sys/cam/ctl/ctl.c

   1 /*-
   2  * SPDX-License-Identifier: BSD-2-Clause
   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 #include <sys/param.h>
  49 #include <sys/systm.h>
  50 #include <sys/ctype.h>
  51 #include <sys/kernel.h>
  52 #include <sys/types.h>
  53 #include <sys/kthread.h>
  54 #include <sys/bio.h>
  55 #include <sys/fcntl.h>
  56 #include <sys/lock.h>
  57 #include <sys/module.h>
  58 #include <sys/mutex.h>
  59 #include <sys/condvar.h>
  60 #include <sys/malloc.h>
  61 #include <sys/conf.h>
  62 #include <sys/ioccom.h>
  63 #include <sys/queue.h>
  64 #include <sys/sbuf.h>
  65 #include <sys/smp.h>
  66 #include <sys/endian.h>
  67 #include <sys/proc.h>
  68 #include <sys/sched.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 | CTLFLAG_MPSAFE, 0,
 408     "CAM Target Layer");
 409 static int worker_threads = -1;
 410 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, worker_threads, CTLFLAG_RDTUN,
 411     &worker_threads, 1, "Number of worker threads");
 412 static int ctl_debug = CTL_DEBUG_NONE;
 413 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, debug, CTLFLAG_RWTUN,
 414     &ctl_debug, 0, "Enabled debug flags");
 415 static int ctl_lun_map_size = 1024;
 416 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, lun_map_size, CTLFLAG_RWTUN,
 417     &ctl_lun_map_size, 0, "Size of per-port LUN map (max LUN + 1)");
 418 #ifdef  CTL_TIME_IO
 419 static int ctl_time_io_secs = CTL_TIME_IO_DEFAULT_SECS;
 420 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, time_io_secs, CTLFLAG_RWTUN,
 421     &ctl_time_io_secs, 0, "Log requests taking more seconds");
 422 #endif
 423
 424 /*
 425  * Maximum number of LUNs we support.  MUST be a power of 2.
 426  */
 427 #define CTL_DEFAULT_MAX_LUNS    1024
 428 static int ctl_max_luns = CTL_DEFAULT_MAX_LUNS;
 429 TUNABLE_INT("kern.cam.ctl.max_luns", &ctl_max_luns);
 430 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, max_luns, CTLFLAG_RDTUN,
 431     &ctl_max_luns, CTL_DEFAULT_MAX_LUNS, "Maximum number of LUNs");
 432
 433 /*
 434  * Maximum number of ports registered at one time.
 435  */
 436 #define CTL_DEFAULT_MAX_PORTS           1024
 437 static int ctl_max_ports = CTL_DEFAULT_MAX_PORTS;
 438 TUNABLE_INT("kern.cam.ctl.max_ports", &ctl_max_ports);
 439 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, max_ports, CTLFLAG_RDTUN,
 440     &ctl_max_ports, CTL_DEFAULT_MAX_LUNS, "Maximum number of ports");
 441
 442 /*
 443  * Maximum number of initiators we support.
 444  */
 445 #define CTL_MAX_INITIATORS      (CTL_MAX_INIT_PER_PORT * ctl_max_ports)
 446
 447 /*
 448  * Supported pages (0x00), Serial number (0x80), Device ID (0x83),
 449  * Extended INQUIRY Data (0x86), Mode Page Policy (0x87),
 450  * SCSI Ports (0x88), Third-party Copy (0x8F), SCSI Feature Sets (0x92),
 451  * Block limits (0xB0), Block Device Characteristics (0xB1) and
 452  * Logical Block Provisioning (0xB2)
 453  */
 454 #define SCSI_EVPD_NUM_SUPPORTED_PAGES   11
 455
 456 static void ctl_isc_event_handler(ctl_ha_channel chanel, ctl_ha_event event,
 457                                   int param);
 458 static void ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest);
 459 static void ctl_copy_sense_data_back(union ctl_io *src, union ctl_ha_msg *dest);
 460 static int ctl_init(void);
 461 static int ctl_shutdown(void);
 462 static int ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td);
 463 static int ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td);
 464 static void ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio);
 465 static void ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
 466                               struct ctl_ooa *ooa_hdr,
 467                               struct ctl_ooa_entry *kern_entries);
 468 static int ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
 469                      struct thread *td);
 470 static int ctl_enable_lun(struct ctl_lun *lun);
 471 static int ctl_disable_lun(struct ctl_lun *lun);
 472 static int ctl_free_lun(struct ctl_lun *lun);
 473
 474 static int ctl_do_mode_select(union ctl_io *io);
 475 static int ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun,
 476                            uint64_t res_key, uint64_t sa_res_key,
 477                            uint8_t type, uint32_t residx,
 478                            struct ctl_scsiio *ctsio,
 479                            struct scsi_per_res_out *cdb,
 480                            struct scsi_per_res_out_parms* param);
 481 static void ctl_pro_preempt_other(struct ctl_lun *lun,
 482                                   union ctl_ha_msg *msg);
 483 static void ctl_hndl_per_res_out_on_other_sc(union ctl_io *io);
 484 static int ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len);
 485 static int ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len);
 486 static int ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len);
 487 static int ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len);
 488 static int ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len);
 489 static int ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio,
 490                                          int alloc_len);
 491 static int ctl_inquiry_evpd_sfs(struct ctl_scsiio *ctsio, int alloc_len);
 492 static int ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio,
 493                                          int alloc_len);
 494 static int ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len);
 495 static int ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len);
 496 static int ctl_inquiry_evpd(struct ctl_scsiio *ctsio);
 497 static int ctl_inquiry_std(struct ctl_scsiio *ctsio);
 498 static int ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len);
 499 static ctl_action ctl_extent_check(union ctl_io *io1, union ctl_io *io2,
 500     bool seq);
 501 static ctl_action ctl_seq_check(union ctl_io *io1, union ctl_io *io2);
 502 static ctl_action ctl_check_for_blockage(struct ctl_lun *lun,
 503     union ctl_io *pending_io, const uint8_t *serialize_row,
 504     union ctl_io *ooa_io);
 505 static ctl_action ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
 506                                 union ctl_io **starting_io);
 507 static void ctl_try_unblock_io(struct ctl_lun *lun, union ctl_io *io,
 508     bool skip);
 509 static void ctl_try_unblock_others(struct ctl_lun *lun, union ctl_io *io,
 510     bool skip);
 511 static int ctl_scsiio_lun_check(struct ctl_lun *lun,
 512                                 const struct ctl_cmd_entry *entry,
 513                                 struct ctl_scsiio *ctsio);
 514 static void ctl_failover_lun(union ctl_io *io);
 515 static void ctl_scsiio_precheck(struct ctl_scsiio *ctsio);
 516 static int ctl_scsiio(struct ctl_scsiio *ctsio);
 517
 518 static int ctl_target_reset(union ctl_io *io);
 519 static void ctl_do_lun_reset(struct ctl_lun *lun, uint32_t initidx,
 520                          ctl_ua_type ua_type);
 521 static int ctl_lun_reset(union ctl_io *io);
 522 static int ctl_abort_task(union ctl_io *io);
 523 static int ctl_abort_task_set(union ctl_io *io);
 524 static int ctl_query_task(union ctl_io *io, int task_set);
 525 static void ctl_i_t_nexus_loss(struct ctl_softc *softc, uint32_t initidx,
 526                               ctl_ua_type ua_type);
 527 static int ctl_i_t_nexus_reset(union ctl_io *io);
 528 static int ctl_query_async_event(union ctl_io *io);
 529 static void ctl_run_task(union ctl_io *io);
 530 #ifdef CTL_IO_DELAY
 531 static void ctl_datamove_timer_wakeup(void *arg);
 532 static void ctl_done_timer_wakeup(void *arg);
 533 #endif /* CTL_IO_DELAY */
 534
 535 static void ctl_send_datamove_done(union ctl_io *io, int have_lock);
 536 static void ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq);
 537 static int ctl_datamove_remote_dm_write_cb(union ctl_io *io, bool samethr);
 538 static void ctl_datamove_remote_write(union ctl_io *io);
 539 static int ctl_datamove_remote_dm_read_cb(union ctl_io *io, bool samethr);
 540 static void ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq);
 541 static int ctl_datamove_remote_sgl_setup(union ctl_io *io);
 542 static int ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
 543                                     ctl_ha_dt_cb callback);
 544 static void ctl_datamove_remote_read(union ctl_io *io);
 545 static void ctl_datamove_remote(union ctl_io *io);
 546 static void ctl_process_done(union ctl_io *io);
 547 static void ctl_thresh_thread(void *arg);
 548 static void ctl_work_thread(void *arg);
 549 static void ctl_enqueue_incoming(union ctl_io *io);
 550 static void ctl_enqueue_rtr(union ctl_io *io);
 551 static void ctl_enqueue_done(union ctl_io *io);
 552 static void ctl_enqueue_isc(union ctl_io *io);
 553 static const struct ctl_cmd_entry *
 554     ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa);
 555 static const struct ctl_cmd_entry *
 556     ctl_validate_command(struct ctl_scsiio *ctsio);
 557 static int ctl_cmd_applicable(uint8_t lun_type,
 558     const struct ctl_cmd_entry *entry);
 559 static int ctl_ha_init(void);
 560 static int ctl_ha_shutdown(void);
 561
 562 static uint64_t ctl_get_prkey(struct ctl_lun *lun, uint32_t residx);
 563 static void ctl_clr_prkey(struct ctl_lun *lun, uint32_t residx);
 564 static void ctl_alloc_prkey(struct ctl_lun *lun, uint32_t residx);
 565 static void ctl_set_prkey(struct ctl_lun *lun, uint32_t residx, uint64_t key);
 566
 567 /*
 568  * Load the serialization table.  This isn't very pretty, but is probably
 569  * the easiest way to do it.
 570  */
 571 #include "ctl_ser_table.c"
 572
 573 /*
 574  * We only need to define open, close and ioctl routines for this driver.
 575  */
 576 static struct cdevsw ctl_cdevsw = {
 577         .d_version =    D_VERSION,
 578         .d_flags =      0,
 579         .d_open =       ctl_open,
 580         .d_close =      ctl_close,
 581         .d_ioctl =      ctl_ioctl,
 582         .d_name =       "ctl",
 583 };
 584
 585 MALLOC_DEFINE(M_CTL, "ctlmem", "Memory used for CTL");
 586
 587 static int ctl_module_event_handler(module_t, int /*modeventtype_t*/, void *);
 588
 589 static moduledata_t ctl_moduledata = {
 590         "ctl",
 591         ctl_module_event_handler,
 592         NULL
 593 };
 594
 595 DECLARE_MODULE(ctl, ctl_moduledata, SI_SUB_CONFIGURE, SI_ORDER_THIRD);
 596 MODULE_VERSION(ctl, 1);
 597
 598 static struct ctl_frontend ha_frontend =
 599 {
 600         .name = "ha",
 601         .init = ctl_ha_init,
 602         .shutdown = ctl_ha_shutdown,
 603 };
 604
 605 static int
 606 ctl_ha_init(void)
 607 {
 608         struct ctl_softc *softc = control_softc;
 609
 610         if (ctl_pool_create(softc, "othersc", CTL_POOL_ENTRIES_OTHER_SC,
 611                             &softc->othersc_pool) != 0)
 612                 return (ENOMEM);
 613         if (ctl_ha_msg_init(softc) != CTL_HA_STATUS_SUCCESS) {
 614                 ctl_pool_free(softc->othersc_pool);
 615                 return (EIO);
 616         }
 617         if (ctl_ha_msg_register(CTL_HA_CHAN_CTL, ctl_isc_event_handler)
 618             != CTL_HA_STATUS_SUCCESS) {
 619                 ctl_ha_msg_destroy(softc);
 620                 ctl_pool_free(softc->othersc_pool);
 621                 return (EIO);
 622         }
 623         return (0);
 624 };
 625
 626 static int
 627 ctl_ha_shutdown(void)
 628 {
 629         struct ctl_softc *softc = control_softc;
 630         struct ctl_port *port;
 631
 632         ctl_ha_msg_shutdown(softc);
 633         if (ctl_ha_msg_deregister(CTL_HA_CHAN_CTL) != CTL_HA_STATUS_SUCCESS)
 634                 return (EIO);
 635         if (ctl_ha_msg_destroy(softc) != CTL_HA_STATUS_SUCCESS)
 636                 return (EIO);
 637         ctl_pool_free(softc->othersc_pool);
 638         while ((port = STAILQ_FIRST(&ha_frontend.port_list)) != NULL) {
 639                 ctl_port_deregister(port);
 640                 free(port->port_name, M_CTL);
 641                 free(port, M_CTL);
 642         }
 643         return (0);
 644 };
 645
 646 static void
 647 ctl_ha_datamove(union ctl_io *io)
 648 {
 649         struct ctl_lun *lun = CTL_LUN(io);
 650         struct ctl_sg_entry *sgl;
 651         union ctl_ha_msg msg;
 652         uint32_t sg_entries_sent;
 653         int do_sg_copy, i, j;
 654
 655         memset(&msg.dt, 0, sizeof(msg.dt));
 656         msg.hdr.msg_type = CTL_MSG_DATAMOVE;
 657         msg.hdr.original_sc = io->io_hdr.remote_io;
 658         msg.hdr.serializing_sc = io;
 659         msg.hdr.nexus = io->io_hdr.nexus;
 660         msg.hdr.status = io->io_hdr.status;
 661         msg.dt.flags = io->io_hdr.flags;
 662
 663         /*
 664          * We convert everything into a S/G list here.  We can't
 665          * pass by reference, only by value between controllers.
 666          * So we can't pass a pointer to the S/G list, only as many
 667          * S/G entries as we can fit in here.  If it's possible for
 668          * us to get more than CTL_HA_MAX_SG_ENTRIES S/G entries,
 669          * then we need to break this up into multiple transfers.
 670          */
 671         if (io->scsiio.kern_sg_entries == 0) {
 672                 msg.dt.kern_sg_entries = 1;
 673 #if 0
 674                 if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
 675                         msg.dt.sg_list[0].addr = io->scsiio.kern_data_ptr;
 676                 } else {
 677                         /* XXX KDM use busdma here! */
 678                         msg.dt.sg_list[0].addr =
 679                             (void *)vtophys(io->scsiio.kern_data_ptr);
 680                 }
 681 #else
 682                 KASSERT((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0,
 683                     ("HA does not support BUS_ADDR"));
 684                 msg.dt.sg_list[0].addr = io->scsiio.kern_data_ptr;
 685 #endif
 686                 msg.dt.sg_list[0].len = io->scsiio.kern_data_len;
 687                 do_sg_copy = 0;
 688         } else {
 689                 msg.dt.kern_sg_entries = io->scsiio.kern_sg_entries;
 690                 do_sg_copy = 1;
 691         }
 692
 693         msg.dt.kern_data_len = io->scsiio.kern_data_len;
 694         msg.dt.kern_total_len = io->scsiio.kern_total_len;
 695         msg.dt.kern_data_resid = io->scsiio.kern_data_resid;
 696         msg.dt.kern_rel_offset = io->scsiio.kern_rel_offset;
 697         msg.dt.sg_sequence = 0;
 698
 699         /*
 700          * Loop until we've sent all of the S/G entries.  On the
 701          * other end, we'll recompose these S/G entries into one
 702          * contiguous list before processing.
 703          */
 704         for (sg_entries_sent = 0; sg_entries_sent < msg.dt.kern_sg_entries;
 705             msg.dt.sg_sequence++) {
 706                 msg.dt.cur_sg_entries = MIN((sizeof(msg.dt.sg_list) /
 707                     sizeof(msg.dt.sg_list[0])),
 708                     msg.dt.kern_sg_entries - sg_entries_sent);
 709                 if (do_sg_copy != 0) {
 710                         sgl = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr;
 711                         for (i = sg_entries_sent, j = 0;
 712                              i < msg.dt.cur_sg_entries; i++, j++) {
 713 #if 0
 714                                 if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
 715                                         msg.dt.sg_list[j].addr = sgl[i].addr;
 716                                 } else {
 717                                         /* XXX KDM use busdma here! */
 718                                         msg.dt.sg_list[j].addr =
 719                                             (void *)vtophys(sgl[i].addr);
 720                                 }
 721 #else
 722                                 KASSERT((io->io_hdr.flags &
 723                                     CTL_FLAG_BUS_ADDR) == 0,
 724                                     ("HA does not support BUS_ADDR"));
 725                                 msg.dt.sg_list[j].addr = sgl[i].addr;
 726 #endif
 727                                 msg.dt.sg_list[j].len = sgl[i].len;
 728                         }
 729                 }
 730
 731                 sg_entries_sent += msg.dt.cur_sg_entries;
 732                 msg.dt.sg_last = (sg_entries_sent >= msg.dt.kern_sg_entries);
 733                 if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
 734                     sizeof(msg.dt) - sizeof(msg.dt.sg_list) +
 735                     sizeof(struct ctl_sg_entry) * msg.dt.cur_sg_entries,
 736                     M_WAITOK) > CTL_HA_STATUS_SUCCESS) {
 737                         io->io_hdr.port_status = 31341;
 738                         ctl_datamove_done(io, true);
 739                         return;
 740                 }
 741                 msg.dt.sent_sg_entries = sg_entries_sent;
 742         }
 743
 744         /*
 745          * Officially handover the request from us to peer.
 746          * If failover has just happened, then we must return error.
 747          * If failover happen just after, then it is not our problem.
 748          */
 749         if (lun)
 750                 mtx_lock(&lun->lun_lock);
 751         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
 752                 if (lun)
 753                         mtx_unlock(&lun->lun_lock);
 754                 io->io_hdr.port_status = 31342;
 755                 ctl_datamove_done(io, true);
 756                 return;
 757         }
 758         io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
 759         io->io_hdr.flags |= CTL_FLAG_DMA_INPROG;
 760         if (lun)
 761                 mtx_unlock(&lun->lun_lock);
 762 }
 763
 764 static void
 765 ctl_ha_done(union ctl_io *io)
 766 {
 767         union ctl_ha_msg msg;
 768
 769         if (io->io_hdr.io_type == CTL_IO_SCSI) {
 770                 memset(&msg, 0, sizeof(msg));
 771                 msg.hdr.msg_type = CTL_MSG_FINISH_IO;
 772                 msg.hdr.original_sc = io->io_hdr.remote_io;
 773                 msg.hdr.nexus = io->io_hdr.nexus;
 774                 msg.hdr.status = io->io_hdr.status;
 775                 msg.scsi.scsi_status = io->scsiio.scsi_status;
 776                 msg.scsi.tag_num = io->scsiio.tag_num;
 777                 msg.scsi.tag_type = io->scsiio.tag_type;
 778                 msg.scsi.sense_len = io->scsiio.sense_len;
 779                 memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
 780                     io->scsiio.sense_len);
 781                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
 782                     sizeof(msg.scsi) - sizeof(msg.scsi.sense_data) +
 783                     msg.scsi.sense_len, M_WAITOK);
 784         }
 785         ctl_free_io(io);
 786 }
 787
 788 static void
 789 ctl_isc_handler_finish_xfer(struct ctl_softc *ctl_softc,
 790                             union ctl_ha_msg *msg_info)
 791 {
 792         struct ctl_scsiio *ctsio;
 793
 794         if (msg_info->hdr.original_sc == NULL) {
 795                 printf("%s: original_sc == NULL!\n", __func__);
 796                 /* XXX KDM now what? */
 797                 return;
 798         }
 799
 800         ctsio = &msg_info->hdr.original_sc->scsiio;
 801         ctsio->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
 802         ctsio->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
 803         ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
 804         ctsio->io_hdr.status = msg_info->hdr.status;
 805         ctsio->scsi_status = msg_info->scsi.scsi_status;
 806         ctsio->sense_len = msg_info->scsi.sense_len;
 807         memcpy(&ctsio->sense_data, &msg_info->scsi.sense_data,
 808                msg_info->scsi.sense_len);
 809         ctl_enqueue_isc((union ctl_io *)ctsio);
 810 }
 811
 812 static void
 813 ctl_isc_handler_finish_ser_only(struct ctl_softc *ctl_softc,
 814                                 union ctl_ha_msg *msg_info)
 815 {
 816         struct ctl_scsiio *ctsio;
 817
 818         if (msg_info->hdr.serializing_sc == NULL) {
 819                 printf("%s: serializing_sc == NULL!\n", __func__);
 820                 /* XXX KDM now what? */
 821                 return;
 822         }
 823
 824         ctsio = &msg_info->hdr.serializing_sc->scsiio;
 825         ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
 826         ctl_enqueue_isc((union ctl_io *)ctsio);
 827 }
 828
 829 void
 830 ctl_isc_announce_lun(struct ctl_lun *lun)
 831 {
 832         struct ctl_softc *softc = lun->ctl_softc;
 833         union ctl_ha_msg *msg;
 834         struct ctl_ha_msg_lun_pr_key pr_key;
 835         int i, k;
 836
 837         if (softc->ha_link != CTL_HA_LINK_ONLINE)
 838                 return;
 839         mtx_lock(&lun->lun_lock);
 840         i = sizeof(msg->lun);
 841         if (lun->lun_devid)
 842                 i += lun->lun_devid->len;
 843         i += sizeof(pr_key) * lun->pr_key_count;
 844 alloc:
 845         mtx_unlock(&lun->lun_lock);
 846         msg = malloc(i, M_CTL, M_WAITOK);
 847         mtx_lock(&lun->lun_lock);
 848         k = sizeof(msg->lun);
 849         if (lun->lun_devid)
 850                 k += lun->lun_devid->len;
 851         k += sizeof(pr_key) * lun->pr_key_count;
 852         if (i < k) {
 853                 free(msg, M_CTL);
 854                 i = k;
 855                 goto alloc;
 856         }
 857         bzero(&msg->lun, sizeof(msg->lun));
 858         msg->hdr.msg_type = CTL_MSG_LUN_SYNC;
 859         msg->hdr.nexus.targ_lun = lun->lun;
 860         msg->hdr.nexus.targ_mapped_lun = lun->lun;
 861         msg->lun.flags = lun->flags;
 862         msg->lun.pr_generation = lun->pr_generation;
 863         msg->lun.pr_res_idx = lun->pr_res_idx;
 864         msg->lun.pr_res_type = lun->pr_res_type;
 865         msg->lun.pr_key_count = lun->pr_key_count;
 866         i = 0;
 867         if (lun->lun_devid) {
 868                 msg->lun.lun_devid_len = lun->lun_devid->len;
 869                 memcpy(&msg->lun.data[i], lun->lun_devid->data,
 870                     msg->lun.lun_devid_len);
 871                 i += msg->lun.lun_devid_len;
 872         }
 873         for (k = 0; k < CTL_MAX_INITIATORS; k++) {
 874                 if ((pr_key.pr_key = ctl_get_prkey(lun, k)) == 0)
 875                         continue;
 876                 pr_key.pr_iid = k;
 877                 memcpy(&msg->lun.data[i], &pr_key, sizeof(pr_key));
 878                 i += sizeof(pr_key);
 879         }
 880         mtx_unlock(&lun->lun_lock);
 881         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->lun, sizeof(msg->lun) + i,
 882             M_WAITOK);
 883         free(msg, M_CTL);
 884
 885         if (lun->flags & CTL_LUN_PRIMARY_SC) {
 886                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
 887                         ctl_isc_announce_mode(lun, -1,
 888                             lun->mode_pages.index[i].page_code & SMPH_PC_MASK,
 889                             lun->mode_pages.index[i].subpage);
 890                 }
 891         }
 892 }
 893
 894 void
 895 ctl_isc_announce_port(struct ctl_port *port)
 896 {
 897         struct ctl_softc *softc = port->ctl_softc;
 898         union ctl_ha_msg *msg;
 899         int i;
 900
 901         if (port->targ_port < softc->port_min ||
 902             port->targ_port >= softc->port_max ||
 903             softc->ha_link != CTL_HA_LINK_ONLINE)
 904                 return;
 905         i = sizeof(msg->port) + strlen(port->port_name) + 1;
 906         if (port->lun_map)
 907                 i += port->lun_map_size * sizeof(uint32_t);
 908         if (port->port_devid)
 909                 i += port->port_devid->len;
 910         if (port->target_devid)
 911                 i += port->target_devid->len;
 912         if (port->init_devid)
 913                 i += port->init_devid->len;
 914         msg = malloc(i, M_CTL, M_WAITOK);
 915         bzero(&msg->port, sizeof(msg->port));
 916         msg->hdr.msg_type = CTL_MSG_PORT_SYNC;
 917         msg->hdr.nexus.targ_port = port->targ_port;
 918         msg->port.port_type = port->port_type;
 919         msg->port.physical_port = port->physical_port;
 920         msg->port.virtual_port = port->virtual_port;
 921         msg->port.status = port->status;
 922         i = 0;
 923         msg->port.name_len = sprintf(&msg->port.data[i],
 924             "%d:%s", softc->ha_id, port->port_name) + 1;
 925         i += msg->port.name_len;
 926         if (port->lun_map) {
 927                 msg->port.lun_map_len = port->lun_map_size * sizeof(uint32_t);
 928                 memcpy(&msg->port.data[i], port->lun_map,
 929                     msg->port.lun_map_len);
 930                 i += msg->port.lun_map_len;
 931         }
 932         if (port->port_devid) {
 933                 msg->port.port_devid_len = port->port_devid->len;
 934                 memcpy(&msg->port.data[i], port->port_devid->data,
 935                     msg->port.port_devid_len);
 936                 i += msg->port.port_devid_len;
 937         }
 938         if (port->target_devid) {
 939                 msg->port.target_devid_len = port->target_devid->len;
 940                 memcpy(&msg->port.data[i], port->target_devid->data,
 941                     msg->port.target_devid_len);
 942                 i += msg->port.target_devid_len;
 943         }
 944         if (port->init_devid) {
 945                 msg->port.init_devid_len = port->init_devid->len;
 946                 memcpy(&msg->port.data[i], port->init_devid->data,
 947                     msg->port.init_devid_len);
 948                 i += msg->port.init_devid_len;
 949         }
 950         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->port, sizeof(msg->port) + i,
 951             M_WAITOK);
 952         free(msg, M_CTL);
 953 }
 954
 955 void
 956 ctl_isc_announce_iid(struct ctl_port *port, int iid)
 957 {
 958         struct ctl_softc *softc = port->ctl_softc;
 959         union ctl_ha_msg *msg;
 960         int i, l;
 961
 962         if (port->targ_port < softc->port_min ||
 963             port->targ_port >= softc->port_max ||
 964             softc->ha_link != CTL_HA_LINK_ONLINE)
 965                 return;
 966         mtx_lock(&softc->ctl_lock);
 967         i = sizeof(msg->iid);
 968         l = 0;
 969         if (port->wwpn_iid[iid].name)
 970                 l = strlen(port->wwpn_iid[iid].name) + 1;
 971         i += l;
 972         msg = malloc(i, M_CTL, M_NOWAIT);
 973         if (msg == NULL) {
 974                 mtx_unlock(&softc->ctl_lock);
 975                 return;
 976         }
 977         bzero(&msg->iid, sizeof(msg->iid));
 978         msg->hdr.msg_type = CTL_MSG_IID_SYNC;
 979         msg->hdr.nexus.targ_port = port->targ_port;
 980         msg->hdr.nexus.initid = iid;
 981         msg->iid.in_use = port->wwpn_iid[iid].in_use;
 982         msg->iid.name_len = l;
 983         msg->iid.wwpn = port->wwpn_iid[iid].wwpn;
 984         if (port->wwpn_iid[iid].name)
 985                 strlcpy(msg->iid.data, port->wwpn_iid[iid].name, l);
 986         mtx_unlock(&softc->ctl_lock);
 987         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->iid, i, M_NOWAIT);
 988         free(msg, M_CTL);
 989 }
 990
 991 void
 992 ctl_isc_announce_mode(struct ctl_lun *lun, uint32_t initidx,
 993     uint8_t page, uint8_t subpage)
 994 {
 995         struct ctl_softc *softc = lun->ctl_softc;
 996         union ctl_ha_msg *msg;
 997         u_int i, l;
 998
 999         if (softc->ha_link != CTL_HA_LINK_ONLINE)
1000                 return;
1001         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
1002                 if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) ==
1003                     page && lun->mode_pages.index[i].subpage == subpage)
1004                         break;
1005         }
1006         if (i == CTL_NUM_MODE_PAGES)
1007                 return;
1008
1009         /* Don't try to replicate pages not present on this device. */
1010         if (lun->mode_pages.index[i].page_data == NULL)
1011                 return;
1012
1013         l = sizeof(msg->mode) + lun->mode_pages.index[i].page_len;
1014         msg = malloc(l, M_CTL, M_WAITOK | M_ZERO);
1015         msg->hdr.msg_type = CTL_MSG_MODE_SYNC;
1016         msg->hdr.nexus.targ_port = initidx / CTL_MAX_INIT_PER_PORT;
1017         msg->hdr.nexus.initid = initidx % CTL_MAX_INIT_PER_PORT;
1018         msg->hdr.nexus.targ_lun = lun->lun;
1019         msg->hdr.nexus.targ_mapped_lun = lun->lun;
1020         msg->mode.page_code = page;
1021         msg->mode.subpage = subpage;
1022         msg->mode.page_len = lun->mode_pages.index[i].page_len;
1023         memcpy(msg->mode.data, lun->mode_pages.index[i].page_data,
1024             msg->mode.page_len);
1025         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->mode, l, M_WAITOK);
1026         free(msg, M_CTL);
1027 }
1028
1029 static void
1030 ctl_isc_ha_link_up(struct ctl_softc *softc)
1031 {
1032         struct ctl_port *port;
1033         struct ctl_lun *lun;
1034         union ctl_ha_msg msg;
1035         int i;
1036
1037         /* Announce this node parameters to peer for validation. */
1038         msg.login.msg_type = CTL_MSG_LOGIN;
1039         msg.login.version = CTL_HA_VERSION;
1040         msg.login.ha_mode = softc->ha_mode;
1041         msg.login.ha_id = softc->ha_id;
1042         msg.login.max_luns = ctl_max_luns;
1043         msg.login.max_ports = ctl_max_ports;
1044         msg.login.max_init_per_port = CTL_MAX_INIT_PER_PORT;
1045         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg.login, sizeof(msg.login),
1046             M_WAITOK);
1047
1048         STAILQ_FOREACH(port, &softc->port_list, links) {
1049                 ctl_isc_announce_port(port);
1050                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1051                         if (port->wwpn_iid[i].in_use)
1052                                 ctl_isc_announce_iid(port, i);
1053                 }
1054         }
1055         STAILQ_FOREACH(lun, &softc->lun_list, links)
1056                 ctl_isc_announce_lun(lun);
1057 }
1058
1059 static void
1060 ctl_isc_ha_link_down(struct ctl_softc *softc)
1061 {
1062         struct ctl_port *port;
1063         struct ctl_lun *lun;
1064         union ctl_io *io;
1065         int i;
1066
1067         mtx_lock(&softc->ctl_lock);
1068         STAILQ_FOREACH(lun, &softc->lun_list, links) {
1069                 mtx_lock(&lun->lun_lock);
1070                 if (lun->flags & CTL_LUN_PEER_SC_PRIMARY) {
1071                         lun->flags &= ~CTL_LUN_PEER_SC_PRIMARY;
1072                         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
1073                 }
1074                 mtx_unlock(&lun->lun_lock);
1075
1076                 mtx_unlock(&softc->ctl_lock);
1077                 io = ctl_alloc_io(softc->othersc_pool);
1078                 mtx_lock(&softc->ctl_lock);
1079                 ctl_zero_io(io);
1080                 io->io_hdr.msg_type = CTL_MSG_FAILOVER;
1081                 io->io_hdr.nexus.targ_mapped_lun = lun->lun;
1082                 ctl_enqueue_isc(io);
1083         }
1084
1085         STAILQ_FOREACH(port, &softc->port_list, links) {
1086                 if (port->targ_port >= softc->port_min &&
1087                     port->targ_port < softc->port_max)
1088                         continue;
1089                 port->status &= ~CTL_PORT_STATUS_ONLINE;
1090                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1091                         port->wwpn_iid[i].in_use = 0;
1092                         free(port->wwpn_iid[i].name, M_CTL);
1093                         port->wwpn_iid[i].name = NULL;
1094                 }
1095         }
1096         mtx_unlock(&softc->ctl_lock);
1097 }
1098
1099 static void
1100 ctl_isc_ua(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1101 {
1102         struct ctl_lun *lun;
1103         uint32_t iid;
1104
1105         if (len < sizeof(msg->ua)) {
1106                 printf("%s: Received truncated message %d < %zu\n",
1107                     __func__, len, sizeof(msg->ua));
1108                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1109                 return;
1110         }
1111
1112         mtx_lock(&softc->ctl_lock);
1113         if (msg->hdr.nexus.targ_mapped_lun >= ctl_max_luns ||
1114             (lun = softc->ctl_luns[msg->hdr.nexus.targ_mapped_lun]) == NULL) {
1115                 mtx_unlock(&softc->ctl_lock);
1116                 return;
1117         }
1118         mtx_lock(&lun->lun_lock);
1119         mtx_unlock(&softc->ctl_lock);
1120         if (msg->ua.ua_type == CTL_UA_THIN_PROV_THRES && msg->ua.ua_set)
1121                 memcpy(lun->ua_tpt_info, msg->ua.ua_info, 8);
1122         iid = ctl_get_initindex(&msg->hdr.nexus);
1123         if (msg->ua.ua_all) {
1124                 if (msg->ua.ua_set)
1125                         ctl_est_ua_all(lun, iid, msg->ua.ua_type);
1126                 else
1127                         ctl_clr_ua_all(lun, iid, msg->ua.ua_type);
1128         } else {
1129                 if (msg->ua.ua_set)
1130                         ctl_est_ua(lun, iid, msg->ua.ua_type);
1131                 else
1132                         ctl_clr_ua(lun, iid, msg->ua.ua_type);
1133         }
1134         mtx_unlock(&lun->lun_lock);
1135 }
1136
1137 static void
1138 ctl_isc_lun_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1139 {
1140         struct ctl_lun *lun;
1141         struct ctl_ha_msg_lun_pr_key pr_key;
1142         int i, k;
1143         ctl_lun_flags oflags;
1144         uint32_t targ_lun;
1145
1146         if (len < offsetof(struct ctl_ha_msg_lun, data[0])) {
1147                 printf("%s: Received truncated message %d < %zu\n",
1148                     __func__, len, offsetof(struct ctl_ha_msg_lun, data[0]));
1149                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1150                 return;
1151         }
1152         i = msg->lun.lun_devid_len + msg->lun.pr_key_count * sizeof(pr_key);
1153         if (len < offsetof(struct ctl_ha_msg_lun, data[i])) {
1154                 printf("%s: Received truncated message data %d < %zu\n",
1155                     __func__, len, offsetof(struct ctl_ha_msg_lun, data[i]));
1156                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1157                 return;
1158         }
1159
1160         targ_lun = msg->hdr.nexus.targ_mapped_lun;
1161         mtx_lock(&softc->ctl_lock);
1162         if (targ_lun >= ctl_max_luns ||
1163             (lun = softc->ctl_luns[targ_lun]) == NULL) {
1164                 mtx_unlock(&softc->ctl_lock);
1165                 return;
1166         }
1167         mtx_lock(&lun->lun_lock);
1168         mtx_unlock(&softc->ctl_lock);
1169         if (lun->flags & CTL_LUN_DISABLED) {
1170                 mtx_unlock(&lun->lun_lock);
1171                 return;
1172         }
1173         i = (lun->lun_devid != NULL) ? lun->lun_devid->len : 0;
1174         if (msg->lun.lun_devid_len != i || (i > 0 &&
1175             memcmp(&msg->lun.data[0], lun->lun_devid->data, i) != 0)) {
1176                 mtx_unlock(&lun->lun_lock);
1177                 printf("%s: Received conflicting HA LUN %d\n",
1178                     __func__, targ_lun);
1179                 return;
1180         } else {
1181                 /* Record whether peer is primary. */
1182                 oflags = lun->flags;
1183                 if ((msg->lun.flags & CTL_LUN_PRIMARY_SC) &&
1184                     (msg->lun.flags & CTL_LUN_DISABLED) == 0)
1185                         lun->flags |= CTL_LUN_PEER_SC_PRIMARY;
1186                 else
1187                         lun->flags &= ~CTL_LUN_PEER_SC_PRIMARY;
1188                 if (oflags != lun->flags)
1189                         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
1190
1191                 /* If peer is primary and we are not -- use data */
1192                 if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
1193                     (lun->flags & CTL_LUN_PEER_SC_PRIMARY)) {
1194                         lun->pr_generation = msg->lun.pr_generation;
1195                         lun->pr_res_idx = msg->lun.pr_res_idx;
1196                         lun->pr_res_type = msg->lun.pr_res_type;
1197                         lun->pr_key_count = msg->lun.pr_key_count;
1198                         for (k = 0; k < CTL_MAX_INITIATORS; k++)
1199                                 ctl_clr_prkey(lun, k);
1200                         for (k = 0; k < msg->lun.pr_key_count; k++) {
1201                                 memcpy(&pr_key, &msg->lun.data[i],
1202                                     sizeof(pr_key));
1203                                 ctl_alloc_prkey(lun, pr_key.pr_iid);
1204                                 ctl_set_prkey(lun, pr_key.pr_iid,
1205                                     pr_key.pr_key);
1206                                 i += sizeof(pr_key);
1207                         }
1208                 }
1209
1210                 mtx_unlock(&lun->lun_lock);
1211                 CTL_DEBUG_PRINT(("%s: Known LUN %d, peer is %s\n",
1212                     __func__, targ_lun,
1213                     (msg->lun.flags & CTL_LUN_PRIMARY_SC) ?
1214                     "primary" : "secondary"));
1215
1216                 /* If we are primary but peer doesn't know -- notify */
1217                 if ((lun->flags & CTL_LUN_PRIMARY_SC) &&
1218                     (msg->lun.flags & CTL_LUN_PEER_SC_PRIMARY) == 0)
1219                         ctl_isc_announce_lun(lun);
1220         }
1221 }
1222
1223 static void
1224 ctl_isc_port_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1225 {
1226         struct ctl_port *port;
1227         struct ctl_lun *lun;
1228         int i, new;
1229
1230         if (len < offsetof(struct ctl_ha_msg_port, data[0])) {
1231                 printf("%s: Received truncated message %d < %zu\n",
1232                     __func__, len, offsetof(struct ctl_ha_msg_port, data[0]));
1233                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1234                 return;
1235         }
1236         i = msg->port.name_len + msg->port.lun_map_len +
1237             msg->port.port_devid_len + msg->port.target_devid_len +
1238             msg->port.init_devid_len;
1239         if (len < offsetof(struct ctl_ha_msg_port, data[i])) {
1240                 printf("%s: Received truncated message data %d < %zu\n",
1241                     __func__, len, offsetof(struct ctl_ha_msg_port, data[i]));
1242                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1243                 return;
1244         }
1245
1246         port = softc->ctl_ports[msg->hdr.nexus.targ_port];
1247         if (port == NULL) {
1248                 CTL_DEBUG_PRINT(("%s: New port %d\n", __func__,
1249                     msg->hdr.nexus.targ_port));
1250                 new = 1;
1251                 port = malloc(sizeof(*port), M_CTL, M_WAITOK | M_ZERO);
1252                 port->frontend = &ha_frontend;
1253                 port->targ_port = msg->hdr.nexus.targ_port;
1254                 port->fe_datamove = ctl_ha_datamove;
1255                 port->fe_done = ctl_ha_done;
1256         } else if (port->frontend == &ha_frontend) {
1257                 CTL_DEBUG_PRINT(("%s: Updated port %d\n", __func__,
1258                     msg->hdr.nexus.targ_port));
1259                 new = 0;
1260         } else {
1261                 printf("%s: Received conflicting HA port %d\n",
1262                     __func__, msg->hdr.nexus.targ_port);
1263                 return;
1264         }
1265         port->port_type = msg->port.port_type;
1266         port->physical_port = msg->port.physical_port;
1267         port->virtual_port = msg->port.virtual_port;
1268         port->status = msg->port.status;
1269         i = 0;
1270         free(port->port_name, M_CTL);
1271         port->port_name = strndup(&msg->port.data[i], msg->port.name_len,
1272             M_CTL);
1273         i += msg->port.name_len;
1274         if (msg->port.lun_map_len != 0) {
1275                 if (port->lun_map == NULL ||
1276                     port->lun_map_size * sizeof(uint32_t) <
1277                     msg->port.lun_map_len) {
1278                         port->lun_map_size = 0;
1279                         free(port->lun_map, M_CTL);
1280                         port->lun_map = malloc(msg->port.lun_map_len,
1281                             M_CTL, M_WAITOK);
1282                 }
1283                 memcpy(port->lun_map, &msg->port.data[i], msg->port.lun_map_len);
1284                 port->lun_map_size = msg->port.lun_map_len / sizeof(uint32_t);
1285                 i += msg->port.lun_map_len;
1286         } else {
1287                 port->lun_map_size = 0;
1288                 free(port->lun_map, M_CTL);
1289                 port->lun_map = NULL;
1290         }
1291         if (msg->port.port_devid_len != 0) {
1292                 if (port->port_devid == NULL ||
1293                     port->port_devid->len < msg->port.port_devid_len) {
1294                         free(port->port_devid, M_CTL);
1295                         port->port_devid = malloc(sizeof(struct ctl_devid) +
1296                             msg->port.port_devid_len, M_CTL, M_WAITOK);
1297                 }
1298                 memcpy(port->port_devid->data, &msg->port.data[i],
1299                     msg->port.port_devid_len);
1300                 port->port_devid->len = msg->port.port_devid_len;
1301                 i += msg->port.port_devid_len;
1302         } else {
1303                 free(port->port_devid, M_CTL);
1304                 port->port_devid = NULL;
1305         }
1306         if (msg->port.target_devid_len != 0) {
1307                 if (port->target_devid == NULL ||
1308                     port->target_devid->len < msg->port.target_devid_len) {
1309                         free(port->target_devid, M_CTL);
1310                         port->target_devid = malloc(sizeof(struct ctl_devid) +
1311                             msg->port.target_devid_len, M_CTL, M_WAITOK);
1312                 }
1313                 memcpy(port->target_devid->data, &msg->port.data[i],
1314                     msg->port.target_devid_len);
1315                 port->target_devid->len = msg->port.target_devid_len;
1316                 i += msg->port.target_devid_len;
1317         } else {
1318                 free(port->target_devid, M_CTL);
1319                 port->target_devid = NULL;
1320         }
1321         if (msg->port.init_devid_len != 0) {
1322                 if (port->init_devid == NULL ||
1323                     port->init_devid->len < msg->port.init_devid_len) {
1324                         free(port->init_devid, M_CTL);
1325                         port->init_devid = malloc(sizeof(struct ctl_devid) +
1326                             msg->port.init_devid_len, M_CTL, M_WAITOK);
1327                 }
1328                 memcpy(port->init_devid->data, &msg->port.data[i],
1329                     msg->port.init_devid_len);
1330                 port->init_devid->len = msg->port.init_devid_len;
1331                 i += msg->port.init_devid_len;
1332         } else {
1333                 free(port->init_devid, M_CTL);
1334                 port->init_devid = NULL;
1335         }
1336         if (new) {
1337                 if (ctl_port_register(port) != 0) {
1338                         printf("%s: ctl_port_register() failed with error\n",
1339                             __func__);
1340                 }
1341         }
1342         mtx_lock(&softc->ctl_lock);
1343         STAILQ_FOREACH(lun, &softc->lun_list, links) {
1344                 if (ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
1345                         continue;
1346                 mtx_lock(&lun->lun_lock);
1347                 ctl_est_ua_all(lun, -1, CTL_UA_INQ_CHANGE);
1348                 mtx_unlock(&lun->lun_lock);
1349         }
1350         mtx_unlock(&softc->ctl_lock);
1351 }
1352
1353 static void
1354 ctl_isc_iid_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1355 {
1356         struct ctl_port *port;
1357         int i, iid;
1358
1359         if (len < offsetof(struct ctl_ha_msg_iid, data[0])) {
1360                 printf("%s: Received truncated message %d < %zu\n",
1361                     __func__, len, offsetof(struct ctl_ha_msg_iid, data[0]));
1362                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1363                 return;
1364         }
1365         i = msg->iid.name_len;
1366         if (len < offsetof(struct ctl_ha_msg_iid, data[i])) {
1367                 printf("%s: Received truncated message data %d < %zu\n",
1368                     __func__, len, offsetof(struct ctl_ha_msg_iid, data[i]));
1369                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1370                 return;
1371         }
1372
1373         port = softc->ctl_ports[msg->hdr.nexus.targ_port];
1374         if (port == NULL) {
1375                 printf("%s: Received IID for unknown port %d\n",
1376                     __func__, msg->hdr.nexus.targ_port);
1377                 return;
1378         }
1379         iid = msg->hdr.nexus.initid;
1380         if (port->wwpn_iid[iid].in_use != 0 &&
1381             msg->iid.in_use == 0)
1382                 ctl_i_t_nexus_loss(softc, iid, CTL_UA_POWERON);
1383         port->wwpn_iid[iid].in_use = msg->iid.in_use;
1384         port->wwpn_iid[iid].wwpn = msg->iid.wwpn;
1385         free(port->wwpn_iid[iid].name, M_CTL);
1386         if (msg->iid.name_len) {
1387                 port->wwpn_iid[iid].name = strndup(&msg->iid.data[0],
1388                     msg->iid.name_len, M_CTL);
1389         } else
1390                 port->wwpn_iid[iid].name = NULL;
1391 }
1392
1393 static void
1394 ctl_isc_login(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1395 {
1396
1397         if (len < sizeof(msg->login)) {
1398                 printf("%s: Received truncated message %d < %zu\n",
1399                     __func__, len, sizeof(msg->login));
1400                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1401                 return;
1402         }
1403
1404         if (msg->login.version != CTL_HA_VERSION) {
1405                 printf("CTL HA peers have different versions %d != %d\n",
1406                     msg->login.version, CTL_HA_VERSION);
1407                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1408                 return;
1409         }
1410         if (msg->login.ha_mode != softc->ha_mode) {
1411                 printf("CTL HA peers have different ha_mode %d != %d\n",
1412                     msg->login.ha_mode, softc->ha_mode);
1413                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1414                 return;
1415         }
1416         if (msg->login.ha_id == softc->ha_id) {
1417                 printf("CTL HA peers have same ha_id %d\n", msg->login.ha_id);
1418                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1419                 return;
1420         }
1421         if (msg->login.max_luns != ctl_max_luns ||
1422             msg->login.max_ports != ctl_max_ports ||
1423             msg->login.max_init_per_port != CTL_MAX_INIT_PER_PORT) {
1424                 printf("CTL HA peers have different limits\n");
1425                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1426                 return;
1427         }
1428 }
1429
1430 static void
1431 ctl_isc_mode_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1432 {
1433         struct ctl_lun *lun;
1434         u_int i;
1435         uint32_t initidx, targ_lun;
1436
1437         if (len < offsetof(struct ctl_ha_msg_mode, data[0])) {
1438                 printf("%s: Received truncated message %d < %zu\n",
1439                     __func__, len, offsetof(struct ctl_ha_msg_mode, data[0]));
1440                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1441                 return;
1442         }
1443         i = msg->mode.page_len;
1444         if (len < offsetof(struct ctl_ha_msg_mode, data[i])) {
1445                 printf("%s: Received truncated message data %d < %zu\n",
1446                     __func__, len, offsetof(struct ctl_ha_msg_mode, data[i]));
1447                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1448                 return;
1449         }
1450
1451         targ_lun = msg->hdr.nexus.targ_mapped_lun;
1452         mtx_lock(&softc->ctl_lock);
1453         if (targ_lun >= ctl_max_luns ||
1454             (lun = softc->ctl_luns[targ_lun]) == NULL) {
1455                 mtx_unlock(&softc->ctl_lock);
1456                 return;
1457         }
1458         mtx_lock(&lun->lun_lock);
1459         mtx_unlock(&softc->ctl_lock);
1460         if (lun->flags & CTL_LUN_DISABLED) {
1461                 mtx_unlock(&lun->lun_lock);
1462                 return;
1463         }
1464         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
1465                 if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) ==
1466                     msg->mode.page_code &&
1467                     lun->mode_pages.index[i].subpage == msg->mode.subpage)
1468                         break;
1469         }
1470         if (i == CTL_NUM_MODE_PAGES) {
1471                 mtx_unlock(&lun->lun_lock);
1472                 return;
1473         }
1474         memcpy(lun->mode_pages.index[i].page_data, msg->mode.data,
1475             min(lun->mode_pages.index[i].page_len, msg->mode.page_len));
1476         initidx = ctl_get_initindex(&msg->hdr.nexus);
1477         if (initidx != -1)
1478                 ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
1479         mtx_unlock(&lun->lun_lock);
1480 }
1481
1482 /*
1483  * ISC (Inter Shelf Communication) event handler.  Events from the HA
1484  * subsystem come in here.
1485  */
1486 static void
1487 ctl_isc_event_handler(ctl_ha_channel channel, ctl_ha_event event, int param)
1488 {
1489         struct ctl_softc *softc = control_softc;
1490         union ctl_io *io;
1491         struct ctl_prio *presio;
1492         ctl_ha_status isc_status;
1493
1494         CTL_DEBUG_PRINT(("CTL: Isc Msg event %d\n", event));
1495         if (event == CTL_HA_EVT_MSG_RECV) {
1496                 union ctl_ha_msg *msg, msgbuf;
1497
1498                 if (param > sizeof(msgbuf))
1499                         msg = malloc(param, M_CTL, M_WAITOK);
1500                 else
1501                         msg = &msgbuf;
1502                 isc_status = ctl_ha_msg_recv(CTL_HA_CHAN_CTL, msg, param,
1503                     M_WAITOK);
1504                 if (isc_status != CTL_HA_STATUS_SUCCESS) {
1505                         printf("%s: Error receiving message: %d\n",
1506                             __func__, isc_status);
1507                         if (msg != &msgbuf)
1508                                 free(msg, M_CTL);
1509                         return;
1510                 }
1511
1512                 CTL_DEBUG_PRINT(("CTL: msg_type %d len %d\n",
1513                     msg->hdr.msg_type, param));
1514                 switch (msg->hdr.msg_type) {
1515                 case CTL_MSG_SERIALIZE:
1516                         io = ctl_alloc_io(softc->othersc_pool);
1517                         ctl_zero_io(io);
1518                         // populate ctsio from msg
1519                         io->io_hdr.io_type = CTL_IO_SCSI;
1520                         io->io_hdr.msg_type = CTL_MSG_SERIALIZE;
1521                         io->io_hdr.remote_io = msg->hdr.original_sc;
1522                         io->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC |
1523                                             CTL_FLAG_IO_ACTIVE;
1524                         /*
1525                          * If we're in serialization-only mode, we don't
1526                          * want to go through full done processing.  Thus
1527                          * the COPY flag.
1528                          *
1529                          * XXX KDM add another flag that is more specific.
1530                          */
1531                         if (softc->ha_mode != CTL_HA_MODE_XFER)
1532                                 io->io_hdr.flags |= CTL_FLAG_INT_COPY;
1533                         io->io_hdr.nexus = msg->hdr.nexus;
1534                         io->scsiio.priority = msg->scsi.priority;
1535                         io->scsiio.tag_num = msg->scsi.tag_num;
1536                         io->scsiio.tag_type = msg->scsi.tag_type;
1537 #ifdef CTL_TIME_IO
1538                         io->io_hdr.start_time = time_uptime;
1539                         getbinuptime(&io->io_hdr.start_bt);
1540 #endif /* CTL_TIME_IO */
1541                         io->scsiio.cdb_len = msg->scsi.cdb_len;
1542                         memcpy(io->scsiio.cdb, msg->scsi.cdb,
1543                                CTL_MAX_CDBLEN);
1544                         if (softc->ha_mode == CTL_HA_MODE_XFER) {
1545                                 const struct ctl_cmd_entry *entry;
1546
1547                                 entry = ctl_get_cmd_entry(&io->scsiio, NULL);
1548                                 io->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
1549                                 io->io_hdr.flags |=
1550                                         entry->flags & CTL_FLAG_DATA_MASK;
1551                         }
1552                         ctl_enqueue_isc(io);
1553                         break;
1554
1555                 /* Performed on the Originating SC, XFER mode only */
1556                 case CTL_MSG_DATAMOVE: {
1557                         struct ctl_sg_entry *sgl;
1558                         int i, j;
1559
1560                         io = msg->hdr.original_sc;
1561                         if (io == NULL) {
1562                                 printf("%s: original_sc == NULL!\n", __func__);
1563                                 /* XXX KDM do something here */
1564                                 break;
1565                         }
1566                         io->io_hdr.msg_type = CTL_MSG_DATAMOVE;
1567                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1568                         /*
1569                          * Keep track of this, we need to send it back over
1570                          * when the datamove is complete.
1571                          */
1572                         io->io_hdr.remote_io = msg->hdr.serializing_sc;
1573                         if (msg->hdr.status == CTL_SUCCESS)
1574                                 io->io_hdr.status = msg->hdr.status;
1575
1576                         if (msg->dt.sg_sequence == 0) {
1577 #ifdef CTL_TIME_IO
1578                                 getbinuptime(&io->io_hdr.dma_start_bt);
1579 #endif
1580                                 i = msg->dt.kern_sg_entries +
1581                                     msg->dt.kern_data_len /
1582                                     CTL_HA_DATAMOVE_SEGMENT + 1;
1583                                 sgl = malloc(sizeof(*sgl) * i, M_CTL,
1584                                     M_WAITOK | M_ZERO);
1585                                 CTL_RSGL(io) = sgl;
1586                                 CTL_LSGL(io) = &sgl[msg->dt.kern_sg_entries];
1587
1588                                 io->scsiio.kern_data_ptr = (uint8_t *)sgl;
1589
1590                                 io->scsiio.kern_sg_entries =
1591                                         msg->dt.kern_sg_entries;
1592                                 io->scsiio.rem_sg_entries =
1593                                         msg->dt.kern_sg_entries;
1594                                 io->scsiio.kern_data_len =
1595                                         msg->dt.kern_data_len;
1596                                 io->scsiio.kern_total_len =
1597                                         msg->dt.kern_total_len;
1598                                 io->scsiio.kern_data_resid =
1599                                         msg->dt.kern_data_resid;
1600                                 io->scsiio.kern_rel_offset =
1601                                         msg->dt.kern_rel_offset;
1602                                 io->io_hdr.flags &= ~CTL_FLAG_BUS_ADDR;
1603                                 io->io_hdr.flags |= msg->dt.flags &
1604                                     CTL_FLAG_BUS_ADDR;
1605                         } else
1606                                 sgl = (struct ctl_sg_entry *)
1607                                         io->scsiio.kern_data_ptr;
1608
1609                         for (i = msg->dt.sent_sg_entries, j = 0;
1610                              i < (msg->dt.sent_sg_entries +
1611                              msg->dt.cur_sg_entries); i++, j++) {
1612                                 sgl[i].addr = msg->dt.sg_list[j].addr;
1613                                 sgl[i].len = msg->dt.sg_list[j].len;
1614                         }
1615
1616                         /*
1617                          * If this is the last piece of the I/O, we've got
1618                          * the full S/G list.  Queue processing in the thread.
1619                          * Otherwise wait for the next piece.
1620                          */
1621                         if (msg->dt.sg_last != 0)
1622                                 ctl_enqueue_isc(io);
1623                         break;
1624                 }
1625                 /* Performed on the Serializing (primary) SC, XFER mode only */
1626                 case CTL_MSG_DATAMOVE_DONE: {
1627                         if (msg->hdr.serializing_sc == NULL) {
1628                                 printf("%s: serializing_sc == NULL!\n",
1629                                        __func__);
1630                                 /* XXX KDM now what? */
1631                                 break;
1632                         }
1633                         /*
1634                          * We grab the sense information here in case
1635                          * there was a failure, so we can return status
1636                          * back to the initiator.
1637                          */
1638                         io = msg->hdr.serializing_sc;
1639                         io->io_hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
1640                         io->io_hdr.flags &= ~CTL_FLAG_DMA_INPROG;
1641                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1642                         io->io_hdr.port_status = msg->scsi.port_status;
1643                         io->scsiio.kern_data_resid = msg->scsi.kern_data_resid;
1644                         if (msg->hdr.status != CTL_STATUS_NONE) {
1645                                 io->io_hdr.status = msg->hdr.status;
1646                                 io->scsiio.scsi_status = msg->scsi.scsi_status;
1647                                 io->scsiio.sense_len = msg->scsi.sense_len;
1648                                 memcpy(&io->scsiio.sense_data,
1649                                     &msg->scsi.sense_data,
1650                                     msg->scsi.sense_len);
1651                                 if (msg->hdr.status == CTL_SUCCESS)
1652                                         io->io_hdr.flags |= CTL_FLAG_STATUS_SENT;
1653                         }
1654                         ctl_enqueue_isc(io);
1655                         break;
1656                 }
1657
1658                 /* Preformed on Originating SC, SER_ONLY mode */
1659                 case CTL_MSG_R2R:
1660                         io = msg->hdr.original_sc;
1661                         if (io == NULL) {
1662                                 printf("%s: original_sc == NULL!\n",
1663                                     __func__);
1664                                 break;
1665                         }
1666                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1667                         io->io_hdr.msg_type = CTL_MSG_R2R;
1668                         io->io_hdr.remote_io = msg->hdr.serializing_sc;
1669                         ctl_enqueue_isc(io);
1670                         break;
1671
1672                 /*
1673                  * Performed on Serializing(i.e. primary SC) SC in SER_ONLY
1674                  * mode.
1675                  * Performed on the Originating (i.e. secondary) SC in XFER
1676                  * mode
1677                  */
1678                 case CTL_MSG_FINISH_IO:
1679                         if (softc->ha_mode == CTL_HA_MODE_XFER)
1680                                 ctl_isc_handler_finish_xfer(softc, msg);
1681                         else
1682                                 ctl_isc_handler_finish_ser_only(softc, msg);
1683                         break;
1684
1685                 /* Preformed on Originating SC */
1686                 case CTL_MSG_BAD_JUJU:
1687                         io = msg->hdr.original_sc;
1688                         if (io == NULL) {
1689                                 printf("%s: Bad JUJU!, original_sc is NULL!\n",
1690                                        __func__);
1691                                 break;
1692                         }
1693                         ctl_copy_sense_data(msg, io);
1694                         /*
1695                          * IO should have already been cleaned up on other
1696                          * SC so clear this flag so we won't send a message
1697                          * back to finish the IO there.
1698                          */
1699                         io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
1700                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1701
1702                         /* io = msg->hdr.serializing_sc; */
1703                         io->io_hdr.msg_type = CTL_MSG_BAD_JUJU;
1704                         ctl_enqueue_isc(io);
1705                         break;
1706
1707                 /* Handle resets sent from the other side */
1708                 case CTL_MSG_MANAGE_TASKS: {
1709                         struct ctl_taskio *taskio;
1710                         taskio = (struct ctl_taskio *)ctl_alloc_io(
1711                             softc->othersc_pool);
1712                         ctl_zero_io((union ctl_io *)taskio);
1713                         taskio->io_hdr.io_type = CTL_IO_TASK;
1714                         taskio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
1715                         taskio->io_hdr.nexus = msg->hdr.nexus;
1716                         taskio->task_action = msg->task.task_action;
1717                         taskio->tag_num = msg->task.tag_num;
1718                         taskio->tag_type = msg->task.tag_type;
1719 #ifdef CTL_TIME_IO
1720                         taskio->io_hdr.start_time = time_uptime;
1721                         getbinuptime(&taskio->io_hdr.start_bt);
1722 #endif /* CTL_TIME_IO */
1723                         ctl_run_task((union ctl_io *)taskio);
1724                         break;
1725                 }
1726                 /* Persistent Reserve action which needs attention */
1727                 case CTL_MSG_PERS_ACTION:
1728                         presio = (struct ctl_prio *)ctl_alloc_io(
1729                             softc->othersc_pool);
1730                         ctl_zero_io((union ctl_io *)presio);
1731                         presio->io_hdr.msg_type = CTL_MSG_PERS_ACTION;
1732                         presio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
1733                         presio->io_hdr.nexus = msg->hdr.nexus;
1734                         presio->pr_msg = msg->pr;
1735                         ctl_enqueue_isc((union ctl_io *)presio);
1736                         break;
1737                 case CTL_MSG_UA:
1738                         ctl_isc_ua(softc, msg, param);
1739                         break;
1740                 case CTL_MSG_PORT_SYNC:
1741                         ctl_isc_port_sync(softc, msg, param);
1742                         break;
1743                 case CTL_MSG_LUN_SYNC:
1744                         ctl_isc_lun_sync(softc, msg, param);
1745                         break;
1746                 case CTL_MSG_IID_SYNC:
1747                         ctl_isc_iid_sync(softc, msg, param);
1748                         break;
1749                 case CTL_MSG_LOGIN:
1750                         ctl_isc_login(softc, msg, param);
1751                         break;
1752                 case CTL_MSG_MODE_SYNC:
1753                         ctl_isc_mode_sync(softc, msg, param);
1754                         break;
1755                 default:
1756                         printf("Received HA message of unknown type %d\n",
1757                             msg->hdr.msg_type);
1758                         ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1759                         break;
1760                 }
1761                 if (msg != &msgbuf)
1762                         free(msg, M_CTL);
1763         } else if (event == CTL_HA_EVT_LINK_CHANGE) {
1764                 printf("CTL: HA link status changed from %d to %d\n",
1765                     softc->ha_link, param);
1766                 if (param == softc->ha_link)
1767                         return;
1768                 if (softc->ha_link == CTL_HA_LINK_ONLINE) {
1769                         softc->ha_link = param;
1770                         ctl_isc_ha_link_down(softc);
1771                 } else {
1772                         softc->ha_link = param;
1773                         if (softc->ha_link == CTL_HA_LINK_ONLINE)
1774                                 ctl_isc_ha_link_up(softc);
1775                 }
1776                 return;
1777         } else {
1778                 printf("ctl_isc_event_handler: Unknown event %d\n", event);
1779                 return;
1780         }
1781 }
1782
1783 static void
1784 ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest)
1785 {
1786
1787         memcpy(&dest->scsiio.sense_data, &src->scsi.sense_data,
1788             src->scsi.sense_len);
1789         dest->scsiio.scsi_status = src->scsi.scsi_status;
1790         dest->scsiio.sense_len = src->scsi.sense_len;
1791         dest->io_hdr.status = src->hdr.status;
1792 }
1793
1794 static void
1795 ctl_copy_sense_data_back(union ctl_io *src, union ctl_ha_msg *dest)
1796 {
1797
1798         memcpy(&dest->scsi.sense_data, &src->scsiio.sense_data,
1799             src->scsiio.sense_len);
1800         dest->scsi.scsi_status = src->scsiio.scsi_status;
1801         dest->scsi.sense_len = src->scsiio.sense_len;
1802         dest->hdr.status = src->io_hdr.status;
1803 }
1804
1805 void
1806 ctl_est_ua(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua)
1807 {
1808         struct ctl_softc *softc = lun->ctl_softc;
1809         ctl_ua_type *pu;
1810
1811         if (initidx < softc->init_min || initidx >= softc->init_max)
1812                 return;
1813         mtx_assert(&lun->lun_lock, MA_OWNED);
1814         pu = lun->pending_ua[initidx / CTL_MAX_INIT_PER_PORT];
1815         if (pu == NULL)
1816                 return;
1817         pu[initidx % CTL_MAX_INIT_PER_PORT] |= ua;
1818 }
1819
1820 void
1821 ctl_est_ua_port(struct ctl_lun *lun, int port, uint32_t except, ctl_ua_type ua)
1822 {
1823         int i;
1824
1825         mtx_assert(&lun->lun_lock, MA_OWNED);
1826         if (lun->pending_ua[port] == NULL)
1827                 return;
1828         for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1829                 if (port * CTL_MAX_INIT_PER_PORT + i == except)
1830                         continue;
1831                 lun->pending_ua[port][i] |= ua;
1832         }
1833 }
1834
1835 void
1836 ctl_est_ua_all(struct ctl_lun *lun, uint32_t except, ctl_ua_type ua)
1837 {
1838         struct ctl_softc *softc = lun->ctl_softc;
1839         int i;
1840
1841         mtx_assert(&lun->lun_lock, MA_OWNED);
1842         for (i = softc->port_min; i < softc->port_max; i++)
1843                 ctl_est_ua_port(lun, i, except, ua);
1844 }
1845
1846 void
1847 ctl_clr_ua(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua)
1848 {
1849         struct ctl_softc *softc = lun->ctl_softc;
1850         ctl_ua_type *pu;
1851
1852         if (initidx < softc->init_min || initidx >= softc->init_max)
1853                 return;
1854         mtx_assert(&lun->lun_lock, MA_OWNED);
1855         pu = lun->pending_ua[initidx / CTL_MAX_INIT_PER_PORT];
1856         if (pu == NULL)
1857                 return;
1858         pu[initidx % CTL_MAX_INIT_PER_PORT] &= ~ua;
1859 }
1860
1861 void
1862 ctl_clr_ua_all(struct ctl_lun *lun, uint32_t except, ctl_ua_type ua)
1863 {
1864         struct ctl_softc *softc = lun->ctl_softc;
1865         int i, j;
1866
1867         mtx_assert(&lun->lun_lock, MA_OWNED);
1868         for (i = softc->port_min; i < softc->port_max; i++) {
1869                 if (lun->pending_ua[i] == NULL)
1870                         continue;
1871                 for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
1872                         if (i * CTL_MAX_INIT_PER_PORT + j == except)
1873                                 continue;
1874                         lun->pending_ua[i][j] &= ~ua;
1875                 }
1876         }
1877 }
1878
1879 void
1880 ctl_clr_ua_allluns(struct ctl_softc *ctl_softc, uint32_t initidx,
1881     ctl_ua_type ua_type)
1882 {
1883         struct ctl_lun *lun;
1884
1885         mtx_assert(&ctl_softc->ctl_lock, MA_OWNED);
1886         STAILQ_FOREACH(lun, &ctl_softc->lun_list, links) {
1887                 mtx_lock(&lun->lun_lock);
1888                 ctl_clr_ua(lun, initidx, ua_type);
1889                 mtx_unlock(&lun->lun_lock);
1890         }
1891 }
1892
1893 static int
1894 ctl_ha_role_sysctl(SYSCTL_HANDLER_ARGS)
1895 {
1896         struct ctl_softc *softc = (struct ctl_softc *)arg1;
1897         struct ctl_lun *lun;
1898         struct ctl_lun_req ireq;
1899         int error, value;
1900
1901         value = (softc->flags & CTL_FLAG_ACTIVE_SHELF) ? 0 : 1;
1902         error = sysctl_handle_int(oidp, &value, 0, req);
1903         if ((error != 0) || (req->newptr == NULL))
1904                 return (error);
1905
1906         mtx_lock(&softc->ctl_lock);
1907         if (value == 0)
1908                 softc->flags |= CTL_FLAG_ACTIVE_SHELF;
1909         else
1910                 softc->flags &= ~CTL_FLAG_ACTIVE_SHELF;
1911         STAILQ_FOREACH(lun, &softc->lun_list, links) {
1912                 mtx_unlock(&softc->ctl_lock);
1913                 bzero(&ireq, sizeof(ireq));
1914                 ireq.reqtype = CTL_LUNREQ_MODIFY;
1915                 ireq.reqdata.modify.lun_id = lun->lun;
1916                 lun->backend->ioctl(NULL, CTL_LUN_REQ, (caddr_t)&ireq, 0,
1917                     curthread);
1918                 if (ireq.status != CTL_LUN_OK) {
1919                         printf("%s: CTL_LUNREQ_MODIFY returned %d '%s'\n",
1920                             __func__, ireq.status, ireq.error_str);
1921                 }
1922                 mtx_lock(&softc->ctl_lock);
1923         }
1924         mtx_unlock(&softc->ctl_lock);
1925         return (0);
1926 }
1927
1928 static int
1929 ctl_init(void)
1930 {
1931         struct make_dev_args args;
1932         struct ctl_softc *softc;
1933         int i, error;
1934
1935         softc = control_softc = malloc(sizeof(*control_softc), M_DEVBUF,
1936                                M_WAITOK | M_ZERO);
1937
1938         make_dev_args_init(&args);
1939         args.mda_devsw = &ctl_cdevsw;
1940         args.mda_uid = UID_ROOT;
1941         args.mda_gid = GID_OPERATOR;
1942         args.mda_mode = 0600;
1943         args.mda_si_drv1 = softc;
1944         args.mda_si_drv2 = NULL;
1945         error = make_dev_s(&args, &softc->dev, "cam/ctl");
1946         if (error != 0) {
1947                 free(softc, M_DEVBUF);
1948                 control_softc = NULL;
1949                 return (error);
1950         }
1951
1952         sysctl_ctx_init(&softc->sysctl_ctx);
1953         softc->sysctl_tree = SYSCTL_ADD_NODE(&softc->sysctl_ctx,
1954                 SYSCTL_STATIC_CHILDREN(_kern_cam), OID_AUTO, "ctl",
1955                 CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "CAM Target Layer");
1956
1957         if (softc->sysctl_tree == NULL) {
1958                 printf("%s: unable to allocate sysctl tree\n", __func__);
1959                 destroy_dev(softc->dev);
1960                 free(softc, M_DEVBUF);
1961                 control_softc = NULL;
1962                 return (ENOMEM);
1963         }
1964
1965         mtx_init(&softc->ctl_lock, "CTL mutex", NULL, MTX_DEF);
1966         softc->io_zone = uma_zcreate("CTL IO", sizeof(union ctl_io),
1967             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1968         softc->flags = 0;
1969
1970         SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1971             OID_AUTO, "ha_mode", CTLFLAG_RDTUN, (int *)&softc->ha_mode, 0,
1972             "HA mode (0 - act/stby, 1 - serialize only, 2 - xfer)");
1973
1974         if (ctl_max_luns <= 0 || powerof2(ctl_max_luns) == 0) {
1975                 printf("Bad value %d for kern.cam.ctl.max_luns, must be a power of two, using %d\n",
1976                     ctl_max_luns, CTL_DEFAULT_MAX_LUNS);
1977                 ctl_max_luns = CTL_DEFAULT_MAX_LUNS;
1978         }
1979         softc->ctl_luns = malloc(sizeof(struct ctl_lun *) * ctl_max_luns,
1980             M_DEVBUF, M_WAITOK | M_ZERO);
1981         softc->ctl_lun_mask = malloc(sizeof(uint32_t) *
1982             ((ctl_max_luns + 31) / 32), M_DEVBUF, M_WAITOK | M_ZERO);
1983         if (ctl_max_ports <= 0 || powerof2(ctl_max_ports) == 0) {
1984                 printf("Bad value %d for kern.cam.ctl.max_ports, must be a power of two, using %d\n",
1985                     ctl_max_ports, CTL_DEFAULT_MAX_PORTS);
1986                 ctl_max_ports = CTL_DEFAULT_MAX_PORTS;
1987         }
1988         softc->ctl_port_mask = malloc(sizeof(uint32_t) *
1989           ((ctl_max_ports + 31) / 32), M_DEVBUF, M_WAITOK | M_ZERO);
1990         softc->ctl_ports = malloc(sizeof(struct ctl_port *) * ctl_max_ports,
1991              M_DEVBUF, M_WAITOK | M_ZERO);
1992
1993         /*
1994          * In Copan's HA scheme, the "master" and "slave" roles are
1995          * figured out through the slot the controller is in.  Although it
1996          * is an active/active system, someone has to be in charge.
1997          */
1998         SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1999             OID_AUTO, "ha_id", CTLFLAG_RDTUN, &softc->ha_id, 0,
2000             "HA head ID (0 - no HA)");
2001         if (softc->ha_id == 0 || softc->ha_id > NUM_HA_SHELVES) {
2002                 softc->flags |= CTL_FLAG_ACTIVE_SHELF;
2003                 softc->is_single = 1;
2004                 softc->port_cnt = ctl_max_ports;
2005                 softc->port_min = 0;
2006         } else {
2007                 softc->port_cnt = ctl_max_ports / NUM_HA_SHELVES;
2008                 softc->port_min = (softc->ha_id - 1) * softc->port_cnt;
2009         }
2010         softc->port_max = softc->port_min + softc->port_cnt;
2011         softc->init_min = softc->port_min * CTL_MAX_INIT_PER_PORT;
2012         softc->init_max = softc->port_max * CTL_MAX_INIT_PER_PORT;
2013
2014         SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
2015             OID_AUTO, "ha_link", CTLFLAG_RD, (int *)&softc->ha_link, 0,
2016             "HA link state (0 - offline, 1 - unknown, 2 - online)");
2017
2018         STAILQ_INIT(&softc->lun_list);
2019         STAILQ_INIT(&softc->fe_list);
2020         STAILQ_INIT(&softc->port_list);
2021         STAILQ_INIT(&softc->be_list);
2022         ctl_tpc_init(softc);
2023
2024         if (worker_threads <= 0)
2025                 worker_threads = max(1, mp_ncpus / 4);
2026         if (worker_threads > CTL_MAX_THREADS)
2027                 worker_threads = CTL_MAX_THREADS;
2028
2029         for (i = 0; i < worker_threads; i++) {
2030                 struct ctl_thread *thr = &softc->threads[i];
2031
2032                 mtx_init(&thr->queue_lock, "CTL queue mutex", NULL, MTX_DEF);
2033                 thr->ctl_softc = softc;
2034                 STAILQ_INIT(&thr->incoming_queue);
2035                 STAILQ_INIT(&thr->rtr_queue);
2036                 STAILQ_INIT(&thr->done_queue);
2037                 STAILQ_INIT(&thr->isc_queue);
2038
2039                 error = kproc_kthread_add(ctl_work_thread, thr,
2040                     &softc->ctl_proc, &thr->thread, 0, 0, "ctl", "work%d", i);
2041                 if (error != 0) {
2042                         printf("error creating CTL work thread!\n");
2043                         return (error);
2044                 }
2045         }
2046         error = kproc_kthread_add(ctl_thresh_thread, softc,
2047             &softc->ctl_proc, &softc->thresh_thread, 0, 0, "ctl", "thresh");
2048         if (error != 0) {
2049                 printf("error creating CTL threshold thread!\n");
2050                 return (error);
2051         }
2052
2053         SYSCTL_ADD_PROC(&softc->sysctl_ctx,SYSCTL_CHILDREN(softc->sysctl_tree),
2054             OID_AUTO, "ha_role",
2055             CTLTYPE_INT | CTLFLAG_RWTUN | CTLFLAG_MPSAFE,
2056             softc, 0, ctl_ha_role_sysctl, "I", "HA role for this head");
2057
2058         if (softc->is_single == 0) {
2059                 if (ctl_frontend_register(&ha_frontend) != 0)
2060                         softc->is_single = 1;
2061         }
2062         return (0);
2063 }
2064
2065 static int
2066 ctl_shutdown(void)
2067 {
2068         struct ctl_softc *softc = control_softc;
2069         int i;
2070
2071         if (softc->is_single == 0)
2072                 ctl_frontend_deregister(&ha_frontend);
2073
2074         destroy_dev(softc->dev);
2075
2076         /* Shutdown CTL threads. */
2077         softc->shutdown = 1;
2078         for (i = 0; i < worker_threads; i++) {
2079                 struct ctl_thread *thr = &softc->threads[i];
2080                 while (thr->thread != NULL) {
2081                         wakeup(thr);
2082                         if (thr->thread != NULL)
2083                                 pause("CTL thr shutdown", 1);
2084                 }
2085                 mtx_destroy(&thr->queue_lock);
2086         }
2087         while (softc->thresh_thread != NULL) {
2088                 wakeup(softc->thresh_thread);
2089                 if (softc->thresh_thread != NULL)
2090                         pause("CTL thr shutdown", 1);
2091         }
2092
2093         ctl_tpc_shutdown(softc);
2094         uma_zdestroy(softc->io_zone);
2095         mtx_destroy(&softc->ctl_lock);
2096
2097         free(softc->ctl_luns, M_DEVBUF);
2098         free(softc->ctl_lun_mask, M_DEVBUF);
2099         free(softc->ctl_port_mask, M_DEVBUF);
2100         free(softc->ctl_ports, M_DEVBUF);
2101
2102         sysctl_ctx_free(&softc->sysctl_ctx);
2103
2104         free(softc, M_DEVBUF);
2105         control_softc = NULL;
2106         return (0);
2107 }
2108
2109 static int
2110 ctl_module_event_handler(module_t mod, int what, void *arg)
2111 {
2112
2113         switch (what) {
2114         case MOD_LOAD:
2115                 return (ctl_init());
2116         case MOD_UNLOAD:
2117                 return (ctl_shutdown());
2118         default:
2119                 return (EOPNOTSUPP);
2120         }
2121 }
2122
2123 /*
2124  * XXX KDM should we do some access checks here?  Bump a reference count to
2125  * prevent a CTL module from being unloaded while someone has it open?
2126  */
2127 static int
2128 ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td)
2129 {
2130         return (0);
2131 }
2132
2133 static int
2134 ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td)
2135 {
2136         return (0);
2137 }
2138
2139 /*
2140  * Remove an initiator by port number and initiator ID.
2141  * Returns 0 for success, -1 for failure.
2142  */
2143 int
2144 ctl_remove_initiator(struct ctl_port *port, int iid)
2145 {
2146         struct ctl_softc *softc = port->ctl_softc;
2147         int last;
2148
2149         mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
2150
2151         if (iid > CTL_MAX_INIT_PER_PORT) {
2152                 printf("%s: initiator ID %u > maximun %u!\n",
2153                        __func__, iid, CTL_MAX_INIT_PER_PORT);
2154                 return (-1);
2155         }
2156
2157         mtx_lock(&softc->ctl_lock);
2158         last = (--port->wwpn_iid[iid].in_use == 0);
2159         port->wwpn_iid[iid].last_use = time_uptime;
2160         mtx_unlock(&softc->ctl_lock);
2161         if (last)
2162                 ctl_i_t_nexus_loss(softc, iid, CTL_UA_POWERON);
2163         ctl_isc_announce_iid(port, iid);
2164
2165         return (0);
2166 }
2167
2168 /*
2169  * Add an initiator to the initiator map.
2170  * Returns iid for success, < 0 for failure.
2171  */
2172 int
2173 ctl_add_initiator(struct ctl_port *port, int iid, uint64_t wwpn, char *name)
2174 {
2175         struct ctl_softc *softc = port->ctl_softc;
2176         time_t best_time;
2177         int i, best;
2178
2179         mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
2180
2181         if (iid >= CTL_MAX_INIT_PER_PORT) {
2182                 printf("%s: WWPN %#jx initiator ID %u > maximum %u!\n",
2183                        __func__, wwpn, iid, CTL_MAX_INIT_PER_PORT);
2184                 free(name, M_CTL);
2185                 return (-1);
2186         }
2187
2188         mtx_lock(&softc->ctl_lock);
2189
2190         if (iid < 0 && (wwpn != 0 || name != NULL)) {
2191                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
2192                         if (wwpn != 0 && wwpn == port->wwpn_iid[i].wwpn) {
2193                                 iid = i;
2194                                 break;
2195                         }
2196                         if (name != NULL && port->wwpn_iid[i].name != NULL &&
2197                             strcmp(name, port->wwpn_iid[i].name) == 0) {
2198                                 iid = i;
2199                                 break;
2200                         }
2201                 }
2202         }
2203
2204         if (iid < 0) {
2205                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
2206                         if (port->wwpn_iid[i].in_use == 0 &&
2207                             port->wwpn_iid[i].wwpn == 0 &&
2208                             port->wwpn_iid[i].name == NULL) {
2209                                 iid = i;
2210                                 break;
2211                         }
2212                 }
2213         }
2214
2215         if (iid < 0) {
2216                 best = -1;
2217                 best_time = INT32_MAX;
2218                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
2219                         if (port->wwpn_iid[i].in_use == 0) {
2220                                 if (port->wwpn_iid[i].last_use < best_time) {
2221                                         best = i;
2222                                         best_time = port->wwpn_iid[i].last_use;
2223                                 }
2224                         }
2225                 }
2226                 iid = best;
2227         }
2228
2229         if (iid < 0) {
2230                 mtx_unlock(&softc->ctl_lock);
2231                 free(name, M_CTL);
2232                 return (-2);
2233         }
2234
2235         if (port->wwpn_iid[iid].in_use > 0 && (wwpn != 0 || name != NULL)) {
2236                 /*
2237                  * This is not an error yet.
2238                  */
2239                 if (wwpn != 0 && wwpn == port->wwpn_iid[iid].wwpn) {
2240 #if 0
2241                         printf("%s: port %d iid %u WWPN %#jx arrived"
2242                             " again\n", __func__, port->targ_port,
2243                             iid, (uintmax_t)wwpn);
2244 #endif
2245                         goto take;
2246                 }
2247                 if (name != NULL && port->wwpn_iid[iid].name != NULL &&
2248                     strcmp(name, port->wwpn_iid[iid].name) == 0) {
2249 #if 0
2250                         printf("%s: port %d iid %u name '%s' arrived"
2251                             " again\n", __func__, port->targ_port,
2252                             iid, name);
2253 #endif
2254                         goto take;
2255                 }
2256
2257                 /*
2258                  * This is an error, but what do we do about it?  The
2259                  * driver is telling us we have a new WWPN for this
2260                  * initiator ID, so we pretty much need to use it.
2261                  */
2262                 printf("%s: port %d iid %u WWPN %#jx '%s' arrived,"
2263                     " but WWPN %#jx '%s' is still at that address\n",
2264                     __func__, port->targ_port, iid, wwpn, name,
2265                     (uintmax_t)port->wwpn_iid[iid].wwpn,
2266                     port->wwpn_iid[iid].name);
2267         }
2268 take:
2269         free(port->wwpn_iid[iid].name, M_CTL);
2270         port->wwpn_iid[iid].name = name;
2271         port->wwpn_iid[iid].wwpn = wwpn;
2272         port->wwpn_iid[iid].in_use++;
2273         mtx_unlock(&softc->ctl_lock);
2274         ctl_isc_announce_iid(port, iid);
2275
2276         return (iid);
2277 }
2278
2279 static int
2280 ctl_create_iid(struct ctl_port *port, int iid, uint8_t *buf)
2281 {
2282         int len;
2283
2284         switch (port->port_type) {
2285         case CTL_PORT_FC:
2286         {
2287                 struct scsi_transportid_fcp *id =
2288                     (struct scsi_transportid_fcp *)buf;
2289                 if (port->wwpn_iid[iid].wwpn == 0)
2290                         return (0);
2291                 memset(id, 0, sizeof(*id));
2292                 id->format_protocol = SCSI_PROTO_FC;
2293                 scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->n_port_name);
2294                 return (sizeof(*id));
2295         }
2296         case CTL_PORT_ISCSI:
2297         {
2298                 struct scsi_transportid_iscsi_port *id =
2299                     (struct scsi_transportid_iscsi_port *)buf;
2300                 if (port->wwpn_iid[iid].name == NULL)
2301                         return (0);
2302                 memset(id, 0, 256);
2303                 id->format_protocol = SCSI_TRN_ISCSI_FORMAT_PORT |
2304                     SCSI_PROTO_ISCSI;
2305                 len = strlcpy(id->iscsi_name, port->wwpn_iid[iid].name, 252) + 1;
2306                 len = roundup2(min(len, 252), 4);
2307                 scsi_ulto2b(len, id->additional_length);
2308                 return (sizeof(*id) + len);
2309         }
2310         case CTL_PORT_SAS:
2311         {
2312                 struct scsi_transportid_sas *id =
2313                     (struct scsi_transportid_sas *)buf;
2314                 if (port->wwpn_iid[iid].wwpn == 0)
2315                         return (0);
2316                 memset(id, 0, sizeof(*id));
2317                 id->format_protocol = SCSI_PROTO_SAS;
2318                 scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->sas_address);
2319                 return (sizeof(*id));
2320         }
2321         default:
2322         {
2323                 struct scsi_transportid_spi *id =
2324                     (struct scsi_transportid_spi *)buf;
2325                 memset(id, 0, sizeof(*id));
2326                 id->format_protocol = SCSI_PROTO_SPI;
2327                 scsi_ulto2b(iid, id->scsi_addr);
2328                 scsi_ulto2b(port->targ_port, id->rel_trgt_port_id);
2329                 return (sizeof(*id));
2330         }
2331         }
2332 }
2333
2334 /*
2335  * Serialize a command that went down the "wrong" side, and so was sent to
2336  * this controller for execution.  The logic is a little different than the
2337  * standard case in ctl_scsiio_precheck().  Errors in this case need to get
2338  * sent back to the other side, but in the success case, we execute the
2339  * command on this side (XFER mode) or tell the other side to execute it
2340  * (SER_ONLY mode).
2341  */
2342 static void
2343 ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio)
2344 {
2345         struct ctl_softc *softc = CTL_SOFTC(ctsio);
2346         struct ctl_port *port = CTL_PORT(ctsio);
2347         union ctl_ha_msg msg_info;
2348         struct ctl_lun *lun;
2349         const struct ctl_cmd_entry *entry;
2350         union ctl_io *bio;
2351         uint32_t targ_lun;
2352
2353         targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
2354
2355         /* Make sure that we know about this port. */
2356         if (port == NULL || (port->status & CTL_PORT_STATUS_ONLINE) == 0) {
2357                 ctl_set_internal_failure(ctsio, /*sks_valid*/ 0,
2358                                          /*retry_count*/ 1);
2359                 goto badjuju;
2360         }
2361
2362         /* Make sure that we know about this LUN. */
2363         mtx_lock(&softc->ctl_lock);
2364         if (targ_lun >= ctl_max_luns ||
2365             (lun = softc->ctl_luns[targ_lun]) == NULL) {
2366                 mtx_unlock(&softc->ctl_lock);
2367
2368                 /*
2369                  * The other node would not send this request to us unless
2370                  * received announce that we are primary node for this LUN.
2371                  * If this LUN does not exist now, it is probably result of
2372                  * a race, so respond to initiator in the most opaque way.
2373                  */
2374                 ctl_set_busy(ctsio);
2375                 goto badjuju;
2376         }
2377         mtx_lock(&lun->lun_lock);
2378         mtx_unlock(&softc->ctl_lock);
2379
2380         /*
2381          * If the LUN is invalid, pretend that it doesn't exist.
2382          * It will go away as soon as all pending I/Os completed.
2383          */
2384         if (lun->flags & CTL_LUN_DISABLED) {
2385                 mtx_unlock(&lun->lun_lock);
2386                 ctl_set_busy(ctsio);
2387                 goto badjuju;
2388         }
2389
2390         entry = ctl_get_cmd_entry(ctsio, NULL);
2391         ctsio->seridx = entry->seridx;
2392         if (ctl_scsiio_lun_check(lun, entry, ctsio) != 0) {
2393                 mtx_unlock(&lun->lun_lock);
2394                 goto badjuju;
2395         }
2396
2397         CTL_LUN(ctsio) = lun;
2398         CTL_BACKEND_LUN(ctsio) = lun->be_lun;
2399
2400         /*
2401          * Every I/O goes into the OOA queue for a
2402          * particular LUN, and stays there until completion.
2403          */
2404 #ifdef CTL_TIME_IO
2405         if (LIST_EMPTY(&lun->ooa_queue))
2406                 lun->idle_time += getsbinuptime() - lun->last_busy;
2407 #endif
2408         LIST_INSERT_HEAD(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2409
2410         bio = (union ctl_io *)LIST_NEXT(&ctsio->io_hdr, ooa_links);
2411         switch (ctl_check_ooa(lun, (union ctl_io *)ctsio, &bio)) {
2412         case CTL_ACTION_PASS:
2413         case CTL_ACTION_SKIP:
2414                 if (softc->ha_mode == CTL_HA_MODE_XFER) {
2415                         ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
2416                         ctl_enqueue_rtr((union ctl_io *)ctsio);
2417                         mtx_unlock(&lun->lun_lock);
2418                 } else {
2419                         ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
2420                         mtx_unlock(&lun->lun_lock);
2421
2422                         /* send msg back to other side */
2423                         msg_info.hdr.original_sc = ctsio->io_hdr.remote_io;
2424                         msg_info.hdr.serializing_sc = (union ctl_io *)ctsio;
2425                         msg_info.hdr.msg_type = CTL_MSG_R2R;
2426                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
2427                             sizeof(msg_info.hdr), M_WAITOK);
2428                 }
2429                 break;
2430         case CTL_ACTION_BLOCK:
2431                 ctsio->io_hdr.blocker = bio;
2432                 TAILQ_INSERT_TAIL(&bio->io_hdr.blocked_queue, &ctsio->io_hdr,
2433                                   blocked_links);
2434                 mtx_unlock(&lun->lun_lock);
2435                 break;
2436         case CTL_ACTION_OVERLAP:
2437                 LIST_REMOVE(&ctsio->io_hdr, ooa_links);
2438                 mtx_unlock(&lun->lun_lock);
2439                 ctl_set_overlapped_cmd(ctsio);
2440                 goto badjuju;
2441         case CTL_ACTION_OVERLAP_TAG:
2442                 LIST_REMOVE(&ctsio->io_hdr, ooa_links);
2443                 mtx_unlock(&lun->lun_lock);
2444                 ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
2445 badjuju:
2446                 ctl_copy_sense_data_back((union ctl_io *)ctsio, &msg_info);
2447                 msg_info.hdr.original_sc = ctsio->io_hdr.remote_io;
2448                 msg_info.hdr.serializing_sc = NULL;
2449                 msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
2450                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
2451                     sizeof(msg_info.scsi), M_WAITOK);
2452                 ctl_free_io((union ctl_io *)ctsio);
2453                 break;
2454         default:
2455                 __assert_unreachable();
2456         }
2457 }
2458
2459 /*
2460  * Returns 0 for success, errno for failure.
2461  */
2462 static void
2463 ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
2464                    struct ctl_ooa *ooa_hdr, struct ctl_ooa_entry *kern_entries)
2465 {
2466         struct ctl_io_hdr *ioh;
2467
2468         mtx_lock(&lun->lun_lock);
2469         ioh = LIST_FIRST(&lun->ooa_queue);
2470         if (ioh == NULL) {
2471                 mtx_unlock(&lun->lun_lock);
2472                 return;
2473         }
2474         while (LIST_NEXT(ioh, ooa_links) != NULL)
2475                 ioh = LIST_NEXT(ioh, ooa_links);
2476         for ( ; ioh; ioh = LIST_PREV(ioh, &lun->ooa_queue, ctl_io_hdr, ooa_links)) {
2477                 union ctl_io *io = (union ctl_io *)ioh;
2478                 struct ctl_ooa_entry *entry;
2479
2480                 /*
2481                  * If we've got more than we can fit, just count the
2482                  * remaining entries.
2483                  */
2484                 if (*cur_fill_num >= ooa_hdr->alloc_num) {
2485                         (*cur_fill_num)++;
2486                         continue;
2487                 }
2488
2489                 entry = &kern_entries[*cur_fill_num];
2490
2491                 entry->tag_num = io->scsiio.tag_num;
2492                 entry->tag_type = io->scsiio.tag_type;
2493                 entry->lun_num = lun->lun;
2494 #ifdef CTL_TIME_IO
2495                 entry->start_bt = io->io_hdr.start_bt;
2496 #endif
2497                 bcopy(io->scsiio.cdb, entry->cdb, io->scsiio.cdb_len);
2498                 entry->cdb_len = io->scsiio.cdb_len;
2499                 if (io->io_hdr.blocker != NULL)
2500                         entry->cmd_flags |= CTL_OOACMD_FLAG_BLOCKED;
2501
2502                 if (io->io_hdr.flags & CTL_FLAG_DMA_INPROG)
2503                         entry->cmd_flags |= CTL_OOACMD_FLAG_DMA;
2504
2505                 if (io->io_hdr.flags & CTL_FLAG_ABORT)
2506                         entry->cmd_flags |= CTL_OOACMD_FLAG_ABORT;
2507
2508                 if (io->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR)
2509                         entry->cmd_flags |= CTL_OOACMD_FLAG_RTR;
2510
2511                 if (io->io_hdr.flags & CTL_FLAG_DMA_QUEUED)
2512                         entry->cmd_flags |= CTL_OOACMD_FLAG_DMA_QUEUED;
2513
2514                 if (io->io_hdr.flags & CTL_FLAG_STATUS_QUEUED)
2515                         entry->cmd_flags |= CTL_OOACMD_FLAG_STATUS_QUEUED;
2516
2517                 if (io->io_hdr.flags & CTL_FLAG_STATUS_SENT)
2518                         entry->cmd_flags |= CTL_OOACMD_FLAG_STATUS_SENT;
2519                 (*cur_fill_num)++;
2520         }
2521         mtx_unlock(&lun->lun_lock);
2522 }
2523
2524 /*
2525  * Escape characters that are illegal or not recommended in XML.
2526  */
2527 int
2528 ctl_sbuf_printf_esc(struct sbuf *sb, char *str, int size)
2529 {
2530         char *end = str + size;
2531         int retval;
2532
2533         retval = 0;
2534
2535         for (; *str && str < end; str++) {
2536                 switch (*str) {
2537                 case '&':
2538                         retval = sbuf_printf(sb, "&amp;");
2539                         break;
2540                 case '>':
2541                         retval = sbuf_printf(sb, "&gt;");
2542                         break;
2543                 case '<':
2544                         retval = sbuf_printf(sb, "&lt;");
2545                         break;
2546                 default:
2547                         retval = sbuf_putc(sb, *str);
2548                         break;
2549                 }
2550
2551                 if (retval != 0)
2552                         break;
2553         }
2554
2555         return (retval);
2556 }
2557
2558 static void
2559 ctl_id_sbuf(struct ctl_devid *id, struct sbuf *sb)
2560 {
2561         struct scsi_vpd_id_descriptor *desc;
2562         int i;
2563
2564         if (id == NULL || id->len < 4)
2565                 return;
2566         desc = (struct scsi_vpd_id_descriptor *)id->data;
2567         switch (desc->id_type & SVPD_ID_TYPE_MASK) {
2568         case SVPD_ID_TYPE_T10:
2569                 sbuf_printf(sb, "t10.");
2570                 break;
2571         case SVPD_ID_TYPE_EUI64:
2572                 sbuf_printf(sb, "eui.");
2573                 break;
2574         case SVPD_ID_TYPE_NAA:
2575                 sbuf_printf(sb, "naa.");
2576                 break;
2577         case SVPD_ID_TYPE_SCSI_NAME:
2578                 break;
2579         }
2580         switch (desc->proto_codeset & SVPD_ID_CODESET_MASK) {
2581         case SVPD_ID_CODESET_BINARY:
2582                 for (i = 0; i < desc->length; i++)
2583                         sbuf_printf(sb, "%02x", desc->identifier[i]);
2584                 break;
2585         case SVPD_ID_CODESET_ASCII:
2586                 sbuf_printf(sb, "%.*s", (int)desc->length,
2587                     (char *)desc->identifier);
2588                 break;
2589         case SVPD_ID_CODESET_UTF8:
2590                 sbuf_printf(sb, "%s", (char *)desc->identifier);
2591                 break;
2592         }
2593 }
2594
2595 static int
2596 ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
2597           struct thread *td)
2598 {
2599         struct ctl_softc *softc = dev->si_drv1;
2600         struct ctl_port *port;
2601         struct ctl_lun *lun;
2602         int retval;
2603
2604         retval = 0;
2605
2606         switch (cmd) {
2607         case CTL_IO:
2608                 retval = ctl_ioctl_io(dev, cmd, addr, flag, td);
2609                 break;
2610         case CTL_ENABLE_PORT:
2611         case CTL_DISABLE_PORT:
2612         case CTL_SET_PORT_WWNS: {
2613                 struct ctl_port *port;
2614                 struct ctl_port_entry *entry;
2615
2616                 entry = (struct ctl_port_entry *)addr;
2617
2618                 mtx_lock(&softc->ctl_lock);
2619                 STAILQ_FOREACH(port, &softc->port_list, links) {
2620                         int action, done;
2621
2622                         if (port->targ_port < softc->port_min ||
2623                             port->targ_port >= softc->port_max)
2624                                 continue;
2625
2626                         action = 0;
2627                         done = 0;
2628                         if ((entry->port_type == CTL_PORT_NONE)
2629                          && (entry->targ_port == port->targ_port)) {
2630                                 /*
2631                                  * If the user only wants to enable or
2632                                  * disable or set WWNs on a specific port,
2633                                  * do the operation and we're done.
2634                                  */
2635                                 action = 1;
2636                                 done = 1;
2637                         } else if (entry->port_type & port->port_type) {
2638                                 /*
2639                                  * Compare the user's type mask with the
2640                                  * particular frontend type to see if we
2641                                  * have a match.
2642                                  */
2643                                 action = 1;
2644                                 done = 0;
2645
2646                                 /*
2647                                  * Make sure the user isn't trying to set
2648                                  * WWNs on multiple ports at the same time.
2649                                  */
2650                                 if (cmd == CTL_SET_PORT_WWNS) {
2651                                         printf("%s: Can't set WWNs on "
2652                                                "multiple ports\n", __func__);
2653                                         retval = EINVAL;
2654                                         break;
2655                                 }
2656                         }
2657                         if (action == 0)
2658                                 continue;
2659
2660                         /*
2661                          * XXX KDM we have to drop the lock here, because
2662                          * the online/offline operations can potentially
2663                          * block.  We need to reference count the frontends
2664                          * so they can't go away,
2665                          */
2666                         if (cmd == CTL_ENABLE_PORT) {
2667                                 mtx_unlock(&softc->ctl_lock);
2668                                 ctl_port_online(port);
2669                                 mtx_lock(&softc->ctl_lock);
2670                         } else if (cmd == CTL_DISABLE_PORT) {
2671                                 mtx_unlock(&softc->ctl_lock);
2672                                 ctl_port_offline(port);
2673                                 mtx_lock(&softc->ctl_lock);
2674                         } else if (cmd == CTL_SET_PORT_WWNS) {
2675                                 ctl_port_set_wwns(port,
2676                                     (entry->flags & CTL_PORT_WWNN_VALID) ?
2677                                     1 : 0, entry->wwnn,
2678                                     (entry->flags & CTL_PORT_WWPN_VALID) ?
2679                                     1 : 0, entry->wwpn);
2680                         }
2681                         if (done != 0)
2682                                 break;
2683                 }
2684                 mtx_unlock(&softc->ctl_lock);
2685                 break;
2686         }
2687         case CTL_GET_OOA: {
2688                 struct ctl_ooa *ooa_hdr;
2689                 struct ctl_ooa_entry *entries;
2690                 uint32_t cur_fill_num;
2691
2692                 ooa_hdr = (struct ctl_ooa *)addr;
2693
2694                 if ((ooa_hdr->alloc_len == 0)
2695                  || (ooa_hdr->alloc_num == 0)) {
2696                         printf("%s: CTL_GET_OOA: alloc len %u and alloc num %u "
2697                                "must be non-zero\n", __func__,
2698                                ooa_hdr->alloc_len, ooa_hdr->alloc_num);
2699                         retval = EINVAL;
2700                         break;
2701                 }
2702
2703                 if (ooa_hdr->alloc_len != (ooa_hdr->alloc_num *
2704                     sizeof(struct ctl_ooa_entry))) {
2705                         printf("%s: CTL_GET_OOA: alloc len %u must be alloc "
2706                                "num %d * sizeof(struct ctl_ooa_entry) %zd\n",
2707                                __func__, ooa_hdr->alloc_len,
2708                                ooa_hdr->alloc_num,sizeof(struct ctl_ooa_entry));
2709                         retval = EINVAL;
2710                         break;
2711                 }
2712
2713                 entries = malloc(ooa_hdr->alloc_len, M_CTL, M_WAITOK | M_ZERO);
2714                 if (entries == NULL) {
2715                         printf("%s: could not allocate %d bytes for OOA "
2716                                "dump\n", __func__, ooa_hdr->alloc_len);
2717                         retval = ENOMEM;
2718                         break;
2719                 }
2720
2721                 mtx_lock(&softc->ctl_lock);
2722                 if ((ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) == 0 &&
2723                     (ooa_hdr->lun_num >= ctl_max_luns ||
2724                      softc->ctl_luns[ooa_hdr->lun_num] == NULL)) {
2725                         mtx_unlock(&softc->ctl_lock);
2726                         free(entries, M_CTL);
2727                         printf("%s: CTL_GET_OOA: invalid LUN %ju\n",
2728                                __func__, (uintmax_t)ooa_hdr->lun_num);
2729                         retval = EINVAL;
2730                         break;
2731                 }
2732
2733                 cur_fill_num = 0;
2734
2735                 if (ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) {
2736                         STAILQ_FOREACH(lun, &softc->lun_list, links) {
2737                                 ctl_ioctl_fill_ooa(lun, &cur_fill_num,
2738                                     ooa_hdr, entries);
2739                         }
2740                 } else {
2741                         lun = softc->ctl_luns[ooa_hdr->lun_num];
2742                         ctl_ioctl_fill_ooa(lun, &cur_fill_num, ooa_hdr,
2743                             entries);
2744                 }
2745                 mtx_unlock(&softc->ctl_lock);
2746
2747                 ooa_hdr->fill_num = min(cur_fill_num, ooa_hdr->alloc_num);
2748                 ooa_hdr->fill_len = ooa_hdr->fill_num *
2749                         sizeof(struct ctl_ooa_entry);
2750                 retval = copyout(entries, ooa_hdr->entries, ooa_hdr->fill_len);
2751                 if (retval != 0) {
2752                         printf("%s: error copying out %d bytes for OOA dump\n",
2753                                __func__, ooa_hdr->fill_len);
2754                 }
2755
2756                 getbinuptime(&ooa_hdr->cur_bt);
2757
2758                 if (cur_fill_num > ooa_hdr->alloc_num) {
2759                         ooa_hdr->dropped_num = cur_fill_num -ooa_hdr->alloc_num;
2760                         ooa_hdr->status = CTL_OOA_NEED_MORE_SPACE;
2761                 } else {
2762                         ooa_hdr->dropped_num = 0;
2763                         ooa_hdr->status = CTL_OOA_OK;
2764                 }
2765
2766                 free(entries, M_CTL);
2767                 break;
2768         }
2769         case CTL_DELAY_IO: {
2770                 struct ctl_io_delay_info *delay_info;
2771
2772                 delay_info = (struct ctl_io_delay_info *)addr;
2773
2774 #ifdef CTL_IO_DELAY
2775                 mtx_lock(&softc->ctl_lock);
2776                 if (delay_info->lun_id >= ctl_max_luns ||
2777                     (lun = softc->ctl_luns[delay_info->lun_id]) == NULL) {
2778                         mtx_unlock(&softc->ctl_lock);
2779                         delay_info->status = CTL_DELAY_STATUS_INVALID_LUN;
2780                         break;
2781                 }
2782                 mtx_lock(&lun->lun_lock);
2783                 mtx_unlock(&softc->ctl_lock);
2784                 delay_info->status = CTL_DELAY_STATUS_OK;
2785                 switch (delay_info->delay_type) {
2786                 case CTL_DELAY_TYPE_CONT:
2787                 case CTL_DELAY_TYPE_ONESHOT:
2788                         break;
2789                 default:
2790                         delay_info->status = CTL_DELAY_STATUS_INVALID_TYPE;
2791                         break;
2792                 }
2793                 switch (delay_info->delay_loc) {
2794                 case CTL_DELAY_LOC_DATAMOVE:
2795                         lun->delay_info.datamove_type = delay_info->delay_type;
2796                         lun->delay_info.datamove_delay = delay_info->delay_secs;
2797                         break;
2798                 case CTL_DELAY_LOC_DONE:
2799                         lun->delay_info.done_type = delay_info->delay_type;
2800                         lun->delay_info.done_delay = delay_info->delay_secs;
2801                         break;
2802                 default:
2803                         delay_info->status = CTL_DELAY_STATUS_INVALID_LOC;
2804                         break;
2805                 }
2806                 mtx_unlock(&lun->lun_lock);
2807 #else
2808                 delay_info->status = CTL_DELAY_STATUS_NOT_IMPLEMENTED;
2809 #endif /* CTL_IO_DELAY */
2810                 break;
2811         }
2812         case CTL_ERROR_INJECT: {
2813                 struct ctl_error_desc *err_desc, *new_err_desc;
2814
2815                 err_desc = (struct ctl_error_desc *)addr;
2816
2817                 new_err_desc = malloc(sizeof(*new_err_desc), M_CTL,
2818                                       M_WAITOK | M_ZERO);
2819                 bcopy(err_desc, new_err_desc, sizeof(*new_err_desc));
2820
2821                 mtx_lock(&softc->ctl_lock);
2822                 if (err_desc->lun_id >= ctl_max_luns ||
2823                     (lun = softc->ctl_luns[err_desc->lun_id]) == NULL) {
2824                         mtx_unlock(&softc->ctl_lock);
2825                         free(new_err_desc, M_CTL);
2826                         printf("%s: CTL_ERROR_INJECT: invalid LUN %ju\n",
2827                                __func__, (uintmax_t)err_desc->lun_id);
2828                         retval = EINVAL;
2829                         break;
2830                 }
2831                 mtx_lock(&lun->lun_lock);
2832                 mtx_unlock(&softc->ctl_lock);
2833
2834                 /*
2835                  * We could do some checking here to verify the validity
2836                  * of the request, but given the complexity of error
2837                  * injection requests, the checking logic would be fairly
2838                  * complex.
2839                  *
2840                  * For now, if the request is invalid, it just won't get
2841                  * executed and might get deleted.
2842                  */
2843                 STAILQ_INSERT_TAIL(&lun->error_list, new_err_desc, links);
2844
2845                 /*
2846                  * XXX KDM check to make sure the serial number is unique,
2847                  * in case we somehow manage to wrap.  That shouldn't
2848                  * happen for a very long time, but it's the right thing to
2849                  * do.
2850                  */
2851                 new_err_desc->serial = lun->error_serial;
2852                 err_desc->serial = lun->error_serial;
2853                 lun->error_serial++;
2854
2855                 mtx_unlock(&lun->lun_lock);
2856                 break;
2857         }
2858         case CTL_ERROR_INJECT_DELETE: {
2859                 struct ctl_error_desc *delete_desc, *desc, *desc2;
2860                 int delete_done;
2861
2862                 delete_desc = (struct ctl_error_desc *)addr;
2863                 delete_done = 0;
2864
2865                 mtx_lock(&softc->ctl_lock);
2866                 if (delete_desc->lun_id >= ctl_max_luns ||
2867                     (lun = softc->ctl_luns[delete_desc->lun_id]) == NULL) {
2868                         mtx_unlock(&softc->ctl_lock);
2869                         printf("%s: CTL_ERROR_INJECT_DELETE: invalid LUN %ju\n",
2870                                __func__, (uintmax_t)delete_desc->lun_id);
2871                         retval = EINVAL;
2872                         break;
2873                 }
2874                 mtx_lock(&lun->lun_lock);
2875                 mtx_unlock(&softc->ctl_lock);
2876                 STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
2877                         if (desc->serial != delete_desc->serial)
2878                                 continue;
2879
2880                         STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc,
2881                                       links);
2882                         free(desc, M_CTL);
2883                         delete_done = 1;
2884                 }
2885                 mtx_unlock(&lun->lun_lock);
2886                 if (delete_done == 0) {
2887                         printf("%s: CTL_ERROR_INJECT_DELETE: can't find "
2888                                "error serial %ju on LUN %u\n", __func__,
2889                                delete_desc->serial, delete_desc->lun_id);
2890                         retval = EINVAL;
2891                         break;
2892                 }
2893                 break;
2894         }
2895         case CTL_DUMP_STRUCTS: {
2896                 int j, k;
2897                 struct ctl_port *port;
2898                 struct ctl_frontend *fe;
2899
2900                 mtx_lock(&softc->ctl_lock);
2901                 printf("CTL Persistent Reservation information start:\n");
2902                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
2903                         mtx_lock(&lun->lun_lock);
2904                         if ((lun->flags & CTL_LUN_DISABLED) != 0) {
2905                                 mtx_unlock(&lun->lun_lock);
2906                                 continue;
2907                         }
2908
2909                         for (j = 0; j < ctl_max_ports; j++) {
2910                                 if (lun->pr_keys[j] == NULL)
2911                                         continue;
2912                                 for (k = 0; k < CTL_MAX_INIT_PER_PORT; k++){
2913                                         if (lun->pr_keys[j][k] == 0)
2914                                                 continue;
2915                                         printf("  LUN %ju port %d iid %d key "
2916                                                "%#jx\n", lun->lun, j, k,
2917                                                (uintmax_t)lun->pr_keys[j][k]);
2918                                 }
2919                         }
2920                         mtx_unlock(&lun->lun_lock);
2921                 }
2922                 printf("CTL Persistent Reservation information end\n");
2923                 printf("CTL Ports:\n");
2924                 STAILQ_FOREACH(port, &softc->port_list, links) {
2925                         printf("  Port %d '%s' Frontend '%s' Type %u pp %d vp %d WWNN "
2926                                "%#jx WWPN %#jx\n", port->targ_port, port->port_name,
2927                                port->frontend->name, port->port_type,
2928                                port->physical_port, port->virtual_port,
2929                                (uintmax_t)port->wwnn, (uintmax_t)port->wwpn);
2930                         for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
2931                                 if (port->wwpn_iid[j].in_use == 0 &&
2932                                     port->wwpn_iid[j].wwpn == 0 &&
2933                                     port->wwpn_iid[j].name == NULL)
2934                                         continue;
2935
2936                                 printf("    iid %u use %d WWPN %#jx '%s'\n",
2937                                     j, port->wwpn_iid[j].in_use,
2938                                     (uintmax_t)port->wwpn_iid[j].wwpn,
2939                                     port->wwpn_iid[j].name);
2940                         }
2941                 }
2942                 printf("CTL Port information end\n");
2943                 mtx_unlock(&softc->ctl_lock);
2944                 /*
2945                  * XXX KDM calling this without a lock.  We'd likely want
2946                  * to drop the lock before calling the frontend's dump
2947                  * routine anyway.
2948                  */
2949                 printf("CTL Frontends:\n");
2950                 STAILQ_FOREACH(fe, &softc->fe_list, links) {
2951                         printf("  Frontend '%s'\n", fe->name);
2952                         if (fe->fe_dump != NULL)
2953                                 fe->fe_dump();
2954                 }
2955                 printf("CTL Frontend information end\n");
2956                 break;
2957         }
2958         case CTL_LUN_REQ: {
2959                 struct ctl_lun_req *lun_req;
2960                 struct ctl_backend_driver *backend;
2961                 void *packed;
2962                 nvlist_t *tmp_args_nvl;
2963                 size_t packed_len;
2964
2965                 lun_req = (struct ctl_lun_req *)addr;
2966                 tmp_args_nvl = lun_req->args_nvl;
2967
2968                 backend = ctl_backend_find(lun_req->backend);
2969                 if (backend == NULL) {
2970                         lun_req->status = CTL_LUN_ERROR;
2971                         snprintf(lun_req->error_str,
2972                                  sizeof(lun_req->error_str),
2973                                  "Backend \"%s\" not found.",
2974                                  lun_req->backend);
2975                         break;
2976                 }
2977
2978                 if (lun_req->args != NULL) {
2979                         if (lun_req->args_len > CTL_MAX_ARGS_LEN) {
2980                                 lun_req->status = CTL_LUN_ERROR;
2981                                 snprintf(lun_req->error_str, sizeof(lun_req->error_str),
2982                                     "Too big args.");
2983                                 break;
2984                         }
2985                         packed = malloc(lun_req->args_len, M_CTL, M_WAITOK);
2986                         if (copyin(lun_req->args, packed, lun_req->args_len) != 0) {
2987                                 free(packed, M_CTL);
2988                                 lun_req->status = CTL_LUN_ERROR;
2989                                 snprintf(lun_req->error_str, sizeof(lun_req->error_str),
2990                                     "Cannot copyin args.");
2991                                 break;
2992                         }
2993                         lun_req->args_nvl = nvlist_unpack(packed,
2994                             lun_req->args_len, 0);
2995                         free(packed, M_CTL);
2996
2997                         if (lun_req->args_nvl == NULL) {
2998                                 lun_req->status = CTL_LUN_ERROR;
2999                                 snprintf(lun_req->error_str, sizeof(lun_req->error_str),
3000                                     "Cannot unpack args nvlist.");
3001                                 break;
3002                         }
3003                 } else
3004                         lun_req->args_nvl = nvlist_create(0);
3005
3006                 lun_req->result_nvl = NULL;
3007                 retval = backend->ioctl(dev, cmd, addr, flag, td);
3008                 nvlist_destroy(lun_req->args_nvl);
3009                 lun_req->args_nvl = tmp_args_nvl;
3010
3011                 if (lun_req->result_nvl != NULL) {
3012                         if (lun_req->result != NULL) {
3013                                 packed = nvlist_pack(lun_req->result_nvl,
3014                                     &packed_len);
3015                                 if (packed == NULL) {
3016                                         lun_req->status = CTL_LUN_ERROR;
3017                                         snprintf(lun_req->error_str,
3018                                             sizeof(lun_req->error_str),
3019                                             "Cannot pack result nvlist.");
3020                                         break;
3021                                 }
3022
3023                                 if (packed_len > lun_req->result_len) {
3024                                         lun_req->status = CTL_LUN_ERROR;
3025                                         snprintf(lun_req->error_str,
3026                                             sizeof(lun_req->error_str),
3027                                             "Result nvlist too large.");
3028                                         free(packed, M_NVLIST);
3029                                         break;
3030                                 }
3031
3032                                 if (copyout(packed, lun_req->result, packed_len)) {
3033                                         lun_req->status = CTL_LUN_ERROR;
3034                                         snprintf(lun_req->error_str,
3035                                             sizeof(lun_req->error_str),
3036                                             "Cannot copyout() the result.");
3037                                         free(packed, M_NVLIST);
3038                                         break;
3039                                 }
3040
3041                                 lun_req->result_len = packed_len;
3042                                 free(packed, M_NVLIST);
3043                         }
3044
3045                         nvlist_destroy(lun_req->result_nvl);
3046                 }
3047                 break;
3048         }
3049         case CTL_LUN_LIST: {
3050                 struct sbuf *sb;
3051                 struct ctl_lun_list *list;
3052                 const char *name, *value;
3053                 void *cookie;
3054                 int type;
3055
3056                 list = (struct ctl_lun_list *)addr;
3057
3058                 /*
3059                  * Allocate a fixed length sbuf here, based on the length
3060                  * of the user's buffer.  We could allocate an auto-extending
3061                  * buffer, and then tell the user how much larger our
3062                  * amount of data is than his buffer, but that presents
3063                  * some problems:
3064                  *
3065                  * 1.  The sbuf(9) routines use a blocking malloc, and so
3066                  *     we can't hold a lock while calling them with an
3067                  *     auto-extending buffer.
3068                  *
3069                  * 2.  There is not currently a LUN reference counting
3070                  *     mechanism, outside of outstanding transactions on
3071                  *     the LUN's OOA queue.  So a LUN could go away on us
3072                  *     while we're getting the LUN number, backend-specific
3073                  *     information, etc.  Thus, given the way things
3074                  *     currently work, we need to hold the CTL lock while
3075                  *     grabbing LUN information.
3076                  *
3077                  * So, from the user's standpoint, the best thing to do is
3078                  * allocate what he thinks is a reasonable buffer length,
3079                  * and then if he gets a CTL_LUN_LIST_NEED_MORE_SPACE error,
3080                  * double the buffer length and try again.  (And repeat
3081                  * that until he succeeds.)
3082                  */
3083                 sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3084                 if (sb == NULL) {
3085                         list->status = CTL_LUN_LIST_ERROR;
3086                         snprintf(list->error_str, sizeof(list->error_str),
3087                                  "Unable to allocate %d bytes for LUN list",
3088                                  list->alloc_len);
3089                         break;
3090                 }
3091
3092                 sbuf_printf(sb, "<ctllunlist>\n");
3093
3094                 mtx_lock(&softc->ctl_lock);
3095                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
3096                         mtx_lock(&lun->lun_lock);
3097                         retval = sbuf_printf(sb, "<lun id=\"%ju\">\n",
3098                                              (uintmax_t)lun->lun);
3099
3100                         /*
3101                          * Bail out as soon as we see that we've overfilled
3102                          * the buffer.
3103                          */
3104                         if (retval != 0)
3105                                 break;
3106
3107                         retval = sbuf_printf(sb, "\t<backend_type>%s"
3108                                              "</backend_type>\n",
3109                                              (lun->backend == NULL) ?  "none" :
3110                                              lun->backend->name);
3111
3112                         if (retval != 0)
3113                                 break;
3114
3115                         retval = sbuf_printf(sb, "\t<lun_type>%d</lun_type>\n",
3116                                              lun->be_lun->lun_type);
3117
3118                         if (retval != 0)
3119                                 break;
3120
3121                         if (lun->backend == NULL) {
3122                                 retval = sbuf_printf(sb, "</lun>\n");
3123                                 if (retval != 0)
3124                                         break;
3125                                 continue;
3126                         }
3127
3128                         retval = sbuf_printf(sb, "\t<size>%ju</size>\n",
3129                                              (lun->be_lun->maxlba > 0) ?
3130                                              lun->be_lun->maxlba + 1 : 0);
3131
3132                         if (retval != 0)
3133                                 break;
3134
3135                         retval = sbuf_printf(sb, "\t<blocksize>%u</blocksize>\n",
3136                                              lun->be_lun->blocksize);
3137
3138                         if (retval != 0)
3139                                 break;
3140
3141                         retval = sbuf_printf(sb, "\t<serial_number>");
3142
3143                         if (retval != 0)
3144                                 break;
3145
3146                         retval = ctl_sbuf_printf_esc(sb,
3147                             lun->be_lun->serial_num,
3148                             sizeof(lun->be_lun->serial_num));
3149
3150                         if (retval != 0)
3151                                 break;
3152
3153                         retval = sbuf_printf(sb, "</serial_number>\n");
3154
3155                         if (retval != 0)
3156                                 break;
3157
3158                         retval = sbuf_printf(sb, "\t<device_id>");
3159
3160                         if (retval != 0)
3161                                 break;
3162
3163                         retval = ctl_sbuf_printf_esc(sb,
3164                             lun->be_lun->device_id,
3165                             sizeof(lun->be_lun->device_id));
3166
3167                         if (retval != 0)
3168                                 break;
3169
3170                         retval = sbuf_printf(sb, "</device_id>\n");
3171
3172                         if (retval != 0)
3173                                 break;
3174
3175                         if (lun->backend->lun_info != NULL) {
3176                                 retval = lun->backend->lun_info(lun->be_lun, sb);
3177                                 if (retval != 0)
3178                                         break;
3179                         }
3180
3181                         cookie = NULL;
3182                         while ((name = nvlist_next(lun->be_lun->options, &type,
3183                             &cookie)) != NULL) {
3184                                 sbuf_printf(sb, "\t<%s>", name);
3185
3186                                 if (type == NV_TYPE_STRING) {
3187                                         value = dnvlist_get_string(
3188                                             lun->be_lun->options, name, NULL);
3189                                         if (value != NULL)
3190                                                 sbuf_printf(sb, "%s", value);
3191                                 }
3192
3193                                 sbuf_printf(sb, "</%s>\n", name);
3194                         }
3195
3196                         retval = sbuf_printf(sb, "</lun>\n");
3197
3198                         if (retval != 0)
3199                                 break;
3200                         mtx_unlock(&lun->lun_lock);
3201                 }
3202                 if (lun != NULL)
3203                         mtx_unlock(&lun->lun_lock);
3204                 mtx_unlock(&softc->ctl_lock);
3205
3206                 if ((retval != 0)
3207                  || ((retval = sbuf_printf(sb, "</ctllunlist>\n")) != 0)) {
3208                         retval = 0;
3209                         sbuf_delete(sb);
3210                         list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3211                         snprintf(list->error_str, sizeof(list->error_str),
3212                                  "Out of space, %d bytes is too small",
3213                                  list->alloc_len);
3214                         break;
3215                 }
3216
3217                 sbuf_finish(sb);
3218
3219                 retval = copyout(sbuf_data(sb), list->lun_xml,
3220                                  sbuf_len(sb) + 1);
3221
3222                 list->fill_len = sbuf_len(sb) + 1;
3223                 list->status = CTL_LUN_LIST_OK;
3224                 sbuf_delete(sb);
3225                 break;
3226         }
3227         case CTL_ISCSI: {
3228                 struct ctl_iscsi *ci;
3229                 struct ctl_frontend *fe;
3230
3231                 ci = (struct ctl_iscsi *)addr;
3232
3233                 fe = ctl_frontend_find("iscsi");
3234                 if (fe == NULL) {
3235                         ci->status = CTL_ISCSI_ERROR;
3236                         snprintf(ci->error_str, sizeof(ci->error_str),
3237                             "Frontend \"iscsi\" not found.");
3238                         break;
3239                 }
3240
3241                 retval = fe->ioctl(dev, cmd, addr, flag, td);
3242                 break;
3243         }
3244         case CTL_PORT_REQ: {
3245                 struct ctl_req *req;
3246                 struct ctl_frontend *fe;
3247                 void *packed;
3248                 nvlist_t *tmp_args_nvl;
3249                 size_t packed_len;
3250
3251                 req = (struct ctl_req *)addr;
3252                 tmp_args_nvl = req->args_nvl;
3253
3254                 fe = ctl_frontend_find(req->driver);
3255                 if (fe == NULL) {
3256                         req->status = CTL_LUN_ERROR;
3257                         snprintf(req->error_str, sizeof(req->error_str),
3258                             "Frontend \"%s\" not found.", req->driver);
3259                         break;
3260                 }
3261
3262                 if (req->args != NULL) {
3263                         if (req->args_len > CTL_MAX_ARGS_LEN) {
3264                                 req->status = CTL_LUN_ERROR;
3265                                 snprintf(req->error_str, sizeof(req->error_str),
3266                                     "Too big args.");
3267                                 break;
3268                         }
3269                         packed = malloc(req->args_len, M_CTL, M_WAITOK);
3270                         if (copyin(req->args, packed, req->args_len) != 0) {
3271                                 free(packed, M_CTL);
3272                                 req->status = CTL_LUN_ERROR;
3273                                 snprintf(req->error_str, sizeof(req->error_str),
3274                                     "Cannot copyin args.");
3275                                 break;
3276                         }
3277                         req->args_nvl = nvlist_unpack(packed,
3278                             req->args_len, 0);
3279                         free(packed, M_CTL);
3280
3281                         if (req->args_nvl == NULL) {
3282                                 req->status = CTL_LUN_ERROR;
3283                                 snprintf(req->error_str, sizeof(req->error_str),
3284                                     "Cannot unpack args nvlist.");
3285                                 break;
3286                         }
3287                 } else
3288                         req->args_nvl = nvlist_create(0);
3289
3290                 req->result_nvl = NULL;
3291                 if (fe->ioctl)
3292                         retval = fe->ioctl(dev, cmd, addr, flag, td);
3293                 else
3294                         retval = ENODEV;
3295
3296                 nvlist_destroy(req->args_nvl);
3297                 req->args_nvl = tmp_args_nvl;
3298
3299                 if (req->result_nvl != NULL) {
3300                         if (req->result != NULL) {
3301                                 packed = nvlist_pack(req->result_nvl,
3302                                     &packed_len);
3303                                 if (packed == NULL) {
3304                                         req->status = CTL_LUN_ERROR;
3305                                         snprintf(req->error_str,
3306                                             sizeof(req->error_str),
3307                                             "Cannot pack result nvlist.");
3308                                         break;
3309                                 }
3310
3311                                 if (packed_len > req->result_len) {
3312                                         req->status = CTL_LUN_ERROR;
3313                                         snprintf(req->error_str,
3314                                             sizeof(req->error_str),
3315                                             "Result nvlist too large.");
3316                                         free(packed, M_NVLIST);
3317                                         break;
3318                                 }
3319
3320                                 if (copyout(packed, req->result, packed_len)) {
3321                                         req->status = CTL_LUN_ERROR;
3322                                         snprintf(req->error_str,
3323                                             sizeof(req->error_str),
3324                                             "Cannot copyout() the result.");
3325                                         free(packed, M_NVLIST);
3326                                         break;
3327                                 }
3328
3329                                 req->result_len = packed_len;
3330                                 free(packed, M_NVLIST);
3331                         }
3332
3333                         nvlist_destroy(req->result_nvl);
3334                 }
3335                 break;
3336         }
3337         case CTL_PORT_LIST: {
3338                 struct sbuf *sb;
3339                 struct ctl_port *port;
3340                 struct ctl_lun_list *list;
3341                 const char *name, *value;
3342                 void *cookie;
3343                 int j, type;
3344                 uint32_t plun;
3345
3346                 list = (struct ctl_lun_list *)addr;
3347
3348                 sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3349                 if (sb == NULL) {
3350                         list->status = CTL_LUN_LIST_ERROR;
3351                         snprintf(list->error_str, sizeof(list->error_str),
3352                                  "Unable to allocate %d bytes for LUN list",
3353                                  list->alloc_len);
3354                         break;
3355                 }
3356
3357                 sbuf_printf(sb, "<ctlportlist>\n");
3358
3359                 mtx_lock(&softc->ctl_lock);
3360                 STAILQ_FOREACH(port, &softc->port_list, links) {
3361                         retval = sbuf_printf(sb, "<targ_port id=\"%ju\">\n",
3362                                              (uintmax_t)port->targ_port);
3363
3364                         /*
3365                          * Bail out as soon as we see that we've overfilled
3366                          * the buffer.
3367                          */
3368                         if (retval != 0)
3369                                 break;
3370
3371                         retval = sbuf_printf(sb, "\t<frontend_type>%s"
3372                             "</frontend_type>\n", port->frontend->name);
3373                         if (retval != 0)
3374                                 break;
3375
3376                         retval = sbuf_printf(sb, "\t<port_type>%d</port_type>\n",
3377                                              port->port_type);
3378                         if (retval != 0)
3379                                 break;
3380
3381                         retval = sbuf_printf(sb, "\t<online>%s</online>\n",
3382                             (port->status & CTL_PORT_STATUS_ONLINE) ? "YES" : "NO");
3383                         if (retval != 0)
3384                                 break;
3385
3386                         retval = sbuf_printf(sb, "\t<port_name>%s</port_name>\n",
3387                             port->port_name);
3388                         if (retval != 0)
3389                                 break;
3390
3391                         retval = sbuf_printf(sb, "\t<physical_port>%d</physical_port>\n",
3392                             port->physical_port);
3393                         if (retval != 0)
3394                                 break;
3395
3396                         retval = sbuf_printf(sb, "\t<virtual_port>%d</virtual_port>\n",
3397                             port->virtual_port);
3398                         if (retval != 0)
3399                                 break;
3400
3401                         if (port->target_devid != NULL) {
3402                                 sbuf_printf(sb, "\t<target>");
3403                                 ctl_id_sbuf(port->target_devid, sb);
3404                                 sbuf_printf(sb, "</target>\n");
3405                         }
3406
3407                         if (port->port_devid != NULL) {
3408                                 sbuf_printf(sb, "\t<port>");
3409                                 ctl_id_sbuf(port->port_devid, sb);
3410                                 sbuf_printf(sb, "</port>\n");
3411                         }
3412
3413                         if (port->port_info != NULL) {
3414                                 retval = port->port_info(port->onoff_arg, sb);
3415                                 if (retval != 0)
3416                                         break;
3417                         }
3418
3419                         cookie = NULL;
3420                         while ((name = nvlist_next(port->options, &type,
3421                             &cookie)) != NULL) {
3422                                 sbuf_printf(sb, "\t<%s>", name);
3423
3424                                 if (type == NV_TYPE_STRING) {
3425                                         value = dnvlist_get_string(port->options,
3426                                             name, NULL);
3427                                         if (value != NULL)
3428                                                 sbuf_printf(sb, "%s", value);
3429                                 }
3430
3431                                 sbuf_printf(sb, "</%s>\n", name);
3432                         }
3433
3434                         if (port->lun_map != NULL) {
3435                                 sbuf_printf(sb, "\t<lun_map>on</lun_map>\n");
3436                                 for (j = 0; j < port->lun_map_size; j++) {
3437                                         plun = ctl_lun_map_from_port(port, j);
3438                                         if (plun == UINT32_MAX)
3439                                                 continue;
3440                                         sbuf_printf(sb,
3441                                             "\t<lun id=\"%u\">%u</lun>\n",
3442                                             j, plun);
3443                                 }
3444                         }
3445
3446                         for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
3447                                 if (port->wwpn_iid[j].in_use == 0 ||
3448                                     (port->wwpn_iid[j].wwpn == 0 &&
3449                                      port->wwpn_iid[j].name == NULL))
3450                                         continue;
3451
3452                                 if (port->wwpn_iid[j].name != NULL)
3453                                         retval = sbuf_printf(sb,
3454                                             "\t<initiator id=\"%u\">%s</initiator>\n",
3455                                             j, port->wwpn_iid[j].name);
3456                                 else
3457                                         retval = sbuf_printf(sb,
3458                                             "\t<initiator id=\"%u\">naa.%08jx</initiator>\n",
3459                                             j, port->wwpn_iid[j].wwpn);
3460                                 if (retval != 0)
3461                                         break;
3462                         }
3463                         if (retval != 0)
3464                                 break;
3465
3466                         retval = sbuf_printf(sb, "</targ_port>\n");
3467                         if (retval != 0)
3468                                 break;
3469                 }
3470                 mtx_unlock(&softc->ctl_lock);
3471
3472                 if ((retval != 0)
3473                  || ((retval = sbuf_printf(sb, "</ctlportlist>\n")) != 0)) {
3474                         retval = 0;
3475                         sbuf_delete(sb);
3476                         list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3477                         snprintf(list->error_str, sizeof(list->error_str),
3478                                  "Out of space, %d bytes is too small",
3479                                  list->alloc_len);
3480                         break;
3481                 }
3482
3483                 sbuf_finish(sb);
3484
3485                 retval = copyout(sbuf_data(sb), list->lun_xml,
3486                                  sbuf_len(sb) + 1);
3487
3488                 list->fill_len = sbuf_len(sb) + 1;
3489                 list->status = CTL_LUN_LIST_OK;
3490                 sbuf_delete(sb);
3491                 break;
3492         }
3493         case CTL_LUN_MAP: {
3494                 struct ctl_lun_map *lm  = (struct ctl_lun_map *)addr;
3495                 struct ctl_port *port;
3496
3497                 mtx_lock(&softc->ctl_lock);
3498                 if (lm->port < softc->port_min ||
3499                     lm->port >= softc->port_max ||
3500                     (port = softc->ctl_ports[lm->port]) == NULL) {
3501                         mtx_unlock(&softc->ctl_lock);
3502                         return (ENXIO);
3503                 }
3504                 if (port->status & CTL_PORT_STATUS_ONLINE) {
3505                         STAILQ_FOREACH(lun, &softc->lun_list, links) {
3506                                 if (ctl_lun_map_to_port(port, lun->lun) ==
3507                                     UINT32_MAX)
3508                                         continue;
3509                                 mtx_lock(&lun->lun_lock);
3510                                 ctl_est_ua_port(lun, lm->port, -1,
3511                                     CTL_UA_LUN_CHANGE);
3512                                 mtx_unlock(&lun->lun_lock);
3513                         }
3514                 }
3515                 mtx_unlock(&softc->ctl_lock); // XXX: port_enable sleeps
3516                 if (lm->plun != UINT32_MAX) {
3517                         if (lm->lun == UINT32_MAX)
3518                                 retval = ctl_lun_map_unset(port, lm->plun);
3519                         else if (lm->lun < ctl_max_luns &&
3520                             softc->ctl_luns[lm->lun] != NULL)
3521                                 retval = ctl_lun_map_set(port, lm->plun, lm->lun);
3522                         else
3523                                 return (ENXIO);
3524                 } else {
3525                         if (lm->lun == UINT32_MAX)
3526                                 retval = ctl_lun_map_deinit(port);
3527                         else
3528                                 retval = ctl_lun_map_init(port);
3529                 }
3530                 if (port->status & CTL_PORT_STATUS_ONLINE)
3531                         ctl_isc_announce_port(port);
3532                 break;
3533         }
3534         case CTL_GET_LUN_STATS: {
3535                 struct ctl_get_io_stats *stats = (struct ctl_get_io_stats *)addr;
3536                 int i;
3537
3538                 /*
3539                  * XXX KDM no locking here.  If the LUN list changes,
3540                  * things can blow up.
3541                  */
3542                 i = 0;
3543                 stats->status = CTL_SS_OK;
3544                 stats->fill_len = 0;
3545                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
3546                         if (lun->lun < stats->first_item)
3547                                 continue;
3548                         if (stats->fill_len + sizeof(lun->stats) >
3549                             stats->alloc_len) {
3550                                 stats->status = CTL_SS_NEED_MORE_SPACE;
3551                                 break;
3552                         }
3553                         retval = copyout(&lun->stats, &stats->stats[i++],
3554                                          sizeof(lun->stats));
3555                         if (retval != 0)
3556                                 break;
3557                         stats->fill_len += sizeof(lun->stats);
3558                 }
3559                 stats->num_items = softc->num_luns;
3560                 stats->flags = CTL_STATS_FLAG_NONE;
3561 #ifdef CTL_TIME_IO
3562                 stats->flags |= CTL_STATS_FLAG_TIME_VALID;
3563 #endif
3564                 getnanouptime(&stats->timestamp);
3565                 break;
3566         }
3567         case CTL_GET_PORT_STATS: {
3568                 struct ctl_get_io_stats *stats = (struct ctl_get_io_stats *)addr;
3569                 int i;
3570
3571                 /*
3572                  * XXX KDM no locking here.  If the LUN list changes,
3573                  * things can blow up.
3574                  */
3575                 i = 0;
3576                 stats->status = CTL_SS_OK;
3577                 stats->fill_len = 0;
3578                 STAILQ_FOREACH(port, &softc->port_list, links) {
3579                         if (port->targ_port < stats->first_item)
3580                                 continue;
3581                         if (stats->fill_len + sizeof(port->stats) >
3582                             stats->alloc_len) {
3583                                 stats->status = CTL_SS_NEED_MORE_SPACE;
3584                                 break;
3585                         }
3586                         retval = copyout(&port->stats, &stats->stats[i++],
3587                                          sizeof(port->stats));
3588                         if (retval != 0)
3589                                 break;
3590                         stats->fill_len += sizeof(port->stats);
3591                 }
3592                 stats->num_items = softc->num_ports;
3593                 stats->flags = CTL_STATS_FLAG_NONE;
3594 #ifdef CTL_TIME_IO
3595                 stats->flags |= CTL_STATS_FLAG_TIME_VALID;
3596 #endif
3597                 getnanouptime(&stats->timestamp);
3598                 break;
3599         }
3600         default: {
3601                 /* XXX KDM should we fix this? */
3602 #if 0
3603                 struct ctl_backend_driver *backend;
3604                 unsigned int type;
3605                 int found;
3606
3607                 found = 0;
3608
3609                 /*
3610                  * We encode the backend type as the ioctl type for backend
3611                  * ioctls.  So parse it out here, and then search for a
3612                  * backend of this type.
3613                  */
3614                 type = _IOC_TYPE(cmd);
3615
3616                 STAILQ_FOREACH(backend, &softc->be_list, links) {
3617                         if (backend->type == type) {
3618                                 found = 1;
3619                                 break;
3620                         }
3621                 }
3622                 if (found == 0) {
3623                         printf("ctl: unknown ioctl command %#lx or backend "
3624                                "%d\n", cmd, type);
3625                         retval = EINVAL;
3626                         break;
3627                 }
3628                 retval = backend->ioctl(dev, cmd, addr, flag, td);
3629 #endif
3630                 retval = ENOTTY;
3631                 break;
3632         }
3633         }
3634         return (retval);
3635 }
3636
3637 uint32_t
3638 ctl_get_initindex(struct ctl_nexus *nexus)
3639 {
3640         return (nexus->initid + (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3641 }
3642
3643 int
3644 ctl_lun_map_init(struct ctl_port *port)
3645 {
3646         struct ctl_softc *softc = port->ctl_softc;
3647         struct ctl_lun *lun;
3648         int size = ctl_lun_map_size;
3649         uint32_t i;
3650
3651         if (port->lun_map == NULL || port->lun_map_size < size) {
3652                 port->lun_map_size = 0;
3653                 free(port->lun_map, M_CTL);
3654                 port->lun_map = malloc(size * sizeof(uint32_t),
3655                     M_CTL, M_NOWAIT);
3656         }
3657         if (port->lun_map == NULL)
3658                 return (ENOMEM);
3659         for (i = 0; i < size; i++)
3660                 port->lun_map[i] = UINT32_MAX;
3661         port->lun_map_size = size;
3662         if (port->status & CTL_PORT_STATUS_ONLINE) {
3663                 if (port->lun_disable != NULL) {
3664                         STAILQ_FOREACH(lun, &softc->lun_list, links)
3665                                 port->lun_disable(port->targ_lun_arg, lun->lun);
3666                 }
3667                 ctl_isc_announce_port(port);
3668         }
3669         return (0);
3670 }
3671
3672 int
3673 ctl_lun_map_deinit(struct ctl_port *port)
3674 {
3675         struct ctl_softc *softc = port->ctl_softc;
3676         struct ctl_lun *lun;
3677
3678         if (port->lun_map == NULL)
3679                 return (0);
3680         port->lun_map_size = 0;
3681         free(port->lun_map, M_CTL);
3682         port->lun_map = NULL;
3683         if (port->status & CTL_PORT_STATUS_ONLINE) {
3684                 if (port->lun_enable != NULL) {
3685                         STAILQ_FOREACH(lun, &softc->lun_list, links)
3686                                 port->lun_enable(port->targ_lun_arg, lun->lun);
3687                 }
3688                 ctl_isc_announce_port(port);
3689         }
3690         return (0);
3691 }
3692
3693 int
3694 ctl_lun_map_set(struct ctl_port *port, uint32_t plun, uint32_t glun)
3695 {
3696         int status;
3697         uint32_t old;
3698
3699         if (port->lun_map == NULL) {
3700                 status = ctl_lun_map_init(port);
3701                 if (status != 0)
3702                         return (status);
3703         }
3704         if (plun >= port->lun_map_size)
3705                 return (EINVAL);
3706         old = port->lun_map[plun];
3707         port->lun_map[plun] = glun;
3708         if ((port->status & CTL_PORT_STATUS_ONLINE) && old == UINT32_MAX) {
3709                 if (port->lun_enable != NULL)
3710                         port->lun_enable(port->targ_lun_arg, plun);
3711                 ctl_isc_announce_port(port);
3712         }
3713         return (0);
3714 }
3715
3716 int
3717 ctl_lun_map_unset(struct ctl_port *port, uint32_t plun)
3718 {
3719         uint32_t old;
3720
3721         if (port->lun_map == NULL || plun >= port->lun_map_size)
3722                 return (0);
3723         old = port->lun_map[plun];
3724         port->lun_map[plun] = UINT32_MAX;
3725         if ((port->status & CTL_PORT_STATUS_ONLINE) && old != UINT32_MAX) {
3726                 if (port->lun_disable != NULL)
3727                         port->lun_disable(port->targ_lun_arg, plun);
3728                 ctl_isc_announce_port(port);
3729         }
3730         return (0);
3731 }
3732
3733 uint32_t
3734 ctl_lun_map_from_port(struct ctl_port *port, uint32_t lun_id)
3735 {
3736
3737         if (port == NULL)
3738                 return (UINT32_MAX);
3739         if (port->lun_map == NULL)
3740                 return (lun_id);
3741         if (lun_id > port->lun_map_size)
3742                 return (UINT32_MAX);
3743         return (port->lun_map[lun_id]);
3744 }
3745
3746 uint32_t
3747 ctl_lun_map_to_port(struct ctl_port *port, uint32_t lun_id)
3748 {
3749         uint32_t i;
3750
3751         if (port == NULL)
3752                 return (UINT32_MAX);
3753         if (port->lun_map == NULL)
3754                 return (lun_id);
3755         for (i = 0; i < port->lun_map_size; i++) {
3756                 if (port->lun_map[i] == lun_id)
3757                         return (i);
3758         }
3759         return (UINT32_MAX);
3760 }
3761
3762 uint32_t
3763 ctl_decode_lun(uint64_t encoded)
3764 {
3765         uint8_t lun[8];
3766         uint32_t result = 0xffffffff;
3767
3768         be64enc(lun, encoded);
3769         switch (lun[0] & RPL_LUNDATA_ATYP_MASK) {
3770         case RPL_LUNDATA_ATYP_PERIPH:
3771                 if ((lun[0] & 0x3f) == 0 && lun[2] == 0 && lun[3] == 0 &&
3772                     lun[4] == 0 && lun[5] == 0 && lun[6] == 0 && lun[7] == 0)
3773                         result = lun[1];
3774                 break;
3775         case RPL_LUNDATA_ATYP_FLAT:
3776                 if (lun[2] == 0 && lun[3] == 0 && lun[4] == 0 && lun[5] == 0 &&
3777                     lun[6] == 0 && lun[7] == 0)
3778                         result = ((lun[0] & 0x3f) << 8) + lun[1];
3779                 break;
3780         case RPL_LUNDATA_ATYP_EXTLUN:
3781                 switch (lun[0] & RPL_LUNDATA_EXT_EAM_MASK) {
3782                 case 0x02:
3783                         switch (lun[0] & RPL_LUNDATA_EXT_LEN_MASK) {
3784                         case 0x00:
3785                                 result = lun[1];
3786                                 break;
3787                         case 0x10:
3788                                 result = (lun[1] << 16) + (lun[2] << 8) +
3789                                     lun[3];
3790                                 break;
3791                         case 0x20:
3792                                 if (lun[1] == 0 && lun[6] == 0 && lun[7] == 0)
3793                                         result = (lun[2] << 24) +
3794                                             (lun[3] << 16) + (lun[4] << 8) +
3795                                             lun[5];
3796                                 break;
3797                         }
3798                         break;
3799                 case RPL_LUNDATA_EXT_EAM_NOT_SPEC:
3800                         result = 0xffffffff;
3801                         break;
3802                 }
3803                 break;
3804         }
3805         return (result);
3806 }
3807
3808 uint64_t
3809 ctl_encode_lun(uint32_t decoded)
3810 {
3811         uint64_t l = decoded;
3812
3813         if (l <= 0xff)
3814                 return (((uint64_t)RPL_LUNDATA_ATYP_PERIPH << 56) | (l << 48));
3815         if (l <= 0x3fff)
3816                 return (((uint64_t)RPL_LUNDATA_ATYP_FLAT << 56) | (l << 48));
3817         if (l <= 0xffffff)
3818                 return (((uint64_t)(RPL_LUNDATA_ATYP_EXTLUN | 0x12) << 56) |
3819                     (l << 32));
3820         return ((((uint64_t)RPL_LUNDATA_ATYP_EXTLUN | 0x22) << 56) | (l << 16));
3821 }
3822
3823 int
3824 ctl_ffz(uint32_t *mask, uint32_t first, uint32_t last)
3825 {
3826         int i;
3827
3828         for (i = first; i < last; i++) {
3829                 if ((mask[i / 32] & (1 << (i % 32))) == 0)
3830                         return (i);
3831         }
3832         return (-1);
3833 }
3834
3835 int
3836 ctl_set_mask(uint32_t *mask, uint32_t bit)
3837 {
3838         uint32_t chunk, piece;
3839
3840         chunk = bit >> 5;
3841         piece = bit % (sizeof(uint32_t) * 8);
3842
3843         if ((mask[chunk] & (1 << piece)) != 0)
3844                 return (-1);
3845         else
3846                 mask[chunk] |= (1 << piece);
3847
3848         return (0);
3849 }
3850
3851 int
3852 ctl_clear_mask(uint32_t *mask, uint32_t bit)
3853 {
3854         uint32_t chunk, piece;
3855
3856         chunk = bit >> 5;
3857         piece = bit % (sizeof(uint32_t) * 8);
3858
3859         if ((mask[chunk] & (1 << piece)) == 0)
3860                 return (-1);
3861         else
3862                 mask[chunk] &= ~(1 << piece);
3863
3864         return (0);
3865 }
3866
3867 int
3868 ctl_is_set(uint32_t *mask, uint32_t bit)
3869 {
3870         uint32_t chunk, piece;
3871
3872         chunk = bit >> 5;
3873         piece = bit % (sizeof(uint32_t) * 8);
3874
3875         if ((mask[chunk] & (1 << piece)) == 0)
3876                 return (0);
3877         else
3878                 return (1);
3879 }
3880
3881 static uint64_t
3882 ctl_get_prkey(struct ctl_lun *lun, uint32_t residx)
3883 {
3884         uint64_t *t;
3885
3886         t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3887         if (t == NULL)
3888                 return (0);
3889         return (t[residx % CTL_MAX_INIT_PER_PORT]);
3890 }
3891
3892 static void
3893 ctl_clr_prkey(struct ctl_lun *lun, uint32_t residx)
3894 {
3895         uint64_t *t;
3896
3897         t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3898         if (t == NULL)
3899                 return;
3900         t[residx % CTL_MAX_INIT_PER_PORT] = 0;
3901 }
3902
3903 static void
3904 ctl_alloc_prkey(struct ctl_lun *lun, uint32_t residx)
3905 {
3906         uint64_t *p;
3907         u_int i;
3908
3909         i = residx/CTL_MAX_INIT_PER_PORT;
3910         if (lun->pr_keys[i] != NULL)
3911                 return;
3912         mtx_unlock(&lun->lun_lock);
3913         p = malloc(sizeof(uint64_t) * CTL_MAX_INIT_PER_PORT, M_CTL,
3914             M_WAITOK | M_ZERO);
3915         mtx_lock(&lun->lun_lock);
3916         if (lun->pr_keys[i] == NULL)
3917                 lun->pr_keys[i] = p;
3918         else
3919                 free(p, M_CTL);
3920 }
3921
3922 static void
3923 ctl_set_prkey(struct ctl_lun *lun, uint32_t residx, uint64_t key)
3924 {
3925         uint64_t *t;
3926
3927         t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3928         KASSERT(t != NULL, ("prkey %d is not allocated", residx));
3929         t[residx % CTL_MAX_INIT_PER_PORT] = key;
3930 }
3931
3932 /*
3933  * ctl_softc, pool_name, total_ctl_io are passed in.
3934  * npool is passed out.
3935  */
3936 int
3937 ctl_pool_create(struct ctl_softc *ctl_softc, const char *pool_name,
3938                 uint32_t total_ctl_io, void **npool)
3939 {
3940         struct ctl_io_pool *pool;
3941
3942         pool = (struct ctl_io_pool *)malloc(sizeof(*pool), M_CTL,
3943                                             M_NOWAIT | M_ZERO);
3944         if (pool == NULL)
3945                 return (ENOMEM);
3946
3947         snprintf(pool->name, sizeof(pool->name), "CTL IO %s", pool_name);
3948         pool->ctl_softc = ctl_softc;
3949 #ifdef IO_POOLS
3950         pool->zone = uma_zsecond_create(pool->name, NULL,
3951             NULL, NULL, NULL, ctl_softc->io_zone);
3952         /* uma_prealloc(pool->zone, total_ctl_io); */
3953 #else
3954         pool->zone = ctl_softc->io_zone;
3955 #endif
3956
3957         *npool = pool;
3958         return (0);
3959 }
3960
3961 void
3962 ctl_pool_free(struct ctl_io_pool *pool)
3963 {
3964
3965         if (pool == NULL)
3966                 return;
3967
3968 #ifdef IO_POOLS
3969         uma_zdestroy(pool->zone);
3970 #endif
3971         free(pool, M_CTL);
3972 }
3973
3974 union ctl_io *
3975 ctl_alloc_io(void *pool_ref)
3976 {
3977         struct ctl_io_pool *pool = (struct ctl_io_pool *)pool_ref;
3978         union ctl_io *io;
3979
3980         io = uma_zalloc(pool->zone, M_WAITOK);
3981         if (io != NULL) {
3982                 io->io_hdr.pool = pool_ref;
3983                 CTL_SOFTC(io) = pool->ctl_softc;
3984                 TAILQ_INIT(&io->io_hdr.blocked_queue);
3985         }
3986         return (io);
3987 }
3988
3989 union ctl_io *
3990 ctl_alloc_io_nowait(void *pool_ref)
3991 {
3992         struct ctl_io_pool *pool = (struct ctl_io_pool *)pool_ref;
3993         union ctl_io *io;
3994
3995         io = uma_zalloc(pool->zone, M_NOWAIT);
3996         if (io != NULL) {
3997                 io->io_hdr.pool = pool_ref;
3998                 CTL_SOFTC(io) = pool->ctl_softc;
3999                 TAILQ_INIT(&io->io_hdr.blocked_queue);
4000         }
4001         return (io);
4002 }
4003
4004 void
4005 ctl_free_io(union ctl_io *io)
4006 {
4007         struct ctl_io_pool *pool;
4008
4009         if (io == NULL)
4010                 return;
4011
4012         pool = (struct ctl_io_pool *)io->io_hdr.pool;
4013         uma_zfree(pool->zone, io);
4014 }
4015
4016 void
4017 ctl_zero_io(union ctl_io *io)
4018 {
4019         struct ctl_io_pool *pool;
4020
4021         if (io == NULL)
4022                 return;
4023
4024         /*
4025          * May need to preserve linked list pointers at some point too.
4026          */
4027         pool = io->io_hdr.pool;
4028         memset(io, 0, sizeof(*io));
4029         io->io_hdr.pool = pool;
4030         CTL_SOFTC(io) = pool->ctl_softc;
4031         TAILQ_INIT(&io->io_hdr.blocked_queue);
4032 }
4033
4034 int
4035 ctl_expand_number(const char *buf, uint64_t *num)
4036 {
4037         char *endptr;
4038         uint64_t number;
4039         unsigned shift;
4040
4041         number = strtoq(buf, &endptr, 0);
4042
4043         switch (tolower((unsigned char)*endptr)) {
4044         case 'e':
4045                 shift = 60;
4046                 break;
4047         case 'p':
4048                 shift = 50;
4049                 break;
4050         case 't':
4051                 shift = 40;
4052                 break;
4053         case 'g':
4054                 shift = 30;
4055                 break;
4056         case 'm':
4057                 shift = 20;
4058                 break;
4059         case 'k':
4060                 shift = 10;
4061                 break;
4062         case 'b':
4063         case '\0': /* No unit. */
4064                 *num = number;
4065                 return (0);
4066         default:
4067                 /* Unrecognized unit. */
4068                 return (-1);
4069         }
4070
4071         if ((number << shift) >> shift != number) {
4072                 /* Overflow */
4073                 return (-1);
4074         }
4075         *num = number << shift;
4076         return (0);
4077 }
4078
4079 /*
4080  * This routine could be used in the future to load default and/or saved
4081  * mode page parameters for a particuar lun.
4082  */
4083 static int
4084 ctl_init_page_index(struct ctl_lun *lun)
4085 {
4086         int i, page_code;
4087         struct ctl_page_index *page_index;
4088         const char *value;
4089         uint64_t ival;
4090
4091         memcpy(&lun->mode_pages.index, page_index_template,
4092                sizeof(page_index_template));
4093
4094         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
4095                 page_index = &lun->mode_pages.index[i];
4096                 if (lun->be_lun->lun_type == T_DIRECT &&
4097                     (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
4098                         continue;
4099                 if (lun->be_lun->lun_type == T_PROCESSOR &&
4100                     (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
4101                         continue;
4102                 if (lun->be_lun->lun_type == T_CDROM &&
4103                     (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
4104                         continue;
4105
4106                 page_code = page_index->page_code & SMPH_PC_MASK;
4107                 switch (page_code) {
4108                 case SMS_RW_ERROR_RECOVERY_PAGE: {
4109                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4110                             ("subpage %#x for page %#x is incorrect!",
4111                             page_index->subpage, page_code));
4112                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_CURRENT],
4113                                &rw_er_page_default,
4114                                sizeof(rw_er_page_default));
4115                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_CHANGEABLE],
4116                                &rw_er_page_changeable,
4117                                sizeof(rw_er_page_changeable));
4118                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_DEFAULT],
4119                                &rw_er_page_default,
4120                                sizeof(rw_er_page_default));
4121                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_SAVED],
4122                                &rw_er_page_default,
4123                                sizeof(rw_er_page_default));
4124                         page_index->page_data =
4125                                 (uint8_t *)lun->mode_pages.rw_er_page;
4126                         break;
4127                 }
4128                 case SMS_FORMAT_DEVICE_PAGE: {
4129                         struct scsi_format_page *format_page;
4130
4131                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4132                             ("subpage %#x for page %#x is incorrect!",
4133                             page_index->subpage, page_code));
4134
4135                         /*
4136                          * Sectors per track are set above.  Bytes per
4137                          * sector need to be set here on a per-LUN basis.
4138                          */
4139                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_CURRENT],
4140                                &format_page_default,
4141                                sizeof(format_page_default));
4142                         memcpy(&lun->mode_pages.format_page[
4143                                CTL_PAGE_CHANGEABLE], &format_page_changeable,
4144                                sizeof(format_page_changeable));
4145                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_DEFAULT],
4146                                &format_page_default,
4147                                sizeof(format_page_default));
4148                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_SAVED],
4149                                &format_page_default,
4150                                sizeof(format_page_default));
4151
4152                         format_page = &lun->mode_pages.format_page[
4153                                 CTL_PAGE_CURRENT];
4154                         scsi_ulto2b(lun->be_lun->blocksize,
4155                                     format_page->bytes_per_sector);
4156
4157                         format_page = &lun->mode_pages.format_page[
4158                                 CTL_PAGE_DEFAULT];
4159                         scsi_ulto2b(lun->be_lun->blocksize,
4160                                     format_page->bytes_per_sector);
4161
4162                         format_page = &lun->mode_pages.format_page[
4163                                 CTL_PAGE_SAVED];
4164                         scsi_ulto2b(lun->be_lun->blocksize,
4165                                     format_page->bytes_per_sector);
4166
4167                         page_index->page_data =
4168                                 (uint8_t *)lun->mode_pages.format_page;
4169                         break;
4170                 }
4171                 case SMS_RIGID_DISK_PAGE: {
4172                         struct scsi_rigid_disk_page *rigid_disk_page;
4173                         uint32_t sectors_per_cylinder;
4174                         uint64_t cylinders;
4175 #ifndef __XSCALE__
4176                         int shift;
4177 #endif /* !__XSCALE__ */
4178
4179                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4180                             ("subpage %#x for page %#x is incorrect!",
4181                             page_index->subpage, page_code));
4182
4183                         /*
4184                          * Rotation rate and sectors per track are set
4185                          * above.  We calculate the cylinders here based on
4186                          * capacity.  Due to the number of heads and
4187                          * sectors per track we're using, smaller arrays
4188                          * may turn out to have 0 cylinders.  Linux and
4189                          * FreeBSD don't pay attention to these mode pages
4190                          * to figure out capacity, but Solaris does.  It
4191                          * seems to deal with 0 cylinders just fine, and
4192                          * works out a fake geometry based on the capacity.
4193                          */
4194                         memcpy(&lun->mode_pages.rigid_disk_page[
4195                                CTL_PAGE_DEFAULT], &rigid_disk_page_default,
4196                                sizeof(rigid_disk_page_default));
4197                         memcpy(&lun->mode_pages.rigid_disk_page[
4198                                CTL_PAGE_CHANGEABLE],&rigid_disk_page_changeable,
4199                                sizeof(rigid_disk_page_changeable));
4200
4201                         sectors_per_cylinder = CTL_DEFAULT_SECTORS_PER_TRACK *
4202                                 CTL_DEFAULT_HEADS;
4203
4204                         /*
4205                          * The divide method here will be more accurate,
4206                          * probably, but results in floating point being
4207                          * used in the kernel on i386 (__udivdi3()).  On the
4208                          * XScale, though, __udivdi3() is implemented in
4209                          * software.
4210                          *
4211                          * The shift method for cylinder calculation is
4212                          * accurate if sectors_per_cylinder is a power of
4213                          * 2.  Otherwise it might be slightly off -- you
4214                          * might have a bit of a truncation problem.
4215                          */
4216 #ifdef  __XSCALE__
4217                         cylinders = (lun->be_lun->maxlba + 1) /
4218                                 sectors_per_cylinder;
4219 #else
4220                         for (shift = 31; shift > 0; shift--) {
4221                                 if (sectors_per_cylinder & (1 << shift))
4222                                         break;
4223                         }
4224                         cylinders = (lun->be_lun->maxlba + 1) >> shift;
4225 #endif
4226
4227                         /*
4228                          * We've basically got 3 bytes, or 24 bits for the
4229                          * cylinder size in the mode page.  If we're over,
4230                          * just round down to 2^24.
4231                          */
4232                         if (cylinders > 0xffffff)
4233                                 cylinders = 0xffffff;
4234
4235                         rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4236                                 CTL_PAGE_DEFAULT];
4237                         scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4238
4239                         if ((value = dnvlist_get_string(lun->be_lun->options,
4240                             "rpm", NULL)) != NULL) {
4241                                 scsi_ulto2b(strtol(value, NULL, 0),
4242                                      rigid_disk_page->rotation_rate);
4243                         }
4244
4245                         memcpy(&lun->mode_pages.rigid_disk_page[CTL_PAGE_CURRENT],
4246                                &lun->mode_pages.rigid_disk_page[CTL_PAGE_DEFAULT],
4247                                sizeof(rigid_disk_page_default));
4248                         memcpy(&lun->mode_pages.rigid_disk_page[CTL_PAGE_SAVED],
4249                                &lun->mode_pages.rigid_disk_page[CTL_PAGE_DEFAULT],
4250                                sizeof(rigid_disk_page_default));
4251
4252                         page_index->page_data =
4253                                 (uint8_t *)lun->mode_pages.rigid_disk_page;
4254                         break;
4255                 }
4256                 case SMS_VERIFY_ERROR_RECOVERY_PAGE: {
4257                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4258                             ("subpage %#x for page %#x is incorrect!",
4259                             page_index->subpage, page_code));
4260                         memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_CURRENT],
4261                                &verify_er_page_default,
4262                                sizeof(verify_er_page_default));
4263                         memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_CHANGEABLE],
4264                                &verify_er_page_changeable,
4265                                sizeof(verify_er_page_changeable));
4266                         memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_DEFAULT],
4267                                &verify_er_page_default,
4268                                sizeof(verify_er_page_default));
4269                         memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_SAVED],
4270                                &verify_er_page_default,
4271                                sizeof(verify_er_page_default));
4272                         page_index->page_data =
4273                                 (uint8_t *)lun->mode_pages.verify_er_page;
4274                         break;
4275                 }
4276                 case SMS_CACHING_PAGE: {
4277                         struct scsi_caching_page *caching_page;
4278
4279                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4280                             ("subpage %#x for page %#x is incorrect!",
4281                             page_index->subpage, page_code));
4282                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_DEFAULT],
4283                                &caching_page_default,
4284                                sizeof(caching_page_default));
4285                         memcpy(&lun->mode_pages.caching_page[
4286                                CTL_PAGE_CHANGEABLE], &caching_page_changeable,
4287                                sizeof(caching_page_changeable));
4288                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4289                                &caching_page_default,
4290                                sizeof(caching_page_default));
4291                         caching_page = &lun->mode_pages.caching_page[
4292                             CTL_PAGE_SAVED];
4293                         value = dnvlist_get_string(lun->be_lun->options,
4294                             "writecache", NULL);
4295                         if (value != NULL && strcmp(value, "off") == 0)
4296                                 caching_page->flags1 &= ~SCP_WCE;
4297                         value = dnvlist_get_string(lun->be_lun->options,
4298                             "readcache", NULL);
4299                         if (value != NULL && strcmp(value, "off") == 0)
4300                                 caching_page->flags1 |= SCP_RCD;
4301                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_CURRENT],
4302                                &lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4303                                sizeof(caching_page_default));
4304                         page_index->page_data =
4305                                 (uint8_t *)lun->mode_pages.caching_page;
4306                         break;
4307                 }
4308                 case SMS_CONTROL_MODE_PAGE: {
4309                         switch (page_index->subpage) {
4310                         case SMS_SUBPAGE_PAGE_0: {
4311                                 struct scsi_control_page *control_page;
4312
4313                                 memcpy(&lun->mode_pages.control_page[
4314                                     CTL_PAGE_DEFAULT],
4315                                        &control_page_default,
4316                                        sizeof(control_page_default));
4317                                 memcpy(&lun->mode_pages.control_page[
4318                                     CTL_PAGE_CHANGEABLE],
4319                                        &control_page_changeable,
4320                                        sizeof(control_page_changeable));
4321                                 memcpy(&lun->mode_pages.control_page[
4322                                     CTL_PAGE_SAVED],
4323                                        &control_page_default,
4324                                        sizeof(control_page_default));
4325                                 control_page = &lun->mode_pages.control_page[
4326                                     CTL_PAGE_SAVED];
4327                                 value = dnvlist_get_string(lun->be_lun->options,
4328                                     "reordering", NULL);
4329                                 if (value != NULL &&
4330                                     strcmp(value, "unrestricted") == 0) {
4331                                         control_page->queue_flags &=
4332                                             ~SCP_QUEUE_ALG_MASK;
4333                                         control_page->queue_flags |=
4334                                             SCP_QUEUE_ALG_UNRESTRICTED;
4335                                 }
4336                                 memcpy(&lun->mode_pages.control_page[
4337                                     CTL_PAGE_CURRENT],
4338                                        &lun->mode_pages.control_page[
4339                                     CTL_PAGE_SAVED],
4340                                        sizeof(control_page_default));
4341                                 page_index->page_data =
4342                                     (uint8_t *)lun->mode_pages.control_page;
4343                                 break;
4344                         }
4345                         case 0x01:
4346                                 memcpy(&lun->mode_pages.control_ext_page[
4347                                     CTL_PAGE_DEFAULT],
4348                                        &control_ext_page_default,
4349                                        sizeof(control_ext_page_default));
4350                                 memcpy(&lun->mode_pages.control_ext_page[
4351                                     CTL_PAGE_CHANGEABLE],
4352                                        &control_ext_page_changeable,
4353                                        sizeof(control_ext_page_changeable));
4354                                 memcpy(&lun->mode_pages.control_ext_page[
4355                                     CTL_PAGE_SAVED],
4356                                        &control_ext_page_default,
4357                                        sizeof(control_ext_page_default));
4358                                 memcpy(&lun->mode_pages.control_ext_page[
4359                                     CTL_PAGE_CURRENT],
4360                                        &lun->mode_pages.control_ext_page[
4361                                     CTL_PAGE_SAVED],
4362                                        sizeof(control_ext_page_default));
4363                                 page_index->page_data =
4364                                     (uint8_t *)lun->mode_pages.control_ext_page;
4365                                 break;
4366                         default:
4367                                 panic("subpage %#x for page %#x is incorrect!",
4368                                       page_index->subpage, page_code);
4369                         }
4370                         break;
4371                 }
4372                 case SMS_INFO_EXCEPTIONS_PAGE: {
4373                         switch (page_index->subpage) {
4374                         case SMS_SUBPAGE_PAGE_0:
4375                                 memcpy(&lun->mode_pages.ie_page[CTL_PAGE_CURRENT],
4376                                        &ie_page_default,
4377                                        sizeof(ie_page_default));
4378                                 memcpy(&lun->mode_pages.ie_page[
4379                                        CTL_PAGE_CHANGEABLE], &ie_page_changeable,
4380                                        sizeof(ie_page_changeable));
4381                                 memcpy(&lun->mode_pages.ie_page[CTL_PAGE_DEFAULT],
4382                                        &ie_page_default,
4383                                        sizeof(ie_page_default));
4384                                 memcpy(&lun->mode_pages.ie_page[CTL_PAGE_SAVED],
4385                                        &ie_page_default,
4386                                        sizeof(ie_page_default));
4387                                 page_index->page_data =
4388                                         (uint8_t *)lun->mode_pages.ie_page;
4389                                 break;
4390                         case 0x02: {
4391                                 struct ctl_logical_block_provisioning_page *page;
4392
4393                                 memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_DEFAULT],
4394                                        &lbp_page_default,
4395                                        sizeof(lbp_page_default));
4396                                 memcpy(&lun->mode_pages.lbp_page[
4397                                        CTL_PAGE_CHANGEABLE], &lbp_page_changeable,
4398                                        sizeof(lbp_page_changeable));
4399                                 memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_SAVED],
4400                                        &lbp_page_default,
4401                                        sizeof(lbp_page_default));
4402                                 page = &lun->mode_pages.lbp_page[CTL_PAGE_SAVED];
4403                                 value = dnvlist_get_string(lun->be_lun->options,
4404                                     "avail-threshold", NULL);
4405                                 if (value != NULL &&
4406                                     ctl_expand_number(value, &ival) == 0) {
4407                                         page->descr[0].flags |= SLBPPD_ENABLED |
4408                                             SLBPPD_ARMING_DEC;
4409                                         if (lun->be_lun->blocksize)
4410                                                 ival /= lun->be_lun->blocksize;
4411                                         else
4412                                                 ival /= 512;
4413                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4414                                             page->descr[0].count);
4415                                 }
4416                                 value = dnvlist_get_string(lun->be_lun->options,
4417                                     "used-threshold", NULL);
4418                                 if (value != NULL &&
4419                                     ctl_expand_number(value, &ival) == 0) {
4420                                         page->descr[1].flags |= SLBPPD_ENABLED |
4421                                             SLBPPD_ARMING_INC;
4422                                         if (lun->be_lun->blocksize)
4423                                                 ival /= lun->be_lun->blocksize;
4424                                         else
4425                                                 ival /= 512;
4426                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4427                                             page->descr[1].count);
4428                                 }
4429                                 value = dnvlist_get_string(lun->be_lun->options,
4430                                     "pool-avail-threshold", NULL);
4431                                 if (value != NULL &&
4432                                     ctl_expand_number(value, &ival) == 0) {
4433                                         page->descr[2].flags |= SLBPPD_ENABLED |
4434                                             SLBPPD_ARMING_DEC;
4435                                         if (lun->be_lun->blocksize)
4436                                                 ival /= lun->be_lun->blocksize;
4437                                         else
4438                                                 ival /= 512;
4439                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4440                                             page->descr[2].count);
4441                                 }
4442                                 value = dnvlist_get_string(lun->be_lun->options,
4443                                     "pool-used-threshold", NULL);
4444                                 if (value != NULL &&
4445                                     ctl_expand_number(value, &ival) == 0) {
4446                                         page->descr[3].flags |= SLBPPD_ENABLED |
4447                                             SLBPPD_ARMING_INC;
4448                                         if (lun->be_lun->blocksize)
4449                                                 ival /= lun->be_lun->blocksize;
4450                                         else
4451                                                 ival /= 512;
4452                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4453                                             page->descr[3].count);
4454                                 }
4455                                 memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_CURRENT],
4456                                        &lun->mode_pages.lbp_page[CTL_PAGE_SAVED],
4457                                        sizeof(lbp_page_default));
4458                                 page_index->page_data =
4459                                         (uint8_t *)lun->mode_pages.lbp_page;
4460                                 break;
4461                         }
4462                         default:
4463                                 panic("subpage %#x for page %#x is incorrect!",
4464                                       page_index->subpage, page_code);
4465                         }
4466                         break;
4467                 }
4468                 case SMS_CDDVD_CAPS_PAGE:{
4469                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4470                             ("subpage %#x for page %#x is incorrect!",
4471                             page_index->subpage, page_code));
4472                         memcpy(&lun->mode_pages.cddvd_page[CTL_PAGE_DEFAULT],
4473                                &cddvd_page_default,
4474                                sizeof(cddvd_page_default));
4475                         memcpy(&lun->mode_pages.cddvd_page[
4476                                CTL_PAGE_CHANGEABLE], &cddvd_page_changeable,
4477                                sizeof(cddvd_page_changeable));
4478                         memcpy(&lun->mode_pages.cddvd_page[CTL_PAGE_SAVED],
4479                                &cddvd_page_default,
4480                                sizeof(cddvd_page_default));
4481                         memcpy(&lun->mode_pages.cddvd_page[CTL_PAGE_CURRENT],
4482                                &lun->mode_pages.cddvd_page[CTL_PAGE_SAVED],
4483                                sizeof(cddvd_page_default));
4484                         page_index->page_data =
4485                                 (uint8_t *)lun->mode_pages.cddvd_page;
4486                         break;
4487                 }
4488                 default:
4489                         panic("invalid page code value %#x", page_code);
4490                 }
4491         }
4492
4493         return (CTL_RETVAL_COMPLETE);
4494 }
4495
4496 static int
4497 ctl_init_log_page_index(struct ctl_lun *lun)
4498 {
4499         struct ctl_page_index *page_index;
4500         int i, j, k, prev;
4501
4502         memcpy(&lun->log_pages.index, log_page_index_template,
4503                sizeof(log_page_index_template));
4504
4505         prev = -1;
4506         for (i = 0, j = 0, k = 0; i < CTL_NUM_LOG_PAGES; i++) {
4507                 page_index = &lun->log_pages.index[i];
4508                 if (lun->be_lun->lun_type == T_DIRECT &&
4509                     (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
4510                         continue;
4511                 if (lun->be_lun->lun_type == T_PROCESSOR &&
4512                     (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
4513                         continue;
4514                 if (lun->be_lun->lun_type == T_CDROM &&
4515                     (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
4516                         continue;
4517
4518                 if (page_index->page_code == SLS_LOGICAL_BLOCK_PROVISIONING &&
4519                     lun->backend->lun_attr == NULL)
4520                         continue;
4521
4522                 if (page_index->page_code != prev) {
4523                         lun->log_pages.pages_page[j] = page_index->page_code;
4524                         prev = page_index->page_code;
4525                         j++;
4526                 }
4527                 lun->log_pages.subpages_page[k*2] = page_index->page_code;
4528                 lun->log_pages.subpages_page[k*2+1] = page_index->subpage;
4529                 k++;
4530         }
4531         lun->log_pages.index[0].page_data = &lun->log_pages.pages_page[0];
4532         lun->log_pages.index[0].page_len = j;
4533         lun->log_pages.index[1].page_data = &lun->log_pages.subpages_page[0];
4534         lun->log_pages.index[1].page_len = k * 2;
4535         lun->log_pages.index[2].page_data = (uint8_t *)&lun->log_pages.temp_page;
4536         lun->log_pages.index[2].page_len = sizeof(lun->log_pages.temp_page);
4537         lun->log_pages.index[3].page_data = &lun->log_pages.lbp_page[0];
4538         lun->log_pages.index[3].page_len = 12*CTL_NUM_LBP_PARAMS;
4539         lun->log_pages.index[4].page_data = (uint8_t *)&lun->log_pages.stat_page;
4540         lun->log_pages.index[4].page_len = sizeof(lun->log_pages.stat_page);
4541         lun->log_pages.index[5].page_data = (uint8_t *)&lun->log_pages.ie_page;
4542         lun->log_pages.index[5].page_len = sizeof(lun->log_pages.ie_page);
4543
4544         return (CTL_RETVAL_COMPLETE);
4545 }
4546
4547 static int
4548 hex2bin(const char *str, uint8_t *buf, int buf_size)
4549 {
4550         int i;
4551         u_char c;
4552
4553         memset(buf, 0, buf_size);
4554         while (isspace(str[0]))
4555                 str++;
4556         if (str[0] == '0' && (str[1] == 'x' || str[1] == 'X'))
4557                 str += 2;
4558         buf_size *= 2;
4559         for (i = 0; str[i] != 0 && i < buf_size; i++) {
4560                 while (str[i] == '-')   /* Skip dashes in UUIDs. */
4561                         str++;
4562                 c = str[i];
4563                 if (isdigit(c))
4564                         c -= '0';
4565                 else if (isalpha(c))
4566                         c -= isupper(c) ? 'A' - 10 : 'a' - 10;
4567                 else
4568                         break;
4569                 if (c >= 16)
4570                         break;
4571                 if ((i & 1) == 0)
4572                         buf[i / 2] |= (c << 4);
4573                 else
4574                         buf[i / 2] |= c;
4575         }
4576         return ((i + 1) / 2);
4577 }
4578
4579 /*
4580  * Add LUN.
4581  *
4582  * Returns 0 for success, non-zero (errno) for failure.
4583  */
4584 int
4585 ctl_add_lun(struct ctl_be_lun *be_lun)
4586 {
4587         struct ctl_softc *ctl_softc = control_softc;
4588         struct ctl_lun *nlun, *lun;
4589         struct scsi_vpd_id_descriptor *desc;
4590         struct scsi_vpd_id_t10 *t10id;
4591         const char *eui, *naa, *scsiname, *uuid, *vendor, *value;
4592         int lun_number;
4593         int devidlen, idlen1, idlen2 = 0, len;
4594
4595         /*
4596          * We support only Direct Access, CD-ROM or Processor LUN types.
4597          */
4598         switch (be_lun->lun_type) {
4599         case T_DIRECT:
4600         case T_PROCESSOR:
4601         case T_CDROM:
4602                 break;
4603         case T_SEQUENTIAL:
4604         case T_CHANGER:
4605         default:
4606                 return (EINVAL);
4607         }
4608         lun = malloc(sizeof(*lun), M_CTL, M_WAITOK | M_ZERO);
4609
4610         lun->pending_sense = malloc(sizeof(struct scsi_sense_data *) *
4611             ctl_max_ports, M_DEVBUF, M_WAITOK | M_ZERO);
4612         lun->pending_ua = malloc(sizeof(ctl_ua_type *) * ctl_max_ports,
4613             M_DEVBUF, M_WAITOK | M_ZERO);
4614         lun->pr_keys = malloc(sizeof(uint64_t *) * ctl_max_ports,
4615             M_DEVBUF, M_WAITOK | M_ZERO);
4616
4617         /* Generate LUN ID. */
4618         devidlen = max(CTL_DEVID_MIN_LEN,
4619             strnlen(be_lun->device_id, CTL_DEVID_LEN));
4620         idlen1 = sizeof(*t10id) + devidlen;
4621         len = sizeof(struct scsi_vpd_id_descriptor) + idlen1;
4622         scsiname = dnvlist_get_string(be_lun->options, "scsiname", NULL);
4623         if (scsiname != NULL) {
4624                 idlen2 = roundup2(strlen(scsiname) + 1, 4);
4625                 len += sizeof(struct scsi_vpd_id_descriptor) + idlen2;
4626         }
4627         eui = dnvlist_get_string(be_lun->options, "eui", NULL);
4628         if (eui != NULL) {
4629                 len += sizeof(struct scsi_vpd_id_descriptor) + 16;
4630         }
4631         naa = dnvlist_get_string(be_lun->options, "naa", NULL);
4632         if (naa != NULL) {
4633                 len += sizeof(struct scsi_vpd_id_descriptor) + 16;
4634         }
4635         uuid = dnvlist_get_string(be_lun->options, "uuid", NULL);
4636         if (uuid != NULL) {
4637                 len += sizeof(struct scsi_vpd_id_descriptor) + 18;
4638         }
4639         lun->lun_devid = malloc(sizeof(struct ctl_devid) + len,
4640             M_CTL, M_WAITOK | M_ZERO);
4641         desc = (struct scsi_vpd_id_descriptor *)lun->lun_devid->data;
4642         desc->proto_codeset = SVPD_ID_CODESET_ASCII;
4643         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_T10;
4644         desc->length = idlen1;
4645         t10id = (struct scsi_vpd_id_t10 *)&desc->identifier[0];
4646         memset(t10id->vendor, ' ', sizeof(t10id->vendor));
4647         if ((vendor = dnvlist_get_string(be_lun->options, "vendor", NULL)) == NULL) {
4648                 strncpy((char *)t10id->vendor, CTL_VENDOR, sizeof(t10id->vendor));
4649         } else {
4650                 strncpy(t10id->vendor, vendor,
4651                     min(sizeof(t10id->vendor), strlen(vendor)));
4652         }
4653         strncpy((char *)t10id->vendor_spec_id,
4654             (char *)be_lun->device_id, devidlen);
4655         if (scsiname != NULL) {
4656                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4657                     desc->length);
4658                 desc->proto_codeset = SVPD_ID_CODESET_UTF8;
4659                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4660                     SVPD_ID_TYPE_SCSI_NAME;
4661                 desc->length = idlen2;
4662                 strlcpy(desc->identifier, scsiname, idlen2);
4663         }
4664         if (eui != NULL) {
4665                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4666                     desc->length);
4667                 desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4668                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4669                     SVPD_ID_TYPE_EUI64;
4670                 desc->length = hex2bin(eui, desc->identifier, 16);
4671                 desc->length = desc->length > 12 ? 16 :
4672                     (desc->length > 8 ? 12 : 8);
4673                 len -= 16 - desc->length;
4674         }
4675         if (naa != NULL) {
4676                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4677                     desc->length);
4678                 desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4679                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4680                     SVPD_ID_TYPE_NAA;
4681                 desc->length = hex2bin(naa, desc->identifier, 16);
4682                 desc->length = desc->length > 8 ? 16 : 8;
4683                 len -= 16 - desc->length;
4684         }
4685         if (uuid != NULL) {
4686                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4687                     desc->length);
4688                 desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4689                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4690                     SVPD_ID_TYPE_UUID;
4691                 desc->identifier[0] = 0x10;
4692                 hex2bin(uuid, &desc->identifier[2], 16);
4693                 desc->length = 18;
4694         }
4695         lun->lun_devid->len = len;
4696
4697         mtx_lock(&ctl_softc->ctl_lock);
4698         /*
4699          * See if the caller requested a particular LUN number.  If so, see
4700          * if it is available.  Otherwise, allocate the first available LUN.
4701          */
4702         if (be_lun->flags & CTL_LUN_FLAG_ID_REQ) {
4703                 if ((be_lun->req_lun_id > (ctl_max_luns - 1))
4704                  || (ctl_is_set(ctl_softc->ctl_lun_mask, be_lun->req_lun_id))) {
4705                         mtx_unlock(&ctl_softc->ctl_lock);
4706                         if (be_lun->req_lun_id > (ctl_max_luns - 1)) {
4707                                 printf("ctl: requested LUN ID %d is higher "
4708                                        "than ctl_max_luns - 1 (%d)\n",
4709                                        be_lun->req_lun_id, ctl_max_luns - 1);
4710                         } else {
4711                                 /*
4712                                  * XXX KDM return an error, or just assign
4713                                  * another LUN ID in this case??
4714                                  */
4715                                 printf("ctl: requested LUN ID %d is already "
4716                                        "in use\n", be_lun->req_lun_id);
4717                         }
4718 fail:
4719                         free(lun->lun_devid, M_CTL);
4720                         free(lun, M_CTL);
4721                         return (ENOSPC);
4722                 }
4723                 lun_number = be_lun->req_lun_id;
4724         } else {
4725                 lun_number = ctl_ffz(ctl_softc->ctl_lun_mask, 0, ctl_max_luns);
4726                 if (lun_number == -1) {
4727                         mtx_unlock(&ctl_softc->ctl_lock);
4728                         printf("ctl: can't allocate LUN, out of LUNs\n");
4729                         goto fail;
4730                 }
4731         }
4732         ctl_set_mask(ctl_softc->ctl_lun_mask, lun_number);
4733         mtx_unlock(&ctl_softc->ctl_lock);
4734
4735         mtx_init(&lun->lun_lock, "CTL LUN", NULL, MTX_DEF);
4736         lun->lun = lun_number;
4737         lun->be_lun = be_lun;
4738         /*
4739          * The processor LUN is always enabled.  Disk LUNs come on line
4740          * disabled, and must be enabled by the backend.
4741          */
4742         lun->flags |= CTL_LUN_DISABLED;
4743         lun->backend = be_lun->be;
4744         be_lun->ctl_lun = lun;
4745         be_lun->lun_id = lun_number;
4746         if (be_lun->flags & CTL_LUN_FLAG_EJECTED)
4747                 lun->flags |= CTL_LUN_EJECTED;
4748         if (be_lun->flags & CTL_LUN_FLAG_NO_MEDIA)
4749                 lun->flags |= CTL_LUN_NO_MEDIA;
4750         if (be_lun->flags & CTL_LUN_FLAG_STOPPED)
4751                 lun->flags |= CTL_LUN_STOPPED;
4752
4753         if (be_lun->flags & CTL_LUN_FLAG_PRIMARY)
4754                 lun->flags |= CTL_LUN_PRIMARY_SC;
4755
4756         value = dnvlist_get_string(be_lun->options, "removable", NULL);
4757         if (value != NULL) {
4758                 if (strcmp(value, "on") == 0)
4759                         lun->flags |= CTL_LUN_REMOVABLE;
4760         } else if (be_lun->lun_type == T_CDROM)
4761                 lun->flags |= CTL_LUN_REMOVABLE;
4762
4763         lun->ctl_softc = ctl_softc;
4764 #ifdef CTL_TIME_IO
4765         lun->last_busy = getsbinuptime();
4766 #endif
4767         LIST_INIT(&lun->ooa_queue);
4768         STAILQ_INIT(&lun->error_list);
4769         lun->ie_reported = 1;
4770         callout_init_mtx(&lun->ie_callout, &lun->lun_lock, 0);
4771         ctl_tpc_lun_init(lun);
4772         if (lun->flags & CTL_LUN_REMOVABLE) {
4773                 lun->prevent = malloc((CTL_MAX_INITIATORS + 31) / 32 * 4,
4774                     M_CTL, M_WAITOK);
4775         }
4776
4777         /*
4778          * Initialize the mode and log page index.
4779          */
4780         ctl_init_page_index(lun);
4781         ctl_init_log_page_index(lun);
4782
4783         /* Setup statistics gathering */
4784         lun->stats.item = lun_number;
4785
4786         /*
4787          * Now, before we insert this lun on the lun list, set the lun
4788          * inventory changed UA for all other luns.
4789          */
4790         mtx_lock(&ctl_softc->ctl_lock);
4791         STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) {
4792                 mtx_lock(&nlun->lun_lock);
4793                 ctl_est_ua_all(nlun, -1, CTL_UA_LUN_CHANGE);
4794                 mtx_unlock(&nlun->lun_lock);
4795         }
4796         STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links);
4797         ctl_softc->ctl_luns[lun_number] = lun;
4798         ctl_softc->num_luns++;
4799         mtx_unlock(&ctl_softc->ctl_lock);
4800
4801         /*
4802          * We successfully added the LUN, attempt to enable it.
4803          */
4804         if (ctl_enable_lun(lun) != 0) {
4805                 printf("%s: ctl_enable_lun() failed!\n", __func__);
4806                 mtx_lock(&ctl_softc->ctl_lock);
4807                 STAILQ_REMOVE(&ctl_softc->lun_list, lun, ctl_lun, links);
4808                 ctl_clear_mask(ctl_softc->ctl_lun_mask, lun_number);
4809                 ctl_softc->ctl_luns[lun_number] = NULL;
4810                 ctl_softc->num_luns--;
4811                 mtx_unlock(&ctl_softc->ctl_lock);
4812                 free(lun->lun_devid, M_CTL);
4813                 free(lun, M_CTL);
4814                 return (EIO);
4815         }
4816
4817         return (0);
4818 }
4819
4820 /*
4821  * Free LUN that has no active requests.
4822  */
4823 static int
4824 ctl_free_lun(struct ctl_lun *lun)
4825 {
4826         struct ctl_softc *softc = lun->ctl_softc;
4827         struct ctl_lun *nlun;
4828         int i;
4829
4830         KASSERT(LIST_EMPTY(&lun->ooa_queue),
4831             ("Freeing a LUN %p with outstanding I/O!\n", lun));
4832
4833         mtx_lock(&softc->ctl_lock);
4834         STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links);
4835         ctl_clear_mask(softc->ctl_lun_mask, lun->lun);
4836         softc->ctl_luns[lun->lun] = NULL;
4837         softc->num_luns--;
4838         STAILQ_FOREACH(nlun, &softc->lun_list, links) {
4839                 mtx_lock(&nlun->lun_lock);
4840                 ctl_est_ua_all(nlun, -1, CTL_UA_LUN_CHANGE);
4841                 mtx_unlock(&nlun->lun_lock);
4842         }
4843         mtx_unlock(&softc->ctl_lock);
4844
4845         /*
4846          * Tell the backend to free resources, if this LUN has a backend.
4847          */
4848         lun->be_lun->lun_shutdown(lun->be_lun);
4849
4850         lun->ie_reportcnt = UINT32_MAX;
4851         callout_drain(&lun->ie_callout);
4852         ctl_tpc_lun_shutdown(lun);
4853         mtx_destroy(&lun->lun_lock);
4854         free(lun->lun_devid, M_CTL);
4855         for (i = 0; i < ctl_max_ports; i++)
4856                 free(lun->pending_ua[i], M_CTL);
4857         free(lun->pending_ua, M_DEVBUF);
4858         for (i = 0; i < ctl_max_ports; i++)
4859                 free(lun->pr_keys[i], M_CTL);
4860         free(lun->pr_keys, M_DEVBUF);
4861         free(lun->write_buffer, M_CTL);
4862         free(lun->prevent, M_CTL);
4863         free(lun, M_CTL);
4864
4865         return (0);
4866 }
4867
4868 static int
4869 ctl_enable_lun(struct ctl_lun *lun)
4870 {
4871         struct ctl_softc *softc;
4872         struct ctl_port *port, *nport;
4873         int retval;
4874
4875         softc = lun->ctl_softc;
4876
4877         mtx_lock(&softc->ctl_lock);
4878         mtx_lock(&lun->lun_lock);
4879         KASSERT((lun->flags & CTL_LUN_DISABLED) != 0,
4880             ("%s: LUN not disabled", __func__));
4881         lun->flags &= ~CTL_LUN_DISABLED;
4882         mtx_unlock(&lun->lun_lock);
4883
4884         STAILQ_FOREACH_SAFE(port, &softc->port_list, links, nport) {
4885                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0 ||
4886                     port->lun_map != NULL || port->lun_enable == NULL)
4887                         continue;
4888
4889                 /*
4890                  * Drop the lock while we call the FETD's enable routine.
4891                  * This can lead to a callback into CTL (at least in the
4892                  * case of the internal initiator frontend.
4893                  */
4894                 mtx_unlock(&softc->ctl_lock);
4895                 retval = port->lun_enable(port->targ_lun_arg, lun->lun);
4896                 mtx_lock(&softc->ctl_lock);
4897                 if (retval != 0) {
4898                         printf("%s: FETD %s port %d returned error "
4899                                "%d for lun_enable on lun %jd\n",
4900                                __func__, port->port_name, port->targ_port,
4901                                retval, (intmax_t)lun->lun);
4902                 }
4903         }
4904
4905         mtx_unlock(&softc->ctl_lock);
4906         ctl_isc_announce_lun(lun);
4907
4908         return (0);
4909 }
4910
4911 static int
4912 ctl_disable_lun(struct ctl_lun *lun)
4913 {
4914         struct ctl_softc *softc;
4915         struct ctl_port *port;
4916         int retval;
4917
4918         softc = lun->ctl_softc;
4919
4920         mtx_lock(&softc->ctl_lock);
4921         mtx_lock(&lun->lun_lock);
4922         KASSERT((lun->flags & CTL_LUN_DISABLED) == 0,
4923             ("%s: LUN not enabled", __func__));
4924         lun->flags |= CTL_LUN_DISABLED;
4925         mtx_unlock(&lun->lun_lock);
4926
4927         STAILQ_FOREACH(port, &softc->port_list, links) {
4928                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0 ||
4929                     port->lun_map != NULL || port->lun_disable == NULL)
4930                         continue;
4931
4932                 /*
4933                  * Drop the lock before we call the frontend's disable
4934                  * routine, to avoid lock order reversals.
4935                  *
4936                  * XXX KDM what happens if the frontend list changes while
4937                  * we're traversing it?  It's unlikely, but should be handled.
4938                  */
4939                 mtx_unlock(&softc->ctl_lock);
4940                 retval = port->lun_disable(port->targ_lun_arg, lun->lun);
4941                 mtx_lock(&softc->ctl_lock);
4942                 if (retval != 0) {
4943                         printf("%s: FETD %s port %d returned error "
4944                                "%d for lun_disable on lun %jd\n",
4945                                __func__, port->port_name, port->targ_port,
4946                                retval, (intmax_t)lun->lun);
4947                 }
4948         }
4949
4950         mtx_unlock(&softc->ctl_lock);
4951         ctl_isc_announce_lun(lun);
4952
4953         return (0);
4954 }
4955
4956 int
4957 ctl_start_lun(struct ctl_be_lun *be_lun)
4958 {
4959         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4960
4961         mtx_lock(&lun->lun_lock);
4962         lun->flags &= ~CTL_LUN_STOPPED;
4963         mtx_unlock(&lun->lun_lock);
4964         return (0);
4965 }
4966
4967 int
4968 ctl_stop_lun(struct ctl_be_lun *be_lun)
4969 {
4970         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4971
4972         mtx_lock(&lun->lun_lock);
4973         lun->flags |= CTL_LUN_STOPPED;
4974         mtx_unlock(&lun->lun_lock);
4975         return (0);
4976 }
4977
4978 int
4979 ctl_lun_no_media(struct ctl_be_lun *be_lun)
4980 {
4981         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4982
4983         mtx_lock(&lun->lun_lock);
4984         lun->flags |= CTL_LUN_NO_MEDIA;
4985         mtx_unlock(&lun->lun_lock);
4986         return (0);
4987 }
4988
4989 int
4990 ctl_lun_has_media(struct ctl_be_lun *be_lun)
4991 {
4992         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4993         union ctl_ha_msg msg;
4994
4995         mtx_lock(&lun->lun_lock);
4996         lun->flags &= ~(CTL_LUN_NO_MEDIA | CTL_LUN_EJECTED);
4997         if (lun->flags & CTL_LUN_REMOVABLE)
4998                 ctl_est_ua_all(lun, -1, CTL_UA_MEDIUM_CHANGE);
4999         mtx_unlock(&lun->lun_lock);
5000         if ((lun->flags & CTL_LUN_REMOVABLE) &&
5001             lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
5002                 bzero(&msg.ua, sizeof(msg.ua));
5003                 msg.hdr.msg_type = CTL_MSG_UA;
5004                 msg.hdr.nexus.initid = -1;
5005                 msg.hdr.nexus.targ_port = -1;
5006                 msg.hdr.nexus.targ_lun = lun->lun;
5007                 msg.hdr.nexus.targ_mapped_lun = lun->lun;
5008                 msg.ua.ua_all = 1;
5009                 msg.ua.ua_set = 1;
5010                 msg.ua.ua_type = CTL_UA_MEDIUM_CHANGE;
5011                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg.ua),
5012                     M_WAITOK);
5013         }
5014         return (0);
5015 }
5016
5017 int
5018 ctl_lun_ejected(struct ctl_be_lun *be_lun)
5019 {
5020         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
5021
5022         mtx_lock(&lun->lun_lock);
5023         lun->flags |= CTL_LUN_EJECTED;
5024         mtx_unlock(&lun->lun_lock);
5025         return (0);
5026 }
5027
5028 int
5029 ctl_lun_primary(struct ctl_be_lun *be_lun)
5030 {
5031         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
5032
5033         mtx_lock(&lun->lun_lock);
5034         lun->flags |= CTL_LUN_PRIMARY_SC;
5035         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
5036         mtx_unlock(&lun->lun_lock);
5037         ctl_isc_announce_lun(lun);
5038         return (0);
5039 }
5040
5041 int
5042 ctl_lun_secondary(struct ctl_be_lun *be_lun)
5043 {
5044         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
5045
5046         mtx_lock(&lun->lun_lock);
5047         lun->flags &= ~CTL_LUN_PRIMARY_SC;
5048         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
5049         mtx_unlock(&lun->lun_lock);
5050         ctl_isc_announce_lun(lun);
5051         return (0);
5052 }
5053
5054 /*
5055  * Remove LUN.  If there are active requests, wait for completion.
5056  *
5057  * Returns 0 for success, non-zero (errno) for failure.
5058  * Completion is reported to backed via the lun_shutdown() method.
5059  */
5060 int
5061 ctl_remove_lun(struct ctl_be_lun *be_lun)
5062 {
5063         struct ctl_lun *lun;
5064
5065         lun = (struct ctl_lun *)be_lun->ctl_lun;
5066
5067         ctl_disable_lun(lun);
5068
5069         mtx_lock(&lun->lun_lock);
5070         lun->flags |= CTL_LUN_INVALID;
5071
5072         /*
5073          * If there is nothing in the OOA queue, go ahead and free the LUN.
5074          * If we have something in the OOA queue, we'll free it when the
5075          * last I/O completes.
5076          */
5077         if (LIST_EMPTY(&lun->ooa_queue)) {
5078                 mtx_unlock(&lun->lun_lock);
5079                 ctl_free_lun(lun);
5080         } else
5081                 mtx_unlock(&lun->lun_lock);
5082
5083         return (0);
5084 }
5085
5086 void
5087 ctl_lun_capacity_changed(struct ctl_be_lun *be_lun)
5088 {
5089         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
5090         union ctl_ha_msg msg;
5091
5092         mtx_lock(&lun->lun_lock);
5093         ctl_est_ua_all(lun, -1, CTL_UA_CAPACITY_CHANGE);
5094         mtx_unlock(&lun->lun_lock);
5095         if (lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
5096                 /* Send msg to other side. */
5097                 bzero(&msg.ua, sizeof(msg.ua));
5098                 msg.hdr.msg_type = CTL_MSG_UA;
5099                 msg.hdr.nexus.initid = -1;
5100                 msg.hdr.nexus.targ_port = -1;
5101                 msg.hdr.nexus.targ_lun = lun->lun;
5102                 msg.hdr.nexus.targ_mapped_lun = lun->lun;
5103                 msg.ua.ua_all = 1;
5104                 msg.ua.ua_set = 1;
5105                 msg.ua.ua_type = CTL_UA_CAPACITY_CHANGE;
5106                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg.ua),
5107                     M_WAITOK);
5108         }
5109 }
5110
5111 /*
5112  * Backend "memory move is complete" callback for requests that never
5113  * make it down to say RAIDCore's configuration code.
5114  */
5115 int
5116 ctl_config_move_done(union ctl_io *io, bool samethr)
5117 {
5118         int retval;
5119
5120         CTL_DEBUG_PRINT(("ctl_config_move_done\n"));
5121         KASSERT(io->io_hdr.io_type == CTL_IO_SCSI,
5122             ("%s: unexpected I/O type %x", __func__, io->io_hdr.io_type));
5123
5124         if (ctl_debug & CTL_DEBUG_CDB_DATA)
5125                 ctl_data_print(io);
5126         if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN) ||
5127             ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
5128              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS) ||
5129             ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) {
5130                 /*
5131                  * XXX KDM just assuming a single pointer here, and not a
5132                  * S/G list.  If we start using S/G lists for config data,
5133                  * we'll need to know how to clean them up here as well.
5134                  */
5135                 if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5136                         free(io->scsiio.kern_data_ptr, M_CTL);
5137                 ctl_done(io);
5138                 retval = CTL_RETVAL_COMPLETE;
5139         } else {
5140                 /*
5141                  * XXX KDM now we need to continue data movement.  Some
5142                  * options:
5143                  * - call ctl_scsiio() again?  We don't do this for data
5144                  *   writes, because for those at least we know ahead of
5145                  *   time where the write will go and how long it is.  For
5146                  *   config writes, though, that information is largely
5147                  *   contained within the write itself, thus we need to
5148                  *   parse out the data again.
5149                  *
5150                  * - Call some other function once the data is in?
5151                  */
5152
5153                 /*
5154                  * XXX KDM call ctl_scsiio() again for now, and check flag
5155                  * bits to see whether we're allocated or not.
5156                  */
5157                 retval = ctl_scsiio(&io->scsiio);
5158         }
5159         return (retval);
5160 }
5161
5162 /*
5163  * This gets called by a backend driver when it is done with a
5164  * data_submit method.
5165  */
5166 void
5167 ctl_data_submit_done(union ctl_io *io)
5168 {
5169         /*
5170          * If the IO_CONT flag is set, we need to call the supplied
5171          * function to continue processing the I/O, instead of completing
5172          * the I/O just yet.
5173          *
5174          * If there is an error, though, we don't want to keep processing.
5175          * Instead, just send status back to the initiator.
5176          */
5177         if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5178             (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5179             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5180              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5181                 io->scsiio.io_cont(io);
5182                 return;
5183         }
5184         ctl_done(io);
5185 }
5186
5187 /*
5188  * This gets called by a backend driver when it is done with a
5189  * configuration write.
5190  */
5191 void
5192 ctl_config_write_done(union ctl_io *io)
5193 {
5194         uint8_t *buf;
5195
5196         /*
5197          * If the IO_CONT flag is set, we need to call the supplied
5198          * function to continue processing the I/O, instead of completing
5199          * the I/O just yet.
5200          *
5201          * If there is an error, though, we don't want to keep processing.
5202          * Instead, just send status back to the initiator.
5203          */
5204         if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5205             (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5206             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5207              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5208                 io->scsiio.io_cont(io);
5209                 return;
5210         }
5211         /*
5212          * Since a configuration write can be done for commands that actually
5213          * have data allocated, like write buffer, and commands that have
5214          * no data, like start/stop unit, we need to check here.
5215          */
5216         if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5217                 buf = io->scsiio.kern_data_ptr;
5218         else
5219                 buf = NULL;
5220         ctl_done(io);
5221         if (buf)
5222                 free(buf, M_CTL);
5223 }
5224
5225 void
5226 ctl_config_read_done(union ctl_io *io)
5227 {
5228         uint8_t *buf;
5229
5230         /*
5231          * If there is some error -- we are done, skip data transfer.
5232          */
5233         if ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0 ||
5234             ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
5235              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) {
5236                 if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5237                         buf = io->scsiio.kern_data_ptr;
5238                 else
5239                         buf = NULL;
5240                 ctl_done(io);
5241                 if (buf)
5242                         free(buf, M_CTL);
5243                 return;
5244         }
5245
5246         /*
5247          * If the IO_CONT flag is set, we need to call the supplied
5248          * function to continue processing the I/O, instead of completing
5249          * the I/O just yet.
5250          */
5251         if (io->io_hdr.flags & CTL_FLAG_IO_CONT) {
5252                 io->scsiio.io_cont(io);
5253                 return;
5254         }
5255
5256         ctl_datamove(io);
5257 }
5258
5259 /*
5260  * SCSI release command.
5261  */
5262 int
5263 ctl_scsi_release(struct ctl_scsiio *ctsio)
5264 {
5265         struct ctl_lun *lun = CTL_LUN(ctsio);
5266         uint32_t residx;
5267
5268         CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
5269
5270         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5271
5272         /*
5273          * XXX KDM right now, we only support LUN reservation.  We don't
5274          * support 3rd party reservations, or extent reservations, which
5275          * might actually need the parameter list.  If we've gotten this
5276          * far, we've got a LUN reservation.  Anything else got kicked out
5277          * above.  So, according to SPC, ignore the length.
5278          */
5279
5280         mtx_lock(&lun->lun_lock);
5281
5282         /*
5283          * According to SPC, it is not an error for an intiator to attempt
5284          * to release a reservation on a LUN that isn't reserved, or that
5285          * is reserved by another initiator.  The reservation can only be
5286          * released, though, by the initiator who made it or by one of
5287          * several reset type events.
5288          */
5289         if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == residx))
5290                         lun->flags &= ~CTL_LUN_RESERVED;
5291
5292         mtx_unlock(&lun->lun_lock);
5293
5294         ctl_set_success(ctsio);
5295         ctl_done((union ctl_io *)ctsio);
5296         return (CTL_RETVAL_COMPLETE);
5297 }
5298
5299 int
5300 ctl_scsi_reserve(struct ctl_scsiio *ctsio)
5301 {
5302         struct ctl_lun *lun = CTL_LUN(ctsio);
5303         uint32_t residx;
5304
5305         CTL_DEBUG_PRINT(("ctl_reserve\n"));
5306
5307         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5308
5309         /*
5310          * XXX KDM right now, we only support LUN reservation.  We don't
5311          * support 3rd party reservations, or extent reservations, which
5312          * might actually need the parameter list.  If we've gotten this
5313          * far, we've got a LUN reservation.  Anything else got kicked out
5314          * above.  So, according to SPC, ignore the length.
5315          */
5316
5317         mtx_lock(&lun->lun_lock);
5318         if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx != residx)) {
5319                 ctl_set_reservation_conflict(ctsio);
5320                 goto bailout;
5321         }
5322
5323         /* SPC-3 exceptions to SPC-2 RESERVE and RELEASE behavior. */
5324         if (lun->flags & CTL_LUN_PR_RESERVED) {
5325                 ctl_set_success(ctsio);
5326                 goto bailout;
5327         }
5328
5329         lun->flags |= CTL_LUN_RESERVED;
5330         lun->res_idx = residx;
5331         ctl_set_success(ctsio);
5332
5333 bailout:
5334         mtx_unlock(&lun->lun_lock);
5335         ctl_done((union ctl_io *)ctsio);
5336         return (CTL_RETVAL_COMPLETE);
5337 }
5338
5339 int
5340 ctl_start_stop(struct ctl_scsiio *ctsio)
5341 {
5342         struct ctl_lun *lun = CTL_LUN(ctsio);
5343         struct scsi_start_stop_unit *cdb;
5344         int retval;
5345
5346         CTL_DEBUG_PRINT(("ctl_start_stop\n"));
5347
5348         cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
5349
5350         if ((cdb->how & SSS_PC_MASK) == 0) {
5351                 if ((lun->flags & CTL_LUN_PR_RESERVED) &&
5352                     (cdb->how & SSS_START) == 0) {
5353                         uint32_t residx;
5354
5355                         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5356                         if (ctl_get_prkey(lun, residx) == 0 ||
5357                             (lun->pr_res_idx != residx && lun->pr_res_type < 4)) {
5358                                 ctl_set_reservation_conflict(ctsio);
5359                                 ctl_done((union ctl_io *)ctsio);
5360                                 return (CTL_RETVAL_COMPLETE);
5361                         }
5362                 }
5363
5364                 if ((cdb->how & SSS_LOEJ) &&
5365                     (lun->flags & CTL_LUN_REMOVABLE) == 0) {
5366                         ctl_set_invalid_field(ctsio,
5367                                               /*sks_valid*/ 1,
5368                                               /*command*/ 1,
5369                                               /*field*/ 4,
5370                                               /*bit_valid*/ 1,
5371                                               /*bit*/ 1);
5372                         ctl_done((union ctl_io *)ctsio);
5373                         return (CTL_RETVAL_COMPLETE);
5374                 }
5375
5376                 if ((cdb->how & SSS_START) == 0 && (cdb->how & SSS_LOEJ) &&
5377                     lun->prevent_count > 0) {
5378                         /* "Medium removal prevented" */
5379                         ctl_set_sense(ctsio, /*current_error*/ 1,
5380                             /*sense_key*/(lun->flags & CTL_LUN_NO_MEDIA) ?
5381                              SSD_KEY_NOT_READY : SSD_KEY_ILLEGAL_REQUEST,
5382                             /*asc*/ 0x53, /*ascq*/ 0x02, SSD_ELEM_NONE);
5383                         ctl_done((union ctl_io *)ctsio);
5384                         return (CTL_RETVAL_COMPLETE);
5385                 }
5386         }
5387
5388         retval = lun->backend->config_write((union ctl_io *)ctsio);
5389         return (retval);
5390 }
5391
5392 int
5393 ctl_prevent_allow(struct ctl_scsiio *ctsio)
5394 {
5395         struct ctl_lun *lun = CTL_LUN(ctsio);
5396         struct scsi_prevent *cdb;
5397         int retval;
5398         uint32_t initidx;
5399
5400         CTL_DEBUG_PRINT(("ctl_prevent_allow\n"));
5401
5402         cdb = (struct scsi_prevent *)ctsio->cdb;
5403
5404         if ((lun->flags & CTL_LUN_REMOVABLE) == 0 || lun->prevent == NULL) {
5405                 ctl_set_invalid_opcode(ctsio);
5406                 ctl_done((union ctl_io *)ctsio);
5407                 return (CTL_RETVAL_COMPLETE);
5408         }
5409
5410         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5411         mtx_lock(&lun->lun_lock);
5412         if ((cdb->how & PR_PREVENT) &&
5413             ctl_is_set(lun->prevent, initidx) == 0) {
5414                 ctl_set_mask(lun->prevent, initidx);
5415                 lun->prevent_count++;
5416         } else if ((cdb->how & PR_PREVENT) == 0 &&
5417             ctl_is_set(lun->prevent, initidx)) {
5418                 ctl_clear_mask(lun->prevent, initidx);
5419                 lun->prevent_count--;
5420         }
5421         mtx_unlock(&lun->lun_lock);
5422         retval = lun->backend->config_write((union ctl_io *)ctsio);
5423         return (retval);
5424 }
5425
5426 /*
5427  * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
5428  * we don't really do anything with the LBA and length fields if the user
5429  * passes them in.  Instead we'll just flush out the cache for the entire
5430  * LUN.
5431  */
5432 int
5433 ctl_sync_cache(struct ctl_scsiio *ctsio)
5434 {
5435         struct ctl_lun *lun = CTL_LUN(ctsio);
5436         struct ctl_lba_len_flags *lbalen;
5437         uint64_t starting_lba;
5438         uint32_t block_count;
5439         int retval;
5440         uint8_t byte2;
5441
5442         CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
5443
5444         retval = 0;
5445
5446         switch (ctsio->cdb[0]) {
5447         case SYNCHRONIZE_CACHE: {
5448                 struct scsi_sync_cache *cdb;
5449                 cdb = (struct scsi_sync_cache *)ctsio->cdb;
5450
5451                 starting_lba = scsi_4btoul(cdb->begin_lba);
5452                 block_count = scsi_2btoul(cdb->lb_count);
5453                 byte2 = cdb->byte2;
5454                 break;
5455         }
5456         case SYNCHRONIZE_CACHE_16: {
5457                 struct scsi_sync_cache_16 *cdb;
5458                 cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5459
5460                 starting_lba = scsi_8btou64(cdb->begin_lba);
5461                 block_count = scsi_4btoul(cdb->lb_count);
5462                 byte2 = cdb->byte2;
5463                 break;
5464         }
5465         default:
5466                 ctl_set_invalid_opcode(ctsio);
5467                 ctl_done((union ctl_io *)ctsio);
5468                 goto bailout;
5469                 break; /* NOTREACHED */
5470         }
5471
5472         /*
5473          * We check the LBA and length, but don't do anything with them.
5474          * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5475          * get flushed.  This check will just help satisfy anyone who wants
5476          * to see an error for an out of range LBA.
5477          */
5478         if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5479                 ctl_set_lba_out_of_range(ctsio,
5480                     MAX(starting_lba, lun->be_lun->maxlba + 1));
5481                 ctl_done((union ctl_io *)ctsio);
5482                 goto bailout;
5483         }
5484
5485         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5486         lbalen->lba = starting_lba;
5487         lbalen->len = block_count;
5488         lbalen->flags = byte2;
5489         retval = lun->backend->config_write((union ctl_io *)ctsio);
5490
5491 bailout:
5492         return (retval);
5493 }
5494
5495 int
5496 ctl_format(struct ctl_scsiio *ctsio)
5497 {
5498         struct scsi_format *cdb;
5499         int length, defect_list_len;
5500
5501         CTL_DEBUG_PRINT(("ctl_format\n"));
5502
5503         cdb = (struct scsi_format *)ctsio->cdb;
5504
5505         length = 0;
5506         if (cdb->byte2 & SF_FMTDATA) {
5507                 if (cdb->byte2 & SF_LONGLIST)
5508                         length = sizeof(struct scsi_format_header_long);
5509                 else
5510                         length = sizeof(struct scsi_format_header_short);
5511         }
5512
5513         if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5514          && (length > 0)) {
5515                 ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5516                 ctsio->kern_data_len = length;
5517                 ctsio->kern_total_len = length;
5518                 ctsio->kern_rel_offset = 0;
5519                 ctsio->kern_sg_entries = 0;
5520                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5521                 ctsio->be_move_done = ctl_config_move_done;
5522                 ctl_datamove((union ctl_io *)ctsio);
5523
5524                 return (CTL_RETVAL_COMPLETE);
5525         }
5526
5527         defect_list_len = 0;
5528
5529         if (cdb->byte2 & SF_FMTDATA) {
5530                 if (cdb->byte2 & SF_LONGLIST) {
5531                         struct scsi_format_header_long *header;
5532
5533                         header = (struct scsi_format_header_long *)
5534                                 ctsio->kern_data_ptr;
5535
5536                         defect_list_len = scsi_4btoul(header->defect_list_len);
5537                         if (defect_list_len != 0) {
5538                                 ctl_set_invalid_field(ctsio,
5539                                                       /*sks_valid*/ 1,
5540                                                       /*command*/ 0,
5541                                                       /*field*/ 2,
5542                                                       /*bit_valid*/ 0,
5543                                                       /*bit*/ 0);
5544                                 goto bailout;
5545                         }
5546                 } else {
5547                         struct scsi_format_header_short *header;
5548
5549                         header = (struct scsi_format_header_short *)
5550                                 ctsio->kern_data_ptr;
5551
5552                         defect_list_len = scsi_2btoul(header->defect_list_len);
5553                         if (defect_list_len != 0) {
5554                                 ctl_set_invalid_field(ctsio,
5555                                                       /*sks_valid*/ 1,
5556                                                       /*command*/ 0,
5557                                                       /*field*/ 2,
5558                                                       /*bit_valid*/ 0,
5559                                                       /*bit*/ 0);
5560                                 goto bailout;
5561                         }
5562                 }
5563         }
5564
5565         ctl_set_success(ctsio);
5566 bailout:
5567
5568         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5569                 free(ctsio->kern_data_ptr, M_CTL);
5570                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5571         }
5572
5573         ctl_done((union ctl_io *)ctsio);
5574         return (CTL_RETVAL_COMPLETE);
5575 }
5576
5577 int
5578 ctl_read_buffer(struct ctl_scsiio *ctsio)
5579 {
5580         struct ctl_lun *lun = CTL_LUN(ctsio);
5581         uint64_t buffer_offset;
5582         uint32_t len;
5583         uint8_t byte2;
5584         static uint8_t descr[4];
5585         static uint8_t echo_descr[4] = { 0 };
5586
5587         CTL_DEBUG_PRINT(("ctl_read_buffer\n"));
5588
5589         switch (ctsio->cdb[0]) {
5590         case READ_BUFFER: {
5591                 struct scsi_read_buffer *cdb;
5592
5593                 cdb = (struct scsi_read_buffer *)ctsio->cdb;
5594                 buffer_offset = scsi_3btoul(cdb->offset);
5595                 len = scsi_3btoul(cdb->length);
5596                 byte2 = cdb->byte2;
5597                 break;
5598         }
5599         case READ_BUFFER_16: {
5600                 struct scsi_read_buffer_16 *cdb;
5601
5602                 cdb = (struct scsi_read_buffer_16 *)ctsio->cdb;
5603                 buffer_offset = scsi_8btou64(cdb->offset);
5604                 len = scsi_4btoul(cdb->length);
5605                 byte2 = cdb->byte2;
5606                 break;
5607         }
5608         default: /* This shouldn't happen. */
5609                 ctl_set_invalid_opcode(ctsio);
5610                 ctl_done((union ctl_io *)ctsio);
5611                 return (CTL_RETVAL_COMPLETE);
5612         }
5613
5614         if (buffer_offset > CTL_WRITE_BUFFER_SIZE ||
5615             buffer_offset + len > CTL_WRITE_BUFFER_SIZE) {
5616                 ctl_set_invalid_field(ctsio,
5617                                       /*sks_valid*/ 1,
5618                                       /*command*/ 1,
5619                                       /*field*/ 6,
5620                                       /*bit_valid*/ 0,
5621                                       /*bit*/ 0);
5622                 ctl_done((union ctl_io *)ctsio);
5623                 return (CTL_RETVAL_COMPLETE);
5624         }
5625
5626         if ((byte2 & RWB_MODE) == RWB_MODE_DESCR) {
5627                 descr[0] = 0;
5628                 scsi_ulto3b(CTL_WRITE_BUFFER_SIZE, &descr[1]);
5629                 ctsio->kern_data_ptr = descr;
5630                 len = min(len, sizeof(descr));
5631         } else if ((byte2 & RWB_MODE) == RWB_MODE_ECHO_DESCR) {
5632                 ctsio->kern_data_ptr = echo_descr;
5633                 len = min(len, sizeof(echo_descr));
5634         } else {
5635                 if (lun->write_buffer == NULL) {
5636                         lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE,
5637                             M_CTL, M_WAITOK);
5638                 }
5639                 ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5640         }
5641         ctsio->kern_data_len = len;
5642         ctsio->kern_total_len = len;
5643         ctsio->kern_rel_offset = 0;
5644         ctsio->kern_sg_entries = 0;
5645         ctl_set_success(ctsio);
5646         ctsio->be_move_done = ctl_config_move_done;
5647         ctl_datamove((union ctl_io *)ctsio);
5648         return (CTL_RETVAL_COMPLETE);
5649 }
5650
5651 int
5652 ctl_write_buffer(struct ctl_scsiio *ctsio)
5653 {
5654         struct ctl_lun *lun = CTL_LUN(ctsio);
5655         struct scsi_write_buffer *cdb;
5656         int buffer_offset, len;
5657
5658         CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5659
5660         cdb = (struct scsi_write_buffer *)ctsio->cdb;
5661
5662         len = scsi_3btoul(cdb->length);
5663         buffer_offset = scsi_3btoul(cdb->offset);
5664
5665         if (buffer_offset + len > CTL_WRITE_BUFFER_SIZE) {
5666                 ctl_set_invalid_field(ctsio,
5667                                       /*sks_valid*/ 1,
5668                                       /*command*/ 1,
5669                                       /*field*/ 6,
5670                                       /*bit_valid*/ 0,
5671                                       /*bit*/ 0);
5672                 ctl_done((union ctl_io *)ctsio);
5673                 return (CTL_RETVAL_COMPLETE);
5674         }
5675
5676         /*
5677          * If we've got a kernel request that hasn't been malloced yet,
5678          * malloc it and tell the caller the data buffer is here.
5679          */
5680         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5681                 if (lun->write_buffer == NULL) {
5682                         lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE,
5683                             M_CTL, M_WAITOK);
5684                 }
5685                 ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5686                 ctsio->kern_data_len = len;
5687                 ctsio->kern_total_len = len;
5688                 ctsio->kern_rel_offset = 0;
5689                 ctsio->kern_sg_entries = 0;
5690                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5691                 ctsio->be_move_done = ctl_config_move_done;
5692                 ctl_datamove((union ctl_io *)ctsio);
5693
5694                 return (CTL_RETVAL_COMPLETE);
5695         }
5696
5697         ctl_set_success(ctsio);
5698         ctl_done((union ctl_io *)ctsio);
5699         return (CTL_RETVAL_COMPLETE);
5700 }
5701
5702 static int
5703 ctl_write_same_cont(union ctl_io *io)
5704 {
5705         struct ctl_lun *lun = CTL_LUN(io);
5706         struct ctl_scsiio *ctsio;
5707         struct ctl_lba_len_flags *lbalen;
5708         int retval;
5709
5710         ctsio = &io->scsiio;
5711         ctsio->io_hdr.status = CTL_STATUS_NONE;
5712         lbalen = (struct ctl_lba_len_flags *)
5713             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5714         lbalen->lba += lbalen->len;
5715         if ((lun->be_lun->maxlba + 1) - lbalen->lba <= UINT32_MAX) {
5716                 ctsio->io_hdr.flags &= ~CTL_FLAG_IO_CONT;
5717                 lbalen->len = (lun->be_lun->maxlba + 1) - lbalen->lba;
5718         }
5719
5720         CTL_DEBUG_PRINT(("ctl_write_same_cont: calling config_write()\n"));
5721         retval = lun->backend->config_write((union ctl_io *)ctsio);
5722         return (retval);
5723 }
5724
5725 int
5726 ctl_write_same(struct ctl_scsiio *ctsio)
5727 {
5728         struct ctl_lun *lun = CTL_LUN(ctsio);
5729         struct ctl_lba_len_flags *lbalen;
5730         const char *val;
5731         uint64_t lba, ival;
5732         uint32_t num_blocks;
5733         int len, retval;
5734         uint8_t byte2;
5735
5736         CTL_DEBUG_PRINT(("ctl_write_same\n"));
5737
5738         switch (ctsio->cdb[0]) {
5739         case WRITE_SAME_10: {
5740                 struct scsi_write_same_10 *cdb;
5741
5742                 cdb = (struct scsi_write_same_10 *)ctsio->cdb;
5743
5744                 lba = scsi_4btoul(cdb->addr);
5745                 num_blocks = scsi_2btoul(cdb->length);
5746                 byte2 = cdb->byte2;
5747                 break;
5748         }
5749         case WRITE_SAME_16: {
5750                 struct scsi_write_same_16 *cdb;
5751
5752                 cdb = (struct scsi_write_same_16 *)ctsio->cdb;
5753
5754                 lba = scsi_8btou64(cdb->addr);
5755                 num_blocks = scsi_4btoul(cdb->length);
5756                 byte2 = cdb->byte2;
5757                 break;
5758         }
5759         default:
5760                 /*
5761                  * We got a command we don't support.  This shouldn't
5762                  * happen, commands should be filtered out above us.
5763                  */
5764                 ctl_set_invalid_opcode(ctsio);
5765                 ctl_done((union ctl_io *)ctsio);
5766
5767                 return (CTL_RETVAL_COMPLETE);
5768                 break; /* NOTREACHED */
5769         }
5770
5771         /* ANCHOR flag can be used only together with UNMAP */
5772         if ((byte2 & SWS_UNMAP) == 0 && (byte2 & SWS_ANCHOR) != 0) {
5773                 ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
5774                     /*command*/ 1, /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 0);
5775                 ctl_done((union ctl_io *)ctsio);
5776                 return (CTL_RETVAL_COMPLETE);
5777         }
5778
5779         /*
5780          * The first check is to make sure we're in bounds, the second
5781          * check is to catch wrap-around problems.  If the lba + num blocks
5782          * is less than the lba, then we've wrapped around and the block
5783          * range is invalid anyway.
5784          */
5785         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5786          || ((lba + num_blocks) < lba)) {
5787                 ctl_set_lba_out_of_range(ctsio,
5788                     MAX(lba, lun->be_lun->maxlba + 1));
5789                 ctl_done((union ctl_io *)ctsio);
5790                 return (CTL_RETVAL_COMPLETE);
5791         }
5792
5793         /* Zero number of blocks means "to the last logical block" */
5794         if (num_blocks == 0) {
5795                 ival = UINT64_MAX;
5796                 val = dnvlist_get_string(lun->be_lun->options,
5797                     "write_same_max_lba", NULL);
5798                 if (val != NULL)
5799                         ctl_expand_number(val, &ival);
5800                 if ((lun->be_lun->maxlba + 1) - lba > ival) {
5801                         ctl_set_invalid_field(ctsio,
5802                             /*sks_valid*/ 1, /*command*/ 1,
5803                             /*field*/ ctsio->cdb[0] == WRITE_SAME_10 ? 7 : 10,
5804                             /*bit_valid*/ 0, /*bit*/ 0);
5805                         ctl_done((union ctl_io *)ctsio);
5806                         return (CTL_RETVAL_COMPLETE);
5807                 }
5808                 if ((lun->be_lun->maxlba + 1) - lba > UINT32_MAX) {
5809                         ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
5810                         ctsio->io_cont = ctl_write_same_cont;
5811                         num_blocks = 1 << 31;
5812                 } else
5813                         num_blocks = (lun->be_lun->maxlba + 1) - lba;
5814         }
5815
5816         len = lun->be_lun->blocksize;
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 ((byte2 & SWS_NDOB) == 0 &&
5823             (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5824                 ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);
5825                 ctsio->kern_data_len = len;
5826                 ctsio->kern_total_len = len;
5827                 ctsio->kern_rel_offset = 0;
5828                 ctsio->kern_sg_entries = 0;
5829                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5830                 ctsio->be_move_done = ctl_config_move_done;
5831                 ctl_datamove((union ctl_io *)ctsio);
5832
5833                 return (CTL_RETVAL_COMPLETE);
5834         }
5835
5836         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5837         lbalen->lba = lba;
5838         lbalen->len = num_blocks;
5839         lbalen->flags = byte2;
5840         retval = lun->backend->config_write((union ctl_io *)ctsio);
5841
5842         return (retval);
5843 }
5844
5845 int
5846 ctl_unmap(struct ctl_scsiio *ctsio)
5847 {
5848         struct ctl_lun *lun = CTL_LUN(ctsio);
5849         struct scsi_unmap *cdb;
5850         struct ctl_ptr_len_flags *ptrlen;
5851         struct scsi_unmap_header *hdr;
5852         struct scsi_unmap_desc *buf, *end, *endnz, *range;
5853         uint64_t lba;
5854         uint32_t num_blocks;
5855         int len, retval;
5856         uint8_t byte2;
5857
5858         CTL_DEBUG_PRINT(("ctl_unmap\n"));
5859
5860         cdb = (struct scsi_unmap *)ctsio->cdb;
5861         len = scsi_2btoul(cdb->length);
5862         byte2 = cdb->byte2;
5863
5864         /*
5865          * If we've got a kernel request that hasn't been malloced yet,
5866          * malloc it and tell the caller the data buffer is here.
5867          */
5868         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5869                 ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);
5870                 ctsio->kern_data_len = len;
5871                 ctsio->kern_total_len = len;
5872                 ctsio->kern_rel_offset = 0;
5873                 ctsio->kern_sg_entries = 0;
5874                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5875                 ctsio->be_move_done = ctl_config_move_done;
5876                 ctl_datamove((union ctl_io *)ctsio);
5877
5878                 return (CTL_RETVAL_COMPLETE);
5879         }
5880
5881         len = ctsio->kern_total_len - ctsio->kern_data_resid;
5882         hdr = (struct scsi_unmap_header *)ctsio->kern_data_ptr;
5883         if (len < sizeof (*hdr) ||
5884             len < (scsi_2btoul(hdr->length) + sizeof(hdr->length)) ||
5885             len < (scsi_2btoul(hdr->desc_length) + sizeof (*hdr)) ||
5886             scsi_2btoul(hdr->desc_length) % sizeof(*buf) != 0) {
5887                 ctl_set_invalid_field(ctsio,
5888                                       /*sks_valid*/ 0,
5889                                       /*command*/ 0,
5890                                       /*field*/ 0,
5891                                       /*bit_valid*/ 0,
5892                                       /*bit*/ 0);
5893                 goto done;
5894         }
5895         len = scsi_2btoul(hdr->desc_length);
5896         buf = (struct scsi_unmap_desc *)(hdr + 1);
5897         end = buf + len / sizeof(*buf);
5898
5899         endnz = buf;
5900         for (range = buf; range < end; range++) {
5901                 lba = scsi_8btou64(range->lba);
5902                 num_blocks = scsi_4btoul(range->length);
5903                 if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5904                  || ((lba + num_blocks) < lba)) {
5905                         ctl_set_lba_out_of_range(ctsio,
5906                             MAX(lba, lun->be_lun->maxlba + 1));
5907                         ctl_done((union ctl_io *)ctsio);
5908                         return (CTL_RETVAL_COMPLETE);
5909                 }
5910                 if (num_blocks != 0)
5911                         endnz = range + 1;
5912         }
5913
5914         /*
5915          * Block backend can not handle zero last range.
5916          * Filter it out and return if there is nothing left.
5917          */
5918         len = (uint8_t *)endnz - (uint8_t *)buf;
5919         if (len == 0) {
5920                 ctl_set_success(ctsio);
5921                 goto done;
5922         }
5923
5924         mtx_lock(&lun->lun_lock);
5925         ptrlen = (struct ctl_ptr_len_flags *)
5926             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5927         ptrlen->ptr = (void *)buf;
5928         ptrlen->len = len;
5929         ptrlen->flags = byte2;
5930         ctl_try_unblock_others(lun, (union ctl_io *)ctsio, FALSE);
5931         mtx_unlock(&lun->lun_lock);
5932
5933         retval = lun->backend->config_write((union ctl_io *)ctsio);
5934         return (retval);
5935
5936 done:
5937         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5938                 free(ctsio->kern_data_ptr, M_CTL);
5939                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5940         }
5941         ctl_done((union ctl_io *)ctsio);
5942         return (CTL_RETVAL_COMPLETE);
5943 }
5944
5945 int
5946 ctl_default_page_handler(struct ctl_scsiio *ctsio,
5947                          struct ctl_page_index *page_index, uint8_t *page_ptr)
5948 {
5949         struct ctl_lun *lun = CTL_LUN(ctsio);
5950         uint8_t *current_cp;
5951         int set_ua;
5952         uint32_t initidx;
5953
5954         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5955         set_ua = 0;
5956
5957         current_cp = (page_index->page_data + (page_index->page_len *
5958             CTL_PAGE_CURRENT));
5959
5960         mtx_lock(&lun->lun_lock);
5961         if (memcmp(current_cp, page_ptr, page_index->page_len)) {
5962                 memcpy(current_cp, page_ptr, page_index->page_len);
5963                 set_ua = 1;
5964         }
5965         if (set_ua != 0)
5966                 ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
5967         mtx_unlock(&lun->lun_lock);
5968         if (set_ua) {
5969                 ctl_isc_announce_mode(lun,
5970                     ctl_get_initindex(&ctsio->io_hdr.nexus),
5971                     page_index->page_code, page_index->subpage);
5972         }
5973         return (CTL_RETVAL_COMPLETE);
5974 }
5975
5976 static void
5977 ctl_ie_timer(void *arg)
5978 {
5979         struct ctl_lun *lun = arg;
5980         uint64_t t;
5981
5982         if (lun->ie_asc == 0)
5983                 return;
5984
5985         if (lun->MODE_IE.mrie == SIEP_MRIE_UA)
5986                 ctl_est_ua_all(lun, -1, CTL_UA_IE);
5987         else
5988                 lun->ie_reported = 0;
5989
5990         if (lun->ie_reportcnt < scsi_4btoul(lun->MODE_IE.report_count)) {
5991                 lun->ie_reportcnt++;
5992                 t = scsi_4btoul(lun->MODE_IE.interval_timer);
5993                 if (t == 0 || t == UINT32_MAX)
5994                         t = 3000;  /* 5 min */
5995                 callout_schedule_sbt(&lun->ie_callout, SBT_1S / 10 * t,
5996                     SBT_1S / 10, 0);
5997         }
5998 }
5999
6000 int
6001 ctl_ie_page_handler(struct ctl_scsiio *ctsio,
6002                          struct ctl_page_index *page_index, uint8_t *page_ptr)
6003 {
6004         struct ctl_lun *lun = CTL_LUN(ctsio);
6005         struct scsi_info_exceptions_page *pg;
6006         uint64_t t;
6007
6008         (void)ctl_default_page_handler(ctsio, page_index, page_ptr);
6009
6010         pg = (struct scsi_info_exceptions_page *)page_ptr;
6011         mtx_lock(&lun->lun_lock);
6012         if (pg->info_flags & SIEP_FLAGS_TEST) {
6013                 lun->ie_asc = 0x5d;
6014                 lun->ie_ascq = 0xff;
6015                 if (pg->mrie == SIEP_MRIE_UA) {
6016                         ctl_est_ua_all(lun, -1, CTL_UA_IE);
6017                         lun->ie_reported = 1;
6018                 } else {
6019                         ctl_clr_ua_all(lun, -1, CTL_UA_IE);
6020                         lun->ie_reported = -1;
6021                 }
6022                 lun->ie_reportcnt = 1;
6023                 if (lun->ie_reportcnt < scsi_4btoul(pg->report_count)) {
6024                         lun->ie_reportcnt++;
6025                         t = scsi_4btoul(pg->interval_timer);
6026                         if (t == 0 || t == UINT32_MAX)
6027                                 t = 3000;  /* 5 min */
6028                         callout_reset_sbt(&lun->ie_callout, SBT_1S / 10 * t,
6029                             SBT_1S / 10, ctl_ie_timer, lun, 0);
6030                 }
6031         } else {
6032                 lun->ie_asc = 0;
6033                 lun->ie_ascq = 0;
6034                 lun->ie_reported = 1;
6035                 ctl_clr_ua_all(lun, -1, CTL_UA_IE);
6036                 lun->ie_reportcnt = UINT32_MAX;
6037                 callout_stop(&lun->ie_callout);
6038         }
6039         mtx_unlock(&lun->lun_lock);
6040         return (CTL_RETVAL_COMPLETE);
6041 }
6042
6043 static int
6044 ctl_do_mode_select(union ctl_io *io)
6045 {
6046         struct ctl_lun *lun = CTL_LUN(io);
6047         struct scsi_mode_page_header *page_header;
6048         struct ctl_page_index *page_index;
6049         struct ctl_scsiio *ctsio;
6050         int page_len, page_len_offset, page_len_size;
6051         union ctl_modepage_info *modepage_info;
6052         uint16_t *len_left, *len_used;
6053         int retval, i;
6054
6055         ctsio = &io->scsiio;
6056         page_index = NULL;
6057         page_len = 0;
6058
6059         modepage_info = (union ctl_modepage_info *)
6060                 ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6061         len_left = &modepage_info->header.len_left;
6062         len_used = &modepage_info->header.len_used;
6063
6064 do_next_page:
6065
6066         page_header = (struct scsi_mode_page_header *)
6067                 (ctsio->kern_data_ptr + *len_used);
6068
6069         if (*len_left == 0) {
6070                 free(ctsio->kern_data_ptr, M_CTL);
6071                 ctl_set_success(ctsio);
6072                 ctl_done((union ctl_io *)ctsio);
6073                 return (CTL_RETVAL_COMPLETE);
6074         } else if (*len_left < sizeof(struct scsi_mode_page_header)) {
6075                 free(ctsio->kern_data_ptr, M_CTL);
6076                 ctl_set_param_len_error(ctsio);
6077                 ctl_done((union ctl_io *)ctsio);
6078                 return (CTL_RETVAL_COMPLETE);
6079
6080         } else if ((page_header->page_code & SMPH_SPF)
6081                 && (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
6082                 free(ctsio->kern_data_ptr, M_CTL);
6083                 ctl_set_param_len_error(ctsio);
6084                 ctl_done((union ctl_io *)ctsio);
6085                 return (CTL_RETVAL_COMPLETE);
6086         }
6087
6088         /*
6089          * XXX KDM should we do something with the block descriptor?
6090          */
6091         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6092                 page_index = &lun->mode_pages.index[i];
6093                 if (lun->be_lun->lun_type == T_DIRECT &&
6094                     (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6095                         continue;
6096                 if (lun->be_lun->lun_type == T_PROCESSOR &&
6097                     (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6098                         continue;
6099                 if (lun->be_lun->lun_type == T_CDROM &&
6100                     (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6101                         continue;
6102
6103                 if ((page_index->page_code & SMPH_PC_MASK) !=
6104                     (page_header->page_code & SMPH_PC_MASK))
6105                         continue;
6106
6107                 /*
6108                  * If neither page has a subpage code, then we've got a
6109                  * match.
6110                  */
6111                 if (((page_index->page_code & SMPH_SPF) == 0)
6112                  && ((page_header->page_code & SMPH_SPF) == 0)) {
6113                         page_len = page_header->page_length;
6114                         break;
6115                 }
6116
6117                 /*
6118                  * If both pages have subpages, then the subpage numbers
6119                  * have to match.
6120                  */
6121                 if ((page_index->page_code & SMPH_SPF)
6122                   && (page_header->page_code & SMPH_SPF)) {
6123                         struct scsi_mode_page_header_sp *sph;
6124
6125                         sph = (struct scsi_mode_page_header_sp *)page_header;
6126                         if (page_index->subpage == sph->subpage) {
6127                                 page_len = scsi_2btoul(sph->page_length);
6128                                 break;
6129                         }
6130                 }
6131         }
6132
6133         /*
6134          * If we couldn't find the page, or if we don't have a mode select
6135          * handler for it, send back an error to the user.
6136          */
6137         if ((i >= CTL_NUM_MODE_PAGES)
6138          || (page_index->select_handler == NULL)) {
6139                 ctl_set_invalid_field(ctsio,
6140                                       /*sks_valid*/ 1,
6141                                       /*command*/ 0,
6142                                       /*field*/ *len_used,
6143                                       /*bit_valid*/ 0,
6144                                       /*bit*/ 0);
6145                 free(ctsio->kern_data_ptr, M_CTL);
6146                 ctl_done((union ctl_io *)ctsio);
6147                 return (CTL_RETVAL_COMPLETE);
6148         }
6149
6150         if (page_index->page_code & SMPH_SPF) {
6151                 page_len_offset = 2;
6152                 page_len_size = 2;
6153         } else {
6154                 page_len_size = 1;
6155                 page_len_offset = 1;
6156         }
6157
6158         /*
6159          * If the length the initiator gives us isn't the one we specify in
6160          * the mode page header, or if they didn't specify enough data in
6161          * the CDB to avoid truncating this page, kick out the request.
6162          */
6163         if (page_len != page_index->page_len - page_len_offset - page_len_size) {
6164                 ctl_set_invalid_field(ctsio,
6165                                       /*sks_valid*/ 1,
6166                                       /*command*/ 0,
6167                                       /*field*/ *len_used + page_len_offset,
6168                                       /*bit_valid*/ 0,
6169                                       /*bit*/ 0);
6170                 free(ctsio->kern_data_ptr, M_CTL);
6171                 ctl_done((union ctl_io *)ctsio);
6172                 return (CTL_RETVAL_COMPLETE);
6173         }
6174         if (*len_left < page_index->page_len) {
6175                 free(ctsio->kern_data_ptr, M_CTL);
6176                 ctl_set_param_len_error(ctsio);
6177                 ctl_done((union ctl_io *)ctsio);
6178                 return (CTL_RETVAL_COMPLETE);
6179         }
6180
6181         /*
6182          * Run through the mode page, checking to make sure that the bits
6183          * the user changed are actually legal for him to change.
6184          */
6185         for (i = 0; i < page_index->page_len; i++) {
6186                 uint8_t *user_byte, *change_mask, *current_byte;
6187                 int bad_bit;
6188                 int j;
6189
6190                 user_byte = (uint8_t *)page_header + i;
6191                 change_mask = page_index->page_data +
6192                               (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
6193                 current_byte = page_index->page_data +
6194                                (page_index->page_len * CTL_PAGE_CURRENT) + i;
6195
6196                 /*
6197                  * Check to see whether the user set any bits in this byte
6198                  * that he is not allowed to set.
6199                  */
6200                 if ((*user_byte & ~(*change_mask)) ==
6201                     (*current_byte & ~(*change_mask)))
6202                         continue;
6203
6204                 /*
6205                  * Go through bit by bit to determine which one is illegal.
6206                  */
6207                 bad_bit = 0;
6208                 for (j = 7; j >= 0; j--) {
6209                         if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
6210                             (((1 << i) & ~(*change_mask)) & *current_byte)) {
6211                                 bad_bit = i;
6212                                 break;
6213                         }
6214                 }
6215                 ctl_set_invalid_field(ctsio,
6216                                       /*sks_valid*/ 1,
6217                                       /*command*/ 0,
6218                                       /*field*/ *len_used + i,
6219                                       /*bit_valid*/ 1,
6220                                       /*bit*/ bad_bit);
6221                 free(ctsio->kern_data_ptr, M_CTL);
6222                 ctl_done((union ctl_io *)ctsio);
6223                 return (CTL_RETVAL_COMPLETE);
6224         }
6225
6226         /*
6227          * Decrement these before we call the page handler, since we may
6228          * end up getting called back one way or another before the handler
6229          * returns to this context.
6230          */
6231         *len_left -= page_index->page_len;
6232         *len_used += page_index->page_len;
6233
6234         retval = page_index->select_handler(ctsio, page_index,
6235                                             (uint8_t *)page_header);
6236
6237         /*
6238          * If the page handler returns CTL_RETVAL_QUEUED, then we need to
6239          * wait until this queued command completes to finish processing
6240          * the mode page.  If it returns anything other than
6241          * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
6242          * already set the sense information, freed the data pointer, and
6243          * completed the io for us.
6244          */
6245         if (retval != CTL_RETVAL_COMPLETE)
6246                 goto bailout_no_done;
6247
6248         /*
6249          * If the initiator sent us more than one page, parse the next one.
6250          */
6251         if (*len_left > 0)
6252                 goto do_next_page;
6253
6254         ctl_set_success(ctsio);
6255         free(ctsio->kern_data_ptr, M_CTL);
6256         ctl_done((union ctl_io *)ctsio);
6257
6258 bailout_no_done:
6259
6260         return (CTL_RETVAL_COMPLETE);
6261
6262 }
6263
6264 int
6265 ctl_mode_select(struct ctl_scsiio *ctsio)
6266 {
6267         struct ctl_lun *lun = CTL_LUN(ctsio);
6268         union ctl_modepage_info *modepage_info;
6269         int bd_len, i, header_size, param_len, rtd;
6270         uint32_t initidx;
6271
6272         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
6273         switch (ctsio->cdb[0]) {
6274         case MODE_SELECT_6: {
6275                 struct scsi_mode_select_6 *cdb;
6276
6277                 cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
6278
6279                 rtd = (cdb->byte2 & SMS_RTD) ? 1 : 0;
6280                 param_len = cdb->length;
6281                 header_size = sizeof(struct scsi_mode_header_6);
6282                 break;
6283         }
6284         case MODE_SELECT_10: {
6285                 struct scsi_mode_select_10 *cdb;
6286
6287                 cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
6288
6289                 rtd = (cdb->byte2 & SMS_RTD) ? 1 : 0;
6290                 param_len = scsi_2btoul(cdb->length);
6291                 header_size = sizeof(struct scsi_mode_header_10);
6292                 break;
6293         }
6294         default:
6295                 ctl_set_invalid_opcode(ctsio);
6296                 ctl_done((union ctl_io *)ctsio);
6297                 return (CTL_RETVAL_COMPLETE);
6298         }
6299
6300         if (rtd) {
6301                 if (param_len != 0) {
6302                         ctl_set_invalid_field(ctsio, /*sks_valid*/ 0,
6303                             /*command*/ 1, /*field*/ 0,
6304                             /*bit_valid*/ 0, /*bit*/ 0);
6305                         ctl_done((union ctl_io *)ctsio);
6306                         return (CTL_RETVAL_COMPLETE);
6307                 }
6308
6309                 /* Revert to defaults. */
6310                 ctl_init_page_index(lun);
6311                 mtx_lock(&lun->lun_lock);
6312                 ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
6313                 mtx_unlock(&lun->lun_lock);
6314                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6315                         ctl_isc_announce_mode(lun, -1,
6316                             lun->mode_pages.index[i].page_code & SMPH_PC_MASK,
6317                             lun->mode_pages.index[i].subpage);
6318                 }
6319                 ctl_set_success(ctsio);
6320                 ctl_done((union ctl_io *)ctsio);
6321                 return (CTL_RETVAL_COMPLETE);
6322         }
6323
6324         /*
6325          * From SPC-3:
6326          * "A parameter list length of zero indicates that the Data-Out Buffer
6327          * shall be empty. This condition shall not be considered as an error."
6328          */
6329         if (param_len == 0) {
6330                 ctl_set_success(ctsio);
6331                 ctl_done((union ctl_io *)ctsio);
6332                 return (CTL_RETVAL_COMPLETE);
6333         }
6334
6335         /*
6336          * Since we'll hit this the first time through, prior to
6337          * allocation, we don't need to free a data buffer here.
6338          */
6339         if (param_len < header_size) {
6340                 ctl_set_param_len_error(ctsio);
6341                 ctl_done((union ctl_io *)ctsio);
6342                 return (CTL_RETVAL_COMPLETE);
6343         }
6344
6345         /*
6346          * Allocate the data buffer and grab the user's data.  In theory,
6347          * we shouldn't have to sanity check the parameter list length here
6348          * because the maximum size is 64K.  We should be able to malloc
6349          * that much without too many problems.
6350          */
6351         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6352                 ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6353                 ctsio->kern_data_len = param_len;
6354                 ctsio->kern_total_len = param_len;
6355                 ctsio->kern_rel_offset = 0;
6356                 ctsio->kern_sg_entries = 0;
6357                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6358                 ctsio->be_move_done = ctl_config_move_done;
6359                 ctl_datamove((union ctl_io *)ctsio);
6360
6361                 return (CTL_RETVAL_COMPLETE);
6362         }
6363
6364         switch (ctsio->cdb[0]) {
6365         case MODE_SELECT_6: {
6366                 struct scsi_mode_header_6 *mh6;
6367
6368                 mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6369                 bd_len = mh6->blk_desc_len;
6370                 break;
6371         }
6372         case MODE_SELECT_10: {
6373                 struct scsi_mode_header_10 *mh10;
6374
6375                 mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6376                 bd_len = scsi_2btoul(mh10->blk_desc_len);
6377                 break;
6378         }
6379         default:
6380                 panic("%s: Invalid CDB type %#x", __func__, ctsio->cdb[0]);
6381         }
6382
6383         if (param_len < (header_size + bd_len)) {
6384                 free(ctsio->kern_data_ptr, M_CTL);
6385                 ctl_set_param_len_error(ctsio);
6386                 ctl_done((union ctl_io *)ctsio);
6387                 return (CTL_RETVAL_COMPLETE);
6388         }
6389
6390         /*
6391          * Set the IO_CONT flag, so that if this I/O gets passed to
6392          * ctl_config_write_done(), it'll get passed back to
6393          * ctl_do_mode_select() for further processing, or completion if
6394          * we're all done.
6395          */
6396         ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6397         ctsio->io_cont = ctl_do_mode_select;
6398
6399         modepage_info = (union ctl_modepage_info *)
6400                 ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6401         memset(modepage_info, 0, sizeof(*modepage_info));
6402         modepage_info->header.len_left = param_len - header_size - bd_len;
6403         modepage_info->header.len_used = header_size + bd_len;
6404
6405         return (ctl_do_mode_select((union ctl_io *)ctsio));
6406 }
6407
6408 int
6409 ctl_mode_sense(struct ctl_scsiio *ctsio)
6410 {
6411         struct ctl_lun *lun = CTL_LUN(ctsio);
6412         int pc, page_code, llba, subpage;
6413         int alloc_len, page_len, header_len, bd_len, total_len;
6414         void *block_desc;
6415         struct ctl_page_index *page_index;
6416
6417         llba = 0;
6418
6419         CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6420
6421         switch (ctsio->cdb[0]) {
6422         case MODE_SENSE_6: {
6423                 struct scsi_mode_sense_6 *cdb;
6424
6425                 cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6426
6427                 header_len = sizeof(struct scsi_mode_hdr_6);
6428                 if (cdb->byte2 & SMS_DBD)
6429                         bd_len = 0;
6430                 else
6431                         bd_len = sizeof(struct scsi_mode_block_descr);
6432                 header_len += bd_len;
6433
6434                 pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6435                 page_code = cdb->page & SMS_PAGE_CODE;
6436                 subpage = cdb->subpage;
6437                 alloc_len = cdb->length;
6438                 break;
6439         }
6440         case MODE_SENSE_10: {
6441                 struct scsi_mode_sense_10 *cdb;
6442
6443                 cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6444
6445                 header_len = sizeof(struct scsi_mode_hdr_10);
6446                 if (cdb->byte2 & SMS_DBD) {
6447                         bd_len = 0;
6448                 } else if (lun->be_lun->lun_type == T_DIRECT) {
6449                         if (cdb->byte2 & SMS10_LLBAA) {
6450                                 llba = 1;
6451                                 bd_len = sizeof(struct scsi_mode_block_descr_dlong);
6452                         } else
6453                                 bd_len = sizeof(struct scsi_mode_block_descr_dshort);
6454                 } else
6455                         bd_len = sizeof(struct scsi_mode_block_descr);
6456                 header_len += bd_len;
6457
6458                 pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6459                 page_code = cdb->page & SMS_PAGE_CODE;
6460                 subpage = cdb->subpage;
6461                 alloc_len = scsi_2btoul(cdb->length);
6462                 break;
6463         }
6464         default:
6465                 ctl_set_invalid_opcode(ctsio);
6466                 ctl_done((union ctl_io *)ctsio);
6467                 return (CTL_RETVAL_COMPLETE);
6468                 break; /* NOTREACHED */
6469         }
6470
6471         /*
6472          * We have to make a first pass through to calculate the size of
6473          * the pages that match the user's query.  Then we allocate enough
6474          * memory to hold it, and actually copy the data into the buffer.
6475          */
6476         switch (page_code) {
6477         case SMS_ALL_PAGES_PAGE: {
6478                 u_int i;
6479
6480                 page_len = 0;
6481
6482                 /*
6483                  * At the moment, values other than 0 and 0xff here are
6484                  * reserved according to SPC-3.
6485                  */
6486                 if ((subpage != SMS_SUBPAGE_PAGE_0)
6487                  && (subpage != SMS_SUBPAGE_ALL)) {
6488                         ctl_set_invalid_field(ctsio,
6489                                               /*sks_valid*/ 1,
6490                                               /*command*/ 1,
6491                                               /*field*/ 3,
6492                                               /*bit_valid*/ 0,
6493                                               /*bit*/ 0);
6494                         ctl_done((union ctl_io *)ctsio);
6495                         return (CTL_RETVAL_COMPLETE);
6496                 }
6497
6498                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6499                         page_index = &lun->mode_pages.index[i];
6500
6501                         /* Make sure the page is supported for this dev type */
6502                         if (lun->be_lun->lun_type == T_DIRECT &&
6503                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6504                                 continue;
6505                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6506                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6507                                 continue;
6508                         if (lun->be_lun->lun_type == T_CDROM &&
6509                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6510                                 continue;
6511
6512                         /*
6513                          * We don't use this subpage if the user didn't
6514                          * request all subpages.
6515                          */
6516                         if ((page_index->subpage != 0)
6517                          && (subpage == SMS_SUBPAGE_PAGE_0))
6518                                 continue;
6519
6520                         page_len += page_index->page_len;
6521                 }
6522                 break;
6523         }
6524         default: {
6525                 u_int i;
6526
6527                 page_len = 0;
6528
6529                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6530                         page_index = &lun->mode_pages.index[i];
6531
6532                         /* Make sure the page is supported for this dev type */
6533                         if (lun->be_lun->lun_type == T_DIRECT &&
6534                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6535                                 continue;
6536                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6537                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6538                                 continue;
6539                         if (lun->be_lun->lun_type == T_CDROM &&
6540                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6541                                 continue;
6542
6543                         /* Look for the right page code */
6544                         if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6545                                 continue;
6546
6547                         /* Look for the right subpage or the subpage wildcard*/
6548                         if ((page_index->subpage != subpage)
6549                          && (subpage != SMS_SUBPAGE_ALL))
6550                                 continue;
6551
6552                         page_len += page_index->page_len;
6553                 }
6554
6555                 if (page_len == 0) {
6556                         ctl_set_invalid_field(ctsio,
6557                                               /*sks_valid*/ 1,
6558                                               /*command*/ 1,
6559                                               /*field*/ 2,
6560                                               /*bit_valid*/ 1,
6561                                               /*bit*/ 5);
6562                         ctl_done((union ctl_io *)ctsio);
6563                         return (CTL_RETVAL_COMPLETE);
6564                 }
6565                 break;
6566         }
6567         }
6568
6569         total_len = header_len + page_len;
6570
6571         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6572         ctsio->kern_sg_entries = 0;
6573         ctsio->kern_rel_offset = 0;
6574         ctsio->kern_data_len = min(total_len, alloc_len);
6575         ctsio->kern_total_len = ctsio->kern_data_len;
6576
6577         switch (ctsio->cdb[0]) {
6578         case MODE_SENSE_6: {
6579                 struct scsi_mode_hdr_6 *header;
6580
6581                 header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
6582
6583                 header->datalen = MIN(total_len - 1, 254);
6584                 if (lun->be_lun->lun_type == T_DIRECT) {
6585                         header->dev_specific = 0x10; /* DPOFUA */
6586                         if ((lun->be_lun->flags & CTL_LUN_FLAG_READONLY) ||
6587                             (lun->MODE_CTRL.eca_and_aen & SCP_SWP) != 0)
6588                                 header->dev_specific |= 0x80; /* WP */
6589                 }
6590                 header->block_descr_len = bd_len;
6591                 block_desc = &header[1];
6592                 break;
6593         }
6594         case MODE_SENSE_10: {
6595                 struct scsi_mode_hdr_10 *header;
6596                 int datalen;
6597
6598                 header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
6599
6600                 datalen = MIN(total_len - 2, 65533);
6601                 scsi_ulto2b(datalen, header->datalen);
6602                 if (lun->be_lun->lun_type == T_DIRECT) {
6603                         header->dev_specific = 0x10; /* DPOFUA */
6604                         if ((lun->be_lun->flags & CTL_LUN_FLAG_READONLY) ||
6605                             (lun->MODE_CTRL.eca_and_aen & SCP_SWP) != 0)
6606                                 header->dev_specific |= 0x80; /* WP */
6607                 }
6608                 if (llba)
6609                         header->flags |= SMH_LONGLBA;
6610                 scsi_ulto2b(bd_len, header->block_descr_len);
6611                 block_desc = &header[1];
6612                 break;
6613         }
6614         default:
6615                 panic("%s: Invalid CDB type %#x", __func__, ctsio->cdb[0]);
6616         }
6617
6618         /*
6619          * If we've got a disk, use its blocksize in the block
6620          * descriptor.  Otherwise, just set it to 0.
6621          */
6622         if (bd_len > 0) {
6623                 if (lun->be_lun->lun_type == T_DIRECT) {
6624                         if (llba) {
6625                                 struct scsi_mode_block_descr_dlong *bd = block_desc;
6626                                 if (lun->be_lun->maxlba != 0)
6627                                         scsi_u64to8b(lun->be_lun->maxlba + 1,
6628                                             bd->num_blocks);
6629                                 scsi_ulto4b(lun->be_lun->blocksize,
6630                                     bd->block_len);
6631                         } else {
6632                                 struct scsi_mode_block_descr_dshort *bd = block_desc;
6633                                 if (lun->be_lun->maxlba != 0)
6634                                         scsi_ulto4b(MIN(lun->be_lun->maxlba+1,
6635                                             UINT32_MAX), bd->num_blocks);
6636                                 scsi_ulto3b(lun->be_lun->blocksize,
6637                                     bd->block_len);
6638                         }
6639                 } else {
6640                         struct scsi_mode_block_descr *bd = block_desc;
6641                         scsi_ulto3b(0, bd->block_len);
6642                 }
6643         }
6644
6645         switch (page_code) {
6646         case SMS_ALL_PAGES_PAGE: {
6647                 int i, data_used;
6648
6649                 data_used = header_len;
6650                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6651                         struct ctl_page_index *page_index;
6652
6653                         page_index = &lun->mode_pages.index[i];
6654                         if (lun->be_lun->lun_type == T_DIRECT &&
6655                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6656                                 continue;
6657                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6658                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6659                                 continue;
6660                         if (lun->be_lun->lun_type == T_CDROM &&
6661                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6662                                 continue;
6663
6664                         /*
6665                          * We don't use this subpage if the user didn't
6666                          * request all subpages.  We already checked (above)
6667                          * to make sure the user only specified a subpage
6668                          * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case.
6669                          */
6670                         if ((page_index->subpage != 0)
6671                          && (subpage == SMS_SUBPAGE_PAGE_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         default: {
6690                 int i, data_used;
6691
6692                 data_used = header_len;
6693
6694                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6695                         struct ctl_page_index *page_index;
6696
6697                         page_index = &lun->mode_pages.index[i];
6698
6699                         /* Look for the right page code */
6700                         if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6701                                 continue;
6702
6703                         /* Look for the right subpage or the subpage wildcard*/
6704                         if ((page_index->subpage != subpage)
6705                          && (subpage != SMS_SUBPAGE_ALL))
6706                                 continue;
6707
6708                         /* Make sure the page is supported for this dev type */
6709                         if (lun->be_lun->lun_type == T_DIRECT &&
6710                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6711                                 continue;
6712                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6713                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6714                                 continue;
6715                         if (lun->be_lun->lun_type == T_CDROM &&
6716                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6717                                 continue;
6718
6719                         /*
6720                          * Call the handler, if it exists, to update the
6721                          * page to the latest values.
6722                          */
6723                         if (page_index->sense_handler != NULL)
6724                                 page_index->sense_handler(ctsio, page_index,pc);
6725
6726                         memcpy(ctsio->kern_data_ptr + data_used,
6727                                page_index->page_data +
6728                                (page_index->page_len * pc),
6729                                page_index->page_len);
6730                         data_used += page_index->page_len;
6731                 }
6732                 break;
6733         }
6734         }
6735
6736         ctl_set_success(ctsio);
6737         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6738         ctsio->be_move_done = ctl_config_move_done;
6739         ctl_datamove((union ctl_io *)ctsio);
6740         return (CTL_RETVAL_COMPLETE);
6741 }
6742
6743 int
6744 ctl_temp_log_sense_handler(struct ctl_scsiio *ctsio,
6745                                struct ctl_page_index *page_index,
6746                                int pc)
6747 {
6748         struct ctl_lun *lun = CTL_LUN(ctsio);
6749         struct scsi_log_temperature *data;
6750         const char *value;
6751
6752         data = (struct scsi_log_temperature *)page_index->page_data;
6753
6754         scsi_ulto2b(SLP_TEMPERATURE, data->hdr.param_code);
6755         data->hdr.param_control = SLP_LBIN;
6756         data->hdr.param_len = sizeof(struct scsi_log_temperature) -
6757             sizeof(struct scsi_log_param_header);
6758         if ((value = dnvlist_get_string(lun->be_lun->options, "temperature",
6759             NULL)) != NULL)
6760                 data->temperature = strtol(value, NULL, 0);
6761         else
6762                 data->temperature = 0xff;
6763         data++;
6764
6765         scsi_ulto2b(SLP_REFTEMPERATURE, data->hdr.param_code);
6766         data->hdr.param_control = SLP_LBIN;
6767         data->hdr.param_len = sizeof(struct scsi_log_temperature) -
6768             sizeof(struct scsi_log_param_header);
6769         if ((value = dnvlist_get_string(lun->be_lun->options, "reftemperature",
6770             NULL)) != NULL)
6771                 data->temperature = strtol(value, NULL, 0);
6772         else
6773                 data->temperature = 0xff;
6774         return (0);
6775 }
6776
6777 int
6778 ctl_lbp_log_sense_handler(struct ctl_scsiio *ctsio,
6779                                struct ctl_page_index *page_index,
6780                                int pc)
6781 {
6782         struct ctl_lun *lun = CTL_LUN(ctsio);
6783         struct scsi_log_param_header *phdr;
6784         uint8_t *data;
6785         uint64_t val;
6786
6787         data = page_index->page_data;
6788
6789         if (lun->backend->lun_attr != NULL &&
6790             (val = lun->backend->lun_attr(lun->be_lun, "blocksavail"))
6791              != UINT64_MAX) {
6792                 phdr = (struct scsi_log_param_header *)data;
6793                 scsi_ulto2b(0x0001, phdr->param_code);
6794                 phdr->param_control = SLP_LBIN | SLP_LP;
6795                 phdr->param_len = 8;
6796                 data = (uint8_t *)(phdr + 1);
6797                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6798                 data[4] = 0x02; /* per-pool */
6799                 data += phdr->param_len;
6800         }
6801
6802         if (lun->backend->lun_attr != NULL &&
6803             (val = lun->backend->lun_attr(lun->be_lun, "blocksused"))
6804              != UINT64_MAX) {
6805                 phdr = (struct scsi_log_param_header *)data;
6806                 scsi_ulto2b(0x0002, phdr->param_code);
6807                 phdr->param_control = SLP_LBIN | SLP_LP;
6808                 phdr->param_len = 8;
6809                 data = (uint8_t *)(phdr + 1);
6810                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6811                 data[4] = 0x01; /* per-LUN */
6812                 data += phdr->param_len;
6813         }
6814
6815         if (lun->backend->lun_attr != NULL &&
6816             (val = lun->backend->lun_attr(lun->be_lun, "poolblocksavail"))
6817              != UINT64_MAX) {
6818                 phdr = (struct scsi_log_param_header *)data;
6819                 scsi_ulto2b(0x00f1, phdr->param_code);
6820                 phdr->param_control = SLP_LBIN | SLP_LP;
6821                 phdr->param_len = 8;
6822                 data = (uint8_t *)(phdr + 1);
6823                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6824                 data[4] = 0x02; /* per-pool */
6825                 data += phdr->param_len;
6826         }
6827
6828         if (lun->backend->lun_attr != NULL &&
6829             (val = lun->backend->lun_attr(lun->be_lun, "poolblocksused"))
6830              != UINT64_MAX) {
6831                 phdr = (struct scsi_log_param_header *)data;
6832                 scsi_ulto2b(0x00f2, phdr->param_code);
6833                 phdr->param_control = SLP_LBIN | SLP_LP;
6834                 phdr->param_len = 8;
6835                 data = (uint8_t *)(phdr + 1);
6836                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6837                 data[4] = 0x02; /* per-pool */
6838                 data += phdr->param_len;
6839         }
6840
6841         page_index->page_len = data - page_index->page_data;
6842         return (0);
6843 }
6844
6845 int
6846 ctl_sap_log_sense_handler(struct ctl_scsiio *ctsio,
6847                                struct ctl_page_index *page_index,
6848                                int pc)
6849 {
6850         struct ctl_lun *lun = CTL_LUN(ctsio);
6851         struct stat_page *data;
6852         struct bintime *t;
6853
6854         data = (struct stat_page *)page_index->page_data;
6855
6856         scsi_ulto2b(SLP_SAP, data->sap.hdr.param_code);
6857         data->sap.hdr.param_control = SLP_LBIN;
6858         data->sap.hdr.param_len = sizeof(struct scsi_log_stat_and_perf) -
6859             sizeof(struct scsi_log_param_header);
6860         scsi_u64to8b(lun->stats.operations[CTL_STATS_READ],
6861             data->sap.read_num);
6862         scsi_u64to8b(lun->stats.operations[CTL_STATS_WRITE],
6863             data->sap.write_num);
6864         if (lun->be_lun->blocksize > 0) {
6865                 scsi_u64to8b(lun->stats.bytes[CTL_STATS_WRITE] /
6866                     lun->be_lun->blocksize, data->sap.recvieved_lba);
6867                 scsi_u64to8b(lun->stats.bytes[CTL_STATS_READ] /
6868                     lun->be_lun->blocksize, data->sap.transmitted_lba);
6869         }
6870         t = &lun->stats.time[CTL_STATS_READ];
6871         scsi_u64to8b((uint64_t)t->sec * 1000 + t->frac / (UINT64_MAX / 1000),
6872             data->sap.read_int);
6873         t = &lun->stats.time[CTL_STATS_WRITE];
6874         scsi_u64to8b((uint64_t)t->sec * 1000 + t->frac / (UINT64_MAX / 1000),
6875             data->sap.write_int);
6876         scsi_u64to8b(0, data->sap.weighted_num);
6877         scsi_u64to8b(0, data->sap.weighted_int);
6878         scsi_ulto2b(SLP_IT, data->it.hdr.param_code);
6879         data->it.hdr.param_control = SLP_LBIN;
6880         data->it.hdr.param_len = sizeof(struct scsi_log_idle_time) -
6881             sizeof(struct scsi_log_param_header);
6882 #ifdef CTL_TIME_IO
6883         scsi_u64to8b(lun->idle_time / SBT_1MS, data->it.idle_int);
6884 #endif
6885         scsi_ulto2b(SLP_TI, data->ti.hdr.param_code);
6886         data->it.hdr.param_control = SLP_LBIN;
6887         data->ti.hdr.param_len = sizeof(struct scsi_log_time_interval) -
6888             sizeof(struct scsi_log_param_header);
6889         scsi_ulto4b(3, data->ti.exponent);
6890         scsi_ulto4b(1, data->ti.integer);
6891         return (0);
6892 }
6893
6894 int
6895 ctl_ie_log_sense_handler(struct ctl_scsiio *ctsio,
6896                                struct ctl_page_index *page_index,
6897                                int pc)
6898 {
6899         struct ctl_lun *lun = CTL_LUN(ctsio);
6900         struct scsi_log_informational_exceptions *data;
6901         const char *value;
6902
6903         data = (struct scsi_log_informational_exceptions *)page_index->page_data;
6904
6905         scsi_ulto2b(SLP_IE_GEN, data->hdr.param_code);
6906         data->hdr.param_control = SLP_LBIN;
6907         data->hdr.param_len = sizeof(struct scsi_log_informational_exceptions) -
6908             sizeof(struct scsi_log_param_header);
6909         data->ie_asc = lun->ie_asc;
6910         data->ie_ascq = lun->ie_ascq;
6911         if ((value = dnvlist_get_string(lun->be_lun->options, "temperature",
6912             NULL)) != NULL)
6913                 data->temperature = strtol(value, NULL, 0);
6914         else
6915                 data->temperature = 0xff;
6916         return (0);
6917 }
6918
6919 int
6920 ctl_log_sense(struct ctl_scsiio *ctsio)
6921 {
6922         struct ctl_lun *lun = CTL_LUN(ctsio);
6923         int i, pc, page_code, subpage;
6924         int alloc_len, total_len;
6925         struct ctl_page_index *page_index;
6926         struct scsi_log_sense *cdb;
6927         struct scsi_log_header *header;
6928
6929         CTL_DEBUG_PRINT(("ctl_log_sense\n"));
6930
6931         cdb = (struct scsi_log_sense *)ctsio->cdb;
6932         pc = (cdb->page & SLS_PAGE_CTRL_MASK) >> 6;
6933         page_code = cdb->page & SLS_PAGE_CODE;
6934         subpage = cdb->subpage;
6935         alloc_len = scsi_2btoul(cdb->length);
6936
6937         page_index = NULL;
6938         for (i = 0; i < CTL_NUM_LOG_PAGES; i++) {
6939                 page_index = &lun->log_pages.index[i];
6940
6941                 /* Look for the right page code */
6942                 if ((page_index->page_code & SL_PAGE_CODE) != page_code)
6943                         continue;
6944
6945                 /* Look for the right subpage or the subpage wildcard*/
6946                 if (page_index->subpage != subpage)
6947                         continue;
6948
6949                 break;
6950         }
6951         if (i >= CTL_NUM_LOG_PAGES) {
6952                 ctl_set_invalid_field(ctsio,
6953                                       /*sks_valid*/ 1,
6954                                       /*command*/ 1,
6955                                       /*field*/ 2,
6956                                       /*bit_valid*/ 0,
6957                                       /*bit*/ 0);
6958                 ctl_done((union ctl_io *)ctsio);
6959                 return (CTL_RETVAL_COMPLETE);
6960         }
6961
6962         total_len = sizeof(struct scsi_log_header) + page_index->page_len;
6963
6964         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6965         ctsio->kern_sg_entries = 0;
6966         ctsio->kern_rel_offset = 0;
6967         ctsio->kern_data_len = min(total_len, alloc_len);
6968         ctsio->kern_total_len = ctsio->kern_data_len;
6969
6970         header = (struct scsi_log_header *)ctsio->kern_data_ptr;
6971         header->page = page_index->page_code;
6972         if (page_index->page_code == SLS_LOGICAL_BLOCK_PROVISIONING)
6973                 header->page |= SL_DS;
6974         if (page_index->subpage) {
6975                 header->page |= SL_SPF;
6976                 header->subpage = page_index->subpage;
6977         }
6978         scsi_ulto2b(page_index->page_len, header->datalen);
6979
6980         /*
6981          * Call the handler, if it exists, to update the
6982          * page to the latest values.
6983          */
6984         if (page_index->sense_handler != NULL)
6985                 page_index->sense_handler(ctsio, page_index, pc);
6986
6987         memcpy(header + 1, page_index->page_data, page_index->page_len);
6988
6989         ctl_set_success(ctsio);
6990         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6991         ctsio->be_move_done = ctl_config_move_done;
6992         ctl_datamove((union ctl_io *)ctsio);
6993         return (CTL_RETVAL_COMPLETE);
6994 }
6995
6996 int
6997 ctl_read_capacity(struct ctl_scsiio *ctsio)
6998 {
6999         struct ctl_lun *lun = CTL_LUN(ctsio);
7000         struct scsi_read_capacity *cdb;
7001         struct scsi_read_capacity_data *data;
7002         uint32_t lba;
7003
7004         CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
7005
7006         cdb = (struct scsi_read_capacity *)ctsio->cdb;
7007
7008         lba = scsi_4btoul(cdb->addr);
7009         if (((cdb->pmi & SRC_PMI) == 0)
7010          && (lba != 0)) {
7011                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7012                                       /*sks_valid*/ 1,
7013                                       /*command*/ 1,
7014                                       /*field*/ 2,
7015                                       /*bit_valid*/ 0,
7016                                       /*bit*/ 0);
7017                 ctl_done((union ctl_io *)ctsio);
7018                 return (CTL_RETVAL_COMPLETE);
7019         }
7020
7021         ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7022         data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
7023         ctsio->kern_data_len = sizeof(*data);
7024         ctsio->kern_total_len = sizeof(*data);
7025         ctsio->kern_rel_offset = 0;
7026         ctsio->kern_sg_entries = 0;
7027
7028         /*
7029          * If the maximum LBA is greater than 0xfffffffe, the user must
7030          * issue a SERVICE ACTION IN (16) command, with the read capacity
7031          * serivce action set.
7032          */
7033         if (lun->be_lun->maxlba > 0xfffffffe)
7034                 scsi_ulto4b(0xffffffff, data->addr);
7035         else
7036                 scsi_ulto4b(lun->be_lun->maxlba, data->addr);
7037
7038         /*
7039          * XXX KDM this may not be 512 bytes...
7040          */
7041         scsi_ulto4b(lun->be_lun->blocksize, data->length);
7042
7043         ctl_set_success(ctsio);
7044         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7045         ctsio->be_move_done = ctl_config_move_done;
7046         ctl_datamove((union ctl_io *)ctsio);
7047         return (CTL_RETVAL_COMPLETE);
7048 }
7049
7050 int
7051 ctl_read_capacity_16(struct ctl_scsiio *ctsio)
7052 {
7053         struct ctl_lun *lun = CTL_LUN(ctsio);
7054         struct scsi_read_capacity_16 *cdb;
7055         struct scsi_read_capacity_data_long *data;
7056         uint64_t lba;
7057         uint32_t alloc_len;
7058
7059         CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
7060
7061         cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
7062
7063         alloc_len = scsi_4btoul(cdb->alloc_len);
7064         lba = scsi_8btou64(cdb->addr);
7065
7066         if ((cdb->reladr & SRC16_PMI)
7067          && (lba != 0)) {
7068                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7069                                       /*sks_valid*/ 1,
7070                                       /*command*/ 1,
7071                                       /*field*/ 2,
7072                                       /*bit_valid*/ 0,
7073                                       /*bit*/ 0);
7074                 ctl_done((union ctl_io *)ctsio);
7075                 return (CTL_RETVAL_COMPLETE);
7076         }
7077
7078         ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7079         data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
7080         ctsio->kern_rel_offset = 0;
7081         ctsio->kern_sg_entries = 0;
7082         ctsio->kern_data_len = min(sizeof(*data), alloc_len);
7083         ctsio->kern_total_len = ctsio->kern_data_len;
7084
7085         scsi_u64to8b(lun->be_lun->maxlba, data->addr);
7086         /* XXX KDM this may not be 512 bytes... */
7087         scsi_ulto4b(lun->be_lun->blocksize, data->length);
7088         data->prot_lbppbe = lun->be_lun->pblockexp & SRC16_LBPPBE;
7089         scsi_ulto2b(lun->be_lun->pblockoff & SRC16_LALBA_A, data->lalba_lbp);
7090         if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
7091                 data->lalba_lbp[0] |= SRC16_LBPME | SRC16_LBPRZ;
7092
7093         ctl_set_success(ctsio);
7094         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7095         ctsio->be_move_done = ctl_config_move_done;
7096         ctl_datamove((union ctl_io *)ctsio);
7097         return (CTL_RETVAL_COMPLETE);
7098 }
7099
7100 int
7101 ctl_get_lba_status(struct ctl_scsiio *ctsio)
7102 {
7103         struct ctl_lun *lun = CTL_LUN(ctsio);
7104         struct scsi_get_lba_status *cdb;
7105         struct scsi_get_lba_status_data *data;
7106         struct ctl_lba_len_flags *lbalen;
7107         uint64_t lba;
7108         uint32_t alloc_len, total_len;
7109         int retval;
7110
7111         CTL_DEBUG_PRINT(("ctl_get_lba_status\n"));
7112
7113         cdb = (struct scsi_get_lba_status *)ctsio->cdb;
7114         lba = scsi_8btou64(cdb->addr);
7115         alloc_len = scsi_4btoul(cdb->alloc_len);
7116
7117         if (lba > lun->be_lun->maxlba) {
7118                 ctl_set_lba_out_of_range(ctsio, lba);
7119                 ctl_done((union ctl_io *)ctsio);
7120                 return (CTL_RETVAL_COMPLETE);
7121         }
7122
7123         total_len = sizeof(*data) + sizeof(data->descr[0]);
7124         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7125         data = (struct scsi_get_lba_status_data *)ctsio->kern_data_ptr;
7126         ctsio->kern_rel_offset = 0;
7127         ctsio->kern_sg_entries = 0;
7128         ctsio->kern_data_len = min(total_len, alloc_len);
7129         ctsio->kern_total_len = ctsio->kern_data_len;
7130
7131         /* Fill dummy data in case backend can't tell anything. */
7132         scsi_ulto4b(4 + sizeof(data->descr[0]), data->length);
7133         scsi_u64to8b(lba, data->descr[0].addr);
7134         scsi_ulto4b(MIN(UINT32_MAX, lun->be_lun->maxlba + 1 - lba),
7135             data->descr[0].length);
7136         data->descr[0].status = 0; /* Mapped or unknown. */
7137
7138         ctl_set_success(ctsio);
7139         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7140         ctsio->be_move_done = ctl_config_move_done;
7141
7142         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
7143         lbalen->lba = lba;
7144         lbalen->len = total_len;
7145         lbalen->flags = 0;
7146         retval = lun->backend->config_read((union ctl_io *)ctsio);
7147         return (retval);
7148 }
7149
7150 int
7151 ctl_read_defect(struct ctl_scsiio *ctsio)
7152 {
7153         struct scsi_read_defect_data_10 *ccb10;
7154         struct scsi_read_defect_data_12 *ccb12;
7155         struct scsi_read_defect_data_hdr_10 *data10;
7156         struct scsi_read_defect_data_hdr_12 *data12;
7157         uint32_t alloc_len, data_len;
7158         uint8_t format;
7159
7160         CTL_DEBUG_PRINT(("ctl_read_defect\n"));
7161
7162         if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
7163                 ccb10 = (struct scsi_read_defect_data_10 *)&ctsio->cdb;
7164                 format = ccb10->format;
7165                 alloc_len = scsi_2btoul(ccb10->alloc_length);
7166                 data_len = sizeof(*data10);
7167         } else {
7168                 ccb12 = (struct scsi_read_defect_data_12 *)&ctsio->cdb;
7169                 format = ccb12->format;
7170                 alloc_len = scsi_4btoul(ccb12->alloc_length);
7171                 data_len = sizeof(*data12);
7172         }
7173         if (alloc_len == 0) {
7174                 ctl_set_success(ctsio);
7175                 ctl_done((union ctl_io *)ctsio);
7176                 return (CTL_RETVAL_COMPLETE);
7177         }
7178
7179         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
7180         ctsio->kern_rel_offset = 0;
7181         ctsio->kern_sg_entries = 0;
7182         ctsio->kern_data_len = min(data_len, alloc_len);
7183         ctsio->kern_total_len = ctsio->kern_data_len;
7184
7185         if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
7186                 data10 = (struct scsi_read_defect_data_hdr_10 *)
7187                     ctsio->kern_data_ptr;
7188                 data10->format = format;
7189                 scsi_ulto2b(0, data10->length);
7190         } else {
7191                 data12 = (struct scsi_read_defect_data_hdr_12 *)
7192                     ctsio->kern_data_ptr;
7193                 data12->format = format;
7194                 scsi_ulto2b(0, data12->generation);
7195                 scsi_ulto4b(0, data12->length);
7196         }
7197
7198         ctl_set_success(ctsio);
7199         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7200         ctsio->be_move_done = ctl_config_move_done;
7201         ctl_datamove((union ctl_io *)ctsio);
7202         return (CTL_RETVAL_COMPLETE);
7203 }
7204
7205 int
7206 ctl_report_ident_info(struct ctl_scsiio *ctsio)
7207 {
7208         struct ctl_lun *lun = CTL_LUN(ctsio);
7209         struct scsi_report_ident_info *cdb;
7210         struct scsi_report_ident_info_data *rii_ptr;
7211         struct scsi_report_ident_info_descr *riid_ptr;
7212         const char *oii, *otii;
7213         int retval, alloc_len, total_len = 0, len = 0;
7214
7215         CTL_DEBUG_PRINT(("ctl_report_ident_info\n"));
7216
7217         cdb = (struct scsi_report_ident_info *)ctsio->cdb;
7218         retval = CTL_RETVAL_COMPLETE;
7219
7220         total_len = sizeof(struct scsi_report_ident_info_data);
7221         switch (cdb->type) {
7222         case RII_LUII:
7223                 oii = dnvlist_get_string(lun->be_lun->options,
7224                     "ident_info", NULL);
7225                 if (oii)
7226                         len = strlen(oii);      /* Approximately */
7227                 break;
7228         case RII_LUTII:
7229                 otii = dnvlist_get_string(lun->be_lun->options,
7230                     "text_ident_info", NULL);
7231                 if (otii)
7232                         len = strlen(otii) + 1; /* NULL-terminated */
7233                 break;
7234         case RII_IIS:
7235                 len = 2 * sizeof(struct scsi_report_ident_info_descr);
7236                 break;
7237         default:
7238                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7239                                       /*sks_valid*/ 1,
7240                                       /*command*/ 1,
7241                                       /*field*/ 11,
7242                                       /*bit_valid*/ 1,
7243                                       /*bit*/ 2);
7244                 ctl_done((union ctl_io *)ctsio);
7245                 return(retval);
7246         }
7247         total_len += len;
7248         alloc_len = scsi_4btoul(cdb->length);
7249
7250         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7251         ctsio->kern_sg_entries = 0;
7252         ctsio->kern_rel_offset = 0;
7253         ctsio->kern_data_len = min(total_len, alloc_len);
7254         ctsio->kern_total_len = ctsio->kern_data_len;
7255
7256         rii_ptr = (struct scsi_report_ident_info_data *)ctsio->kern_data_ptr;
7257         switch (cdb->type) {
7258         case RII_LUII:
7259                 if (oii) {
7260                         if (oii[0] == '0' && oii[1] == 'x')
7261                                 len = hex2bin(oii, (uint8_t *)(rii_ptr + 1), len);
7262                         else
7263                                 strncpy((uint8_t *)(rii_ptr + 1), oii, len);
7264                 }
7265                 break;
7266         case RII_LUTII:
7267                 if (otii)
7268                         strlcpy((uint8_t *)(rii_ptr + 1), otii, len);
7269                 break;
7270         case RII_IIS:
7271                 riid_ptr = (struct scsi_report_ident_info_descr *)(rii_ptr + 1);
7272                 riid_ptr->type = RII_LUII;
7273                 scsi_ulto2b(0xffff, riid_ptr->length);
7274                 riid_ptr++;
7275                 riid_ptr->type = RII_LUTII;
7276                 scsi_ulto2b(0xffff, riid_ptr->length);
7277         }
7278         scsi_ulto2b(len, rii_ptr->length);
7279
7280         ctl_set_success(ctsio);
7281         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7282         ctsio->be_move_done = ctl_config_move_done;
7283         ctl_datamove((union ctl_io *)ctsio);
7284         return(retval);
7285 }
7286
7287 int
7288 ctl_report_tagret_port_groups(struct ctl_scsiio *ctsio)
7289 {
7290         struct ctl_softc *softc = CTL_SOFTC(ctsio);
7291         struct ctl_lun *lun = CTL_LUN(ctsio);
7292         struct scsi_maintenance_in *cdb;
7293         int retval;
7294         int alloc_len, ext, total_len = 0, g, pc, pg, ts, os;
7295         int num_ha_groups, num_target_ports, shared_group;
7296         struct ctl_port *port;
7297         struct scsi_target_group_data *rtg_ptr;
7298         struct scsi_target_group_data_extended *rtg_ext_ptr;
7299         struct scsi_target_port_group_descriptor *tpg_desc;
7300
7301         CTL_DEBUG_PRINT(("ctl_report_tagret_port_groups\n"));
7302
7303         cdb = (struct scsi_maintenance_in *)ctsio->cdb;
7304         retval = CTL_RETVAL_COMPLETE;
7305
7306         switch (cdb->byte2 & STG_PDF_MASK) {
7307         case STG_PDF_LENGTH:
7308                 ext = 0;
7309                 break;
7310         case STG_PDF_EXTENDED:
7311                 ext = 1;
7312                 break;
7313         default:
7314                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7315                                       /*sks_valid*/ 1,
7316                                       /*command*/ 1,
7317                                       /*field*/ 2,
7318                                       /*bit_valid*/ 1,
7319                                       /*bit*/ 5);
7320                 ctl_done((union ctl_io *)ctsio);
7321                 return(retval);
7322         }
7323
7324         num_target_ports = 0;
7325         shared_group = (softc->is_single != 0);
7326         mtx_lock(&softc->ctl_lock);
7327         STAILQ_FOREACH(port, &softc->port_list, links) {
7328                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7329                         continue;
7330                 if (ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
7331                         continue;
7332                 num_target_ports++;
7333                 if (port->status & CTL_PORT_STATUS_HA_SHARED)
7334                         shared_group = 1;
7335         }
7336         mtx_unlock(&softc->ctl_lock);
7337         num_ha_groups = (softc->is_single) ? 0 : NUM_HA_SHELVES;
7338
7339         if (ext)
7340                 total_len = sizeof(struct scsi_target_group_data_extended);
7341         else
7342                 total_len = sizeof(struct scsi_target_group_data);
7343         total_len += sizeof(struct scsi_target_port_group_descriptor) *
7344                 (shared_group + num_ha_groups) +
7345             sizeof(struct scsi_target_port_descriptor) * num_target_ports;
7346
7347         alloc_len = scsi_4btoul(cdb->length);
7348
7349         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7350         ctsio->kern_sg_entries = 0;
7351         ctsio->kern_rel_offset = 0;
7352         ctsio->kern_data_len = min(total_len, alloc_len);
7353         ctsio->kern_total_len = ctsio->kern_data_len;
7354
7355         if (ext) {
7356                 rtg_ext_ptr = (struct scsi_target_group_data_extended *)
7357                     ctsio->kern_data_ptr;
7358                 scsi_ulto4b(total_len - 4, rtg_ext_ptr->length);
7359                 rtg_ext_ptr->format_type = 0x10;
7360                 rtg_ext_ptr->implicit_transition_time = 0;
7361                 tpg_desc = &rtg_ext_ptr->groups[0];
7362         } else {
7363                 rtg_ptr = (struct scsi_target_group_data *)
7364                     ctsio->kern_data_ptr;
7365                 scsi_ulto4b(total_len - 4, rtg_ptr->length);
7366                 tpg_desc = &rtg_ptr->groups[0];
7367         }
7368
7369         mtx_lock(&softc->ctl_lock);
7370         pg = softc->port_min / softc->port_cnt;
7371         if (lun->flags & (CTL_LUN_PRIMARY_SC | CTL_LUN_PEER_SC_PRIMARY)) {
7372                 /* Some shelf is known to be primary. */
7373                 if (softc->ha_link == CTL_HA_LINK_OFFLINE)
7374                         os = TPG_ASYMMETRIC_ACCESS_UNAVAILABLE;
7375                 else if (softc->ha_link == CTL_HA_LINK_UNKNOWN)
7376                         os = TPG_ASYMMETRIC_ACCESS_TRANSITIONING;
7377                 else if (softc->ha_mode == CTL_HA_MODE_ACT_STBY)
7378                         os = TPG_ASYMMETRIC_ACCESS_STANDBY;
7379                 else
7380                         os = TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7381                 if (lun->flags & CTL_LUN_PRIMARY_SC) {
7382                         ts = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7383                 } else {
7384                         ts = os;
7385                         os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7386                 }
7387         } else {
7388                 /* No known primary shelf. */
7389                 if (softc->ha_link == CTL_HA_LINK_OFFLINE) {
7390                         ts = TPG_ASYMMETRIC_ACCESS_UNAVAILABLE;
7391                         os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7392                 } else if (softc->ha_link == CTL_HA_LINK_UNKNOWN) {
7393                         ts = TPG_ASYMMETRIC_ACCESS_TRANSITIONING;
7394                         os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7395                 } else {
7396                         ts = os = TPG_ASYMMETRIC_ACCESS_TRANSITIONING;
7397                 }
7398         }
7399         if (shared_group) {
7400                 tpg_desc->pref_state = ts;
7401                 tpg_desc->support = TPG_AO_SUP | TPG_AN_SUP | TPG_S_SUP |
7402                     TPG_U_SUP | TPG_T_SUP;
7403                 scsi_ulto2b(1, tpg_desc->target_port_group);
7404                 tpg_desc->status = TPG_IMPLICIT;
7405                 pc = 0;
7406                 STAILQ_FOREACH(port, &softc->port_list, links) {
7407                         if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7408                                 continue;
7409                         if (!softc->is_single &&
7410                             (port->status & CTL_PORT_STATUS_HA_SHARED) == 0)
7411                                 continue;
7412                         if (ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
7413                                 continue;
7414                         scsi_ulto2b(port->targ_port, tpg_desc->descriptors[pc].
7415                             relative_target_port_identifier);
7416                         pc++;
7417                 }
7418                 tpg_desc->target_port_count = pc;
7419                 tpg_desc = (struct scsi_target_port_group_descriptor *)
7420                     &tpg_desc->descriptors[pc];
7421         }
7422         for (g = 0; g < num_ha_groups; g++) {
7423                 tpg_desc->pref_state = (g == pg) ? ts : os;
7424                 tpg_desc->support = TPG_AO_SUP | TPG_AN_SUP | TPG_S_SUP |
7425                     TPG_U_SUP | TPG_T_SUP;
7426                 scsi_ulto2b(2 + g, tpg_desc->target_port_group);
7427                 tpg_desc->status = TPG_IMPLICIT;
7428                 pc = 0;
7429                 STAILQ_FOREACH(port, &softc->port_list, links) {
7430                         if (port->targ_port < g * softc->port_cnt ||
7431                             port->targ_port >= (g + 1) * softc->port_cnt)
7432                                 continue;
7433                         if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7434                                 continue;
7435                         if (port->status & CTL_PORT_STATUS_HA_SHARED)
7436                                 continue;
7437                         if (ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
7438                                 continue;
7439                         scsi_ulto2b(port->targ_port, tpg_desc->descriptors[pc].
7440                             relative_target_port_identifier);
7441                         pc++;
7442                 }
7443                 tpg_desc->target_port_count = pc;
7444                 tpg_desc = (struct scsi_target_port_group_descriptor *)
7445                     &tpg_desc->descriptors[pc];
7446         }
7447         mtx_unlock(&softc->ctl_lock);
7448
7449         ctl_set_success(ctsio);
7450         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7451         ctsio->be_move_done = ctl_config_move_done;
7452         ctl_datamove((union ctl_io *)ctsio);
7453         return(retval);
7454 }
7455
7456 int
7457 ctl_report_supported_opcodes(struct ctl_scsiio *ctsio)
7458 {
7459         struct ctl_lun *lun = CTL_LUN(ctsio);
7460         struct scsi_report_supported_opcodes *cdb;
7461         const struct ctl_cmd_entry *entry, *sentry;
7462         struct scsi_report_supported_opcodes_all *all;
7463         struct scsi_report_supported_opcodes_descr *descr;
7464         struct scsi_report_supported_opcodes_one *one;
7465         int retval;
7466         int alloc_len, total_len;
7467         int opcode, service_action, i, j, num;
7468
7469         CTL_DEBUG_PRINT(("ctl_report_supported_opcodes\n"));
7470
7471         cdb = (struct scsi_report_supported_opcodes *)ctsio->cdb;
7472         retval = CTL_RETVAL_COMPLETE;
7473
7474         opcode = cdb->requested_opcode;
7475         service_action = scsi_2btoul(cdb->requested_service_action);
7476         switch (cdb->options & RSO_OPTIONS_MASK) {
7477         case RSO_OPTIONS_ALL:
7478                 num = 0;
7479                 for (i = 0; i < 256; i++) {
7480                         entry = &ctl_cmd_table[i];
7481                         if (entry->flags & CTL_CMD_FLAG_SA5) {
7482                                 for (j = 0; j < 32; j++) {
7483                                         sentry = &((const struct ctl_cmd_entry *)
7484                                             entry->execute)[j];
7485                                         if (ctl_cmd_applicable(
7486                                             lun->be_lun->lun_type, sentry))
7487                                                 num++;
7488                                 }
7489                         } else {
7490                                 if (ctl_cmd_applicable(lun->be_lun->lun_type,
7491                                     entry))
7492                                         num++;
7493                         }
7494                 }
7495                 total_len = sizeof(struct scsi_report_supported_opcodes_all) +
7496                     num * sizeof(struct scsi_report_supported_opcodes_descr);
7497                 break;
7498         case RSO_OPTIONS_OC:
7499                 if (ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) {
7500                         ctl_set_invalid_field(/*ctsio*/ ctsio,
7501                                               /*sks_valid*/ 1,
7502                                               /*command*/ 1,
7503                                               /*field*/ 2,
7504                                               /*bit_valid*/ 1,
7505                                               /*bit*/ 2);
7506                         ctl_done((union ctl_io *)ctsio);
7507                         return (CTL_RETVAL_COMPLETE);
7508                 }
7509                 total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7510                 break;
7511         case RSO_OPTIONS_OC_SA:
7512                 if ((ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) == 0 ||
7513                     service_action >= 32) {
7514                         ctl_set_invalid_field(/*ctsio*/ ctsio,
7515                                               /*sks_valid*/ 1,
7516                                               /*command*/ 1,
7517                                               /*field*/ 2,
7518                                               /*bit_valid*/ 1,
7519                                               /*bit*/ 2);
7520                         ctl_done((union ctl_io *)ctsio);
7521                         return (CTL_RETVAL_COMPLETE);
7522                 }
7523                 /* FALLTHROUGH */
7524         case RSO_OPTIONS_OC_ASA:
7525                 total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7526                 break;
7527         default:
7528                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7529                                       /*sks_valid*/ 1,
7530                                       /*command*/ 1,
7531                                       /*field*/ 2,
7532                                       /*bit_valid*/ 1,
7533                                       /*bit*/ 2);
7534                 ctl_done((union ctl_io *)ctsio);
7535                 return (CTL_RETVAL_COMPLETE);
7536         }
7537
7538         alloc_len = scsi_4btoul(cdb->length);
7539
7540         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7541         ctsio->kern_sg_entries = 0;
7542         ctsio->kern_rel_offset = 0;
7543         ctsio->kern_data_len = min(total_len, alloc_len);
7544         ctsio->kern_total_len = ctsio->kern_data_len;
7545
7546         switch (cdb->options & RSO_OPTIONS_MASK) {
7547         case RSO_OPTIONS_ALL:
7548                 all = (struct scsi_report_supported_opcodes_all *)
7549                     ctsio->kern_data_ptr;
7550                 num = 0;
7551                 for (i = 0; i < 256; i++) {
7552                         entry = &ctl_cmd_table[i];
7553                         if (entry->flags & CTL_CMD_FLAG_SA5) {
7554                                 for (j = 0; j < 32; j++) {
7555                                         sentry = &((const struct ctl_cmd_entry *)
7556                                             entry->execute)[j];
7557                                         if (!ctl_cmd_applicable(
7558                                             lun->be_lun->lun_type, sentry))
7559                                                 continue;
7560                                         descr = &all->descr[num++];
7561                                         descr->opcode = i;
7562                                         scsi_ulto2b(j, descr->service_action);
7563                                         descr->flags = RSO_SERVACTV;
7564                                         scsi_ulto2b(sentry->length,
7565                                             descr->cdb_length);
7566                                 }
7567                         } else {
7568                                 if (!ctl_cmd_applicable(lun->be_lun->lun_type,
7569                                     entry))
7570                                         continue;
7571                                 descr = &all->descr[num++];
7572                                 descr->opcode = i;
7573                                 scsi_ulto2b(0, descr->service_action);
7574                                 descr->flags = 0;
7575                                 scsi_ulto2b(entry->length, descr->cdb_length);
7576                         }
7577                 }
7578                 scsi_ulto4b(
7579                     num * sizeof(struct scsi_report_supported_opcodes_descr),
7580                     all->length);
7581                 break;
7582         case RSO_OPTIONS_OC:
7583                 one = (struct scsi_report_supported_opcodes_one *)
7584                     ctsio->kern_data_ptr;
7585                 entry = &ctl_cmd_table[opcode];
7586                 goto fill_one;
7587         case RSO_OPTIONS_OC_SA:
7588                 one = (struct scsi_report_supported_opcodes_one *)
7589                     ctsio->kern_data_ptr;
7590                 entry = &ctl_cmd_table[opcode];
7591                 entry = &((const struct ctl_cmd_entry *)
7592                     entry->execute)[service_action];
7593 fill_one:
7594                 if (ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
7595                         one->support = 3;
7596                         scsi_ulto2b(entry->length, one->cdb_length);
7597                         one->cdb_usage[0] = opcode;
7598                         memcpy(&one->cdb_usage[1], entry->usage,
7599                             entry->length - 1);
7600                 } else
7601                         one->support = 1;
7602                 break;
7603         case RSO_OPTIONS_OC_ASA:
7604                 one = (struct scsi_report_supported_opcodes_one *)
7605                     ctsio->kern_data_ptr;
7606                 entry = &ctl_cmd_table[opcode];
7607                 if (entry->flags & CTL_CMD_FLAG_SA5) {
7608                         entry = &((const struct ctl_cmd_entry *)
7609                             entry->execute)[service_action];
7610                 } else if (service_action != 0) {
7611                         one->support = 1;
7612                         break;
7613                 }
7614                 goto fill_one;
7615         }
7616
7617         ctl_set_success(ctsio);
7618         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7619         ctsio->be_move_done = ctl_config_move_done;
7620         ctl_datamove((union ctl_io *)ctsio);
7621         return(retval);
7622 }
7623
7624 int
7625 ctl_report_supported_tmf(struct ctl_scsiio *ctsio)
7626 {
7627         struct scsi_report_supported_tmf *cdb;
7628         struct scsi_report_supported_tmf_ext_data *data;
7629         int retval;
7630         int alloc_len, total_len;
7631
7632         CTL_DEBUG_PRINT(("ctl_report_supported_tmf\n"));
7633
7634         cdb = (struct scsi_report_supported_tmf *)ctsio->cdb;
7635
7636         retval = CTL_RETVAL_COMPLETE;
7637
7638         if (cdb->options & RST_REPD)
7639                 total_len = sizeof(struct scsi_report_supported_tmf_ext_data);
7640         else
7641                 total_len = sizeof(struct scsi_report_supported_tmf_data);
7642         alloc_len = scsi_4btoul(cdb->length);
7643
7644         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7645         ctsio->kern_sg_entries = 0;
7646         ctsio->kern_rel_offset = 0;
7647         ctsio->kern_data_len = min(total_len, alloc_len);
7648         ctsio->kern_total_len = ctsio->kern_data_len;
7649
7650         data = (struct scsi_report_supported_tmf_ext_data *)ctsio->kern_data_ptr;
7651         data->byte1 |= RST_ATS | RST_ATSS | RST_CTSS | RST_LURS | RST_QTS |
7652             RST_TRS;
7653         data->byte2 |= RST_QAES | RST_QTSS | RST_ITNRS;
7654         data->length = total_len - 4;
7655
7656         ctl_set_success(ctsio);
7657         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7658         ctsio->be_move_done = ctl_config_move_done;
7659         ctl_datamove((union ctl_io *)ctsio);
7660         return (retval);
7661 }
7662
7663 int
7664 ctl_report_timestamp(struct ctl_scsiio *ctsio)
7665 {
7666         struct scsi_report_timestamp *cdb;
7667         struct scsi_report_timestamp_data *data;
7668         struct timeval tv;
7669         int64_t timestamp;
7670         int retval;
7671         int alloc_len, total_len;
7672
7673         CTL_DEBUG_PRINT(("ctl_report_timestamp\n"));
7674
7675         cdb = (struct scsi_report_timestamp *)ctsio->cdb;
7676
7677         retval = CTL_RETVAL_COMPLETE;
7678
7679         total_len = sizeof(struct scsi_report_timestamp_data);
7680         alloc_len = scsi_4btoul(cdb->length);
7681
7682         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7683         ctsio->kern_sg_entries = 0;
7684         ctsio->kern_rel_offset = 0;
7685         ctsio->kern_data_len = min(total_len, alloc_len);
7686         ctsio->kern_total_len = ctsio->kern_data_len;
7687
7688         data = (struct scsi_report_timestamp_data *)ctsio->kern_data_ptr;
7689         scsi_ulto2b(sizeof(*data) - 2, data->length);
7690         data->origin = RTS_ORIG_OUTSIDE;
7691         getmicrotime(&tv);
7692         timestamp = (int64_t)tv.tv_sec * 1000 + tv.tv_usec / 1000;
7693         scsi_ulto4b(timestamp >> 16, data->timestamp);
7694         scsi_ulto2b(timestamp & 0xffff, &data->timestamp[4]);
7695
7696         ctl_set_success(ctsio);
7697         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7698         ctsio->be_move_done = ctl_config_move_done;
7699         ctl_datamove((union ctl_io *)ctsio);
7700         return (retval);
7701 }
7702
7703 int
7704 ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7705 {
7706         struct ctl_softc *softc = CTL_SOFTC(ctsio);
7707         struct ctl_lun *lun = CTL_LUN(ctsio);
7708         struct scsi_per_res_in *cdb;
7709         int alloc_len, total_len = 0;
7710         /* struct scsi_per_res_in_rsrv in_data; */
7711         uint64_t key;
7712
7713         CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7714
7715         cdb = (struct scsi_per_res_in *)ctsio->cdb;
7716
7717         alloc_len = scsi_2btoul(cdb->length);
7718
7719 retry:
7720         mtx_lock(&lun->lun_lock);
7721         switch (cdb->action) {
7722         case SPRI_RK: /* read keys */
7723                 total_len = sizeof(struct scsi_per_res_in_keys) +
7724                         lun->pr_key_count *
7725                         sizeof(struct scsi_per_res_key);
7726                 break;
7727         case SPRI_RR: /* read reservation */
7728                 if (lun->flags & CTL_LUN_PR_RESERVED)
7729                         total_len = sizeof(struct scsi_per_res_in_rsrv);
7730                 else
7731                         total_len = sizeof(struct scsi_per_res_in_header);
7732                 break;
7733         case SPRI_RC: /* report capabilities */
7734                 total_len = sizeof(struct scsi_per_res_cap);
7735                 break;
7736         case SPRI_RS: /* read full status */
7737                 total_len = sizeof(struct scsi_per_res_in_header) +
7738                     (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7739                     lun->pr_key_count;
7740                 break;
7741         default:
7742                 panic("%s: Invalid PR type %#x", __func__, cdb->action);
7743         }
7744         mtx_unlock(&lun->lun_lock);
7745
7746         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7747         ctsio->kern_rel_offset = 0;
7748         ctsio->kern_sg_entries = 0;
7749         ctsio->kern_data_len = min(total_len, alloc_len);
7750         ctsio->kern_total_len = ctsio->kern_data_len;
7751
7752         mtx_lock(&lun->lun_lock);
7753         switch (cdb->action) {
7754         case SPRI_RK: { // read keys
7755         struct scsi_per_res_in_keys *res_keys;
7756                 int i, key_count;
7757
7758                 res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7759
7760                 /*
7761                  * We had to drop the lock to allocate our buffer, which
7762                  * leaves time for someone to come in with another
7763                  * persistent reservation.  (That is unlikely, though,
7764                  * since this should be the only persistent reservation
7765                  * command active right now.)
7766                  */
7767                 if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7768                     (lun->pr_key_count *
7769                      sizeof(struct scsi_per_res_key)))){
7770                         mtx_unlock(&lun->lun_lock);
7771                         free(ctsio->kern_data_ptr, M_CTL);
7772                         printf("%s: reservation length changed, retrying\n",
7773                                __func__);
7774                         goto retry;
7775                 }
7776
7777                 scsi_ulto4b(lun->pr_generation, res_keys->header.generation);
7778
7779                 scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7780                              lun->pr_key_count, res_keys->header.length);
7781
7782                 for (i = 0, key_count = 0; i < CTL_MAX_INITIATORS; i++) {
7783                         if ((key = ctl_get_prkey(lun, i)) == 0)
7784                                 continue;
7785
7786                         /*
7787                          * We used lun->pr_key_count to calculate the
7788                          * size to allocate.  If it turns out the number of
7789                          * initiators with the registered flag set is
7790                          * larger than that (i.e. they haven't been kept in
7791                          * sync), we've got a problem.
7792                          */
7793                         if (key_count >= lun->pr_key_count) {
7794                                 key_count++;
7795                                 continue;
7796                         }
7797                         scsi_u64to8b(key, res_keys->keys[key_count].key);
7798                         key_count++;
7799                 }
7800                 break;
7801         }
7802         case SPRI_RR: { // read reservation
7803                 struct scsi_per_res_in_rsrv *res;
7804                 int tmp_len, header_only;
7805
7806                 res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7807
7808                 scsi_ulto4b(lun->pr_generation, res->header.generation);
7809
7810                 if (lun->flags & CTL_LUN_PR_RESERVED)
7811                 {
7812                         tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7813                         scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7814                                     res->header.length);
7815                         header_only = 0;
7816                 } else {
7817                         tmp_len = sizeof(struct scsi_per_res_in_header);
7818                         scsi_ulto4b(0, res->header.length);
7819                         header_only = 1;
7820                 }
7821
7822                 /*
7823                  * We had to drop the lock to allocate our buffer, which
7824                  * leaves time for someone to come in with another
7825                  * persistent reservation.  (That is unlikely, though,
7826                  * since this should be the only persistent reservation
7827                  * command active right now.)
7828                  */
7829                 if (tmp_len != total_len) {
7830                         mtx_unlock(&lun->lun_lock);
7831                         free(ctsio->kern_data_ptr, M_CTL);
7832                         printf("%s: reservation status changed, retrying\n",
7833                                __func__);
7834                         goto retry;
7835                 }
7836
7837                 /*
7838                  * No reservation held, so we're done.
7839                  */
7840                 if (header_only != 0)
7841                         break;
7842
7843                 /*
7844                  * If the registration is an All Registrants type, the key
7845                  * is 0, since it doesn't really matter.
7846                  */
7847                 if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7848                         scsi_u64to8b(ctl_get_prkey(lun, lun->pr_res_idx),
7849                             res->data.reservation);
7850                 }
7851                 res->data.scopetype = lun->pr_res_type;
7852                 break;
7853         }
7854         case SPRI_RC:     //report capabilities
7855         {
7856                 struct scsi_per_res_cap *res_cap;
7857                 uint16_t type_mask;
7858
7859                 res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7860                 scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7861                 res_cap->flags1 = SPRI_CRH;
7862                 res_cap->flags2 = SPRI_TMV | SPRI_ALLOW_5;
7863                 type_mask = SPRI_TM_WR_EX_AR |
7864                             SPRI_TM_EX_AC_RO |
7865                             SPRI_TM_WR_EX_RO |
7866                             SPRI_TM_EX_AC |
7867                             SPRI_TM_WR_EX |
7868                             SPRI_TM_EX_AC_AR;
7869                 scsi_ulto2b(type_mask, res_cap->type_mask);
7870                 break;
7871         }
7872         case SPRI_RS: { // read full status
7873                 struct scsi_per_res_in_full *res_status;
7874                 struct scsi_per_res_in_full_desc *res_desc;
7875                 struct ctl_port *port;
7876                 int i, len;
7877
7878                 res_status = (struct scsi_per_res_in_full*)ctsio->kern_data_ptr;
7879
7880                 /*
7881                  * We had to drop the lock to allocate our buffer, which
7882                  * leaves time for someone to come in with another
7883                  * persistent reservation.  (That is unlikely, though,
7884                  * since this should be the only persistent reservation
7885                  * command active right now.)
7886                  */
7887                 if (total_len < (sizeof(struct scsi_per_res_in_header) +
7888                     (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7889                      lun->pr_key_count)){
7890                         mtx_unlock(&lun->lun_lock);
7891                         free(ctsio->kern_data_ptr, M_CTL);
7892                         printf("%s: reservation length changed, retrying\n",
7893                                __func__);
7894                         goto retry;
7895                 }
7896
7897                 scsi_ulto4b(lun->pr_generation, res_status->header.generation);
7898
7899                 res_desc = &res_status->desc[0];
7900                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
7901                         if ((key = ctl_get_prkey(lun, i)) == 0)
7902                                 continue;
7903
7904                         scsi_u64to8b(key, res_desc->res_key.key);
7905                         if ((lun->flags & CTL_LUN_PR_RESERVED) &&
7906                             (lun->pr_res_idx == i ||
7907                              lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS)) {
7908                                 res_desc->flags = SPRI_FULL_R_HOLDER;
7909                                 res_desc->scopetype = lun->pr_res_type;
7910                         }
7911                         scsi_ulto2b(i / CTL_MAX_INIT_PER_PORT,
7912                             res_desc->rel_trgt_port_id);
7913                         len = 0;
7914                         port = softc->ctl_ports[i / CTL_MAX_INIT_PER_PORT];
7915                         if (port != NULL)
7916                                 len = ctl_create_iid(port,
7917                                     i % CTL_MAX_INIT_PER_PORT,
7918                                     res_desc->transport_id);
7919                         scsi_ulto4b(len, res_desc->additional_length);
7920                         res_desc = (struct scsi_per_res_in_full_desc *)
7921                             &res_desc->transport_id[len];
7922                 }
7923                 scsi_ulto4b((uint8_t *)res_desc - (uint8_t *)&res_status->desc[0],
7924                     res_status->header.length);
7925                 break;
7926         }
7927         default:
7928                 panic("%s: Invalid PR type %#x", __func__, cdb->action);
7929         }
7930         mtx_unlock(&lun->lun_lock);
7931
7932         ctl_set_success(ctsio);
7933         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7934         ctsio->be_move_done = ctl_config_move_done;
7935         ctl_datamove((union ctl_io *)ctsio);
7936         return (CTL_RETVAL_COMPLETE);
7937 }
7938
7939 /*
7940  * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
7941  * it should return.
7942  */
7943 static int
7944 ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
7945                 uint64_t sa_res_key, uint8_t type, uint32_t residx,
7946                 struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
7947                 struct scsi_per_res_out_parms* param)
7948 {
7949         union ctl_ha_msg persis_io;
7950         int i;
7951
7952         mtx_lock(&lun->lun_lock);
7953         if (sa_res_key == 0) {
7954                 if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
7955                         /* validate scope and type */
7956                         if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7957                              SPR_LU_SCOPE) {
7958                                 mtx_unlock(&lun->lun_lock);
7959                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7960                                                       /*sks_valid*/ 1,
7961                                                       /*command*/ 1,
7962                                                       /*field*/ 2,
7963                                                       /*bit_valid*/ 1,
7964                                                       /*bit*/ 4);
7965                                 ctl_done((union ctl_io *)ctsio);
7966                                 return (1);
7967                         }
7968
7969                         if (type>8 || type==2 || type==4 || type==0) {
7970                                 mtx_unlock(&lun->lun_lock);
7971                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7972                                                       /*sks_valid*/ 1,
7973                                                       /*command*/ 1,
7974                                                       /*field*/ 2,
7975                                                       /*bit_valid*/ 1,
7976                                                       /*bit*/ 0);
7977                                 ctl_done((union ctl_io *)ctsio);
7978                                 return (1);
7979                         }
7980
7981                         /*
7982                          * Unregister everybody else and build UA for
7983                          * them
7984                          */
7985                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
7986                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
7987                                         continue;
7988
7989                                 ctl_clr_prkey(lun, i);
7990                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7991                         }
7992                         lun->pr_key_count = 1;
7993                         lun->pr_res_type = type;
7994                         if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
7995                             lun->pr_res_type != SPR_TYPE_EX_AC_AR)
7996                                 lun->pr_res_idx = residx;
7997                         lun->pr_generation++;
7998                         mtx_unlock(&lun->lun_lock);
7999
8000                         /* send msg to other side */
8001                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8002                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8003                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8004                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8005                         persis_io.pr.pr_info.res_type = type;
8006                         memcpy(persis_io.pr.pr_info.sa_res_key,
8007                                param->serv_act_res_key,
8008                                sizeof(param->serv_act_res_key));
8009                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8010                             sizeof(persis_io.pr), M_WAITOK);
8011                 } else {
8012                         /* not all registrants */
8013                         mtx_unlock(&lun->lun_lock);
8014                         free(ctsio->kern_data_ptr, M_CTL);
8015                         ctl_set_invalid_field(ctsio,
8016                                               /*sks_valid*/ 1,
8017                                               /*command*/ 0,
8018                                               /*field*/ 8,
8019                                               /*bit_valid*/ 0,
8020                                               /*bit*/ 0);
8021                         ctl_done((union ctl_io *)ctsio);
8022                         return (1);
8023                 }
8024         } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8025                 || !(lun->flags & CTL_LUN_PR_RESERVED)) {
8026                 int found = 0;
8027
8028                 if (res_key == sa_res_key) {
8029                         /* special case */
8030                         /*
8031                          * The spec implies this is not good but doesn't
8032                          * say what to do. There are two choices either
8033                          * generate a res conflict or check condition
8034                          * with illegal field in parameter data. Since
8035                          * that is what is done when the sa_res_key is
8036                          * zero I'll take that approach since this has
8037                          * to do with the sa_res_key.
8038                          */
8039                         mtx_unlock(&lun->lun_lock);
8040                         free(ctsio->kern_data_ptr, M_CTL);
8041                         ctl_set_invalid_field(ctsio,
8042                                               /*sks_valid*/ 1,
8043                                               /*command*/ 0,
8044                                               /*field*/ 8,
8045                                               /*bit_valid*/ 0,
8046                                               /*bit*/ 0);
8047                         ctl_done((union ctl_io *)ctsio);
8048                         return (1);
8049                 }
8050
8051                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8052                         if (ctl_get_prkey(lun, i) != sa_res_key)
8053                                 continue;
8054
8055                         found = 1;
8056                         ctl_clr_prkey(lun, i);
8057                         lun->pr_key_count--;
8058                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8059                 }
8060                 if (!found) {
8061                         mtx_unlock(&lun->lun_lock);
8062                         free(ctsio->kern_data_ptr, M_CTL);
8063                         ctl_set_reservation_conflict(ctsio);
8064                         ctl_done((union ctl_io *)ctsio);
8065                         return (CTL_RETVAL_COMPLETE);
8066                 }
8067                 lun->pr_generation++;
8068                 mtx_unlock(&lun->lun_lock);
8069
8070                 /* send msg to other side */
8071                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8072                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8073                 persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8074                 persis_io.pr.pr_info.residx = lun->pr_res_idx;
8075                 persis_io.pr.pr_info.res_type = type;
8076                 memcpy(persis_io.pr.pr_info.sa_res_key,
8077                        param->serv_act_res_key,
8078                        sizeof(param->serv_act_res_key));
8079                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8080                     sizeof(persis_io.pr), M_WAITOK);
8081         } else {
8082                 /* Reserved but not all registrants */
8083                 /* sa_res_key is res holder */
8084                 if (sa_res_key == ctl_get_prkey(lun, lun->pr_res_idx)) {
8085                         /* validate scope and type */
8086                         if ((cdb->scope_type & SPR_SCOPE_MASK) !=
8087                              SPR_LU_SCOPE) {
8088                                 mtx_unlock(&lun->lun_lock);
8089                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
8090                                                       /*sks_valid*/ 1,
8091                                                       /*command*/ 1,
8092                                                       /*field*/ 2,
8093                                                       /*bit_valid*/ 1,
8094                                                       /*bit*/ 4);
8095                                 ctl_done((union ctl_io *)ctsio);
8096                                 return (1);
8097                         }
8098
8099                         if (type>8 || type==2 || type==4 || type==0) {
8100                                 mtx_unlock(&lun->lun_lock);
8101                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
8102                                                       /*sks_valid*/ 1,
8103                                                       /*command*/ 1,
8104                                                       /*field*/ 2,
8105                                                       /*bit_valid*/ 1,
8106                                                       /*bit*/ 0);
8107                                 ctl_done((union ctl_io *)ctsio);
8108                                 return (1);
8109                         }
8110
8111                         /*
8112                          * Do the following:
8113                          * if sa_res_key != res_key remove all
8114                          * registrants w/sa_res_key and generate UA
8115                          * for these registrants(Registrations
8116                          * Preempted) if it wasn't an exclusive
8117                          * reservation generate UA(Reservations
8118                          * Preempted) for all other registered nexuses
8119                          * if the type has changed. Establish the new
8120                          * reservation and holder. If res_key and
8121                          * sa_res_key are the same do the above
8122                          * except don't unregister the res holder.
8123                          */
8124
8125                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
8126                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
8127                                         continue;
8128
8129                                 if (sa_res_key == ctl_get_prkey(lun, i)) {
8130                                         ctl_clr_prkey(lun, i);
8131                                         lun->pr_key_count--;
8132                                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8133                                 } else if (type != lun->pr_res_type &&
8134                                     (lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8135                                      lun->pr_res_type == SPR_TYPE_EX_AC_RO)) {
8136                                         ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8137                                 }
8138                         }
8139                         lun->pr_res_type = type;
8140                         if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
8141                             lun->pr_res_type != SPR_TYPE_EX_AC_AR)
8142                                 lun->pr_res_idx = residx;
8143                         else
8144                                 lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8145                         lun->pr_generation++;
8146                         mtx_unlock(&lun->lun_lock);
8147
8148                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8149                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8150                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8151                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8152                         persis_io.pr.pr_info.res_type = type;
8153                         memcpy(persis_io.pr.pr_info.sa_res_key,
8154                                param->serv_act_res_key,
8155                                sizeof(param->serv_act_res_key));
8156                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8157                             sizeof(persis_io.pr), M_WAITOK);
8158                 } else {
8159                         /*
8160                          * sa_res_key is not the res holder just
8161                          * remove registrants
8162                          */
8163                         int found=0;
8164
8165                         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8166                                 if (sa_res_key != ctl_get_prkey(lun, i))
8167                                         continue;
8168
8169                                 found = 1;
8170                                 ctl_clr_prkey(lun, i);
8171                                 lun->pr_key_count--;
8172                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8173                         }
8174
8175                         if (!found) {
8176                                 mtx_unlock(&lun->lun_lock);
8177                                 free(ctsio->kern_data_ptr, M_CTL);
8178                                 ctl_set_reservation_conflict(ctsio);
8179                                 ctl_done((union ctl_io *)ctsio);
8180                                 return (1);
8181                         }
8182                         lun->pr_generation++;
8183                         mtx_unlock(&lun->lun_lock);
8184
8185                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8186                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8187                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8188                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8189                         persis_io.pr.pr_info.res_type = type;
8190                         memcpy(persis_io.pr.pr_info.sa_res_key,
8191                                param->serv_act_res_key,
8192                                sizeof(param->serv_act_res_key));
8193                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8194                             sizeof(persis_io.pr), M_WAITOK);
8195                 }
8196         }
8197         return (0);
8198 }
8199
8200 static void
8201 ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
8202 {
8203         uint64_t sa_res_key;
8204         int i;
8205
8206         sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
8207
8208         if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8209          || lun->pr_res_idx == CTL_PR_NO_RESERVATION
8210          || sa_res_key != ctl_get_prkey(lun, lun->pr_res_idx)) {
8211                 if (sa_res_key == 0) {
8212                         /*
8213                          * Unregister everybody else and build UA for
8214                          * them
8215                          */
8216                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
8217                                 if (i == msg->pr.pr_info.residx ||
8218                                     ctl_get_prkey(lun, i) == 0)
8219                                         continue;
8220
8221                                 ctl_clr_prkey(lun, i);
8222                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8223                         }
8224
8225                         lun->pr_key_count = 1;
8226                         lun->pr_res_type = msg->pr.pr_info.res_type;
8227                         if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
8228                             lun->pr_res_type != SPR_TYPE_EX_AC_AR)
8229                                 lun->pr_res_idx = msg->pr.pr_info.residx;
8230                 } else {
8231                         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8232                                 if (sa_res_key == ctl_get_prkey(lun, i))
8233                                         continue;
8234
8235                                 ctl_clr_prkey(lun, i);
8236                                 lun->pr_key_count--;
8237                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8238                         }
8239                 }
8240         } else {
8241                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8242                         if (i == msg->pr.pr_info.residx ||
8243                             ctl_get_prkey(lun, i) == 0)
8244                                 continue;
8245
8246                         if (sa_res_key == ctl_get_prkey(lun, i)) {
8247                                 ctl_clr_prkey(lun, i);
8248                                 lun->pr_key_count--;
8249                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8250                         } else if (msg->pr.pr_info.res_type != lun->pr_res_type
8251                             && (lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8252                              lun->pr_res_type == SPR_TYPE_EX_AC_RO)) {
8253                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8254                         }
8255                 }
8256                 lun->pr_res_type = msg->pr.pr_info.res_type;
8257                 if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
8258                     lun->pr_res_type != SPR_TYPE_EX_AC_AR)
8259                         lun->pr_res_idx = msg->pr.pr_info.residx;
8260                 else
8261                         lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8262         }
8263         lun->pr_generation++;
8264
8265 }
8266
8267 int
8268 ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
8269 {
8270         struct ctl_softc *softc = CTL_SOFTC(ctsio);
8271         struct ctl_lun *lun = CTL_LUN(ctsio);
8272         int retval;
8273         uint32_t param_len;
8274         struct scsi_per_res_out *cdb;
8275         struct scsi_per_res_out_parms* param;
8276         uint32_t residx;
8277         uint64_t res_key, sa_res_key, key;
8278         uint8_t type;
8279         union ctl_ha_msg persis_io;
8280         int    i;
8281
8282         CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n"));
8283
8284         cdb = (struct scsi_per_res_out *)ctsio->cdb;
8285         retval = CTL_RETVAL_COMPLETE;
8286
8287         /*
8288          * We only support whole-LUN scope.  The scope & type are ignored for
8289          * register, register and ignore existing key and clear.
8290          * We sometimes ignore scope and type on preempts too!!
8291          * Verify reservation type here as well.
8292          */
8293         type = cdb->scope_type & SPR_TYPE_MASK;
8294         if ((cdb->action == SPRO_RESERVE)
8295          || (cdb->action == SPRO_RELEASE)) {
8296                 if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
8297                         ctl_set_invalid_field(/*ctsio*/ ctsio,
8298                                               /*sks_valid*/ 1,
8299                                               /*command*/ 1,
8300                                               /*field*/ 2,
8301                                               /*bit_valid*/ 1,
8302                                               /*bit*/ 4);
8303                         ctl_done((union ctl_io *)ctsio);
8304                         return (CTL_RETVAL_COMPLETE);
8305                 }
8306
8307                 if (type>8 || type==2 || type==4 || type==0) {
8308                         ctl_set_invalid_field(/*ctsio*/ ctsio,
8309                                               /*sks_valid*/ 1,
8310                                               /*command*/ 1,
8311                                               /*field*/ 2,
8312                                               /*bit_valid*/ 1,
8313                                               /*bit*/ 0);
8314                         ctl_done((union ctl_io *)ctsio);
8315                         return (CTL_RETVAL_COMPLETE);
8316                 }
8317         }
8318
8319         param_len = scsi_4btoul(cdb->length);
8320
8321         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
8322                 ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
8323                 ctsio->kern_data_len = param_len;
8324                 ctsio->kern_total_len = param_len;
8325                 ctsio->kern_rel_offset = 0;
8326                 ctsio->kern_sg_entries = 0;
8327                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
8328                 ctsio->be_move_done = ctl_config_move_done;
8329                 ctl_datamove((union ctl_io *)ctsio);
8330
8331                 return (CTL_RETVAL_COMPLETE);
8332         }
8333
8334         param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
8335
8336         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
8337         res_key = scsi_8btou64(param->res_key.key);
8338         sa_res_key = scsi_8btou64(param->serv_act_res_key);
8339
8340         /*
8341          * Validate the reservation key here except for SPRO_REG_IGNO
8342          * This must be done for all other service actions
8343          */
8344         if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
8345                 mtx_lock(&lun->lun_lock);
8346                 if ((key = ctl_get_prkey(lun, residx)) != 0) {
8347                         if (res_key != key) {
8348                                 /*
8349                                  * The current key passed in doesn't match
8350                                  * the one the initiator previously
8351                                  * registered.
8352                                  */
8353                                 mtx_unlock(&lun->lun_lock);
8354                                 free(ctsio->kern_data_ptr, M_CTL);
8355                                 ctl_set_reservation_conflict(ctsio);
8356                                 ctl_done((union ctl_io *)ctsio);
8357                                 return (CTL_RETVAL_COMPLETE);
8358                         }
8359                 } else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
8360                         /*
8361                          * We are not registered
8362                          */
8363                         mtx_unlock(&lun->lun_lock);
8364                         free(ctsio->kern_data_ptr, M_CTL);
8365                         ctl_set_reservation_conflict(ctsio);
8366                         ctl_done((union ctl_io *)ctsio);
8367                         return (CTL_RETVAL_COMPLETE);
8368                 } else if (res_key != 0) {
8369                         /*
8370                          * We are not registered and trying to register but
8371                          * the register key isn't zero.
8372                          */
8373                         mtx_unlock(&lun->lun_lock);
8374                         free(ctsio->kern_data_ptr, M_CTL);
8375                         ctl_set_reservation_conflict(ctsio);
8376                         ctl_done((union ctl_io *)ctsio);
8377                         return (CTL_RETVAL_COMPLETE);
8378                 }
8379                 mtx_unlock(&lun->lun_lock);
8380         }
8381
8382         switch (cdb->action & SPRO_ACTION_MASK) {
8383         case SPRO_REGISTER:
8384         case SPRO_REG_IGNO: {
8385                 /*
8386                  * We don't support any of these options, as we report in
8387                  * the read capabilities request (see
8388                  * ctl_persistent_reserve_in(), above).
8389                  */
8390                 if ((param->flags & SPR_SPEC_I_PT)
8391                  || (param->flags & SPR_ALL_TG_PT)
8392                  || (param->flags & SPR_APTPL)) {
8393                         int bit_ptr;
8394
8395                         if (param->flags & SPR_APTPL)
8396                                 bit_ptr = 0;
8397                         else if (param->flags & SPR_ALL_TG_PT)
8398                                 bit_ptr = 2;
8399                         else /* SPR_SPEC_I_PT */
8400                                 bit_ptr = 3;
8401
8402                         free(ctsio->kern_data_ptr, M_CTL);
8403                         ctl_set_invalid_field(ctsio,
8404                                               /*sks_valid*/ 1,
8405                                               /*command*/ 0,
8406                                               /*field*/ 20,
8407                                               /*bit_valid*/ 1,
8408                                               /*bit*/ bit_ptr);
8409                         ctl_done((union ctl_io *)ctsio);
8410                         return (CTL_RETVAL_COMPLETE);
8411                 }
8412
8413                 mtx_lock(&lun->lun_lock);
8414
8415                 /*
8416                  * The initiator wants to clear the
8417                  * key/unregister.
8418                  */
8419                 if (sa_res_key == 0) {
8420                         if ((res_key == 0
8421                           && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
8422                          || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
8423                           && ctl_get_prkey(lun, residx) == 0)) {
8424                                 mtx_unlock(&lun->lun_lock);
8425                                 goto done;
8426                         }
8427
8428                         ctl_clr_prkey(lun, residx);
8429                         lun->pr_key_count--;
8430
8431                         if (residx == lun->pr_res_idx) {
8432                                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8433                                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8434
8435                                 if ((lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8436                                      lun->pr_res_type == SPR_TYPE_EX_AC_RO) &&
8437                                     lun->pr_key_count) {
8438                                         /*
8439                                          * If the reservation is a registrants
8440                                          * only type we need to generate a UA
8441                                          * for other registered inits.  The
8442                                          * sense code should be RESERVATIONS
8443                                          * RELEASED
8444                                          */
8445
8446                                         for (i = softc->init_min; i < softc->init_max; i++){
8447                                                 if (ctl_get_prkey(lun, i) == 0)
8448                                                         continue;
8449                                                 ctl_est_ua(lun, i,
8450                                                     CTL_UA_RES_RELEASE);
8451                                         }
8452                                 }
8453                                 lun->pr_res_type = 0;
8454                         } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8455                                 if (lun->pr_key_count==0) {
8456                                         lun->flags &= ~CTL_LUN_PR_RESERVED;
8457                                         lun->pr_res_type = 0;
8458                                         lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8459                                 }
8460                         }
8461                         lun->pr_generation++;
8462                         mtx_unlock(&lun->lun_lock);
8463
8464                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8465                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8466                         persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
8467                         persis_io.pr.pr_info.residx = residx;
8468                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8469                             sizeof(persis_io.pr), M_WAITOK);
8470                 } else /* sa_res_key != 0 */ {
8471                         /*
8472                          * If we aren't registered currently then increment
8473                          * the key count and set the registered flag.
8474                          */
8475                         ctl_alloc_prkey(lun, residx);
8476                         if (ctl_get_prkey(lun, residx) == 0)
8477                                 lun->pr_key_count++;
8478                         ctl_set_prkey(lun, residx, sa_res_key);
8479                         lun->pr_generation++;
8480                         mtx_unlock(&lun->lun_lock);
8481
8482                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8483                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8484                         persis_io.pr.pr_info.action = CTL_PR_REG_KEY;
8485                         persis_io.pr.pr_info.residx = residx;
8486                         memcpy(persis_io.pr.pr_info.sa_res_key,
8487                                param->serv_act_res_key,
8488                                sizeof(param->serv_act_res_key));
8489                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8490                             sizeof(persis_io.pr), M_WAITOK);
8491                 }
8492
8493                 break;
8494         }
8495         case SPRO_RESERVE:
8496                 mtx_lock(&lun->lun_lock);
8497                 if (lun->flags & CTL_LUN_PR_RESERVED) {
8498                         /*
8499                          * if this isn't the reservation holder and it's
8500                          * not a "all registrants" type or if the type is
8501                          * different then we have a conflict
8502                          */
8503                         if ((lun->pr_res_idx != residx
8504                           && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8505                          || lun->pr_res_type != type) {
8506                                 mtx_unlock(&lun->lun_lock);
8507                                 free(ctsio->kern_data_ptr, M_CTL);
8508                                 ctl_set_reservation_conflict(ctsio);
8509                                 ctl_done((union ctl_io *)ctsio);
8510                                 return (CTL_RETVAL_COMPLETE);
8511                         }
8512                         mtx_unlock(&lun->lun_lock);
8513                 } else /* create a reservation */ {
8514                         /*
8515                          * If it's not an "all registrants" type record
8516                          * reservation holder
8517                          */
8518                         if (type != SPR_TYPE_WR_EX_AR
8519                          && type != SPR_TYPE_EX_AC_AR)
8520                                 lun->pr_res_idx = residx; /* Res holder */
8521                         else
8522                                 lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8523
8524                         lun->flags |= CTL_LUN_PR_RESERVED;
8525                         lun->pr_res_type = type;
8526
8527                         mtx_unlock(&lun->lun_lock);
8528
8529                         /* send msg to other side */
8530                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8531                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8532                         persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8533                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8534                         persis_io.pr.pr_info.res_type = type;
8535                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8536                             sizeof(persis_io.pr), M_WAITOK);
8537                 }
8538                 break;
8539
8540         case SPRO_RELEASE:
8541                 mtx_lock(&lun->lun_lock);
8542                 if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8543                         /* No reservation exists return good status */
8544                         mtx_unlock(&lun->lun_lock);
8545                         goto done;
8546                 }
8547                 /*
8548                  * Is this nexus a reservation holder?
8549                  */
8550                 if (lun->pr_res_idx != residx
8551                  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8552                         /*
8553                          * not a res holder return good status but
8554                          * do nothing
8555                          */
8556                         mtx_unlock(&lun->lun_lock);
8557                         goto done;
8558                 }
8559
8560                 if (lun->pr_res_type != type) {
8561                         mtx_unlock(&lun->lun_lock);
8562                         free(ctsio->kern_data_ptr, M_CTL);
8563                         ctl_set_illegal_pr_release(ctsio);
8564                         ctl_done((union ctl_io *)ctsio);
8565                         return (CTL_RETVAL_COMPLETE);
8566                 }
8567
8568                 /* okay to release */
8569                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8570                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8571                 lun->pr_res_type = 0;
8572
8573                 /*
8574                  * If this isn't an exclusive access reservation and NUAR
8575                  * is not set, generate UA for all other registrants.
8576                  */
8577                 if (type != SPR_TYPE_EX_AC && 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                 mtx_unlock(&lun->lun_lock);
8586
8587                 /* Send msg to other side */
8588                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8589                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8590                 persis_io.pr.pr_info.action = CTL_PR_RELEASE;
8591                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8592                      sizeof(persis_io.pr), M_WAITOK);
8593                 break;
8594
8595         case SPRO_CLEAR:
8596                 /* send msg to other side */
8597
8598                 mtx_lock(&lun->lun_lock);
8599                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8600                 lun->pr_res_type = 0;
8601                 lun->pr_key_count = 0;
8602                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8603
8604                 ctl_clr_prkey(lun, residx);
8605                 for (i = 0; i < CTL_MAX_INITIATORS; i++)
8606                         if (ctl_get_prkey(lun, i) != 0) {
8607                                 ctl_clr_prkey(lun, i);
8608                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8609                         }
8610                 lun->pr_generation++;
8611                 mtx_unlock(&lun->lun_lock);
8612
8613                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8614                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8615                 persis_io.pr.pr_info.action = CTL_PR_CLEAR;
8616                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8617                      sizeof(persis_io.pr), M_WAITOK);
8618                 break;
8619
8620         case SPRO_PREEMPT:
8621         case SPRO_PRE_ABO: {
8622                 int nretval;
8623
8624                 nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8625                                           residx, ctsio, cdb, param);
8626                 if (nretval != 0)
8627                         return (CTL_RETVAL_COMPLETE);
8628                 break;
8629         }
8630         default:
8631                 panic("%s: Invalid PR type %#x", __func__, cdb->action);
8632         }
8633
8634 done:
8635         free(ctsio->kern_data_ptr, M_CTL);
8636         ctl_set_success(ctsio);
8637         ctl_done((union ctl_io *)ctsio);
8638
8639         return (retval);
8640 }
8641
8642 /*
8643  * This routine is for handling a message from the other SC pertaining to
8644  * persistent reserve out. All the error checking will have been done
8645  * so only performing the action need be done here to keep the two
8646  * in sync.
8647  */
8648 static void
8649 ctl_hndl_per_res_out_on_other_sc(union ctl_io *io)
8650 {
8651         struct ctl_softc *softc = CTL_SOFTC(io);
8652         union ctl_ha_msg *msg = (union ctl_ha_msg *)&io->presio.pr_msg;
8653         struct ctl_lun *lun;
8654         int i;
8655         uint32_t residx, targ_lun;
8656
8657         targ_lun = msg->hdr.nexus.targ_mapped_lun;
8658         mtx_lock(&softc->ctl_lock);
8659         if (targ_lun >= ctl_max_luns ||
8660             (lun = softc->ctl_luns[targ_lun]) == NULL) {
8661                 mtx_unlock(&softc->ctl_lock);
8662                 return;
8663         }
8664         mtx_lock(&lun->lun_lock);
8665         mtx_unlock(&softc->ctl_lock);
8666         if (lun->flags & CTL_LUN_DISABLED) {
8667                 mtx_unlock(&lun->lun_lock);
8668                 return;
8669         }
8670         residx = ctl_get_initindex(&msg->hdr.nexus);
8671         switch(msg->pr.pr_info.action) {
8672         case CTL_PR_REG_KEY:
8673                 ctl_alloc_prkey(lun, msg->pr.pr_info.residx);
8674                 if (ctl_get_prkey(lun, msg->pr.pr_info.residx) == 0)
8675                         lun->pr_key_count++;
8676                 ctl_set_prkey(lun, msg->pr.pr_info.residx,
8677                     scsi_8btou64(msg->pr.pr_info.sa_res_key));
8678                 lun->pr_generation++;
8679                 break;
8680
8681         case CTL_PR_UNREG_KEY:
8682                 ctl_clr_prkey(lun, msg->pr.pr_info.residx);
8683                 lun->pr_key_count--;
8684
8685                 /* XXX Need to see if the reservation has been released */
8686                 /* if so do we need to generate UA? */
8687                 if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8688                         lun->flags &= ~CTL_LUN_PR_RESERVED;
8689                         lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8690
8691                         if ((lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8692                              lun->pr_res_type == SPR_TYPE_EX_AC_RO) &&
8693                             lun->pr_key_count) {
8694                                 /*
8695                                  * If the reservation is a registrants
8696                                  * only type we need to generate a UA
8697                                  * for other registered inits.  The
8698                                  * sense code should be RESERVATIONS
8699                                  * RELEASED
8700                                  */
8701
8702                                 for (i = softc->init_min; i < softc->init_max; i++) {
8703                                         if (ctl_get_prkey(lun, i) == 0)
8704                                                 continue;
8705
8706                                         ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8707                                 }
8708                         }
8709                         lun->pr_res_type = 0;
8710                 } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8711                         if (lun->pr_key_count==0) {
8712                                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8713                                 lun->pr_res_type = 0;
8714                                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8715                         }
8716                 }
8717                 lun->pr_generation++;
8718                 break;
8719
8720         case CTL_PR_RESERVE:
8721                 lun->flags |= CTL_LUN_PR_RESERVED;
8722                 lun->pr_res_type = msg->pr.pr_info.res_type;
8723                 lun->pr_res_idx = msg->pr.pr_info.residx;
8724
8725                 break;
8726
8727         case CTL_PR_RELEASE:
8728                 /*
8729                  * If this isn't an exclusive access reservation and NUAR
8730                  * is not set, generate UA for all other registrants.
8731                  */
8732                 if (lun->pr_res_type != SPR_TYPE_EX_AC &&
8733                     lun->pr_res_type != SPR_TYPE_WR_EX &&
8734                     (lun->MODE_CTRL.queue_flags & SCP_NUAR) == 0) {
8735                         for (i = softc->init_min; i < softc->init_max; i++) {
8736                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
8737                                         continue;
8738                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8739                         }
8740                 }
8741
8742                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8743                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8744                 lun->pr_res_type = 0;
8745                 break;
8746
8747         case CTL_PR_PREEMPT:
8748                 ctl_pro_preempt_other(lun, msg);
8749                 break;
8750         case CTL_PR_CLEAR:
8751                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8752                 lun->pr_res_type = 0;
8753                 lun->pr_key_count = 0;
8754                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8755
8756                 for (i=0; i < CTL_MAX_INITIATORS; i++) {
8757                         if (ctl_get_prkey(lun, i) == 0)
8758                                 continue;
8759                         ctl_clr_prkey(lun, i);
8760                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8761                 }
8762                 lun->pr_generation++;
8763                 break;
8764         }
8765
8766         mtx_unlock(&lun->lun_lock);
8767 }
8768
8769 int
8770 ctl_read_write(struct ctl_scsiio *ctsio)
8771 {
8772         struct ctl_lun *lun = CTL_LUN(ctsio);
8773         struct ctl_lba_len_flags *lbalen;
8774         uint64_t lba;
8775         uint32_t num_blocks;
8776         int flags, retval;
8777         int isread;
8778
8779         CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
8780
8781         flags = 0;
8782         isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
8783               || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
8784         switch (ctsio->cdb[0]) {
8785         case READ_6:
8786         case WRITE_6: {
8787                 struct scsi_rw_6 *cdb;
8788
8789                 cdb = (struct scsi_rw_6 *)ctsio->cdb;
8790
8791                 lba = scsi_3btoul(cdb->addr);
8792                 /* only 5 bits are valid in the most significant address byte */
8793                 lba &= 0x1fffff;
8794                 num_blocks = cdb->length;
8795                 /*
8796                  * This is correct according to SBC-2.
8797                  */
8798                 if (num_blocks == 0)
8799                         num_blocks = 256;
8800                 break;
8801         }
8802         case READ_10:
8803         case WRITE_10: {
8804                 struct scsi_rw_10 *cdb;
8805
8806                 cdb = (struct scsi_rw_10 *)ctsio->cdb;
8807                 if (cdb->byte2 & SRW10_FUA)
8808                         flags |= CTL_LLF_FUA;
8809                 if (cdb->byte2 & SRW10_DPO)
8810                         flags |= CTL_LLF_DPO;
8811                 lba = scsi_4btoul(cdb->addr);
8812                 num_blocks = scsi_2btoul(cdb->length);
8813                 break;
8814         }
8815         case WRITE_VERIFY_10: {
8816                 struct scsi_write_verify_10 *cdb;
8817
8818                 cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
8819                 flags |= CTL_LLF_FUA;
8820                 if (cdb->byte2 & SWV_DPO)
8821                         flags |= CTL_LLF_DPO;
8822                 lba = scsi_4btoul(cdb->addr);
8823                 num_blocks = scsi_2btoul(cdb->length);
8824                 break;
8825         }
8826         case READ_12:
8827         case WRITE_12: {
8828                 struct scsi_rw_12 *cdb;
8829
8830                 cdb = (struct scsi_rw_12 *)ctsio->cdb;
8831                 if (cdb->byte2 & SRW12_FUA)
8832                         flags |= CTL_LLF_FUA;
8833                 if (cdb->byte2 & SRW12_DPO)
8834                         flags |= CTL_LLF_DPO;
8835                 lba = scsi_4btoul(cdb->addr);
8836                 num_blocks = scsi_4btoul(cdb->length);
8837                 break;
8838         }
8839         case WRITE_VERIFY_12: {
8840                 struct scsi_write_verify_12 *cdb;
8841
8842                 cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
8843                 flags |= CTL_LLF_FUA;
8844                 if (cdb->byte2 & SWV_DPO)
8845                         flags |= CTL_LLF_DPO;
8846                 lba = scsi_4btoul(cdb->addr);
8847                 num_blocks = scsi_4btoul(cdb->length);
8848                 break;
8849         }
8850         case READ_16:
8851         case WRITE_16: {
8852                 struct scsi_rw_16 *cdb;
8853
8854                 cdb = (struct scsi_rw_16 *)ctsio->cdb;
8855                 if (cdb->byte2 & SRW12_FUA)
8856                         flags |= CTL_LLF_FUA;
8857                 if (cdb->byte2 & SRW12_DPO)
8858                         flags |= CTL_LLF_DPO;
8859                 lba = scsi_8btou64(cdb->addr);
8860                 num_blocks = scsi_4btoul(cdb->length);
8861                 break;
8862         }
8863         case WRITE_ATOMIC_16: {
8864                 struct scsi_write_atomic_16 *cdb;
8865
8866                 if (lun->be_lun->atomicblock == 0) {
8867                         ctl_set_invalid_opcode(ctsio);
8868                         ctl_done((union ctl_io *)ctsio);
8869                         return (CTL_RETVAL_COMPLETE);
8870                 }
8871
8872                 cdb = (struct scsi_write_atomic_16 *)ctsio->cdb;
8873                 if (cdb->byte2 & SRW12_FUA)
8874                         flags |= CTL_LLF_FUA;
8875                 if (cdb->byte2 & SRW12_DPO)
8876                         flags |= CTL_LLF_DPO;
8877                 lba = scsi_8btou64(cdb->addr);
8878                 num_blocks = scsi_2btoul(cdb->length);
8879                 if (num_blocks > lun->be_lun->atomicblock) {
8880                         ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
8881                             /*command*/ 1, /*field*/ 12, /*bit_valid*/ 0,
8882                             /*bit*/ 0);
8883                         ctl_done((union ctl_io *)ctsio);
8884                         return (CTL_RETVAL_COMPLETE);
8885                 }
8886                 break;
8887         }
8888         case WRITE_VERIFY_16: {
8889                 struct scsi_write_verify_16 *cdb;
8890
8891                 cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
8892                 flags |= CTL_LLF_FUA;
8893                 if (cdb->byte2 & SWV_DPO)
8894                         flags |= CTL_LLF_DPO;
8895                 lba = scsi_8btou64(cdb->addr);
8896                 num_blocks = scsi_4btoul(cdb->length);
8897                 break;
8898         }
8899         default:
8900                 /*
8901                  * We got a command we don't support.  This shouldn't
8902                  * happen, commands should be filtered out above us.
8903                  */
8904                 ctl_set_invalid_opcode(ctsio);
8905                 ctl_done((union ctl_io *)ctsio);
8906
8907                 return (CTL_RETVAL_COMPLETE);
8908                 break; /* NOTREACHED */
8909         }
8910
8911         /*
8912          * The first check is to make sure we're in bounds, the second
8913          * check is to catch wrap-around problems.  If the lba + num blocks
8914          * is less than the lba, then we've wrapped around and the block
8915          * range is invalid anyway.
8916          */
8917         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8918          || ((lba + num_blocks) < lba)) {
8919                 ctl_set_lba_out_of_range(ctsio,
8920                     MAX(lba, lun->be_lun->maxlba + 1));
8921                 ctl_done((union ctl_io *)ctsio);
8922                 return (CTL_RETVAL_COMPLETE);
8923         }
8924
8925         /*
8926          * According to SBC-3, a transfer length of 0 is not an error.
8927          * Note that this cannot happen with WRITE(6) or READ(6), since 0
8928          * translates to 256 blocks for those commands.
8929          */
8930         if (num_blocks == 0) {
8931                 ctl_set_success(ctsio);
8932                 ctl_done((union ctl_io *)ctsio);
8933                 return (CTL_RETVAL_COMPLETE);
8934         }
8935
8936         /* Set FUA and/or DPO if caches are disabled. */
8937         if (isread) {
8938                 if ((lun->MODE_CACHING.flags1 & SCP_RCD) != 0)
8939                         flags |= CTL_LLF_FUA | CTL_LLF_DPO;
8940         } else {
8941                 if ((lun->MODE_CACHING.flags1 & SCP_WCE) == 0)
8942                         flags |= CTL_LLF_FUA;
8943         }
8944
8945         lbalen = (struct ctl_lba_len_flags *)
8946             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8947         lbalen->lba = lba;
8948         lbalen->len = num_blocks;
8949         lbalen->flags = (isread ? CTL_LLF_READ : CTL_LLF_WRITE) | flags;
8950
8951         ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
8952         ctsio->kern_rel_offset = 0;
8953
8954         CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
8955
8956         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8957         return (retval);
8958 }
8959
8960 static int
8961 ctl_cnw_cont(union ctl_io *io)
8962 {
8963         struct ctl_lun *lun = CTL_LUN(io);
8964         struct ctl_scsiio *ctsio;
8965         struct ctl_lba_len_flags *lbalen;
8966         int retval;
8967
8968         ctsio = &io->scsiio;
8969         ctsio->io_hdr.status = CTL_STATUS_NONE;
8970         ctsio->io_hdr.flags &= ~CTL_FLAG_IO_CONT;
8971         lbalen = (struct ctl_lba_len_flags *)
8972             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8973         lbalen->flags &= ~CTL_LLF_COMPARE;
8974         lbalen->flags |= CTL_LLF_WRITE;
8975
8976         CTL_DEBUG_PRINT(("ctl_cnw_cont: calling data_submit()\n"));
8977         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8978         return (retval);
8979 }
8980
8981 int
8982 ctl_cnw(struct ctl_scsiio *ctsio)
8983 {
8984         struct ctl_lun *lun = CTL_LUN(ctsio);
8985         struct ctl_lba_len_flags *lbalen;
8986         uint64_t lba;
8987         uint32_t num_blocks;
8988         int flags, retval;
8989
8990         CTL_DEBUG_PRINT(("ctl_cnw: command: %#x\n", ctsio->cdb[0]));
8991
8992         flags = 0;
8993         switch (ctsio->cdb[0]) {
8994         case COMPARE_AND_WRITE: {
8995                 struct scsi_compare_and_write *cdb;
8996
8997                 cdb = (struct scsi_compare_and_write *)ctsio->cdb;
8998                 if (cdb->byte2 & SRW10_FUA)
8999                         flags |= CTL_LLF_FUA;
9000                 if (cdb->byte2 & SRW10_DPO)
9001                         flags |= CTL_LLF_DPO;
9002                 lba = scsi_8btou64(cdb->addr);
9003                 num_blocks = cdb->length;
9004                 break;
9005         }
9006         default:
9007                 /*
9008                  * We got a command we don't support.  This shouldn't
9009                  * happen, commands should be filtered out above us.
9010                  */
9011                 ctl_set_invalid_opcode(ctsio);
9012                 ctl_done((union ctl_io *)ctsio);
9013
9014                 return (CTL_RETVAL_COMPLETE);
9015                 break; /* NOTREACHED */
9016         }
9017
9018         /*
9019          * The first check is to make sure we're in bounds, the second
9020          * check is to catch wrap-around problems.  If the lba + num blocks
9021          * is less than the lba, then we've wrapped around and the block
9022          * range is invalid anyway.
9023          */
9024         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9025          || ((lba + num_blocks) < lba)) {
9026                 ctl_set_lba_out_of_range(ctsio,
9027                     MAX(lba, lun->be_lun->maxlba + 1));
9028                 ctl_done((union ctl_io *)ctsio);
9029                 return (CTL_RETVAL_COMPLETE);
9030         }
9031
9032         /*
9033          * According to SBC-3, a transfer length of 0 is not an error.
9034          */
9035         if (num_blocks == 0) {
9036                 ctl_set_success(ctsio);
9037                 ctl_done((union ctl_io *)ctsio);
9038                 return (CTL_RETVAL_COMPLETE);
9039         }
9040
9041         /* Set FUA if write cache is disabled. */
9042         if ((lun->MODE_CACHING.flags1 & SCP_WCE) == 0)
9043                 flags |= CTL_LLF_FUA;
9044
9045         ctsio->kern_total_len = 2 * num_blocks * lun->be_lun->blocksize;
9046         ctsio->kern_rel_offset = 0;
9047
9048         /*
9049          * Set the IO_CONT flag, so that if this I/O gets passed to
9050          * ctl_data_submit_done(), it'll get passed back to
9051          * ctl_ctl_cnw_cont() for further processing.
9052          */
9053         ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
9054         ctsio->io_cont = ctl_cnw_cont;
9055
9056         lbalen = (struct ctl_lba_len_flags *)
9057             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9058         lbalen->lba = lba;
9059         lbalen->len = num_blocks;
9060         lbalen->flags = CTL_LLF_COMPARE | flags;
9061
9062         CTL_DEBUG_PRINT(("ctl_cnw: calling data_submit()\n"));
9063         retval = lun->backend->data_submit((union ctl_io *)ctsio);
9064         return (retval);
9065 }
9066
9067 int
9068 ctl_verify(struct ctl_scsiio *ctsio)
9069 {
9070         struct ctl_lun *lun = CTL_LUN(ctsio);
9071         struct ctl_lba_len_flags *lbalen;
9072         uint64_t lba;
9073         uint32_t num_blocks;
9074         int bytchk, flags;
9075         int retval;
9076
9077         CTL_DEBUG_PRINT(("ctl_verify: command: %#x\n", ctsio->cdb[0]));
9078
9079         bytchk = 0;
9080         flags = CTL_LLF_FUA;
9081         switch (ctsio->cdb[0]) {
9082         case VERIFY_10: {
9083                 struct scsi_verify_10 *cdb;
9084
9085                 cdb = (struct scsi_verify_10 *)ctsio->cdb;
9086                 if (cdb->byte2 & SVFY_BYTCHK)
9087                         bytchk = 1;
9088                 if (cdb->byte2 & SVFY_DPO)
9089                         flags |= CTL_LLF_DPO;
9090                 lba = scsi_4btoul(cdb->addr);
9091                 num_blocks = scsi_2btoul(cdb->length);
9092                 break;
9093         }
9094         case VERIFY_12: {
9095                 struct scsi_verify_12 *cdb;
9096
9097                 cdb = (struct scsi_verify_12 *)ctsio->cdb;
9098                 if (cdb->byte2 & SVFY_BYTCHK)
9099                         bytchk = 1;
9100                 if (cdb->byte2 & SVFY_DPO)
9101                         flags |= CTL_LLF_DPO;
9102                 lba = scsi_4btoul(cdb->addr);
9103                 num_blocks = scsi_4btoul(cdb->length);
9104                 break;
9105         }
9106         case VERIFY_16: {
9107                 struct scsi_rw_16 *cdb;
9108
9109                 cdb = (struct scsi_rw_16 *)ctsio->cdb;
9110                 if (cdb->byte2 & SVFY_BYTCHK)
9111                         bytchk = 1;
9112                 if (cdb->byte2 & SVFY_DPO)
9113                         flags |= CTL_LLF_DPO;
9114                 lba = scsi_8btou64(cdb->addr);
9115                 num_blocks = scsi_4btoul(cdb->length);
9116                 break;
9117         }
9118         default:
9119                 /*
9120                  * We got a command we don't support.  This shouldn't
9121                  * happen, commands should be filtered out above us.
9122                  */
9123                 ctl_set_invalid_opcode(ctsio);
9124                 ctl_done((union ctl_io *)ctsio);
9125                 return (CTL_RETVAL_COMPLETE);
9126         }
9127
9128         /*
9129          * The first check is to make sure we're in bounds, the second
9130          * check is to catch wrap-around problems.  If the lba + num blocks
9131          * is less than the lba, then we've wrapped around and the block
9132          * range is invalid anyway.
9133          */
9134         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9135          || ((lba + num_blocks) < lba)) {
9136                 ctl_set_lba_out_of_range(ctsio,
9137                     MAX(lba, lun->be_lun->maxlba + 1));
9138                 ctl_done((union ctl_io *)ctsio);
9139                 return (CTL_RETVAL_COMPLETE);
9140         }
9141
9142         /*
9143          * According to SBC-3, a transfer length of 0 is not an error.
9144          */
9145         if (num_blocks == 0) {
9146                 ctl_set_success(ctsio);
9147                 ctl_done((union ctl_io *)ctsio);
9148                 return (CTL_RETVAL_COMPLETE);
9149         }
9150
9151         lbalen = (struct ctl_lba_len_flags *)
9152             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9153         lbalen->lba = lba;
9154         lbalen->len = num_blocks;
9155         if (bytchk) {
9156                 lbalen->flags = CTL_LLF_COMPARE | flags;
9157                 ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9158         } else {
9159                 lbalen->flags = CTL_LLF_VERIFY | flags;
9160                 ctsio->kern_total_len = 0;
9161         }
9162         ctsio->kern_rel_offset = 0;
9163
9164         CTL_DEBUG_PRINT(("ctl_verify: calling data_submit()\n"));
9165         retval = lun->backend->data_submit((union ctl_io *)ctsio);
9166         return (retval);
9167 }
9168
9169 int
9170 ctl_report_luns(struct ctl_scsiio *ctsio)
9171 {
9172         struct ctl_softc *softc = CTL_SOFTC(ctsio);
9173         struct ctl_port *port = CTL_PORT(ctsio);
9174         struct ctl_lun *lun, *request_lun = CTL_LUN(ctsio);
9175         struct scsi_report_luns *cdb;
9176         struct scsi_report_luns_data *lun_data;
9177         int num_filled, num_luns, num_port_luns, retval;
9178         uint32_t alloc_len, lun_datalen;
9179         uint32_t initidx, targ_lun_id, lun_id;
9180
9181         retval = CTL_RETVAL_COMPLETE;
9182         cdb = (struct scsi_report_luns *)ctsio->cdb;
9183
9184         CTL_DEBUG_PRINT(("ctl_report_luns\n"));
9185
9186         num_luns = 0;
9187         num_port_luns = port->lun_map ? port->lun_map_size : ctl_max_luns;
9188         mtx_lock(&softc->ctl_lock);
9189         for (targ_lun_id = 0; targ_lun_id < num_port_luns; targ_lun_id++) {
9190                 if (ctl_lun_map_from_port(port, targ_lun_id) != UINT32_MAX)
9191                         num_luns++;
9192         }
9193         mtx_unlock(&softc->ctl_lock);
9194
9195         switch (cdb->select_report) {
9196         case RPL_REPORT_DEFAULT:
9197         case RPL_REPORT_ALL:
9198         case RPL_REPORT_NONSUBSID:
9199                 break;
9200         case RPL_REPORT_WELLKNOWN:
9201         case RPL_REPORT_ADMIN:
9202         case RPL_REPORT_CONGLOM:
9203                 num_luns = 0;
9204                 break;
9205         default:
9206                 ctl_set_invalid_field(ctsio,
9207                                       /*sks_valid*/ 1,
9208                                       /*command*/ 1,
9209                                       /*field*/ 2,
9210                                       /*bit_valid*/ 0,
9211                                       /*bit*/ 0);
9212                 ctl_done((union ctl_io *)ctsio);
9213                 return (retval);
9214                 break; /* NOTREACHED */
9215         }
9216
9217         alloc_len = scsi_4btoul(cdb->length);
9218         /*
9219          * The initiator has to allocate at least 16 bytes for this request,
9220          * so he can at least get the header and the first LUN.  Otherwise
9221          * we reject the request (per SPC-3 rev 14, section 6.21).
9222          */
9223         if (alloc_len < (sizeof(struct scsi_report_luns_data) +
9224             sizeof(struct scsi_report_luns_lundata))) {
9225                 ctl_set_invalid_field(ctsio,
9226                                       /*sks_valid*/ 1,
9227                                       /*command*/ 1,
9228                                       /*field*/ 6,
9229                                       /*bit_valid*/ 0,
9230                                       /*bit*/ 0);
9231                 ctl_done((union ctl_io *)ctsio);
9232                 return (retval);
9233         }
9234
9235         lun_datalen = sizeof(*lun_data) +
9236                 (num_luns * sizeof(struct scsi_report_luns_lundata));
9237
9238         ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO);
9239         lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
9240         ctsio->kern_sg_entries = 0;
9241
9242         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9243
9244         mtx_lock(&softc->ctl_lock);
9245         for (targ_lun_id = 0, num_filled = 0;
9246             targ_lun_id < num_port_luns && num_filled < num_luns;
9247             targ_lun_id++) {
9248                 lun_id = ctl_lun_map_from_port(port, targ_lun_id);
9249                 if (lun_id == UINT32_MAX)
9250                         continue;
9251                 lun = softc->ctl_luns[lun_id];
9252                 if (lun == NULL)
9253                         continue;
9254
9255                 be64enc(lun_data->luns[num_filled++].lundata,
9256                     ctl_encode_lun(targ_lun_id));
9257
9258                 /*
9259                  * According to SPC-3, rev 14 section 6.21:
9260                  *
9261                  * "The execution of a REPORT LUNS command to any valid and
9262                  * installed logical unit shall clear the REPORTED LUNS DATA
9263                  * HAS CHANGED unit attention condition for all logical
9264                  * units of that target with respect to the requesting
9265                  * initiator. A valid and installed logical unit is one
9266                  * having a PERIPHERAL QUALIFIER of 000b in the standard
9267                  * INQUIRY data (see 6.4.2)."
9268                  *
9269                  * If request_lun is NULL, the LUN this report luns command
9270                  * was issued to is either disabled or doesn't exist. In that
9271                  * case, we shouldn't clear any pending lun change unit
9272                  * attention.
9273                  */
9274                 if (request_lun != NULL) {
9275                         mtx_lock(&lun->lun_lock);
9276                         ctl_clr_ua(lun, initidx, CTL_UA_LUN_CHANGE);
9277                         mtx_unlock(&lun->lun_lock);
9278                 }
9279         }
9280         mtx_unlock(&softc->ctl_lock);
9281
9282         /*
9283          * It's quite possible that we've returned fewer LUNs than we allocated
9284          * space for.  Trim it.
9285          */
9286         lun_datalen = sizeof(*lun_data) +
9287                 (num_filled * sizeof(struct scsi_report_luns_lundata));
9288         ctsio->kern_rel_offset = 0;
9289         ctsio->kern_sg_entries = 0;
9290         ctsio->kern_data_len = min(lun_datalen, alloc_len);
9291         ctsio->kern_total_len = ctsio->kern_data_len;
9292
9293         /*
9294          * We set this to the actual data length, regardless of how much
9295          * space we actually have to return results.  If the user looks at
9296          * this value, he'll know whether or not he allocated enough space
9297          * and reissue the command if necessary.  We don't support well
9298          * known logical units, so if the user asks for that, return none.
9299          */
9300         scsi_ulto4b(lun_datalen - 8, lun_data->length);
9301
9302         /*
9303          * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
9304          * this request.
9305          */
9306         ctl_set_success(ctsio);
9307         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9308         ctsio->be_move_done = ctl_config_move_done;
9309         ctl_datamove((union ctl_io *)ctsio);
9310         return (retval);
9311 }
9312
9313 int
9314 ctl_request_sense(struct ctl_scsiio *ctsio)
9315 {
9316         struct ctl_softc *softc = CTL_SOFTC(ctsio);
9317         struct ctl_lun *lun = CTL_LUN(ctsio);
9318         struct scsi_request_sense *cdb;
9319         struct scsi_sense_data *sense_ptr, *ps;
9320         uint32_t initidx;
9321         int have_error;
9322         u_int sense_len = SSD_FULL_SIZE;
9323         scsi_sense_data_type sense_format;
9324         ctl_ua_type ua_type;
9325         uint8_t asc = 0, ascq = 0;
9326
9327         cdb = (struct scsi_request_sense *)ctsio->cdb;
9328
9329         CTL_DEBUG_PRINT(("ctl_request_sense\n"));
9330
9331         /*
9332          * Determine which sense format the user wants.
9333          */
9334         if (cdb->byte2 & SRS_DESC)
9335                 sense_format = SSD_TYPE_DESC;
9336         else
9337                 sense_format = SSD_TYPE_FIXED;
9338
9339         ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
9340         sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
9341         ctsio->kern_sg_entries = 0;
9342         ctsio->kern_rel_offset = 0;
9343
9344         /*
9345          * struct scsi_sense_data, which is currently set to 256 bytes, is
9346          * larger than the largest allowed value for the length field in the
9347          * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
9348          */
9349         ctsio->kern_data_len = cdb->length;
9350         ctsio->kern_total_len = cdb->length;
9351
9352         /*
9353          * If we don't have a LUN, we don't have any pending sense.
9354          */
9355         if (lun == NULL ||
9356             ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
9357              softc->ha_link < CTL_HA_LINK_UNKNOWN)) {
9358                 /* "Logical unit not supported" */
9359                 ctl_set_sense_data(sense_ptr, &sense_len, NULL, sense_format,
9360                     /*current_error*/ 1,
9361                     /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
9362                     /*asc*/ 0x25,
9363                     /*ascq*/ 0x00,
9364                     SSD_ELEM_NONE);
9365                 goto send;
9366         }
9367
9368         have_error = 0;
9369         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9370         /*
9371          * Check for pending sense, and then for pending unit attentions.
9372          * Pending sense gets returned first, then pending unit attentions.
9373          */
9374         mtx_lock(&lun->lun_lock);
9375         ps = lun->pending_sense[initidx / CTL_MAX_INIT_PER_PORT];
9376         if (ps != NULL)
9377                 ps += initidx % CTL_MAX_INIT_PER_PORT;
9378         if (ps != NULL && ps->error_code != 0) {
9379                 scsi_sense_data_type stored_format;
9380
9381                 /*
9382                  * Check to see which sense format was used for the stored
9383                  * sense data.
9384                  */
9385                 stored_format = scsi_sense_type(ps);
9386
9387                 /*
9388                  * If the user requested a different sense format than the
9389                  * one we stored, then we need to convert it to the other
9390                  * format.  If we're going from descriptor to fixed format
9391                  * sense data, we may lose things in translation, depending
9392                  * on what options were used.
9393                  *
9394                  * If the stored format is SSD_TYPE_NONE (i.e. invalid),
9395                  * for some reason we'll just copy it out as-is.
9396                  */
9397                 if ((stored_format == SSD_TYPE_FIXED)
9398                  && (sense_format == SSD_TYPE_DESC))
9399                         ctl_sense_to_desc((struct scsi_sense_data_fixed *)
9400                             ps, (struct scsi_sense_data_desc *)sense_ptr);
9401                 else if ((stored_format == SSD_TYPE_DESC)
9402                       && (sense_format == SSD_TYPE_FIXED))
9403                         ctl_sense_to_fixed((struct scsi_sense_data_desc *)
9404                             ps, (struct scsi_sense_data_fixed *)sense_ptr);
9405                 else
9406                         memcpy(sense_ptr, ps, sizeof(*sense_ptr));
9407
9408                 ps->error_code = 0;
9409                 have_error = 1;
9410         } else {
9411                 ua_type = ctl_build_ua(lun, initidx, sense_ptr, &sense_len,
9412                     sense_format);
9413                 if (ua_type != CTL_UA_NONE)
9414                         have_error = 1;
9415         }
9416         if (have_error == 0) {
9417                 /*
9418                  * Report informational exception if have one and allowed.
9419                  */
9420                 if (lun->MODE_IE.mrie != SIEP_MRIE_NO) {
9421                         asc = lun->ie_asc;
9422                         ascq = lun->ie_ascq;
9423                 }
9424                 ctl_set_sense_data(sense_ptr, &sense_len, lun, sense_format,
9425                     /*current_error*/ 1,
9426                     /*sense_key*/ SSD_KEY_NO_SENSE,
9427                     /*asc*/ asc,
9428                     /*ascq*/ ascq,
9429                     SSD_ELEM_NONE);
9430         }
9431         mtx_unlock(&lun->lun_lock);
9432
9433 send:
9434         /*
9435          * We report the SCSI status as OK, since the status of the command
9436          * itself is OK.  We're reporting sense as parameter data.
9437          */
9438         ctl_set_success(ctsio);
9439         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9440         ctsio->be_move_done = ctl_config_move_done;
9441         ctl_datamove((union ctl_io *)ctsio);
9442         return (CTL_RETVAL_COMPLETE);
9443 }
9444
9445 int
9446 ctl_tur(struct ctl_scsiio *ctsio)
9447 {
9448
9449         CTL_DEBUG_PRINT(("ctl_tur\n"));
9450
9451         ctl_set_success(ctsio);
9452         ctl_done((union ctl_io *)ctsio);
9453
9454         return (CTL_RETVAL_COMPLETE);
9455 }
9456
9457 /*
9458  * SCSI VPD page 0x00, the Supported VPD Pages page.
9459  */
9460 static int
9461 ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
9462 {
9463         struct ctl_lun *lun = CTL_LUN(ctsio);
9464         struct scsi_vpd_supported_pages *pages;
9465         int sup_page_size;
9466         int p;
9467
9468         sup_page_size = sizeof(struct scsi_vpd_supported_pages) *
9469             SCSI_EVPD_NUM_SUPPORTED_PAGES;
9470         ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO);
9471         pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
9472         ctsio->kern_rel_offset = 0;
9473         ctsio->kern_sg_entries = 0;
9474         ctsio->kern_data_len = min(sup_page_size, alloc_len);
9475         ctsio->kern_total_len = ctsio->kern_data_len;
9476
9477         /*
9478          * The control device is always connected.  The disk device, on the
9479          * other hand, may not be online all the time.  Need to change this
9480          * to figure out whether the disk device is actually online or not.
9481          */
9482         if (lun != NULL)
9483                 pages->device = (SID_QUAL_LU_CONNECTED << 5) |
9484                                 lun->be_lun->lun_type;
9485         else
9486                 pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9487
9488         p = 0;
9489         /* Supported VPD pages */
9490         pages->page_list[p++] = SVPD_SUPPORTED_PAGES;
9491         /* Serial Number */
9492         pages->page_list[p++] = SVPD_UNIT_SERIAL_NUMBER;
9493         /* Device Identification */
9494         pages->page_list[p++] = SVPD_DEVICE_ID;
9495         /* Extended INQUIRY Data */
9496         pages->page_list[p++] = SVPD_EXTENDED_INQUIRY_DATA;
9497         /* Mode Page Policy */
9498         pages->page_list[p++] = SVPD_MODE_PAGE_POLICY;
9499         /* SCSI Ports */
9500         pages->page_list[p++] = SVPD_SCSI_PORTS;
9501         /* Third-party Copy */
9502         pages->page_list[p++] = SVPD_SCSI_TPC;
9503         /* SCSI Feature Sets */
9504         pages->page_list[p++] = SVPD_SCSI_SFS;
9505         if (lun != NULL && lun->be_lun->lun_type == T_DIRECT) {
9506                 /* Block limits */
9507                 pages->page_list[p++] = SVPD_BLOCK_LIMITS;
9508                 /* Block Device Characteristics */
9509                 pages->page_list[p++] = SVPD_BDC;
9510                 /* Logical Block Provisioning */
9511                 pages->page_list[p++] = SVPD_LBP;
9512         }
9513         pages->length = p;
9514
9515         ctl_set_success(ctsio);
9516         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9517         ctsio->be_move_done = ctl_config_move_done;
9518         ctl_datamove((union ctl_io *)ctsio);
9519         return (CTL_RETVAL_COMPLETE);
9520 }
9521
9522 /*
9523  * SCSI VPD page 0x80, the Unit Serial Number page.
9524  */
9525 static int
9526 ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
9527 {
9528         struct ctl_lun *lun = CTL_LUN(ctsio);
9529         struct scsi_vpd_unit_serial_number *sn_ptr;
9530         int data_len;
9531
9532         data_len = 4 + CTL_SN_LEN;
9533         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9534         sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
9535         ctsio->kern_rel_offset = 0;
9536         ctsio->kern_sg_entries = 0;
9537         ctsio->kern_data_len = min(data_len, alloc_len);
9538         ctsio->kern_total_len = ctsio->kern_data_len;
9539
9540         /*
9541          * The control device is always connected.  The disk device, on the
9542          * other hand, may not be online all the time.  Need to change this
9543          * to figure out whether the disk device is actually online or not.
9544          */
9545         if (lun != NULL)
9546                 sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9547                                   lun->be_lun->lun_type;
9548         else
9549                 sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9550
9551         sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9552         sn_ptr->length = CTL_SN_LEN;
9553         /*
9554          * If we don't have a LUN, we just leave the serial number as
9555          * all spaces.
9556          */
9557         if (lun != NULL) {
9558                 strncpy((char *)sn_ptr->serial_num,
9559                         (char *)lun->be_lun->serial_num, CTL_SN_LEN);
9560         } else
9561                 memset(sn_ptr->serial_num, 0x20, CTL_SN_LEN);
9562
9563         ctl_set_success(ctsio);
9564         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9565         ctsio->be_move_done = ctl_config_move_done;
9566         ctl_datamove((union ctl_io *)ctsio);
9567         return (CTL_RETVAL_COMPLETE);
9568 }
9569
9570 /*
9571  * SCSI VPD page 0x86, the Extended INQUIRY Data page.
9572  */
9573 static int
9574 ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len)
9575 {
9576         struct ctl_lun *lun = CTL_LUN(ctsio);
9577         struct scsi_vpd_extended_inquiry_data *eid_ptr;
9578         int data_len;
9579
9580         data_len = sizeof(struct scsi_vpd_extended_inquiry_data);
9581         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9582         eid_ptr = (struct scsi_vpd_extended_inquiry_data *)ctsio->kern_data_ptr;
9583         ctsio->kern_sg_entries = 0;
9584         ctsio->kern_rel_offset = 0;
9585         ctsio->kern_data_len = min(data_len, alloc_len);
9586         ctsio->kern_total_len = ctsio->kern_data_len;
9587
9588         /*
9589          * The control device is always connected.  The disk device, on the
9590          * other hand, may not be online all the time.
9591          */
9592         if (lun != NULL)
9593                 eid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9594                                      lun->be_lun->lun_type;
9595         else
9596                 eid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9597         eid_ptr->page_code = SVPD_EXTENDED_INQUIRY_DATA;
9598         scsi_ulto2b(data_len - 4, eid_ptr->page_length);
9599         /*
9600          * We support head of queue, ordered and simple tags.
9601          */
9602         eid_ptr->flags2 = SVPD_EID_HEADSUP | SVPD_EID_ORDSUP | SVPD_EID_SIMPSUP;
9603         /*
9604          * Volatile cache supported.
9605          */
9606         eid_ptr->flags3 = SVPD_EID_V_SUP;
9607
9608         /*
9609          * This means that we clear the REPORTED LUNS DATA HAS CHANGED unit
9610          * attention for a particular IT nexus on all LUNs once we report
9611          * it to that nexus once.  This bit is required as of SPC-4.
9612          */
9613         eid_ptr->flags4 = SVPD_EID_LUICLR;
9614
9615         /*
9616          * We support revert to defaults (RTD) bit in MODE SELECT.
9617          */
9618         eid_ptr->flags5 = SVPD_EID_RTD_SUP;
9619
9620         /*
9621          * XXX KDM in order to correctly answer this, we would need
9622          * information from the SIM to determine how much sense data it
9623          * can send.  So this would really be a path inquiry field, most
9624          * likely.  This can be set to a maximum of 252 according to SPC-4,
9625          * but the hardware may or may not be able to support that much.
9626          * 0 just means that the maximum sense data length is not reported.
9627          */
9628         eid_ptr->max_sense_length = 0;
9629
9630         ctl_set_success(ctsio);
9631         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9632         ctsio->be_move_done = ctl_config_move_done;
9633         ctl_datamove((union ctl_io *)ctsio);
9634         return (CTL_RETVAL_COMPLETE);
9635 }
9636
9637 static int
9638 ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len)
9639 {
9640         struct ctl_lun *lun = CTL_LUN(ctsio);
9641         struct scsi_vpd_mode_page_policy *mpp_ptr;
9642         int data_len;
9643
9644         data_len = sizeof(struct scsi_vpd_mode_page_policy) +
9645             sizeof(struct scsi_vpd_mode_page_policy_descr);
9646
9647         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9648         mpp_ptr = (struct scsi_vpd_mode_page_policy *)ctsio->kern_data_ptr;
9649         ctsio->kern_rel_offset = 0;
9650         ctsio->kern_sg_entries = 0;
9651         ctsio->kern_data_len = min(data_len, alloc_len);
9652         ctsio->kern_total_len = ctsio->kern_data_len;
9653
9654         /*
9655          * The control device is always connected.  The disk device, on the
9656          * other hand, may not be online all the time.
9657          */
9658         if (lun != NULL)
9659                 mpp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9660                                      lun->be_lun->lun_type;
9661         else
9662                 mpp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9663         mpp_ptr->page_code = SVPD_MODE_PAGE_POLICY;
9664         scsi_ulto2b(data_len - 4, mpp_ptr->page_length);
9665         mpp_ptr->descr[0].page_code = 0x3f;
9666         mpp_ptr->descr[0].subpage_code = 0xff;
9667         mpp_ptr->descr[0].policy = SVPD_MPP_SHARED;
9668
9669         ctl_set_success(ctsio);
9670         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9671         ctsio->be_move_done = ctl_config_move_done;
9672         ctl_datamove((union ctl_io *)ctsio);
9673         return (CTL_RETVAL_COMPLETE);
9674 }
9675
9676 /*
9677  * SCSI VPD page 0x83, the Device Identification page.
9678  */
9679 static int
9680 ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
9681 {
9682         struct ctl_softc *softc = CTL_SOFTC(ctsio);
9683         struct ctl_port *port = CTL_PORT(ctsio);
9684         struct ctl_lun *lun = CTL_LUN(ctsio);
9685         struct scsi_vpd_device_id *devid_ptr;
9686         struct scsi_vpd_id_descriptor *desc;
9687         int data_len, g;
9688         uint8_t proto;
9689
9690         data_len = sizeof(struct scsi_vpd_device_id) +
9691             sizeof(struct scsi_vpd_id_descriptor) +
9692                 sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
9693             sizeof(struct scsi_vpd_id_descriptor) +
9694                 sizeof(struct scsi_vpd_id_trgt_port_grp_id);
9695         if (lun && lun->lun_devid)
9696                 data_len += lun->lun_devid->len;
9697         if (port && port->port_devid)
9698                 data_len += port->port_devid->len;
9699         if (port && port->target_devid)
9700                 data_len += port->target_devid->len;
9701
9702         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9703         devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
9704         ctsio->kern_sg_entries = 0;
9705         ctsio->kern_rel_offset = 0;
9706         ctsio->kern_sg_entries = 0;
9707         ctsio->kern_data_len = min(data_len, alloc_len);
9708         ctsio->kern_total_len = ctsio->kern_data_len;
9709
9710         /*
9711          * The control device is always connected.  The disk device, on the
9712          * other hand, may not be online all the time.
9713          */
9714         if (lun != NULL)
9715                 devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9716                                      lun->be_lun->lun_type;
9717         else
9718                 devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9719         devid_ptr->page_code = SVPD_DEVICE_ID;
9720         scsi_ulto2b(data_len - 4, devid_ptr->length);
9721
9722         if (port && port->port_type == CTL_PORT_FC)
9723                 proto = SCSI_PROTO_FC << 4;
9724         else if (port && port->port_type == CTL_PORT_SAS)
9725                 proto = SCSI_PROTO_SAS << 4;
9726         else if (port && port->port_type == CTL_PORT_ISCSI)
9727                 proto = SCSI_PROTO_ISCSI << 4;
9728         else
9729                 proto = SCSI_PROTO_SPI << 4;
9730         desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
9731
9732         /*
9733          * We're using a LUN association here.  i.e., this device ID is a
9734          * per-LUN identifier.
9735          */
9736         if (lun && lun->lun_devid) {
9737                 memcpy(desc, lun->lun_devid->data, lun->lun_devid->len);
9738                 desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9739                     lun->lun_devid->len);
9740         }
9741
9742         /*
9743          * This is for the WWPN which is a port association.
9744          */
9745         if (port && port->port_devid) {
9746                 memcpy(desc, port->port_devid->data, port->port_devid->len);
9747                 desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9748                     port->port_devid->len);
9749         }
9750
9751         /*
9752          * This is for the Relative Target Port(type 4h) identifier
9753          */
9754         desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9755         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9756             SVPD_ID_TYPE_RELTARG;
9757         desc->length = 4;
9758         scsi_ulto2b(ctsio->io_hdr.nexus.targ_port, &desc->identifier[2]);
9759         desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9760             sizeof(struct scsi_vpd_id_rel_trgt_port_id));
9761
9762         /*
9763          * This is for the Target Port Group(type 5h) identifier
9764          */
9765         desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9766         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9767             SVPD_ID_TYPE_TPORTGRP;
9768         desc->length = 4;
9769         if (softc->is_single ||
9770             (port && port->status & CTL_PORT_STATUS_HA_SHARED))
9771                 g = 1;
9772         else
9773                 g = 2 + ctsio->io_hdr.nexus.targ_port / softc->port_cnt;
9774         scsi_ulto2b(g, &desc->identifier[2]);
9775         desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9776             sizeof(struct scsi_vpd_id_trgt_port_grp_id));
9777
9778         /*
9779          * This is for the Target identifier
9780          */
9781         if (port && port->target_devid) {
9782                 memcpy(desc, port->target_devid->data, port->target_devid->len);
9783         }
9784
9785         ctl_set_success(ctsio);
9786         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9787         ctsio->be_move_done = ctl_config_move_done;
9788         ctl_datamove((union ctl_io *)ctsio);
9789         return (CTL_RETVAL_COMPLETE);
9790 }
9791
9792 static int
9793 ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio, int alloc_len)
9794 {
9795         struct ctl_softc *softc = CTL_SOFTC(ctsio);
9796         struct ctl_lun *lun = CTL_LUN(ctsio);
9797         struct scsi_vpd_scsi_ports *sp;
9798         struct scsi_vpd_port_designation *pd;
9799         struct scsi_vpd_port_designation_cont *pdc;
9800         struct ctl_port *port;
9801         int data_len, num_target_ports, iid_len, id_len;
9802
9803         num_target_ports = 0;
9804         iid_len = 0;
9805         id_len = 0;
9806         mtx_lock(&softc->ctl_lock);
9807         STAILQ_FOREACH(port, &softc->port_list, links) {
9808                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
9809                         continue;
9810                 if (lun != NULL &&
9811                     ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
9812                         continue;
9813                 num_target_ports++;
9814                 if (port->init_devid)
9815                         iid_len += port->init_devid->len;
9816                 if (port->port_devid)
9817                         id_len += port->port_devid->len;
9818         }
9819         mtx_unlock(&softc->ctl_lock);
9820
9821         data_len = sizeof(struct scsi_vpd_scsi_ports) +
9822             num_target_ports * (sizeof(struct scsi_vpd_port_designation) +
9823              sizeof(struct scsi_vpd_port_designation_cont)) + iid_len + id_len;
9824         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9825         sp = (struct scsi_vpd_scsi_ports *)ctsio->kern_data_ptr;
9826         ctsio->kern_sg_entries = 0;
9827         ctsio->kern_rel_offset = 0;
9828         ctsio->kern_sg_entries = 0;
9829         ctsio->kern_data_len = min(data_len, alloc_len);
9830         ctsio->kern_total_len = ctsio->kern_data_len;
9831
9832         /*
9833          * The control device is always connected.  The disk device, on the
9834          * other hand, may not be online all the time.  Need to change this
9835          * to figure out whether the disk device is actually online or not.
9836          */
9837         if (lun != NULL)
9838                 sp->device = (SID_QUAL_LU_CONNECTED << 5) |
9839                                   lun->be_lun->lun_type;
9840         else
9841                 sp->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9842
9843         sp->page_code = SVPD_SCSI_PORTS;
9844         scsi_ulto2b(data_len - sizeof(struct scsi_vpd_scsi_ports),
9845             sp->page_length);
9846         pd = &sp->design[0];
9847
9848         mtx_lock(&softc->ctl_lock);
9849         STAILQ_FOREACH(port, &softc->port_list, links) {
9850                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
9851                         continue;
9852                 if (lun != NULL &&
9853                     ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
9854                         continue;
9855                 scsi_ulto2b(port->targ_port, pd->relative_port_id);
9856                 if (port->init_devid) {
9857                         iid_len = port->init_devid->len;
9858                         memcpy(pd->initiator_transportid,
9859                             port->init_devid->data, port->init_devid->len);
9860                 } else
9861                         iid_len = 0;
9862                 scsi_ulto2b(iid_len, pd->initiator_transportid_length);
9863                 pdc = (struct scsi_vpd_port_designation_cont *)
9864                     (&pd->initiator_transportid[iid_len]);
9865                 if (port->port_devid) {
9866                         id_len = port->port_devid->len;
9867                         memcpy(pdc->target_port_descriptors,
9868                             port->port_devid->data, port->port_devid->len);
9869                 } else
9870                         id_len = 0;
9871                 scsi_ulto2b(id_len, pdc->target_port_descriptors_length);
9872                 pd = (struct scsi_vpd_port_designation *)
9873                     ((uint8_t *)pdc->target_port_descriptors + id_len);
9874         }
9875         mtx_unlock(&softc->ctl_lock);
9876
9877         ctl_set_success(ctsio);
9878         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9879         ctsio->be_move_done = ctl_config_move_done;
9880         ctl_datamove((union ctl_io *)ctsio);
9881         return (CTL_RETVAL_COMPLETE);
9882 }
9883
9884 static int
9885 ctl_inquiry_evpd_sfs(struct ctl_scsiio *ctsio, int alloc_len)
9886 {
9887         struct ctl_lun *lun = CTL_LUN(ctsio);
9888         struct scsi_vpd_sfs *sfs_ptr;
9889         int sfs_page_size, n;
9890
9891         sfs_page_size = sizeof(*sfs_ptr) + 5 * 2;
9892         ctsio->kern_data_ptr = malloc(sfs_page_size, M_CTL, M_WAITOK | M_ZERO);
9893         sfs_ptr = (struct scsi_vpd_sfs *)ctsio->kern_data_ptr;
9894         ctsio->kern_sg_entries = 0;
9895         ctsio->kern_rel_offset = 0;
9896         ctsio->kern_sg_entries = 0;
9897         ctsio->kern_data_len = min(sfs_page_size, alloc_len);
9898         ctsio->kern_total_len = ctsio->kern_data_len;
9899
9900         /*
9901          * The control device is always connected.  The disk device, on the
9902          * other hand, may not be online all the time.  Need to change this
9903          * to figure out whether the disk device is actually online or not.
9904          */
9905         if (lun != NULL)
9906                 sfs_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9907                                   lun->be_lun->lun_type;
9908         else
9909                 sfs_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9910
9911         sfs_ptr->page_code = SVPD_SCSI_SFS;
9912         n = 0;
9913         /* Discovery 2016 */
9914         scsi_ulto2b(0x0001, &sfs_ptr->codes[2 * n++]);
9915         if (lun != NULL && lun->be_lun->lun_type == T_DIRECT) {
9916                  /* SBC Base 2016 */
9917                 scsi_ulto2b(0x0101, &sfs_ptr->codes[2 * n++]);
9918                  /* SBC Base 2010 */
9919                 scsi_ulto2b(0x0102, &sfs_ptr->codes[2 * n++]);
9920                 if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
9921                         /* Basic Provisioning 2016 */
9922                         scsi_ulto2b(0x0103, &sfs_ptr->codes[2 * n++]);
9923                 }
9924                 /* Drive Maintenance 2016 */
9925                 //scsi_ulto2b(0x0104, &sfs_ptr->codes[2 * n++]);
9926         }
9927         scsi_ulto2b(4 + 2 * n, sfs_ptr->page_length);
9928
9929         ctl_set_success(ctsio);
9930         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9931         ctsio->be_move_done = ctl_config_move_done;
9932         ctl_datamove((union ctl_io *)ctsio);
9933         return (CTL_RETVAL_COMPLETE);
9934 }
9935
9936 static int
9937 ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio, int alloc_len)
9938 {
9939         struct ctl_lun *lun = CTL_LUN(ctsio);
9940         struct scsi_vpd_block_limits *bl_ptr;
9941         const char *val;
9942         uint64_t ival;
9943
9944         ctsio->kern_data_ptr = malloc(sizeof(*bl_ptr), M_CTL, M_WAITOK | M_ZERO);
9945         bl_ptr = (struct scsi_vpd_block_limits *)ctsio->kern_data_ptr;
9946         ctsio->kern_sg_entries = 0;
9947         ctsio->kern_rel_offset = 0;
9948         ctsio->kern_sg_entries = 0;
9949         ctsio->kern_data_len = min(sizeof(*bl_ptr), alloc_len);
9950         ctsio->kern_total_len = ctsio->kern_data_len;
9951
9952         /*
9953          * The control device is always connected.  The disk device, on the
9954          * other hand, may not be online all the time.  Need to change this
9955          * to figure out whether the disk device is actually online or not.
9956          */
9957         if (lun != NULL)
9958                 bl_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9959                                   lun->be_lun->lun_type;
9960         else
9961                 bl_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9962
9963         bl_ptr->page_code = SVPD_BLOCK_LIMITS;
9964         scsi_ulto2b(sizeof(*bl_ptr) - 4, bl_ptr->page_length);
9965         bl_ptr->max_cmp_write_len = 0xff;
9966         scsi_ulto4b(0xffffffff, bl_ptr->max_txfer_len);
9967         if (lun != NULL) {
9968                 scsi_ulto4b(lun->be_lun->opttxferlen, bl_ptr->opt_txfer_len);
9969                 if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
9970                         ival = 0xffffffff;
9971                         val = dnvlist_get_string(lun->be_lun->options,
9972                             "unmap_max_lba", NULL);
9973                         if (val != NULL)
9974                                 ctl_expand_number(val, &ival);
9975                         scsi_ulto4b(ival, bl_ptr->max_unmap_lba_cnt);
9976                         ival = 0xffffffff;
9977                         val = dnvlist_get_string(lun->be_lun->options,
9978                             "unmap_max_descr", NULL);
9979                         if (val != NULL)
9980                                 ctl_expand_number(val, &ival);
9981                         scsi_ulto4b(ival, bl_ptr->max_unmap_blk_cnt);
9982                         if (lun->be_lun->ublockexp != 0) {
9983                                 scsi_ulto4b((1 << lun->be_lun->ublockexp),
9984                                     bl_ptr->opt_unmap_grain);
9985                                 scsi_ulto4b(0x80000000 | lun->be_lun->ublockoff,
9986                                     bl_ptr->unmap_grain_align);
9987                         }
9988                 }
9989                 scsi_ulto4b(lun->be_lun->atomicblock,
9990                     bl_ptr->max_atomic_transfer_length);
9991                 scsi_ulto4b(0, bl_ptr->atomic_alignment);
9992                 scsi_ulto4b(0, bl_ptr->atomic_transfer_length_granularity);
9993                 scsi_ulto4b(0, bl_ptr->max_atomic_transfer_length_with_atomic_boundary);
9994                 scsi_ulto4b(0, bl_ptr->max_atomic_boundary_size);
9995                 ival = UINT64_MAX;
9996                 val = dnvlist_get_string(lun->be_lun->options,
9997                     "write_same_max_lba", NULL);
9998                 if (val != NULL)
9999                         ctl_expand_number(val, &ival);
10000                 scsi_u64to8b(ival, bl_ptr->max_write_same_length);
10001                 if (lun->be_lun->maxlba + 1 > ival)
10002                         bl_ptr->flags |= SVPD_BL_WSNZ;
10003         }
10004
10005         ctl_set_success(ctsio);
10006         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10007         ctsio->be_move_done = ctl_config_move_done;
10008         ctl_datamove((union ctl_io *)ctsio);
10009         return (CTL_RETVAL_COMPLETE);
10010 }
10011
10012 static int
10013 ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len)
10014 {
10015         struct ctl_lun *lun = CTL_LUN(ctsio);
10016         struct scsi_vpd_block_device_characteristics *bdc_ptr;
10017         const char *value;
10018         u_int i;
10019
10020         ctsio->kern_data_ptr = malloc(sizeof(*bdc_ptr), M_CTL, M_WAITOK | M_ZERO);
10021         bdc_ptr = (struct scsi_vpd_block_device_characteristics *)ctsio->kern_data_ptr;
10022         ctsio->kern_sg_entries = 0;
10023         ctsio->kern_rel_offset = 0;
10024         ctsio->kern_data_len = min(sizeof(*bdc_ptr), alloc_len);
10025         ctsio->kern_total_len = ctsio->kern_data_len;
10026
10027         /*
10028          * The control device is always connected.  The disk device, on the
10029          * other hand, may not be online all the time.  Need to change this
10030          * to figure out whether the disk device is actually online or not.
10031          */
10032         if (lun != NULL)
10033                 bdc_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10034                                   lun->be_lun->lun_type;
10035         else
10036                 bdc_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10037         bdc_ptr->page_code = SVPD_BDC;
10038         scsi_ulto2b(sizeof(*bdc_ptr) - 4, bdc_ptr->page_length);
10039         if (lun != NULL &&
10040             (value = dnvlist_get_string(lun->be_lun->options, "rpm", NULL)) != NULL)
10041                 i = strtol(value, NULL, 0);
10042         else
10043                 i = CTL_DEFAULT_ROTATION_RATE;
10044         scsi_ulto2b(i, bdc_ptr->medium_rotation_rate);
10045         if (lun != NULL &&
10046             (value = dnvlist_get_string(lun->be_lun->options, "formfactor", NULL)) != NULL)
10047                 i = strtol(value, NULL, 0);
10048         else
10049                 i = 0;
10050         bdc_ptr->wab_wac_ff = (i & 0x0f);
10051         bdc_ptr->flags = SVPD_RBWZ | SVPD_FUAB | SVPD_VBULS;
10052
10053         ctl_set_success(ctsio);
10054         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10055         ctsio->be_move_done = ctl_config_move_done;
10056         ctl_datamove((union ctl_io *)ctsio);
10057         return (CTL_RETVAL_COMPLETE);
10058 }
10059
10060 static int
10061 ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len)
10062 {
10063         struct ctl_lun *lun = CTL_LUN(ctsio);
10064         struct scsi_vpd_logical_block_prov *lbp_ptr;
10065         const char *value;
10066
10067         ctsio->kern_data_ptr = malloc(sizeof(*lbp_ptr), M_CTL, M_WAITOK | M_ZERO);
10068         lbp_ptr = (struct scsi_vpd_logical_block_prov *)ctsio->kern_data_ptr;
10069         ctsio->kern_sg_entries = 0;
10070         ctsio->kern_rel_offset = 0;
10071         ctsio->kern_data_len = min(sizeof(*lbp_ptr), alloc_len);
10072         ctsio->kern_total_len = ctsio->kern_data_len;
10073
10074         /*
10075          * The control device is always connected.  The disk device, on the
10076          * other hand, may not be online all the time.  Need to change this
10077          * to figure out whether the disk device is actually online or not.
10078          */
10079         if (lun != NULL)
10080                 lbp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10081                                   lun->be_lun->lun_type;
10082         else
10083                 lbp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10084
10085         lbp_ptr->page_code = SVPD_LBP;
10086         scsi_ulto2b(sizeof(*lbp_ptr) - 4, lbp_ptr->page_length);
10087         lbp_ptr->threshold_exponent = CTL_LBP_EXPONENT;
10088         if (lun != NULL && lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
10089                 lbp_ptr->flags = SVPD_LBP_UNMAP | SVPD_LBP_WS16 |
10090                     SVPD_LBP_WS10 | SVPD_LBP_RZ | SVPD_LBP_ANC_SUP;
10091                 value = dnvlist_get_string(lun->be_lun->options,
10092                     "provisioning_type", NULL);
10093                 if (value != NULL) {
10094                         if (strcmp(value, "resource") == 0)
10095                                 lbp_ptr->prov_type = SVPD_LBP_RESOURCE;
10096                         else if (strcmp(value, "thin") == 0)
10097                                 lbp_ptr->prov_type = SVPD_LBP_THIN;
10098                 } else
10099                         lbp_ptr->prov_type = SVPD_LBP_THIN;
10100         }
10101
10102         ctl_set_success(ctsio);
10103         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10104         ctsio->be_move_done = ctl_config_move_done;
10105         ctl_datamove((union ctl_io *)ctsio);
10106         return (CTL_RETVAL_COMPLETE);
10107 }
10108
10109 /*
10110  * INQUIRY with the EVPD bit set.
10111  */
10112 static int
10113 ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
10114 {
10115         struct ctl_lun *lun = CTL_LUN(ctsio);
10116         struct scsi_inquiry *cdb;
10117         int alloc_len, retval;
10118
10119         cdb = (struct scsi_inquiry *)ctsio->cdb;
10120         alloc_len = scsi_2btoul(cdb->length);
10121
10122         switch (cdb->page_code) {
10123         case SVPD_SUPPORTED_PAGES:
10124                 retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
10125                 break;
10126         case SVPD_UNIT_SERIAL_NUMBER:
10127                 retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
10128                 break;
10129         case SVPD_DEVICE_ID:
10130                 retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
10131                 break;
10132         case SVPD_EXTENDED_INQUIRY_DATA:
10133                 retval = ctl_inquiry_evpd_eid(ctsio, alloc_len);
10134                 break;
10135         case SVPD_MODE_PAGE_POLICY:
10136                 retval = ctl_inquiry_evpd_mpp(ctsio, alloc_len);
10137                 break;
10138         case SVPD_SCSI_PORTS:
10139                 retval = ctl_inquiry_evpd_scsi_ports(ctsio, alloc_len);
10140                 break;
10141         case SVPD_SCSI_TPC:
10142                 retval = ctl_inquiry_evpd_tpc(ctsio, alloc_len);
10143                 break;
10144         case SVPD_SCSI_SFS:
10145                 retval = ctl_inquiry_evpd_sfs(ctsio, alloc_len);
10146                 break;
10147         case SVPD_BLOCK_LIMITS:
10148                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
10149                         goto err;
10150                 retval = ctl_inquiry_evpd_block_limits(ctsio, alloc_len);
10151                 break;
10152         case SVPD_BDC:
10153                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
10154                         goto err;
10155                 retval = ctl_inquiry_evpd_bdc(ctsio, alloc_len);
10156                 break;
10157         case SVPD_LBP:
10158                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
10159                         goto err;
10160                 retval = ctl_inquiry_evpd_lbp(ctsio, alloc_len);
10161                 break;
10162         default:
10163 err:
10164                 ctl_set_invalid_field(ctsio,
10165                                       /*sks_valid*/ 1,
10166                                       /*command*/ 1,
10167                                       /*field*/ 2,
10168                                       /*bit_valid*/ 0,
10169                                       /*bit*/ 0);
10170                 ctl_done((union ctl_io *)ctsio);
10171                 retval = CTL_RETVAL_COMPLETE;
10172                 break;
10173         }
10174
10175         return (retval);
10176 }
10177
10178 /*
10179  * Standard INQUIRY data.
10180  */
10181 static int
10182 ctl_inquiry_std(struct ctl_scsiio *ctsio)
10183 {
10184         struct ctl_softc *softc = CTL_SOFTC(ctsio);
10185         struct ctl_port *port = CTL_PORT(ctsio);
10186         struct ctl_lun *lun = CTL_LUN(ctsio);
10187         struct scsi_inquiry_data *inq_ptr;
10188         struct scsi_inquiry *cdb;
10189         const char *val;
10190         uint32_t alloc_len, data_len;
10191         ctl_port_type port_type;
10192
10193         port_type = port->port_type;
10194         if (port_type == CTL_PORT_IOCTL || port_type == CTL_PORT_INTERNAL)
10195                 port_type = CTL_PORT_SCSI;
10196
10197         cdb = (struct scsi_inquiry *)ctsio->cdb;
10198         alloc_len = scsi_2btoul(cdb->length);
10199
10200         /*
10201          * We malloc the full inquiry data size here and fill it
10202          * in.  If the user only asks for less, we'll give him
10203          * that much.
10204          */
10205         data_len = offsetof(struct scsi_inquiry_data, vendor_specific1);
10206         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10207         inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
10208         ctsio->kern_sg_entries = 0;
10209         ctsio->kern_rel_offset = 0;
10210         ctsio->kern_data_len = min(data_len, alloc_len);
10211         ctsio->kern_total_len = ctsio->kern_data_len;
10212
10213         if (lun != NULL) {
10214                 if ((lun->flags & CTL_LUN_PRIMARY_SC) ||
10215                     softc->ha_link >= CTL_HA_LINK_UNKNOWN) {
10216                         inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10217                             lun->be_lun->lun_type;
10218                 } else {
10219                         inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) |
10220                             lun->be_lun->lun_type;
10221                 }
10222                 if (lun->flags & CTL_LUN_REMOVABLE)
10223                         inq_ptr->dev_qual2 |= SID_RMB;
10224         } else
10225                 inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE;
10226
10227         /* RMB in byte 2 is 0 */
10228         inq_ptr->version = SCSI_REV_SPC5;
10229
10230         /*
10231          * According to SAM-3, even if a device only supports a single
10232          * level of LUN addressing, it should still set the HISUP bit:
10233          *
10234          * 4.9.1 Logical unit numbers overview
10235          *
10236          * All logical unit number formats described in this standard are
10237          * hierarchical in structure even when only a single level in that
10238          * hierarchy is used. The HISUP bit shall be set to one in the
10239          * standard INQUIRY data (see SPC-2) when any logical unit number
10240          * format described in this standard is used.  Non-hierarchical
10241          * formats are outside the scope of this standard.
10242          *
10243          * Therefore we set the HiSup bit here.
10244          *
10245          * The response format is 2, per SPC-3.
10246          */
10247         inq_ptr->response_format = SID_HiSup | 2;
10248
10249         inq_ptr->additional_length = data_len -
10250             (offsetof(struct scsi_inquiry_data, additional_length) + 1);
10251         CTL_DEBUG_PRINT(("additional_length = %d\n",
10252                          inq_ptr->additional_length));
10253
10254         inq_ptr->spc3_flags = SPC3_SID_3PC | SPC3_SID_TPGS_IMPLICIT;
10255         if (port_type == CTL_PORT_SCSI)
10256                 inq_ptr->spc2_flags = SPC2_SID_ADDR16;
10257         inq_ptr->spc2_flags |= SPC2_SID_MultiP;
10258         inq_ptr->flags = SID_CmdQue;
10259         if (port_type == CTL_PORT_SCSI)
10260                 inq_ptr->flags |= SID_WBus16 | SID_Sync;
10261
10262         /*
10263          * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
10264          * We have 8 bytes for the vendor name, and 16 bytes for the device
10265          * name and 4 bytes for the revision.
10266          */
10267         if (lun == NULL || (val = dnvlist_get_string(lun->be_lun->options,
10268             "vendor", NULL)) == NULL) {
10269                 strncpy(inq_ptr->vendor, CTL_VENDOR, sizeof(inq_ptr->vendor));
10270         } else {
10271                 memset(inq_ptr->vendor, ' ', sizeof(inq_ptr->vendor));
10272                 strncpy(inq_ptr->vendor, val,
10273                     min(sizeof(inq_ptr->vendor), strlen(val)));
10274         }
10275         if (lun == NULL) {
10276                 strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10277                     sizeof(inq_ptr->product));
10278         } else if ((val = dnvlist_get_string(lun->be_lun->options, "product",
10279             NULL)) == NULL) {
10280                 switch (lun->be_lun->lun_type) {
10281                 case T_DIRECT:
10282                         strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10283                             sizeof(inq_ptr->product));
10284                         break;
10285                 case T_PROCESSOR:
10286                         strncpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT,
10287                             sizeof(inq_ptr->product));
10288                         break;
10289                 case T_CDROM:
10290                         strncpy(inq_ptr->product, CTL_CDROM_PRODUCT,
10291                             sizeof(inq_ptr->product));
10292                         break;
10293                 default:
10294                         strncpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT,
10295                             sizeof(inq_ptr->product));
10296                         break;
10297                 }
10298         } else {
10299                 memset(inq_ptr->product, ' ', sizeof(inq_ptr->product));
10300                 strncpy(inq_ptr->product, val,
10301                     min(sizeof(inq_ptr->product), strlen(val)));
10302         }
10303
10304         /*
10305          * XXX make this a macro somewhere so it automatically gets
10306          * incremented when we make changes.
10307          */
10308         if (lun == NULL || (val = dnvlist_get_string(lun->be_lun->options,
10309             "revision", NULL)) == NULL) {
10310                 strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
10311         } else {
10312                 memset(inq_ptr->revision, ' ', sizeof(inq_ptr->revision));
10313                 strncpy(inq_ptr->revision, val,
10314                     min(sizeof(inq_ptr->revision), strlen(val)));
10315         }
10316
10317         /*
10318          * For parallel SCSI, we support double transition and single
10319          * transition clocking.  We also support QAS (Quick Arbitration
10320          * and Selection) and Information Unit transfers on both the
10321          * control and array devices.
10322          */
10323         if (port_type == CTL_PORT_SCSI)
10324                 inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
10325                                     SID_SPI_IUS;
10326
10327         /* SAM-6 (no version claimed) */
10328         scsi_ulto2b(0x00C0, inq_ptr->version1);
10329         /* SPC-5 (no version claimed) */
10330         scsi_ulto2b(0x05C0, inq_ptr->version2);
10331         if (port_type == CTL_PORT_FC) {
10332                 /* FCP-2 ANSI INCITS.350:2003 */
10333                 scsi_ulto2b(0x0917, inq_ptr->version3);
10334         } else if (port_type == CTL_PORT_SCSI) {
10335                 /* SPI-4 ANSI INCITS.362:200x */
10336                 scsi_ulto2b(0x0B56, inq_ptr->version3);
10337         } else if (port_type == CTL_PORT_ISCSI) {
10338                 /* iSCSI (no version claimed) */
10339                 scsi_ulto2b(0x0960, inq_ptr->version3);
10340         } else if (port_type == CTL_PORT_SAS) {
10341                 /* SAS (no version claimed) */
10342                 scsi_ulto2b(0x0BE0, inq_ptr->version3);
10343         } else if (port_type == CTL_PORT_UMASS) {
10344                 /* USB Mass Storage Class Bulk-Only Transport, Revision 1.0 */
10345                 scsi_ulto2b(0x1730, inq_ptr->version3);
10346         }
10347
10348         if (lun == NULL) {
10349                 /* SBC-4 (no version claimed) */
10350                 scsi_ulto2b(0x0600, inq_ptr->version4);
10351         } else {
10352                 switch (lun->be_lun->lun_type) {
10353                 case T_DIRECT:
10354                         /* SBC-4 (no version claimed) */
10355                         scsi_ulto2b(0x0600, inq_ptr->version4);
10356                         break;
10357                 case T_PROCESSOR:
10358                         break;
10359                 case T_CDROM:
10360                         /* MMC-6 (no version claimed) */
10361                         scsi_ulto2b(0x04E0, inq_ptr->version4);
10362                         break;
10363                 default:
10364                         break;
10365                 }
10366         }
10367
10368         ctl_set_success(ctsio);
10369         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10370         ctsio->be_move_done = ctl_config_move_done;
10371         ctl_datamove((union ctl_io *)ctsio);
10372         return (CTL_RETVAL_COMPLETE);
10373 }
10374
10375 int
10376 ctl_inquiry(struct ctl_scsiio *ctsio)
10377 {
10378         struct scsi_inquiry *cdb;
10379         int retval;
10380
10381         CTL_DEBUG_PRINT(("ctl_inquiry\n"));
10382
10383         cdb = (struct scsi_inquiry *)ctsio->cdb;
10384         if (cdb->byte2 & SI_EVPD)
10385                 retval = ctl_inquiry_evpd(ctsio);
10386         else if (cdb->page_code == 0)
10387                 retval = ctl_inquiry_std(ctsio);
10388         else {
10389                 ctl_set_invalid_field(ctsio,
10390                                       /*sks_valid*/ 1,
10391                                       /*command*/ 1,
10392                                       /*field*/ 2,
10393                                       /*bit_valid*/ 0,
10394                                       /*bit*/ 0);
10395                 ctl_done((union ctl_io *)ctsio);
10396                 return (CTL_RETVAL_COMPLETE);
10397         }
10398
10399         return (retval);
10400 }
10401
10402 int
10403 ctl_get_config(struct ctl_scsiio *ctsio)
10404 {
10405         struct ctl_lun *lun = CTL_LUN(ctsio);
10406         struct scsi_get_config_header *hdr;
10407         struct scsi_get_config_feature *feature;
10408         struct scsi_get_config *cdb;
10409         uint32_t alloc_len, data_len;
10410         int rt, starting;
10411
10412         cdb = (struct scsi_get_config *)ctsio->cdb;
10413         rt = (cdb->rt & SGC_RT_MASK);
10414         starting = scsi_2btoul(cdb->starting_feature);
10415         alloc_len = scsi_2btoul(cdb->length);
10416
10417         data_len = sizeof(struct scsi_get_config_header) +
10418             sizeof(struct scsi_get_config_feature) + 8 +
10419             sizeof(struct scsi_get_config_feature) + 8 +
10420             sizeof(struct scsi_get_config_feature) + 4 +
10421             sizeof(struct scsi_get_config_feature) + 4 +
10422             sizeof(struct scsi_get_config_feature) + 8 +
10423             sizeof(struct scsi_get_config_feature) +
10424             sizeof(struct scsi_get_config_feature) + 4 +
10425             sizeof(struct scsi_get_config_feature) + 4 +
10426             sizeof(struct scsi_get_config_feature) + 4 +
10427             sizeof(struct scsi_get_config_feature) + 4 +
10428             sizeof(struct scsi_get_config_feature) + 4 +
10429             sizeof(struct scsi_get_config_feature) + 4;
10430         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10431         ctsio->kern_sg_entries = 0;
10432         ctsio->kern_rel_offset = 0;
10433
10434         hdr = (struct scsi_get_config_header *)ctsio->kern_data_ptr;
10435         if (lun->flags & CTL_LUN_NO_MEDIA)
10436                 scsi_ulto2b(0x0000, hdr->current_profile);
10437         else
10438                 scsi_ulto2b(0x0010, hdr->current_profile);
10439         feature = (struct scsi_get_config_feature *)(hdr + 1);
10440
10441         if (starting > 0x003b)
10442                 goto done;
10443         if (starting > 0x003a)
10444                 goto f3b;
10445         if (starting > 0x002b)
10446                 goto f3a;
10447         if (starting > 0x002a)
10448                 goto f2b;
10449         if (starting > 0x001f)
10450                 goto f2a;
10451         if (starting > 0x001e)
10452                 goto f1f;
10453         if (starting > 0x001d)
10454                 goto f1e;
10455         if (starting > 0x0010)
10456                 goto f1d;
10457         if (starting > 0x0003)
10458                 goto f10;
10459         if (starting > 0x0002)
10460                 goto f3;
10461         if (starting > 0x0001)
10462                 goto f2;
10463         if (starting > 0x0000)
10464                 goto f1;
10465
10466         /* Profile List */
10467         scsi_ulto2b(0x0000, feature->feature_code);
10468         feature->flags = SGC_F_PERSISTENT | SGC_F_CURRENT;
10469         feature->add_length = 8;
10470         scsi_ulto2b(0x0008, &feature->feature_data[0]); /* CD-ROM */
10471         feature->feature_data[2] = 0x00;
10472         scsi_ulto2b(0x0010, &feature->feature_data[4]); /* DVD-ROM */
10473         feature->feature_data[6] = 0x01;
10474         feature = (struct scsi_get_config_feature *)
10475             &feature->feature_data[feature->add_length];
10476
10477 f1:     /* Core */
10478         scsi_ulto2b(0x0001, feature->feature_code);
10479         feature->flags = 0x08 | SGC_F_PERSISTENT | SGC_F_CURRENT;
10480         feature->add_length = 8;
10481         scsi_ulto4b(0x00000000, &feature->feature_data[0]);
10482         feature->feature_data[4] = 0x03;
10483         feature = (struct scsi_get_config_feature *)
10484             &feature->feature_data[feature->add_length];
10485
10486 f2:     /* Morphing */
10487         scsi_ulto2b(0x0002, feature->feature_code);
10488         feature->flags = 0x04 | SGC_F_PERSISTENT | SGC_F_CURRENT;
10489         feature->add_length = 4;
10490         feature->feature_data[0] = 0x02;
10491         feature = (struct scsi_get_config_feature *)
10492             &feature->feature_data[feature->add_length];
10493
10494 f3:     /* Removable Medium */
10495         scsi_ulto2b(0x0003, feature->feature_code);
10496         feature->flags = 0x04 | SGC_F_PERSISTENT | SGC_F_CURRENT;
10497         feature->add_length = 4;
10498         feature->feature_data[0] = 0x39;
10499         feature = (struct scsi_get_config_feature *)
10500             &feature->feature_data[feature->add_length];
10501
10502         if (rt == SGC_RT_CURRENT && (lun->flags & CTL_LUN_NO_MEDIA))
10503                 goto done;
10504
10505 f10:    /* Random Read */
10506         scsi_ulto2b(0x0010, feature->feature_code);
10507         feature->flags = 0x00;
10508         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10509                 feature->flags |= SGC_F_CURRENT;
10510         feature->add_length = 8;
10511         scsi_ulto4b(lun->be_lun->blocksize, &feature->feature_data[0]);
10512         scsi_ulto2b(1, &feature->feature_data[4]);
10513         feature->feature_data[6] = 0x00;
10514         feature = (struct scsi_get_config_feature *)
10515             &feature->feature_data[feature->add_length];
10516
10517 f1d:    /* Multi-Read */
10518         scsi_ulto2b(0x001D, feature->feature_code);
10519         feature->flags = 0x00;
10520         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10521                 feature->flags |= SGC_F_CURRENT;
10522         feature->add_length = 0;
10523         feature = (struct scsi_get_config_feature *)
10524             &feature->feature_data[feature->add_length];
10525
10526 f1e:    /* CD Read */
10527         scsi_ulto2b(0x001E, feature->feature_code);
10528         feature->flags = 0x00;
10529         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10530                 feature->flags |= SGC_F_CURRENT;
10531         feature->add_length = 4;
10532         feature->feature_data[0] = 0x00;
10533         feature = (struct scsi_get_config_feature *)
10534             &feature->feature_data[feature->add_length];
10535
10536 f1f:    /* DVD Read */
10537         scsi_ulto2b(0x001F, feature->feature_code);
10538         feature->flags = 0x08;
10539         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10540                 feature->flags |= SGC_F_CURRENT;
10541         feature->add_length = 4;
10542         feature->feature_data[0] = 0x01;
10543         feature->feature_data[2] = 0x03;
10544         feature = (struct scsi_get_config_feature *)
10545             &feature->feature_data[feature->add_length];
10546
10547 f2a:    /* DVD+RW */
10548         scsi_ulto2b(0x002A, feature->feature_code);
10549         feature->flags = 0x04;
10550         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10551                 feature->flags |= SGC_F_CURRENT;
10552         feature->add_length = 4;
10553         feature->feature_data[0] = 0x00;
10554         feature->feature_data[1] = 0x00;
10555         feature = (struct scsi_get_config_feature *)
10556             &feature->feature_data[feature->add_length];
10557
10558 f2b:    /* DVD+R */
10559         scsi_ulto2b(0x002B, feature->feature_code);
10560         feature->flags = 0x00;
10561         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10562                 feature->flags |= SGC_F_CURRENT;
10563         feature->add_length = 4;
10564         feature->feature_data[0] = 0x00;
10565         feature = (struct scsi_get_config_feature *)
10566             &feature->feature_data[feature->add_length];
10567
10568 f3a:    /* DVD+RW Dual Layer */
10569         scsi_ulto2b(0x003A, feature->feature_code);
10570         feature->flags = 0x00;
10571         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10572                 feature->flags |= SGC_F_CURRENT;
10573         feature->add_length = 4;
10574         feature->feature_data[0] = 0x00;
10575         feature->feature_data[1] = 0x00;
10576         feature = (struct scsi_get_config_feature *)
10577             &feature->feature_data[feature->add_length];
10578
10579 f3b:    /* DVD+R Dual Layer */
10580         scsi_ulto2b(0x003B, feature->feature_code);
10581         feature->flags = 0x00;
10582         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10583                 feature->flags |= SGC_F_CURRENT;
10584         feature->add_length = 4;
10585         feature->feature_data[0] = 0x00;
10586         feature = (struct scsi_get_config_feature *)
10587             &feature->feature_data[feature->add_length];
10588
10589 done:
10590         data_len = (uint8_t *)feature - (uint8_t *)hdr;
10591         if (rt == SGC_RT_SPECIFIC && data_len > 4) {
10592                 feature = (struct scsi_get_config_feature *)(hdr + 1);
10593                 if (scsi_2btoul(feature->feature_code) == starting)
10594                         feature = (struct scsi_get_config_feature *)
10595                             &feature->feature_data[feature->add_length];
10596                 data_len = (uint8_t *)feature - (uint8_t *)hdr;
10597         }
10598         scsi_ulto4b(data_len - 4, hdr->data_length);
10599         ctsio->kern_data_len = min(data_len, alloc_len);
10600         ctsio->kern_total_len = ctsio->kern_data_len;
10601
10602         ctl_set_success(ctsio);
10603         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10604         ctsio->be_move_done = ctl_config_move_done;
10605         ctl_datamove((union ctl_io *)ctsio);
10606         return (CTL_RETVAL_COMPLETE);
10607 }
10608
10609 int
10610 ctl_get_event_status(struct ctl_scsiio *ctsio)
10611 {
10612         struct scsi_get_event_status_header *hdr;
10613         struct scsi_get_event_status *cdb;
10614         uint32_t alloc_len, data_len;
10615
10616         cdb = (struct scsi_get_event_status *)ctsio->cdb;
10617         if ((cdb->byte2 & SGESN_POLLED) == 0) {
10618                 ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 1,
10619                     /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 0);
10620                 ctl_done((union ctl_io *)ctsio);
10621                 return (CTL_RETVAL_COMPLETE);
10622         }
10623         alloc_len = scsi_2btoul(cdb->length);
10624
10625         data_len = sizeof(struct scsi_get_event_status_header);
10626         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10627         ctsio->kern_sg_entries = 0;
10628         ctsio->kern_rel_offset = 0;
10629         ctsio->kern_data_len = min(data_len, alloc_len);
10630         ctsio->kern_total_len = ctsio->kern_data_len;
10631
10632         hdr = (struct scsi_get_event_status_header *)ctsio->kern_data_ptr;
10633         scsi_ulto2b(0, hdr->descr_length);
10634         hdr->nea_class = SGESN_NEA;
10635         hdr->supported_class = 0;
10636
10637         ctl_set_success(ctsio);
10638         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10639         ctsio->be_move_done = ctl_config_move_done;
10640         ctl_datamove((union ctl_io *)ctsio);
10641         return (CTL_RETVAL_COMPLETE);
10642 }
10643
10644 int
10645 ctl_mechanism_status(struct ctl_scsiio *ctsio)
10646 {
10647         struct scsi_mechanism_status_header *hdr;
10648         struct scsi_mechanism_status *cdb;
10649         uint32_t alloc_len, data_len;
10650
10651         cdb = (struct scsi_mechanism_status *)ctsio->cdb;
10652         alloc_len = scsi_2btoul(cdb->length);
10653
10654         data_len = sizeof(struct scsi_mechanism_status_header);
10655         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10656         ctsio->kern_sg_entries = 0;
10657         ctsio->kern_rel_offset = 0;
10658         ctsio->kern_data_len = min(data_len, alloc_len);
10659         ctsio->kern_total_len = ctsio->kern_data_len;
10660
10661         hdr = (struct scsi_mechanism_status_header *)ctsio->kern_data_ptr;
10662         hdr->state1 = 0x00;
10663         hdr->state2 = 0xe0;
10664         scsi_ulto3b(0, hdr->lba);
10665         hdr->slots_num = 0;
10666         scsi_ulto2b(0, hdr->slots_length);
10667
10668         ctl_set_success(ctsio);
10669         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10670         ctsio->be_move_done = ctl_config_move_done;
10671         ctl_datamove((union ctl_io *)ctsio);
10672         return (CTL_RETVAL_COMPLETE);
10673 }
10674
10675 static void
10676 ctl_ultomsf(uint32_t lba, uint8_t *buf)
10677 {
10678
10679         lba += 150;
10680         buf[0] = 0;
10681         buf[1] = bin2bcd((lba / 75) / 60);
10682         buf[2] = bin2bcd((lba / 75) % 60);
10683         buf[3] = bin2bcd(lba % 75);
10684 }
10685
10686 int
10687 ctl_read_toc(struct ctl_scsiio *ctsio)
10688 {
10689         struct ctl_lun *lun = CTL_LUN(ctsio);
10690         struct scsi_read_toc_hdr *hdr;
10691         struct scsi_read_toc_type01_descr *descr;
10692         struct scsi_read_toc *cdb;
10693         uint32_t alloc_len, data_len;
10694         int format, msf;
10695
10696         cdb = (struct scsi_read_toc *)ctsio->cdb;
10697         msf = (cdb->byte2 & CD_MSF) != 0;
10698         format = cdb->format;
10699         alloc_len = scsi_2btoul(cdb->data_len);
10700
10701         data_len = sizeof(struct scsi_read_toc_hdr);
10702         if (format == 0)
10703                 data_len += 2 * sizeof(struct scsi_read_toc_type01_descr);
10704         else
10705                 data_len += sizeof(struct scsi_read_toc_type01_descr);
10706         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10707         ctsio->kern_sg_entries = 0;
10708         ctsio->kern_rel_offset = 0;
10709         ctsio->kern_data_len = min(data_len, alloc_len);
10710         ctsio->kern_total_len = ctsio->kern_data_len;
10711
10712         hdr = (struct scsi_read_toc_hdr *)ctsio->kern_data_ptr;
10713         if (format == 0) {
10714                 scsi_ulto2b(0x12, hdr->data_length);
10715                 hdr->first = 1;
10716                 hdr->last = 1;
10717                 descr = (struct scsi_read_toc_type01_descr *)(hdr + 1);
10718                 descr->addr_ctl = 0x14;
10719                 descr->track_number = 1;
10720                 if (msf)
10721                         ctl_ultomsf(0, descr->track_start);
10722                 else
10723                         scsi_ulto4b(0, descr->track_start);
10724                 descr++;
10725                 descr->addr_ctl = 0x14;
10726                 descr->track_number = 0xaa;
10727                 if (msf)
10728                         ctl_ultomsf(lun->be_lun->maxlba+1, descr->track_start);
10729                 else
10730                         scsi_ulto4b(lun->be_lun->maxlba+1, descr->track_start);
10731         } else {
10732                 scsi_ulto2b(0x0a, hdr->data_length);
10733                 hdr->first = 1;
10734                 hdr->last = 1;
10735                 descr = (struct scsi_read_toc_type01_descr *)(hdr + 1);
10736                 descr->addr_ctl = 0x14;
10737                 descr->track_number = 1;
10738                 if (msf)
10739                         ctl_ultomsf(0, descr->track_start);
10740                 else
10741                         scsi_ulto4b(0, descr->track_start);
10742         }
10743
10744         ctl_set_success(ctsio);
10745         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10746         ctsio->be_move_done = ctl_config_move_done;
10747         ctl_datamove((union ctl_io *)ctsio);
10748         return (CTL_RETVAL_COMPLETE);
10749 }
10750
10751 /*
10752  * For known CDB types, parse the LBA and length.
10753  */
10754 static int
10755 ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len)
10756 {
10757
10758         KASSERT(io->io_hdr.io_type == CTL_IO_SCSI,
10759             ("%s: unexpected I/O type %x", __func__, io->io_hdr.io_type));
10760
10761         switch (io->scsiio.cdb[0]) {
10762         case COMPARE_AND_WRITE: {
10763                 struct scsi_compare_and_write *cdb;
10764
10765                 cdb = (struct scsi_compare_and_write *)io->scsiio.cdb;
10766
10767                 *lba = scsi_8btou64(cdb->addr);
10768                 *len = cdb->length;
10769                 break;
10770         }
10771         case READ_6:
10772         case WRITE_6: {
10773                 struct scsi_rw_6 *cdb;
10774
10775                 cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
10776
10777                 *lba = scsi_3btoul(cdb->addr);
10778                 /* only 5 bits are valid in the most significant address byte */
10779                 *lba &= 0x1fffff;
10780                 *len = cdb->length;
10781                 break;
10782         }
10783         case READ_10:
10784         case WRITE_10: {
10785                 struct scsi_rw_10 *cdb;
10786
10787                 cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
10788
10789                 *lba = scsi_4btoul(cdb->addr);
10790                 *len = scsi_2btoul(cdb->length);
10791                 break;
10792         }
10793         case WRITE_VERIFY_10: {
10794                 struct scsi_write_verify_10 *cdb;
10795
10796                 cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
10797
10798                 *lba = scsi_4btoul(cdb->addr);
10799                 *len = scsi_2btoul(cdb->length);
10800                 break;
10801         }
10802         case READ_12:
10803         case WRITE_12: {
10804                 struct scsi_rw_12 *cdb;
10805
10806                 cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
10807
10808                 *lba = scsi_4btoul(cdb->addr);
10809                 *len = scsi_4btoul(cdb->length);
10810                 break;
10811         }
10812         case WRITE_VERIFY_12: {
10813                 struct scsi_write_verify_12 *cdb;
10814
10815                 cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
10816
10817                 *lba = scsi_4btoul(cdb->addr);
10818                 *len = scsi_4btoul(cdb->length);
10819                 break;
10820         }
10821         case READ_16:
10822         case WRITE_16: {
10823                 struct scsi_rw_16 *cdb;
10824
10825                 cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
10826
10827                 *lba = scsi_8btou64(cdb->addr);
10828                 *len = scsi_4btoul(cdb->length);
10829                 break;
10830         }
10831         case WRITE_ATOMIC_16: {
10832                 struct scsi_write_atomic_16 *cdb;
10833
10834                 cdb = (struct scsi_write_atomic_16 *)io->scsiio.cdb;
10835
10836                 *lba = scsi_8btou64(cdb->addr);
10837                 *len = scsi_2btoul(cdb->length);
10838                 break;
10839         }
10840         case WRITE_VERIFY_16: {
10841                 struct scsi_write_verify_16 *cdb;
10842
10843                 cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
10844
10845                 *lba = scsi_8btou64(cdb->addr);
10846                 *len = scsi_4btoul(cdb->length);
10847                 break;
10848         }
10849         case WRITE_SAME_10: {
10850                 struct scsi_write_same_10 *cdb;
10851
10852                 cdb = (struct scsi_write_same_10 *)io->scsiio.cdb;
10853
10854                 *lba = scsi_4btoul(cdb->addr);
10855                 *len = scsi_2btoul(cdb->length);
10856                 break;
10857         }
10858         case WRITE_SAME_16: {
10859                 struct scsi_write_same_16 *cdb;
10860
10861                 cdb = (struct scsi_write_same_16 *)io->scsiio.cdb;
10862
10863                 *lba = scsi_8btou64(cdb->addr);
10864                 *len = scsi_4btoul(cdb->length);
10865                 break;
10866         }
10867         case VERIFY_10: {
10868                 struct scsi_verify_10 *cdb;
10869
10870                 cdb = (struct scsi_verify_10 *)io->scsiio.cdb;
10871
10872                 *lba = scsi_4btoul(cdb->addr);
10873                 *len = scsi_2btoul(cdb->length);
10874                 break;
10875         }
10876         case VERIFY_12: {
10877                 struct scsi_verify_12 *cdb;
10878
10879                 cdb = (struct scsi_verify_12 *)io->scsiio.cdb;
10880
10881                 *lba = scsi_4btoul(cdb->addr);
10882                 *len = scsi_4btoul(cdb->length);
10883                 break;
10884         }
10885         case VERIFY_16: {
10886                 struct scsi_verify_16 *cdb;
10887
10888                 cdb = (struct scsi_verify_16 *)io->scsiio.cdb;
10889
10890                 *lba = scsi_8btou64(cdb->addr);
10891                 *len = scsi_4btoul(cdb->length);
10892                 break;
10893         }
10894         case UNMAP: {
10895                 *lba = 0;
10896                 *len = UINT64_MAX;
10897                 break;
10898         }
10899         case SERVICE_ACTION_IN: {       /* GET LBA STATUS */
10900                 struct scsi_get_lba_status *cdb;
10901
10902                 cdb = (struct scsi_get_lba_status *)io->scsiio.cdb;
10903                 *lba = scsi_8btou64(cdb->addr);
10904                 *len = UINT32_MAX;
10905                 break;
10906         }
10907         default:
10908                 *lba = 0;
10909                 *len = UINT64_MAX;
10910                 return (1);
10911         }
10912
10913         return (0);
10914 }
10915
10916 static ctl_action
10917 ctl_extent_check_lba(uint64_t lba1, uint64_t len1, uint64_t lba2, uint64_t len2,
10918     bool seq)
10919 {
10920         uint64_t endlba1, endlba2;
10921
10922         endlba1 = lba1 + len1 - (seq ? 0 : 1);
10923         endlba2 = lba2 + len2 - 1;
10924
10925         if ((endlba1 < lba2) || (endlba2 < lba1))
10926                 return (CTL_ACTION_PASS);
10927         else
10928                 return (CTL_ACTION_BLOCK);
10929 }
10930
10931 static int
10932 ctl_extent_check_unmap(union ctl_io *io, uint64_t lba2, uint64_t len2)
10933 {
10934         struct ctl_ptr_len_flags *ptrlen;
10935         struct scsi_unmap_desc *buf, *end, *range;
10936         uint64_t lba;
10937         uint32_t len;
10938
10939         KASSERT(io->io_hdr.io_type == CTL_IO_SCSI,
10940             ("%s: unexpected I/O type %x", __func__, io->io_hdr.io_type));
10941
10942         /* If not UNMAP -- go other way. */
10943         if (io->scsiio.cdb[0] != UNMAP)
10944                 return (CTL_ACTION_SKIP);
10945
10946         /* If UNMAP without data -- block and wait for data. */
10947         ptrlen = (struct ctl_ptr_len_flags *)
10948             &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
10949         if ((io->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0 ||
10950             ptrlen->ptr == NULL)
10951                 return (CTL_ACTION_BLOCK);
10952
10953         /* UNMAP with data -- check for collision. */
10954         buf = (struct scsi_unmap_desc *)ptrlen->ptr;
10955         end = buf + ptrlen->len / sizeof(*buf);
10956         for (range = buf; range < end; range++) {
10957                 lba = scsi_8btou64(range->lba);
10958                 len = scsi_4btoul(range->length);
10959                 if ((lba < lba2 + len2) && (lba + len > lba2))
10960                         return (CTL_ACTION_BLOCK);
10961         }
10962         return (CTL_ACTION_PASS);
10963 }
10964
10965 static ctl_action
10966 ctl_extent_check(union ctl_io *io1, union ctl_io *io2, bool seq)
10967 {
10968         uint64_t lba1, lba2;
10969         uint64_t len1, len2;
10970         int retval;
10971
10972         retval = ctl_get_lba_len(io2, &lba2, &len2);
10973         KASSERT(retval == 0, ("ctl_get_lba_len() error"));
10974
10975         retval = ctl_extent_check_unmap(io1, lba2, len2);
10976         if (retval != CTL_ACTION_SKIP)
10977                 return (retval);
10978
10979         retval = ctl_get_lba_len(io1, &lba1, &len1);
10980         KASSERT(retval == 0, ("ctl_get_lba_len() error"));
10981
10982         if (seq && (io1->io_hdr.flags & CTL_FLAG_SERSEQ_DONE))
10983                 seq = FALSE;
10984         return (ctl_extent_check_lba(lba1, len1, lba2, len2, seq));
10985 }
10986
10987 static ctl_action
10988 ctl_seq_check(union ctl_io *io1, union ctl_io *io2)
10989 {
10990         uint64_t lba1, lba2;
10991         uint64_t len1, len2;
10992         int retval __diagused;
10993
10994         if (io1->io_hdr.flags & CTL_FLAG_SERSEQ_DONE)
10995                 return (CTL_ACTION_PASS);
10996         retval = ctl_get_lba_len(io1, &lba1, &len1);
10997         KASSERT(retval == 0, ("ctl_get_lba_len() error"));
10998         retval = ctl_get_lba_len(io2, &lba2, &len2);
10999         KASSERT(retval == 0, ("ctl_get_lba_len() error"));
11000
11001         if (lba1 + len1 == lba2)
11002                 return (CTL_ACTION_BLOCK);
11003         return (CTL_ACTION_PASS);
11004 }
11005
11006 static ctl_action
11007 ctl_check_for_blockage(struct ctl_lun *lun, union ctl_io *pending_io,
11008     const uint8_t *serialize_row, union ctl_io *ooa_io)
11009 {
11010
11011         /*
11012          * The initiator attempted multiple untagged commands at the same
11013          * time.  Can't do that.
11014          */
11015         if (__predict_false(pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
11016          && __predict_false(ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
11017          && ((pending_io->io_hdr.nexus.targ_port ==
11018               ooa_io->io_hdr.nexus.targ_port)
11019           && (pending_io->io_hdr.nexus.initid ==
11020               ooa_io->io_hdr.nexus.initid))
11021          && ((ooa_io->io_hdr.flags & (CTL_FLAG_ABORT |
11022               CTL_FLAG_STATUS_SENT)) == 0))
11023                 return (CTL_ACTION_OVERLAP);
11024
11025         /*
11026          * The initiator attempted to send multiple tagged commands with
11027          * the same ID.  (It's fine if different initiators have the same
11028          * tag ID.)
11029          *
11030          * Even if all of those conditions are true, we don't kill the I/O
11031          * if the command ahead of us has been aborted.  We won't end up
11032          * sending it to the FETD, and it's perfectly legal to resend a
11033          * command with the same tag number as long as the previous
11034          * instance of this tag number has been aborted somehow.
11035          */
11036         if (__predict_true(pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
11037          && __predict_true(ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
11038          && __predict_false(pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
11039          && ((pending_io->io_hdr.nexus.targ_port ==
11040               ooa_io->io_hdr.nexus.targ_port)
11041           && (pending_io->io_hdr.nexus.initid ==
11042               ooa_io->io_hdr.nexus.initid))
11043          && ((ooa_io->io_hdr.flags & (CTL_FLAG_ABORT |
11044               CTL_FLAG_STATUS_SENT)) == 0))
11045                 return (CTL_ACTION_OVERLAP_TAG);
11046
11047         /*
11048          * If we get a head of queue tag, SAM-3 says that we should
11049          * immediately execute it.
11050          *
11051          * What happens if this command would normally block for some other
11052          * reason?  e.g. a request sense with a head of queue tag
11053          * immediately after a write.  Normally that would block, but this
11054          * will result in its getting executed immediately...
11055          *
11056          * We currently return "pass" instead of "skip", so we'll end up
11057          * going through the rest of the queue to check for overlapped tags.
11058          *
11059          * XXX KDM check for other types of blockage first??
11060          */
11061         if (__predict_false(pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE))
11062                 return (CTL_ACTION_PASS);
11063
11064         /*
11065          * Simple tags get blocked until all head of queue and ordered tags
11066          * ahead of them have completed.  I'm lumping untagged commands in
11067          * with simple tags here.  XXX KDM is that the right thing to do?
11068          */
11069         if (__predict_false(ooa_io->scsiio.tag_type == CTL_TAG_ORDERED) ||
11070             __predict_false(ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE))
11071                 return (CTL_ACTION_BLOCK);
11072
11073         /* Unsupported command in OOA queue. */
11074         if (__predict_false(ooa_io->scsiio.seridx == CTL_SERIDX_INVLD))
11075                 return (CTL_ACTION_PASS);
11076
11077         switch (serialize_row[ooa_io->scsiio.seridx]) {
11078         case CTL_SER_SEQ:
11079                 if (lun->be_lun->serseq != CTL_LUN_SERSEQ_OFF)
11080                         return (ctl_seq_check(ooa_io, pending_io));
11081                 /* FALLTHROUGH */
11082         case CTL_SER_PASS:
11083                 return (CTL_ACTION_PASS);
11084         case CTL_SER_EXTENTOPT:
11085                 if ((lun->MODE_CTRL.queue_flags & SCP_QUEUE_ALG_MASK) ==
11086                     SCP_QUEUE_ALG_UNRESTRICTED)
11087                         return (CTL_ACTION_PASS);
11088                 /* FALLTHROUGH */
11089         case CTL_SER_EXTENT:
11090                 return (ctl_extent_check(ooa_io, pending_io,
11091                     (lun->be_lun->serseq == CTL_LUN_SERSEQ_ON)));
11092         case CTL_SER_BLOCKOPT:
11093                 if ((lun->MODE_CTRL.queue_flags & SCP_QUEUE_ALG_MASK) ==
11094                     SCP_QUEUE_ALG_UNRESTRICTED)
11095                         return (CTL_ACTION_PASS);
11096                 /* FALLTHROUGH */
11097         case CTL_SER_BLOCK:
11098                 return (CTL_ACTION_BLOCK);
11099         default:
11100                 __assert_unreachable();
11101         }
11102 }
11103
11104 /*
11105  * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
11106  * Assumptions:
11107  * - pending_io is generally either incoming, or on the blocked queue
11108  * - starting I/O is the I/O we want to start the check with.
11109  */
11110 static ctl_action
11111 ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
11112               union ctl_io **starting_io)
11113 {
11114         union ctl_io *ooa_io = *starting_io;
11115         const uint8_t *serialize_row;
11116         ctl_action action;
11117
11118         mtx_assert(&lun->lun_lock, MA_OWNED);
11119
11120         /*
11121          * Aborted commands are not going to be executed and may even
11122          * not report completion, so we don't care about their order.
11123          * Let them complete ASAP to clean the OOA queue.
11124          */
11125         if (__predict_false(pending_io->io_hdr.flags & CTL_FLAG_ABORT))
11126                 return (CTL_ACTION_SKIP);
11127
11128         /*
11129          * Ordered tags have to block until all items ahead of them have
11130          * completed.  If we get called with an ordered tag, we always
11131          * block, if something else is ahead of us in the queue.
11132          */
11133         if ((pending_io->scsiio.tag_type == CTL_TAG_ORDERED) &&
11134             (ooa_io != NULL))
11135                 return (CTL_ACTION_BLOCK);
11136
11137         serialize_row = ctl_serialize_table[pending_io->scsiio.seridx];
11138
11139         /*
11140          * Run back along the OOA queue, starting with the current
11141          * blocked I/O and going through every I/O before it on the
11142          * queue.  If starting_io is NULL, we'll just end up returning
11143          * CTL_ACTION_PASS.
11144          */
11145         for (; ooa_io != NULL;
11146              ooa_io = (union ctl_io *)LIST_NEXT(&ooa_io->io_hdr, ooa_links)) {
11147                 action = ctl_check_for_blockage(lun, pending_io, serialize_row,
11148                     ooa_io);
11149                 if (action != CTL_ACTION_PASS) {
11150                         *starting_io = ooa_io;
11151                         return (action);
11152                 }
11153         }
11154
11155         *starting_io = NULL;
11156         return (CTL_ACTION_PASS);
11157 }
11158
11159 /*
11160  * Try to unblock the specified I/O.
11161  *
11162  * skip parameter allows explicitly skip present blocker of the I/O,
11163  * starting from the previous one on OOA queue.  It can be used when
11164  * we know for sure that the blocker I/O does no longer count.
11165  */
11166 static void
11167 ctl_try_unblock_io(struct ctl_lun *lun, union ctl_io *io, bool skip)
11168 {
11169         struct ctl_softc *softc = lun->ctl_softc;
11170         union ctl_io *bio, *obio;
11171         const struct ctl_cmd_entry *entry;
11172         union ctl_ha_msg msg_info;
11173         ctl_action action;
11174
11175         mtx_assert(&lun->lun_lock, MA_OWNED);
11176
11177         if (io->io_hdr.blocker == NULL)
11178                 return;
11179
11180         obio = bio = io->io_hdr.blocker;
11181         if (skip)
11182                 bio = (union ctl_io *)LIST_NEXT(&bio->io_hdr, ooa_links);
11183         action = ctl_check_ooa(lun, io, &bio);
11184         if (action == CTL_ACTION_BLOCK) {
11185                 /* Still blocked, but may be by different I/O now. */
11186                 if (bio != obio) {
11187                         TAILQ_REMOVE(&obio->io_hdr.blocked_queue,
11188                             &io->io_hdr, blocked_links);
11189                         TAILQ_INSERT_TAIL(&bio->io_hdr.blocked_queue,
11190                             &io->io_hdr, blocked_links);
11191                         io->io_hdr.blocker = bio;
11192                 }
11193                 return;
11194         }
11195
11196         /* No longer blocked, one way or another. */
11197         TAILQ_REMOVE(&obio->io_hdr.blocked_queue, &io->io_hdr, blocked_links);
11198         io->io_hdr.blocker = NULL;
11199
11200         switch (action) {
11201         case CTL_ACTION_PASS:
11202         case CTL_ACTION_SKIP:
11203
11204                 /* Serializing commands from the other SC retire there. */
11205                 if ((io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) &&
11206                     (softc->ha_mode != CTL_HA_MODE_XFER)) {
11207                         io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11208                         msg_info.hdr.original_sc = io->io_hdr.remote_io;
11209                         msg_info.hdr.serializing_sc = io;
11210                         msg_info.hdr.msg_type = CTL_MSG_R2R;
11211                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11212                             sizeof(msg_info.hdr), M_NOWAIT);
11213                         break;
11214                 }
11215
11216                 /*
11217                  * Check this I/O for LUN state changes that may have happened
11218                  * while this command was blocked. The LUN state may have been
11219                  * changed by a command ahead of us in the queue.
11220                  */
11221                 entry = ctl_get_cmd_entry(&io->scsiio, NULL);
11222                 if (ctl_scsiio_lun_check(lun, entry, &io->scsiio) != 0) {
11223                         ctl_done(io);
11224                         break;
11225                 }
11226
11227                 io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11228                 ctl_enqueue_rtr(io);
11229                 break;
11230         default:
11231                 __assert_unreachable();
11232         case CTL_ACTION_OVERLAP:
11233                 ctl_set_overlapped_cmd(&io->scsiio);
11234                 goto error;
11235         case CTL_ACTION_OVERLAP_TAG:
11236                 ctl_set_overlapped_tag(&io->scsiio,
11237                     io->scsiio.tag_num & 0xff);
11238 error:
11239                 /* Serializing commands from the other SC are done here. */
11240                 if ((io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) &&
11241                     (softc->ha_mode != CTL_HA_MODE_XFER)) {
11242                         ctl_try_unblock_others(lun, io, TRUE);
11243                         LIST_REMOVE(&io->io_hdr, ooa_links);
11244
11245                         ctl_copy_sense_data_back(io, &msg_info);
11246                         msg_info.hdr.original_sc = io->io_hdr.remote_io;
11247                         msg_info.hdr.serializing_sc = NULL;
11248                         msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
11249                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11250                             sizeof(msg_info.scsi), M_WAITOK);
11251                         ctl_free_io(io);
11252                         break;
11253                 }
11254
11255                 ctl_done(io);
11256                 break;
11257         }
11258 }
11259
11260 /*
11261  * Try to unblock I/Os blocked by the specified I/O.
11262  *
11263  * skip parameter allows explicitly skip the specified I/O as blocker,
11264  * starting from the previous one on the OOA queue.  It can be used when
11265  * we know for sure that the specified I/O does no longer count (done).
11266  * It has to be still on OOA queue though so that we know where to start.
11267  */
11268 static void
11269 ctl_try_unblock_others(struct ctl_lun *lun, union ctl_io *bio, bool skip)
11270 {
11271         union ctl_io *io, *next_io;
11272
11273         mtx_assert(&lun->lun_lock, MA_OWNED);
11274
11275         for (io = (union ctl_io *)TAILQ_FIRST(&bio->io_hdr.blocked_queue);
11276              io != NULL; io = next_io) {
11277                 next_io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr, blocked_links);
11278
11279                 KASSERT(io->io_hdr.blocker != NULL,
11280                     ("I/O %p on blocked list without blocker", io));
11281                 ctl_try_unblock_io(lun, io, skip);
11282         }
11283         KASSERT(!skip || TAILQ_EMPTY(&bio->io_hdr.blocked_queue),
11284             ("blocked_queue is not empty after skipping %p", bio));
11285 }
11286
11287 /*
11288  * This routine (with one exception) checks LUN flags that can be set by
11289  * commands ahead of us in the OOA queue.  These flags have to be checked
11290  * when a command initially comes in, and when we pull a command off the
11291  * blocked queue and are preparing to execute it.  The reason we have to
11292  * check these flags for commands on the blocked queue is that the LUN
11293  * state may have been changed by a command ahead of us while we're on the
11294  * blocked queue.
11295  *
11296  * Ordering is somewhat important with these checks, so please pay
11297  * careful attention to the placement of any new checks.
11298  */
11299 static int
11300 ctl_scsiio_lun_check(struct ctl_lun *lun,
11301     const struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
11302 {
11303         struct ctl_softc *softc = lun->ctl_softc;
11304         int retval;
11305         uint32_t residx;
11306
11307         retval = 0;
11308
11309         mtx_assert(&lun->lun_lock, MA_OWNED);
11310
11311         /*
11312          * If this shelf is a secondary shelf controller, we may have to
11313          * reject some commands disallowed by HA mode and link state.
11314          */
11315         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
11316                 if (softc->ha_link == CTL_HA_LINK_OFFLINE &&
11317                     (entry->flags & CTL_CMD_FLAG_OK_ON_UNAVAIL) == 0) {
11318                         ctl_set_lun_unavail(ctsio);
11319                         retval = 1;
11320                         goto bailout;
11321                 }
11322                 if ((lun->flags & CTL_LUN_PEER_SC_PRIMARY) == 0 &&
11323                     (entry->flags & CTL_CMD_FLAG_OK_ON_UNAVAIL) == 0) {
11324                         ctl_set_lun_transit(ctsio);
11325                         retval = 1;
11326                         goto bailout;
11327                 }
11328                 if (softc->ha_mode == CTL_HA_MODE_ACT_STBY &&
11329                     (entry->flags & CTL_CMD_FLAG_OK_ON_STANDBY) == 0) {
11330                         ctl_set_lun_standby(ctsio);
11331                         retval = 1;
11332                         goto bailout;
11333                 }
11334
11335                 /* The rest of checks are only done on executing side */
11336                 if (softc->ha_mode == CTL_HA_MODE_XFER)
11337                         goto bailout;
11338         }
11339
11340         if (entry->pattern & CTL_LUN_PAT_WRITE) {
11341                 if (lun->be_lun->flags & CTL_LUN_FLAG_READONLY) {
11342                         ctl_set_hw_write_protected(ctsio);
11343                         retval = 1;
11344                         goto bailout;
11345                 }
11346                 if ((lun->MODE_CTRL.eca_and_aen & SCP_SWP) != 0) {
11347                         ctl_set_sense(ctsio, /*current_error*/ 1,
11348                             /*sense_key*/ SSD_KEY_DATA_PROTECT,
11349                             /*asc*/ 0x27, /*ascq*/ 0x02, SSD_ELEM_NONE);
11350                         retval = 1;
11351                         goto bailout;
11352                 }
11353         }
11354
11355         /*
11356          * Check for a reservation conflict.  If this command isn't allowed
11357          * even on reserved LUNs, and if this initiator isn't the one who
11358          * reserved us, reject the command with a reservation conflict.
11359          */
11360         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11361         if ((lun->flags & CTL_LUN_RESERVED)
11362          && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
11363                 if (lun->res_idx != residx) {
11364                         ctl_set_reservation_conflict(ctsio);
11365                         retval = 1;
11366                         goto bailout;
11367                 }
11368         }
11369
11370         if ((lun->flags & CTL_LUN_PR_RESERVED) == 0 ||
11371             (entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV)) {
11372                 /* No reservation or command is allowed. */;
11373         } else if ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_WRESV) &&
11374             (lun->pr_res_type == SPR_TYPE_WR_EX ||
11375              lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
11376              lun->pr_res_type == SPR_TYPE_WR_EX_AR)) {
11377                 /* The command is allowed for Write Exclusive resv. */;
11378         } else {
11379                 /*
11380                  * if we aren't registered or it's a res holder type
11381                  * reservation and this isn't the res holder then set a
11382                  * conflict.
11383                  */
11384                 if (ctl_get_prkey(lun, residx) == 0 ||
11385                     (residx != lun->pr_res_idx && lun->pr_res_type < 4)) {
11386                         ctl_set_reservation_conflict(ctsio);
11387                         retval = 1;
11388                         goto bailout;
11389                 }
11390         }
11391
11392         if ((entry->flags & CTL_CMD_FLAG_OK_ON_NO_MEDIA) == 0) {
11393                 if (lun->flags & CTL_LUN_EJECTED)
11394                         ctl_set_lun_ejected(ctsio);
11395                 else if (lun->flags & CTL_LUN_NO_MEDIA) {
11396                         if (lun->flags & CTL_LUN_REMOVABLE)
11397                                 ctl_set_lun_no_media(ctsio);
11398                         else
11399                                 ctl_set_lun_int_reqd(ctsio);
11400                 } else if (lun->flags & CTL_LUN_STOPPED)
11401                         ctl_set_lun_stopped(ctsio);
11402                 else
11403                         goto bailout;
11404                 retval = 1;
11405                 goto bailout;
11406         }
11407
11408 bailout:
11409         return (retval);
11410 }
11411
11412 static void
11413 ctl_failover_io(union ctl_io *io, int have_lock)
11414 {
11415         ctl_set_busy(&io->scsiio);
11416         ctl_done(io);
11417 }
11418
11419 static void
11420 ctl_failover_lun(union ctl_io *rio)
11421 {
11422         struct ctl_softc *softc = CTL_SOFTC(rio);
11423         struct ctl_lun *lun;
11424         struct ctl_io_hdr *io, *next_io;
11425         uint32_t targ_lun;
11426
11427         targ_lun = rio->io_hdr.nexus.targ_mapped_lun;
11428         CTL_DEBUG_PRINT(("FAILOVER for lun %u\n", targ_lun));
11429
11430         /* Find and lock the LUN. */
11431         mtx_lock(&softc->ctl_lock);
11432         if (targ_lun > ctl_max_luns ||
11433             (lun = softc->ctl_luns[targ_lun]) == NULL) {
11434                 mtx_unlock(&softc->ctl_lock);
11435                 return;
11436         }
11437         mtx_lock(&lun->lun_lock);
11438         mtx_unlock(&softc->ctl_lock);
11439         if (lun->flags & CTL_LUN_DISABLED) {
11440                 mtx_unlock(&lun->lun_lock);
11441                 return;
11442         }
11443
11444         if (softc->ha_mode == CTL_HA_MODE_XFER) {
11445                 LIST_FOREACH_SAFE(io, &lun->ooa_queue, ooa_links, next_io) {
11446                         /* We are master */
11447                         if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
11448                                 if (io->flags & CTL_FLAG_IO_ACTIVE) {
11449                                         io->flags |= CTL_FLAG_ABORT |
11450                                             CTL_FLAG_FAILOVER;
11451                                         ctl_try_unblock_io(lun,
11452                                             (union ctl_io *)io, FALSE);
11453                                 } else { /* This can be only due to DATAMOVE */
11454                                         io->msg_type = CTL_MSG_DATAMOVE_DONE;
11455                                         io->flags &= ~CTL_FLAG_DMA_INPROG;
11456                                         io->flags |= CTL_FLAG_IO_ACTIVE;
11457                                         io->port_status = 31340;
11458                                         ctl_enqueue_isc((union ctl_io *)io);
11459                                 }
11460                         } else
11461                         /* We are slave */
11462                         if (io->flags & CTL_FLAG_SENT_2OTHER_SC) {
11463                                 io->flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11464                                 if (io->flags & CTL_FLAG_IO_ACTIVE) {
11465                                         io->flags |= CTL_FLAG_FAILOVER;
11466                                 } else {
11467                                         ctl_set_busy(&((union ctl_io *)io)->
11468                                             scsiio);
11469                                         ctl_done((union ctl_io *)io);
11470                                 }
11471                         }
11472                 }
11473         } else { /* SERIALIZE modes */
11474                 LIST_FOREACH_SAFE(io, &lun->ooa_queue, ooa_links, next_io) {
11475                         /* We are master */
11476                         if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
11477                                 if (io->blocker != NULL) {
11478                                         TAILQ_REMOVE(&io->blocker->io_hdr.blocked_queue,
11479                                             io, blocked_links);
11480                                         io->blocker = NULL;
11481                                 }
11482                                 ctl_try_unblock_others(lun, (union ctl_io *)io,
11483                                     TRUE);
11484                                 LIST_REMOVE(io, ooa_links);
11485                                 ctl_free_io((union ctl_io *)io);
11486                         } else
11487                         /* We are slave */
11488                         if (io->flags & CTL_FLAG_SENT_2OTHER_SC) {
11489                                 io->flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11490                                 if (!(io->flags & CTL_FLAG_IO_ACTIVE)) {
11491                                         ctl_set_busy(&((union ctl_io *)io)->
11492                                             scsiio);
11493                                         ctl_done((union ctl_io *)io);
11494                                 }
11495                         }
11496                 }
11497         }
11498         mtx_unlock(&lun->lun_lock);
11499 }
11500
11501 static void
11502 ctl_scsiio_precheck(struct ctl_scsiio *ctsio)
11503 {
11504         struct ctl_softc *softc = CTL_SOFTC(ctsio);
11505         struct ctl_lun *lun;
11506         const struct ctl_cmd_entry *entry;
11507         union ctl_io *bio;
11508         uint32_t initidx, targ_lun;
11509
11510         lun = NULL;
11511         targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
11512         if (targ_lun < ctl_max_luns)
11513                 lun = softc->ctl_luns[targ_lun];
11514         if (lun) {
11515                 /*
11516                  * If the LUN is invalid, pretend that it doesn't exist.
11517                  * It will go away as soon as all pending I/O has been
11518                  * completed.
11519                  */
11520                 mtx_lock(&lun->lun_lock);
11521                 if (lun->flags & CTL_LUN_DISABLED) {
11522                         mtx_unlock(&lun->lun_lock);
11523                         lun = NULL;
11524                 }
11525         }
11526         CTL_LUN(ctsio) = lun;
11527         if (lun) {
11528                 CTL_BACKEND_LUN(ctsio) = lun->be_lun;
11529
11530                 /*
11531                  * Every I/O goes into the OOA queue for a particular LUN,
11532                  * and stays there until completion.
11533                  */
11534 #ifdef CTL_TIME_IO
11535                 if (LIST_EMPTY(&lun->ooa_queue))
11536                         lun->idle_time += getsbinuptime() - lun->last_busy;
11537 #endif
11538                 LIST_INSERT_HEAD(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
11539         }
11540
11541         /* Get command entry and return error if it is unsuppotyed. */
11542         entry = ctl_validate_command(ctsio);
11543         if (entry == NULL) {
11544                 if (lun)
11545                         mtx_unlock(&lun->lun_lock);
11546                 return;
11547         }
11548
11549         ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
11550         ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
11551
11552         /*
11553          * Check to see whether we can send this command to LUNs that don't
11554          * exist.  This should pretty much only be the case for inquiry
11555          * and request sense.  Further checks, below, really require having
11556          * a LUN, so we can't really check the command anymore.  Just put
11557          * it on the rtr queue.
11558          */
11559         if (lun == NULL) {
11560                 if (entry->flags & CTL_CMD_FLAG_OK_ON_NO_LUN) {
11561                         ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11562                         ctl_enqueue_rtr((union ctl_io *)ctsio);
11563                         return;
11564                 }
11565
11566                 ctl_set_unsupported_lun(ctsio);
11567                 ctl_done((union ctl_io *)ctsio);
11568                 CTL_DEBUG_PRINT(("ctl_scsiio_precheck: bailing out due to invalid LUN\n"));
11569                 return;
11570         } else {
11571                 /*
11572                  * Make sure we support this particular command on this LUN.
11573                  * e.g., we don't support writes to the control LUN.
11574                  */
11575                 if (!ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
11576                         mtx_unlock(&lun->lun_lock);
11577                         ctl_set_invalid_opcode(ctsio);
11578                         ctl_done((union ctl_io *)ctsio);
11579                         return;
11580                 }
11581         }
11582
11583         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11584
11585         /*
11586          * If we've got a request sense, it'll clear the contingent
11587          * allegiance condition.  Otherwise, if we have a CA condition for
11588          * this initiator, clear it, because it sent down a command other
11589          * than request sense.
11590          */
11591         if (ctsio->cdb[0] != REQUEST_SENSE) {
11592                 struct scsi_sense_data *ps;
11593
11594                 ps = lun->pending_sense[initidx / CTL_MAX_INIT_PER_PORT];
11595                 if (ps != NULL)
11596                         ps[initidx % CTL_MAX_INIT_PER_PORT].error_code = 0;
11597         }
11598
11599         /*
11600          * If the command has this flag set, it handles its own unit
11601          * attention reporting, we shouldn't do anything.  Otherwise we
11602          * check for any pending unit attentions, and send them back to the
11603          * initiator.  We only do this when a command initially comes in,
11604          * not when we pull it off the blocked queue.
11605          *
11606          * According to SAM-3, section 5.3.2, the order that things get
11607          * presented back to the host is basically unit attentions caused
11608          * by some sort of reset event, busy status, reservation conflicts
11609          * or task set full, and finally any other status.
11610          *
11611          * One issue here is that some of the unit attentions we report
11612          * don't fall into the "reset" category (e.g. "reported luns data
11613          * has changed").  So reporting it here, before the reservation
11614          * check, may be technically wrong.  I guess the only thing to do
11615          * would be to check for and report the reset events here, and then
11616          * check for the other unit attention types after we check for a
11617          * reservation conflict.
11618          *
11619          * XXX KDM need to fix this
11620          */
11621         if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
11622                 ctl_ua_type ua_type;
11623                 u_int sense_len = 0;
11624
11625                 ua_type = ctl_build_ua(lun, initidx, &ctsio->sense_data,
11626                     &sense_len, SSD_TYPE_NONE);
11627                 if (ua_type != CTL_UA_NONE) {
11628                         mtx_unlock(&lun->lun_lock);
11629                         ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
11630                         ctsio->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
11631                         ctsio->sense_len = sense_len;
11632                         ctl_done((union ctl_io *)ctsio);
11633                         return;
11634                 }
11635         }
11636
11637         if (ctl_scsiio_lun_check(lun, entry, ctsio) != 0) {
11638                 mtx_unlock(&lun->lun_lock);
11639                 ctl_done((union ctl_io *)ctsio);
11640                 return;
11641         }
11642
11643         /*
11644          * XXX CHD this is where we want to send IO to other side if
11645          * this LUN is secondary on this SC. We will need to make a copy
11646          * of the IO and flag the IO on this side as SENT_2OTHER and the flag
11647          * the copy we send as FROM_OTHER.
11648          * We also need to stuff the address of the original IO so we can
11649          * find it easily. Something similar will need be done on the other
11650          * side so when we are done we can find the copy.
11651          */
11652         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
11653             (lun->flags & CTL_LUN_PEER_SC_PRIMARY) != 0 &&
11654             (entry->flags & CTL_CMD_FLAG_RUN_HERE) == 0) {
11655                 union ctl_ha_msg msg_info;
11656                 int isc_retval;
11657
11658                 ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11659                 ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11660                 mtx_unlock(&lun->lun_lock);
11661
11662                 msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
11663                 msg_info.hdr.original_sc = (union ctl_io *)ctsio;
11664                 msg_info.hdr.serializing_sc = NULL;
11665                 msg_info.hdr.nexus = ctsio->io_hdr.nexus;
11666                 msg_info.scsi.tag_num = ctsio->tag_num;
11667                 msg_info.scsi.tag_type = ctsio->tag_type;
11668                 memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
11669                 msg_info.scsi.cdb_len = ctsio->cdb_len;
11670                 msg_info.scsi.priority = ctsio->priority;
11671
11672                 if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11673                     sizeof(msg_info.scsi) - sizeof(msg_info.scsi.sense_data),
11674                     M_WAITOK)) > CTL_HA_STATUS_SUCCESS) {
11675                         ctsio->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11676                         ctsio->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
11677                         ctl_set_busy(ctsio);
11678                         ctl_done((union ctl_io *)ctsio);
11679                         return;
11680                 }
11681                 return;
11682         }
11683
11684         bio = (union ctl_io *)LIST_NEXT(&ctsio->io_hdr, ooa_links);
11685         switch (ctl_check_ooa(lun, (union ctl_io *)ctsio, &bio)) {
11686         case CTL_ACTION_PASS:
11687         case CTL_ACTION_SKIP:
11688                 ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11689                 mtx_unlock(&lun->lun_lock);
11690                 ctl_enqueue_rtr((union ctl_io *)ctsio);
11691                 break;
11692         case CTL_ACTION_BLOCK:
11693                 ctsio->io_hdr.blocker = bio;
11694                 TAILQ_INSERT_TAIL(&bio->io_hdr.blocked_queue, &ctsio->io_hdr,
11695                                   blocked_links);
11696                 mtx_unlock(&lun->lun_lock);
11697                 break;
11698         case CTL_ACTION_OVERLAP:
11699                 mtx_unlock(&lun->lun_lock);
11700                 ctl_set_overlapped_cmd(ctsio);
11701                 ctl_done((union ctl_io *)ctsio);
11702                 break;
11703         case CTL_ACTION_OVERLAP_TAG:
11704                 mtx_unlock(&lun->lun_lock);
11705                 ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
11706                 ctl_done((union ctl_io *)ctsio);
11707                 break;
11708         default:
11709                 __assert_unreachable();
11710         }
11711 }
11712
11713 const struct ctl_cmd_entry *
11714 ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa)
11715 {
11716         const struct ctl_cmd_entry *entry;
11717         int service_action;
11718
11719         entry = &ctl_cmd_table[ctsio->cdb[0]];
11720         if (sa)
11721                 *sa = ((entry->flags & CTL_CMD_FLAG_SA5) != 0);
11722         if (entry->flags & CTL_CMD_FLAG_SA5) {
11723                 service_action = ctsio->cdb[1] & SERVICE_ACTION_MASK;
11724                 entry = &((const struct ctl_cmd_entry *)
11725                     entry->execute)[service_action];
11726         }
11727         return (entry);
11728 }
11729
11730 const struct ctl_cmd_entry *
11731 ctl_validate_command(struct ctl_scsiio *ctsio)
11732 {
11733         const struct ctl_cmd_entry *entry;
11734         int i, sa;
11735         uint8_t diff;
11736
11737         entry = ctl_get_cmd_entry(ctsio, &sa);
11738         ctsio->seridx = entry->seridx;
11739         if (entry->execute == NULL) {
11740                 if (sa)
11741                         ctl_set_invalid_field(ctsio,
11742                                               /*sks_valid*/ 1,
11743                                               /*command*/ 1,
11744                                               /*field*/ 1,
11745                                               /*bit_valid*/ 1,
11746                                               /*bit*/ 4);
11747                 else
11748                         ctl_set_invalid_opcode(ctsio);
11749                 ctl_done((union ctl_io *)ctsio);
11750                 return (NULL);
11751         }
11752         KASSERT(entry->length > 0,
11753             ("Not defined length for command 0x%02x/0x%02x",
11754              ctsio->cdb[0], ctsio->cdb[1]));
11755         for (i = 1; i < entry->length; i++) {
11756                 diff = ctsio->cdb[i] & ~entry->usage[i - 1];
11757                 if (diff == 0)
11758                         continue;
11759                 ctl_set_invalid_field(ctsio,
11760                                       /*sks_valid*/ 1,
11761                                       /*command*/ 1,
11762                                       /*field*/ i,
11763                                       /*bit_valid*/ 1,
11764                                       /*bit*/ fls(diff) - 1);
11765                 ctl_done((union ctl_io *)ctsio);
11766                 return (NULL);
11767         }
11768         return (entry);
11769 }
11770
11771 static int
11772 ctl_cmd_applicable(uint8_t lun_type, const struct ctl_cmd_entry *entry)
11773 {
11774
11775         switch (lun_type) {
11776         case T_DIRECT:
11777                 if ((entry->flags & CTL_CMD_FLAG_OK_ON_DIRECT) == 0)
11778                         return (0);
11779                 break;
11780         case T_PROCESSOR:
11781                 if ((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0)
11782                         return (0);
11783                 break;
11784         case T_CDROM:
11785                 if ((entry->flags & CTL_CMD_FLAG_OK_ON_CDROM) == 0)
11786                         return (0);
11787                 break;
11788         default:
11789                 return (0);
11790         }
11791         return (1);
11792 }
11793
11794 static int
11795 ctl_scsiio(struct ctl_scsiio *ctsio)
11796 {
11797         int retval;
11798         const struct ctl_cmd_entry *entry;
11799
11800         retval = CTL_RETVAL_COMPLETE;
11801
11802         CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
11803
11804         entry = ctl_get_cmd_entry(ctsio, NULL);
11805
11806         /*
11807          * If this I/O has been aborted, just send it straight to
11808          * ctl_done() without executing it.
11809          */
11810         if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
11811                 ctl_done((union ctl_io *)ctsio);
11812                 goto bailout;
11813         }
11814
11815         /*
11816          * All the checks should have been handled by ctl_scsiio_precheck().
11817          * We should be clear now to just execute the I/O.
11818          */
11819         retval = entry->execute(ctsio);
11820
11821 bailout:
11822         return (retval);
11823 }
11824
11825 static int
11826 ctl_target_reset(union ctl_io *io)
11827 {
11828         struct ctl_softc *softc = CTL_SOFTC(io);
11829         struct ctl_port *port = CTL_PORT(io);
11830         struct ctl_lun *lun;
11831         uint32_t initidx;
11832         ctl_ua_type ua_type;
11833
11834         if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11835                 union ctl_ha_msg msg_info;
11836
11837                 msg_info.hdr.nexus = io->io_hdr.nexus;
11838                 msg_info.task.task_action = io->taskio.task_action;
11839                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11840                 msg_info.hdr.original_sc = NULL;
11841                 msg_info.hdr.serializing_sc = NULL;
11842                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11843                     sizeof(msg_info.task), M_WAITOK);
11844         }
11845
11846         initidx = ctl_get_initindex(&io->io_hdr.nexus);
11847         if (io->taskio.task_action == CTL_TASK_TARGET_RESET)
11848                 ua_type = CTL_UA_TARG_RESET;
11849         else
11850                 ua_type = CTL_UA_BUS_RESET;
11851         mtx_lock(&softc->ctl_lock);
11852         STAILQ_FOREACH(lun, &softc->lun_list, links) {
11853                 if (port != NULL &&
11854                     ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
11855                         continue;
11856                 ctl_do_lun_reset(lun, initidx, ua_type);
11857         }
11858         mtx_unlock(&softc->ctl_lock);
11859         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11860         return (0);
11861 }
11862
11863 /*
11864  * The LUN should always be set.  The I/O is optional, and is used to
11865  * distinguish between I/Os sent by this initiator, and by other
11866  * initiators.  We set unit attention for initiators other than this one.
11867  * SAM-3 is vague on this point.  It does say that a unit attention should
11868  * be established for other initiators when a LUN is reset (see section
11869  * 5.7.3), but it doesn't specifically say that the unit attention should
11870  * be established for this particular initiator when a LUN is reset.  Here
11871  * is the relevant text, from SAM-3 rev 8:
11872  *
11873  * 5.7.2 When a SCSI initiator port aborts its own tasks
11874  *
11875  * When a SCSI initiator port causes its own task(s) to be aborted, no
11876  * notification that the task(s) have been aborted shall be returned to
11877  * the SCSI initiator port other than the completion response for the
11878  * command or task management function action that caused the task(s) to
11879  * be aborted and notification(s) associated with related effects of the
11880  * action (e.g., a reset unit attention condition).
11881  *
11882  * XXX KDM for now, we're setting unit attention for all initiators.
11883  */
11884 static void
11885 ctl_do_lun_reset(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua_type)
11886 {
11887         struct ctl_io_hdr *xioh;
11888         int i;
11889
11890         mtx_lock(&lun->lun_lock);
11891         /* Abort tasks. */
11892         LIST_FOREACH(xioh, &lun->ooa_queue, ooa_links) {
11893                 xioh->flags |= CTL_FLAG_ABORT | CTL_FLAG_ABORT_STATUS;
11894                 ctl_try_unblock_io(lun, (union ctl_io *)xioh, FALSE);
11895         }
11896         /* Clear CA. */
11897         for (i = 0; i < ctl_max_ports; i++) {
11898                 free(lun->pending_sense[i], M_CTL);
11899                 lun->pending_sense[i] = NULL;
11900         }
11901         /* Clear reservation. */
11902         lun->flags &= ~CTL_LUN_RESERVED;
11903         /* Clear prevent media removal. */
11904         if (lun->prevent) {
11905                 for (i = 0; i < CTL_MAX_INITIATORS; i++)
11906                         ctl_clear_mask(lun->prevent, i);
11907                 lun->prevent_count = 0;
11908         }
11909         /* Clear TPC status */
11910         ctl_tpc_lun_clear(lun, -1);
11911         /* Establish UA. */
11912 #if 0
11913         ctl_est_ua_all(lun, initidx, ua_type);
11914 #else
11915         ctl_est_ua_all(lun, -1, ua_type);
11916 #endif
11917         mtx_unlock(&lun->lun_lock);
11918 }
11919
11920 static int
11921 ctl_lun_reset(union ctl_io *io)
11922 {
11923         struct ctl_softc *softc = CTL_SOFTC(io);
11924         struct ctl_lun *lun;
11925         uint32_t targ_lun, initidx;
11926
11927         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11928         initidx = ctl_get_initindex(&io->io_hdr.nexus);
11929         mtx_lock(&softc->ctl_lock);
11930         if (targ_lun >= ctl_max_luns ||
11931             (lun = softc->ctl_luns[targ_lun]) == NULL) {
11932                 mtx_unlock(&softc->ctl_lock);
11933                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
11934                 return (1);
11935         }
11936         ctl_do_lun_reset(lun, initidx, CTL_UA_LUN_RESET);
11937         mtx_unlock(&softc->ctl_lock);
11938         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11939
11940         if ((io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) == 0) {
11941                 union ctl_ha_msg msg_info;
11942
11943                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11944                 msg_info.hdr.nexus = io->io_hdr.nexus;
11945                 msg_info.task.task_action = CTL_TASK_LUN_RESET;
11946                 msg_info.hdr.original_sc = NULL;
11947                 msg_info.hdr.serializing_sc = NULL;
11948                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11949                     sizeof(msg_info.task), M_WAITOK);
11950         }
11951         return (0);
11952 }
11953
11954 static void
11955 ctl_abort_tasks_lun(struct ctl_lun *lun, uint32_t targ_port, uint32_t init_id,
11956     int other_sc)
11957 {
11958         struct ctl_io_hdr *xioh;
11959
11960         mtx_assert(&lun->lun_lock, MA_OWNED);
11961
11962         /*
11963          * Run through the OOA queue and attempt to find the given I/O.
11964          * The target port, initiator ID, tag type and tag number have to
11965          * match the values that we got from the initiator.  If we have an
11966          * untagged command to abort, simply abort the first untagged command
11967          * we come to.  We only allow one untagged command at a time of course.
11968          */
11969         LIST_FOREACH(xioh, &lun->ooa_queue, ooa_links) {
11970                 union ctl_io *xio = (union ctl_io *)xioh;
11971                 if ((targ_port == UINT32_MAX ||
11972                      targ_port == xioh->nexus.targ_port) &&
11973                     (init_id == UINT32_MAX ||
11974                      init_id == xioh->nexus.initid)) {
11975                         if (targ_port != xioh->nexus.targ_port ||
11976                             init_id != xioh->nexus.initid)
11977                                 xioh->flags |= CTL_FLAG_ABORT_STATUS;
11978                         xioh->flags |= CTL_FLAG_ABORT;
11979                         if (!other_sc && !(lun->flags & CTL_LUN_PRIMARY_SC)) {
11980                                 union ctl_ha_msg msg_info;
11981
11982                                 msg_info.hdr.nexus = xioh->nexus;
11983                                 msg_info.task.task_action = CTL_TASK_ABORT_TASK;
11984                                 msg_info.task.tag_num = xio->scsiio.tag_num;
11985                                 msg_info.task.tag_type = xio->scsiio.tag_type;
11986                                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11987                                 msg_info.hdr.original_sc = NULL;
11988                                 msg_info.hdr.serializing_sc = NULL;
11989                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11990                                     sizeof(msg_info.task), M_NOWAIT);
11991                         }
11992                         ctl_try_unblock_io(lun, xio, FALSE);
11993                 }
11994         }
11995 }
11996
11997 static int
11998 ctl_abort_task_set(union ctl_io *io)
11999 {
12000         struct ctl_softc *softc = CTL_SOFTC(io);
12001         struct ctl_lun *lun;
12002         uint32_t targ_lun;
12003
12004         /*
12005          * Look up the LUN.
12006          */
12007         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12008         mtx_lock(&softc->ctl_lock);
12009         if (targ_lun >= ctl_max_luns ||
12010             (lun = softc->ctl_luns[targ_lun]) == NULL) {
12011                 mtx_unlock(&softc->ctl_lock);
12012                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
12013                 return (1);
12014         }
12015
12016         mtx_lock(&lun->lun_lock);
12017         mtx_unlock(&softc->ctl_lock);
12018         if (io->taskio.task_action == CTL_TASK_ABORT_TASK_SET) {
12019                 ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
12020                     io->io_hdr.nexus.initid,
12021                     (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12022         } else { /* CTL_TASK_CLEAR_TASK_SET */
12023                 ctl_abort_tasks_lun(lun, UINT32_MAX, UINT32_MAX,
12024                     (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12025         }
12026         mtx_unlock(&lun->lun_lock);
12027         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12028         return (0);
12029 }
12030
12031 static void
12032 ctl_i_t_nexus_loss(struct ctl_softc *softc, uint32_t initidx,
12033     ctl_ua_type ua_type)
12034 {
12035         struct ctl_lun *lun;
12036         struct scsi_sense_data *ps;
12037         uint32_t p, i;
12038
12039         p = initidx / CTL_MAX_INIT_PER_PORT;
12040         i = initidx % CTL_MAX_INIT_PER_PORT;
12041         mtx_lock(&softc->ctl_lock);
12042         STAILQ_FOREACH(lun, &softc->lun_list, links) {
12043                 mtx_lock(&lun->lun_lock);
12044                 /* Abort tasks. */
12045                 ctl_abort_tasks_lun(lun, p, i, 1);
12046                 /* Clear CA. */
12047                 ps = lun->pending_sense[p];
12048                 if (ps != NULL)
12049                         ps[i].error_code = 0;
12050                 /* Clear reservation. */
12051                 if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == initidx))
12052                         lun->flags &= ~CTL_LUN_RESERVED;
12053                 /* Clear prevent media removal. */
12054                 if (lun->prevent && ctl_is_set(lun->prevent, initidx)) {
12055                         ctl_clear_mask(lun->prevent, initidx);
12056                         lun->prevent_count--;
12057                 }
12058                 /* Clear TPC status */
12059                 ctl_tpc_lun_clear(lun, initidx);
12060                 /* Establish UA. */
12061                 ctl_est_ua(lun, initidx, ua_type);
12062                 mtx_unlock(&lun->lun_lock);
12063         }
12064         mtx_unlock(&softc->ctl_lock);
12065 }
12066
12067 static int
12068 ctl_i_t_nexus_reset(union ctl_io *io)
12069 {
12070         struct ctl_softc *softc = CTL_SOFTC(io);
12071         uint32_t initidx;
12072
12073         if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
12074                 union ctl_ha_msg msg_info;
12075
12076                 msg_info.hdr.nexus = io->io_hdr.nexus;
12077                 msg_info.task.task_action = CTL_TASK_I_T_NEXUS_RESET;
12078                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
12079                 msg_info.hdr.original_sc = NULL;
12080                 msg_info.hdr.serializing_sc = NULL;
12081                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
12082                     sizeof(msg_info.task), M_WAITOK);
12083         }
12084
12085         initidx = ctl_get_initindex(&io->io_hdr.nexus);
12086         ctl_i_t_nexus_loss(softc, initidx, CTL_UA_I_T_NEXUS_LOSS);
12087         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12088         return (0);
12089 }
12090
12091 static int
12092 ctl_abort_task(union ctl_io *io)
12093 {
12094         struct ctl_softc *softc = CTL_SOFTC(io);
12095         struct ctl_io_hdr *xioh;
12096         struct ctl_lun *lun;
12097         uint32_t targ_lun;
12098
12099         /*
12100          * Look up the LUN.
12101          */
12102         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12103         mtx_lock(&softc->ctl_lock);
12104         if (targ_lun >= ctl_max_luns ||
12105             (lun = softc->ctl_luns[targ_lun]) == NULL) {
12106                 mtx_unlock(&softc->ctl_lock);
12107                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
12108                 return (1);
12109         }
12110
12111         mtx_lock(&lun->lun_lock);
12112         mtx_unlock(&softc->ctl_lock);
12113         /*
12114          * Run through the OOA queue and attempt to find the given I/O.
12115          * The target port, initiator ID, tag type and tag number have to
12116          * match the values that we got from the initiator.  If we have an
12117          * untagged command to abort, simply abort the first untagged command
12118          * we come to.  We only allow one untagged command at a time of course.
12119          */
12120         LIST_FOREACH(xioh, &lun->ooa_queue, ooa_links) {
12121                 union ctl_io *xio = (union ctl_io *)xioh;
12122                 if ((xioh->nexus.targ_port != io->io_hdr.nexus.targ_port)
12123                  || (xioh->nexus.initid != io->io_hdr.nexus.initid)
12124                  || (xioh->flags & CTL_FLAG_ABORT))
12125                         continue;
12126
12127                 /*
12128                  * If the abort says that the task is untagged, the
12129                  * task in the queue must be untagged.  Otherwise,
12130                  * we just check to see whether the tag numbers
12131                  * match.  This is because the QLogic firmware
12132                  * doesn't pass back the tag type in an abort
12133                  * request.
12134                  */
12135 #if 0
12136                 if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
12137                   && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
12138                  || (xio->scsiio.tag_num == io->taskio.tag_num)) {
12139 #else
12140                 /*
12141                  * XXX KDM we've got problems with FC, because it
12142                  * doesn't send down a tag type with aborts.  So we
12143                  * can only really go by the tag number...
12144                  * This may cause problems with parallel SCSI.
12145                  * Need to figure that out!!
12146                  */
12147                 if (xio->scsiio.tag_num == io->taskio.tag_num) {
12148 #endif
12149                         xioh->flags |= CTL_FLAG_ABORT;
12150                         if ((io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) == 0 &&
12151                             !(lun->flags & CTL_LUN_PRIMARY_SC)) {
12152                                 union ctl_ha_msg msg_info;
12153
12154                                 msg_info.hdr.nexus = io->io_hdr.nexus;
12155                                 msg_info.task.task_action = CTL_TASK_ABORT_TASK;
12156                                 msg_info.task.tag_num = io->taskio.tag_num;
12157                                 msg_info.task.tag_type = io->taskio.tag_type;
12158                                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
12159                                 msg_info.hdr.original_sc = NULL;
12160                                 msg_info.hdr.serializing_sc = NULL;
12161                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
12162                                     sizeof(msg_info.task), M_NOWAIT);
12163                         }
12164                         ctl_try_unblock_io(lun, xio, FALSE);
12165                 }
12166         }
12167         mtx_unlock(&lun->lun_lock);
12168         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12169         return (0);
12170 }
12171
12172 static int
12173 ctl_query_task(union ctl_io *io, int task_set)
12174 {
12175         struct ctl_softc *softc = CTL_SOFTC(io);
12176         struct ctl_io_hdr *xioh;
12177         struct ctl_lun *lun;
12178         int found = 0;
12179         uint32_t targ_lun;
12180
12181         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12182         mtx_lock(&softc->ctl_lock);
12183         if (targ_lun >= ctl_max_luns ||
12184             (lun = softc->ctl_luns[targ_lun]) == NULL) {
12185                 mtx_unlock(&softc->ctl_lock);
12186                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
12187                 return (1);
12188         }
12189         mtx_lock(&lun->lun_lock);
12190         mtx_unlock(&softc->ctl_lock);
12191         LIST_FOREACH(xioh, &lun->ooa_queue, ooa_links) {
12192                 union ctl_io *xio = (union ctl_io *)xioh;
12193                 if ((xioh->nexus.targ_port != io->io_hdr.nexus.targ_port)
12194                  || (xioh->nexus.initid != io->io_hdr.nexus.initid)
12195                  || (xioh->flags & CTL_FLAG_ABORT))
12196                         continue;
12197
12198                 if (task_set || xio->scsiio.tag_num == io->taskio.tag_num) {
12199                         found = 1;
12200                         break;
12201                 }
12202         }
12203         mtx_unlock(&lun->lun_lock);
12204         if (found)
12205                 io->taskio.task_status = CTL_TASK_FUNCTION_SUCCEEDED;
12206         else
12207                 io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12208         return (0);
12209 }
12210
12211 static int
12212 ctl_query_async_event(union ctl_io *io)
12213 {
12214         struct ctl_softc *softc = CTL_SOFTC(io);
12215         struct ctl_lun *lun;
12216         ctl_ua_type ua;
12217         uint32_t targ_lun, initidx;
12218
12219         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12220         mtx_lock(&softc->ctl_lock);
12221         if (targ_lun >= ctl_max_luns ||
12222             (lun = softc->ctl_luns[targ_lun]) == NULL) {
12223                 mtx_unlock(&softc->ctl_lock);
12224                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
12225                 return (1);
12226         }
12227         mtx_lock(&lun->lun_lock);
12228         mtx_unlock(&softc->ctl_lock);
12229         initidx = ctl_get_initindex(&io->io_hdr.nexus);
12230         ua = ctl_build_qae(lun, initidx, io->taskio.task_resp);
12231         mtx_unlock(&lun->lun_lock);
12232         if (ua != CTL_UA_NONE)
12233                 io->taskio.task_status = CTL_TASK_FUNCTION_SUCCEEDED;
12234         else
12235                 io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12236         return (0);
12237 }
12238
12239 static void
12240 ctl_run_task(union ctl_io *io)
12241 {
12242         int retval = 1;
12243
12244         CTL_DEBUG_PRINT(("ctl_run_task\n"));
12245         KASSERT(io->io_hdr.io_type == CTL_IO_TASK,
12246             ("ctl_run_task: Unextected io_type %d\n", io->io_hdr.io_type));
12247         io->taskio.task_status = CTL_TASK_FUNCTION_NOT_SUPPORTED;
12248         bzero(io->taskio.task_resp, sizeof(io->taskio.task_resp));
12249         switch (io->taskio.task_action) {
12250         case CTL_TASK_ABORT_TASK:
12251                 retval = ctl_abort_task(io);
12252                 break;
12253         case CTL_TASK_ABORT_TASK_SET:
12254         case CTL_TASK_CLEAR_TASK_SET:
12255                 retval = ctl_abort_task_set(io);
12256                 break;
12257         case CTL_TASK_CLEAR_ACA:
12258                 break;
12259         case CTL_TASK_I_T_NEXUS_RESET:
12260                 retval = ctl_i_t_nexus_reset(io);
12261                 break;
12262         case CTL_TASK_LUN_RESET:
12263                 retval = ctl_lun_reset(io);
12264                 break;
12265         case CTL_TASK_TARGET_RESET:
12266         case CTL_TASK_BUS_RESET:
12267                 retval = ctl_target_reset(io);
12268                 break;
12269         case CTL_TASK_PORT_LOGIN:
12270                 break;
12271         case CTL_TASK_PORT_LOGOUT:
12272                 break;
12273         case CTL_TASK_QUERY_TASK:
12274                 retval = ctl_query_task(io, 0);
12275                 break;
12276         case CTL_TASK_QUERY_TASK_SET:
12277                 retval = ctl_query_task(io, 1);
12278                 break;
12279         case CTL_TASK_QUERY_ASYNC_EVENT:
12280                 retval = ctl_query_async_event(io);
12281                 break;
12282         default:
12283                 printf("%s: got unknown task management event %d\n",
12284                        __func__, io->taskio.task_action);
12285                 break;
12286         }
12287         if (retval == 0)
12288                 io->io_hdr.status = CTL_SUCCESS;
12289         else
12290                 io->io_hdr.status = CTL_ERROR;
12291         ctl_done(io);
12292 }
12293
12294 /*
12295  * For HA operation.  Handle commands that come in from the other
12296  * controller.
12297  */
12298 static void
12299 ctl_handle_isc(union ctl_io *io)
12300 {
12301         struct ctl_softc *softc = CTL_SOFTC(io);
12302         struct ctl_lun *lun;
12303         const struct ctl_cmd_entry *entry;
12304         uint32_t targ_lun;
12305
12306         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12307         switch (io->io_hdr.msg_type) {
12308         case CTL_MSG_SERIALIZE:
12309                 ctl_serialize_other_sc_cmd(&io->scsiio);
12310                 break;
12311         case CTL_MSG_R2R:               /* Only used in SER_ONLY mode. */
12312                 entry = ctl_get_cmd_entry(&io->scsiio, NULL);
12313                 if (targ_lun >= ctl_max_luns ||
12314                     (lun = softc->ctl_luns[targ_lun]) == NULL) {
12315                         ctl_done(io);
12316                         break;
12317                 }
12318                 mtx_lock(&lun->lun_lock);
12319                 if (ctl_scsiio_lun_check(lun, entry, &io->scsiio) != 0) {
12320                         mtx_unlock(&lun->lun_lock);
12321                         ctl_done(io);
12322                         break;
12323                 }
12324                 io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
12325                 mtx_unlock(&lun->lun_lock);
12326                 ctl_enqueue_rtr(io);
12327                 break;
12328         case CTL_MSG_FINISH_IO:
12329                 if (softc->ha_mode == CTL_HA_MODE_XFER) {
12330                         ctl_done(io);
12331                         break;
12332                 }
12333                 if (targ_lun >= ctl_max_luns ||
12334                     (lun = softc->ctl_luns[targ_lun]) == NULL) {
12335                         ctl_free_io(io);
12336                         break;
12337                 }
12338                 mtx_lock(&lun->lun_lock);
12339                 ctl_try_unblock_others(lun, io, TRUE);
12340                 LIST_REMOVE(&io->io_hdr, ooa_links);
12341                 mtx_unlock(&lun->lun_lock);
12342                 ctl_free_io(io);
12343                 break;
12344         case CTL_MSG_PERS_ACTION:
12345                 ctl_hndl_per_res_out_on_other_sc(io);
12346                 ctl_free_io(io);
12347                 break;
12348         case CTL_MSG_BAD_JUJU:
12349                 ctl_done(io);
12350                 break;
12351         case CTL_MSG_DATAMOVE:          /* Only used in XFER mode */
12352                 ctl_datamove_remote(io);
12353                 break;
12354         case CTL_MSG_DATAMOVE_DONE:     /* Only used in XFER mode */
12355                 ctl_datamove_done(io, false);
12356                 break;
12357         case CTL_MSG_FAILOVER:
12358                 ctl_failover_lun(io);
12359                 ctl_free_io(io);
12360                 break;
12361         default:
12362                 printf("%s: Invalid message type %d\n",
12363                        __func__, io->io_hdr.msg_type);
12364                 ctl_free_io(io);
12365                 break;
12366         }
12367
12368 }
12369
12370 /*
12371  * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
12372  * there is no match.
12373  */
12374 static ctl_lun_error_pattern
12375 ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
12376 {
12377         const struct ctl_cmd_entry *entry;
12378         ctl_lun_error_pattern filtered_pattern, pattern;
12379
12380         pattern = desc->error_pattern;
12381
12382         /*
12383          * XXX KDM we need more data passed into this function to match a
12384          * custom pattern, and we actually need to implement custom pattern
12385          * matching.
12386          */
12387         if (pattern & CTL_LUN_PAT_CMD)
12388                 return (CTL_LUN_PAT_CMD);
12389
12390         if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
12391                 return (CTL_LUN_PAT_ANY);
12392
12393         entry = ctl_get_cmd_entry(ctsio, NULL);
12394
12395         filtered_pattern = entry->pattern & pattern;
12396
12397         /*
12398          * If the user requested specific flags in the pattern (e.g.
12399          * CTL_LUN_PAT_RANGE), make sure the command supports all of those
12400          * flags.
12401          *
12402          * If the user did not specify any flags, it doesn't matter whether
12403          * or not the command supports the flags.
12404          */
12405         if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
12406              (pattern & ~CTL_LUN_PAT_MASK))
12407                 return (CTL_LUN_PAT_NONE);
12408
12409         /*
12410          * If the user asked for a range check, see if the requested LBA
12411          * range overlaps with this command's LBA range.
12412          */
12413         if (filtered_pattern & CTL_LUN_PAT_RANGE) {
12414                 uint64_t lba1;
12415                 uint64_t len1;
12416                 ctl_action action;
12417                 int retval;
12418
12419                 retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
12420                 if (retval != 0)
12421                         return (CTL_LUN_PAT_NONE);
12422
12423                 action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
12424                                               desc->lba_range.len, FALSE);
12425                 /*
12426                  * A "pass" means that the LBA ranges don't overlap, so
12427                  * this doesn't match the user's range criteria.
12428                  */
12429                 if (action == CTL_ACTION_PASS)
12430                         return (CTL_LUN_PAT_NONE);
12431         }
12432
12433         return (filtered_pattern);
12434 }
12435
12436 static void
12437 ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
12438 {
12439         struct ctl_error_desc *desc, *desc2;
12440
12441         mtx_assert(&lun->lun_lock, MA_OWNED);
12442
12443         STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
12444                 ctl_lun_error_pattern pattern;
12445                 /*
12446                  * Check to see whether this particular command matches
12447                  * the pattern in the descriptor.
12448                  */
12449                 pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
12450                 if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
12451                         continue;
12452
12453                 switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
12454                 case CTL_LUN_INJ_ABORTED:
12455                         ctl_set_aborted(&io->scsiio);
12456                         break;
12457                 case CTL_LUN_INJ_MEDIUM_ERR:
12458                         ctl_set_medium_error(&io->scsiio,
12459                             (io->io_hdr.flags & CTL_FLAG_DATA_MASK) !=
12460                              CTL_FLAG_DATA_OUT);
12461                         break;
12462                 case CTL_LUN_INJ_UA:
12463                         /* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
12464                          * OCCURRED */
12465                         ctl_set_ua(&io->scsiio, 0x29, 0x00);
12466                         break;
12467                 case CTL_LUN_INJ_CUSTOM:
12468                         /*
12469                          * We're assuming the user knows what he is doing.
12470                          * Just copy the sense information without doing
12471                          * checks.
12472                          */
12473                         bcopy(&desc->custom_sense, &io->scsiio.sense_data,
12474                               MIN(sizeof(desc->custom_sense),
12475                                   sizeof(io->scsiio.sense_data)));
12476                         io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
12477                         io->scsiio.sense_len = SSD_FULL_SIZE;
12478                         io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
12479                         break;
12480                 case CTL_LUN_INJ_NONE:
12481                 default:
12482                         /*
12483                          * If this is an error injection type we don't know
12484                          * about, clear the continuous flag (if it is set)
12485                          * so it will get deleted below.
12486                          */
12487                         desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
12488                         break;
12489                 }
12490                 /*
12491                  * By default, each error injection action is a one-shot
12492                  */
12493                 if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
12494                         continue;
12495
12496                 STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
12497
12498                 free(desc, M_CTL);
12499         }
12500 }
12501
12502 #ifdef CTL_IO_DELAY
12503 static void
12504 ctl_datamove_timer_wakeup(void *arg)
12505 {
12506         union ctl_io *io;
12507
12508         io = (union ctl_io *)arg;
12509
12510         ctl_datamove(io);
12511 }
12512 #endif /* CTL_IO_DELAY */
12513
12514 static void
12515 ctl_datamove_done_process(union ctl_io *io)
12516 {
12517 #ifdef CTL_TIME_IO
12518         struct bintime cur_bt;
12519 #endif
12520
12521         KASSERT(io->io_hdr.io_type == CTL_IO_SCSI,
12522             ("%s: unexpected I/O type %x", __func__, io->io_hdr.io_type));
12523
12524 #ifdef CTL_TIME_IO
12525         getbinuptime(&cur_bt);
12526         bintime_sub(&cur_bt, &io->io_hdr.dma_start_bt);
12527         bintime_add(&io->io_hdr.dma_bt, &cur_bt);
12528 #endif
12529         io->io_hdr.num_dmas++;
12530
12531         if ((io->io_hdr.port_status != 0) &&
12532             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
12533              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
12534                 ctl_set_internal_failure(&io->scsiio, /*sks_valid*/ 1,
12535                     /*retry_count*/ io->io_hdr.port_status);
12536         } else if (io->scsiio.kern_data_resid != 0 &&
12537             (io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT &&
12538             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
12539              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
12540                 ctl_set_invalid_field_ciu(&io->scsiio);
12541         } else if (ctl_debug & CTL_DEBUG_CDB_DATA)
12542                 ctl_data_print(io);
12543 }
12544
12545 void
12546 ctl_datamove_done(union ctl_io *io, bool samethr)
12547 {
12548
12549         ctl_datamove_done_process(io);
12550         io->scsiio.be_move_done(io, samethr);
12551 }
12552
12553 void
12554 ctl_datamove(union ctl_io *io)
12555 {
12556         void (*fe_datamove)(union ctl_io *io);
12557
12558         mtx_assert(&((struct ctl_softc *)CTL_SOFTC(io))->ctl_lock, MA_NOTOWNED);
12559
12560         CTL_DEBUG_PRINT(("ctl_datamove\n"));
12561
12562         /* No data transferred yet.  Frontend must update this when done. */
12563         io->scsiio.kern_data_resid = io->scsiio.kern_data_len;
12564
12565 #ifdef CTL_TIME_IO
12566         getbinuptime(&io->io_hdr.dma_start_bt);
12567 #endif /* CTL_TIME_IO */
12568
12569 #ifdef CTL_IO_DELAY
12570         if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
12571                 io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
12572         } else {
12573                 struct ctl_lun *lun;
12574
12575                 lun = CTL_LUN(io);
12576                 if ((lun != NULL)
12577                  && (lun->delay_info.datamove_delay > 0)) {
12578                         callout_init(&io->io_hdr.delay_callout, /*mpsafe*/ 1);
12579                         io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
12580                         callout_reset(&io->io_hdr.delay_callout,
12581                                       lun->delay_info.datamove_delay * hz,
12582                                       ctl_datamove_timer_wakeup, io);
12583                         if (lun->delay_info.datamove_type ==
12584                             CTL_DELAY_TYPE_ONESHOT)
12585                                 lun->delay_info.datamove_delay = 0;
12586                         return;
12587                 }
12588         }
12589 #endif
12590
12591         /*
12592          * This command has been aborted.  Set the port status, so we fail
12593          * the data move.
12594          */
12595         if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12596                 printf("ctl_datamove: tag 0x%jx on (%u:%u:%u) aborted\n",
12597                        io->scsiio.tag_num, io->io_hdr.nexus.initid,
12598                        io->io_hdr.nexus.targ_port,
12599                        io->io_hdr.nexus.targ_lun);
12600                 io->io_hdr.port_status = 31337;
12601                 ctl_datamove_done_process(io);
12602                 io->scsiio.be_move_done(io, true);
12603                 return;
12604         }
12605
12606         /* Don't confuse frontend with zero length data move. */
12607         if (io->scsiio.kern_data_len == 0) {
12608                 ctl_datamove_done_process(io);
12609                 io->scsiio.be_move_done(io, true);
12610                 return;
12611         }
12612
12613         fe_datamove = CTL_PORT(io)->fe_datamove;
12614         fe_datamove(io);
12615 }
12616
12617 static void
12618 ctl_send_datamove_done(union ctl_io *io, int have_lock)
12619 {
12620         union ctl_ha_msg msg;
12621 #ifdef CTL_TIME_IO
12622         struct bintime cur_bt;
12623 #endif
12624
12625         memset(&msg, 0, sizeof(msg));
12626         msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
12627         msg.hdr.original_sc = io;
12628         msg.hdr.serializing_sc = io->io_hdr.remote_io;
12629         msg.hdr.nexus = io->io_hdr.nexus;
12630         msg.hdr.status = io->io_hdr.status;
12631         msg.scsi.kern_data_resid = io->scsiio.kern_data_resid;
12632         msg.scsi.tag_num = io->scsiio.tag_num;
12633         msg.scsi.tag_type = io->scsiio.tag_type;
12634         msg.scsi.scsi_status = io->scsiio.scsi_status;
12635         memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
12636                io->scsiio.sense_len);
12637         msg.scsi.sense_len = io->scsiio.sense_len;
12638         msg.scsi.port_status = io->io_hdr.port_status;
12639         io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12640         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12641                 ctl_failover_io(io, /*have_lock*/ have_lock);
12642                 return;
12643         }
12644         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12645             sizeof(msg.scsi) - sizeof(msg.scsi.sense_data) +
12646             msg.scsi.sense_len, M_WAITOK);
12647
12648 #ifdef CTL_TIME_IO
12649         getbinuptime(&cur_bt);
12650         bintime_sub(&cur_bt, &io->io_hdr.dma_start_bt);
12651         bintime_add(&io->io_hdr.dma_bt, &cur_bt);
12652 #endif
12653         io->io_hdr.num_dmas++;
12654 }
12655
12656 /*
12657  * The DMA to the remote side is done, now we need to tell the other side
12658  * we're done so it can continue with its data movement.
12659  */
12660 static void
12661 ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
12662 {
12663         union ctl_io *io;
12664         uint32_t i;
12665
12666         io = rq->context;
12667
12668         if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12669                 printf("%s: ISC DMA write failed with error %d", __func__,
12670                        rq->ret);
12671                 ctl_set_internal_failure(&io->scsiio,
12672                                          /*sks_valid*/ 1,
12673                                          /*retry_count*/ rq->ret);
12674         }
12675
12676         ctl_dt_req_free(rq);
12677
12678         for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12679                 free(CTL_LSGLT(io)[i].addr, M_CTL);
12680         free(CTL_RSGL(io), M_CTL);
12681         CTL_RSGL(io) = NULL;
12682         CTL_LSGL(io) = NULL;
12683
12684         /*
12685          * The data is in local and remote memory, so now we need to send
12686          * status (good or back) back to the other side.
12687          */
12688         ctl_send_datamove_done(io, /*have_lock*/ 0);
12689 }
12690
12691 /*
12692  * We've moved the data from the host/controller into local memory.  Now we
12693  * need to push it over to the remote controller's memory.
12694  */
12695 static int
12696 ctl_datamove_remote_dm_write_cb(union ctl_io *io, bool samethr)
12697 {
12698         int retval;
12699
12700         retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
12701                                           ctl_datamove_remote_write_cb);
12702         return (retval);
12703 }
12704
12705 static void
12706 ctl_datamove_remote_write(union ctl_io *io)
12707 {
12708         int retval;
12709         void (*fe_datamove)(union ctl_io *io);
12710
12711         /*
12712          * - Get the data from the host/HBA into local memory.
12713          * - DMA memory from the local controller to the remote controller.
12714          * - Send status back to the remote controller.
12715          */
12716
12717         retval = ctl_datamove_remote_sgl_setup(io);
12718         if (retval != 0)
12719                 return;
12720
12721         /* Switch the pointer over so the FETD knows what to do */
12722         io->scsiio.kern_data_ptr = (uint8_t *)CTL_LSGL(io);
12723
12724         /*
12725          * Use a custom move done callback, since we need to send completion
12726          * back to the other controller, not to the backend on this side.
12727          */
12728         io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
12729
12730         fe_datamove = CTL_PORT(io)->fe_datamove;
12731         fe_datamove(io);
12732 }
12733
12734 static int
12735 ctl_datamove_remote_dm_read_cb(union ctl_io *io, bool samethr)
12736 {
12737         uint32_t i;
12738
12739         for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12740                 free(CTL_LSGLT(io)[i].addr, M_CTL);
12741         free(CTL_RSGL(io), M_CTL);
12742         CTL_RSGL(io) = NULL;
12743         CTL_LSGL(io) = NULL;
12744
12745         /*
12746          * The read is done, now we need to send status (good or bad) back
12747          * to the other side.
12748          */
12749         ctl_send_datamove_done(io, /*have_lock*/ 0);
12750
12751         return (0);
12752 }
12753
12754 static void
12755 ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
12756 {
12757         union ctl_io *io;
12758         void (*fe_datamove)(union ctl_io *io);
12759
12760         io = rq->context;
12761
12762         if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12763                 printf("%s: ISC DMA read failed with error %d\n", __func__,
12764                        rq->ret);
12765                 ctl_set_internal_failure(&io->scsiio,
12766                                          /*sks_valid*/ 1,
12767                                          /*retry_count*/ rq->ret);
12768         }
12769
12770         ctl_dt_req_free(rq);
12771
12772         /* Switch the pointer over so the FETD knows what to do */
12773         io->scsiio.kern_data_ptr = (uint8_t *)CTL_LSGL(io);
12774
12775         /*
12776          * Use a custom move done callback, since we need to send completion
12777          * back to the other controller, not to the backend on this side.
12778          */
12779         io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
12780
12781         /* XXX KDM add checks like the ones in ctl_datamove? */
12782
12783         fe_datamove = CTL_PORT(io)->fe_datamove;
12784         fe_datamove(io);
12785 }
12786
12787 static int
12788 ctl_datamove_remote_sgl_setup(union ctl_io *io)
12789 {
12790         struct ctl_sg_entry *local_sglist;
12791         uint32_t len_to_go;
12792         int retval;
12793         int i;
12794
12795         retval = 0;
12796         local_sglist = CTL_LSGL(io);
12797         len_to_go = io->scsiio.kern_data_len;
12798
12799         /*
12800          * The difficult thing here is that the size of the various
12801          * S/G segments may be different than the size from the
12802          * remote controller.  That'll make it harder when DMAing
12803          * the data back to the other side.
12804          */
12805         for (i = 0; len_to_go > 0; i++) {
12806                 local_sglist[i].len = MIN(len_to_go, CTL_HA_DATAMOVE_SEGMENT);
12807                 local_sglist[i].addr =
12808                     malloc(local_sglist[i].len, M_CTL, M_WAITOK);
12809
12810                 len_to_go -= local_sglist[i].len;
12811         }
12812         /*
12813          * Reset the number of S/G entries accordingly.  The original
12814          * number of S/G entries is available in rem_sg_entries.
12815          */
12816         io->scsiio.kern_sg_entries = i;
12817
12818         return (retval);
12819 }
12820
12821 static int
12822 ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
12823                          ctl_ha_dt_cb callback)
12824 {
12825         struct ctl_ha_dt_req *rq;
12826         struct ctl_sg_entry *remote_sglist, *local_sglist;
12827         uint32_t local_used, remote_used, total_used;
12828         int i, j, isc_ret;
12829
12830         rq = ctl_dt_req_alloc();
12831
12832         /*
12833          * If we failed to allocate the request, and if the DMA didn't fail
12834          * anyway, set busy status.  This is just a resource allocation
12835          * failure.
12836          */
12837         if ((rq == NULL)
12838          && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
12839              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS))
12840                 ctl_set_busy(&io->scsiio);
12841
12842         if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
12843             (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS) {
12844                 if (rq != NULL)
12845                         ctl_dt_req_free(rq);
12846
12847                 /*
12848                  * The data move failed.  We need to return status back
12849                  * to the other controller.  No point in trying to DMA
12850                  * data to the remote controller.
12851                  */
12852
12853                 ctl_send_datamove_done(io, /*have_lock*/ 0);
12854
12855                 return (1);
12856         }
12857
12858         local_sglist = CTL_LSGL(io);
12859         remote_sglist = CTL_RSGL(io);
12860         local_used = 0;
12861         remote_used = 0;
12862         total_used = 0;
12863
12864         /*
12865          * Pull/push the data over the wire from/to the other controller.
12866          * This takes into account the possibility that the local and
12867          * remote sglists may not be identical in terms of the size of
12868          * the elements and the number of elements.
12869          *
12870          * One fundamental assumption here is that the length allocated for
12871          * both the local and remote sglists is identical.  Otherwise, we've
12872          * essentially got a coding error of some sort.
12873          */
12874         isc_ret = CTL_HA_STATUS_SUCCESS;
12875         for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
12876                 uint32_t cur_len;
12877                 uint8_t *tmp_ptr;
12878
12879                 rq->command = command;
12880                 rq->context = io;
12881
12882                 /*
12883                  * Both pointers should be aligned.  But it is possible
12884                  * that the allocation length is not.  They should both
12885                  * also have enough slack left over at the end, though,
12886                  * to round up to the next 8 byte boundary.
12887                  */
12888                 cur_len = MIN(local_sglist[i].len - local_used,
12889                               remote_sglist[j].len - remote_used);
12890                 rq->size = cur_len;
12891
12892                 tmp_ptr = (uint8_t *)local_sglist[i].addr;
12893                 tmp_ptr += local_used;
12894
12895 #if 0
12896                 /* Use physical addresses when talking to ISC hardware */
12897                 if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
12898                         /* XXX KDM use busdma */
12899                         rq->local = vtophys(tmp_ptr);
12900                 } else
12901                         rq->local = tmp_ptr;
12902 #else
12903                 KASSERT((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0,
12904                     ("HA does not support BUS_ADDR"));
12905                 rq->local = tmp_ptr;
12906 #endif
12907
12908                 tmp_ptr = (uint8_t *)remote_sglist[j].addr;
12909                 tmp_ptr += remote_used;
12910                 rq->remote = tmp_ptr;
12911
12912                 rq->callback = NULL;
12913
12914                 local_used += cur_len;
12915                 if (local_used >= local_sglist[i].len) {
12916                         i++;
12917                         local_used = 0;
12918                 }
12919
12920                 remote_used += cur_len;
12921                 if (remote_used >= remote_sglist[j].len) {
12922                         j++;
12923                         remote_used = 0;
12924                 }
12925                 total_used += cur_len;
12926
12927                 if (total_used >= io->scsiio.kern_data_len)
12928                         rq->callback = callback;
12929
12930                 isc_ret = ctl_dt_single(rq);
12931                 if (isc_ret > CTL_HA_STATUS_SUCCESS)
12932                         break;
12933         }
12934         if (isc_ret != CTL_HA_STATUS_WAIT) {
12935                 rq->ret = isc_ret;
12936                 callback(rq);
12937         }
12938
12939         return (0);
12940 }
12941
12942 static void
12943 ctl_datamove_remote_read(union ctl_io *io)
12944 {
12945         int retval;
12946         uint32_t i;
12947
12948         /*
12949          * This will send an error to the other controller in the case of a
12950          * failure.
12951          */
12952         retval = ctl_datamove_remote_sgl_setup(io);
12953         if (retval != 0)
12954                 return;
12955
12956         retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
12957                                           ctl_datamove_remote_read_cb);
12958         if (retval != 0) {
12959                 /*
12960                  * Make sure we free memory if there was an error..  The
12961                  * ctl_datamove_remote_xfer() function will send the
12962                  * datamove done message, or call the callback with an
12963                  * error if there is a problem.
12964                  */
12965                 for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12966                         free(CTL_LSGLT(io)[i].addr, M_CTL);
12967                 free(CTL_RSGL(io), M_CTL);
12968                 CTL_RSGL(io) = NULL;
12969                 CTL_LSGL(io) = NULL;
12970         }
12971 }
12972
12973 /*
12974  * Process a datamove request from the other controller.  This is used for
12975  * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
12976  * first.  Once that is complete, the data gets DMAed into the remote
12977  * controller's memory.  For reads, we DMA from the remote controller's
12978  * memory into our memory first, and then move it out to the FETD.
12979  */
12980 static void
12981 ctl_datamove_remote(union ctl_io *io)
12982 {
12983
12984         mtx_assert(&((struct ctl_softc *)CTL_SOFTC(io))->ctl_lock, MA_NOTOWNED);
12985
12986         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12987                 ctl_failover_io(io, /*have_lock*/ 0);
12988                 return;
12989         }
12990
12991         /*
12992          * Note that we look for an aborted I/O here, but don't do some of
12993          * the other checks that ctl_datamove() normally does.
12994          * We don't need to run the datamove delay code, since that should
12995          * have been done if need be on the other controller.
12996          */
12997         if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12998                 printf("%s: tag 0x%jx on (%u:%u:%u) aborted\n", __func__,
12999                        io->scsiio.tag_num, io->io_hdr.nexus.initid,
13000                        io->io_hdr.nexus.targ_port,
13001                        io->io_hdr.nexus.targ_lun);
13002                 io->io_hdr.port_status = 31338;
13003                 ctl_send_datamove_done(io, /*have_lock*/ 0);
13004                 return;
13005         }
13006
13007         if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT)
13008                 ctl_datamove_remote_write(io);
13009         else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN)
13010                 ctl_datamove_remote_read(io);
13011         else {
13012                 io->io_hdr.port_status = 31339;
13013                 ctl_send_datamove_done(io, /*have_lock*/ 0);
13014         }
13015 }
13016
13017 static void
13018 ctl_process_done(union ctl_io *io)
13019 {
13020         struct ctl_softc *softc = CTL_SOFTC(io);
13021         struct ctl_port *port = CTL_PORT(io);
13022         struct ctl_lun *lun = CTL_LUN(io);
13023         void (*fe_done)(union ctl_io *io);
13024         union ctl_ha_msg msg;
13025
13026         CTL_DEBUG_PRINT(("ctl_process_done\n"));
13027         fe_done = port->fe_done;
13028
13029 #ifdef CTL_TIME_IO
13030         if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
13031                 char str[256];
13032                 char path_str[64];
13033                 struct sbuf sb;
13034
13035                 ctl_scsi_path_string(io, path_str, sizeof(path_str));
13036                 sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
13037
13038                 sbuf_cat(&sb, path_str);
13039                 switch (io->io_hdr.io_type) {
13040                 case CTL_IO_SCSI:
13041                         ctl_scsi_command_string(&io->scsiio, NULL, &sb);
13042                         sbuf_printf(&sb, "\n");
13043                         sbuf_cat(&sb, path_str);
13044                         sbuf_printf(&sb, "Tag: 0x%jx/%d, Prio: %d\n",
13045                                     io->scsiio.tag_num, io->scsiio.tag_type,
13046                                     io->scsiio.priority);
13047                         break;
13048                 case CTL_IO_TASK:
13049                         sbuf_printf(&sb, "Task Action: %d Tag: 0x%jx/%d\n",
13050                                     io->taskio.task_action,
13051                                     io->taskio.tag_num, io->taskio.tag_type);
13052                         break;
13053                 default:
13054                         panic("%s: Invalid CTL I/O type %d\n",
13055                             __func__, io->io_hdr.io_type);
13056                 }
13057                 sbuf_cat(&sb, path_str);
13058                 sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
13059                             (intmax_t)time_uptime - io->io_hdr.start_time);
13060                 sbuf_finish(&sb);
13061                 printf("%s", sbuf_data(&sb));
13062         }
13063 #endif /* CTL_TIME_IO */
13064
13065         switch (io->io_hdr.io_type) {
13066         case CTL_IO_SCSI:
13067                 break;
13068         case CTL_IO_TASK:
13069                 if (ctl_debug & CTL_DEBUG_INFO)
13070                         ctl_io_error_print(io, NULL);
13071                 fe_done(io);
13072                 return;
13073         default:
13074                 panic("%s: Invalid CTL I/O type %d\n",
13075                     __func__, io->io_hdr.io_type);
13076         }
13077
13078         if (lun == NULL) {
13079                 CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
13080                                  io->io_hdr.nexus.targ_mapped_lun));
13081                 goto bailout;
13082         }
13083
13084         mtx_lock(&lun->lun_lock);
13085
13086         /*
13087          * Check to see if we have any informational exception and status
13088          * of this command can be modified to report it in form of either
13089          * RECOVERED ERROR or NO SENSE, depending on MRIE mode page field.
13090          */
13091         if (lun->ie_reported == 0 && lun->ie_asc != 0 &&
13092             io->io_hdr.status == CTL_SUCCESS &&
13093             (io->io_hdr.flags & CTL_FLAG_STATUS_SENT) == 0) {
13094                 uint8_t mrie = lun->MODE_IE.mrie;
13095                 uint8_t per = ((lun->MODE_RWER.byte3 & SMS_RWER_PER) ||
13096                     (lun->MODE_VER.byte3 & SMS_VER_PER));
13097                 if (((mrie == SIEP_MRIE_REC_COND && per) ||
13098                      mrie == SIEP_MRIE_REC_UNCOND ||
13099                      mrie == SIEP_MRIE_NO_SENSE) &&
13100                     (ctl_get_cmd_entry(&io->scsiio, NULL)->flags &
13101                      CTL_CMD_FLAG_NO_SENSE) == 0) {
13102                         ctl_set_sense(&io->scsiio,
13103                               /*current_error*/ 1,
13104                               /*sense_key*/ (mrie == SIEP_MRIE_NO_SENSE) ?
13105                                 SSD_KEY_NO_SENSE : SSD_KEY_RECOVERED_ERROR,
13106                               /*asc*/ lun->ie_asc,
13107                               /*ascq*/ lun->ie_ascq,
13108                               SSD_ELEM_NONE);
13109                         lun->ie_reported = 1;
13110                 }
13111         } else if (lun->ie_reported < 0)
13112                 lun->ie_reported = 0;
13113
13114         /*
13115          * Check to see if we have any errors to inject here.  We only
13116          * inject errors for commands that don't already have errors set.
13117          */
13118         if (!STAILQ_EMPTY(&lun->error_list) &&
13119             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS) &&
13120             ((io->io_hdr.flags & CTL_FLAG_STATUS_SENT) == 0))
13121                 ctl_inject_error(lun, io);
13122
13123         /*
13124          * XXX KDM how do we treat commands that aren't completed
13125          * successfully?
13126          *
13127          * XXX KDM should we also track I/O latency?
13128          */
13129         if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS &&
13130             io->io_hdr.io_type == CTL_IO_SCSI) {
13131                 int type;
13132 #ifdef CTL_TIME_IO
13133                 struct bintime bt;
13134
13135                 getbinuptime(&bt);
13136                 bintime_sub(&bt, &io->io_hdr.start_bt);
13137 #endif
13138                 if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13139                     CTL_FLAG_DATA_IN)
13140                         type = CTL_STATS_READ;
13141                 else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13142                     CTL_FLAG_DATA_OUT)
13143                         type = CTL_STATS_WRITE;
13144                 else
13145                         type = CTL_STATS_NO_IO;
13146
13147                 lun->stats.bytes[type] += io->scsiio.kern_total_len;
13148                 lun->stats.operations[type] ++;
13149                 lun->stats.dmas[type] += io->io_hdr.num_dmas;
13150 #ifdef CTL_TIME_IO
13151                 bintime_add(&lun->stats.dma_time[type], &io->io_hdr.dma_bt);
13152                 bintime_add(&lun->stats.time[type], &bt);
13153 #endif
13154
13155                 mtx_lock(&port->port_lock);
13156                 port->stats.bytes[type] += io->scsiio.kern_total_len;
13157                 port->stats.operations[type] ++;
13158                 port->stats.dmas[type] += io->io_hdr.num_dmas;
13159 #ifdef CTL_TIME_IO
13160                 bintime_add(&port->stats.dma_time[type], &io->io_hdr.dma_bt);
13161                 bintime_add(&port->stats.time[type], &bt);
13162 #endif
13163                 mtx_unlock(&port->port_lock);
13164         }
13165
13166         /*
13167          * Run through the blocked queue of this I/O and see if anything
13168          * can be unblocked, now that this I/O is done and will be removed.
13169          * We need to do it before removal to have OOA position to start.
13170          */
13171         ctl_try_unblock_others(lun, io, TRUE);
13172
13173         /*
13174          * Remove this from the OOA queue.
13175          */
13176         LIST_REMOVE(&io->io_hdr, ooa_links);
13177 #ifdef CTL_TIME_IO
13178         if (LIST_EMPTY(&lun->ooa_queue))
13179                 lun->last_busy = getsbinuptime();
13180 #endif
13181
13182         /*
13183          * If the LUN has been invalidated, free it if there is nothing
13184          * left on its OOA queue.
13185          */
13186         if ((lun->flags & CTL_LUN_INVALID)
13187          && LIST_EMPTY(&lun->ooa_queue)) {
13188                 mtx_unlock(&lun->lun_lock);
13189                 ctl_free_lun(lun);
13190         } else
13191                 mtx_unlock(&lun->lun_lock);
13192
13193 bailout:
13194
13195         /*
13196          * If this command has been aborted, make sure we set the status
13197          * properly.  The FETD is responsible for freeing the I/O and doing
13198          * whatever it needs to do to clean up its state.
13199          */
13200         if (io->io_hdr.flags & CTL_FLAG_ABORT)
13201                 ctl_set_task_aborted(&io->scsiio);
13202
13203         /*
13204          * If enabled, print command error status.
13205          */
13206         if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS &&
13207             (ctl_debug & CTL_DEBUG_INFO) != 0)
13208                 ctl_io_error_print(io, NULL);
13209
13210         /*
13211          * Tell the FETD or the other shelf controller we're done with this
13212          * command.  Note that only SCSI commands get to this point.  Task
13213          * management commands are completed above.
13214          */
13215         if ((softc->ha_mode != CTL_HA_MODE_XFER) &&
13216             (io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)) {
13217                 memset(&msg, 0, sizeof(msg));
13218                 msg.hdr.msg_type = CTL_MSG_FINISH_IO;
13219                 msg.hdr.serializing_sc = io->io_hdr.remote_io;
13220                 msg.hdr.nexus = io->io_hdr.nexus;
13221                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13222                     sizeof(msg.scsi) - sizeof(msg.scsi.sense_data),
13223                     M_WAITOK);
13224         }
13225
13226         fe_done(io);
13227 }
13228
13229 /*
13230  * Front end should call this if it doesn't do autosense.  When the request
13231  * sense comes back in from the initiator, we'll dequeue this and send it.
13232  */
13233 int
13234 ctl_queue_sense(union ctl_io *io)
13235 {
13236         struct ctl_softc *softc = CTL_SOFTC(io);
13237         struct ctl_port *port = CTL_PORT(io);
13238         struct ctl_lun *lun;
13239         struct scsi_sense_data *ps;
13240         uint32_t initidx, p, targ_lun;
13241
13242         CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
13243
13244         targ_lun = ctl_lun_map_from_port(port, io->io_hdr.nexus.targ_lun);
13245
13246         /*
13247          * LUN lookup will likely move to the ctl_work_thread() once we
13248          * have our new queueing infrastructure (that doesn't put things on
13249          * a per-LUN queue initially).  That is so that we can handle
13250          * things like an INQUIRY to a LUN that we don't have enabled.  We
13251          * can't deal with that right now.
13252          * If we don't have a LUN for this, just toss the sense information.
13253          */
13254         mtx_lock(&softc->ctl_lock);
13255         if (targ_lun >= ctl_max_luns ||
13256             (lun = softc->ctl_luns[targ_lun]) == NULL) {
13257                 mtx_unlock(&softc->ctl_lock);
13258                 goto bailout;
13259         }
13260         mtx_lock(&lun->lun_lock);
13261         mtx_unlock(&softc->ctl_lock);
13262
13263         initidx = ctl_get_initindex(&io->io_hdr.nexus);
13264         p = initidx / CTL_MAX_INIT_PER_PORT;
13265         if (lun->pending_sense[p] == NULL) {
13266                 lun->pending_sense[p] = malloc(sizeof(*ps) * CTL_MAX_INIT_PER_PORT,
13267                     M_CTL, M_NOWAIT | M_ZERO);
13268         }
13269         if ((ps = lun->pending_sense[p]) != NULL) {
13270                 ps += initidx % CTL_MAX_INIT_PER_PORT;
13271                 memset(ps, 0, sizeof(*ps));
13272                 memcpy(ps, &io->scsiio.sense_data, io->scsiio.sense_len);
13273         }
13274         mtx_unlock(&lun->lun_lock);
13275
13276 bailout:
13277         ctl_free_io(io);
13278         return (CTL_RETVAL_COMPLETE);
13279 }
13280
13281 /*
13282  * Primary command inlet from frontend ports.  All SCSI and task I/O
13283  * requests must go through this function.
13284  */
13285 int
13286 ctl_queue(union ctl_io *io)
13287 {
13288         struct ctl_port *port = CTL_PORT(io);
13289
13290         CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
13291
13292 #ifdef CTL_TIME_IO
13293         io->io_hdr.start_time = time_uptime;
13294         getbinuptime(&io->io_hdr.start_bt);
13295 #endif /* CTL_TIME_IO */
13296
13297         /* Map FE-specific LUN ID into global one. */
13298         io->io_hdr.nexus.targ_mapped_lun =
13299             ctl_lun_map_from_port(port, io->io_hdr.nexus.targ_lun);
13300
13301         switch (io->io_hdr.io_type) {
13302         case CTL_IO_SCSI:
13303         case CTL_IO_TASK:
13304                 if (ctl_debug & CTL_DEBUG_CDB)
13305                         ctl_io_print(io);
13306                 ctl_enqueue_incoming(io);
13307                 break;
13308         default:
13309                 printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
13310                 return (EINVAL);
13311         }
13312
13313         return (CTL_RETVAL_COMPLETE);
13314 }
13315
13316 int
13317 ctl_run(union ctl_io *io)
13318 {
13319         struct ctl_port *port = CTL_PORT(io);
13320
13321         CTL_DEBUG_PRINT(("ctl_run cdb[0]=%02X\n", io->scsiio.cdb[0]));
13322
13323 #ifdef CTL_TIME_IO
13324         io->io_hdr.start_time = time_uptime;
13325         getbinuptime(&io->io_hdr.start_bt);
13326 #endif /* CTL_TIME_IO */
13327
13328         /* Map FE-specific LUN ID into global one. */
13329         io->io_hdr.nexus.targ_mapped_lun =
13330             ctl_lun_map_from_port(port, io->io_hdr.nexus.targ_lun);
13331
13332         switch (io->io_hdr.io_type) {
13333         case CTL_IO_SCSI:
13334                 if (ctl_debug & CTL_DEBUG_CDB)
13335                         ctl_io_print(io);
13336                 ctl_scsiio_precheck(&io->scsiio);
13337                 break;
13338         case CTL_IO_TASK:
13339                 if (ctl_debug & CTL_DEBUG_CDB)
13340                         ctl_io_print(io);
13341                 ctl_run_task(io);
13342                 break;
13343         default:
13344                 printf("ctl_run: unknown I/O type %d\n", io->io_hdr.io_type);
13345                 return (EINVAL);
13346         }
13347
13348         return (CTL_RETVAL_COMPLETE);
13349 }
13350
13351 #ifdef CTL_IO_DELAY
13352 static void
13353 ctl_done_timer_wakeup(void *arg)
13354 {
13355         union ctl_io *io;
13356
13357         io = (union ctl_io *)arg;
13358         ctl_done(io);
13359 }
13360 #endif /* CTL_IO_DELAY */
13361
13362 void
13363 ctl_serseq_done(union ctl_io *io)
13364 {
13365         struct ctl_lun *lun = CTL_LUN(io);
13366
13367         /* This is racy, but should not be a problem. */
13368         if (!TAILQ_EMPTY(&io->io_hdr.blocked_queue)) {
13369                 mtx_lock(&lun->lun_lock);
13370                 io->io_hdr.flags |= CTL_FLAG_SERSEQ_DONE;
13371                 ctl_try_unblock_others(lun, io, FALSE);
13372                 mtx_unlock(&lun->lun_lock);
13373         } else
13374                 io->io_hdr.flags |= CTL_FLAG_SERSEQ_DONE;
13375 }
13376
13377 void
13378 ctl_done(union ctl_io *io)
13379 {
13380
13381         /*
13382          * Enable this to catch duplicate completion issues.
13383          */
13384 #if 0
13385         if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
13386                 printf("%s: type %d msg %d cdb %x iptl: "
13387                        "%u:%u:%u tag 0x%04x "
13388                        "flag %#x status %x\n",
13389                         __func__,
13390                         io->io_hdr.io_type,
13391                         io->io_hdr.msg_type,
13392                         io->scsiio.cdb[0],
13393                         io->io_hdr.nexus.initid,
13394                         io->io_hdr.nexus.targ_port,
13395                         io->io_hdr.nexus.targ_lun,
13396                         (io->io_hdr.io_type ==
13397                         CTL_IO_TASK) ?
13398                         io->taskio.tag_num :
13399                         io->scsiio.tag_num,
13400                         io->io_hdr.flags,
13401                         io->io_hdr.status);
13402         } else
13403                 io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
13404 #endif
13405
13406         /*
13407          * This is an internal copy of an I/O, and should not go through
13408          * the normal done processing logic.
13409          */
13410         if (io->io_hdr.flags & CTL_FLAG_INT_COPY)
13411                 return;
13412
13413 #ifdef CTL_IO_DELAY
13414         if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
13415                 io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
13416         } else {
13417                 struct ctl_lun *lun = CTL_LUN(io);
13418
13419                 if ((lun != NULL)
13420                  && (lun->delay_info.done_delay > 0)) {
13421                         callout_init(&io->io_hdr.delay_callout, /*mpsafe*/ 1);
13422                         io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
13423                         callout_reset(&io->io_hdr.delay_callout,
13424                                       lun->delay_info.done_delay * hz,
13425                                       ctl_done_timer_wakeup, io);
13426                         if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
13427                                 lun->delay_info.done_delay = 0;
13428                         return;
13429                 }
13430         }
13431 #endif /* CTL_IO_DELAY */
13432
13433         ctl_enqueue_done(io);
13434 }
13435
13436 static void
13437 ctl_work_thread(void *arg)
13438 {
13439         struct ctl_thread *thr = (struct ctl_thread *)arg;
13440         struct ctl_softc *softc = thr->ctl_softc;
13441         union ctl_io *io;
13442         int retval;
13443
13444         CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
13445         thread_lock(curthread);
13446         sched_prio(curthread, PUSER - 1);
13447         thread_unlock(curthread);
13448
13449         while (!softc->shutdown) {
13450                 /*
13451                  * We handle the queues in this order:
13452                  * - ISC
13453                  * - done queue (to free up resources, unblock other commands)
13454                  * - incoming queue
13455                  * - RtR queue
13456                  *
13457                  * If those queues are empty, we break out of the loop and
13458                  * go to sleep.
13459                  */
13460                 mtx_lock(&thr->queue_lock);
13461                 io = (union ctl_io *)STAILQ_FIRST(&thr->isc_queue);
13462                 if (io != NULL) {
13463                         STAILQ_REMOVE_HEAD(&thr->isc_queue, links);
13464                         mtx_unlock(&thr->queue_lock);
13465                         ctl_handle_isc(io);
13466                         continue;
13467                 }
13468                 io = (union ctl_io *)STAILQ_FIRST(&thr->done_queue);
13469                 if (io != NULL) {
13470                         STAILQ_REMOVE_HEAD(&thr->done_queue, links);
13471                         /* clear any blocked commands, call fe_done */
13472                         mtx_unlock(&thr->queue_lock);
13473                         ctl_process_done(io);
13474                         continue;
13475                 }
13476                 io = (union ctl_io *)STAILQ_FIRST(&thr->incoming_queue);
13477                 if (io != NULL) {
13478                         STAILQ_REMOVE_HEAD(&thr->incoming_queue, links);
13479                         mtx_unlock(&thr->queue_lock);
13480                         if (io->io_hdr.io_type == CTL_IO_TASK)
13481                                 ctl_run_task(io);
13482                         else
13483                                 ctl_scsiio_precheck(&io->scsiio);
13484                         continue;
13485                 }
13486                 io = (union ctl_io *)STAILQ_FIRST(&thr->rtr_queue);
13487                 if (io != NULL) {
13488                         STAILQ_REMOVE_HEAD(&thr->rtr_queue, links);
13489                         mtx_unlock(&thr->queue_lock);
13490                         retval = ctl_scsiio(&io->scsiio);
13491                         if (retval != CTL_RETVAL_COMPLETE)
13492                                 CTL_DEBUG_PRINT(("ctl_scsiio failed\n"));
13493                         continue;
13494                 }
13495
13496                 /* Sleep until we have something to do. */
13497                 mtx_sleep(thr, &thr->queue_lock, PDROP, "-", 0);
13498         }
13499         thr->thread = NULL;
13500         kthread_exit();
13501 }
13502
13503 static void
13504 ctl_thresh_thread(void *arg)
13505 {
13506         struct ctl_softc *softc = (struct ctl_softc *)arg;
13507         struct ctl_lun *lun;
13508         struct ctl_logical_block_provisioning_page *page;
13509         const char *attr;
13510         union ctl_ha_msg msg;
13511         uint64_t thres, val;
13512         int i, e, set;
13513
13514         CTL_DEBUG_PRINT(("ctl_thresh_thread starting\n"));
13515         thread_lock(curthread);
13516         sched_prio(curthread, PUSER - 1);
13517         thread_unlock(curthread);
13518
13519         while (!softc->shutdown) {
13520                 mtx_lock(&softc->ctl_lock);
13521                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
13522                         if ((lun->flags & CTL_LUN_DISABLED) ||
13523                             (lun->flags & CTL_LUN_NO_MEDIA) ||
13524                             lun->backend->lun_attr == NULL)
13525                                 continue;
13526                         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
13527                             softc->ha_mode == CTL_HA_MODE_XFER)
13528                                 continue;
13529                         if ((lun->MODE_RWER.byte8 & SMS_RWER_LBPERE) == 0)
13530                                 continue;
13531                         e = 0;
13532                         page = &lun->MODE_LBP;
13533                         for (i = 0; i < CTL_NUM_LBP_THRESH; i++) {
13534                                 if ((page->descr[i].flags & SLBPPD_ENABLED) == 0)
13535                                         continue;
13536                                 thres = scsi_4btoul(page->descr[i].count);
13537                                 thres <<= CTL_LBP_EXPONENT;
13538                                 switch (page->descr[i].resource) {
13539                                 case 0x01:
13540                                         attr = "blocksavail";
13541                                         break;
13542                                 case 0x02:
13543                                         attr = "blocksused";
13544                                         break;
13545                                 case 0xf1:
13546                                         attr = "poolblocksavail";
13547                                         break;
13548                                 case 0xf2:
13549                                         attr = "poolblocksused";
13550                                         break;
13551                                 default:
13552                                         continue;
13553                                 }
13554                                 mtx_unlock(&softc->ctl_lock); // XXX
13555                                 val = lun->backend->lun_attr(lun->be_lun, attr);
13556                                 mtx_lock(&softc->ctl_lock);
13557                                 if (val == UINT64_MAX)
13558                                         continue;
13559                                 if ((page->descr[i].flags & SLBPPD_ARMING_MASK)
13560                                     == SLBPPD_ARMING_INC)
13561                                         e = (val >= thres);
13562                                 else
13563                                         e = (val <= thres);
13564                                 if (e)
13565                                         break;
13566                         }
13567                         mtx_lock(&lun->lun_lock);
13568                         if (e) {
13569                                 scsi_u64to8b((uint8_t *)&page->descr[i] -
13570                                     (uint8_t *)page, lun->ua_tpt_info);
13571                                 if (lun->lasttpt == 0 ||
13572                                     time_uptime - lun->lasttpt >= CTL_LBP_UA_PERIOD) {
13573                                         lun->lasttpt = time_uptime;
13574                                         ctl_est_ua_all(lun, -1, CTL_UA_THIN_PROV_THRES);
13575                                         set = 1;
13576                                 } else
13577                                         set = 0;
13578                         } else {
13579                                 lun->lasttpt = 0;
13580                                 ctl_clr_ua_all(lun, -1, CTL_UA_THIN_PROV_THRES);
13581                                 set = -1;
13582                         }
13583                         mtx_unlock(&lun->lun_lock);
13584                         if (set != 0 &&
13585                             lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
13586                                 /* Send msg to other side. */
13587                                 bzero(&msg.ua, sizeof(msg.ua));
13588                                 msg.hdr.msg_type = CTL_MSG_UA;
13589                                 msg.hdr.nexus.initid = -1;
13590                                 msg.hdr.nexus.targ_port = -1;
13591                                 msg.hdr.nexus.targ_lun = lun->lun;
13592                                 msg.hdr.nexus.targ_mapped_lun = lun->lun;
13593                                 msg.ua.ua_all = 1;
13594                                 msg.ua.ua_set = (set > 0);
13595                                 msg.ua.ua_type = CTL_UA_THIN_PROV_THRES;
13596                                 memcpy(msg.ua.ua_info, lun->ua_tpt_info, 8);
13597                                 mtx_unlock(&softc->ctl_lock); // XXX
13598                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13599                                     sizeof(msg.ua), M_WAITOK);
13600                                 mtx_lock(&softc->ctl_lock);
13601                         }
13602                 }
13603                 mtx_sleep(&softc->thresh_thread, &softc->ctl_lock,
13604                     PDROP, "-", CTL_LBP_PERIOD * hz);
13605         }
13606         softc->thresh_thread = NULL;
13607         kthread_exit();
13608 }
13609
13610 static void
13611 ctl_enqueue_incoming(union ctl_io *io)
13612 {
13613         struct ctl_softc *softc = CTL_SOFTC(io);
13614         struct ctl_thread *thr;
13615         u_int idx;
13616
13617         idx = (io->io_hdr.nexus.targ_port * 127 +
13618                io->io_hdr.nexus.initid) % worker_threads;
13619         thr = &softc->threads[idx];
13620         mtx_lock(&thr->queue_lock);
13621         STAILQ_INSERT_TAIL(&thr->incoming_queue, &io->io_hdr, links);
13622         mtx_unlock(&thr->queue_lock);
13623         wakeup(thr);
13624 }
13625
13626 static void
13627 ctl_enqueue_rtr(union ctl_io *io)
13628 {
13629         struct ctl_softc *softc = CTL_SOFTC(io);
13630         struct ctl_thread *thr;
13631
13632         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13633         mtx_lock(&thr->queue_lock);
13634         STAILQ_INSERT_TAIL(&thr->rtr_queue, &io->io_hdr, links);
13635         mtx_unlock(&thr->queue_lock);
13636         wakeup(thr);
13637 }
13638
13639 static void
13640 ctl_enqueue_done(union ctl_io *io)
13641 {
13642         struct ctl_softc *softc = CTL_SOFTC(io);
13643         struct ctl_thread *thr;
13644
13645         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13646         mtx_lock(&thr->queue_lock);
13647         STAILQ_INSERT_TAIL(&thr->done_queue, &io->io_hdr, links);
13648         mtx_unlock(&thr->queue_lock);
13649         wakeup(thr);
13650 }
13651
13652 static void
13653 ctl_enqueue_isc(union ctl_io *io)
13654 {
13655         struct ctl_softc *softc = CTL_SOFTC(io);
13656         struct ctl_thread *thr;
13657
13658         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13659         mtx_lock(&thr->queue_lock);
13660         STAILQ_INSERT_TAIL(&thr->isc_queue, &io->io_hdr, links);
13661         mtx_unlock(&thr->queue_lock);
13662         wakeup(thr);
13663 }
13664
13665 /*
13666  *  vim: ts=8
13667  */