9381
9382         /*
9383          * The control device is always connected.  The disk device, on the
9384          * other hand, may not be online all the time.  Need to change this
9385          * to figure out whether the disk device is actually online or not.
9386          */
9387         if (lun != NULL)
9388                 sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9389                                   lun->be_lun->lun_type;
9390         else
9391                 sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9392
9393         sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9394         sn_ptr->length = CTL_SN_LEN;
9395         /*
9396          * If we don't have a LUN, we just leave the serial number as
9397          * all spaces.
9398          */
9399         if (lun != NULL) {
9400                 strncpy((char *)sn_ptr->serial_num,
9401                         (char *)lun->be_lun->serial_num, CTL_SN_LEN);
9402         } else
9403                 memset(sn_ptr->serial_num, 0x20, CTL_SN_LEN);
9404
9405         ctl_set_success(ctsio);
9406         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9407         ctsio->be_move_done = ctl_config_move_done;
9408         ctl_datamove((union ctl_io *)ctsio);
9409         return (CTL_RETVAL_COMPLETE);
9410 }
9411
9412
9413 /*
9414  * SCSI VPD page 0x86, the Extended INQUIRY Data page.
9415  */
9416 static int
9417 ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len)
9418 {
9419         struct ctl_lun *lun = CTL_LUN(ctsio);
9420         struct scsi_vpd_extended_inquiry_data *eid_ptr;
9421         int data_len;
9422
9423         data_len = sizeof(struct scsi_vpd_extended_inquiry_data);
9424         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9425         eid_ptr = (struct scsi_vpd_extended_inquiry_data *)ctsio->kern_data_ptr;
9426         ctsio->kern_sg_entries = 0;
9427         ctsio->kern_rel_offset = 0;
9428         ctsio->kern_data_len = min(data_len, alloc_len);
9429         ctsio->kern_total_len = ctsio->kern_data_len;
9430
9431         /*
9432          * The control device is always connected.  The disk device, on the
9433          * other hand, may not be online all the time.
9434          */
9435         if (lun != NULL)
9436                 eid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9437                                      lun->be_lun->lun_type;
9438         else
9439                 eid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9440         eid_ptr->page_code = SVPD_EXTENDED_INQUIRY_DATA;
9441         scsi_ulto2b(data_len - 4, eid_ptr->page_length);
9442         /*
9443          * We support head of queue, ordered and simple tags.
9444          */
9445         eid_ptr->flags2 = SVPD_EID_HEADSUP | SVPD_EID_ORDSUP | SVPD_EID_SIMPSUP;
9446         /*
9447          * Volatile cache supported.
9448          */
9449         eid_ptr->flags3 = SVPD_EID_V_SUP;
9450
9451         /*
9452          * This means that we clear the REPORTED LUNS DATA HAS CHANGED unit
9453          * attention for a particular IT nexus on all LUNs once we report
9454          * it to that nexus once.  This bit is required as of SPC-4.
9455          */
9456         eid_ptr->flags4 = SVPD_EID_LUICLR;
9457
9458         /*
9459          * We support revert to defaults (RTD) bit in MODE SELECT.
9460          */
9461         eid_ptr->flags5 = SVPD_EID_RTD_SUP;
9462
9463         /*
9464          * XXX KDM in order to correctly answer this, we would need
9465          * information from the SIM to determine how much sense data it
9466          * can send.  So this would really be a path inquiry field, most
9467          * likely.  This can be set to a maximum of 252 according to SPC-4,
9468          * but the hardware may or may not be able to support that much.
9469          * 0 just means that the maximum sense data length is not reported.
9470          */
9471         eid_ptr->max_sense_length = 0;
9472
9473         ctl_set_success(ctsio);
9474         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9475         ctsio->be_move_done = ctl_config_move_done;
9476         ctl_datamove((union ctl_io *)ctsio);
9477         return (CTL_RETVAL_COMPLETE);
9478 }
9479
9480 static int
9481 ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len)
9482 {
9483         struct ctl_lun *lun = CTL_LUN(ctsio);
9484         struct scsi_vpd_mode_page_policy *mpp_ptr;
9485         int data_len;
9486
9487         data_len = sizeof(struct scsi_vpd_mode_page_policy) +
9488             sizeof(struct scsi_vpd_mode_page_policy_descr);
9489
9490         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9491         mpp_ptr = (struct scsi_vpd_mode_page_policy *)ctsio->kern_data_ptr;
9492         ctsio->kern_rel_offset = 0;
9493         ctsio->kern_sg_entries = 0;
9494         ctsio->kern_data_len = min(data_len, alloc_len);
9495         ctsio->kern_total_len = ctsio->kern_data_len;
9496
9497         /*
9498          * The control device is always connected.  The disk device, on the
9499          * other hand, may not be online all the time.
9500          */
9501         if (lun != NULL)
9502                 mpp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9503                                      lun->be_lun->lun_type;
9504         else
9505                 mpp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9506         mpp_ptr->page_code = SVPD_MODE_PAGE_POLICY;
9507         scsi_ulto2b(data_len - 4, mpp_ptr->page_length);
9508         mpp_ptr->descr[0].page_code = 0x3f;
9509         mpp_ptr->descr[0].subpage_code = 0xff;
9510         mpp_ptr->descr[0].policy = SVPD_MPP_SHARED;
9511
9512         ctl_set_success(ctsio);
9513         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9514         ctsio->be_move_done = ctl_config_move_done;
9515         ctl_datamove((union ctl_io *)ctsio);
9516         return (CTL_RETVAL_COMPLETE);
9517 }
9518
9519 /*
9520  * SCSI VPD page 0x83, the Device Identification page.
9521  */
9522 static int
9523 ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
9524 {
9525         struct ctl_softc *softc = CTL_SOFTC(ctsio);
9526         struct ctl_port *port = CTL_PORT(ctsio);
9527         struct ctl_lun *lun = CTL_LUN(ctsio);
9528         struct scsi_vpd_device_id *devid_ptr;
9529         struct scsi_vpd_id_descriptor *desc;
9530         int data_len, g;
9531         uint8_t proto;
9532
9533         data_len = sizeof(struct scsi_vpd_device_id) +
9534             sizeof(struct scsi_vpd_id_descriptor) +
9535                 sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
9536             sizeof(struct scsi_vpd_id_descriptor) +
9537                 sizeof(struct scsi_vpd_id_trgt_port_grp_id);
9538         if (lun && lun->lun_devid)
9539                 data_len += lun->lun_devid->len;
9540         if (port && port->port_devid)
9541                 data_len += port->port_devid->len;
9542         if (port && port->target_devid)
9543                 data_len += port->target_devid->len;
9544
9545         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9546         devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
9547         ctsio->kern_sg_entries = 0;
9548         ctsio->kern_rel_offset = 0;
9549         ctsio->kern_sg_entries = 0;
9550         ctsio->kern_data_len = min(data_len, alloc_len);
9551         ctsio->kern_total_len = ctsio->kern_data_len;
9552
9553         /*
9554          * The control device is always connected.  The disk device, on the
9555          * other hand, may not be online all the time.
9556          */
9557         if (lun != NULL)
9558                 devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9559                                      lun->be_lun->lun_type;
9560         else
9561                 devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9562         devid_ptr->page_code = SVPD_DEVICE_ID;
9563         scsi_ulto2b(data_len - 4, devid_ptr->length);
9564
9565         if (port && port->port_type == CTL_PORT_FC)
9566                 proto = SCSI_PROTO_FC << 4;
9567         else if (port && port->port_type == CTL_PORT_SAS)
9568                 proto = SCSI_PROTO_SAS << 4;
9569         else if (port && port->port_type == CTL_PORT_ISCSI)
9570                 proto = SCSI_PROTO_ISCSI << 4;
9571         else
9572                 proto = SCSI_PROTO_SPI << 4;
9573         desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
9574
9575         /*
9576          * We're using a LUN association here.  i.e., this device ID is a
9577          * per-LUN identifier.
9578          */
9579         if (lun && lun->lun_devid) {
9580                 memcpy(desc, lun->lun_devid->data, lun->lun_devid->len);
9581                 desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9582                     lun->lun_devid->len);
9583         }
9584
9585         /*
9586          * This is for the WWPN which is a port association.
9587          */
9588         if (port && port->port_devid) {
9589                 memcpy(desc, port->port_devid->data, port->port_devid->len);
9590                 desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9591                     port->port_devid->len);
9592         }
9593
9594         /*
9595          * This is for the Relative Target Port(type 4h) identifier
9596          */
9597         desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9598         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9599             SVPD_ID_TYPE_RELTARG;
9600         desc->length = 4;
9601         scsi_ulto2b(ctsio->io_hdr.nexus.targ_port, &desc->identifier[2]);
9602         desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9603             sizeof(struct scsi_vpd_id_rel_trgt_port_id));
9604
9605         /*
9606          * This is for the Target Port Group(type 5h) identifier
9607          */
9608         desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9609         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9610             SVPD_ID_TYPE_TPORTGRP;
9611         desc->length = 4;
9612         if (softc->is_single ||
9613             (port && port->status & CTL_PORT_STATUS_HA_SHARED))
9614                 g = 1;
9615         else
9616                 g = 2 + ctsio->io_hdr.nexus.targ_port / softc->port_cnt;
9617         scsi_ulto2b(g, &desc->identifier[2]);
9618         desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9619             sizeof(struct scsi_vpd_id_trgt_port_grp_id));
9620
9621         /*
9622          * This is for the Target identifier
9623          */
9624         if (port && port->target_devid) {
9625                 memcpy(desc, port->target_devid->data, port->target_devid->len);
9626         }
9627
9628         ctl_set_success(ctsio);
9629         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9630         ctsio->be_move_done = ctl_config_move_done;
9631         ctl_datamove((union ctl_io *)ctsio);
9632         return (CTL_RETVAL_COMPLETE);
9633 }
9634
9635 static int
9636 ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio, int alloc_len)
9637 {
9638         struct ctl_softc *softc = CTL_SOFTC(ctsio);
9639         struct ctl_lun *lun = CTL_LUN(ctsio);
9640         struct scsi_vpd_scsi_ports *sp;
9641         struct scsi_vpd_port_designation *pd;
9642         struct scsi_vpd_port_designation_cont *pdc;
9643         struct ctl_port *port;
9644         int data_len, num_target_ports, iid_len, id_len;
9645
9646         num_target_ports = 0;
9647         iid_len = 0;
9648         id_len = 0;
9649         mtx_lock(&softc->ctl_lock);
9650         STAILQ_FOREACH(port, &softc->port_list, links) {
9651                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
9652                         continue;
9653                 if (lun != NULL &&
9654                     ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
9655                         continue;
9656                 num_target_ports++;
9657                 if (port->init_devid)
9658                         iid_len += port->init_devid->len;
9659                 if (port->port_devid)
9660                         id_len += port->port_devid->len;
9661         }
9662         mtx_unlock(&softc->ctl_lock);
9663
9664         data_len = sizeof(struct scsi_vpd_scsi_ports) +
9665             num_target_ports * (sizeof(struct scsi_vpd_port_designation) +
9666              sizeof(struct scsi_vpd_port_designation_cont)) + iid_len + id_len;
9667         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9668         sp = (struct scsi_vpd_scsi_ports *)ctsio->kern_data_ptr;
9669         ctsio->kern_sg_entries = 0;
9670         ctsio->kern_rel_offset = 0;
9671         ctsio->kern_sg_entries = 0;
9672         ctsio->kern_data_len = min(data_len, alloc_len);
9673         ctsio->kern_total_len = ctsio->kern_data_len;
9674
9675         /*
9676          * The control device is always connected.  The disk device, on the
9677          * other hand, may not be online all the time.  Need to change this
9678          * to figure out whether the disk device is actually online or not.
9679          */
9680         if (lun != NULL)
9681                 sp->device = (SID_QUAL_LU_CONNECTED << 5) |
9682                                   lun->be_lun->lun_type;
9683         else
9684                 sp->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9685
9686         sp->page_code = SVPD_SCSI_PORTS;
9687         scsi_ulto2b(data_len - sizeof(struct scsi_vpd_scsi_ports),
9688             sp->page_length);
9689         pd = &sp->design[0];
9690
9691         mtx_lock(&softc->ctl_lock);
9692         STAILQ_FOREACH(port, &softc->port_list, links) {
9693                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
9694                         continue;
9695                 if (lun != NULL &&
9696                     ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
9697                         continue;
9698                 scsi_ulto2b(port->targ_port, pd->relative_port_id);
9699                 if (port->init_devid) {
9700                         iid_len = port->init_devid->len;
9701                         memcpy(pd->initiator_transportid,
9702                             port->init_devid->data, port->init_devid->len);
9703                 } else
9704                         iid_len = 0;
9705                 scsi_ulto2b(iid_len, pd->initiator_transportid_length);
9706                 pdc = (struct scsi_vpd_port_designation_cont *)
9707                     (&pd->initiator_transportid[iid_len]);
9708                 if (port->port_devid) {
9709                         id_len = port->port_devid->len;
9710                         memcpy(pdc->target_port_descriptors,
9711                             port->port_devid->data, port->port_devid->len);
9712                 } else
9713                         id_len = 0;
9714                 scsi_ulto2b(id_len, pdc->target_port_descriptors_length);
9715                 pd = (struct scsi_vpd_port_designation *)
9716                     ((uint8_t *)pdc->target_port_descriptors + id_len);
9717         }
9718         mtx_unlock(&softc->ctl_lock);
9719
9720         ctl_set_success(ctsio);
9721         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9722         ctsio->be_move_done = ctl_config_move_done;
9723         ctl_datamove((union ctl_io *)ctsio);
9724         return (CTL_RETVAL_COMPLETE);
9725 }
9726
9727 static int
9728 ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio, int alloc_len)
9729 {
9730         struct ctl_lun *lun = CTL_LUN(ctsio);
9731         struct scsi_vpd_block_limits *bl_ptr;
9732         const char *val;
9733         uint64_t ival;
9734
9735         ctsio->kern_data_ptr = malloc(sizeof(*bl_ptr), M_CTL, M_WAITOK | M_ZERO);
9736         bl_ptr = (struct scsi_vpd_block_limits *)ctsio->kern_data_ptr;
9737         ctsio->kern_sg_entries = 0;
9738         ctsio->kern_rel_offset = 0;
9739         ctsio->kern_sg_entries = 0;
9740         ctsio->kern_data_len = min(sizeof(*bl_ptr), alloc_len);
9741         ctsio->kern_total_len = ctsio->kern_data_len;
9742
9743         /*
9744          * The control device is always connected.  The disk device, on the
9745          * other hand, may not be online all the time.  Need to change this
9746          * to figure out whether the disk device is actually online or not.
9747          */
9748         if (lun != NULL)
9749                 bl_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9750                                   lun->be_lun->lun_type;
9751         else
9752                 bl_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9753
9754         bl_ptr->page_code = SVPD_BLOCK_LIMITS;
9755         scsi_ulto2b(sizeof(*bl_ptr) - 4, bl_ptr->page_length);
9756         bl_ptr->max_cmp_write_len = 0xff;
9757         scsi_ulto4b(0xffffffff, bl_ptr->max_txfer_len);
9758         if (lun != NULL) {
9759                 scsi_ulto4b(lun->be_lun->opttxferlen, bl_ptr->opt_txfer_len);
9760                 if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
9761                         ival = 0xffffffff;
9762                         val = dnvlist_get_string(lun->be_lun->options,
9763                             "unmap_max_lba", NULL);
9764                         if (val != NULL)
9765                                 ctl_expand_number(val, &ival);
9766                         scsi_ulto4b(ival, bl_ptr->max_unmap_lba_cnt);
9767                         ival = 0xffffffff;
9768                         val = dnvlist_get_string(lun->be_lun->options,
9769                             "unmap_max_descr", NULL);
9770                         if (val != NULL)
9771                                 ctl_expand_number(val, &ival);
9772                         scsi_ulto4b(ival, bl_ptr->max_unmap_blk_cnt);
9773                         if (lun->be_lun->ublockexp != 0) {
9774                                 scsi_ulto4b((1 << lun->be_lun->ublockexp),
9775                                     bl_ptr->opt_unmap_grain);
9776                                 scsi_ulto4b(0x80000000 | lun->be_lun->ublockoff,
9777                                     bl_ptr->unmap_grain_align);
9778                         }
9779                 }
9780                 scsi_ulto4b(lun->be_lun->atomicblock,
9781                     bl_ptr->max_atomic_transfer_length);
9782                 scsi_ulto4b(0, bl_ptr->atomic_alignment);
9783                 scsi_ulto4b(0, bl_ptr->atomic_transfer_length_granularity);
9784                 scsi_ulto4b(0, bl_ptr->max_atomic_transfer_length_with_atomic_boundary);
9785                 scsi_ulto4b(0, bl_ptr->max_atomic_boundary_size);
9786                 ival = UINT64_MAX;
9787                 val = dnvlist_get_string(lun->be_lun->options,
9788                     "write_same_max_lba", NULL);
9789                 if (val != NULL)
9790                         ctl_expand_number(val, &ival);
9791                 scsi_u64to8b(ival, bl_ptr->max_write_same_length);
9792         }
9793
9794         ctl_set_success(ctsio);
9795         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9796         ctsio->be_move_done = ctl_config_move_done;
9797         ctl_datamove((union ctl_io *)ctsio);
9798         return (CTL_RETVAL_COMPLETE);
9799 }
9800
9801 static int
9802 ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len)
9803 {
9804         struct ctl_lun *lun = CTL_LUN(ctsio);
9805         struct scsi_vpd_block_device_characteristics *bdc_ptr;
9806         const char *value;
9807         u_int i;
9808
9809         ctsio->kern_data_ptr = malloc(sizeof(*bdc_ptr), M_CTL, M_WAITOK | M_ZERO);
9810         bdc_ptr = (struct scsi_vpd_block_device_characteristics *)ctsio->kern_data_ptr;
9811         ctsio->kern_sg_entries = 0;
9812         ctsio->kern_rel_offset = 0;
9813         ctsio->kern_data_len = min(sizeof(*bdc_ptr), alloc_len);
9814         ctsio->kern_total_len = ctsio->kern_data_len;
9815
9816         /*
9817          * The control device is always connected.  The disk device, on the
9818          * other hand, may not be online all the time.  Need to change this
9819          * to figure out whether the disk device is actually online or not.
9820          */
9821         if (lun != NULL)
9822                 bdc_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9823                                   lun->be_lun->lun_type;
9824         else
9825                 bdc_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9826         bdc_ptr->page_code = SVPD_BDC;
9827         scsi_ulto2b(sizeof(*bdc_ptr) - 4, bdc_ptr->page_length);
9828         if (lun != NULL &&
9829             (value = dnvlist_get_string(lun->be_lun->options, "rpm", NULL)) != NULL)
9830                 i = strtol(value, NULL, 0);
9831         else
9832                 i = CTL_DEFAULT_ROTATION_RATE;
9833         scsi_ulto2b(i, bdc_ptr->medium_rotation_rate);
9834         if (lun != NULL &&
9835             (value = dnvlist_get_string(lun->be_lun->options, "formfactor", NULL)) != NULL)
9836                 i = strtol(value, NULL, 0);
9837         else
9838                 i = 0;
9839         bdc_ptr->wab_wac_ff = (i & 0x0f);
9840         bdc_ptr->flags = SVPD_FUAB | SVPD_VBULS;
9841
9842         ctl_set_success(ctsio);
9843         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9844         ctsio->be_move_done = ctl_config_move_done;
9845         ctl_datamove((union ctl_io *)ctsio);
9846         return (CTL_RETVAL_COMPLETE);
9847 }
9848
9849 static int
9850 ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len)
9851 {
9852         struct ctl_lun *lun = CTL_LUN(ctsio);
9853         struct scsi_vpd_logical_block_prov *lbp_ptr;
9854         const char *value;
9855
9856         ctsio->kern_data_ptr = malloc(sizeof(*lbp_ptr), M_CTL, M_WAITOK | M_ZERO);
9857         lbp_ptr = (struct scsi_vpd_logical_block_prov *)ctsio->kern_data_ptr;
9858         ctsio->kern_sg_entries = 0;
9859         ctsio->kern_rel_offset = 0;
9860         ctsio->kern_data_len = min(sizeof(*lbp_ptr), alloc_len);
9861         ctsio->kern_total_len = ctsio->kern_data_len;
9862
9863         /*
9864          * The control device is always connected.  The disk device, on the
9865          * other hand, may not be online all the time.  Need to change this
9866          * to figure out whether the disk device is actually online or not.
9867          */
9868         if (lun != NULL)
9869                 lbp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9870                                   lun->be_lun->lun_type;
9871         else
9872                 lbp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9873
9874         lbp_ptr->page_code = SVPD_LBP;
9875         scsi_ulto2b(sizeof(*lbp_ptr) - 4, lbp_ptr->page_length);
9876         lbp_ptr->threshold_exponent = CTL_LBP_EXPONENT;
9877         if (lun != NULL && lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
9878                 lbp_ptr->flags = SVPD_LBP_UNMAP | SVPD_LBP_WS16 |
9879                     SVPD_LBP_WS10 | SVPD_LBP_RZ | SVPD_LBP_ANC_SUP;
9880                 value = dnvlist_get_string(lun->be_lun->options,
9881                     "provisioning_type", NULL);
9882                 if (value != NULL) {
9883                         if (strcmp(value, "resource") == 0)
9884                                 lbp_ptr->prov_type = SVPD_LBP_RESOURCE;
9885                         else if (strcmp(value, "thin") == 0)
9886                                 lbp_ptr->prov_type = SVPD_LBP_THIN;
9887                 } else
9888                         lbp_ptr->prov_type = SVPD_LBP_THIN;
9889         }
9890
9891         ctl_set_success(ctsio);
9892         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9893         ctsio->be_move_done = ctl_config_move_done;
9894         ctl_datamove((union ctl_io *)ctsio);
9895         return (CTL_RETVAL_COMPLETE);
9896 }
9897
9898 /*
9899  * INQUIRY with the EVPD bit set.
9900  */
9901 static int
9902 ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
9903 {
9904         struct ctl_lun *lun = CTL_LUN(ctsio);
9905         struct scsi_inquiry *cdb;
9906         int alloc_len, retval;
9907
9908         cdb = (struct scsi_inquiry *)ctsio->cdb;
9909         alloc_len = scsi_2btoul(cdb->length);
9910
9911         switch (cdb->page_code) {
9912         case SVPD_SUPPORTED_PAGES:
9913                 retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
9914                 break;
9915         case SVPD_UNIT_SERIAL_NUMBER:
9916                 retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
9917                 break;
9918         case SVPD_DEVICE_ID:
9919                 retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
9920                 break;
9921         case SVPD_EXTENDED_INQUIRY_DATA:
9922                 retval = ctl_inquiry_evpd_eid(ctsio, alloc_len);
9923                 break;
9924         case SVPD_MODE_PAGE_POLICY:
9925                 retval = ctl_inquiry_evpd_mpp(ctsio, alloc_len);
9926                 break;
9927         case SVPD_SCSI_PORTS:
9928                 retval = ctl_inquiry_evpd_scsi_ports(ctsio, alloc_len);
9929                 break;
9930         case SVPD_SCSI_TPC:
9931                 retval = ctl_inquiry_evpd_tpc(ctsio, alloc_len);
9932                 break;
9933         case SVPD_BLOCK_LIMITS:
9934                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
9935                         goto err;
9936                 retval = ctl_inquiry_evpd_block_limits(ctsio, alloc_len);
9937                 break;
9938         case SVPD_BDC:
9939                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
9940                         goto err;
9941                 retval = ctl_inquiry_evpd_bdc(ctsio, alloc_len);
9942                 break;
9943         case SVPD_LBP:
9944                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
9945                         goto err;
9946                 retval = ctl_inquiry_evpd_lbp(ctsio, alloc_len);
9947                 break;
9948         default:
9949 err:
9950                 ctl_set_invalid_field(ctsio,
9951                                       /*sks_valid*/ 1,
9952                                       /*command*/ 1,
9953                                       /*field*/ 2,
9954                                       /*bit_valid*/ 0,
9955                                       /*bit*/ 0);
9956                 ctl_done((union ctl_io *)ctsio);
9957                 retval = CTL_RETVAL_COMPLETE;
9958                 break;
9959         }
9960
9961         return (retval);
9962 }
9963
9964 /*
9965  * Standard INQUIRY data.
9966  */
9967 static int
9968 ctl_inquiry_std(struct ctl_scsiio *ctsio)
9969 {
9970         struct ctl_softc *softc = CTL_SOFTC(ctsio);
9971         struct ctl_port *port = CTL_PORT(ctsio);
9972         struct ctl_lun *lun = CTL_LUN(ctsio);
9973         struct scsi_inquiry_data *inq_ptr;
9974         struct scsi_inquiry *cdb;
9975         const char *val;
9976         uint32_t alloc_len, data_len;
9977         ctl_port_type port_type;
9978
9979         port_type = port->port_type;
9980         if (port_type == CTL_PORT_IOCTL || port_type == CTL_PORT_INTERNAL)
9981                 port_type = CTL_PORT_SCSI;
9982
9983         cdb = (struct scsi_inquiry *)ctsio->cdb;
9984         alloc_len = scsi_2btoul(cdb->length);
9985
9986         /*
9987          * We malloc the full inquiry data size here and fill it
9988          * in.  If the user only asks for less, we'll give him
9989          * that much.
9990          */
9991         data_len = offsetof(struct scsi_inquiry_data, vendor_specific1);
9992         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9993         inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
9994         ctsio->kern_sg_entries = 0;
9995         ctsio->kern_rel_offset = 0;
9996         ctsio->kern_data_len = min(data_len, alloc_len);
9997         ctsio->kern_total_len = ctsio->kern_data_len;
9998
9999         if (lun != NULL) {
10000                 if ((lun->flags & CTL_LUN_PRIMARY_SC) ||
10001                     softc->ha_link >= CTL_HA_LINK_UNKNOWN) {
10002                         inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10003                             lun->be_lun->lun_type;
10004                 } else {
10005                         inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) |
10006                             lun->be_lun->lun_type;
10007                 }
10008                 if (lun->flags & CTL_LUN_REMOVABLE)
10009                         inq_ptr->dev_qual2 |= SID_RMB;
10010         } else
10011                 inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE;
10012
10013         /* RMB in byte 2 is 0 */
10014         inq_ptr->version = SCSI_REV_SPC5;
10015
10016         /*
10017          * According to SAM-3, even if a device only supports a single
10018          * level of LUN addressing, it should still set the HISUP bit:
10019          *
10020          * 4.9.1 Logical unit numbers overview
10021          *
10022          * All logical unit number formats described in this standard are
10023          * hierarchical in structure even when only a single level in that
10024          * hierarchy is used. The HISUP bit shall be set to one in the
10025          * standard INQUIRY data (see SPC-2) when any logical unit number
10026          * format described in this standard is used.  Non-hierarchical
10027          * formats are outside the scope of this standard.
10028          *
10029          * Therefore we set the HiSup bit here.
10030          *
10031          * The response format is 2, per SPC-3.
10032          */
10033         inq_ptr->response_format = SID_HiSup | 2;
10034
10035         inq_ptr->additional_length = data_len -
10036             (offsetof(struct scsi_inquiry_data, additional_length) + 1);
10037         CTL_DEBUG_PRINT(("additional_length = %d\n",
10038                          inq_ptr->additional_length));
10039
10040         inq_ptr->spc3_flags = SPC3_SID_3PC | SPC3_SID_TPGS_IMPLICIT;
10041         if (port_type == CTL_PORT_SCSI)
10042                 inq_ptr->spc2_flags = SPC2_SID_ADDR16;
10043         inq_ptr->spc2_flags |= SPC2_SID_MultiP;
10044         inq_ptr->flags = SID_CmdQue;
10045         if (port_type == CTL_PORT_SCSI)
10046                 inq_ptr->flags |= SID_WBus16 | SID_Sync;
10047
10048         /*
10049          * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
10050          * We have 8 bytes for the vendor name, and 16 bytes for the device
10051          * name and 4 bytes for the revision.
10052          */
10053         if (lun == NULL || (val = dnvlist_get_string(lun->be_lun->options,
10054             "vendor", NULL)) == NULL) {
10055                 strncpy(inq_ptr->vendor, CTL_VENDOR, sizeof(inq_ptr->vendor));
10056         } else {
10057                 memset(inq_ptr->vendor, ' ', sizeof(inq_ptr->vendor));
10058                 strncpy(inq_ptr->vendor, val,
10059                     min(sizeof(inq_ptr->vendor), strlen(val)));
10060         }
10061         if (lun == NULL) {
10062                 strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10063                     sizeof(inq_ptr->product));
10064         } else if ((val = dnvlist_get_string(lun->be_lun->options, "product",
10065             NULL)) == NULL) {
10066                 switch (lun->be_lun->lun_type) {
10067                 case T_DIRECT:
10068                         strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10069                             sizeof(inq_ptr->product));
10070                         break;
10071                 case T_PROCESSOR:
10072                         strncpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT,
10073                             sizeof(inq_ptr->product));
10074                         break;
10075                 case T_CDROM:
10076                         strncpy(inq_ptr->product, CTL_CDROM_PRODUCT,
10077                             sizeof(inq_ptr->product));
10078                         break;
10079                 default:
10080                         strncpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT,
10081                             sizeof(inq_ptr->product));
10082                         break;
10083                 }
10084         } else {
10085                 memset(inq_ptr->product, ' ', sizeof(inq_ptr->product));
10086                 strncpy(inq_ptr->product, val,
10087                     min(sizeof(inq_ptr->product), strlen(val)));
10088         }
10089
10090         /*
10091          * XXX make this a macro somewhere so it automatically gets
10092          * incremented when we make changes.
10093          */
10094         if (lun == NULL || (val = dnvlist_get_string(lun->be_lun->options,
10095             "revision", NULL)) == NULL) {
10096                 strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
10097         } else {
10098                 memset(inq_ptr->revision, ' ', sizeof(inq_ptr->revision));
10099                 strncpy(inq_ptr->revision, val,
10100                     min(sizeof(inq_ptr->revision), strlen(val)));
10101         }
10102
10103         /*
10104          * For parallel SCSI, we support double transition and single
10105          * transition clocking.  We also support QAS (Quick Arbitration
10106          * and Selection) and Information Unit transfers on both the
10107          * control and array devices.
10108          */
10109         if (port_type == CTL_PORT_SCSI)
10110                 inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
10111                                     SID_SPI_IUS;
10112
10113         /* SAM-6 (no version claimed) */
10114         scsi_ulto2b(0x00C0, inq_ptr->version1);
10115         /* SPC-5 (no version claimed) */
10116         scsi_ulto2b(0x05C0, inq_ptr->version2);
10117         if (port_type == CTL_PORT_FC) {
10118                 /* FCP-2 ANSI INCITS.350:2003 */
10119                 scsi_ulto2b(0x0917, inq_ptr->version3);
10120         } else if (port_type == CTL_PORT_SCSI) {
10121                 /* SPI-4 ANSI INCITS.362:200x */
10122                 scsi_ulto2b(0x0B56, inq_ptr->version3);
10123         } else if (port_type == CTL_PORT_ISCSI) {
10124                 /* iSCSI (no version claimed) */
10125                 scsi_ulto2b(0x0960, inq_ptr->version3);
10126         } else if (port_type == CTL_PORT_SAS) {
10127                 /* SAS (no version claimed) */
10128                 scsi_ulto2b(0x0BE0, inq_ptr->version3);
10129         } else if (port_type == CTL_PORT_UMASS) {
10130                 /* USB Mass Storage Class Bulk-Only Transport, Revision 1.0 */
10131                 scsi_ulto2b(0x1730, inq_ptr->version3);
10132         }
10133
10134         if (lun == NULL) {
10135                 /* SBC-4 (no version claimed) */
10136                 scsi_ulto2b(0x0600, inq_ptr->version4);
10137         } else {
10138                 switch (lun->be_lun->lun_type) {
10139                 case T_DIRECT:
10140                         /* SBC-4 (no version claimed) */
10141                         scsi_ulto2b(0x0600, inq_ptr->version4);
10142                         break;
10143                 case T_PROCESSOR:
10144                         break;
10145                 case T_CDROM:
10146                         /* MMC-6 (no version claimed) */
10147                         scsi_ulto2b(0x04E0, inq_ptr->version4);
10148                         break;
10149                 default:
10150                         break;
10151                 }
10152         }
10153
10154         ctl_set_success(ctsio);
10155         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10156         ctsio->be_move_done = ctl_config_move_done;
10157         ctl_datamove((union ctl_io *)ctsio);
10158         return (CTL_RETVAL_COMPLETE);
10159 }
10160
10161 int
10162 ctl_inquiry(struct ctl_scsiio *ctsio)
10163 {
10164         struct scsi_inquiry *cdb;
10165         int retval;
10166
10167         CTL_DEBUG_PRINT(("ctl_inquiry\n"));
10168
10169         cdb = (struct scsi_inquiry *)ctsio->cdb;
10170         if (cdb->byte2 & SI_EVPD)
10171                 retval = ctl_inquiry_evpd(ctsio);
10172         else if (cdb->page_code == 0)
10173                 retval = ctl_inquiry_std(ctsio);
10174         else {
10175                 ctl_set_invalid_field(ctsio,
10176                                       /*sks_valid*/ 1,
10177                                       /*command*/ 1,
10178                                       /*field*/ 2,
10179                                       /*bit_valid*/ 0,
10180                                       /*bit*/ 0);
10181                 ctl_done((union ctl_io *)ctsio);
10182                 return (CTL_RETVAL_COMPLETE);
10183         }
10184
10185         return (retval);
10186 }
10187
10188 int
10189 ctl_get_config(struct ctl_scsiio *ctsio)
10190 {
10191         struct ctl_lun *lun = CTL_LUN(ctsio);
10192         struct scsi_get_config_header *hdr;
10193         struct scsi_get_config_feature *feature;
10194         struct scsi_get_config *cdb;
10195         uint32_t alloc_len, data_len;
10196         int rt, starting;
10197
10198         cdb = (struct scsi_get_config *)ctsio->cdb;
10199         rt = (cdb->rt & SGC_RT_MASK);
10200         starting = scsi_2btoul(cdb->starting_feature);
10201         alloc_len = scsi_2btoul(cdb->length);
10202
10203         data_len = sizeof(struct scsi_get_config_header) +
10204             sizeof(struct scsi_get_config_feature) + 8 +
10205             sizeof(struct scsi_get_config_feature) + 8 +
10206             sizeof(struct scsi_get_config_feature) + 4 +
10207             sizeof(struct scsi_get_config_feature) + 4 +
10208             sizeof(struct scsi_get_config_feature) + 8 +
10209             sizeof(struct scsi_get_config_feature) +
10210             sizeof(struct scsi_get_config_feature) + 4 +
10211             sizeof(struct scsi_get_config_feature) + 4 +
10212             sizeof(struct scsi_get_config_feature) + 4 +
10213             sizeof(struct scsi_get_config_feature) + 4 +
10214             sizeof(struct scsi_get_config_feature) + 4 +
10215             sizeof(struct scsi_get_config_feature) + 4;
10216         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10217         ctsio->kern_sg_entries = 0;
10218         ctsio->kern_rel_offset = 0;
10219
10220         hdr = (struct scsi_get_config_header *)ctsio->kern_data_ptr;
10221         if (lun->flags & CTL_LUN_NO_MEDIA)
10222                 scsi_ulto2b(0x0000, hdr->current_profile);
10223         else
10224                 scsi_ulto2b(0x0010, hdr->current_profile);
10225         feature = (struct scsi_get_config_feature *)(hdr + 1);
10226
10227         if (starting > 0x003b)
10228                 goto done;
10229         if (starting > 0x003a)
10230                 goto f3b;
10231         if (starting > 0x002b)
10232                 goto f3a;
10233         if (starting > 0x002a)
10234                 goto f2b;
10235         if (starting > 0x001f)
10236                 goto f2a;
10237         if (starting > 0x001e)
10238                 goto f1f;
10239         if (starting > 0x001d)
10240                 goto f1e;
10241         if (starting > 0x0010)
10242                 goto f1d;
10243         if (starting > 0x0003)
10244                 goto f10;
10245         if (starting > 0x0002)
10246                 goto f3;
10247         if (starting > 0x0001)
10248                 goto f2;
10249         if (starting > 0x0000)
10250                 goto f1;
10251
10252         /* Profile List */
10253         scsi_ulto2b(0x0000, feature->feature_code);
10254         feature->flags = SGC_F_PERSISTENT | SGC_F_CURRENT;
10255         feature->add_length = 8;
10256         scsi_ulto2b(0x0008, &feature->feature_data[0]); /* CD-ROM */
10257         feature->feature_data[2] = 0x00;
10258         scsi_ulto2b(0x0010, &feature->feature_data[4]); /* DVD-ROM */
10259         feature->feature_data[6] = 0x01;
10260         feature = (struct scsi_get_config_feature *)
10261             &feature->feature_data[feature->add_length];
10262
10263 f1:     /* Core */
10264         scsi_ulto2b(0x0001, feature->feature_code);
10265         feature->flags = 0x08 | SGC_F_PERSISTENT | SGC_F_CURRENT;
10266         feature->add_length = 8;
10267         scsi_ulto4b(0x00000000, &feature->feature_data[0]);
10268         feature->feature_data[4] = 0x03;
10269         feature = (struct scsi_get_config_feature *)
10270             &feature->feature_data[feature->add_length];
10271
10272 f2:     /* Morphing */
10273         scsi_ulto2b(0x0002, feature->feature_code);
10274         feature->flags = 0x04 | SGC_F_PERSISTENT | SGC_F_CURRENT;
10275         feature->add_length = 4;
10276         feature->feature_data[0] = 0x02;
10277         feature = (struct scsi_get_config_feature *)
10278             &feature->feature_data[feature->add_length];
10279
10280 f3:     /* Removable Medium */
10281         scsi_ulto2b(0x0003, feature->feature_code);
10282         feature->flags = 0x04 | SGC_F_PERSISTENT | SGC_F_CURRENT;
10283         feature->add_length = 4;
10284         feature->feature_data[0] = 0x39;
10285         feature = (struct scsi_get_config_feature *)
10286             &feature->feature_data[feature->add_length];
10287
10288         if (rt == SGC_RT_CURRENT && (lun->flags & CTL_LUN_NO_MEDIA))
10289                 goto done;
10290
10291 f10:    /* Random Read */
10292         scsi_ulto2b(0x0010, feature->feature_code);
10293         feature->flags = 0x00;
10294         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10295                 feature->flags |= SGC_F_CURRENT;
10296         feature->add_length = 8;
10297         scsi_ulto4b(lun->be_lun->blocksize, &feature->feature_data[0]);
10298         scsi_ulto2b(1, &feature->feature_data[4]);
10299         feature->feature_data[6] = 0x00;
10300         feature = (struct scsi_get_config_feature *)
10301             &feature->feature_data[feature->add_length];
10302
10303 f1d:    /* Multi-Read */
10304         scsi_ulto2b(0x001D, feature->feature_code);
10305         feature->flags = 0x00;
10306         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10307                 feature->flags |= SGC_F_CURRENT;
10308         feature->add_length = 0;
10309         feature = (struct scsi_get_config_feature *)
10310             &feature->feature_data[feature->add_length];
10311
10312 f1e:    /* CD Read */
10313         scsi_ulto2b(0x001E, feature->feature_code);
10314         feature->flags = 0x00;
10315         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10316                 feature->flags |= SGC_F_CURRENT;
10317         feature->add_length = 4;
10318         feature->feature_data[0] = 0x00;
10319         feature = (struct scsi_get_config_feature *)
10320             &feature->feature_data[feature->add_length];
10321
10322 f1f:    /* DVD Read */
10323         scsi_ulto2b(0x001F, feature->feature_code);
10324         feature->flags = 0x08;
10325         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10326                 feature->flags |= SGC_F_CURRENT;
10327         feature->add_length = 4;
10328         feature->feature_data[0] = 0x01;
10329         feature->feature_data[2] = 0x03;
10330         feature = (struct scsi_get_config_feature *)
10331             &feature->feature_data[feature->add_length];
10332
10333 f2a:    /* DVD+RW */
10334         scsi_ulto2b(0x002A, feature->feature_code);
10335         feature->flags = 0x04;
10336         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10337                 feature->flags |= SGC_F_CURRENT;
10338         feature->add_length = 4;
10339         feature->feature_data[0] = 0x00;
10340         feature->feature_data[1] = 0x00;
10341         feature = (struct scsi_get_config_feature *)
10342             &feature->feature_data[feature->add_length];
10343
10344 f2b:    /* DVD+R */
10345         scsi_ulto2b(0x002B, feature->feature_code);
10346         feature->flags = 0x00;
10347         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10348                 feature->flags |= SGC_F_CURRENT;
10349         feature->add_length = 4;
10350         feature->feature_data[0] = 0x00;
10351         feature = (struct scsi_get_config_feature *)
10352             &feature->feature_data[feature->add_length];
10353
10354 f3a:    /* DVD+RW Dual Layer */
10355         scsi_ulto2b(0x003A, feature->feature_code);
10356         feature->flags = 0x00;
10357         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10358                 feature->flags |= SGC_F_CURRENT;
10359         feature->add_length = 4;
10360         feature->feature_data[0] = 0x00;
10361         feature->feature_data[1] = 0x00;
10362         feature = (struct scsi_get_config_feature *)
10363             &feature->feature_data[feature->add_length];
10364
10365 f3b:    /* DVD+R Dual Layer */
10366         scsi_ulto2b(0x003B, feature->feature_code);
10367         feature->flags = 0x00;
10368         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10369                 feature->flags |= SGC_F_CURRENT;
10370         feature->add_length = 4;
10371         feature->feature_data[0] = 0x00;
10372         feature = (struct scsi_get_config_feature *)
10373             &feature->feature_data[feature->add_length];
10374
10375 done:
10376         data_len = (uint8_t *)feature - (uint8_t *)hdr;
10377         if (rt == SGC_RT_SPECIFIC && data_len > 4) {
10378                 feature = (struct scsi_get_config_feature *)(hdr + 1);
10379                 if (scsi_2btoul(feature->feature_code) == starting)
10380                         feature = (struct scsi_get_config_feature *)
10381                             &feature->feature_data[feature->add_length];
10382                 data_len = (uint8_t *)feature - (uint8_t *)hdr;
10383         }
10384         scsi_ulto4b(data_len - 4, hdr->data_length);
10385         ctsio->kern_data_len = min(data_len, alloc_len);
10386         ctsio->kern_total_len = ctsio->kern_data_len;
10387
10388         ctl_set_success(ctsio);
10389         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10390         ctsio->be_move_done = ctl_config_move_done;
10391         ctl_datamove((union ctl_io *)ctsio);
10392         return (CTL_RETVAL_COMPLETE);
10393 }
10394
10395 int
10396 ctl_get_event_status(struct ctl_scsiio *ctsio)
10397 {
10398         struct scsi_get_event_status_header *hdr;
10399         struct scsi_get_event_status *cdb;
10400         uint32_t alloc_len, data_len;
10401
10402         cdb = (struct scsi_get_event_status *)ctsio->cdb;
10403         if ((cdb->byte2 & SGESN_POLLED) == 0) {
10404                 ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 1,
10405                     /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 0);
10406                 ctl_done((union ctl_io *)ctsio);
10407                 return (CTL_RETVAL_COMPLETE);
10408         }
10409         alloc_len = scsi_2btoul(cdb->length);
10410
10411         data_len = sizeof(struct scsi_get_event_status_header);
10412         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10413         ctsio->kern_sg_entries = 0;
10414         ctsio->kern_rel_offset = 0;
10415         ctsio->kern_data_len = min(data_len, alloc_len);
10416         ctsio->kern_total_len = ctsio->kern_data_len;
10417
10418         hdr = (struct scsi_get_event_status_header *)ctsio->kern_data_ptr;
10419         scsi_ulto2b(0, hdr->descr_length);
10420         hdr->nea_class = SGESN_NEA;
10421         hdr->supported_class = 0;
10422
10423         ctl_set_success(ctsio);
10424         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10425         ctsio->be_move_done = ctl_config_move_done;
10426         ctl_datamove((union ctl_io *)ctsio);
10427         return (CTL_RETVAL_COMPLETE);
10428 }
10429
10430 int
10431 ctl_mechanism_status(struct ctl_scsiio *ctsio)
10432 {
10433         struct scsi_mechanism_status_header *hdr;
10434         struct scsi_mechanism_status *cdb;
10435         uint32_t alloc_len, data_len;
10436
10437         cdb = (struct scsi_mechanism_status *)ctsio->cdb;
10438         alloc_len = scsi_2btoul(cdb->length);
10439
10440         data_len = sizeof(struct scsi_mechanism_status_header);
10441         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10442         ctsio->kern_sg_entries = 0;
10443         ctsio->kern_rel_offset = 0;
10444         ctsio->kern_data_len = min(data_len, alloc_len);
10445         ctsio->kern_total_len = ctsio->kern_data_len;
10446
10447         hdr = (struct scsi_mechanism_status_header *)ctsio->kern_data_ptr;
10448         hdr->state1 = 0x00;
10449         hdr->state2 = 0xe0;
10450         scsi_ulto3b(0, hdr->lba);
10451         hdr->slots_num = 0;
10452         scsi_ulto2b(0, hdr->slots_length);
10453
10454         ctl_set_success(ctsio);
10455         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10456         ctsio->be_move_done = ctl_config_move_done;
10457         ctl_datamove((union ctl_io *)ctsio);
10458         return (CTL_RETVAL_COMPLETE);
10459 }
10460
10461 static void
10462 ctl_ultomsf(uint32_t lba, uint8_t *buf)
10463 {
10464
10465         lba += 150;
10466         buf[0] = 0;
10467         buf[1] = bin2bcd((lba / 75) / 60);
10468         buf[2] = bin2bcd((lba / 75) % 60);
10469         buf[3] = bin2bcd(lba % 75);
10470 }
10471
10472 int
10473 ctl_read_toc(struct ctl_scsiio *ctsio)
10474 {
10475         struct ctl_lun *lun = CTL_LUN(ctsio);
10476         struct scsi_read_toc_hdr *hdr;
10477         struct scsi_read_toc_type01_descr *descr;
10478         struct scsi_read_toc *cdb;
10479         uint32_t alloc_len, data_len;
10480         int format, msf;
10481
10482         cdb = (struct scsi_read_toc *)ctsio->cdb;
10483         msf = (cdb->byte2 & CD_MSF) != 0;
10484         format = cdb->format;
10485         alloc_len = scsi_2btoul(cdb->data_len);
10486
10487         data_len = sizeof(struct scsi_read_toc_hdr);
10488         if (format == 0)
10489                 data_len += 2 * sizeof(struct scsi_read_toc_type01_descr);
10490         else
10491                 data_len += sizeof(struct scsi_read_toc_type01_descr);
10492         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10493         ctsio->kern_sg_entries = 0;
10494         ctsio->kern_rel_offset = 0;
10495         ctsio->kern_data_len = min(data_len, alloc_len);
10496         ctsio->kern_total_len = ctsio->kern_data_len;
10497
10498         hdr = (struct scsi_read_toc_hdr *)ctsio->kern_data_ptr;
10499         if (format == 0) {
10500                 scsi_ulto2b(0x12, hdr->data_length);
10501                 hdr->first = 1;
10502                 hdr->last = 1;
10503                 descr = (struct scsi_read_toc_type01_descr *)(hdr + 1);
10504                 descr->addr_ctl = 0x14;
10505                 descr->track_number = 1;
10506                 if (msf)
10507                         ctl_ultomsf(0, descr->track_start);
10508                 else
10509                         scsi_ulto4b(0, descr->track_start);
10510                 descr++;
10511                 descr->addr_ctl = 0x14;
10512                 descr->track_number = 0xaa;
10513                 if (msf)
10514                         ctl_ultomsf(lun->be_lun->maxlba+1, descr->track_start);
10515                 else
10516                         scsi_ulto4b(lun->be_lun->maxlba+1, descr->track_start);
10517         } else {
10518                 scsi_ulto2b(0x0a, hdr->data_length);
10519                 hdr->first = 1;
10520                 hdr->last = 1;
10521                 descr = (struct scsi_read_toc_type01_descr *)(hdr + 1);
10522                 descr->addr_ctl = 0x14;
10523                 descr->track_number = 1;
10524                 if (msf)
10525                         ctl_ultomsf(0, descr->track_start);
10526                 else
10527                         scsi_ulto4b(0, descr->track_start);
10528         }
10529
10530         ctl_set_success(ctsio);
10531         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10532         ctsio->be_move_done = ctl_config_move_done;
10533         ctl_datamove((union ctl_io *)ctsio);
10534         return (CTL_RETVAL_COMPLETE);
10535 }
10536
10537 /*
10538  * For known CDB types, parse the LBA and length.
10539  */
10540 static int
10541 ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len)
10542 {
10543         if (io->io_hdr.io_type != CTL_IO_SCSI)
10544                 return (1);
10545
10546         switch (io->scsiio.cdb[0]) {
10547         case COMPARE_AND_WRITE: {
10548                 struct scsi_compare_and_write *cdb;
10549
10550                 cdb = (struct scsi_compare_and_write *)io->scsiio.cdb;
10551
10552                 *lba = scsi_8btou64(cdb->addr);
10553                 *len = cdb->length;
10554                 break;
10555         }
10556         case READ_6:
10557         case WRITE_6: {
10558                 struct scsi_rw_6 *cdb;
10559
10560                 cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
10561
10562                 *lba = scsi_3btoul(cdb->addr);
10563                 /* only 5 bits are valid in the most significant address byte */
10564                 *lba &= 0x1fffff;
10565                 *len = cdb->length;
10566                 break;
10567         }
10568         case READ_10:
10569         case WRITE_10: {
10570                 struct scsi_rw_10 *cdb;
10571
10572                 cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
10573
10574                 *lba = scsi_4btoul(cdb->addr);
10575                 *len = scsi_2btoul(cdb->length);
10576                 break;
10577         }
10578         case WRITE_VERIFY_10: {
10579                 struct scsi_write_verify_10 *cdb;
10580
10581                 cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
10582
10583                 *lba = scsi_4btoul(cdb->addr);
10584                 *len = scsi_2btoul(cdb->length);
10585                 break;
10586         }
10587         case READ_12:
10588         case WRITE_12: {
10589                 struct scsi_rw_12 *cdb;
10590
10591                 cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
10592
10593                 *lba = scsi_4btoul(cdb->addr);
10594                 *len = scsi_4btoul(cdb->length);
10595                 break;
10596         }
10597         case WRITE_VERIFY_12: {
10598                 struct scsi_write_verify_12 *cdb;
10599
10600                 cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
10601
10602                 *lba = scsi_4btoul(cdb->addr);
10603                 *len = scsi_4btoul(cdb->length);
10604                 break;
10605         }
10606         case READ_16:
10607         case WRITE_16: {
10608                 struct scsi_rw_16 *cdb;
10609
10610                 cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
10611
10612                 *lba = scsi_8btou64(cdb->addr);
10613                 *len = scsi_4btoul(cdb->length);
10614                 break;
10615         }
10616         case WRITE_ATOMIC_16: {
10617                 struct scsi_write_atomic_16 *cdb;
10618
10619                 cdb = (struct scsi_write_atomic_16 *)io->scsiio.cdb;
10620
10621                 *lba = scsi_8btou64(cdb->addr);
10622                 *len = scsi_2btoul(cdb->length);
10623                 break;
10624         }
10625         case WRITE_VERIFY_16: {
10626                 struct scsi_write_verify_16 *cdb;
10627
10628                 cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
10629
10630                 *lba = scsi_8btou64(cdb->addr);
10631                 *len = scsi_4btoul(cdb->length);
10632                 break;
10633         }
10634         case WRITE_SAME_10: {
10635                 struct scsi_write_same_10 *cdb;
10636
10637                 cdb = (struct scsi_write_same_10 *)io->scsiio.cdb;
10638
10639                 *lba = scsi_4btoul(cdb->addr);
10640                 *len = scsi_2btoul(cdb->length);
10641                 break;
10642         }
10643         case WRITE_SAME_16: {
10644                 struct scsi_write_same_16 *cdb;
10645
10646                 cdb = (struct scsi_write_same_16 *)io->scsiio.cdb;
10647
10648                 *lba = scsi_8btou64(cdb->addr);
10649                 *len = scsi_4btoul(cdb->length);
10650                 break;
10651         }
10652         case VERIFY_10: {
10653                 struct scsi_verify_10 *cdb;
10654
10655                 cdb = (struct scsi_verify_10 *)io->scsiio.cdb;
10656
10657                 *lba = scsi_4btoul(cdb->addr);
10658                 *len = scsi_2btoul(cdb->length);
10659                 break;
10660         }
10661         case VERIFY_12: {
10662                 struct scsi_verify_12 *cdb;
10663
10664                 cdb = (struct scsi_verify_12 *)io->scsiio.cdb;
10665
10666                 *lba = scsi_4btoul(cdb->addr);
10667                 *len = scsi_4btoul(cdb->length);
10668                 break;
10669         }
10670         case VERIFY_16: {
10671                 struct scsi_verify_16 *cdb;
10672
10673                 cdb = (struct scsi_verify_16 *)io->scsiio.cdb;
10674
10675                 *lba = scsi_8btou64(cdb->addr);
10676                 *len = scsi_4btoul(cdb->length);
10677                 break;
10678         }
10679         case UNMAP: {
10680                 *lba = 0;
10681                 *len = UINT64_MAX;
10682                 break;
10683         }
10684         case SERVICE_ACTION_IN: {       /* GET LBA STATUS */
10685                 struct scsi_get_lba_status *cdb;
10686
10687                 cdb = (struct scsi_get_lba_status *)io->scsiio.cdb;
10688                 *lba = scsi_8btou64(cdb->addr);
10689                 *len = UINT32_MAX;
10690                 break;
10691         }
10692         default:
10693                 return (1);
10694                 break; /* NOTREACHED */
10695         }
10696
10697         return (0);
10698 }
10699
10700 static ctl_action
10701 ctl_extent_check_lba(uint64_t lba1, uint64_t len1, uint64_t lba2, uint64_t len2,
10702     bool seq)
10703 {
10704         uint64_t endlba1, endlba2;
10705
10706         endlba1 = lba1 + len1 - (seq ? 0 : 1);
10707         endlba2 = lba2 + len2 - 1;
10708
10709         if ((endlba1 < lba2) || (endlba2 < lba1))
10710                 return (CTL_ACTION_PASS);
10711         else
10712                 return (CTL_ACTION_BLOCK);
10713 }
10714
10715 static int
10716 ctl_extent_check_unmap(union ctl_io *io, uint64_t lba2, uint64_t len2)
10717 {
10718         struct ctl_ptr_len_flags *ptrlen;
10719         struct scsi_unmap_desc *buf, *end, *range;
10720         uint64_t lba;
10721         uint32_t len;
10722
10723         /* If not UNMAP -- go other way. */
10724         if (io->io_hdr.io_type != CTL_IO_SCSI ||
10725             io->scsiio.cdb[0] != UNMAP)
10726                 return (CTL_ACTION_ERROR);
10727
10728         /* If UNMAP without data -- block and wait for data. */
10729         ptrlen = (struct ctl_ptr_len_flags *)
10730             &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
10731         if ((io->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0 ||
10732             ptrlen->ptr == NULL)
10733                 return (CTL_ACTION_BLOCK);
10734
10735         /* UNMAP with data -- check for collision. */
10736         buf = (struct scsi_unmap_desc *)ptrlen->ptr;
10737         end = buf + ptrlen->len / sizeof(*buf);
10738         for (range = buf; range < end; range++) {
10739                 lba = scsi_8btou64(range->lba);
10740                 len = scsi_4btoul(range->length);
10741                 if ((lba < lba2 + len2) && (lba + len > lba2))
10742                         return (CTL_ACTION_BLOCK);
10743         }
10744         return (CTL_ACTION_PASS);
10745 }
10746
10747 static ctl_action
10748 ctl_extent_check(union ctl_io *io1, union ctl_io *io2, bool seq)
10749 {
10750         uint64_t lba1, lba2;
10751         uint64_t len1, len2;
10752         int retval;
10753
10754         if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
10755                 return (CTL_ACTION_ERROR);
10756
10757         retval = ctl_extent_check_unmap(io1, lba2, len2);
10758         if (retval != CTL_ACTION_ERROR)
10759                 return (retval);
10760
10761         if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
10762                 return (CTL_ACTION_ERROR);
10763
10764         if (io1->io_hdr.flags & CTL_FLAG_SERSEQ_DONE)
10765                 seq = FALSE;
10766         return (ctl_extent_check_lba(lba1, len1, lba2, len2, seq));
10767 }
10768
10769 static ctl_action
10770 ctl_extent_check_seq(union ctl_io *io1, union ctl_io *io2)
10771 {
10772         uint64_t lba1, lba2;
10773         uint64_t len1, len2;
10774
10775         if (io1->io_hdr.flags & CTL_FLAG_SERSEQ_DONE)
10776                 return (CTL_ACTION_PASS);
10777         if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
10778                 return (CTL_ACTION_ERROR);
10779         if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
10780                 return (CTL_ACTION_ERROR);
10781
10782         if (lba1 + len1 == lba2)
10783                 return (CTL_ACTION_BLOCK);
10784         return (CTL_ACTION_PASS);
10785 }
10786
10787 static ctl_action
10788 ctl_check_for_blockage(struct ctl_lun *lun, union ctl_io *pending_io,
10789     union ctl_io *ooa_io)
10790 {
10791         const struct ctl_cmd_entry *pending_entry, *ooa_entry;
10792         const ctl_serialize_action *serialize_row;
10793
10794         /*
10795          * The initiator attempted multiple untagged commands at the same
10796          * time.  Can't do that.
10797          */
10798         if ((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10799          && (ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10800          && ((pending_io->io_hdr.nexus.targ_port ==
10801               ooa_io->io_hdr.nexus.targ_port)
10802           && (pending_io->io_hdr.nexus.initid ==
10803               ooa_io->io_hdr.nexus.initid))
10804          && ((ooa_io->io_hdr.flags & (CTL_FLAG_ABORT |
10805               CTL_FLAG_STATUS_SENT)) == 0))
10806                 return (CTL_ACTION_OVERLAP);
10807
10808         /*
10809          * The initiator attempted to send multiple tagged commands with
10810          * the same ID.  (It's fine if different initiators have the same
10811          * tag ID.)
10812          *
10813          * Even if all of those conditions are true, we don't kill the I/O
10814          * if the command ahead of us has been aborted.  We won't end up
10815          * sending it to the FETD, and it's perfectly legal to resend a
10816          * command with the same tag number as long as the previous
10817          * instance of this tag number has been aborted somehow.
10818          */
10819         if ((pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10820          && (ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10821          && (pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
10822          && ((pending_io->io_hdr.nexus.targ_port ==
10823               ooa_io->io_hdr.nexus.targ_port)
10824           && (pending_io->io_hdr.nexus.initid ==
10825               ooa_io->io_hdr.nexus.initid))
10826          && ((ooa_io->io_hdr.flags & (CTL_FLAG_ABORT |
10827               CTL_FLAG_STATUS_SENT)) == 0))
10828                 return (CTL_ACTION_OVERLAP_TAG);
10829
10830         /*
10831          * If we get a head of queue tag, SAM-3 says that we should
10832          * immediately execute it.
10833          *
10834          * What happens if this command would normally block for some other
10835          * reason?  e.g. a request sense with a head of queue tag
10836          * immediately after a write.  Normally that would block, but this
10837          * will result in its getting executed immediately...
10838          *
10839          * We currently return "pass" instead of "skip", so we'll end up
10840          * going through the rest of the queue to check for overlapped tags.
10841          *
10842          * XXX KDM check for other types of blockage first??
10843          */
10844         if (pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10845                 return (CTL_ACTION_PASS);
10846
10847         /*
10848          * Ordered tags have to block until all items ahead of them
10849          * have completed.  If we get called with an ordered tag, we always
10850          * block, if something else is ahead of us in the queue.
10851          */
10852         if (pending_io->scsiio.tag_type == CTL_TAG_ORDERED)
10853                 return (CTL_ACTION_BLOCK);
10854
10855         /*
10856          * Simple tags get blocked until all head of queue and ordered tags
10857          * ahead of them have completed.  I'm lumping untagged commands in
10858          * with simple tags here.  XXX KDM is that the right thing to do?
10859          */
10860         if (((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10861           || (pending_io->scsiio.tag_type == CTL_TAG_SIMPLE))
10862          && ((ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10863           || (ooa_io->scsiio.tag_type == CTL_TAG_ORDERED)))
10864                 return (CTL_ACTION_BLOCK);
10865
10866         pending_entry = ctl_get_cmd_entry(&pending_io->scsiio, NULL);
10867         KASSERT(pending_entry->seridx < CTL_SERIDX_COUNT,
10868             ("%s: Invalid seridx %d for pending CDB %02x %02x @ %p",
10869              __func__, pending_entry->seridx, pending_io->scsiio.cdb[0],
10870              pending_io->scsiio.cdb[1], pending_io));
10871         ooa_entry = ctl_get_cmd_entry(&ooa_io->scsiio, NULL);
10872         if (ooa_entry->seridx == CTL_SERIDX_INVLD)
10873                 return (CTL_ACTION_PASS); /* Unsupported command in OOA queue */
10874         KASSERT(ooa_entry->seridx < CTL_SERIDX_COUNT,
10875             ("%s: Invalid seridx %d for ooa CDB %02x %02x @ %p",
10876              __func__, ooa_entry->seridx, ooa_io->scsiio.cdb[0],
10877              ooa_io->scsiio.cdb[1], ooa_io));
10878
10879         serialize_row = ctl_serialize_table[ooa_entry->seridx];
10880
10881         switch (serialize_row[pending_entry->seridx]) {
10882         case CTL_SER_BLOCK:
10883                 return (CTL_ACTION_BLOCK);
10884         case CTL_SER_EXTENT:
10885                 return (ctl_extent_check(ooa_io, pending_io,
10886                     (lun->be_lun && lun->be_lun->serseq == CTL_LUN_SERSEQ_ON)));
10887         case CTL_SER_EXTENTOPT:
10888                 if ((lun->MODE_CTRL.queue_flags & SCP_QUEUE_ALG_MASK) !=
10889                     SCP_QUEUE_ALG_UNRESTRICTED)
10890                         return (ctl_extent_check(ooa_io, pending_io,
10891                             (lun->be_lun &&
10892                              lun->be_lun->serseq == CTL_LUN_SERSEQ_ON)));
10893                 return (CTL_ACTION_PASS);
10894         case CTL_SER_EXTENTSEQ:
10895                 if (lun->be_lun && lun->be_lun->serseq != CTL_LUN_SERSEQ_OFF)
10896                         return (ctl_extent_check_seq(ooa_io, pending_io));
10897                 return (CTL_ACTION_PASS);
10898         case CTL_SER_PASS:
10899                 return (CTL_ACTION_PASS);
10900         case CTL_SER_BLOCKOPT:
10901                 if ((lun->MODE_CTRL.queue_flags & SCP_QUEUE_ALG_MASK) !=
10902                     SCP_QUEUE_ALG_UNRESTRICTED)
10903                         return (CTL_ACTION_BLOCK);
10904                 return (CTL_ACTION_PASS);
10905         case CTL_SER_SKIP:
10906                 return (CTL_ACTION_SKIP);
10907         default:
10908                 panic("%s: Invalid serialization value %d for %d => %d",
10909                     __func__, serialize_row[pending_entry->seridx],
10910                     pending_entry->seridx, ooa_entry->seridx);
10911         }
10912
10913         return (CTL_ACTION_ERROR);
10914 }
10915
10916 /*
10917  * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
10918  * Assumptions:
10919  * - pending_io is generally either incoming, or on the blocked queue
10920  * - starting I/O is the I/O we want to start the check with.
10921  */
10922 static ctl_action
10923 ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
10924               union ctl_io *starting_io)
10925 {
10926         union ctl_io *ooa_io;
10927         ctl_action action;
10928
10929         mtx_assert(&lun->lun_lock, MA_OWNED);
10930
10931         /*
10932          * Run back along the OOA queue, starting with the current
10933          * blocked I/O and going through every I/O before it on the
10934          * queue.  If starting_io is NULL, we'll just end up returning
10935          * CTL_ACTION_PASS.
10936          */
10937         for (ooa_io = starting_io; ooa_io != NULL;
10938              ooa_io = (union ctl_io *)TAILQ_PREV(&ooa_io->io_hdr, ctl_ooaq,
10939              ooa_links)){
10940
10941                 /*
10942                  * This routine just checks to see whether
10943                  * cur_blocked is blocked by ooa_io, which is ahead
10944                  * of it in the queue.  It doesn't queue/dequeue
10945                  * cur_blocked.
10946                  */
10947                 action = ctl_check_for_blockage(lun, pending_io, ooa_io);
10948                 switch (action) {
10949                 case CTL_ACTION_BLOCK:
10950                 case CTL_ACTION_OVERLAP:
10951                 case CTL_ACTION_OVERLAP_TAG:
10952                 case CTL_ACTION_SKIP:
10953                 case CTL_ACTION_ERROR:
10954                         return (action);
10955                         break; /* NOTREACHED */
10956                 case CTL_ACTION_PASS:
10957                         break;
10958                 default:
10959                         panic("%s: Invalid action %d\n", __func__, action);
10960                 }
10961         }
10962
10963         return (CTL_ACTION_PASS);
10964 }
10965
10966 /*
10967  * Assumptions:
10968  * - An I/O has just completed, and has been removed from the per-LUN OOA
10969  *   queue, so some items on the blocked queue may now be unblocked.
10970  */
10971 static int
10972 ctl_check_blocked(struct ctl_lun *lun)
10973 {
10974         struct ctl_softc *softc = lun->ctl_softc;
10975         union ctl_io *cur_blocked, *next_blocked;
10976
10977         mtx_assert(&lun->lun_lock, MA_OWNED);
10978
10979         /*
10980          * Run forward from the head of the blocked queue, checking each
10981          * entry against the I/Os prior to it on the OOA queue to see if
10982          * there is still any blockage.
10983          *
10984          * We cannot use the TAILQ_FOREACH() macro, because it can't deal
10985          * with our removing a variable on it while it is traversing the
10986          * list.
10987          */
10988         for (cur_blocked = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue);
10989              cur_blocked != NULL; cur_blocked = next_blocked) {
10990                 union ctl_io *prev_ooa;
10991                 ctl_action action;
10992
10993                 next_blocked = (union ctl_io *)TAILQ_NEXT(&cur_blocked->io_hdr,
10994                                                           blocked_links);
10995
10996                 prev_ooa = (union ctl_io *)TAILQ_PREV(&cur_blocked->io_hdr,
10997                                                       ctl_ooaq, ooa_links);
10998
10999                 /*
11000                  * If cur_blocked happens to be the first item in the OOA
11001                  * queue now, prev_ooa will be NULL, and the action
11002                  * returned will just be CTL_ACTION_PASS.
11003                  */
11004                 action = ctl_check_ooa(lun, cur_blocked, prev_ooa);
11005
11006                 switch (action) {
11007                 case CTL_ACTION_BLOCK:
11008                         /* Nothing to do here, still blocked */
11009                         break;
11010                 case CTL_ACTION_OVERLAP:
11011                 case CTL_ACTION_OVERLAP_TAG:
11012                         /*
11013                          * This shouldn't happen!  In theory we've already
11014                          * checked this command for overlap...
11015                          */
11016                         break;
11017                 case CTL_ACTION_PASS:
11018                 case CTL_ACTION_SKIP: {
11019                         const struct ctl_cmd_entry *entry;
11020
11021                         /*
11022                          * The skip case shouldn't happen, this transaction
11023                          * should have never made it onto the blocked queue.
11024                          */
11025                         /*
11026                          * This I/O is no longer blocked, we can remove it
11027                          * from the blocked queue.  Since this is a TAILQ
11028                          * (doubly linked list), we can do O(1) removals
11029                          * from any place on the list.
11030                          */
11031                         TAILQ_REMOVE(&lun->blocked_queue, &cur_blocked->io_hdr,
11032                                      blocked_links);
11033                         cur_blocked->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
11034
11035                         if ((softc->ha_mode != CTL_HA_MODE_XFER) &&
11036                             (cur_blocked->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)){
11037                                 /*
11038                                  * Need to send IO back to original side to
11039                                  * run
11040                                  */
11041                                 union ctl_ha_msg msg_info;
11042
11043                                 cur_blocked->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11044                                 msg_info.hdr.original_sc =
11045                                         cur_blocked->io_hdr.remote_io;
11046                                 msg_info.hdr.serializing_sc = cur_blocked;
11047                                 msg_info.hdr.msg_type = CTL_MSG_R2R;
11048                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11049                                     sizeof(msg_info.hdr), M_NOWAIT);
11050                                 break;
11051                         }
11052                         entry = ctl_get_cmd_entry(&cur_blocked->scsiio, NULL);
11053
11054                         /*
11055                          * Check this I/O for LUN state changes that may
11056                          * have happened while this command was blocked.
11057                          * The LUN state may have been changed by a command
11058                          * ahead of us in the queue, so we need to re-check
11059                          * for any states that can be caused by SCSI
11060                          * commands.
11061                          */
11062                         if (ctl_scsiio_lun_check(lun, entry,
11063                                                  &cur_blocked->scsiio) == 0) {
11064                                 cur_blocked->io_hdr.flags |=
11065                                                       CTL_FLAG_IS_WAS_ON_RTR;
11066                                 ctl_enqueue_rtr(cur_blocked);
11067                         } else
11068                                 ctl_done(cur_blocked);
11069                         break;
11070                 }
11071                 default:
11072                         /*
11073                          * This probably shouldn't happen -- we shouldn't
11074                          * get CTL_ACTION_ERROR, or anything else.
11075                          */
11076                         break;
11077                 }
11078         }
11079
11080         return (CTL_RETVAL_COMPLETE);
11081 }
11082
11083 /*
11084  * This routine (with one exception) checks LUN flags that can be set by
11085  * commands ahead of us in the OOA queue.  These flags have to be checked
11086  * when a command initially comes in, and when we pull a command off the
11087  * blocked queue and are preparing to execute it.  The reason we have to
11088  * check these flags for commands on the blocked queue is that the LUN
11089  * state may have been changed by a command ahead of us while we're on the
11090  * blocked queue.
11091  *
11092  * Ordering is somewhat important with these checks, so please pay
11093  * careful attention to the placement of any new checks.
11094  */
11095 static int
11096 ctl_scsiio_lun_check(struct ctl_lun *lun,
11097     const struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
11098 {
11099         struct ctl_softc *softc = lun->ctl_softc;
11100         int retval;
11101         uint32_t residx;
11102
11103         retval = 0;
11104
11105         mtx_assert(&lun->lun_lock, MA_OWNED);
11106
11107         /*
11108          * If this shelf is a secondary shelf controller, we may have to
11109          * reject some commands disallowed by HA mode and link state.
11110          */
11111         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
11112                 if (softc->ha_link == CTL_HA_LINK_OFFLINE &&
11113                     (entry->flags & CTL_CMD_FLAG_OK_ON_UNAVAIL) == 0) {
11114                         ctl_set_lun_unavail(ctsio);
11115                         retval = 1;
11116                         goto bailout;
11117                 }
11118                 if ((lun->flags & CTL_LUN_PEER_SC_PRIMARY) == 0 &&
11119                     (entry->flags & CTL_CMD_FLAG_OK_ON_UNAVAIL) == 0) {
11120                         ctl_set_lun_transit(ctsio);
11121                         retval = 1;
11122                         goto bailout;
11123                 }
11124                 if (softc->ha_mode == CTL_HA_MODE_ACT_STBY &&
11125                     (entry->flags & CTL_CMD_FLAG_OK_ON_STANDBY) == 0) {
11126                         ctl_set_lun_standby(ctsio);
11127                         retval = 1;
11128                         goto bailout;
11129                 }
11130
11131                 /* The rest of checks are only done on executing side */
11132                 if (softc->ha_mode == CTL_HA_MODE_XFER)
11133                         goto bailout;
11134         }
11135
11136         if (entry->pattern & CTL_LUN_PAT_WRITE) {
11137                 if (lun->be_lun &&
11138                     lun->be_lun->flags & CTL_LUN_FLAG_READONLY) {
11139                         ctl_set_hw_write_protected(ctsio);
11140                         retval = 1;
11141                         goto bailout;
11142                 }
11143                 if ((lun->MODE_CTRL.eca_and_aen & SCP_SWP) != 0) {
11144                         ctl_set_sense(ctsio, /*current_error*/ 1,
11145                             /*sense_key*/ SSD_KEY_DATA_PROTECT,
11146                             /*asc*/ 0x27, /*ascq*/ 0x02, SSD_ELEM_NONE);
11147                         retval = 1;
11148                         goto bailout;
11149                 }
11150         }
11151
11152         /*
11153          * Check for a reservation conflict.  If this command isn't allowed
11154          * even on reserved LUNs, and if this initiator isn't the one who
11155          * reserved us, reject the command with a reservation conflict.
11156          */
11157         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11158         if ((lun->flags & CTL_LUN_RESERVED)
11159          && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
11160                 if (lun->res_idx != residx) {
11161                         ctl_set_reservation_conflict(ctsio);
11162                         retval = 1;
11163                         goto bailout;
11164                 }
11165         }
11166
11167         if ((lun->flags & CTL_LUN_PR_RESERVED) == 0 ||
11168             (entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV)) {
11169                 /* No reservation or command is allowed. */;
11170         } else if ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_WRESV) &&
11171             (lun->pr_res_type == SPR_TYPE_WR_EX ||
11172              lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
11173              lun->pr_res_type == SPR_TYPE_WR_EX_AR)) {
11174                 /* The command is allowed for Write Exclusive resv. */;
11175         } else {
11176                 /*
11177                  * if we aren't registered or it's a res holder type
11178                  * reservation and this isn't the res holder then set a
11179                  * conflict.
11180                  */
11181                 if (ctl_get_prkey(lun, residx) == 0 ||
11182                     (residx != lun->pr_res_idx && lun->pr_res_type < 4)) {
11183                         ctl_set_reservation_conflict(ctsio);
11184                         retval = 1;
11185                         goto bailout;
11186                 }
11187         }
11188
11189         if ((entry->flags & CTL_CMD_FLAG_OK_ON_NO_MEDIA) == 0) {
11190                 if (lun->flags & CTL_LUN_EJECTED)
11191                         ctl_set_lun_ejected(ctsio);
11192                 else if (lun->flags & CTL_LUN_NO_MEDIA) {
11193                         if (lun->flags & CTL_LUN_REMOVABLE)
11194                                 ctl_set_lun_no_media(ctsio);
11195                         else
11196                                 ctl_set_lun_int_reqd(ctsio);
11197                 } else if (lun->flags & CTL_LUN_STOPPED)
11198                         ctl_set_lun_stopped(ctsio);
11199                 else
11200                         goto bailout;
11201                 retval = 1;
11202                 goto bailout;
11203         }
11204
11205 bailout:
11206         return (retval);
11207 }
11208
11209 static void
11210 ctl_failover_io(union ctl_io *io, int have_lock)
11211 {
11212         ctl_set_busy(&io->scsiio);
11213         ctl_done(io);
11214 }
11215
11216 static void
11217 ctl_failover_lun(union ctl_io *rio)
11218 {
11219         struct ctl_softc *softc = CTL_SOFTC(rio);
11220         struct ctl_lun *lun;
11221         struct ctl_io_hdr *io, *next_io;
11222         uint32_t targ_lun;
11223
11224         targ_lun = rio->io_hdr.nexus.targ_mapped_lun;
11225         CTL_DEBUG_PRINT(("FAILOVER for lun %ju\n", targ_lun));
11226
11227         /* Find and lock the LUN. */
11228         mtx_lock(&softc->ctl_lock);
11229         if (targ_lun > ctl_max_luns ||
11230             (lun = softc->ctl_luns[targ_lun]) == NULL) {
11231                 mtx_unlock(&softc->ctl_lock);
11232                 return;
11233         }
11234         mtx_lock(&lun->lun_lock);
11235         mtx_unlock(&softc->ctl_lock);
11236         if (lun->flags & CTL_LUN_DISABLED) {
11237                 mtx_unlock(&lun->lun_lock);
11238                 return;
11239         }
11240
11241         if (softc->ha_mode == CTL_HA_MODE_XFER) {
11242                 TAILQ_FOREACH_SAFE(io, &lun->ooa_queue, ooa_links, next_io) {
11243                         /* We are master */
11244                         if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
11245                                 if (io->flags & CTL_FLAG_IO_ACTIVE) {
11246                                         io->flags |= CTL_FLAG_ABORT;
11247                                         io->flags |= CTL_FLAG_FAILOVER;
11248                                 } else { /* This can be only due to DATAMOVE */
11249                                         io->msg_type = CTL_MSG_DATAMOVE_DONE;
11250                                         io->flags &= ~CTL_FLAG_DMA_INPROG;
11251                                         io->flags |= CTL_FLAG_IO_ACTIVE;
11252                                         io->port_status = 31340;
11253                                         ctl_enqueue_isc((union ctl_io *)io);
11254                                 }
11255                         }
11256                         /* We are slave */
11257                         if (io->flags & CTL_FLAG_SENT_2OTHER_SC) {
11258                                 io->flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11259                                 if (io->flags & CTL_FLAG_IO_ACTIVE) {
11260                                         io->flags |= CTL_FLAG_FAILOVER;
11261                                 } else {
11262                                         ctl_set_busy(&((union ctl_io *)io)->
11263                                             scsiio);
11264                                         ctl_done((union ctl_io *)io);
11265                                 }
11266                         }
11267                 }
11268         } else { /* SERIALIZE modes */
11269                 TAILQ_FOREACH_SAFE(io, &lun->blocked_queue, blocked_links,
11270                     next_io) {
11271                         /* We are master */
11272                         if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
11273                                 TAILQ_REMOVE(&lun->blocked_queue, io,
11274                                     blocked_links);
11275                                 io->flags &= ~CTL_FLAG_BLOCKED;
11276                                 TAILQ_REMOVE(&lun->ooa_queue, io, ooa_links);
11277                                 ctl_free_io((union ctl_io *)io);
11278                         }
11279                 }
11280                 TAILQ_FOREACH_SAFE(io, &lun->ooa_queue, ooa_links, next_io) {
11281                         /* We are master */
11282                         if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
11283                                 TAILQ_REMOVE(&lun->ooa_queue, io, ooa_links);
11284                                 ctl_free_io((union ctl_io *)io);
11285                         }
11286                         /* We are slave */
11287                         if (io->flags & CTL_FLAG_SENT_2OTHER_SC) {
11288                                 io->flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11289                                 if (!(io->flags & CTL_FLAG_IO_ACTIVE)) {
11290                                         ctl_set_busy(&((union ctl_io *)io)->
11291                                             scsiio);
11292                                         ctl_done((union ctl_io *)io);
11293                                 }
11294                         }
11295                 }
11296                 ctl_check_blocked(lun);
11297         }
11298         mtx_unlock(&lun->lun_lock);
11299 }
11300
11301 static int
11302 ctl_scsiio_precheck(struct ctl_softc *softc, struct ctl_scsiio *ctsio)
11303 {
11304         struct ctl_lun *lun;
11305         const struct ctl_cmd_entry *entry;
11306         uint32_t initidx, targ_lun;
11307         int retval = 0;
11308
11309         lun = NULL;
11310         targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
11311         if (targ_lun < ctl_max_luns)
11312                 lun = softc->ctl_luns[targ_lun];
11313         if (lun) {
11314                 /*
11315                  * If the LUN is invalid, pretend that it doesn't exist.
11316                  * It will go away as soon as all pending I/O has been
11317                  * completed.
11318                  */
11319                 mtx_lock(&lun->lun_lock);
11320                 if (lun->flags & CTL_LUN_DISABLED) {
11321                         mtx_unlock(&lun->lun_lock);
11322                         lun = NULL;
11323                 }
11324         }
11325         CTL_LUN(ctsio) = lun;
11326         if (lun) {
11327                 CTL_BACKEND_LUN(ctsio) = lun->be_lun;
11328
11329                 /*
11330                  * Every I/O goes into the OOA queue for a particular LUN,
11331                  * and stays there until completion.
11332                  */
11333 #ifdef CTL_TIME_IO
11334                 if (TAILQ_EMPTY(&lun->ooa_queue))
11335                         lun->idle_time += getsbinuptime() - lun->last_busy;
11336 #endif
11337                 TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
11338         }
11339
11340         /* Get command entry and return error if it is unsuppotyed. */
11341         entry = ctl_validate_command(ctsio);
11342         if (entry == NULL) {
11343                 if (lun)
11344                         mtx_unlock(&lun->lun_lock);
11345                 return (retval);
11346         }
11347
11348         ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
11349         ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
11350
11351         /*
11352          * Check to see whether we can send this command to LUNs that don't
11353          * exist.  This should pretty much only be the case for inquiry
11354          * and request sense.  Further checks, below, really require having
11355          * a LUN, so we can't really check the command anymore.  Just put
11356          * it on the rtr queue.
11357          */
11358         if (lun == NULL) {
11359                 if (entry->flags & CTL_CMD_FLAG_OK_ON_NO_LUN) {
11360                         ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11361                         ctl_enqueue_rtr((union ctl_io *)ctsio);
11362                         return (retval);
11363                 }
11364
11365                 ctl_set_unsupported_lun(ctsio);
11366                 ctl_done((union ctl_io *)ctsio);
11367                 CTL_DEBUG_PRINT(("ctl_scsiio_precheck: bailing out due to invalid LUN\n"));
11368                 return (retval);
11369         } else {
11370                 /*
11371                  * Make sure we support this particular command on this LUN.
11372                  * e.g., we don't support writes to the control LUN.
11373                  */
11374                 if (!ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
11375                         mtx_unlock(&lun->lun_lock);
11376                         ctl_set_invalid_opcode(ctsio);
11377                         ctl_done((union ctl_io *)ctsio);
11378                         return (retval);
11379                 }
11380         }
11381
11382         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11383
11384         /*
11385          * If we've got a request sense, it'll clear the contingent
11386          * allegiance condition.  Otherwise, if we have a CA condition for
11387          * this initiator, clear it, because it sent down a command other
11388          * than request sense.
11389          */
11390         if (ctsio->cdb[0] != REQUEST_SENSE) {
11391                 struct scsi_sense_data *ps;
11392
11393                 ps = lun->pending_sense[initidx / CTL_MAX_INIT_PER_PORT];
11394                 if (ps != NULL)
11395                         ps[initidx % CTL_MAX_INIT_PER_PORT].error_code = 0;
11396         }
11397
11398         /*
11399          * If the command has this flag set, it handles its own unit
11400          * attention reporting, we shouldn't do anything.  Otherwise we
11401          * check for any pending unit attentions, and send them back to the
11402          * initiator.  We only do this when a command initially comes in,
11403          * not when we pull it off the blocked queue.
11404          *
11405          * According to SAM-3, section 5.3.2, the order that things get
11406          * presented back to the host is basically unit attentions caused
11407          * by some sort of reset event, busy status, reservation conflicts
11408          * or task set full, and finally any other status.
11409          *
11410          * One issue here is that some of the unit attentions we report
11411          * don't fall into the "reset" category (e.g. "reported luns data
11412          * has changed").  So reporting it here, before the reservation
11413          * check, may be technically wrong.  I guess the only thing to do
11414          * would be to check for and report the reset events here, and then
11415          * check for the other unit attention types after we check for a
11416          * reservation conflict.
11417          *
11418          * XXX KDM need to fix this
11419          */
11420         if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
11421                 ctl_ua_type ua_type;
11422                 u_int sense_len = 0;
11423
11424                 ua_type = ctl_build_ua(lun, initidx, &ctsio->sense_data,
11425                     &sense_len, SSD_TYPE_NONE);
11426                 if (ua_type != CTL_UA_NONE) {
11427                         mtx_unlock(&lun->lun_lock);
11428                         ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
11429                         ctsio->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
11430                         ctsio->sense_len = sense_len;
11431                         ctl_done((union ctl_io *)ctsio);
11432                         return (retval);
11433                 }
11434         }
11435
11436
11437         if (ctl_scsiio_lun_check(lun, entry, ctsio) != 0) {
11438                 mtx_unlock(&lun->lun_lock);
11439                 ctl_done((union ctl_io *)ctsio);
11440                 return (retval);
11441         }
11442
11443         /*
11444          * XXX CHD this is where we want to send IO to other side if
11445          * this LUN is secondary on this SC. We will need to make a copy
11446          * of the IO and flag the IO on this side as SENT_2OTHER and the flag
11447          * the copy we send as FROM_OTHER.
11448          * We also need to stuff the address of the original IO so we can
11449          * find it easily. Something similar will need be done on the other
11450          * side so when we are done we can find the copy.
11451          */
11452         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
11453             (lun->flags & CTL_LUN_PEER_SC_PRIMARY) != 0 &&
11454             (entry->flags & CTL_CMD_FLAG_RUN_HERE) == 0) {
11455                 union ctl_ha_msg msg_info;
11456                 int isc_retval;
11457
11458                 ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11459                 ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11460                 mtx_unlock(&lun->lun_lock);
11461
11462                 msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
11463                 msg_info.hdr.original_sc = (union ctl_io *)ctsio;
11464                 msg_info.hdr.serializing_sc = NULL;
11465                 msg_info.hdr.nexus = ctsio->io_hdr.nexus;
11466                 msg_info.scsi.tag_num = ctsio->tag_num;
11467                 msg_info.scsi.tag_type = ctsio->tag_type;
11468                 msg_info.scsi.cdb_len = ctsio->cdb_len;
11469                 memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
11470
11471                 if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11472                     sizeof(msg_info.scsi) - sizeof(msg_info.scsi.sense_data),
11473                     M_WAITOK)) > CTL_HA_STATUS_SUCCESS) {
11474                         ctl_set_busy(ctsio);
11475                         ctl_done((union ctl_io *)ctsio);
11476                         return (retval);
11477                 }
11478                 return (retval);
11479         }
11480
11481         switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
11482                               (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr,
11483                               ctl_ooaq, ooa_links))) {
11484         case CTL_ACTION_BLOCK:
11485                 ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
11486                 TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
11487                                   blocked_links);
11488                 mtx_unlock(&lun->lun_lock);
11489                 return (retval);
11490         case CTL_ACTION_PASS:
11491         case CTL_ACTION_SKIP:
11492                 ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11493                 mtx_unlock(&lun->lun_lock);
11494                 ctl_enqueue_rtr((union ctl_io *)ctsio);
11495                 break;
11496         case CTL_ACTION_OVERLAP:
11497                 mtx_unlock(&lun->lun_lock);
11498                 ctl_set_overlapped_cmd(ctsio);
11499                 ctl_done((union ctl_io *)ctsio);
11500                 break;
11501         case CTL_ACTION_OVERLAP_TAG:
11502                 mtx_unlock(&lun->lun_lock);
11503                 ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
11504                 ctl_done((union ctl_io *)ctsio);
11505                 break;
11506         case CTL_ACTION_ERROR:
11507         default:
11508                 mtx_unlock(&lun->lun_lock);
11509                 ctl_set_internal_failure(ctsio,
11510                                          /*sks_valid*/ 0,
11511                                          /*retry_count*/ 0);
11512                 ctl_done((union ctl_io *)ctsio);
11513                 break;
11514         }
11515         return (retval);
11516 }
11517
11518 const struct ctl_cmd_entry *
11519 ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa)
11520 {
11521         const struct ctl_cmd_entry *entry;
11522         int service_action;
11523
11524         entry = &ctl_cmd_table[ctsio->cdb[0]];
11525         if (sa)
11526                 *sa = ((entry->flags & CTL_CMD_FLAG_SA5) != 0);
11527         if (entry->flags & CTL_CMD_FLAG_SA5) {
11528                 service_action = ctsio->cdb[1] & SERVICE_ACTION_MASK;
11529                 entry = &((const struct ctl_cmd_entry *)
11530                     entry->execute)[service_action];
11531         }
11532         return (entry);
11533 }
11534
11535 const struct ctl_cmd_entry *
11536 ctl_validate_command(struct ctl_scsiio *ctsio)
11537 {
11538         const struct ctl_cmd_entry *entry;
11539         int i, sa;
11540         uint8_t diff;
11541
11542         entry = ctl_get_cmd_entry(ctsio, &sa);
11543         if (entry->execute == NULL) {
11544                 if (sa)
11545                         ctl_set_invalid_field(ctsio,
11546                                               /*sks_valid*/ 1,
11547                                               /*command*/ 1,
11548                                               /*field*/ 1,
11549                                               /*bit_valid*/ 1,
11550                                               /*bit*/ 4);
11551                 else
11552                         ctl_set_invalid_opcode(ctsio);
11553                 ctl_done((union ctl_io *)ctsio);
11554                 return (NULL);
11555         }
11556         KASSERT(entry->length > 0,
11557             ("Not defined length for command 0x%02x/0x%02x",
11558              ctsio->cdb[0], ctsio->cdb[1]));
11559         for (i = 1; i < entry->length; i++) {
11560                 diff = ctsio->cdb[i] & ~entry->usage[i - 1];
11561                 if (diff == 0)
11562                         continue;
11563                 ctl_set_invalid_field(ctsio,
11564                                       /*sks_valid*/ 1,
11565                                       /*command*/ 1,
11566                                       /*field*/ i,
11567                                       /*bit_valid*/ 1,
11568                                       /*bit*/ fls(diff) - 1);
11569                 ctl_done((union ctl_io *)ctsio);
11570                 return (NULL);
11571         }
11572         return (entry);
11573 }
11574
11575 static int
11576 ctl_cmd_applicable(uint8_t lun_type, const struct ctl_cmd_entry *entry)
11577 {
11578
11579         switch (lun_type) {
11580         case T_DIRECT:
11581                 if ((entry->flags & CTL_CMD_FLAG_OK_ON_DIRECT) == 0)
11582                         return (0);
11583                 break;
11584         case T_PROCESSOR:
11585                 if ((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0)
11586                         return (0);
11587                 break;
11588         case T_CDROM:
11589                 if ((entry->flags & CTL_CMD_FLAG_OK_ON_CDROM) == 0)
11590                         return (0);
11591                 break;
11592         default:
11593                 return (0);
11594         }
11595         return (1);
11596 }
11597
11598 static int
11599 ctl_scsiio(struct ctl_scsiio *ctsio)
11600 {
11601         int retval;
11602         const struct ctl_cmd_entry *entry;
11603
11604         retval = CTL_RETVAL_COMPLETE;
11605
11606         CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
11607
11608         entry = ctl_get_cmd_entry(ctsio, NULL);
11609
11610         /*
11611          * If this I/O has been aborted, just send it straight to
11612          * ctl_done() without executing it.
11613          */
11614         if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
11615                 ctl_done((union ctl_io *)ctsio);
11616                 goto bailout;
11617         }
11618
11619         /*
11620          * All the checks should have been handled by ctl_scsiio_precheck().
11621          * We should be clear now to just execute the I/O.
11622          */
11623         retval = entry->execute(ctsio);
11624
11625 bailout:
11626         return (retval);
11627 }
11628
11629 static int
11630 ctl_target_reset(union ctl_io *io)
11631 {
11632         struct ctl_softc *softc = CTL_SOFTC(io);
11633         struct ctl_port *port = CTL_PORT(io);
11634         struct ctl_lun *lun;
11635         uint32_t initidx;
11636         ctl_ua_type ua_type;
11637
11638         if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11639                 union ctl_ha_msg msg_info;
11640
11641                 msg_info.hdr.nexus = io->io_hdr.nexus;
11642                 msg_info.task.task_action = io->taskio.task_action;
11643                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11644                 msg_info.hdr.original_sc = NULL;
11645                 msg_info.hdr.serializing_sc = NULL;
11646                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11647                     sizeof(msg_info.task), M_WAITOK);
11648         }
11649
11650         initidx = ctl_get_initindex(&io->io_hdr.nexus);
11651         if (io->taskio.task_action == CTL_TASK_TARGET_RESET)
11652                 ua_type = CTL_UA_TARG_RESET;
11653         else
11654                 ua_type = CTL_UA_BUS_RESET;
11655         mtx_lock(&softc->ctl_lock);
11656         STAILQ_FOREACH(lun, &softc->lun_list, links) {
11657                 if (port != NULL &&
11658                     ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
11659                         continue;
11660                 ctl_do_lun_reset(lun, initidx, ua_type);
11661         }
11662         mtx_unlock(&softc->ctl_lock);
11663         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11664         return (0);
11665 }
11666
11667 /*
11668  * The LUN should always be set.  The I/O is optional, and is used to
11669  * distinguish between I/Os sent by this initiator, and by other
11670  * initiators.  We set unit attention for initiators other than this one.
11671  * SAM-3 is vague on this point.  It does say that a unit attention should
11672  * be established for other initiators when a LUN is reset (see section
11673  * 5.7.3), but it doesn't specifically say that the unit attention should
11674  * be established for this particular initiator when a LUN is reset.  Here
11675  * is the relevant text, from SAM-3 rev 8:
11676  *
11677  * 5.7.2 When a SCSI initiator port aborts its own tasks
11678  *
11679  * When a SCSI initiator port causes its own task(s) to be aborted, no
11680  * notification that the task(s) have been aborted shall be returned to
11681  * the SCSI initiator port other than the completion response for the
11682  * command or task management function action that caused the task(s) to
11683  * be aborted and notification(s) associated with related effects of the
11684  * action (e.g., a reset unit attention condition).
11685  *
11686  * XXX KDM for now, we're setting unit attention for all initiators.
11687  */
11688 static void
11689 ctl_do_lun_reset(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua_type)
11690 {
11691         union ctl_io *xio;
11692         int i;
11693
11694         mtx_lock(&lun->lun_lock);
11695         /* Abort tasks. */
11696         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11697              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11698                 xio->io_hdr.flags |= CTL_FLAG_ABORT | CTL_FLAG_ABORT_STATUS;
11699         }
11700         /* Clear CA. */
11701         for (i = 0; i < ctl_max_ports; i++) {
11702                 free(lun->pending_sense[i], M_CTL);
11703                 lun->pending_sense[i] = NULL;
11704         }
11705         /* Clear reservation. */
11706         lun->flags &= ~CTL_LUN_RESERVED;
11707         /* Clear prevent media removal. */
11708         if (lun->prevent) {
11709                 for (i = 0; i < CTL_MAX_INITIATORS; i++)
11710                         ctl_clear_mask(lun->prevent, i);
11711                 lun->prevent_count = 0;
11712         }
11713         /* Clear TPC status */
11714         ctl_tpc_lun_clear(lun, -1);
11715         /* Establish UA. */
11716 #if 0
11717         ctl_est_ua_all(lun, initidx, ua_type);
11718 #else
11719         ctl_est_ua_all(lun, -1, ua_type);
11720 #endif
11721         mtx_unlock(&lun->lun_lock);
11722 }
11723
11724 static int
11725 ctl_lun_reset(union ctl_io *io)
11726 {
11727         struct ctl_softc *softc = CTL_SOFTC(io);
11728         struct ctl_lun *lun;
11729         uint32_t targ_lun, initidx;
11730
11731         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11732         initidx = ctl_get_initindex(&io->io_hdr.nexus);
11733         mtx_lock(&softc->ctl_lock);
11734         if (targ_lun >= ctl_max_luns ||
11735             (lun = softc->ctl_luns[targ_lun]) == NULL) {
11736                 mtx_unlock(&softc->ctl_lock);
11737                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
11738                 return (1);
11739         }
11740         ctl_do_lun_reset(lun, initidx, CTL_UA_LUN_RESET);
11741         mtx_unlock(&softc->ctl_lock);
11742         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11743
11744         if ((io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) == 0) {
11745                 union ctl_ha_msg msg_info;
11746
11747                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11748                 msg_info.hdr.nexus = io->io_hdr.nexus;
11749                 msg_info.task.task_action = CTL_TASK_LUN_RESET;
11750                 msg_info.hdr.original_sc = NULL;
11751                 msg_info.hdr.serializing_sc = NULL;
11752                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11753                     sizeof(msg_info.task), M_WAITOK);
11754         }
11755         return (0);
11756 }
11757
11758 static void
11759 ctl_abort_tasks_lun(struct ctl_lun *lun, uint32_t targ_port, uint32_t init_id,
11760     int other_sc)
11761 {
11762         union ctl_io *xio;
11763
11764         mtx_assert(&lun->lun_lock, MA_OWNED);
11765
11766         /*
11767          * Run through the OOA queue and attempt to find the given I/O.
11768          * The target port, initiator ID, tag type and tag number have to
11769          * match the values that we got from the initiator.  If we have an
11770          * untagged command to abort, simply abort the first untagged command
11771          * we come to.  We only allow one untagged command at a time of course.
11772          */
11773         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11774              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11775
11776                 if ((targ_port == UINT32_MAX ||
11777                      targ_port == xio->io_hdr.nexus.targ_port) &&
11778                     (init_id == UINT32_MAX ||
11779                      init_id == xio->io_hdr.nexus.initid)) {
11780                         if (targ_port != xio->io_hdr.nexus.targ_port ||
11781                             init_id != xio->io_hdr.nexus.initid)
11782                                 xio->io_hdr.flags |= CTL_FLAG_ABORT_STATUS;
11783                         xio->io_hdr.flags |= CTL_FLAG_ABORT;
11784                         if (!other_sc && !(lun->flags & CTL_LUN_PRIMARY_SC)) {
11785                                 union ctl_ha_msg msg_info;
11786
11787                                 msg_info.hdr.nexus = xio->io_hdr.nexus;
11788                                 msg_info.task.task_action = CTL_TASK_ABORT_TASK;
11789                                 msg_info.task.tag_num = xio->scsiio.tag_num;
11790                                 msg_info.task.tag_type = xio->scsiio.tag_type;
11791                                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11792                                 msg_info.hdr.original_sc = NULL;
11793                                 msg_info.hdr.serializing_sc = NULL;
11794                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11795                                     sizeof(msg_info.task), M_NOWAIT);
11796                         }
11797                 }
11798         }
11799 }
11800
11801 static int
11802 ctl_abort_task_set(union ctl_io *io)
11803 {
11804         struct ctl_softc *softc = CTL_SOFTC(io);
11805         struct ctl_lun *lun;
11806         uint32_t targ_lun;
11807
11808         /*
11809          * Look up the LUN.
11810          */
11811         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11812         mtx_lock(&softc->ctl_lock);
11813         if (targ_lun >= ctl_max_luns ||
11814             (lun = softc->ctl_luns[targ_lun]) == NULL) {
11815                 mtx_unlock(&softc->ctl_lock);
11816                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
11817                 return (1);
11818         }
11819
11820         mtx_lock(&lun->lun_lock);
11821         mtx_unlock(&softc->ctl_lock);
11822         if (io->taskio.task_action == CTL_TASK_ABORT_TASK_SET) {
11823                 ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
11824                     io->io_hdr.nexus.initid,
11825                     (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
11826         } else { /* CTL_TASK_CLEAR_TASK_SET */
11827                 ctl_abort_tasks_lun(lun, UINT32_MAX, UINT32_MAX,
11828                     (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
11829         }
11830         mtx_unlock(&lun->lun_lock);
11831         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11832         return (0);
11833 }
11834
11835 static void
11836 ctl_i_t_nexus_loss(struct ctl_softc *softc, uint32_t initidx,
11837     ctl_ua_type ua_type)
11838 {
11839         struct ctl_lun *lun;
11840         struct scsi_sense_data *ps;
11841         uint32_t p, i;
11842
11843         p = initidx / CTL_MAX_INIT_PER_PORT;
11844         i = initidx % CTL_MAX_INIT_PER_PORT;
11845         mtx_lock(&softc->ctl_lock);
11846         STAILQ_FOREACH(lun, &softc->lun_list, links) {
11847                 mtx_lock(&lun->lun_lock);
11848                 /* Abort tasks. */
11849                 ctl_abort_tasks_lun(lun, p, i, 1);
11850                 /* Clear CA. */
11851                 ps = lun->pending_sense[p];
11852                 if (ps != NULL)
11853                         ps[i].error_code = 0;
11854                 /* Clear reservation. */
11855                 if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == initidx))
11856                         lun->flags &= ~CTL_LUN_RESERVED;
11857                 /* Clear prevent media removal. */
11858                 if (lun->prevent && ctl_is_set(lun->prevent, initidx)) {
11859                         ctl_clear_mask(lun->prevent, initidx);
11860                         lun->prevent_count--;
11861                 }
11862                 /* Clear TPC status */
11863                 ctl_tpc_lun_clear(lun, initidx);
11864                 /* Establish UA. */
11865                 ctl_est_ua(lun, initidx, ua_type);
11866                 mtx_unlock(&lun->lun_lock);
11867         }
11868         mtx_unlock(&softc->ctl_lock);
11869 }
11870
11871 static int
11872 ctl_i_t_nexus_reset(union ctl_io *io)
11873 {
11874         struct ctl_softc *softc = CTL_SOFTC(io);
11875         uint32_t initidx;
11876
11877         if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11878                 union ctl_ha_msg msg_info;
11879
11880                 msg_info.hdr.nexus = io->io_hdr.nexus;
11881                 msg_info.task.task_action = CTL_TASK_I_T_NEXUS_RESET;
11882                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11883                 msg_info.hdr.original_sc = NULL;
11884                 msg_info.hdr.serializing_sc = NULL;
11885                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11886                     sizeof(msg_info.task), M_WAITOK);
11887         }
11888
11889         initidx = ctl_get_initindex(&io->io_hdr.nexus);
11890         ctl_i_t_nexus_loss(softc, initidx, CTL_UA_I_T_NEXUS_LOSS);
11891         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11892         return (0);
11893 }
11894
11895 static int
11896 ctl_abort_task(union ctl_io *io)
11897 {
11898         struct ctl_softc *softc = CTL_SOFTC(io);
11899         union ctl_io *xio;
11900         struct ctl_lun *lun;
11901 #if 0
11902         struct sbuf sb;
11903         char printbuf[128];
11904 #endif
11905         int found;
11906         uint32_t targ_lun;
11907
11908         found = 0;
11909
11910         /*
11911          * Look up the LUN.
11912          */
11913         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11914         mtx_lock(&softc->ctl_lock);
11915         if (targ_lun >= ctl_max_luns ||
11916             (lun = softc->ctl_luns[targ_lun]) == NULL) {
11917                 mtx_unlock(&softc->ctl_lock);
11918                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
11919                 return (1);
11920         }
11921
11922 #if 0
11923         printf("ctl_abort_task: called for lun %lld, tag %d type %d\n",
11924                lun->lun, io->taskio.tag_num, io->taskio.tag_type);
11925 #endif
11926
11927         mtx_lock(&lun->lun_lock);
11928         mtx_unlock(&softc->ctl_lock);
11929         /*
11930          * Run through the OOA queue and attempt to find the given I/O.
11931          * The target port, initiator ID, tag type and tag number have to
11932          * match the values that we got from the initiator.  If we have an
11933          * untagged command to abort, simply abort the first untagged command
11934          * we come to.  We only allow one untagged command at a time of course.
11935          */
11936         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11937              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11938 #if 0
11939                 sbuf_new(&sb, printbuf, sizeof(printbuf), SBUF_FIXEDLEN);
11940
11941                 sbuf_printf(&sb, "LUN %lld tag %d type %d%s%s%s%s: ",
11942                             lun->lun, xio->scsiio.tag_num,
11943                             xio->scsiio.tag_type,
11944                             (xio->io_hdr.blocked_links.tqe_prev
11945                             == NULL) ? "" : " BLOCKED",
11946                             (xio->io_hdr.flags &
11947                             CTL_FLAG_DMA_INPROG) ? " DMA" : "",
11948                             (xio->io_hdr.flags &
11949                             CTL_FLAG_ABORT) ? " ABORT" : "",
11950                             (xio->io_hdr.flags &
11951                             CTL_FLAG_IS_WAS_ON_RTR ? " RTR" : ""));
11952                 ctl_scsi_command_string(&xio->scsiio, NULL, &sb);
11953                 sbuf_finish(&sb);
11954                 printf("%s\n", sbuf_data(&sb));
11955 #endif
11956
11957                 if ((xio->io_hdr.nexus.targ_port != io->io_hdr.nexus.targ_port)
11958                  || (xio->io_hdr.nexus.initid != io->io_hdr.nexus.initid)
11959                  || (xio->io_hdr.flags & CTL_FLAG_ABORT))
11960                         continue;
11961
11962                 /*
11963                  * If the abort says that the task is untagged, the
11964                  * task in the queue must be untagged.  Otherwise,
11965                  * we just check to see whether the tag numbers
11966                  * match.  This is because the QLogic firmware
11967                  * doesn't pass back the tag type in an abort
11968                  * request.
11969                  */
11970 #if 0
11971                 if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
11972                   && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
11973                  || (xio->scsiio.tag_num == io->taskio.tag_num))
11974 #endif
11975                 /*
11976                  * XXX KDM we've got problems with FC, because it
11977                  * doesn't send down a tag type with aborts.  So we
11978                  * can only really go by the tag number...
11979                  * This may cause problems with parallel SCSI.
11980                  * Need to figure that out!!
11981                  */
11982                 if (xio->scsiio.tag_num == io->taskio.tag_num) {
11983                         xio->io_hdr.flags |= CTL_FLAG_ABORT;
11984                         found = 1;
11985                         if ((io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) == 0 &&
11986                             !(lun->flags & CTL_LUN_PRIMARY_SC)) {
11987                                 union ctl_ha_msg msg_info;
11988
11989                                 msg_info.hdr.nexus = io->io_hdr.nexus;
11990                                 msg_info.task.task_action = CTL_TASK_ABORT_TASK;
11991                                 msg_info.task.tag_num = io->taskio.tag_num;
11992                                 msg_info.task.tag_type = io->taskio.tag_type;
11993                                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11994                                 msg_info.hdr.original_sc = NULL;
11995                                 msg_info.hdr.serializing_sc = NULL;
11996 #if 0
11997                                 printf("Sent Abort to other side\n");
11998 #endif
11999                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
12000                                     sizeof(msg_info.task), M_NOWAIT);
12001                         }
12002 #if 0
12003                         printf("ctl_abort_task: found I/O to abort\n");
12004 #endif
12005                 }
12006         }
12007         mtx_unlock(&lun->lun_lock);
12008
12009         if (found == 0) {
12010                 /*
12011                  * This isn't really an error.  It's entirely possible for
12012                  * the abort and command completion to cross on the wire.
12013                  * This is more of an informative/diagnostic error.
12014                  */
12015 #if 0
12016                 printf("ctl_abort_task: ABORT sent for nonexistent I/O: "
12017                        "%u:%u:%u tag %d type %d\n",
12018                        io->io_hdr.nexus.initid,
12019                        io->io_hdr.nexus.targ_port,
12020                        io->io_hdr.nexus.targ_lun, io->taskio.tag_num,
12021                        io->taskio.tag_type);
12022 #endif
12023         }
12024         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12025         return (0);
12026 }
12027
12028 static int
12029 ctl_query_task(union ctl_io *io, int task_set)
12030 {
12031         struct ctl_softc *softc = CTL_SOFTC(io);
12032         union ctl_io *xio;
12033         struct ctl_lun *lun;
12034         int found = 0;
12035         uint32_t targ_lun;
12036
12037         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12038         mtx_lock(&softc->ctl_lock);
12039         if (targ_lun >= ctl_max_luns ||
12040             (lun = softc->ctl_luns[targ_lun]) == NULL) {
12041                 mtx_unlock(&softc->ctl_lock);
12042                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
12043                 return (1);
12044         }
12045         mtx_lock(&lun->lun_lock);
12046         mtx_unlock(&softc->ctl_lock);
12047         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12048              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12049
12050                 if ((xio->io_hdr.nexus.targ_port != io->io_hdr.nexus.targ_port)
12051                  || (xio->io_hdr.nexus.initid != io->io_hdr.nexus.initid)
12052                  || (xio->io_hdr.flags & CTL_FLAG_ABORT))
12053                         continue;
12054
12055                 if (task_set || xio->scsiio.tag_num == io->taskio.tag_num) {
12056                         found = 1;
12057                         break;
12058                 }
12059         }
12060         mtx_unlock(&lun->lun_lock);
12061         if (found)
12062                 io->taskio.task_status = CTL_TASK_FUNCTION_SUCCEEDED;
12063         else
12064                 io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12065         return (0);
12066 }
12067
12068 static int
12069 ctl_query_async_event(union ctl_io *io)
12070 {
12071         struct ctl_softc *softc = CTL_SOFTC(io);
12072         struct ctl_lun *lun;
12073         ctl_ua_type ua;
12074         uint32_t targ_lun, initidx;
12075
12076         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12077         mtx_lock(&softc->ctl_lock);
12078         if (targ_lun >= ctl_max_luns ||
12079             (lun = softc->ctl_luns[targ_lun]) == NULL) {
12080                 mtx_unlock(&softc->ctl_lock);
12081                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
12082                 return (1);
12083         }
12084         mtx_lock(&lun->lun_lock);
12085         mtx_unlock(&softc->ctl_lock);
12086         initidx = ctl_get_initindex(&io->io_hdr.nexus);
12087         ua = ctl_build_qae(lun, initidx, io->taskio.task_resp);
12088         mtx_unlock(&lun->lun_lock);
12089         if (ua != CTL_UA_NONE)
12090                 io->taskio.task_status = CTL_TASK_FUNCTION_SUCCEEDED;
12091         else
12092                 io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12093         return (0);
12094 }
12095
12096 static void
12097 ctl_run_task(union ctl_io *io)
12098 {
12099         int retval = 1;
12100
12101         CTL_DEBUG_PRINT(("ctl_run_task\n"));
12102         KASSERT(io->io_hdr.io_type == CTL_IO_TASK,
12103             ("ctl_run_task: Unextected io_type %d\n", io->io_hdr.io_type));
12104         io->taskio.task_status = CTL_TASK_FUNCTION_NOT_SUPPORTED;
12105         bzero(io->taskio.task_resp, sizeof(io->taskio.task_resp));
12106         switch (io->taskio.task_action) {
12107         case CTL_TASK_ABORT_TASK:
12108                 retval = ctl_abort_task(io);
12109                 break;
12110         case CTL_TASK_ABORT_TASK_SET:
12111         case CTL_TASK_CLEAR_TASK_SET:
12112                 retval = ctl_abort_task_set(io);
12113                 break;
12114         case CTL_TASK_CLEAR_ACA:
12115                 break;
12116         case CTL_TASK_I_T_NEXUS_RESET:
12117                 retval = ctl_i_t_nexus_reset(io);
12118                 break;
12119         case CTL_TASK_LUN_RESET:
12120                 retval = ctl_lun_reset(io);
12121                 break;
12122         case CTL_TASK_TARGET_RESET:
12123         case CTL_TASK_BUS_RESET:
12124                 retval = ctl_target_reset(io);
12125                 break;
12126         case CTL_TASK_PORT_LOGIN:
12127                 break;
12128         case CTL_TASK_PORT_LOGOUT:
12129                 break;
12130         case CTL_TASK_QUERY_TASK:
12131                 retval = ctl_query_task(io, 0);
12132                 break;
12133         case CTL_TASK_QUERY_TASK_SET:
12134                 retval = ctl_query_task(io, 1);
12135                 break;
12136         case CTL_TASK_QUERY_ASYNC_EVENT:
12137                 retval = ctl_query_async_event(io);
12138                 break;
12139         default:
12140                 printf("%s: got unknown task management event %d\n",
12141                        __func__, io->taskio.task_action);
12142                 break;
12143         }
12144         if (retval == 0)
12145                 io->io_hdr.status = CTL_SUCCESS;
12146         else
12147                 io->io_hdr.status = CTL_ERROR;
12148         ctl_done(io);
12149 }
12150
12151 /*
12152  * For HA operation.  Handle commands that come in from the other
12153  * controller.
12154  */
12155 static void
12156 ctl_handle_isc(union ctl_io *io)
12157 {
12158         struct ctl_softc *softc = CTL_SOFTC(io);
12159         struct ctl_lun *lun;
12160         const struct ctl_cmd_entry *entry;
12161         uint32_t targ_lun;
12162
12163         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12164         switch (io->io_hdr.msg_type) {
12165         case CTL_MSG_SERIALIZE:
12166                 ctl_serialize_other_sc_cmd(&io->scsiio);
12167                 break;
12168         case CTL_MSG_R2R:               /* Only used in SER_ONLY mode. */
12169                 entry = ctl_get_cmd_entry(&io->scsiio, NULL);
12170                 if (targ_lun >= ctl_max_luns ||
12171                     (lun = softc->ctl_luns[targ_lun]) == NULL) {
12172                         ctl_done(io);
12173                         break;
12174                 }
12175                 mtx_lock(&lun->lun_lock);
12176                 if (ctl_scsiio_lun_check(lun, entry, &io->scsiio) != 0) {
12177                         mtx_unlock(&lun->lun_lock);
12178                         ctl_done(io);
12179                         break;
12180                 }
12181                 io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
12182                 mtx_unlock(&lun->lun_lock);
12183                 ctl_enqueue_rtr(io);
12184                 break;
12185         case CTL_MSG_FINISH_IO:
12186                 if (softc->ha_mode == CTL_HA_MODE_XFER) {
12187                         ctl_done(io);
12188                         break;
12189                 }
12190                 if (targ_lun >= ctl_max_luns ||
12191                     (lun = softc->ctl_luns[targ_lun]) == NULL) {
12192                         ctl_free_io(io);
12193                         break;
12194                 }
12195                 mtx_lock(&lun->lun_lock);
12196                 TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
12197                 ctl_check_blocked(lun);
12198                 mtx_unlock(&lun->lun_lock);
12199                 ctl_free_io(io);
12200                 break;
12201         case CTL_MSG_PERS_ACTION:
12202                 ctl_hndl_per_res_out_on_other_sc(io);
12203                 ctl_free_io(io);
12204                 break;
12205         case CTL_MSG_BAD_JUJU:
12206                 ctl_done(io);
12207                 break;
12208         case CTL_MSG_DATAMOVE:          /* Only used in XFER mode */
12209                 ctl_datamove_remote(io);
12210                 break;
12211         case CTL_MSG_DATAMOVE_DONE:     /* Only used in XFER mode */
12212                 io->scsiio.be_move_done(io);
12213                 break;
12214         case CTL_MSG_FAILOVER:
12215                 ctl_failover_lun(io);
12216                 ctl_free_io(io);
12217                 break;
12218         default:
12219                 printf("%s: Invalid message type %d\n",
12220                        __func__, io->io_hdr.msg_type);
12221                 ctl_free_io(io);
12222                 break;
12223         }
12224
12225 }
12226
12227
12228 /*
12229  * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
12230  * there is no match.
12231  */
12232 static ctl_lun_error_pattern
12233 ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
12234 {
12235         const struct ctl_cmd_entry *entry;
12236         ctl_lun_error_pattern filtered_pattern, pattern;
12237
12238         pattern = desc->error_pattern;
12239
12240         /*
12241          * XXX KDM we need more data passed into this function to match a
12242          * custom pattern, and we actually need to implement custom pattern
12243          * matching.
12244          */
12245         if (pattern & CTL_LUN_PAT_CMD)
12246                 return (CTL_LUN_PAT_CMD);
12247
12248         if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
12249                 return (CTL_LUN_PAT_ANY);
12250
12251         entry = ctl_get_cmd_entry(ctsio, NULL);
12252
12253         filtered_pattern = entry->pattern & pattern;
12254
12255         /*
12256          * If the user requested specific flags in the pattern (e.g.
12257          * CTL_LUN_PAT_RANGE), make sure the command supports all of those
12258          * flags.
12259          *
12260          * If the user did not specify any flags, it doesn't matter whether
12261          * or not the command supports the flags.
12262          */
12263         if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
12264              (pattern & ~CTL_LUN_PAT_MASK))
12265                 return (CTL_LUN_PAT_NONE);
12266
12267         /*
12268          * If the user asked for a range check, see if the requested LBA
12269          * range overlaps with this command's LBA range.
12270          */
12271         if (filtered_pattern & CTL_LUN_PAT_RANGE) {
12272                 uint64_t lba1;
12273                 uint64_t len1;
12274                 ctl_action action;
12275                 int retval;
12276
12277                 retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
12278                 if (retval != 0)
12279                         return (CTL_LUN_PAT_NONE);
12280
12281                 action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
12282                                               desc->lba_range.len, FALSE);
12283                 /*
12284                  * A "pass" means that the LBA ranges don't overlap, so
12285                  * this doesn't match the user's range criteria.
12286                  */
12287                 if (action == CTL_ACTION_PASS)
12288                         return (CTL_LUN_PAT_NONE);
12289         }
12290
12291         return (filtered_pattern);
12292 }
12293
12294 static void
12295 ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
12296 {
12297         struct ctl_error_desc *desc, *desc2;
12298
12299         mtx_assert(&lun->lun_lock, MA_OWNED);
12300
12301         STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
12302                 ctl_lun_error_pattern pattern;
12303                 /*
12304                  * Check to see whether this particular command matches
12305                  * the pattern in the descriptor.
12306                  */
12307                 pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
12308                 if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
12309                         continue;
12310
12311                 switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
12312                 case CTL_LUN_INJ_ABORTED:
12313                         ctl_set_aborted(&io->scsiio);
12314                         break;
12315                 case CTL_LUN_INJ_MEDIUM_ERR:
12316                         ctl_set_medium_error(&io->scsiio,
12317                             (io->io_hdr.flags & CTL_FLAG_DATA_MASK) !=
12318                              CTL_FLAG_DATA_OUT);
12319                         break;
12320                 case CTL_LUN_INJ_UA:
12321                         /* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
12322                          * OCCURRED */
12323                         ctl_set_ua(&io->scsiio, 0x29, 0x00);
12324                         break;
12325                 case CTL_LUN_INJ_CUSTOM:
12326                         /*
12327                          * We're assuming the user knows what he is doing.
12328                          * Just copy the sense information without doing
12329                          * checks.
12330                          */
12331                         bcopy(&desc->custom_sense, &io->scsiio.sense_data,
12332                               MIN(sizeof(desc->custom_sense),
12333                                   sizeof(io->scsiio.sense_data)));
12334                         io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
12335                         io->scsiio.sense_len = SSD_FULL_SIZE;
12336                         io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
12337                         break;
12338                 case CTL_LUN_INJ_NONE:
12339                 default:
12340                         /*
12341                          * If this is an error injection type we don't know
12342                          * about, clear the continuous flag (if it is set)
12343                          * so it will get deleted below.
12344                          */
12345                         desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
12346                         break;
12347                 }
12348                 /*
12349                  * By default, each error injection action is a one-shot
12350                  */
12351                 if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
12352                         continue;
12353
12354                 STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
12355
12356                 free(desc, M_CTL);
12357         }
12358 }
12359
12360 #ifdef CTL_IO_DELAY
12361 static void
12362 ctl_datamove_timer_wakeup(void *arg)
12363 {
12364         union ctl_io *io;
12365
12366         io = (union ctl_io *)arg;
12367
12368         ctl_datamove(io);
12369 }
12370 #endif /* CTL_IO_DELAY */
12371
12372 void
12373 ctl_datamove(union ctl_io *io)
12374 {
12375         void (*fe_datamove)(union ctl_io *io);
12376
12377         mtx_assert(&((struct ctl_softc *)CTL_SOFTC(io))->ctl_lock, MA_NOTOWNED);
12378
12379         CTL_DEBUG_PRINT(("ctl_datamove\n"));
12380
12381         /* No data transferred yet.  Frontend must update this when done. */
12382         io->scsiio.kern_data_resid = io->scsiio.kern_data_len;
12383
12384 #ifdef CTL_TIME_IO
12385         if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12386                 char str[256];
12387                 char path_str[64];
12388                 struct sbuf sb;
12389
12390                 ctl_scsi_path_string(io, path_str, sizeof(path_str));
12391                 sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12392
12393                 sbuf_cat(&sb, path_str);
12394                 switch (io->io_hdr.io_type) {
12395                 case CTL_IO_SCSI:
12396                         ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12397                         sbuf_printf(&sb, "\n");
12398                         sbuf_cat(&sb, path_str);
12399                         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12400                                     io->scsiio.tag_num, io->scsiio.tag_type);
12401                         break;
12402                 case CTL_IO_TASK:
12403                         sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12404                                     "Tag Type: %d\n", io->taskio.task_action,
12405                                     io->taskio.tag_num, io->taskio.tag_type);
12406                         break;
12407                 default:
12408                         panic("%s: Invalid CTL I/O type %d\n",
12409                             __func__, io->io_hdr.io_type);
12410                 }
12411                 sbuf_cat(&sb, path_str);
12412                 sbuf_printf(&sb, "ctl_datamove: %jd seconds\n",
12413                             (intmax_t)time_uptime - io->io_hdr.start_time);
12414                 sbuf_finish(&sb);
12415                 printf("%s", sbuf_data(&sb));
12416         }
12417 #endif /* CTL_TIME_IO */
12418
12419 #ifdef CTL_IO_DELAY
12420         if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
12421                 io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
12422         } else {
12423                 struct ctl_lun *lun;
12424
12425                 lun = CTL_LUN(io);
12426                 if ((lun != NULL)
12427                  && (lun->delay_info.datamove_delay > 0)) {
12428
12429                         callout_init(&io->io_hdr.delay_callout, /*mpsafe*/ 1);
12430                         io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
12431                         callout_reset(&io->io_hdr.delay_callout,
12432                                       lun->delay_info.datamove_delay * hz,
12433                                       ctl_datamove_timer_wakeup, io);
12434                         if (lun->delay_info.datamove_type ==
12435                             CTL_DELAY_TYPE_ONESHOT)
12436                                 lun->delay_info.datamove_delay = 0;
12437                         return;
12438                 }
12439         }
12440 #endif
12441
12442         /*
12443          * This command has been aborted.  Set the port status, so we fail
12444          * the data move.
12445          */
12446         if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12447                 printf("ctl_datamove: tag 0x%04x on (%u:%u:%u) aborted\n",
12448                        io->scsiio.tag_num, io->io_hdr.nexus.initid,
12449                        io->io_hdr.nexus.targ_port,
12450                        io->io_hdr.nexus.targ_lun);
12451                 io->io_hdr.port_status = 31337;
12452                 /*
12453                  * Note that the backend, in this case, will get the
12454                  * callback in its context.  In other cases it may get
12455                  * called in the frontend's interrupt thread context.
12456                  */
12457                 io->scsiio.be_move_done(io);
12458                 return;
12459         }
12460
12461         /* Don't confuse frontend with zero length data move. */
12462         if (io->scsiio.kern_data_len == 0) {
12463                 io->scsiio.be_move_done(io);
12464                 return;
12465         }
12466
12467         fe_datamove = CTL_PORT(io)->fe_datamove;
12468         fe_datamove(io);
12469 }
12470
12471 static void
12472 ctl_send_datamove_done(union ctl_io *io, int have_lock)
12473 {
12474         union ctl_ha_msg msg;
12475 #ifdef CTL_TIME_IO
12476         struct bintime cur_bt;
12477 #endif
12478
12479         memset(&msg, 0, sizeof(msg));
12480         msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
12481         msg.hdr.original_sc = io;
12482         msg.hdr.serializing_sc = io->io_hdr.remote_io;
12483         msg.hdr.nexus = io->io_hdr.nexus;
12484         msg.hdr.status = io->io_hdr.status;
12485         msg.scsi.kern_data_resid = io->scsiio.kern_data_resid;
12486         msg.scsi.tag_num = io->scsiio.tag_num;
12487         msg.scsi.tag_type = io->scsiio.tag_type;
12488         msg.scsi.scsi_status = io->scsiio.scsi_status;
12489         memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
12490                io->scsiio.sense_len);
12491         msg.scsi.sense_len = io->scsiio.sense_len;
12492         msg.scsi.port_status = io->io_hdr.port_status;
12493         io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12494         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12495                 ctl_failover_io(io, /*have_lock*/ have_lock);
12496                 return;
12497         }
12498         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12499             sizeof(msg.scsi) - sizeof(msg.scsi.sense_data) +
12500             msg.scsi.sense_len, M_WAITOK);
12501
12502 #ifdef CTL_TIME_IO
12503         getbinuptime(&cur_bt);
12504         bintime_sub(&cur_bt, &io->io_hdr.dma_start_bt);
12505         bintime_add(&io->io_hdr.dma_bt, &cur_bt);
12506 #endif
12507         io->io_hdr.num_dmas++;
12508 }
12509
12510 /*
12511  * The DMA to the remote side is done, now we need to tell the other side
12512  * we're done so it can continue with its data movement.
12513  */
12514 static void
12515 ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
12516 {
12517         union ctl_io *io;
12518         uint32_t i;
12519
12520         io = rq->context;
12521
12522         if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12523                 printf("%s: ISC DMA write failed with error %d", __func__,
12524                        rq->ret);
12525                 ctl_set_internal_failure(&io->scsiio,
12526                                          /*sks_valid*/ 1,
12527                                          /*retry_count*/ rq->ret);
12528         }
12529
12530         ctl_dt_req_free(rq);
12531
12532         for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12533                 free(CTL_LSGLT(io)[i].addr, M_CTL);
12534         free(CTL_RSGL(io), M_CTL);
12535         CTL_RSGL(io) = NULL;
12536         CTL_LSGL(io) = NULL;
12537
12538         /*
12539          * The data is in local and remote memory, so now we need to send
12540          * status (good or back) back to the other side.
12541          */
12542         ctl_send_datamove_done(io, /*have_lock*/ 0);
12543 }
12544
12545 /*
12546  * We've moved the data from the host/controller into local memory.  Now we
12547  * need to push it over to the remote controller's memory.
12548  */
12549 static int
12550 ctl_datamove_remote_dm_write_cb(union ctl_io *io)
12551 {
12552         int retval;
12553
12554         retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
12555                                           ctl_datamove_remote_write_cb);
12556         return (retval);
12557 }
12558
12559 static void
12560 ctl_datamove_remote_write(union ctl_io *io)
12561 {
12562         int retval;
12563         void (*fe_datamove)(union ctl_io *io);
12564
12565         /*
12566          * - Get the data from the host/HBA into local memory.
12567          * - DMA memory from the local controller to the remote controller.
12568          * - Send status back to the remote controller.
12569          */
12570
12571         retval = ctl_datamove_remote_sgl_setup(io);
12572         if (retval != 0)
12573                 return;
12574
12575         /* Switch the pointer over so the FETD knows what to do */
12576         io->scsiio.kern_data_ptr = (uint8_t *)CTL_LSGL(io);
12577
12578         /*
12579          * Use a custom move done callback, since we need to send completion
12580          * back to the other controller, not to the backend on this side.
12581          */
12582         io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
12583
12584         fe_datamove = CTL_PORT(io)->fe_datamove;
12585         fe_datamove(io);
12586 }
12587
12588 static int
12589 ctl_datamove_remote_dm_read_cb(union ctl_io *io)
12590 {
12591 #if 0
12592         char str[256];
12593         char path_str[64];
12594         struct sbuf sb;
12595 #endif
12596         uint32_t i;
12597
12598         for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12599                 free(CTL_LSGLT(io)[i].addr, M_CTL);
12600         free(CTL_RSGL(io), M_CTL);
12601         CTL_RSGL(io) = NULL;
12602         CTL_LSGL(io) = NULL;
12603
12604 #if 0
12605         scsi_path_string(io, path_str, sizeof(path_str));
12606         sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12607         sbuf_cat(&sb, path_str);
12608         scsi_command_string(&io->scsiio, NULL, &sb);
12609         sbuf_printf(&sb, "\n");
12610         sbuf_cat(&sb, path_str);
12611         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12612                     io->scsiio.tag_num, io->scsiio.tag_type);
12613         sbuf_cat(&sb, path_str);
12614         sbuf_printf(&sb, "%s: flags %#x, status %#x\n", __func__,
12615                     io->io_hdr.flags, io->io_hdr.status);
12616         sbuf_finish(&sb);
12617         printk("%s", sbuf_data(&sb));
12618 #endif
12619
12620
12621         /*
12622          * The read is done, now we need to send status (good or bad) back
12623          * to the other side.
12624          */
12625         ctl_send_datamove_done(io, /*have_lock*/ 0);
12626
12627         return (0);
12628 }
12629
12630 static void
12631 ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
12632 {
12633         union ctl_io *io;
12634         void (*fe_datamove)(union ctl_io *io);
12635
12636         io = rq->context;
12637
12638         if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12639                 printf("%s: ISC DMA read failed with error %d\n", __func__,
12640                        rq->ret);
12641                 ctl_set_internal_failure(&io->scsiio,
12642                                          /*sks_valid*/ 1,
12643                                          /*retry_count*/ rq->ret);
12644         }
12645
12646         ctl_dt_req_free(rq);
12647
12648         /* Switch the pointer over so the FETD knows what to do */
12649         io->scsiio.kern_data_ptr = (uint8_t *)CTL_LSGL(io);
12650
12651         /*
12652          * Use a custom move done callback, since we need to send completion
12653          * back to the other controller, not to the backend on this side.
12654          */
12655         io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
12656
12657         /* XXX KDM add checks like the ones in ctl_datamove? */
12658
12659         fe_datamove = CTL_PORT(io)->fe_datamove;
12660         fe_datamove(io);
12661 }
12662
12663 static int
12664 ctl_datamove_remote_sgl_setup(union ctl_io *io)
12665 {
12666         struct ctl_sg_entry *local_sglist;
12667         uint32_t len_to_go;
12668         int retval;
12669         int i;
12670
12671         retval = 0;
12672         local_sglist = CTL_LSGL(io);
12673         len_to_go = io->scsiio.kern_data_len;
12674
12675         /*
12676          * The difficult thing here is that the size of the various
12677          * S/G segments may be different than the size from the
12678          * remote controller.  That'll make it harder when DMAing
12679          * the data back to the other side.
12680          */
12681         for (i = 0; len_to_go > 0; i++) {
12682                 local_sglist[i].len = MIN(len_to_go, CTL_HA_DATAMOVE_SEGMENT);
12683                 local_sglist[i].addr =
12684                     malloc(local_sglist[i].len, M_CTL, M_WAITOK);
12685
12686                 len_to_go -= local_sglist[i].len;
12687         }
12688         /*
12689          * Reset the number of S/G entries accordingly.  The original
12690          * number of S/G entries is available in rem_sg_entries.
12691          */
12692         io->scsiio.kern_sg_entries = i;
12693
12694 #if 0
12695         printf("%s: kern_sg_entries = %d\n", __func__,
12696                io->scsiio.kern_sg_entries);
12697         for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12698                 printf("%s: sg[%d] = %p, %lu\n", __func__, i,
12699                        local_sglist[i].addr, local_sglist[i].len);
12700 #endif
12701
12702         return (retval);
12703 }
12704
12705 static int
12706 ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
12707                          ctl_ha_dt_cb callback)
12708 {
12709         struct ctl_ha_dt_req *rq;
12710         struct ctl_sg_entry *remote_sglist, *local_sglist;
12711         uint32_t local_used, remote_used, total_used;
12712         int i, j, isc_ret;
12713
12714         rq = ctl_dt_req_alloc();
12715
12716         /*
12717          * If we failed to allocate the request, and if the DMA didn't fail
12718          * anyway, set busy status.  This is just a resource allocation
12719          * failure.
12720          */
12721         if ((rq == NULL)
12722          && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
12723              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS))
12724                 ctl_set_busy(&io->scsiio);
12725
12726         if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
12727             (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS) {
12728
12729                 if (rq != NULL)
12730                         ctl_dt_req_free(rq);
12731
12732                 /*
12733                  * The data move failed.  We need to return status back
12734                  * to the other controller.  No point in trying to DMA
12735                  * data to the remote controller.
12736                  */
12737
12738                 ctl_send_datamove_done(io, /*have_lock*/ 0);
12739
12740                 return (1);
12741         }
12742
12743         local_sglist = CTL_LSGL(io);
12744         remote_sglist = CTL_RSGL(io);
12745         local_used = 0;
12746         remote_used = 0;
12747         total_used = 0;
12748
12749         /*
12750          * Pull/push the data over the wire from/to the other controller.
12751          * This takes into account the possibility that the local and
12752          * remote sglists may not be identical in terms of the size of
12753          * the elements and the number of elements.
12754          *
12755          * One fundamental assumption here is that the length allocated for
12756          * both the local and remote sglists is identical.  Otherwise, we've
12757          * essentially got a coding error of some sort.
12758          */
12759         isc_ret = CTL_HA_STATUS_SUCCESS;
12760         for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
12761                 uint32_t cur_len;
12762                 uint8_t *tmp_ptr;
12763
12764                 rq->command = command;
12765                 rq->context = io;
12766
12767                 /*
12768                  * Both pointers should be aligned.  But it is possible
12769                  * that the allocation length is not.  They should both
12770                  * also have enough slack left over at the end, though,
12771                  * to round up to the next 8 byte boundary.
12772                  */
12773                 cur_len = MIN(local_sglist[i].len - local_used,
12774                               remote_sglist[j].len - remote_used);
12775                 rq->size = cur_len;
12776
12777                 tmp_ptr = (uint8_t *)local_sglist[i].addr;
12778                 tmp_ptr += local_used;
12779
12780 #if 0
12781                 /* Use physical addresses when talking to ISC hardware */
12782                 if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
12783                         /* XXX KDM use busdma */
12784                         rq->local = vtophys(tmp_ptr);
12785                 } else
12786                         rq->local = tmp_ptr;
12787 #else
12788                 KASSERT((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0,
12789                     ("HA does not support BUS_ADDR"));
12790                 rq->local = tmp_ptr;
12791 #endif
12792
12793                 tmp_ptr = (uint8_t *)remote_sglist[j].addr;
12794                 tmp_ptr += remote_used;
12795                 rq->remote = tmp_ptr;
12796
12797                 rq->callback = NULL;
12798
12799                 local_used += cur_len;
12800                 if (local_used >= local_sglist[i].len) {
12801                         i++;
12802                         local_used = 0;
12803                 }
12804
12805                 remote_used += cur_len;
12806                 if (remote_used >= remote_sglist[j].len) {
12807                         j++;
12808                         remote_used = 0;
12809                 }
12810                 total_used += cur_len;
12811
12812                 if (total_used >= io->scsiio.kern_data_len)
12813                         rq->callback = callback;
12814
12815 #if 0
12816                 printf("%s: %s: local %p remote %p size %d\n", __func__,
12817                        (command == CTL_HA_DT_CMD_WRITE) ? "WRITE" : "READ",
12818                        rq->local, rq->remote, rq->size);
12819 #endif
12820
12821                 isc_ret = ctl_dt_single(rq);
12822                 if (isc_ret > CTL_HA_STATUS_SUCCESS)
12823                         break;
12824         }
12825         if (isc_ret != CTL_HA_STATUS_WAIT) {
12826                 rq->ret = isc_ret;
12827                 callback(rq);
12828         }
12829
12830         return (0);
12831 }
12832
12833 static void
12834 ctl_datamove_remote_read(union ctl_io *io)
12835 {
12836         int retval;
12837         uint32_t i;
12838
12839         /*
12840          * This will send an error to the other controller in the case of a
12841          * failure.
12842          */
12843         retval = ctl_datamove_remote_sgl_setup(io);
12844         if (retval != 0)
12845                 return;
12846
12847         retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
12848                                           ctl_datamove_remote_read_cb);
12849         if (retval != 0) {
12850                 /*
12851                  * Make sure we free memory if there was an error..  The
12852                  * ctl_datamove_remote_xfer() function will send the
12853                  * datamove done message, or call the callback with an
12854                  * error if there is a problem.
12855                  */
12856                 for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12857                         free(CTL_LSGLT(io)[i].addr, M_CTL);
12858                 free(CTL_RSGL(io), M_CTL);
12859                 CTL_RSGL(io) = NULL;
12860                 CTL_LSGL(io) = NULL;
12861         }
12862 }
12863
12864 /*
12865  * Process a datamove request from the other controller.  This is used for
12866  * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
12867  * first.  Once that is complete, the data gets DMAed into the remote
12868  * controller's memory.  For reads, we DMA from the remote controller's
12869  * memory into our memory first, and then move it out to the FETD.
12870  */
12871 static void
12872 ctl_datamove_remote(union ctl_io *io)
12873 {
12874
12875         mtx_assert(&((struct ctl_softc *)CTL_SOFTC(io))->ctl_lock, MA_NOTOWNED);
12876
12877         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12878                 ctl_failover_io(io, /*have_lock*/ 0);
12879                 return;
12880         }
12881
12882         /*
12883          * Note that we look for an aborted I/O here, but don't do some of
12884          * the other checks that ctl_datamove() normally does.
12885          * We don't need to run the datamove delay code, since that should
12886          * have been done if need be on the other controller.
12887          */
12888         if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12889                 printf("%s: tag 0x%04x on (%u:%u:%u) aborted\n", __func__,
12890                        io->scsiio.tag_num, io->io_hdr.nexus.initid,
12891                        io->io_hdr.nexus.targ_port,
12892                        io->io_hdr.nexus.targ_lun);
12893                 io->io_hdr.port_status = 31338;
12894                 ctl_send_datamove_done(io, /*have_lock*/ 0);
12895                 return;
12896         }
12897
12898         if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT)
12899                 ctl_datamove_remote_write(io);
12900         else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN)
12901                 ctl_datamove_remote_read(io);
12902         else {
12903                 io->io_hdr.port_status = 31339;
12904                 ctl_send_datamove_done(io, /*have_lock*/ 0);
12905         }
12906 }
12907
12908 static void
12909 ctl_process_done(union ctl_io *io)
12910 {
12911         struct ctl_softc *softc = CTL_SOFTC(io);
12912         struct ctl_port *port = CTL_PORT(io);
12913         struct ctl_lun *lun = CTL_LUN(io);
12914         void (*fe_done)(union ctl_io *io);
12915         union ctl_ha_msg msg;
12916
12917         CTL_DEBUG_PRINT(("ctl_process_done\n"));
12918         fe_done = port->fe_done;
12919
12920 #ifdef CTL_TIME_IO
12921         if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12922                 char str[256];
12923                 char path_str[64];
12924                 struct sbuf sb;
12925
12926                 ctl_scsi_path_string(io, path_str, sizeof(path_str));
12927                 sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12928
12929                 sbuf_cat(&sb, path_str);
12930                 switch (io->io_hdr.io_type) {
12931                 case CTL_IO_SCSI:
12932                         ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12933                         sbuf_printf(&sb, "\n");
12934                         sbuf_cat(&sb, path_str);
12935                         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12936                                     io->scsiio.tag_num, io->scsiio.tag_type);
12937                         break;
12938                 case CTL_IO_TASK:
12939                         sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12940                                     "Tag Type: %d\n", io->taskio.task_action,
12941                                     io->taskio.tag_num, io->taskio.tag_type);
12942                         break;
12943                 default:
12944                         panic("%s: Invalid CTL I/O type %d\n",
12945                             __func__, io->io_hdr.io_type);
12946                 }
12947                 sbuf_cat(&sb, path_str);
12948                 sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
12949                             (intmax_t)time_uptime - io->io_hdr.start_time);
12950                 sbuf_finish(&sb);
12951                 printf("%s", sbuf_data(&sb));
12952         }
12953 #endif /* CTL_TIME_IO */
12954
12955         switch (io->io_hdr.io_type) {
12956         case CTL_IO_SCSI:
12957                 break;
12958         case CTL_IO_TASK:
12959                 if (ctl_debug & CTL_DEBUG_INFO)
12960                         ctl_io_error_print(io, NULL);
12961                 fe_done(io);
12962                 return;
12963         default:
12964                 panic("%s: Invalid CTL I/O type %d\n",
12965                     __func__, io->io_hdr.io_type);
12966         }
12967
12968         if (lun == NULL) {
12969                 CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
12970                                  io->io_hdr.nexus.targ_mapped_lun));
12971                 goto bailout;
12972         }
12973
12974         mtx_lock(&lun->lun_lock);
12975
12976         /*
12977          * Check to see if we have any informational exception and status
12978          * of this command can be modified to report it in form of either
12979          * RECOVERED ERROR or NO SENSE, depending on MRIE mode page field.
12980          */
12981         if (lun->ie_reported == 0 && lun->ie_asc != 0 &&
12982             io->io_hdr.status == CTL_SUCCESS &&
12983             (io->io_hdr.flags & CTL_FLAG_STATUS_SENT) == 0) {
12984                 uint8_t mrie = lun->MODE_IE.mrie;
12985                 uint8_t per = ((lun->MODE_RWER.byte3 & SMS_RWER_PER) ||
12986                     (lun->MODE_VER.byte3 & SMS_VER_PER));
12987                 if (((mrie == SIEP_MRIE_REC_COND && per) ||
12988                      mrie == SIEP_MRIE_REC_UNCOND ||
12989                      mrie == SIEP_MRIE_NO_SENSE) &&
12990                     (ctl_get_cmd_entry(&io->scsiio, NULL)->flags &
12991                      CTL_CMD_FLAG_NO_SENSE) == 0) {
12992                         ctl_set_sense(&io->scsiio,
12993                               /*current_error*/ 1,
12994                               /*sense_key*/ (mrie == SIEP_MRIE_NO_SENSE) ?
12995                                 SSD_KEY_NO_SENSE : SSD_KEY_RECOVERED_ERROR,
12996                               /*asc*/ lun->ie_asc,
12997                               /*ascq*/ lun->ie_ascq,
12998                               SSD_ELEM_NONE);
12999                         lun->ie_reported = 1;
13000                 }
13001         } else if (lun->ie_reported < 0)
13002                 lun->ie_reported = 0;
13003
13004         /*
13005          * Check to see if we have any errors to inject here.  We only
13006          * inject errors for commands that don't already have errors set.
13007          */
13008         if (!STAILQ_EMPTY(&lun->error_list) &&
13009             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS) &&
13010             ((io->io_hdr.flags & CTL_FLAG_STATUS_SENT) == 0))
13011                 ctl_inject_error(lun, io);
13012
13013         /*
13014          * XXX KDM how do we treat commands that aren't completed
13015          * successfully?
13016          *
13017          * XXX KDM should we also track I/O latency?
13018          */
13019         if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS &&
13020             io->io_hdr.io_type == CTL_IO_SCSI) {
13021                 int type;
13022 #ifdef CTL_TIME_IO
13023                 struct bintime bt;
13024
13025                 getbinuptime(&bt);
13026                 bintime_sub(&bt, &io->io_hdr.start_bt);
13027 #endif
13028                 if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13029                     CTL_FLAG_DATA_IN)
13030                         type = CTL_STATS_READ;
13031                 else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13032                     CTL_FLAG_DATA_OUT)
13033                         type = CTL_STATS_WRITE;
13034                 else
13035                         type = CTL_STATS_NO_IO;
13036
13037                 lun->stats.bytes[type] += io->scsiio.kern_total_len;
13038                 lun->stats.operations[type] ++;
13039                 lun->stats.dmas[type] += io->io_hdr.num_dmas;
13040 #ifdef CTL_TIME_IO
13041                 bintime_add(&lun->stats.dma_time[type], &io->io_hdr.dma_bt);
13042                 bintime_add(&lun->stats.time[type], &bt);
13043 #endif
13044
13045                 mtx_lock(&port->port_lock);
13046                 port->stats.bytes[type] += io->scsiio.kern_total_len;
13047                 port->stats.operations[type] ++;
13048                 port->stats.dmas[type] += io->io_hdr.num_dmas;
13049 #ifdef CTL_TIME_IO
13050                 bintime_add(&port->stats.dma_time[type], &io->io_hdr.dma_bt);
13051                 bintime_add(&port->stats.time[type], &bt);
13052 #endif
13053                 mtx_unlock(&port->port_lock);
13054         }
13055
13056         /*
13057          * Remove this from the OOA queue.
13058          */
13059         TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
13060 #ifdef CTL_TIME_IO
13061         if (TAILQ_EMPTY(&lun->ooa_queue))
13062                 lun->last_busy = getsbinuptime();
13063 #endif
13064
13065         /*
13066          * Run through the blocked queue on this LUN and see if anything
13067          * has become unblocked, now that this transaction is done.
13068          */
13069         ctl_check_blocked(lun);
13070
13071         /*
13072          * If the LUN has been invalidated, free it if there is nothing
13073          * left on its OOA queue.
13074          */
13075         if ((lun->flags & CTL_LUN_INVALID)
13076          && TAILQ_EMPTY(&lun->ooa_queue)) {
13077                 mtx_unlock(&lun->lun_lock);
13078                 ctl_free_lun(lun);
13079         } else
13080                 mtx_unlock(&lun->lun_lock);
13081
13082 bailout:
13083
13084         /*
13085          * If this command has been aborted, make sure we set the status
13086          * properly.  The FETD is responsible for freeing the I/O and doing
13087          * whatever it needs to do to clean up its state.
13088          */
13089         if (io->io_hdr.flags & CTL_FLAG_ABORT)
13090                 ctl_set_task_aborted(&io->scsiio);
13091
13092         /*
13093          * If enabled, print command error status.
13094          */
13095         if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS &&
13096             (ctl_debug & CTL_DEBUG_INFO) != 0)
13097                 ctl_io_error_print(io, NULL);
13098
13099         /*
13100          * Tell the FETD or the other shelf controller we're done with this
13101          * command.  Note that only SCSI commands get to this point.  Task
13102          * management commands are completed above.
13103          */
13104         if ((softc->ha_mode != CTL_HA_MODE_XFER) &&
13105             (io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)) {
13106                 memset(&msg, 0, sizeof(msg));
13107                 msg.hdr.msg_type = CTL_MSG_FINISH_IO;
13108                 msg.hdr.serializing_sc = io->io_hdr.remote_io;
13109                 msg.hdr.nexus = io->io_hdr.nexus;
13110                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13111                     sizeof(msg.scsi) - sizeof(msg.scsi.sense_data),
13112                     M_WAITOK);
13113         }
13114
13115         fe_done(io);
13116 }
13117
13118 /*
13119  * Front end should call this if it doesn't do autosense.  When the request
13120  * sense comes back in from the initiator, we'll dequeue this and send it.
13121  */
13122 int
13123 ctl_queue_sense(union ctl_io *io)
13124 {
13125         struct ctl_softc *softc = CTL_SOFTC(io);
13126         struct ctl_port *port = CTL_PORT(io);
13127         struct ctl_lun *lun;
13128         struct scsi_sense_data *ps;
13129         uint32_t initidx, p, targ_lun;
13130
13131         CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
13132
13133         targ_lun = ctl_lun_map_from_port(port, io->io_hdr.nexus.targ_lun);
13134
13135         /*
13136          * LUN lookup will likely move to the ctl_work_thread() once we
13137          * have our new queueing infrastructure (that doesn't put things on
13138          * a per-LUN queue initially).  That is so that we can handle
13139          * things like an INQUIRY to a LUN that we don't have enabled.  We
13140          * can't deal with that right now.
13141          * If we don't have a LUN for this, just toss the sense information.
13142          */
13143         mtx_lock(&softc->ctl_lock);
13144         if (targ_lun >= ctl_max_luns ||
13145             (lun = softc->ctl_luns[targ_lun]) == NULL) {
13146                 mtx_unlock(&softc->ctl_lock);
13147                 goto bailout;
13148         }
13149         mtx_lock(&lun->lun_lock);
13150         mtx_unlock(&softc->ctl_lock);
13151
13152         initidx = ctl_get_initindex(&io->io_hdr.nexus);
13153         p = initidx / CTL_MAX_INIT_PER_PORT;
13154         if (lun->pending_sense[p] == NULL) {
13155                 lun->pending_sense[p] = malloc(sizeof(*ps) * CTL_MAX_INIT_PER_PORT,
13156                     M_CTL, M_NOWAIT | M_ZERO);
13157         }
13158         if ((ps = lun->pending_sense[p]) != NULL) {
13159                 ps += initidx % CTL_MAX_INIT_PER_PORT;
13160                 memset(ps, 0, sizeof(*ps));
13161                 memcpy(ps, &io->scsiio.sense_data, io->scsiio.sense_len);
13162         }
13163         mtx_unlock(&lun->lun_lock);
13164
13165 bailout:
13166         ctl_free_io(io);
13167         return (CTL_RETVAL_COMPLETE);
13168 }
13169
13170 /*
13171  * Primary command inlet from frontend ports.  All SCSI and task I/O
13172  * requests must go through this function.
13173  */
13174 int
13175 ctl_queue(union ctl_io *io)
13176 {
13177         struct ctl_port *port = CTL_PORT(io);
13178
13179         CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
13180
13181 #ifdef CTL_TIME_IO
13182         io->io_hdr.start_time = time_uptime;
13183         getbinuptime(&io->io_hdr.start_bt);
13184 #endif /* CTL_TIME_IO */
13185
13186         /* Map FE-specific LUN ID into global one. */
13187         io->io_hdr.nexus.targ_mapped_lun =
13188             ctl_lun_map_from_port(port, io->io_hdr.nexus.targ_lun);
13189
13190         switch (io->io_hdr.io_type) {
13191         case CTL_IO_SCSI:
13192         case CTL_IO_TASK:
13193                 if (ctl_debug & CTL_DEBUG_CDB)
13194                         ctl_io_print(io);
13195                 ctl_enqueue_incoming(io);
13196                 break;
13197         default:
13198                 printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
13199                 return (EINVAL);
13200         }
13201
13202         return (CTL_RETVAL_COMPLETE);
13203 }
13204
13205 #ifdef CTL_IO_DELAY
13206 static void
13207 ctl_done_timer_wakeup(void *arg)
13208 {
13209         union ctl_io *io;
13210
13211         io = (union ctl_io *)arg;
13212         ctl_done(io);
13213 }
13214 #endif /* CTL_IO_DELAY */
13215
13216 void
13217 ctl_serseq_done(union ctl_io *io)
13218 {
13219         struct ctl_lun *lun = CTL_LUN(io);;
13220
13221         if (lun->be_lun == NULL ||
13222             lun->be_lun->serseq == CTL_LUN_SERSEQ_OFF)
13223                 return;
13224         mtx_lock(&lun->lun_lock);
13225         io->io_hdr.flags |= CTL_FLAG_SERSEQ_DONE;
13226         ctl_check_blocked(lun);
13227         mtx_unlock(&lun->lun_lock);
13228 }
13229
13230 void
13231 ctl_done(union ctl_io *io)
13232 {
13233
13234         /*
13235          * Enable this to catch duplicate completion issues.
13236          */
13237 #if 0
13238         if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
13239                 printf("%s: type %d msg %d cdb %x iptl: "
13240                        "%u:%u:%u tag 0x%04x "
13241                        "flag %#x status %x\n",
13242                         __func__,
13243                         io->io_hdr.io_type,
13244                         io->io_hdr.msg_type,
13245                         io->scsiio.cdb[0],
13246                         io->io_hdr.nexus.initid,
13247                         io->io_hdr.nexus.targ_port,
13248                         io->io_hdr.nexus.targ_lun,
13249                         (io->io_hdr.io_type ==
13250                         CTL_IO_TASK) ?
13251                         io->taskio.tag_num :
13252                         io->scsiio.tag_num,
13253                         io->io_hdr.flags,
13254                         io->io_hdr.status);
13255         } else
13256                 io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
13257 #endif
13258
13259         /*
13260          * This is an internal copy of an I/O, and should not go through
13261          * the normal done processing logic.
13262          */
13263         if (io->io_hdr.flags & CTL_FLAG_INT_COPY)
13264                 return;
13265
13266 #ifdef CTL_IO_DELAY
13267         if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
13268                 io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
13269         } else {
13270                 struct ctl_lun *lun = CTL_LUN(io);
13271
13272                 if ((lun != NULL)
13273                  && (lun->delay_info.done_delay > 0)) {
13274
13275                         callout_init(&io->io_hdr.delay_callout, /*mpsafe*/ 1);
13276                         io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
13277                         callout_reset(&io->io_hdr.delay_callout,
13278                                       lun->delay_info.done_delay * hz,
13279                                       ctl_done_timer_wakeup, io);
13280                         if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
13281                                 lun->delay_info.done_delay = 0;
13282                         return;
13283                 }
13284         }
13285 #endif /* CTL_IO_DELAY */
13286
13287         ctl_enqueue_done(io);
13288 }
13289
13290 static void
13291 ctl_work_thread(void *arg)
13292 {
13293         struct ctl_thread *thr = (struct ctl_thread *)arg;
13294         struct ctl_softc *softc = thr->ctl_softc;
13295         union ctl_io *io;
13296         int retval;
13297
13298         CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
13299
13300         while (!softc->shutdown) {
13301                 /*
13302                  * We handle the queues in this order:
13303                  * - ISC
13304                  * - done queue (to free up resources, unblock other commands)
13305                  * - incoming queue
13306                  * - RtR queue
13307                  *
13308                  * If those queues are empty, we break out of the loop and
13309                  * go to sleep.
13310                  */
13311                 mtx_lock(&thr->queue_lock);
13312                 io = (union ctl_io *)STAILQ_FIRST(&thr->isc_queue);
13313                 if (io != NULL) {
13314                         STAILQ_REMOVE_HEAD(&thr->isc_queue, links);
13315                         mtx_unlock(&thr->queue_lock);
13316                         ctl_handle_isc(io);
13317                         continue;
13318                 }
13319                 io = (union ctl_io *)STAILQ_FIRST(&thr->done_queue);
13320                 if (io != NULL) {
13321                         STAILQ_REMOVE_HEAD(&thr->done_queue, links);
13322                         /* clear any blocked commands, call fe_done */
13323                         mtx_unlock(&thr->queue_lock);
13324                         ctl_process_done(io);
13325                         continue;
13326                 }
13327                 io = (union ctl_io *)STAILQ_FIRST(&thr->incoming_queue);
13328                 if (io != NULL) {
13329                         STAILQ_REMOVE_HEAD(&thr->incoming_queue, links);
13330                         mtx_unlock(&thr->queue_lock);
13331                         if (io->io_hdr.io_type == CTL_IO_TASK)
13332                                 ctl_run_task(io);
13333                         else
13334                                 ctl_scsiio_precheck(softc, &io->scsiio);
13335                         continue;
13336                 }
13337                 io = (union ctl_io *)STAILQ_FIRST(&thr->rtr_queue);
13338                 if (io != NULL) {
13339                         STAILQ_REMOVE_HEAD(&thr->rtr_queue, links);
13340                         mtx_unlock(&thr->queue_lock);
13341                         retval = ctl_scsiio(&io->scsiio);
13342                         if (retval != CTL_RETVAL_COMPLETE)
13343                                 CTL_DEBUG_PRINT(("ctl_scsiio failed\n"));
13344                         continue;
13345                 }
13346
13347                 /* Sleep until we have something to do. */
13348                 mtx_sleep(thr, &thr->queue_lock, PDROP | PRIBIO, "-", 0);
13349         }
13350         thr->thread = NULL;
13351         kthread_exit();
13352 }
13353
13354 static void
13355 ctl_lun_thread(void *arg)
13356 {
13357         struct ctl_softc *softc = (struct ctl_softc *)arg;
13358         struct ctl_be_lun *be_lun;
13359
13360         CTL_DEBUG_PRINT(("ctl_lun_thread starting\n"));
13361
13362         while (!softc->shutdown) {
13363                 mtx_lock(&softc->ctl_lock);
13364                 be_lun = STAILQ_FIRST(&softc->pending_lun_queue);
13365                 if (be_lun != NULL) {
13366                         STAILQ_REMOVE_HEAD(&softc->pending_lun_queue, links);
13367                         mtx_unlock(&softc->ctl_lock);
13368                         ctl_create_lun(be_lun);
13369                         continue;
13370                 }
13371
13372                 /* Sleep until we have something to do. */
13373                 mtx_sleep(&softc->pending_lun_queue, &softc->ctl_lock,
13374                     PDROP | PRIBIO, "-", 0);
13375         }
13376         softc->lun_thread = NULL;
13377         kthread_exit();
13378 }
13379
13380 static void
13381 ctl_thresh_thread(void *arg)
13382 {
13383         struct ctl_softc *softc = (struct ctl_softc *)arg;
13384         struct ctl_lun *lun;
13385         struct ctl_logical_block_provisioning_page *page;
13386         const char *attr;
13387         union ctl_ha_msg msg;
13388         uint64_t thres, val;
13389         int i, e, set;
13390
13391         CTL_DEBUG_PRINT(("ctl_thresh_thread starting\n"));
13392
13393         while (!softc->shutdown) {
13394                 mtx_lock(&softc->ctl_lock);
13395                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
13396                         if ((lun->flags & CTL_LUN_DISABLED) ||
13397                             (lun->flags & CTL_LUN_NO_MEDIA) ||
13398                             lun->backend->lun_attr == NULL)
13399                                 continue;
13400                         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
13401                             softc->ha_mode == CTL_HA_MODE_XFER)
13402                                 continue;
13403                         if ((lun->MODE_RWER.byte8 & SMS_RWER_LBPERE) == 0)
13404                                 continue;
13405                         e = 0;
13406                         page = &lun->MODE_LBP;
13407                         for (i = 0; i < CTL_NUM_LBP_THRESH; i++) {
13408                                 if ((page->descr[i].flags & SLBPPD_ENABLED) == 0)
13409                                         continue;
13410                                 thres = scsi_4btoul(page->descr[i].count);
13411                                 thres <<= CTL_LBP_EXPONENT;
13412                                 switch (page->descr[i].resource) {
13413                                 case 0x01:
13414                                         attr = "blocksavail";
13415                                         break;
13416                                 case 0x02:
13417                                         attr = "blocksused";
13418                                         break;
13419                                 case 0xf1:
13420                                         attr = "poolblocksavail";
13421                                         break;
13422                                 case 0xf2:
13423                                         attr = "poolblocksused";
13424                                         break;
13425                                 default:
13426                                         continue;
13427                                 }
13428                                 mtx_unlock(&softc->ctl_lock); // XXX
13429                                 val = lun->backend->lun_attr(
13430                                     lun->be_lun->be_lun, attr);
13431                                 mtx_lock(&softc->ctl_lock);
13432                                 if (val == UINT64_MAX)
13433                                         continue;
13434                                 if ((page->descr[i].flags & SLBPPD_ARMING_MASK)
13435                                     == SLBPPD_ARMING_INC)
13436                                         e = (val >= thres);
13437                                 else
13438                                         e = (val <= thres);
13439                                 if (e)
13440                                         break;
13441                         }
13442                         mtx_lock(&lun->lun_lock);
13443                         if (e) {
13444                                 scsi_u64to8b((uint8_t *)&page->descr[i] -
13445                                     (uint8_t *)page, lun->ua_tpt_info);
13446                                 if (lun->lasttpt == 0 ||
13447                                     time_uptime - lun->lasttpt >= CTL_LBP_UA_PERIOD) {
13448                                         lun->lasttpt = time_uptime;
13449                                         ctl_est_ua_all(lun, -1, CTL_UA_THIN_PROV_THRES);
13450                                         set = 1;
13451                                 } else
13452                                         set = 0;
13453                         } else {
13454                                 lun->lasttpt = 0;
13455                                 ctl_clr_ua_all(lun, -1, CTL_UA_THIN_PROV_THRES);
13456                                 set = -1;
13457                         }
13458                         mtx_unlock(&lun->lun_lock);
13459                         if (set != 0 &&
13460                             lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
13461                                 /* Send msg to other side. */
13462                                 bzero(&msg.ua, sizeof(msg.ua));
13463                                 msg.hdr.msg_type = CTL_MSG_UA;
13464                                 msg.hdr.nexus.initid = -1;
13465                                 msg.hdr.nexus.targ_port = -1;
13466                                 msg.hdr.nexus.targ_lun = lun->lun;
13467                                 msg.hdr.nexus.targ_mapped_lun = lun->lun;
13468                                 msg.ua.ua_all = 1;
13469                                 msg.ua.ua_set = (set > 0);
13470                                 msg.ua.ua_type = CTL_UA_THIN_PROV_THRES;
13471                                 memcpy(msg.ua.ua_info, lun->ua_tpt_info, 8);
13472                                 mtx_unlock(&softc->ctl_lock); // XXX
13473                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13474                                     sizeof(msg.ua), M_WAITOK);
13475                                 mtx_lock(&softc->ctl_lock);
13476                         }
13477                 }
13478                 mtx_sleep(&softc->thresh_thread, &softc->ctl_lock,
13479                     PDROP | PRIBIO, "-", CTL_LBP_PERIOD * hz);
13480         }
13481         softc->thresh_thread = NULL;
13482         kthread_exit();
13483 }
13484
13485 static void
13486 ctl_enqueue_incoming(union ctl_io *io)
13487 {
13488         struct ctl_softc *softc = CTL_SOFTC(io);
13489         struct ctl_thread *thr;
13490         u_int idx;
13491
13492         idx = (io->io_hdr.nexus.targ_port * 127 +
13493                io->io_hdr.nexus.initid) % worker_threads;
13494         thr = &softc->threads[idx];
13495         mtx_lock(&thr->queue_lock);
13496         STAILQ_INSERT_TAIL(&thr->incoming_queue, &io->io_hdr, links);
13497         mtx_unlock(&thr->queue_lock);
13498         wakeup(thr);
13499 }
13500
13501 static void
13502 ctl_enqueue_rtr(union ctl_io *io)
13503 {
13504         struct ctl_softc *softc = CTL_SOFTC(io);
13505         struct ctl_thread *thr;
13506
13507         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13508         mtx_lock(&thr->queue_lock);
13509         STAILQ_INSERT_TAIL(&thr->rtr_queue, &io->io_hdr, links);
13510         mtx_unlock(&thr->queue_lock);
13511         wakeup(thr);
13512 }
13513
13514 static void
13515 ctl_enqueue_done(union ctl_io *io)
13516 {
13517         struct ctl_softc *softc = CTL_SOFTC(io);
13518         struct ctl_thread *thr;
13519
13520         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13521         mtx_lock(&thr->queue_lock);
13522         STAILQ_INSERT_TAIL(&thr->done_queue, &io->io_hdr, links);
13523         mtx_unlock(&thr->queue_lock);
13524         wakeup(thr);
13525 }
13526
13527 static void
13528 ctl_enqueue_isc(union ctl_io *io)
13529 {
13530         struct ctl_softc *softc = CTL_SOFTC(io);
13531         struct ctl_thread *thr;
13532
13533         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13534         mtx_lock(&thr->queue_lock);
13535         STAILQ_INSERT_TAIL(&thr->isc_queue, &io->io_hdr, links);
13536         mtx_unlock(&thr->queue_lock);
13537         wakeup(thr);
13538 }
13539
13540 /*
13541  *  vim: ts=8
13542  */