sys/cam/ctl/ctl.c

   1 /*-
   2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
   3  *
   4  * Copyright (c) 2003-2009 Silicon Graphics International Corp.
   5  * Copyright (c) 2012 The FreeBSD Foundation
   6  * Copyright (c) 2014-2017 Alexander Motin <mav@FreeBSD.org>
   7  * Copyright (c) 2017 Jakub Wojciech Klama <jceel@FreeBSD.org>
   8  * Copyright (c) 2018 Marcelo Araujo <araujo@FreeBSD.org>
   9  * All rights reserved.
  10  *
  11  * Portions of this software were developed by Edward Tomasz Napierala
  12  * under sponsorship from the FreeBSD Foundation.
  13  *
  14  * Redistribution and use in source and binary forms, with or without
  15  * modification, are permitted provided that the following conditions
  16  * are met:
  17  * 1. Redistributions of source code must retain the above copyright
  18  *    notice, this list of conditions, and the following disclaimer,
  19  *    without modification.
  20  * 2. Redistributions in binary form must reproduce at minimum a disclaimer
  21  *    substantially similar to the "NO WARRANTY" disclaimer below
  22  *    ("Disclaimer") and any redistribution must be conditioned upon
  23  *    including a substantially similar Disclaimer requirement for further
  24  *    binary redistribution.
  25  *
  26  * NO WARRANTY
  27  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  28  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  29  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR
  30  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  31  * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  32  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  33  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  34  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
  35  * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
  36  * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
  37  * POSSIBILITY OF SUCH DAMAGES.
  38  *
  39  * $Id$
  40  */
  41 /*
  42  * CAM Target Layer, a SCSI device emulation subsystem.
  43  *
  44  * Author: Ken Merry <ken@FreeBSD.org>
  45  */
  46
  47 #include <sys/cdefs.h>
  48 __FBSDID("$FreeBSD$");
  49
  50 #include <sys/param.h>
  51 #include <sys/systm.h>
  52 #include <sys/ctype.h>
  53 #include <sys/kernel.h>
  54 #include <sys/types.h>
  55 #include <sys/kthread.h>
  56 #include <sys/bio.h>
  57 #include <sys/fcntl.h>
  58 #include <sys/lock.h>
  59 #include <sys/module.h>
  60 #include <sys/mutex.h>
  61 #include <sys/condvar.h>
  62 #include <sys/malloc.h>
  63 #include <sys/conf.h>
  64 #include <sys/ioccom.h>
  65 #include <sys/queue.h>
  66 #include <sys/sbuf.h>
  67 #include <sys/smp.h>
  68 #include <sys/endian.h>
  69 #include <sys/sysctl.h>
  70 #include <sys/nv.h>
  71 #include <sys/dnv.h>
  72 #include <vm/uma.h>
  73
  74 #include <cam/cam.h>
  75 #include <cam/scsi/scsi_all.h>
  76 #include <cam/scsi/scsi_cd.h>
  77 #include <cam/scsi/scsi_da.h>
  78 #include <cam/ctl/ctl_io.h>
  79 #include <cam/ctl/ctl.h>
  80 #include <cam/ctl/ctl_frontend.h>
  81 #include <cam/ctl/ctl_util.h>
  82 #include <cam/ctl/ctl_backend.h>
  83 #include <cam/ctl/ctl_ioctl.h>
  84 #include <cam/ctl/ctl_ha.h>
  85 #include <cam/ctl/ctl_private.h>
  86 #include <cam/ctl/ctl_debug.h>
  87 #include <cam/ctl/ctl_scsi_all.h>
  88 #include <cam/ctl/ctl_error.h>
  89
  90 struct ctl_softc *control_softc = NULL;
  91
  92 /*
  93  * Template mode pages.
  94  */
  95
  96 /*
  97  * Note that these are default values only.  The actual values will be
  98  * filled in when the user does a mode sense.
  99  */
 100 const static struct scsi_da_rw_recovery_page rw_er_page_default = {
 101         /*page_code*/SMS_RW_ERROR_RECOVERY_PAGE,
 102         /*page_length*/sizeof(struct scsi_da_rw_recovery_page) - 2,
 103         /*byte3*/SMS_RWER_AWRE|SMS_RWER_ARRE,
 104         /*read_retry_count*/0,
 105         /*correction_span*/0,
 106         /*head_offset_count*/0,
 107         /*data_strobe_offset_cnt*/0,
 108         /*byte8*/SMS_RWER_LBPERE,
 109         /*write_retry_count*/0,
 110         /*reserved2*/0,
 111         /*recovery_time_limit*/{0, 0},
 112 };
 113
 114 const static struct scsi_da_rw_recovery_page rw_er_page_changeable = {
 115         /*page_code*/SMS_RW_ERROR_RECOVERY_PAGE,
 116         /*page_length*/sizeof(struct scsi_da_rw_recovery_page) - 2,
 117         /*byte3*/SMS_RWER_PER,
 118         /*read_retry_count*/0,
 119         /*correction_span*/0,
 120         /*head_offset_count*/0,
 121         /*data_strobe_offset_cnt*/0,
 122         /*byte8*/SMS_RWER_LBPERE,
 123         /*write_retry_count*/0,
 124         /*reserved2*/0,
 125         /*recovery_time_limit*/{0, 0},
 126 };
 127
 128 const static struct scsi_format_page format_page_default = {
 129         /*page_code*/SMS_FORMAT_DEVICE_PAGE,
 130         /*page_length*/sizeof(struct scsi_format_page) - 2,
 131         /*tracks_per_zone*/ {0, 0},
 132         /*alt_sectors_per_zone*/ {0, 0},
 133         /*alt_tracks_per_zone*/ {0, 0},
 134         /*alt_tracks_per_lun*/ {0, 0},
 135         /*sectors_per_track*/ {(CTL_DEFAULT_SECTORS_PER_TRACK >> 8) & 0xff,
 136                                 CTL_DEFAULT_SECTORS_PER_TRACK & 0xff},
 137         /*bytes_per_sector*/ {0, 0},
 138         /*interleave*/ {0, 0},
 139         /*track_skew*/ {0, 0},
 140         /*cylinder_skew*/ {0, 0},
 141         /*flags*/ SFP_HSEC,
 142         /*reserved*/ {0, 0, 0}
 143 };
 144
 145 const static struct scsi_format_page format_page_changeable = {
 146         /*page_code*/SMS_FORMAT_DEVICE_PAGE,
 147         /*page_length*/sizeof(struct scsi_format_page) - 2,
 148         /*tracks_per_zone*/ {0, 0},
 149         /*alt_sectors_per_zone*/ {0, 0},
 150         /*alt_tracks_per_zone*/ {0, 0},
 151         /*alt_tracks_per_lun*/ {0, 0},
 152         /*sectors_per_track*/ {0, 0},
 153         /*bytes_per_sector*/ {0, 0},
 154         /*interleave*/ {0, 0},
 155         /*track_skew*/ {0, 0},
 156         /*cylinder_skew*/ {0, 0},
 157         /*flags*/ 0,
 158         /*reserved*/ {0, 0, 0}
 159 };
 160
 161 const static struct scsi_rigid_disk_page rigid_disk_page_default = {
 162         /*page_code*/SMS_RIGID_DISK_PAGE,
 163         /*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
 164         /*cylinders*/ {0, 0, 0},
 165         /*heads*/ CTL_DEFAULT_HEADS,
 166         /*start_write_precomp*/ {0, 0, 0},
 167         /*start_reduced_current*/ {0, 0, 0},
 168         /*step_rate*/ {0, 0},
 169         /*landing_zone_cylinder*/ {0, 0, 0},
 170         /*rpl*/ SRDP_RPL_DISABLED,
 171         /*rotational_offset*/ 0,
 172         /*reserved1*/ 0,
 173         /*rotation_rate*/ {(CTL_DEFAULT_ROTATION_RATE >> 8) & 0xff,
 174                            CTL_DEFAULT_ROTATION_RATE & 0xff},
 175         /*reserved2*/ {0, 0}
 176 };
 177
 178 const static struct scsi_rigid_disk_page rigid_disk_page_changeable = {
 179         /*page_code*/SMS_RIGID_DISK_PAGE,
 180         /*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
 181         /*cylinders*/ {0, 0, 0},
 182         /*heads*/ 0,
 183         /*start_write_precomp*/ {0, 0, 0},
 184         /*start_reduced_current*/ {0, 0, 0},
 185         /*step_rate*/ {0, 0},
 186         /*landing_zone_cylinder*/ {0, 0, 0},
 187         /*rpl*/ 0,
 188         /*rotational_offset*/ 0,
 189         /*reserved1*/ 0,
 190         /*rotation_rate*/ {0, 0},
 191         /*reserved2*/ {0, 0}
 192 };
 193
 194 const static struct scsi_da_verify_recovery_page verify_er_page_default = {
 195         /*page_code*/SMS_VERIFY_ERROR_RECOVERY_PAGE,
 196         /*page_length*/sizeof(struct scsi_da_verify_recovery_page) - 2,
 197         /*byte3*/0,
 198         /*read_retry_count*/0,
 199         /*reserved*/{ 0, 0, 0, 0, 0, 0 },
 200         /*recovery_time_limit*/{0, 0},
 201 };
 202
 203 const static struct scsi_da_verify_recovery_page verify_er_page_changeable = {
 204         /*page_code*/SMS_VERIFY_ERROR_RECOVERY_PAGE,
 205         /*page_length*/sizeof(struct scsi_da_verify_recovery_page) - 2,
 206         /*byte3*/SMS_VER_PER,
 207         /*read_retry_count*/0,
 208         /*reserved*/{ 0, 0, 0, 0, 0, 0 },
 209         /*recovery_time_limit*/{0, 0},
 210 };
 211
 212 const static struct scsi_caching_page caching_page_default = {
 213         /*page_code*/SMS_CACHING_PAGE,
 214         /*page_length*/sizeof(struct scsi_caching_page) - 2,
 215         /*flags1*/ SCP_DISC | SCP_WCE,
 216         /*ret_priority*/ 0,
 217         /*disable_pf_transfer_len*/ {0xff, 0xff},
 218         /*min_prefetch*/ {0, 0},
 219         /*max_prefetch*/ {0xff, 0xff},
 220         /*max_pf_ceiling*/ {0xff, 0xff},
 221         /*flags2*/ 0,
 222         /*cache_segments*/ 0,
 223         /*cache_seg_size*/ {0, 0},
 224         /*reserved*/ 0,
 225         /*non_cache_seg_size*/ {0, 0, 0}
 226 };
 227
 228 const static struct scsi_caching_page caching_page_changeable = {
 229         /*page_code*/SMS_CACHING_PAGE,
 230         /*page_length*/sizeof(struct scsi_caching_page) - 2,
 231         /*flags1*/ SCP_WCE | SCP_RCD,
 232         /*ret_priority*/ 0,
 233         /*disable_pf_transfer_len*/ {0, 0},
 234         /*min_prefetch*/ {0, 0},
 235         /*max_prefetch*/ {0, 0},
 236         /*max_pf_ceiling*/ {0, 0},
 237         /*flags2*/ 0,
 238         /*cache_segments*/ 0,
 239         /*cache_seg_size*/ {0, 0},
 240         /*reserved*/ 0,
 241         /*non_cache_seg_size*/ {0, 0, 0}
 242 };
 243
 244 const static struct scsi_control_page control_page_default = {
 245         /*page_code*/SMS_CONTROL_MODE_PAGE,
 246         /*page_length*/sizeof(struct scsi_control_page) - 2,
 247         /*rlec*/0,
 248         /*queue_flags*/SCP_QUEUE_ALG_RESTRICTED,
 249         /*eca_and_aen*/0,
 250         /*flags4*/SCP_TAS,
 251         /*aen_holdoff_period*/{0, 0},
 252         /*busy_timeout_period*/{0, 0},
 253         /*extended_selftest_completion_time*/{0, 0}
 254 };
 255
 256 const static struct scsi_control_page control_page_changeable = {
 257         /*page_code*/SMS_CONTROL_MODE_PAGE,
 258         /*page_length*/sizeof(struct scsi_control_page) - 2,
 259         /*rlec*/SCP_DSENSE,
 260         /*queue_flags*/SCP_QUEUE_ALG_MASK | SCP_NUAR,
 261         /*eca_and_aen*/SCP_SWP,
 262         /*flags4*/0,
 263         /*aen_holdoff_period*/{0, 0},
 264         /*busy_timeout_period*/{0, 0},
 265         /*extended_selftest_completion_time*/{0, 0}
 266 };
 267
 268 #define CTL_CEM_LEN     (sizeof(struct scsi_control_ext_page) - 4)
 269
 270 const static struct scsi_control_ext_page control_ext_page_default = {
 271         /*page_code*/SMS_CONTROL_MODE_PAGE | SMPH_SPF,
 272         /*subpage_code*/0x01,
 273         /*page_length*/{CTL_CEM_LEN >> 8, CTL_CEM_LEN},
 274         /*flags*/0,
 275         /*prio*/0,
 276         /*max_sense*/0
 277 };
 278
 279 const static struct scsi_control_ext_page control_ext_page_changeable = {
 280         /*page_code*/SMS_CONTROL_MODE_PAGE | SMPH_SPF,
 281         /*subpage_code*/0x01,
 282         /*page_length*/{CTL_CEM_LEN >> 8, CTL_CEM_LEN},
 283         /*flags*/0,
 284         /*prio*/0,
 285         /*max_sense*/0xff
 286 };
 287
 288 const static struct scsi_info_exceptions_page ie_page_default = {
 289         /*page_code*/SMS_INFO_EXCEPTIONS_PAGE,
 290         /*page_length*/sizeof(struct scsi_info_exceptions_page) - 2,
 291         /*info_flags*/SIEP_FLAGS_EWASC,
 292         /*mrie*/SIEP_MRIE_NO,
 293         /*interval_timer*/{0, 0, 0, 0},
 294         /*report_count*/{0, 0, 0, 1}
 295 };
 296
 297 const static struct scsi_info_exceptions_page ie_page_changeable = {
 298         /*page_code*/SMS_INFO_EXCEPTIONS_PAGE,
 299         /*page_length*/sizeof(struct scsi_info_exceptions_page) - 2,
 300         /*info_flags*/SIEP_FLAGS_EWASC | SIEP_FLAGS_DEXCPT | SIEP_FLAGS_TEST |
 301             SIEP_FLAGS_LOGERR,
 302         /*mrie*/0x0f,
 303         /*interval_timer*/{0xff, 0xff, 0xff, 0xff},
 304         /*report_count*/{0xff, 0xff, 0xff, 0xff}
 305 };
 306
 307 #define CTL_LBPM_LEN    (sizeof(struct ctl_logical_block_provisioning_page) - 4)
 308
 309 const static struct ctl_logical_block_provisioning_page lbp_page_default = {{
 310         /*page_code*/SMS_INFO_EXCEPTIONS_PAGE | SMPH_SPF,
 311         /*subpage_code*/0x02,
 312         /*page_length*/{CTL_LBPM_LEN >> 8, CTL_LBPM_LEN},
 313         /*flags*/0,
 314         /*reserved*/{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
 315         /*descr*/{}},
 316         {{/*flags*/0,
 317           /*resource*/0x01,
 318           /*reserved*/{0, 0},
 319           /*count*/{0, 0, 0, 0}},
 320          {/*flags*/0,
 321           /*resource*/0x02,
 322           /*reserved*/{0, 0},
 323           /*count*/{0, 0, 0, 0}},
 324          {/*flags*/0,
 325           /*resource*/0xf1,
 326           /*reserved*/{0, 0},
 327           /*count*/{0, 0, 0, 0}},
 328          {/*flags*/0,
 329           /*resource*/0xf2,
 330           /*reserved*/{0, 0},
 331           /*count*/{0, 0, 0, 0}}
 332         }
 333 };
 334
 335 const static struct ctl_logical_block_provisioning_page lbp_page_changeable = {{
 336         /*page_code*/SMS_INFO_EXCEPTIONS_PAGE | SMPH_SPF,
 337         /*subpage_code*/0x02,
 338         /*page_length*/{CTL_LBPM_LEN >> 8, CTL_LBPM_LEN},
 339         /*flags*/SLBPP_SITUA,
 340         /*reserved*/{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
 341         /*descr*/{}},
 342         {{/*flags*/0,
 343           /*resource*/0,
 344           /*reserved*/{0, 0},
 345           /*count*/{0, 0, 0, 0}},
 346          {/*flags*/0,
 347           /*resource*/0,
 348           /*reserved*/{0, 0},
 349           /*count*/{0, 0, 0, 0}},
 350          {/*flags*/0,
 351           /*resource*/0,
 352           /*reserved*/{0, 0},
 353           /*count*/{0, 0, 0, 0}},
 354          {/*flags*/0,
 355           /*resource*/0,
 356           /*reserved*/{0, 0},
 357           /*count*/{0, 0, 0, 0}}
 358         }
 359 };
 360
 361 const static struct scsi_cddvd_capabilities_page cddvd_page_default = {
 362         /*page_code*/SMS_CDDVD_CAPS_PAGE,
 363         /*page_length*/sizeof(struct scsi_cddvd_capabilities_page) - 2,
 364         /*caps1*/0x3f,
 365         /*caps2*/0x00,
 366         /*caps3*/0xf0,
 367         /*caps4*/0x00,
 368         /*caps5*/0x29,
 369         /*caps6*/0x00,
 370         /*obsolete*/{0, 0},
 371         /*nvol_levels*/{0, 0},
 372         /*buffer_size*/{8, 0},
 373         /*obsolete2*/{0, 0},
 374         /*reserved*/0,
 375         /*digital*/0,
 376         /*obsolete3*/0,
 377         /*copy_management*/0,
 378         /*reserved2*/0,
 379         /*rotation_control*/0,
 380         /*cur_write_speed*/0,
 381         /*num_speed_descr*/0,
 382 };
 383
 384 const static struct scsi_cddvd_capabilities_page cddvd_page_changeable = {
 385         /*page_code*/SMS_CDDVD_CAPS_PAGE,
 386         /*page_length*/sizeof(struct scsi_cddvd_capabilities_page) - 2,
 387         /*caps1*/0,
 388         /*caps2*/0,
 389         /*caps3*/0,
 390         /*caps4*/0,
 391         /*caps5*/0,
 392         /*caps6*/0,
 393         /*obsolete*/{0, 0},
 394         /*nvol_levels*/{0, 0},
 395         /*buffer_size*/{0, 0},
 396         /*obsolete2*/{0, 0},
 397         /*reserved*/0,
 398         /*digital*/0,
 399         /*obsolete3*/0,
 400         /*copy_management*/0,
 401         /*reserved2*/0,
 402         /*rotation_control*/0,
 403         /*cur_write_speed*/0,
 404         /*num_speed_descr*/0,
 405 };
 406
 407 SYSCTL_NODE(_kern_cam, OID_AUTO, ctl, CTLFLAG_RD, 0, "CAM Target Layer");
 408 static int worker_threads = -1;
 409 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, worker_threads, CTLFLAG_RDTUN,
 410     &worker_threads, 1, "Number of worker threads");
 411 static int ctl_debug = CTL_DEBUG_NONE;
 412 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, debug, CTLFLAG_RWTUN,
 413     &ctl_debug, 0, "Enabled debug flags");
 414 static int ctl_lun_map_size = 1024;
 415 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, lun_map_size, CTLFLAG_RWTUN,
 416     &ctl_lun_map_size, 0, "Size of per-port LUN map (max LUN + 1)");
 417 #ifdef  CTL_TIME_IO
 418 static int ctl_time_io_secs = CTL_TIME_IO_DEFAULT_SECS;
 419 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, time_io_secs, CTLFLAG_RWTUN,
 420     &ctl_time_io_secs, 0, "Log requests taking more seconds");
 421 #endif
 422
 423 /*
 424  * Maximum number of LUNs we support.  MUST be a power of 2.
 425  */
 426 #define CTL_DEFAULT_MAX_LUNS    1024
 427 static int ctl_max_luns = CTL_DEFAULT_MAX_LUNS;
 428 TUNABLE_INT("kern.cam.ctl.max_luns", &ctl_max_luns);
 429 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, max_luns, CTLFLAG_RDTUN,
 430     &ctl_max_luns, CTL_DEFAULT_MAX_LUNS, "Maximum number of LUNs");
 431
 432 /*
 433  * Maximum number of ports registered at one time.
 434  */
 435 #define CTL_DEFAULT_MAX_PORTS           256
 436 static int ctl_max_ports = CTL_DEFAULT_MAX_PORTS;
 437 TUNABLE_INT("kern.cam.ctl.max_ports", &ctl_max_ports);
 438 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, max_ports, CTLFLAG_RDTUN,
 439     &ctl_max_ports, CTL_DEFAULT_MAX_LUNS, "Maximum number of ports");
 440
 441 /*
 442  * Maximum number of initiators we support.
 443  */
 444 #define CTL_MAX_INITIATORS      (CTL_MAX_INIT_PER_PORT * ctl_max_ports)
 445
 446 /*
 447  * Supported pages (0x00), Serial number (0x80), Device ID (0x83),
 448  * Extended INQUIRY Data (0x86), Mode Page Policy (0x87),
 449  * SCSI Ports (0x88), Third-party Copy (0x8F), Block limits (0xB0),
 450  * Block Device Characteristics (0xB1) and Logical Block Provisioning (0xB2)
 451  */
 452 #define SCSI_EVPD_NUM_SUPPORTED_PAGES   10
 453
 454 static void ctl_isc_event_handler(ctl_ha_channel chanel, ctl_ha_event event,
 455                                   int param);
 456 static void ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest);
 457 static void ctl_copy_sense_data_back(union ctl_io *src, union ctl_ha_msg *dest);
 458 static int ctl_init(void);
 459 static int ctl_shutdown(void);
 460 static int ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td);
 461 static int ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td);
 462 static void ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio);
 463 static void ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
 464                               struct ctl_ooa *ooa_hdr,
 465                               struct ctl_ooa_entry *kern_entries);
 466 static int ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
 467                      struct thread *td);
 468 static int ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
 469                          struct ctl_be_lun *be_lun);
 470 static int ctl_free_lun(struct ctl_lun *lun);
 471 static void ctl_create_lun(struct ctl_be_lun *be_lun);
 472
 473 static int ctl_do_mode_select(union ctl_io *io);
 474 static int ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun,
 475                            uint64_t res_key, uint64_t sa_res_key,
 476                            uint8_t type, uint32_t residx,
 477                            struct ctl_scsiio *ctsio,
 478                            struct scsi_per_res_out *cdb,
 479                            struct scsi_per_res_out_parms* param);
 480 static void ctl_pro_preempt_other(struct ctl_lun *lun,
 481                                   union ctl_ha_msg *msg);
 482 static void ctl_hndl_per_res_out_on_other_sc(union ctl_io *io);
 483 static int ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len);
 484 static int ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len);
 485 static int ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len);
 486 static int ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len);
 487 static int ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len);
 488 static int ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio,
 489                                          int alloc_len);
 490 static int ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio,
 491                                          int alloc_len);
 492 static int ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len);
 493 static int ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len);
 494 static int ctl_inquiry_evpd(struct ctl_scsiio *ctsio);
 495 static int ctl_inquiry_std(struct ctl_scsiio *ctsio);
 496 static int ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len);
 497 static ctl_action ctl_extent_check(union ctl_io *io1, union ctl_io *io2,
 498     bool seq);
 499 static ctl_action ctl_extent_check_seq(union ctl_io *io1, union ctl_io *io2);
 500 static ctl_action ctl_check_for_blockage(struct ctl_lun *lun,
 501     union ctl_io *pending_io, union ctl_io *ooa_io);
 502 static ctl_action ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
 503                                 union ctl_io *starting_io);
 504 static int ctl_check_blocked(struct ctl_lun *lun);
 505 static int ctl_scsiio_lun_check(struct ctl_lun *lun,
 506                                 const struct ctl_cmd_entry *entry,
 507                                 struct ctl_scsiio *ctsio);
 508 static void ctl_failover_lun(union ctl_io *io);
 509 static int ctl_scsiio_precheck(struct ctl_softc *ctl_softc,
 510                                struct ctl_scsiio *ctsio);
 511 static int ctl_scsiio(struct ctl_scsiio *ctsio);
 512
 513 static int ctl_target_reset(union ctl_io *io);
 514 static void ctl_do_lun_reset(struct ctl_lun *lun, uint32_t initidx,
 515                          ctl_ua_type ua_type);
 516 static int ctl_lun_reset(union ctl_io *io);
 517 static int ctl_abort_task(union ctl_io *io);
 518 static int ctl_abort_task_set(union ctl_io *io);
 519 static int ctl_query_task(union ctl_io *io, int task_set);
 520 static void ctl_i_t_nexus_loss(struct ctl_softc *softc, uint32_t initidx,
 521                               ctl_ua_type ua_type);
 522 static int ctl_i_t_nexus_reset(union ctl_io *io);
 523 static int ctl_query_async_event(union ctl_io *io);
 524 static void ctl_run_task(union ctl_io *io);
 525 #ifdef CTL_IO_DELAY
 526 static void ctl_datamove_timer_wakeup(void *arg);
 527 static void ctl_done_timer_wakeup(void *arg);
 528 #endif /* CTL_IO_DELAY */
 529
 530 static void ctl_send_datamove_done(union ctl_io *io, int have_lock);
 531 static void ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq);
 532 static int ctl_datamove_remote_dm_write_cb(union ctl_io *io);
 533 static void ctl_datamove_remote_write(union ctl_io *io);
 534 static int ctl_datamove_remote_dm_read_cb(union ctl_io *io);
 535 static void ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq);
 536 static int ctl_datamove_remote_sgl_setup(union ctl_io *io);
 537 static int ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
 538                                     ctl_ha_dt_cb callback);
 539 static void ctl_datamove_remote_read(union ctl_io *io);
 540 static void ctl_datamove_remote(union ctl_io *io);
 541 static void ctl_process_done(union ctl_io *io);
 542 static void ctl_lun_thread(void *arg);
 543 static void ctl_thresh_thread(void *arg);
 544 static void ctl_work_thread(void *arg);
 545 static void ctl_enqueue_incoming(union ctl_io *io);
 546 static void ctl_enqueue_rtr(union ctl_io *io);
 547 static void ctl_enqueue_done(union ctl_io *io);
 548 static void ctl_enqueue_isc(union ctl_io *io);
 549 static const struct ctl_cmd_entry *
 550     ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa);
 551 static const struct ctl_cmd_entry *
 552     ctl_validate_command(struct ctl_scsiio *ctsio);
 553 static int ctl_cmd_applicable(uint8_t lun_type,
 554     const struct ctl_cmd_entry *entry);
 555 static int ctl_ha_init(void);
 556 static int ctl_ha_shutdown(void);
 557
 558 static uint64_t ctl_get_prkey(struct ctl_lun *lun, uint32_t residx);
 559 static void ctl_clr_prkey(struct ctl_lun *lun, uint32_t residx);
 560 static void ctl_alloc_prkey(struct ctl_lun *lun, uint32_t residx);
 561 static void ctl_set_prkey(struct ctl_lun *lun, uint32_t residx, uint64_t key);
 562
 563 /*
 564  * Load the serialization table.  This isn't very pretty, but is probably
 565  * the easiest way to do it.
 566  */
 567 #include "ctl_ser_table.c"
 568
 569 /*
 570  * We only need to define open, close and ioctl routines for this driver.
 571  */
 572 static struct cdevsw ctl_cdevsw = {
 573         .d_version =    D_VERSION,
 574         .d_flags =      0,
 575         .d_open =       ctl_open,
 576         .d_close =      ctl_close,
 577         .d_ioctl =      ctl_ioctl,
 578         .d_name =       "ctl",
 579 };
 580
 581
 582 MALLOC_DEFINE(M_CTL, "ctlmem", "Memory used for CTL");
 583
 584 static int ctl_module_event_handler(module_t, int /*modeventtype_t*/, void *);
 585
 586 static moduledata_t ctl_moduledata = {
 587         "ctl",
 588         ctl_module_event_handler,
 589         NULL
 590 };
 591
 592 DECLARE_MODULE(ctl, ctl_moduledata, SI_SUB_CONFIGURE, SI_ORDER_THIRD);
 593 MODULE_VERSION(ctl, 1);
 594
 595 static struct ctl_frontend ha_frontend =
 596 {
 597         .name = "ha",
 598         .init = ctl_ha_init,
 599         .shutdown = ctl_ha_shutdown,
 600 };
 601
 602 static int
 603 ctl_ha_init(void)
 604 {
 605         struct ctl_softc *softc = control_softc;
 606
 607         if (ctl_pool_create(softc, "othersc", CTL_POOL_ENTRIES_OTHER_SC,
 608                             &softc->othersc_pool) != 0)
 609                 return (ENOMEM);
 610         if (ctl_ha_msg_init(softc) != CTL_HA_STATUS_SUCCESS) {
 611                 ctl_pool_free(softc->othersc_pool);
 612                 return (EIO);
 613         }
 614         if (ctl_ha_msg_register(CTL_HA_CHAN_CTL, ctl_isc_event_handler)
 615             != CTL_HA_STATUS_SUCCESS) {
 616                 ctl_ha_msg_destroy(softc);
 617                 ctl_pool_free(softc->othersc_pool);
 618                 return (EIO);
 619         }
 620         return (0);
 621 };
 622
 623 static int
 624 ctl_ha_shutdown(void)
 625 {
 626         struct ctl_softc *softc = control_softc;
 627         struct ctl_port *port;
 628
 629         ctl_ha_msg_shutdown(softc);
 630         if (ctl_ha_msg_deregister(CTL_HA_CHAN_CTL) != CTL_HA_STATUS_SUCCESS)
 631                 return (EIO);
 632         if (ctl_ha_msg_destroy(softc) != CTL_HA_STATUS_SUCCESS)
 633                 return (EIO);
 634         ctl_pool_free(softc->othersc_pool);
 635         while ((port = STAILQ_FIRST(&ha_frontend.port_list)) != NULL) {
 636                 ctl_port_deregister(port);
 637                 free(port->port_name, M_CTL);
 638                 free(port, M_CTL);
 639         }
 640         return (0);
 641 };
 642
 643 static void
 644 ctl_ha_datamove(union ctl_io *io)
 645 {
 646         struct ctl_lun *lun = CTL_LUN(io);
 647         struct ctl_sg_entry *sgl;
 648         union ctl_ha_msg msg;
 649         uint32_t sg_entries_sent;
 650         int do_sg_copy, i, j;
 651
 652         memset(&msg.dt, 0, sizeof(msg.dt));
 653         msg.hdr.msg_type = CTL_MSG_DATAMOVE;
 654         msg.hdr.original_sc = io->io_hdr.original_sc;
 655         msg.hdr.serializing_sc = io;
 656         msg.hdr.nexus = io->io_hdr.nexus;
 657         msg.hdr.status = io->io_hdr.status;
 658         msg.dt.flags = io->io_hdr.flags;
 659
 660         /*
 661          * We convert everything into a S/G list here.  We can't
 662          * pass by reference, only by value between controllers.
 663          * So we can't pass a pointer to the S/G list, only as many
 664          * S/G entries as we can fit in here.  If it's possible for
 665          * us to get more than CTL_HA_MAX_SG_ENTRIES S/G entries,
 666          * then we need to break this up into multiple transfers.
 667          */
 668         if (io->scsiio.kern_sg_entries == 0) {
 669                 msg.dt.kern_sg_entries = 1;
 670 #if 0
 671                 if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
 672                         msg.dt.sg_list[0].addr = io->scsiio.kern_data_ptr;
 673                 } else {
 674                         /* XXX KDM use busdma here! */
 675                         msg.dt.sg_list[0].addr =
 676                             (void *)vtophys(io->scsiio.kern_data_ptr);
 677                 }
 678 #else
 679                 KASSERT((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0,
 680                     ("HA does not support BUS_ADDR"));
 681                 msg.dt.sg_list[0].addr = io->scsiio.kern_data_ptr;
 682 #endif
 683                 msg.dt.sg_list[0].len = io->scsiio.kern_data_len;
 684                 do_sg_copy = 0;
 685         } else {
 686                 msg.dt.kern_sg_entries = io->scsiio.kern_sg_entries;
 687                 do_sg_copy = 1;
 688         }
 689
 690         msg.dt.kern_data_len = io->scsiio.kern_data_len;
 691         msg.dt.kern_total_len = io->scsiio.kern_total_len;
 692         msg.dt.kern_data_resid = io->scsiio.kern_data_resid;
 693         msg.dt.kern_rel_offset = io->scsiio.kern_rel_offset;
 694         msg.dt.sg_sequence = 0;
 695
 696         /*
 697          * Loop until we've sent all of the S/G entries.  On the
 698          * other end, we'll recompose these S/G entries into one
 699          * contiguous list before processing.
 700          */
 701         for (sg_entries_sent = 0; sg_entries_sent < msg.dt.kern_sg_entries;
 702             msg.dt.sg_sequence++) {
 703                 msg.dt.cur_sg_entries = MIN((sizeof(msg.dt.sg_list) /
 704                     sizeof(msg.dt.sg_list[0])),
 705                     msg.dt.kern_sg_entries - sg_entries_sent);
 706                 if (do_sg_copy != 0) {
 707                         sgl = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr;
 708                         for (i = sg_entries_sent, j = 0;
 709                              i < msg.dt.cur_sg_entries; i++, j++) {
 710 #if 0
 711                                 if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
 712                                         msg.dt.sg_list[j].addr = sgl[i].addr;
 713                                 } else {
 714                                         /* XXX KDM use busdma here! */
 715                                         msg.dt.sg_list[j].addr =
 716                                             (void *)vtophys(sgl[i].addr);
 717                                 }
 718 #else
 719                                 KASSERT((io->io_hdr.flags &
 720                                     CTL_FLAG_BUS_ADDR) == 0,
 721                                     ("HA does not support BUS_ADDR"));
 722                                 msg.dt.sg_list[j].addr = sgl[i].addr;
 723 #endif
 724                                 msg.dt.sg_list[j].len = sgl[i].len;
 725                         }
 726                 }
 727
 728                 sg_entries_sent += msg.dt.cur_sg_entries;
 729                 msg.dt.sg_last = (sg_entries_sent >= msg.dt.kern_sg_entries);
 730                 if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
 731                     sizeof(msg.dt) - sizeof(msg.dt.sg_list) +
 732                     sizeof(struct ctl_sg_entry) * msg.dt.cur_sg_entries,
 733                     M_WAITOK) > CTL_HA_STATUS_SUCCESS) {
 734                         io->io_hdr.port_status = 31341;
 735                         io->scsiio.be_move_done(io);
 736                         return;
 737                 }
 738                 msg.dt.sent_sg_entries = sg_entries_sent;
 739         }
 740
 741         /*
 742          * Officially handover the request from us to peer.
 743          * If failover has just happened, then we must return error.
 744          * If failover happen just after, then it is not our problem.
 745          */
 746         if (lun)
 747                 mtx_lock(&lun->lun_lock);
 748         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
 749                 if (lun)
 750                         mtx_unlock(&lun->lun_lock);
 751                 io->io_hdr.port_status = 31342;
 752                 io->scsiio.be_move_done(io);
 753                 return;
 754         }
 755         io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
 756         io->io_hdr.flags |= CTL_FLAG_DMA_INPROG;
 757         if (lun)
 758                 mtx_unlock(&lun->lun_lock);
 759 }
 760
 761 static void
 762 ctl_ha_done(union ctl_io *io)
 763 {
 764         union ctl_ha_msg msg;
 765
 766         if (io->io_hdr.io_type == CTL_IO_SCSI) {
 767                 memset(&msg, 0, sizeof(msg));
 768                 msg.hdr.msg_type = CTL_MSG_FINISH_IO;
 769                 msg.hdr.original_sc = io->io_hdr.original_sc;
 770                 msg.hdr.nexus = io->io_hdr.nexus;
 771                 msg.hdr.status = io->io_hdr.status;
 772                 msg.scsi.scsi_status = io->scsiio.scsi_status;
 773                 msg.scsi.tag_num = io->scsiio.tag_num;
 774                 msg.scsi.tag_type = io->scsiio.tag_type;
 775                 msg.scsi.sense_len = io->scsiio.sense_len;
 776                 memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
 777                     io->scsiio.sense_len);
 778                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
 779                     sizeof(msg.scsi) - sizeof(msg.scsi.sense_data) +
 780                     msg.scsi.sense_len, M_WAITOK);
 781         }
 782         ctl_free_io(io);
 783 }
 784
 785 static void
 786 ctl_isc_handler_finish_xfer(struct ctl_softc *ctl_softc,
 787                             union ctl_ha_msg *msg_info)
 788 {
 789         struct ctl_scsiio *ctsio;
 790
 791         if (msg_info->hdr.original_sc == NULL) {
 792                 printf("%s: original_sc == NULL!\n", __func__);
 793                 /* XXX KDM now what? */
 794                 return;
 795         }
 796
 797         ctsio = &msg_info->hdr.original_sc->scsiio;
 798         ctsio->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
 799         ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
 800         ctsio->io_hdr.status = msg_info->hdr.status;
 801         ctsio->scsi_status = msg_info->scsi.scsi_status;
 802         ctsio->sense_len = msg_info->scsi.sense_len;
 803         memcpy(&ctsio->sense_data, &msg_info->scsi.sense_data,
 804                msg_info->scsi.sense_len);
 805         ctl_enqueue_isc((union ctl_io *)ctsio);
 806 }
 807
 808 static void
 809 ctl_isc_handler_finish_ser_only(struct ctl_softc *ctl_softc,
 810                                 union ctl_ha_msg *msg_info)
 811 {
 812         struct ctl_scsiio *ctsio;
 813
 814         if (msg_info->hdr.serializing_sc == NULL) {
 815                 printf("%s: serializing_sc == NULL!\n", __func__);
 816                 /* XXX KDM now what? */
 817                 return;
 818         }
 819
 820         ctsio = &msg_info->hdr.serializing_sc->scsiio;
 821         ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
 822         ctl_enqueue_isc((union ctl_io *)ctsio);
 823 }
 824
 825 void
 826 ctl_isc_announce_lun(struct ctl_lun *lun)
 827 {
 828         struct ctl_softc *softc = lun->ctl_softc;
 829         union ctl_ha_msg *msg;
 830         struct ctl_ha_msg_lun_pr_key pr_key;
 831         int i, k;
 832
 833         if (softc->ha_link != CTL_HA_LINK_ONLINE)
 834                 return;
 835         mtx_lock(&lun->lun_lock);
 836         i = sizeof(msg->lun);
 837         if (lun->lun_devid)
 838                 i += lun->lun_devid->len;
 839         i += sizeof(pr_key) * lun->pr_key_count;
 840 alloc:
 841         mtx_unlock(&lun->lun_lock);
 842         msg = malloc(i, M_CTL, M_WAITOK);
 843         mtx_lock(&lun->lun_lock);
 844         k = sizeof(msg->lun);
 845         if (lun->lun_devid)
 846                 k += lun->lun_devid->len;
 847         k += sizeof(pr_key) * lun->pr_key_count;
 848         if (i < k) {
 849                 free(msg, M_CTL);
 850                 i = k;
 851                 goto alloc;
 852         }
 853         bzero(&msg->lun, sizeof(msg->lun));
 854         msg->hdr.msg_type = CTL_MSG_LUN_SYNC;
 855         msg->hdr.nexus.targ_lun = lun->lun;
 856         msg->hdr.nexus.targ_mapped_lun = lun->lun;
 857         msg->lun.flags = lun->flags;
 858         msg->lun.pr_generation = lun->pr_generation;
 859         msg->lun.pr_res_idx = lun->pr_res_idx;
 860         msg->lun.pr_res_type = lun->pr_res_type;
 861         msg->lun.pr_key_count = lun->pr_key_count;
 862         i = 0;
 863         if (lun->lun_devid) {
 864                 msg->lun.lun_devid_len = lun->lun_devid->len;
 865                 memcpy(&msg->lun.data[i], lun->lun_devid->data,
 866                     msg->lun.lun_devid_len);
 867                 i += msg->lun.lun_devid_len;
 868         }
 869         for (k = 0; k < CTL_MAX_INITIATORS; k++) {
 870                 if ((pr_key.pr_key = ctl_get_prkey(lun, k)) == 0)
 871                         continue;
 872                 pr_key.pr_iid = k;
 873                 memcpy(&msg->lun.data[i], &pr_key, sizeof(pr_key));
 874                 i += sizeof(pr_key);
 875         }
 876         mtx_unlock(&lun->lun_lock);
 877         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->port, sizeof(msg->port) + i,
 878             M_WAITOK);
 879         free(msg, M_CTL);
 880
 881         if (lun->flags & CTL_LUN_PRIMARY_SC) {
 882                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
 883                         ctl_isc_announce_mode(lun, -1,
 884                             lun->mode_pages.index[i].page_code & SMPH_PC_MASK,
 885                             lun->mode_pages.index[i].subpage);
 886                 }
 887         }
 888 }
 889
 890 void
 891 ctl_isc_announce_port(struct ctl_port *port)
 892 {
 893         struct ctl_softc *softc = port->ctl_softc;
 894         union ctl_ha_msg *msg;
 895         int i;
 896
 897         if (port->targ_port < softc->port_min ||
 898             port->targ_port >= softc->port_max ||
 899             softc->ha_link != CTL_HA_LINK_ONLINE)
 900                 return;
 901         i = sizeof(msg->port) + strlen(port->port_name) + 1;
 902         if (port->lun_map)
 903                 i += port->lun_map_size * sizeof(uint32_t);
 904         if (port->port_devid)
 905                 i += port->port_devid->len;
 906         if (port->target_devid)
 907                 i += port->target_devid->len;
 908         if (port->init_devid)
 909                 i += port->init_devid->len;
 910         msg = malloc(i, M_CTL, M_WAITOK);
 911         bzero(&msg->port, sizeof(msg->port));
 912         msg->hdr.msg_type = CTL_MSG_PORT_SYNC;
 913         msg->hdr.nexus.targ_port = port->targ_port;
 914         msg->port.port_type = port->port_type;
 915         msg->port.physical_port = port->physical_port;
 916         msg->port.virtual_port = port->virtual_port;
 917         msg->port.status = port->status;
 918         i = 0;
 919         msg->port.name_len = sprintf(&msg->port.data[i],
 920             "%d:%s", softc->ha_id, port->port_name) + 1;
 921         i += msg->port.name_len;
 922         if (port->lun_map) {
 923                 msg->port.lun_map_len = port->lun_map_size * sizeof(uint32_t);
 924                 memcpy(&msg->port.data[i], port->lun_map,
 925                     msg->port.lun_map_len);
 926                 i += msg->port.lun_map_len;
 927         }
 928         if (port->port_devid) {
 929                 msg->port.port_devid_len = port->port_devid->len;
 930                 memcpy(&msg->port.data[i], port->port_devid->data,
 931                     msg->port.port_devid_len);
 932                 i += msg->port.port_devid_len;
 933         }
 934         if (port->target_devid) {
 935                 msg->port.target_devid_len = port->target_devid->len;
 936                 memcpy(&msg->port.data[i], port->target_devid->data,
 937                     msg->port.target_devid_len);
 938                 i += msg->port.target_devid_len;
 939         }
 940         if (port->init_devid) {
 941                 msg->port.init_devid_len = port->init_devid->len;
 942                 memcpy(&msg->port.data[i], port->init_devid->data,
 943                     msg->port.init_devid_len);
 944                 i += msg->port.init_devid_len;
 945         }
 946         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->port, sizeof(msg->port) + i,
 947             M_WAITOK);
 948         free(msg, M_CTL);
 949 }
 950
 951 void
 952 ctl_isc_announce_iid(struct ctl_port *port, int iid)
 953 {
 954         struct ctl_softc *softc = port->ctl_softc;
 955         union ctl_ha_msg *msg;
 956         int i, l;
 957
 958         if (port->targ_port < softc->port_min ||
 959             port->targ_port >= softc->port_max ||
 960             softc->ha_link != CTL_HA_LINK_ONLINE)
 961                 return;
 962         mtx_lock(&softc->ctl_lock);
 963         i = sizeof(msg->iid);
 964         l = 0;
 965         if (port->wwpn_iid[iid].name)
 966                 l = strlen(port->wwpn_iid[iid].name) + 1;
 967         i += l;
 968         msg = malloc(i, M_CTL, M_NOWAIT);
 969         if (msg == NULL) {
 970                 mtx_unlock(&softc->ctl_lock);
 971                 return;
 972         }
 973         bzero(&msg->iid, sizeof(msg->iid));
 974         msg->hdr.msg_type = CTL_MSG_IID_SYNC;
 975         msg->hdr.nexus.targ_port = port->targ_port;
 976         msg->hdr.nexus.initid = iid;
 977         msg->iid.in_use = port->wwpn_iid[iid].in_use;
 978         msg->iid.name_len = l;
 979         msg->iid.wwpn = port->wwpn_iid[iid].wwpn;
 980         if (port->wwpn_iid[iid].name)
 981                 strlcpy(msg->iid.data, port->wwpn_iid[iid].name, l);
 982         mtx_unlock(&softc->ctl_lock);
 983         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->iid, i, M_NOWAIT);
 984         free(msg, M_CTL);
 985 }
 986
 987 void
 988 ctl_isc_announce_mode(struct ctl_lun *lun, uint32_t initidx,
 989     uint8_t page, uint8_t subpage)
 990 {
 991         struct ctl_softc *softc = lun->ctl_softc;
 992         union ctl_ha_msg msg;
 993         u_int i;
 994
 995         if (softc->ha_link != CTL_HA_LINK_ONLINE)
 996                 return;
 997         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
 998                 if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) ==
 999                     page && lun->mode_pages.index[i].subpage == subpage)
1000                         break;
1001         }
1002         if (i == CTL_NUM_MODE_PAGES)
1003                 return;
1004
1005         /* Don't try to replicate pages not present on this device. */
1006         if (lun->mode_pages.index[i].page_data == NULL)
1007                 return;
1008
1009         bzero(&msg.mode, sizeof(msg.mode));
1010         msg.hdr.msg_type = CTL_MSG_MODE_SYNC;
1011         msg.hdr.nexus.targ_port = initidx / CTL_MAX_INIT_PER_PORT;
1012         msg.hdr.nexus.initid = initidx % CTL_MAX_INIT_PER_PORT;
1013         msg.hdr.nexus.targ_lun = lun->lun;
1014         msg.hdr.nexus.targ_mapped_lun = lun->lun;
1015         msg.mode.page_code = page;
1016         msg.mode.subpage = subpage;
1017         msg.mode.page_len = lun->mode_pages.index[i].page_len;
1018         memcpy(msg.mode.data, lun->mode_pages.index[i].page_data,
1019             msg.mode.page_len);
1020         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg.mode, sizeof(msg.mode),
1021             M_WAITOK);
1022 }
1023
1024 static void
1025 ctl_isc_ha_link_up(struct ctl_softc *softc)
1026 {
1027         struct ctl_port *port;
1028         struct ctl_lun *lun;
1029         union ctl_ha_msg msg;
1030         int i;
1031
1032         /* Announce this node parameters to peer for validation. */
1033         msg.login.msg_type = CTL_MSG_LOGIN;
1034         msg.login.version = CTL_HA_VERSION;
1035         msg.login.ha_mode = softc->ha_mode;
1036         msg.login.ha_id = softc->ha_id;
1037         msg.login.max_luns = ctl_max_luns;
1038         msg.login.max_ports = ctl_max_ports;
1039         msg.login.max_init_per_port = CTL_MAX_INIT_PER_PORT;
1040         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg.login, sizeof(msg.login),
1041             M_WAITOK);
1042
1043         STAILQ_FOREACH(port, &softc->port_list, links) {
1044                 ctl_isc_announce_port(port);
1045                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1046                         if (port->wwpn_iid[i].in_use)
1047                                 ctl_isc_announce_iid(port, i);
1048                 }
1049         }
1050         STAILQ_FOREACH(lun, &softc->lun_list, links)
1051                 ctl_isc_announce_lun(lun);
1052 }
1053
1054 static void
1055 ctl_isc_ha_link_down(struct ctl_softc *softc)
1056 {
1057         struct ctl_port *port;
1058         struct ctl_lun *lun;
1059         union ctl_io *io;
1060         int i;
1061
1062         mtx_lock(&softc->ctl_lock);
1063         STAILQ_FOREACH(lun, &softc->lun_list, links) {
1064                 mtx_lock(&lun->lun_lock);
1065                 if (lun->flags & CTL_LUN_PEER_SC_PRIMARY) {
1066                         lun->flags &= ~CTL_LUN_PEER_SC_PRIMARY;
1067                         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
1068                 }
1069                 mtx_unlock(&lun->lun_lock);
1070
1071                 mtx_unlock(&softc->ctl_lock);
1072                 io = ctl_alloc_io(softc->othersc_pool);
1073                 mtx_lock(&softc->ctl_lock);
1074                 ctl_zero_io(io);
1075                 io->io_hdr.msg_type = CTL_MSG_FAILOVER;
1076                 io->io_hdr.nexus.targ_mapped_lun = lun->lun;
1077                 ctl_enqueue_isc(io);
1078         }
1079
1080         STAILQ_FOREACH(port, &softc->port_list, links) {
1081                 if (port->targ_port >= softc->port_min &&
1082                     port->targ_port < softc->port_max)
1083                         continue;
1084                 port->status &= ~CTL_PORT_STATUS_ONLINE;
1085                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1086                         port->wwpn_iid[i].in_use = 0;
1087                         free(port->wwpn_iid[i].name, M_CTL);
1088                         port->wwpn_iid[i].name = NULL;
1089                 }
1090         }
1091         mtx_unlock(&softc->ctl_lock);
1092 }
1093
1094 static void
1095 ctl_isc_ua(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1096 {
1097         struct ctl_lun *lun;
1098         uint32_t iid = ctl_get_initindex(&msg->hdr.nexus);
1099
1100         mtx_lock(&softc->ctl_lock);
1101         if (msg->hdr.nexus.targ_mapped_lun >= ctl_max_luns ||
1102             (lun = softc->ctl_luns[msg->hdr.nexus.targ_mapped_lun]) == NULL) {
1103                 mtx_unlock(&softc->ctl_lock);
1104                 return;
1105         }
1106         mtx_lock(&lun->lun_lock);
1107         mtx_unlock(&softc->ctl_lock);
1108         if (msg->ua.ua_type == CTL_UA_THIN_PROV_THRES && msg->ua.ua_set)
1109                 memcpy(lun->ua_tpt_info, msg->ua.ua_info, 8);
1110         if (msg->ua.ua_all) {
1111                 if (msg->ua.ua_set)
1112                         ctl_est_ua_all(lun, iid, msg->ua.ua_type);
1113                 else
1114                         ctl_clr_ua_all(lun, iid, msg->ua.ua_type);
1115         } else {
1116                 if (msg->ua.ua_set)
1117                         ctl_est_ua(lun, iid, msg->ua.ua_type);
1118                 else
1119                         ctl_clr_ua(lun, iid, msg->ua.ua_type);
1120         }
1121         mtx_unlock(&lun->lun_lock);
1122 }
1123
1124 static void
1125 ctl_isc_lun_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1126 {
1127         struct ctl_lun *lun;
1128         struct ctl_ha_msg_lun_pr_key pr_key;
1129         int i, k;
1130         ctl_lun_flags oflags;
1131         uint32_t targ_lun;
1132
1133         targ_lun = msg->hdr.nexus.targ_mapped_lun;
1134         mtx_lock(&softc->ctl_lock);
1135         if (targ_lun >= ctl_max_luns ||
1136             (lun = softc->ctl_luns[targ_lun]) == NULL) {
1137                 mtx_unlock(&softc->ctl_lock);
1138                 return;
1139         }
1140         mtx_lock(&lun->lun_lock);
1141         mtx_unlock(&softc->ctl_lock);
1142         if (lun->flags & CTL_LUN_DISABLED) {
1143                 mtx_unlock(&lun->lun_lock);
1144                 return;
1145         }
1146         i = (lun->lun_devid != NULL) ? lun->lun_devid->len : 0;
1147         if (msg->lun.lun_devid_len != i || (i > 0 &&
1148             memcmp(&msg->lun.data[0], lun->lun_devid->data, i) != 0)) {
1149                 mtx_unlock(&lun->lun_lock);
1150                 printf("%s: Received conflicting HA LUN %d\n",
1151                     __func__, targ_lun);
1152                 return;
1153         } else {
1154                 /* Record whether peer is primary. */
1155                 oflags = lun->flags;
1156                 if ((msg->lun.flags & CTL_LUN_PRIMARY_SC) &&
1157                     (msg->lun.flags & CTL_LUN_DISABLED) == 0)
1158                         lun->flags |= CTL_LUN_PEER_SC_PRIMARY;
1159                 else
1160                         lun->flags &= ~CTL_LUN_PEER_SC_PRIMARY;
1161                 if (oflags != lun->flags)
1162                         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
1163
1164                 /* If peer is primary and we are not -- use data */
1165                 if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
1166                     (lun->flags & CTL_LUN_PEER_SC_PRIMARY)) {
1167                         lun->pr_generation = msg->lun.pr_generation;
1168                         lun->pr_res_idx = msg->lun.pr_res_idx;
1169                         lun->pr_res_type = msg->lun.pr_res_type;
1170                         lun->pr_key_count = msg->lun.pr_key_count;
1171                         for (k = 0; k < CTL_MAX_INITIATORS; k++)
1172                                 ctl_clr_prkey(lun, k);
1173                         for (k = 0; k < msg->lun.pr_key_count; k++) {
1174                                 memcpy(&pr_key, &msg->lun.data[i],
1175                                     sizeof(pr_key));
1176                                 ctl_alloc_prkey(lun, pr_key.pr_iid);
1177                                 ctl_set_prkey(lun, pr_key.pr_iid,
1178                                     pr_key.pr_key);
1179                                 i += sizeof(pr_key);
1180                         }
1181                 }
1182
1183                 mtx_unlock(&lun->lun_lock);
1184                 CTL_DEBUG_PRINT(("%s: Known LUN %d, peer is %s\n",
1185                     __func__, targ_lun,
1186                     (msg->lun.flags & CTL_LUN_PRIMARY_SC) ?
1187                     "primary" : "secondary"));
1188
1189                 /* If we are primary but peer doesn't know -- notify */
1190                 if ((lun->flags & CTL_LUN_PRIMARY_SC) &&
1191                     (msg->lun.flags & CTL_LUN_PEER_SC_PRIMARY) == 0)
1192                         ctl_isc_announce_lun(lun);
1193         }
1194 }
1195
1196 static void
1197 ctl_isc_port_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1198 {
1199         struct ctl_port *port;
1200         struct ctl_lun *lun;
1201         int i, new;
1202
1203         port = softc->ctl_ports[msg->hdr.nexus.targ_port];
1204         if (port == NULL) {
1205                 CTL_DEBUG_PRINT(("%s: New port %d\n", __func__,
1206                     msg->hdr.nexus.targ_port));
1207                 new = 1;
1208                 port = malloc(sizeof(*port), M_CTL, M_WAITOK | M_ZERO);
1209                 port->frontend = &ha_frontend;
1210                 port->targ_port = msg->hdr.nexus.targ_port;
1211                 port->fe_datamove = ctl_ha_datamove;
1212                 port->fe_done = ctl_ha_done;
1213         } else if (port->frontend == &ha_frontend) {
1214                 CTL_DEBUG_PRINT(("%s: Updated port %d\n", __func__,
1215                     msg->hdr.nexus.targ_port));
1216                 new = 0;
1217         } else {
1218                 printf("%s: Received conflicting HA port %d\n",
1219                     __func__, msg->hdr.nexus.targ_port);
1220                 return;
1221         }
1222         port->port_type = msg->port.port_type;
1223         port->physical_port = msg->port.physical_port;
1224         port->virtual_port = msg->port.virtual_port;
1225         port->status = msg->port.status;
1226         i = 0;
1227         free(port->port_name, M_CTL);
1228         port->port_name = strndup(&msg->port.data[i], msg->port.name_len,
1229             M_CTL);
1230         i += msg->port.name_len;
1231         if (msg->port.lun_map_len != 0) {
1232                 if (port->lun_map == NULL ||
1233                     port->lun_map_size * sizeof(uint32_t) <
1234                     msg->port.lun_map_len) {
1235                         port->lun_map_size = 0;
1236                         free(port->lun_map, M_CTL);
1237                         port->lun_map = malloc(msg->port.lun_map_len,
1238                             M_CTL, M_WAITOK);
1239                 }
1240                 memcpy(port->lun_map, &msg->port.data[i], msg->port.lun_map_len);
1241                 port->lun_map_size = msg->port.lun_map_len / sizeof(uint32_t);
1242                 i += msg->port.lun_map_len;
1243         } else {
1244                 port->lun_map_size = 0;
1245                 free(port->lun_map, M_CTL);
1246                 port->lun_map = NULL;
1247         }
1248         if (msg->port.port_devid_len != 0) {
1249                 if (port->port_devid == NULL ||
1250                     port->port_devid->len < msg->port.port_devid_len) {
1251                         free(port->port_devid, M_CTL);
1252                         port->port_devid = malloc(sizeof(struct ctl_devid) +
1253                             msg->port.port_devid_len, M_CTL, M_WAITOK);
1254                 }
1255                 memcpy(port->port_devid->data, &msg->port.data[i],
1256                     msg->port.port_devid_len);
1257                 port->port_devid->len = msg->port.port_devid_len;
1258                 i += msg->port.port_devid_len;
1259         } else {
1260                 free(port->port_devid, M_CTL);
1261                 port->port_devid = NULL;
1262         }
1263         if (msg->port.target_devid_len != 0) {
1264                 if (port->target_devid == NULL ||
1265                     port->target_devid->len < msg->port.target_devid_len) {
1266                         free(port->target_devid, M_CTL);
1267                         port->target_devid = malloc(sizeof(struct ctl_devid) +
1268                             msg->port.target_devid_len, M_CTL, M_WAITOK);
1269                 }
1270                 memcpy(port->target_devid->data, &msg->port.data[i],
1271                     msg->port.target_devid_len);
1272                 port->target_devid->len = msg->port.target_devid_len;
1273                 i += msg->port.target_devid_len;
1274         } else {
1275                 free(port->target_devid, M_CTL);
1276                 port->target_devid = NULL;
1277         }
1278         if (msg->port.init_devid_len != 0) {
1279                 if (port->init_devid == NULL ||
1280                     port->init_devid->len < msg->port.init_devid_len) {
1281                         free(port->init_devid, M_CTL);
1282                         port->init_devid = malloc(sizeof(struct ctl_devid) +
1283                             msg->port.init_devid_len, M_CTL, M_WAITOK);
1284                 }
1285                 memcpy(port->init_devid->data, &msg->port.data[i],
1286                     msg->port.init_devid_len);
1287                 port->init_devid->len = msg->port.init_devid_len;
1288                 i += msg->port.init_devid_len;
1289         } else {
1290                 free(port->init_devid, M_CTL);
1291                 port->init_devid = NULL;
1292         }
1293         if (new) {
1294                 if (ctl_port_register(port) != 0) {
1295                         printf("%s: ctl_port_register() failed with error\n",
1296                             __func__);
1297                 }
1298         }
1299         mtx_lock(&softc->ctl_lock);
1300         STAILQ_FOREACH(lun, &softc->lun_list, links) {
1301                 if (ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
1302                         continue;
1303                 mtx_lock(&lun->lun_lock);
1304                 ctl_est_ua_all(lun, -1, CTL_UA_INQ_CHANGE);
1305                 mtx_unlock(&lun->lun_lock);
1306         }
1307         mtx_unlock(&softc->ctl_lock);
1308 }
1309
1310 static void
1311 ctl_isc_iid_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1312 {
1313         struct ctl_port *port;
1314         int iid;
1315
1316         port = softc->ctl_ports[msg->hdr.nexus.targ_port];
1317         if (port == NULL) {
1318                 printf("%s: Received IID for unknown port %d\n",
1319                     __func__, msg->hdr.nexus.targ_port);
1320                 return;
1321         }
1322         iid = msg->hdr.nexus.initid;
1323         if (port->wwpn_iid[iid].in_use != 0 &&
1324             msg->iid.in_use == 0)
1325                 ctl_i_t_nexus_loss(softc, iid, CTL_UA_POWERON);
1326         port->wwpn_iid[iid].in_use = msg->iid.in_use;
1327         port->wwpn_iid[iid].wwpn = msg->iid.wwpn;
1328         free(port->wwpn_iid[iid].name, M_CTL);
1329         if (msg->iid.name_len) {
1330                 port->wwpn_iid[iid].name = strndup(&msg->iid.data[0],
1331                     msg->iid.name_len, M_CTL);
1332         } else
1333                 port->wwpn_iid[iid].name = NULL;
1334 }
1335
1336 static void
1337 ctl_isc_login(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1338 {
1339
1340         if (msg->login.version != CTL_HA_VERSION) {
1341                 printf("CTL HA peers have different versions %d != %d\n",
1342                     msg->login.version, CTL_HA_VERSION);
1343                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1344                 return;
1345         }
1346         if (msg->login.ha_mode != softc->ha_mode) {
1347                 printf("CTL HA peers have different ha_mode %d != %d\n",
1348                     msg->login.ha_mode, softc->ha_mode);
1349                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1350                 return;
1351         }
1352         if (msg->login.ha_id == softc->ha_id) {
1353                 printf("CTL HA peers have same ha_id %d\n", msg->login.ha_id);
1354                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1355                 return;
1356         }
1357         if (msg->login.max_luns != ctl_max_luns ||
1358             msg->login.max_ports != ctl_max_ports ||
1359             msg->login.max_init_per_port != CTL_MAX_INIT_PER_PORT) {
1360                 printf("CTL HA peers have different limits\n");
1361                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1362                 return;
1363         }
1364 }
1365
1366 static void
1367 ctl_isc_mode_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1368 {
1369         struct ctl_lun *lun;
1370         u_int i;
1371         uint32_t initidx, targ_lun;
1372
1373         targ_lun = msg->hdr.nexus.targ_mapped_lun;
1374         mtx_lock(&softc->ctl_lock);
1375         if (targ_lun >= ctl_max_luns ||
1376             (lun = softc->ctl_luns[targ_lun]) == NULL) {
1377                 mtx_unlock(&softc->ctl_lock);
1378                 return;
1379         }
1380         mtx_lock(&lun->lun_lock);
1381         mtx_unlock(&softc->ctl_lock);
1382         if (lun->flags & CTL_LUN_DISABLED) {
1383                 mtx_unlock(&lun->lun_lock);
1384                 return;
1385         }
1386         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
1387                 if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) ==
1388                     msg->mode.page_code &&
1389                     lun->mode_pages.index[i].subpage == msg->mode.subpage)
1390                         break;
1391         }
1392         if (i == CTL_NUM_MODE_PAGES) {
1393                 mtx_unlock(&lun->lun_lock);
1394                 return;
1395         }
1396         memcpy(lun->mode_pages.index[i].page_data, msg->mode.data,
1397             lun->mode_pages.index[i].page_len);
1398         initidx = ctl_get_initindex(&msg->hdr.nexus);
1399         if (initidx != -1)
1400                 ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
1401         mtx_unlock(&lun->lun_lock);
1402 }
1403
1404 /*
1405  * ISC (Inter Shelf Communication) event handler.  Events from the HA
1406  * subsystem come in here.
1407  */
1408 static void
1409 ctl_isc_event_handler(ctl_ha_channel channel, ctl_ha_event event, int param)
1410 {
1411         struct ctl_softc *softc = control_softc;
1412         union ctl_io *io;
1413         struct ctl_prio *presio;
1414         ctl_ha_status isc_status;
1415
1416         CTL_DEBUG_PRINT(("CTL: Isc Msg event %d\n", event));
1417         if (event == CTL_HA_EVT_MSG_RECV) {
1418                 union ctl_ha_msg *msg, msgbuf;
1419
1420                 if (param > sizeof(msgbuf))
1421                         msg = malloc(param, M_CTL, M_WAITOK);
1422                 else
1423                         msg = &msgbuf;
1424                 isc_status = ctl_ha_msg_recv(CTL_HA_CHAN_CTL, msg, param,
1425                     M_WAITOK);
1426                 if (isc_status != CTL_HA_STATUS_SUCCESS) {
1427                         printf("%s: Error receiving message: %d\n",
1428                             __func__, isc_status);
1429                         if (msg != &msgbuf)
1430                                 free(msg, M_CTL);
1431                         return;
1432                 }
1433
1434                 CTL_DEBUG_PRINT(("CTL: msg_type %d\n", msg->msg_type));
1435                 switch (msg->hdr.msg_type) {
1436                 case CTL_MSG_SERIALIZE:
1437                         io = ctl_alloc_io(softc->othersc_pool);
1438                         ctl_zero_io(io);
1439                         // populate ctsio from msg
1440                         io->io_hdr.io_type = CTL_IO_SCSI;
1441                         io->io_hdr.msg_type = CTL_MSG_SERIALIZE;
1442                         io->io_hdr.original_sc = msg->hdr.original_sc;
1443                         io->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC |
1444                                             CTL_FLAG_IO_ACTIVE;
1445                         /*
1446                          * If we're in serialization-only mode, we don't
1447                          * want to go through full done processing.  Thus
1448                          * the COPY flag.
1449                          *
1450                          * XXX KDM add another flag that is more specific.
1451                          */
1452                         if (softc->ha_mode != CTL_HA_MODE_XFER)
1453                                 io->io_hdr.flags |= CTL_FLAG_INT_COPY;
1454                         io->io_hdr.nexus = msg->hdr.nexus;
1455 #if 0
1456                         printf("port %u, iid %u, lun %u\n",
1457                                io->io_hdr.nexus.targ_port,
1458                                io->io_hdr.nexus.initid,
1459                                io->io_hdr.nexus.targ_lun);
1460 #endif
1461                         io->scsiio.tag_num = msg->scsi.tag_num;
1462                         io->scsiio.tag_type = msg->scsi.tag_type;
1463 #ifdef CTL_TIME_IO
1464                         io->io_hdr.start_time = time_uptime;
1465                         getbinuptime(&io->io_hdr.start_bt);
1466 #endif /* CTL_TIME_IO */
1467                         io->scsiio.cdb_len = msg->scsi.cdb_len;
1468                         memcpy(io->scsiio.cdb, msg->scsi.cdb,
1469                                CTL_MAX_CDBLEN);
1470                         if (softc->ha_mode == CTL_HA_MODE_XFER) {
1471                                 const struct ctl_cmd_entry *entry;
1472
1473                                 entry = ctl_get_cmd_entry(&io->scsiio, NULL);
1474                                 io->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
1475                                 io->io_hdr.flags |=
1476                                         entry->flags & CTL_FLAG_DATA_MASK;
1477                         }
1478                         ctl_enqueue_isc(io);
1479                         break;
1480
1481                 /* Performed on the Originating SC, XFER mode only */
1482                 case CTL_MSG_DATAMOVE: {
1483                         struct ctl_sg_entry *sgl;
1484                         int i, j;
1485
1486                         io = msg->hdr.original_sc;
1487                         if (io == NULL) {
1488                                 printf("%s: original_sc == NULL!\n", __func__);
1489                                 /* XXX KDM do something here */
1490                                 break;
1491                         }
1492                         io->io_hdr.msg_type = CTL_MSG_DATAMOVE;
1493                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1494                         /*
1495                          * Keep track of this, we need to send it back over
1496                          * when the datamove is complete.
1497                          */
1498                         io->io_hdr.serializing_sc = msg->hdr.serializing_sc;
1499                         if (msg->hdr.status == CTL_SUCCESS)
1500                                 io->io_hdr.status = msg->hdr.status;
1501
1502                         if (msg->dt.sg_sequence == 0) {
1503 #ifdef CTL_TIME_IO
1504                                 getbinuptime(&io->io_hdr.dma_start_bt);
1505 #endif
1506                                 i = msg->dt.kern_sg_entries +
1507                                     msg->dt.kern_data_len /
1508                                     CTL_HA_DATAMOVE_SEGMENT + 1;
1509                                 sgl = malloc(sizeof(*sgl) * i, M_CTL,
1510                                     M_WAITOK | M_ZERO);
1511                                 io->io_hdr.remote_sglist = sgl;
1512                                 io->io_hdr.local_sglist =
1513                                     &sgl[msg->dt.kern_sg_entries];
1514
1515                                 io->scsiio.kern_data_ptr = (uint8_t *)sgl;
1516
1517                                 io->scsiio.kern_sg_entries =
1518                                         msg->dt.kern_sg_entries;
1519                                 io->scsiio.rem_sg_entries =
1520                                         msg->dt.kern_sg_entries;
1521                                 io->scsiio.kern_data_len =
1522                                         msg->dt.kern_data_len;
1523                                 io->scsiio.kern_total_len =
1524                                         msg->dt.kern_total_len;
1525                                 io->scsiio.kern_data_resid =
1526                                         msg->dt.kern_data_resid;
1527                                 io->scsiio.kern_rel_offset =
1528                                         msg->dt.kern_rel_offset;
1529                                 io->io_hdr.flags &= ~CTL_FLAG_BUS_ADDR;
1530                                 io->io_hdr.flags |= msg->dt.flags &
1531                                     CTL_FLAG_BUS_ADDR;
1532                         } else
1533                                 sgl = (struct ctl_sg_entry *)
1534                                         io->scsiio.kern_data_ptr;
1535
1536                         for (i = msg->dt.sent_sg_entries, j = 0;
1537                              i < (msg->dt.sent_sg_entries +
1538                              msg->dt.cur_sg_entries); i++, j++) {
1539                                 sgl[i].addr = msg->dt.sg_list[j].addr;
1540                                 sgl[i].len = msg->dt.sg_list[j].len;
1541
1542 #if 0
1543                                 printf("%s: DATAMOVE: %p,%lu j=%d, i=%d\n",
1544                                     __func__, sgl[i].addr, sgl[i].len, j, i);
1545 #endif
1546                         }
1547
1548                         /*
1549                          * If this is the last piece of the I/O, we've got
1550                          * the full S/G list.  Queue processing in the thread.
1551                          * Otherwise wait for the next piece.
1552                          */
1553                         if (msg->dt.sg_last != 0)
1554                                 ctl_enqueue_isc(io);
1555                         break;
1556                 }
1557                 /* Performed on the Serializing (primary) SC, XFER mode only */
1558                 case CTL_MSG_DATAMOVE_DONE: {
1559                         if (msg->hdr.serializing_sc == NULL) {
1560                                 printf("%s: serializing_sc == NULL!\n",
1561                                        __func__);
1562                                 /* XXX KDM now what? */
1563                                 break;
1564                         }
1565                         /*
1566                          * We grab the sense information here in case
1567                          * there was a failure, so we can return status
1568                          * back to the initiator.
1569                          */
1570                         io = msg->hdr.serializing_sc;
1571                         io->io_hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
1572                         io->io_hdr.flags &= ~CTL_FLAG_DMA_INPROG;
1573                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1574                         io->io_hdr.port_status = msg->scsi.port_status;
1575                         io->scsiio.kern_data_resid = msg->scsi.kern_data_resid;
1576                         if (msg->hdr.status != CTL_STATUS_NONE) {
1577                                 io->io_hdr.status = msg->hdr.status;
1578                                 io->scsiio.scsi_status = msg->scsi.scsi_status;
1579                                 io->scsiio.sense_len = msg->scsi.sense_len;
1580                                 memcpy(&io->scsiio.sense_data,
1581                                     &msg->scsi.sense_data,
1582                                     msg->scsi.sense_len);
1583                                 if (msg->hdr.status == CTL_SUCCESS)
1584                                         io->io_hdr.flags |= CTL_FLAG_STATUS_SENT;
1585                         }
1586                         ctl_enqueue_isc(io);
1587                         break;
1588                 }
1589
1590                 /* Preformed on Originating SC, SER_ONLY mode */
1591                 case CTL_MSG_R2R:
1592                         io = msg->hdr.original_sc;
1593                         if (io == NULL) {
1594                                 printf("%s: original_sc == NULL!\n",
1595                                     __func__);
1596                                 break;
1597                         }
1598                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1599                         io->io_hdr.msg_type = CTL_MSG_R2R;
1600                         io->io_hdr.serializing_sc = msg->hdr.serializing_sc;
1601                         ctl_enqueue_isc(io);
1602                         break;
1603
1604                 /*
1605                  * Performed on Serializing(i.e. primary SC) SC in SER_ONLY
1606                  * mode.
1607                  * Performed on the Originating (i.e. secondary) SC in XFER
1608                  * mode
1609                  */
1610                 case CTL_MSG_FINISH_IO:
1611                         if (softc->ha_mode == CTL_HA_MODE_XFER)
1612                                 ctl_isc_handler_finish_xfer(softc, msg);
1613                         else
1614                                 ctl_isc_handler_finish_ser_only(softc, msg);
1615                         break;
1616
1617                 /* Preformed on Originating SC */
1618                 case CTL_MSG_BAD_JUJU:
1619                         io = msg->hdr.original_sc;
1620                         if (io == NULL) {
1621                                 printf("%s: Bad JUJU!, original_sc is NULL!\n",
1622                                        __func__);
1623                                 break;
1624                         }
1625                         ctl_copy_sense_data(msg, io);
1626                         /*
1627                          * IO should have already been cleaned up on other
1628                          * SC so clear this flag so we won't send a message
1629                          * back to finish the IO there.
1630                          */
1631                         io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
1632                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1633
1634                         /* io = msg->hdr.serializing_sc; */
1635                         io->io_hdr.msg_type = CTL_MSG_BAD_JUJU;
1636                         ctl_enqueue_isc(io);
1637                         break;
1638
1639                 /* Handle resets sent from the other side */
1640                 case CTL_MSG_MANAGE_TASKS: {
1641                         struct ctl_taskio *taskio;
1642                         taskio = (struct ctl_taskio *)ctl_alloc_io(
1643                             softc->othersc_pool);
1644                         ctl_zero_io((union ctl_io *)taskio);
1645                         taskio->io_hdr.io_type = CTL_IO_TASK;
1646                         taskio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
1647                         taskio->io_hdr.nexus = msg->hdr.nexus;
1648                         taskio->task_action = msg->task.task_action;
1649                         taskio->tag_num = msg->task.tag_num;
1650                         taskio->tag_type = msg->task.tag_type;
1651 #ifdef CTL_TIME_IO
1652                         taskio->io_hdr.start_time = time_uptime;
1653                         getbinuptime(&taskio->io_hdr.start_bt);
1654 #endif /* CTL_TIME_IO */
1655                         ctl_run_task((union ctl_io *)taskio);
1656                         break;
1657                 }
1658                 /* Persistent Reserve action which needs attention */
1659                 case CTL_MSG_PERS_ACTION:
1660                         presio = (struct ctl_prio *)ctl_alloc_io(
1661                             softc->othersc_pool);
1662                         ctl_zero_io((union ctl_io *)presio);
1663                         presio->io_hdr.msg_type = CTL_MSG_PERS_ACTION;
1664                         presio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
1665                         presio->io_hdr.nexus = msg->hdr.nexus;
1666                         presio->pr_msg = msg->pr;
1667                         ctl_enqueue_isc((union ctl_io *)presio);
1668                         break;
1669                 case CTL_MSG_UA:
1670                         ctl_isc_ua(softc, msg, param);
1671                         break;
1672                 case CTL_MSG_PORT_SYNC:
1673                         ctl_isc_port_sync(softc, msg, param);
1674                         break;
1675                 case CTL_MSG_LUN_SYNC:
1676                         ctl_isc_lun_sync(softc, msg, param);
1677                         break;
1678                 case CTL_MSG_IID_SYNC:
1679                         ctl_isc_iid_sync(softc, msg, param);
1680                         break;
1681                 case CTL_MSG_LOGIN:
1682                         ctl_isc_login(softc, msg, param);
1683                         break;
1684                 case CTL_MSG_MODE_SYNC:
1685                         ctl_isc_mode_sync(softc, msg, param);
1686                         break;
1687                 default:
1688                         printf("Received HA message of unknown type %d\n",
1689                             msg->hdr.msg_type);
1690                         ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1691                         break;
1692                 }
1693                 if (msg != &msgbuf)
1694                         free(msg, M_CTL);
1695         } else if (event == CTL_HA_EVT_LINK_CHANGE) {
1696                 printf("CTL: HA link status changed from %d to %d\n",
1697                     softc->ha_link, param);
1698                 if (param == softc->ha_link)
1699                         return;
1700                 if (softc->ha_link == CTL_HA_LINK_ONLINE) {
1701                         softc->ha_link = param;
1702                         ctl_isc_ha_link_down(softc);
1703                 } else {
1704                         softc->ha_link = param;
1705                         if (softc->ha_link == CTL_HA_LINK_ONLINE)
1706                                 ctl_isc_ha_link_up(softc);
1707                 }
1708                 return;
1709         } else {
1710                 printf("ctl_isc_event_handler: Unknown event %d\n", event);
1711                 return;
1712         }
1713 }
1714
1715 static void
1716 ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest)
1717 {
1718
1719         memcpy(&dest->scsiio.sense_data, &src->scsi.sense_data,
1720             src->scsi.sense_len);
1721         dest->scsiio.scsi_status = src->scsi.scsi_status;
1722         dest->scsiio.sense_len = src->scsi.sense_len;
1723         dest->io_hdr.status = src->hdr.status;
1724 }
1725
1726 static void
1727 ctl_copy_sense_data_back(union ctl_io *src, union ctl_ha_msg *dest)
1728 {
1729
1730         memcpy(&dest->scsi.sense_data, &src->scsiio.sense_data,
1731             src->scsiio.sense_len);
1732         dest->scsi.scsi_status = src->scsiio.scsi_status;
1733         dest->scsi.sense_len = src->scsiio.sense_len;
1734         dest->hdr.status = src->io_hdr.status;
1735 }
1736
1737 void
1738 ctl_est_ua(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua)
1739 {
1740         struct ctl_softc *softc = lun->ctl_softc;
1741         ctl_ua_type *pu;
1742
1743         if (initidx < softc->init_min || initidx >= softc->init_max)
1744                 return;
1745         mtx_assert(&lun->lun_lock, MA_OWNED);
1746         pu = lun->pending_ua[initidx / CTL_MAX_INIT_PER_PORT];
1747         if (pu == NULL)
1748                 return;
1749         pu[initidx % CTL_MAX_INIT_PER_PORT] |= ua;
1750 }
1751
1752 void
1753 ctl_est_ua_port(struct ctl_lun *lun, int port, uint32_t except, ctl_ua_type ua)
1754 {
1755         int i;
1756
1757         mtx_assert(&lun->lun_lock, MA_OWNED);
1758         if (lun->pending_ua[port] == NULL)
1759                 return;
1760         for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1761                 if (port * CTL_MAX_INIT_PER_PORT + i == except)
1762                         continue;
1763                 lun->pending_ua[port][i] |= ua;
1764         }
1765 }
1766
1767 void
1768 ctl_est_ua_all(struct ctl_lun *lun, uint32_t except, ctl_ua_type ua)
1769 {
1770         struct ctl_softc *softc = lun->ctl_softc;
1771         int i;
1772
1773         mtx_assert(&lun->lun_lock, MA_OWNED);
1774         for (i = softc->port_min; i < softc->port_max; i++)
1775                 ctl_est_ua_port(lun, i, except, ua);
1776 }
1777
1778 void
1779 ctl_clr_ua(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua)
1780 {
1781         struct ctl_softc *softc = lun->ctl_softc;
1782         ctl_ua_type *pu;
1783
1784         if (initidx < softc->init_min || initidx >= softc->init_max)
1785                 return;
1786         mtx_assert(&lun->lun_lock, MA_OWNED);
1787         pu = lun->pending_ua[initidx / CTL_MAX_INIT_PER_PORT];
1788         if (pu == NULL)
1789                 return;
1790         pu[initidx % CTL_MAX_INIT_PER_PORT] &= ~ua;
1791 }
1792
1793 void
1794 ctl_clr_ua_all(struct ctl_lun *lun, uint32_t except, ctl_ua_type ua)
1795 {
1796         struct ctl_softc *softc = lun->ctl_softc;
1797         int i, j;
1798
1799         mtx_assert(&lun->lun_lock, MA_OWNED);
1800         for (i = softc->port_min; i < softc->port_max; i++) {
1801                 if (lun->pending_ua[i] == NULL)
1802                         continue;
1803                 for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
1804                         if (i * CTL_MAX_INIT_PER_PORT + j == except)
1805                                 continue;
1806                         lun->pending_ua[i][j] &= ~ua;
1807                 }
1808         }
1809 }
1810
1811 void
1812 ctl_clr_ua_allluns(struct ctl_softc *ctl_softc, uint32_t initidx,
1813     ctl_ua_type ua_type)
1814 {
1815         struct ctl_lun *lun;
1816
1817         mtx_assert(&ctl_softc->ctl_lock, MA_OWNED);
1818         STAILQ_FOREACH(lun, &ctl_softc->lun_list, links) {
1819                 mtx_lock(&lun->lun_lock);
1820                 ctl_clr_ua(lun, initidx, ua_type);
1821                 mtx_unlock(&lun->lun_lock);
1822         }
1823 }
1824
1825 static int
1826 ctl_ha_role_sysctl(SYSCTL_HANDLER_ARGS)
1827 {
1828         struct ctl_softc *softc = (struct ctl_softc *)arg1;
1829         struct ctl_lun *lun;
1830         struct ctl_lun_req ireq;
1831         int error, value;
1832
1833         value = (softc->flags & CTL_FLAG_ACTIVE_SHELF) ? 0 : 1;
1834         error = sysctl_handle_int(oidp, &value, 0, req);
1835         if ((error != 0) || (req->newptr == NULL))
1836                 return (error);
1837
1838         mtx_lock(&softc->ctl_lock);
1839         if (value == 0)
1840                 softc->flags |= CTL_FLAG_ACTIVE_SHELF;
1841         else
1842                 softc->flags &= ~CTL_FLAG_ACTIVE_SHELF;
1843         STAILQ_FOREACH(lun, &softc->lun_list, links) {
1844                 mtx_unlock(&softc->ctl_lock);
1845                 bzero(&ireq, sizeof(ireq));
1846                 ireq.reqtype = CTL_LUNREQ_MODIFY;
1847                 ireq.reqdata.modify.lun_id = lun->lun;
1848                 lun->backend->ioctl(NULL, CTL_LUN_REQ, (caddr_t)&ireq, 0,
1849                     curthread);
1850                 if (ireq.status != CTL_LUN_OK) {
1851                         printf("%s: CTL_LUNREQ_MODIFY returned %d '%s'\n",
1852                             __func__, ireq.status, ireq.error_str);
1853                 }
1854                 mtx_lock(&softc->ctl_lock);
1855         }
1856         mtx_unlock(&softc->ctl_lock);
1857         return (0);
1858 }
1859
1860 static int
1861 ctl_init(void)
1862 {
1863         struct make_dev_args args;
1864         struct ctl_softc *softc;
1865         int i, error;
1866
1867         softc = control_softc = malloc(sizeof(*control_softc), M_DEVBUF,
1868                                M_WAITOK | M_ZERO);
1869
1870         make_dev_args_init(&args);
1871         args.mda_devsw = &ctl_cdevsw;
1872         args.mda_uid = UID_ROOT;
1873         args.mda_gid = GID_OPERATOR;
1874         args.mda_mode = 0600;
1875         args.mda_si_drv1 = softc;
1876         args.mda_si_drv2 = NULL;
1877         error = make_dev_s(&args, &softc->dev, "cam/ctl");
1878         if (error != 0) {
1879                 free(softc, M_DEVBUF);
1880                 control_softc = NULL;
1881                 return (error);
1882         }
1883
1884         sysctl_ctx_init(&softc->sysctl_ctx);
1885         softc->sysctl_tree = SYSCTL_ADD_NODE(&softc->sysctl_ctx,
1886                 SYSCTL_STATIC_CHILDREN(_kern_cam), OID_AUTO, "ctl",
1887                 CTLFLAG_RD, 0, "CAM Target Layer");
1888
1889         if (softc->sysctl_tree == NULL) {
1890                 printf("%s: unable to allocate sysctl tree\n", __func__);
1891                 destroy_dev(softc->dev);
1892                 free(softc, M_DEVBUF);
1893                 control_softc = NULL;
1894                 return (ENOMEM);
1895         }
1896
1897         mtx_init(&softc->ctl_lock, "CTL mutex", NULL, MTX_DEF);
1898         softc->io_zone = uma_zcreate("CTL IO", sizeof(union ctl_io),
1899             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1900         softc->flags = 0;
1901
1902         SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1903             OID_AUTO, "ha_mode", CTLFLAG_RDTUN, (int *)&softc->ha_mode, 0,
1904             "HA mode (0 - act/stby, 1 - serialize only, 2 - xfer)");
1905
1906         if (ctl_max_luns <= 0 || powerof2(ctl_max_luns) == 0) {
1907                 printf("Bad value %d for kern.cam.ctl.max_luns, must be a power of two, using %d\n",
1908                     ctl_max_luns, CTL_DEFAULT_MAX_LUNS);
1909                 ctl_max_luns = CTL_DEFAULT_MAX_LUNS;
1910         }
1911         softc->ctl_luns = malloc(sizeof(struct ctl_lun *) * ctl_max_luns,
1912             M_DEVBUF, M_WAITOK | M_ZERO);
1913         softc->ctl_lun_mask = malloc(sizeof(uint32_t) *
1914             ((ctl_max_luns + 31) / 32), M_DEVBUF, M_WAITOK | M_ZERO);
1915         if (ctl_max_ports <= 0 || powerof2(ctl_max_ports) == 0) {
1916                 printf("Bad value %d for kern.cam.ctl.max_ports, must be a power of two, using %d\n",
1917                     ctl_max_ports, CTL_DEFAULT_MAX_PORTS);
1918                 ctl_max_ports = CTL_DEFAULT_MAX_PORTS;
1919         }
1920         softc->ctl_port_mask = malloc(sizeof(uint32_t) *
1921           ((ctl_max_ports + 31) / 32), M_DEVBUF, M_WAITOK | M_ZERO);
1922         softc->ctl_ports = malloc(sizeof(struct ctl_port *) * ctl_max_ports,
1923              M_DEVBUF, M_WAITOK | M_ZERO);
1924
1925
1926         /*
1927          * In Copan's HA scheme, the "master" and "slave" roles are
1928          * figured out through the slot the controller is in.  Although it
1929          * is an active/active system, someone has to be in charge.
1930          */
1931         SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1932             OID_AUTO, "ha_id", CTLFLAG_RDTUN, &softc->ha_id, 0,
1933             "HA head ID (0 - no HA)");
1934         if (softc->ha_id == 0 || softc->ha_id > NUM_HA_SHELVES) {
1935                 softc->flags |= CTL_FLAG_ACTIVE_SHELF;
1936                 softc->is_single = 1;
1937                 softc->port_cnt = ctl_max_ports;
1938                 softc->port_min = 0;
1939         } else {
1940                 softc->port_cnt = ctl_max_ports / NUM_HA_SHELVES;
1941                 softc->port_min = (softc->ha_id - 1) * softc->port_cnt;
1942         }
1943         softc->port_max = softc->port_min + softc->port_cnt;
1944         softc->init_min = softc->port_min * CTL_MAX_INIT_PER_PORT;
1945         softc->init_max = softc->port_max * CTL_MAX_INIT_PER_PORT;
1946
1947         SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1948             OID_AUTO, "ha_link", CTLFLAG_RD, (int *)&softc->ha_link, 0,
1949             "HA link state (0 - offline, 1 - unknown, 2 - online)");
1950
1951         STAILQ_INIT(&softc->lun_list);
1952         STAILQ_INIT(&softc->pending_lun_queue);
1953         STAILQ_INIT(&softc->fe_list);
1954         STAILQ_INIT(&softc->port_list);
1955         STAILQ_INIT(&softc->be_list);
1956         ctl_tpc_init(softc);
1957
1958         if (worker_threads <= 0)
1959                 worker_threads = max(1, mp_ncpus / 4);
1960         if (worker_threads > CTL_MAX_THREADS)
1961                 worker_threads = CTL_MAX_THREADS;
1962
1963         for (i = 0; i < worker_threads; i++) {
1964                 struct ctl_thread *thr = &softc->threads[i];
1965
1966                 mtx_init(&thr->queue_lock, "CTL queue mutex", NULL, MTX_DEF);
1967                 thr->ctl_softc = softc;
1968                 STAILQ_INIT(&thr->incoming_queue);
1969                 STAILQ_INIT(&thr->rtr_queue);
1970                 STAILQ_INIT(&thr->done_queue);
1971                 STAILQ_INIT(&thr->isc_queue);
1972
1973                 error = kproc_kthread_add(ctl_work_thread, thr,
1974                     &softc->ctl_proc, &thr->thread, 0, 0, "ctl", "work%d", i);
1975                 if (error != 0) {
1976                         printf("error creating CTL work thread!\n");
1977                         return (error);
1978                 }
1979         }
1980         error = kproc_kthread_add(ctl_lun_thread, softc,
1981             &softc->ctl_proc, &softc->lun_thread, 0, 0, "ctl", "lun");
1982         if (error != 0) {
1983                 printf("error creating CTL lun thread!\n");
1984                 return (error);
1985         }
1986         error = kproc_kthread_add(ctl_thresh_thread, softc,
1987             &softc->ctl_proc, &softc->thresh_thread, 0, 0, "ctl", "thresh");
1988         if (error != 0) {
1989                 printf("error creating CTL threshold thread!\n");
1990                 return (error);
1991         }
1992
1993         SYSCTL_ADD_PROC(&softc->sysctl_ctx,SYSCTL_CHILDREN(softc->sysctl_tree),
1994             OID_AUTO, "ha_role", CTLTYPE_INT | CTLFLAG_RWTUN,
1995             softc, 0, ctl_ha_role_sysctl, "I", "HA role for this head");
1996
1997         if (softc->is_single == 0) {
1998                 if (ctl_frontend_register(&ha_frontend) != 0)
1999                         softc->is_single = 1;
2000         }
2001         return (0);
2002 }
2003
2004 static int
2005 ctl_shutdown(void)
2006 {
2007         struct ctl_softc *softc = control_softc;
2008         int i;
2009
2010         if (softc->is_single == 0)
2011                 ctl_frontend_deregister(&ha_frontend);
2012
2013         destroy_dev(softc->dev);
2014
2015         /* Shutdown CTL threads. */
2016         softc->shutdown = 1;
2017         for (i = 0; i < worker_threads; i++) {
2018                 struct ctl_thread *thr = &softc->threads[i];
2019                 while (thr->thread != NULL) {
2020                         wakeup(thr);
2021                         if (thr->thread != NULL)
2022                                 pause("CTL thr shutdown", 1);
2023                 }
2024                 mtx_destroy(&thr->queue_lock);
2025         }
2026         while (softc->lun_thread != NULL) {
2027                 wakeup(&softc->pending_lun_queue);
2028                 if (softc->lun_thread != NULL)
2029                         pause("CTL thr shutdown", 1);
2030         }
2031         while (softc->thresh_thread != NULL) {
2032                 wakeup(softc->thresh_thread);
2033                 if (softc->thresh_thread != NULL)
2034                         pause("CTL thr shutdown", 1);
2035         }
2036
2037         ctl_tpc_shutdown(softc);
2038         uma_zdestroy(softc->io_zone);
2039         mtx_destroy(&softc->ctl_lock);
2040
2041         free(softc->ctl_luns, M_DEVBUF);
2042         free(softc->ctl_lun_mask, M_DEVBUF);
2043         free(softc->ctl_port_mask, M_DEVBUF);
2044         free(softc->ctl_ports, M_DEVBUF);
2045
2046         sysctl_ctx_free(&softc->sysctl_ctx);
2047
2048         free(softc, M_DEVBUF);
2049         control_softc = NULL;
2050         return (0);
2051 }
2052
2053 static int
2054 ctl_module_event_handler(module_t mod, int what, void *arg)
2055 {
2056
2057         switch (what) {
2058         case MOD_LOAD:
2059                 return (ctl_init());
2060         case MOD_UNLOAD:
2061                 return (ctl_shutdown());
2062         default:
2063                 return (EOPNOTSUPP);
2064         }
2065 }
2066
2067 /*
2068  * XXX KDM should we do some access checks here?  Bump a reference count to
2069  * prevent a CTL module from being unloaded while someone has it open?
2070  */
2071 static int
2072 ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td)
2073 {
2074         return (0);
2075 }
2076
2077 static int
2078 ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td)
2079 {
2080         return (0);
2081 }
2082
2083 /*
2084  * Remove an initiator by port number and initiator ID.
2085  * Returns 0 for success, -1 for failure.
2086  */
2087 int
2088 ctl_remove_initiator(struct ctl_port *port, int iid)
2089 {
2090         struct ctl_softc *softc = port->ctl_softc;
2091         int last;
2092
2093         mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
2094
2095         if (iid > CTL_MAX_INIT_PER_PORT) {
2096                 printf("%s: initiator ID %u > maximun %u!\n",
2097                        __func__, iid, CTL_MAX_INIT_PER_PORT);
2098                 return (-1);
2099         }
2100
2101         mtx_lock(&softc->ctl_lock);
2102         last = (--port->wwpn_iid[iid].in_use == 0);
2103         port->wwpn_iid[iid].last_use = time_uptime;
2104         mtx_unlock(&softc->ctl_lock);
2105         if (last)
2106                 ctl_i_t_nexus_loss(softc, iid, CTL_UA_POWERON);
2107         ctl_isc_announce_iid(port, iid);
2108
2109         return (0);
2110 }
2111
2112 /*
2113  * Add an initiator to the initiator map.
2114  * Returns iid for success, < 0 for failure.
2115  */
2116 int
2117 ctl_add_initiator(struct ctl_port *port, int iid, uint64_t wwpn, char *name)
2118 {
2119         struct ctl_softc *softc = port->ctl_softc;
2120         time_t best_time;
2121         int i, best;
2122
2123         mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
2124
2125         if (iid >= CTL_MAX_INIT_PER_PORT) {
2126                 printf("%s: WWPN %#jx initiator ID %u > maximum %u!\n",
2127                        __func__, wwpn, iid, CTL_MAX_INIT_PER_PORT);
2128                 free(name, M_CTL);
2129                 return (-1);
2130         }
2131
2132         mtx_lock(&softc->ctl_lock);
2133
2134         if (iid < 0 && (wwpn != 0 || name != NULL)) {
2135                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
2136                         if (wwpn != 0 && wwpn == port->wwpn_iid[i].wwpn) {
2137                                 iid = i;
2138                                 break;
2139                         }
2140                         if (name != NULL && port->wwpn_iid[i].name != NULL &&
2141                             strcmp(name, port->wwpn_iid[i].name) == 0) {
2142                                 iid = i;
2143                                 break;
2144                         }
2145                 }
2146         }
2147
2148         if (iid < 0) {
2149                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
2150                         if (port->wwpn_iid[i].in_use == 0 &&
2151                             port->wwpn_iid[i].wwpn == 0 &&
2152                             port->wwpn_iid[i].name == NULL) {
2153                                 iid = i;
2154                                 break;
2155                         }
2156                 }
2157         }
2158
2159         if (iid < 0) {
2160                 best = -1;
2161                 best_time = INT32_MAX;
2162                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
2163                         if (port->wwpn_iid[i].in_use == 0) {
2164                                 if (port->wwpn_iid[i].last_use < best_time) {
2165                                         best = i;
2166                                         best_time = port->wwpn_iid[i].last_use;
2167                                 }
2168                         }
2169                 }
2170                 iid = best;
2171         }
2172
2173         if (iid < 0) {
2174                 mtx_unlock(&softc->ctl_lock);
2175                 free(name, M_CTL);
2176                 return (-2);
2177         }
2178
2179         if (port->wwpn_iid[iid].in_use > 0 && (wwpn != 0 || name != NULL)) {
2180                 /*
2181                  * This is not an error yet.
2182                  */
2183                 if (wwpn != 0 && wwpn == port->wwpn_iid[iid].wwpn) {
2184 #if 0
2185                         printf("%s: port %d iid %u WWPN %#jx arrived"
2186                             " again\n", __func__, port->targ_port,
2187                             iid, (uintmax_t)wwpn);
2188 #endif
2189                         goto take;
2190                 }
2191                 if (name != NULL && port->wwpn_iid[iid].name != NULL &&
2192                     strcmp(name, port->wwpn_iid[iid].name) == 0) {
2193 #if 0
2194                         printf("%s: port %d iid %u name '%s' arrived"
2195                             " again\n", __func__, port->targ_port,
2196                             iid, name);
2197 #endif
2198                         goto take;
2199                 }
2200
2201                 /*
2202                  * This is an error, but what do we do about it?  The
2203                  * driver is telling us we have a new WWPN for this
2204                  * initiator ID, so we pretty much need to use it.
2205                  */
2206                 printf("%s: port %d iid %u WWPN %#jx '%s' arrived,"
2207                     " but WWPN %#jx '%s' is still at that address\n",
2208                     __func__, port->targ_port, iid, wwpn, name,
2209                     (uintmax_t)port->wwpn_iid[iid].wwpn,
2210                     port->wwpn_iid[iid].name);
2211         }
2212 take:
2213         free(port->wwpn_iid[iid].name, M_CTL);
2214         port->wwpn_iid[iid].name = name;
2215         port->wwpn_iid[iid].wwpn = wwpn;
2216         port->wwpn_iid[iid].in_use++;
2217         mtx_unlock(&softc->ctl_lock);
2218         ctl_isc_announce_iid(port, iid);
2219
2220         return (iid);
2221 }
2222
2223 static int
2224 ctl_create_iid(struct ctl_port *port, int iid, uint8_t *buf)
2225 {
2226         int len;
2227
2228         switch (port->port_type) {
2229         case CTL_PORT_FC:
2230         {
2231                 struct scsi_transportid_fcp *id =
2232                     (struct scsi_transportid_fcp *)buf;
2233                 if (port->wwpn_iid[iid].wwpn == 0)
2234                         return (0);
2235                 memset(id, 0, sizeof(*id));
2236                 id->format_protocol = SCSI_PROTO_FC;
2237                 scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->n_port_name);
2238                 return (sizeof(*id));
2239         }
2240         case CTL_PORT_ISCSI:
2241         {
2242                 struct scsi_transportid_iscsi_port *id =
2243                     (struct scsi_transportid_iscsi_port *)buf;
2244                 if (port->wwpn_iid[iid].name == NULL)
2245                         return (0);
2246                 memset(id, 0, 256);
2247                 id->format_protocol = SCSI_TRN_ISCSI_FORMAT_PORT |
2248                     SCSI_PROTO_ISCSI;
2249                 len = strlcpy(id->iscsi_name, port->wwpn_iid[iid].name, 252) + 1;
2250                 len = roundup2(min(len, 252), 4);
2251                 scsi_ulto2b(len, id->additional_length);
2252                 return (sizeof(*id) + len);
2253         }
2254         case CTL_PORT_SAS:
2255         {
2256                 struct scsi_transportid_sas *id =
2257                     (struct scsi_transportid_sas *)buf;
2258                 if (port->wwpn_iid[iid].wwpn == 0)
2259                         return (0);
2260                 memset(id, 0, sizeof(*id));
2261                 id->format_protocol = SCSI_PROTO_SAS;
2262                 scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->sas_address);
2263                 return (sizeof(*id));
2264         }
2265         default:
2266         {
2267                 struct scsi_transportid_spi *id =
2268                     (struct scsi_transportid_spi *)buf;
2269                 memset(id, 0, sizeof(*id));
2270                 id->format_protocol = SCSI_PROTO_SPI;
2271                 scsi_ulto2b(iid, id->scsi_addr);
2272                 scsi_ulto2b(port->targ_port, id->rel_trgt_port_id);
2273                 return (sizeof(*id));
2274         }
2275         }
2276 }
2277
2278 /*
2279  * Serialize a command that went down the "wrong" side, and so was sent to
2280  * this controller for execution.  The logic is a little different than the
2281  * standard case in ctl_scsiio_precheck().  Errors in this case need to get
2282  * sent back to the other side, but in the success case, we execute the
2283  * command on this side (XFER mode) or tell the other side to execute it
2284  * (SER_ONLY mode).
2285  */
2286 static void
2287 ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio)
2288 {
2289         struct ctl_softc *softc = CTL_SOFTC(ctsio);
2290         struct ctl_port *port = CTL_PORT(ctsio);
2291         union ctl_ha_msg msg_info;
2292         struct ctl_lun *lun;
2293         const struct ctl_cmd_entry *entry;
2294         uint32_t targ_lun;
2295
2296         targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
2297
2298         /* Make sure that we know about this port. */
2299         if (port == NULL || (port->status & CTL_PORT_STATUS_ONLINE) == 0) {
2300                 ctl_set_internal_failure(ctsio, /*sks_valid*/ 0,
2301                                          /*retry_count*/ 1);
2302                 goto badjuju;
2303         }
2304
2305         /* Make sure that we know about this LUN. */
2306         mtx_lock(&softc->ctl_lock);
2307         if (targ_lun >= ctl_max_luns ||
2308             (lun = softc->ctl_luns[targ_lun]) == NULL) {
2309                 mtx_unlock(&softc->ctl_lock);
2310
2311                 /*
2312                  * The other node would not send this request to us unless
2313                  * received announce that we are primary node for this LUN.
2314                  * If this LUN does not exist now, it is probably result of
2315                  * a race, so respond to initiator in the most opaque way.
2316                  */
2317                 ctl_set_busy(ctsio);
2318                 goto badjuju;
2319         }
2320         mtx_lock(&lun->lun_lock);
2321         mtx_unlock(&softc->ctl_lock);
2322
2323         /*
2324          * If the LUN is invalid, pretend that it doesn't exist.
2325          * It will go away as soon as all pending I/Os completed.
2326          */
2327         if (lun->flags & CTL_LUN_DISABLED) {
2328                 mtx_unlock(&lun->lun_lock);
2329                 ctl_set_busy(ctsio);
2330                 goto badjuju;
2331         }
2332
2333         entry = ctl_get_cmd_entry(ctsio, NULL);
2334         if (ctl_scsiio_lun_check(lun, entry, ctsio) != 0) {
2335                 mtx_unlock(&lun->lun_lock);
2336                 goto badjuju;
2337         }
2338
2339         CTL_LUN(ctsio) = lun;
2340         CTL_BACKEND_LUN(ctsio) = lun->be_lun;
2341
2342         /*
2343          * Every I/O goes into the OOA queue for a
2344          * particular LUN, and stays there until completion.
2345          */
2346 #ifdef CTL_TIME_IO
2347         if (TAILQ_EMPTY(&lun->ooa_queue))
2348                 lun->idle_time += getsbinuptime() - lun->last_busy;
2349 #endif
2350         TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2351
2352         switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
2353                 (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr, ctl_ooaq,
2354                  ooa_links))) {
2355         case CTL_ACTION_BLOCK:
2356                 ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
2357                 TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
2358                                   blocked_links);
2359                 mtx_unlock(&lun->lun_lock);
2360                 break;
2361         case CTL_ACTION_PASS:
2362         case CTL_ACTION_SKIP:
2363                 if (softc->ha_mode == CTL_HA_MODE_XFER) {
2364                         ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
2365                         ctl_enqueue_rtr((union ctl_io *)ctsio);
2366                         mtx_unlock(&lun->lun_lock);
2367                 } else {
2368                         ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
2369                         mtx_unlock(&lun->lun_lock);
2370
2371                         /* send msg back to other side */
2372                         msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
2373                         msg_info.hdr.serializing_sc = (union ctl_io *)ctsio;
2374                         msg_info.hdr.msg_type = CTL_MSG_R2R;
2375                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
2376                             sizeof(msg_info.hdr), M_WAITOK);
2377                 }
2378                 break;
2379         case CTL_ACTION_OVERLAP:
2380                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2381                 mtx_unlock(&lun->lun_lock);
2382                 ctl_set_overlapped_cmd(ctsio);
2383                 goto badjuju;
2384         case CTL_ACTION_OVERLAP_TAG:
2385                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2386                 mtx_unlock(&lun->lun_lock);
2387                 ctl_set_overlapped_tag(ctsio, ctsio->tag_num);
2388                 goto badjuju;
2389         case CTL_ACTION_ERROR:
2390         default:
2391                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2392                 mtx_unlock(&lun->lun_lock);
2393
2394                 ctl_set_internal_failure(ctsio, /*sks_valid*/ 0,
2395                                          /*retry_count*/ 0);
2396 badjuju:
2397                 ctl_copy_sense_data_back((union ctl_io *)ctsio, &msg_info);
2398                 msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
2399                 msg_info.hdr.serializing_sc = NULL;
2400                 msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
2401                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
2402                     sizeof(msg_info.scsi), M_WAITOK);
2403                 ctl_free_io((union ctl_io *)ctsio);
2404                 break;
2405         }
2406 }
2407
2408 /*
2409  * Returns 0 for success, errno for failure.
2410  */
2411 static void
2412 ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
2413                    struct ctl_ooa *ooa_hdr, struct ctl_ooa_entry *kern_entries)
2414 {
2415         union ctl_io *io;
2416
2417         mtx_lock(&lun->lun_lock);
2418         for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); (io != NULL);
2419              (*cur_fill_num)++, io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2420              ooa_links)) {
2421                 struct ctl_ooa_entry *entry;
2422
2423                 /*
2424                  * If we've got more than we can fit, just count the
2425                  * remaining entries.
2426                  */
2427                 if (*cur_fill_num >= ooa_hdr->alloc_num)
2428                         continue;
2429
2430                 entry = &kern_entries[*cur_fill_num];
2431
2432                 entry->tag_num = io->scsiio.tag_num;
2433                 entry->lun_num = lun->lun;
2434 #ifdef CTL_TIME_IO
2435                 entry->start_bt = io->io_hdr.start_bt;
2436 #endif
2437                 bcopy(io->scsiio.cdb, entry->cdb, io->scsiio.cdb_len);
2438                 entry->cdb_len = io->scsiio.cdb_len;
2439                 if (io->io_hdr.flags & CTL_FLAG_BLOCKED)
2440                         entry->cmd_flags |= CTL_OOACMD_FLAG_BLOCKED;
2441
2442                 if (io->io_hdr.flags & CTL_FLAG_DMA_INPROG)
2443                         entry->cmd_flags |= CTL_OOACMD_FLAG_DMA;
2444
2445                 if (io->io_hdr.flags & CTL_FLAG_ABORT)
2446                         entry->cmd_flags |= CTL_OOACMD_FLAG_ABORT;
2447
2448                 if (io->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR)
2449                         entry->cmd_flags |= CTL_OOACMD_FLAG_RTR;
2450
2451                 if (io->io_hdr.flags & CTL_FLAG_DMA_QUEUED)
2452                         entry->cmd_flags |= CTL_OOACMD_FLAG_DMA_QUEUED;
2453         }
2454         mtx_unlock(&lun->lun_lock);
2455 }
2456
2457 /*
2458  * Escape characters that are illegal or not recommended in XML.
2459  */
2460 int
2461 ctl_sbuf_printf_esc(struct sbuf *sb, char *str, int size)
2462 {
2463         char *end = str + size;
2464         int retval;
2465
2466         retval = 0;
2467
2468         for (; *str && str < end; str++) {
2469                 switch (*str) {
2470                 case '&':
2471                         retval = sbuf_printf(sb, "&amp;");
2472                         break;
2473                 case '>':
2474                         retval = sbuf_printf(sb, "&gt;");
2475                         break;
2476                 case '<':
2477                         retval = sbuf_printf(sb, "&lt;");
2478                         break;
2479                 default:
2480                         retval = sbuf_putc(sb, *str);
2481                         break;
2482                 }
2483
2484                 if (retval != 0)
2485                         break;
2486
2487         }
2488
2489         return (retval);
2490 }
2491
2492 static void
2493 ctl_id_sbuf(struct ctl_devid *id, struct sbuf *sb)
2494 {
2495         struct scsi_vpd_id_descriptor *desc;
2496         int i;
2497
2498         if (id == NULL || id->len < 4)
2499                 return;
2500         desc = (struct scsi_vpd_id_descriptor *)id->data;
2501         switch (desc->id_type & SVPD_ID_TYPE_MASK) {
2502         case SVPD_ID_TYPE_T10:
2503                 sbuf_printf(sb, "t10.");
2504                 break;
2505         case SVPD_ID_TYPE_EUI64:
2506                 sbuf_printf(sb, "eui.");
2507                 break;
2508         case SVPD_ID_TYPE_NAA:
2509                 sbuf_printf(sb, "naa.");
2510                 break;
2511         case SVPD_ID_TYPE_SCSI_NAME:
2512                 break;
2513         }
2514         switch (desc->proto_codeset & SVPD_ID_CODESET_MASK) {
2515         case SVPD_ID_CODESET_BINARY:
2516                 for (i = 0; i < desc->length; i++)
2517                         sbuf_printf(sb, "%02x", desc->identifier[i]);
2518                 break;
2519         case SVPD_ID_CODESET_ASCII:
2520                 sbuf_printf(sb, "%.*s", (int)desc->length,
2521                     (char *)desc->identifier);
2522                 break;
2523         case SVPD_ID_CODESET_UTF8:
2524                 sbuf_printf(sb, "%s", (char *)desc->identifier);
2525                 break;
2526         }
2527 }
2528
2529 static int
2530 ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
2531           struct thread *td)
2532 {
2533         struct ctl_softc *softc = dev->si_drv1;
2534         struct ctl_port *port;
2535         struct ctl_lun *lun;
2536         int retval;
2537
2538         retval = 0;
2539
2540         switch (cmd) {
2541         case CTL_IO:
2542                 retval = ctl_ioctl_io(dev, cmd, addr, flag, td);
2543                 break;
2544         case CTL_ENABLE_PORT:
2545         case CTL_DISABLE_PORT:
2546         case CTL_SET_PORT_WWNS: {
2547                 struct ctl_port *port;
2548                 struct ctl_port_entry *entry;
2549
2550                 entry = (struct ctl_port_entry *)addr;
2551
2552                 mtx_lock(&softc->ctl_lock);
2553                 STAILQ_FOREACH(port, &softc->port_list, links) {
2554                         int action, done;
2555
2556                         if (port->targ_port < softc->port_min ||
2557                             port->targ_port >= softc->port_max)
2558                                 continue;
2559
2560                         action = 0;
2561                         done = 0;
2562                         if ((entry->port_type == CTL_PORT_NONE)
2563                          && (entry->targ_port == port->targ_port)) {
2564                                 /*
2565                                  * If the user only wants to enable or
2566                                  * disable or set WWNs on a specific port,
2567                                  * do the operation and we're done.
2568                                  */
2569                                 action = 1;
2570                                 done = 1;
2571                         } else if (entry->port_type & port->port_type) {
2572                                 /*
2573                                  * Compare the user's type mask with the
2574                                  * particular frontend type to see if we
2575                                  * have a match.
2576                                  */
2577                                 action = 1;
2578                                 done = 0;
2579
2580                                 /*
2581                                  * Make sure the user isn't trying to set
2582                                  * WWNs on multiple ports at the same time.
2583                                  */
2584                                 if (cmd == CTL_SET_PORT_WWNS) {
2585                                         printf("%s: Can't set WWNs on "
2586                                                "multiple ports\n", __func__);
2587                                         retval = EINVAL;
2588                                         break;
2589                                 }
2590                         }
2591                         if (action == 0)
2592                                 continue;
2593
2594                         /*
2595                          * XXX KDM we have to drop the lock here, because
2596                          * the online/offline operations can potentially
2597                          * block.  We need to reference count the frontends
2598                          * so they can't go away,
2599                          */
2600                         if (cmd == CTL_ENABLE_PORT) {
2601                                 mtx_unlock(&softc->ctl_lock);
2602                                 ctl_port_online(port);
2603                                 mtx_lock(&softc->ctl_lock);
2604                         } else if (cmd == CTL_DISABLE_PORT) {
2605                                 mtx_unlock(&softc->ctl_lock);
2606                                 ctl_port_offline(port);
2607                                 mtx_lock(&softc->ctl_lock);
2608                         } else if (cmd == CTL_SET_PORT_WWNS) {
2609                                 ctl_port_set_wwns(port,
2610                                     (entry->flags & CTL_PORT_WWNN_VALID) ?
2611                                     1 : 0, entry->wwnn,
2612                                     (entry->flags & CTL_PORT_WWPN_VALID) ?
2613                                     1 : 0, entry->wwpn);
2614                         }
2615                         if (done != 0)
2616                                 break;
2617                 }
2618                 mtx_unlock(&softc->ctl_lock);
2619                 break;
2620         }
2621         case CTL_GET_OOA: {
2622                 struct ctl_ooa *ooa_hdr;
2623                 struct ctl_ooa_entry *entries;
2624                 uint32_t cur_fill_num;
2625
2626                 ooa_hdr = (struct ctl_ooa *)addr;
2627
2628                 if ((ooa_hdr->alloc_len == 0)
2629                  || (ooa_hdr->alloc_num == 0)) {
2630                         printf("%s: CTL_GET_OOA: alloc len %u and alloc num %u "
2631                                "must be non-zero\n", __func__,
2632                                ooa_hdr->alloc_len, ooa_hdr->alloc_num);
2633                         retval = EINVAL;
2634                         break;
2635                 }
2636
2637                 if (ooa_hdr->alloc_len != (ooa_hdr->alloc_num *
2638                     sizeof(struct ctl_ooa_entry))) {
2639                         printf("%s: CTL_GET_OOA: alloc len %u must be alloc "
2640                                "num %d * sizeof(struct ctl_ooa_entry) %zd\n",
2641                                __func__, ooa_hdr->alloc_len,
2642                                ooa_hdr->alloc_num,sizeof(struct ctl_ooa_entry));
2643                         retval = EINVAL;
2644                         break;
2645                 }
2646
2647                 entries = malloc(ooa_hdr->alloc_len, M_CTL, M_WAITOK | M_ZERO);
2648                 if (entries == NULL) {
2649                         printf("%s: could not allocate %d bytes for OOA "
2650                                "dump\n", __func__, ooa_hdr->alloc_len);
2651                         retval = ENOMEM;
2652                         break;
2653                 }
2654
2655                 mtx_lock(&softc->ctl_lock);
2656                 if ((ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) == 0 &&
2657                     (ooa_hdr->lun_num >= ctl_max_luns ||
2658                      softc->ctl_luns[ooa_hdr->lun_num] == NULL)) {
2659                         mtx_unlock(&softc->ctl_lock);
2660                         free(entries, M_CTL);
2661                         printf("%s: CTL_GET_OOA: invalid LUN %ju\n",
2662                                __func__, (uintmax_t)ooa_hdr->lun_num);
2663                         retval = EINVAL;
2664                         break;
2665                 }
2666
2667                 cur_fill_num = 0;
2668
2669                 if (ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) {
2670                         STAILQ_FOREACH(lun, &softc->lun_list, links) {
2671                                 ctl_ioctl_fill_ooa(lun, &cur_fill_num,
2672                                     ooa_hdr, entries);
2673                         }
2674                 } else {
2675                         lun = softc->ctl_luns[ooa_hdr->lun_num];
2676                         ctl_ioctl_fill_ooa(lun, &cur_fill_num, ooa_hdr,
2677                             entries);
2678                 }
2679                 mtx_unlock(&softc->ctl_lock);
2680
2681                 ooa_hdr->fill_num = min(cur_fill_num, ooa_hdr->alloc_num);
2682                 ooa_hdr->fill_len = ooa_hdr->fill_num *
2683                         sizeof(struct ctl_ooa_entry);
2684                 retval = copyout(entries, ooa_hdr->entries, ooa_hdr->fill_len);
2685                 if (retval != 0) {
2686                         printf("%s: error copying out %d bytes for OOA dump\n",
2687                                __func__, ooa_hdr->fill_len);
2688                 }
2689
2690                 getbinuptime(&ooa_hdr->cur_bt);
2691
2692                 if (cur_fill_num > ooa_hdr->alloc_num) {
2693                         ooa_hdr->dropped_num = cur_fill_num -ooa_hdr->alloc_num;
2694                         ooa_hdr->status = CTL_OOA_NEED_MORE_SPACE;
2695                 } else {
2696                         ooa_hdr->dropped_num = 0;
2697                         ooa_hdr->status = CTL_OOA_OK;
2698                 }
2699
2700                 free(entries, M_CTL);
2701                 break;
2702         }
2703         case CTL_DELAY_IO: {
2704                 struct ctl_io_delay_info *delay_info;
2705
2706                 delay_info = (struct ctl_io_delay_info *)addr;
2707
2708 #ifdef CTL_IO_DELAY
2709                 mtx_lock(&softc->ctl_lock);
2710                 if (delay_info->lun_id >= ctl_max_luns ||
2711                     (lun = softc->ctl_luns[delay_info->lun_id]) == NULL) {
2712                         mtx_unlock(&softc->ctl_lock);
2713                         delay_info->status = CTL_DELAY_STATUS_INVALID_LUN;
2714                         break;
2715                 }
2716                 mtx_lock(&lun->lun_lock);
2717                 mtx_unlock(&softc->ctl_lock);
2718                 delay_info->status = CTL_DELAY_STATUS_OK;
2719                 switch (delay_info->delay_type) {
2720                 case CTL_DELAY_TYPE_CONT:
2721                 case CTL_DELAY_TYPE_ONESHOT:
2722                         break;
2723                 default:
2724                         delay_info->status = CTL_DELAY_STATUS_INVALID_TYPE;
2725                         break;
2726                 }
2727                 switch (delay_info->delay_loc) {
2728                 case CTL_DELAY_LOC_DATAMOVE:
2729                         lun->delay_info.datamove_type = delay_info->delay_type;
2730                         lun->delay_info.datamove_delay = delay_info->delay_secs;
2731                         break;
2732                 case CTL_DELAY_LOC_DONE:
2733                         lun->delay_info.done_type = delay_info->delay_type;
2734                         lun->delay_info.done_delay = delay_info->delay_secs;
2735                         break;
2736                 default:
2737                         delay_info->status = CTL_DELAY_STATUS_INVALID_LOC;
2738                         break;
2739                 }
2740                 mtx_unlock(&lun->lun_lock);
2741 #else
2742                 delay_info->status = CTL_DELAY_STATUS_NOT_IMPLEMENTED;
2743 #endif /* CTL_IO_DELAY */
2744                 break;
2745         }
2746 #ifdef CTL_LEGACY_STATS
2747         case CTL_GETSTATS: {
2748                 struct ctl_stats *stats = (struct ctl_stats *)addr;
2749                 int i;
2750
2751                 /*
2752                  * XXX KDM no locking here.  If the LUN list changes,
2753                  * things can blow up.
2754                  */
2755                 i = 0;
2756                 stats->status = CTL_SS_OK;
2757                 stats->fill_len = 0;
2758                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
2759                         if (stats->fill_len + sizeof(lun->legacy_stats) >
2760                             stats->alloc_len) {
2761                                 stats->status = CTL_SS_NEED_MORE_SPACE;
2762                                 break;
2763                         }
2764                         retval = copyout(&lun->legacy_stats, &stats->lun_stats[i++],
2765                                          sizeof(lun->legacy_stats));
2766                         if (retval != 0)
2767                                 break;
2768                         stats->fill_len += sizeof(lun->legacy_stats);
2769                 }
2770                 stats->num_luns = softc->num_luns;
2771                 stats->flags = CTL_STATS_FLAG_NONE;
2772 #ifdef CTL_TIME_IO
2773                 stats->flags |= CTL_STATS_FLAG_TIME_VALID;
2774 #endif
2775                 getnanouptime(&stats->timestamp);
2776                 break;
2777         }
2778 #endif /* CTL_LEGACY_STATS */
2779         case CTL_ERROR_INJECT: {
2780                 struct ctl_error_desc *err_desc, *new_err_desc;
2781
2782                 err_desc = (struct ctl_error_desc *)addr;
2783
2784                 new_err_desc = malloc(sizeof(*new_err_desc), M_CTL,
2785                                       M_WAITOK | M_ZERO);
2786                 bcopy(err_desc, new_err_desc, sizeof(*new_err_desc));
2787
2788                 mtx_lock(&softc->ctl_lock);
2789                 if (err_desc->lun_id >= ctl_max_luns ||
2790                     (lun = softc->ctl_luns[err_desc->lun_id]) == NULL) {
2791                         mtx_unlock(&softc->ctl_lock);
2792                         free(new_err_desc, M_CTL);
2793                         printf("%s: CTL_ERROR_INJECT: invalid LUN %ju\n",
2794                                __func__, (uintmax_t)err_desc->lun_id);
2795                         retval = EINVAL;
2796                         break;
2797                 }
2798                 mtx_lock(&lun->lun_lock);
2799                 mtx_unlock(&softc->ctl_lock);
2800
2801                 /*
2802                  * We could do some checking here to verify the validity
2803                  * of the request, but given the complexity of error
2804                  * injection requests, the checking logic would be fairly
2805                  * complex.
2806                  *
2807                  * For now, if the request is invalid, it just won't get
2808                  * executed and might get deleted.
2809                  */
2810                 STAILQ_INSERT_TAIL(&lun->error_list, new_err_desc, links);
2811
2812                 /*
2813                  * XXX KDM check to make sure the serial number is unique,
2814                  * in case we somehow manage to wrap.  That shouldn't
2815                  * happen for a very long time, but it's the right thing to
2816                  * do.
2817                  */
2818                 new_err_desc->serial = lun->error_serial;
2819                 err_desc->serial = lun->error_serial;
2820                 lun->error_serial++;
2821
2822                 mtx_unlock(&lun->lun_lock);
2823                 break;
2824         }
2825         case CTL_ERROR_INJECT_DELETE: {
2826                 struct ctl_error_desc *delete_desc, *desc, *desc2;
2827                 int delete_done;
2828
2829                 delete_desc = (struct ctl_error_desc *)addr;
2830                 delete_done = 0;
2831
2832                 mtx_lock(&softc->ctl_lock);
2833                 if (delete_desc->lun_id >= ctl_max_luns ||
2834                     (lun = softc->ctl_luns[delete_desc->lun_id]) == NULL) {
2835                         mtx_unlock(&softc->ctl_lock);
2836                         printf("%s: CTL_ERROR_INJECT_DELETE: invalid LUN %ju\n",
2837                                __func__, (uintmax_t)delete_desc->lun_id);
2838                         retval = EINVAL;
2839                         break;
2840                 }
2841                 mtx_lock(&lun->lun_lock);
2842                 mtx_unlock(&softc->ctl_lock);
2843                 STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
2844                         if (desc->serial != delete_desc->serial)
2845                                 continue;
2846
2847                         STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc,
2848                                       links);
2849                         free(desc, M_CTL);
2850                         delete_done = 1;
2851                 }
2852                 mtx_unlock(&lun->lun_lock);
2853                 if (delete_done == 0) {
2854                         printf("%s: CTL_ERROR_INJECT_DELETE: can't find "
2855                                "error serial %ju on LUN %u\n", __func__,
2856                                delete_desc->serial, delete_desc->lun_id);
2857                         retval = EINVAL;
2858                         break;
2859                 }
2860                 break;
2861         }
2862         case CTL_DUMP_STRUCTS: {
2863                 int j, k;
2864                 struct ctl_port *port;
2865                 struct ctl_frontend *fe;
2866
2867                 mtx_lock(&softc->ctl_lock);
2868                 printf("CTL Persistent Reservation information start:\n");
2869                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
2870                         mtx_lock(&lun->lun_lock);
2871                         if ((lun->flags & CTL_LUN_DISABLED) != 0) {
2872                                 mtx_unlock(&lun->lun_lock);
2873                                 continue;
2874                         }
2875
2876                         for (j = 0; j < ctl_max_ports; j++) {
2877                                 if (lun->pr_keys[j] == NULL)
2878                                         continue;
2879                                 for (k = 0; k < CTL_MAX_INIT_PER_PORT; k++){
2880                                         if (lun->pr_keys[j][k] == 0)
2881                                                 continue;
2882                                         printf("  LUN %ju port %d iid %d key "
2883                                                "%#jx\n", lun->lun, j, k,
2884                                                (uintmax_t)lun->pr_keys[j][k]);
2885                                 }
2886                         }
2887                         mtx_unlock(&lun->lun_lock);
2888                 }
2889                 printf("CTL Persistent Reservation information end\n");
2890                 printf("CTL Ports:\n");
2891                 STAILQ_FOREACH(port, &softc->port_list, links) {
2892                         printf("  Port %d '%s' Frontend '%s' Type %u pp %d vp %d WWNN "
2893                                "%#jx WWPN %#jx\n", port->targ_port, port->port_name,
2894                                port->frontend->name, port->port_type,
2895                                port->physical_port, port->virtual_port,
2896                                (uintmax_t)port->wwnn, (uintmax_t)port->wwpn);
2897                         for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
2898                                 if (port->wwpn_iid[j].in_use == 0 &&
2899                                     port->wwpn_iid[j].wwpn == 0 &&
2900                                     port->wwpn_iid[j].name == NULL)
2901                                         continue;
2902
2903                                 printf("    iid %u use %d WWPN %#jx '%s'\n",
2904                                     j, port->wwpn_iid[j].in_use,
2905                                     (uintmax_t)port->wwpn_iid[j].wwpn,
2906                                     port->wwpn_iid[j].name);
2907                         }
2908                 }
2909                 printf("CTL Port information end\n");
2910                 mtx_unlock(&softc->ctl_lock);
2911                 /*
2912                  * XXX KDM calling this without a lock.  We'd likely want
2913                  * to drop the lock before calling the frontend's dump
2914                  * routine anyway.
2915                  */
2916                 printf("CTL Frontends:\n");
2917                 STAILQ_FOREACH(fe, &softc->fe_list, links) {
2918                         printf("  Frontend '%s'\n", fe->name);
2919                         if (fe->fe_dump != NULL)
2920                                 fe->fe_dump();
2921                 }
2922                 printf("CTL Frontend information end\n");
2923                 break;
2924         }
2925         case CTL_LUN_REQ: {
2926                 struct ctl_lun_req *lun_req;
2927                 struct ctl_backend_driver *backend;
2928                 void *packed;
2929                 nvlist_t *tmp_args_nvl;
2930                 size_t packed_len;
2931
2932                 lun_req = (struct ctl_lun_req *)addr;
2933                 tmp_args_nvl = lun_req->args_nvl;
2934
2935                 backend = ctl_backend_find(lun_req->backend);
2936                 if (backend == NULL) {
2937                         lun_req->status = CTL_LUN_ERROR;
2938                         snprintf(lun_req->error_str,
2939                                  sizeof(lun_req->error_str),
2940                                  "Backend \"%s\" not found.",
2941                                  lun_req->backend);
2942                         break;
2943                 }
2944
2945                 if (lun_req->args != NULL) {
2946                         packed = malloc(lun_req->args_len, M_CTL, M_WAITOK);
2947                         if (copyin(lun_req->args, packed, lun_req->args_len) != 0) {
2948                                 free(packed, M_CTL);
2949                                 lun_req->status = CTL_LUN_ERROR;
2950                                 snprintf(lun_req->error_str, sizeof(lun_req->error_str),
2951                                     "Cannot copyin args.");
2952                                 break;
2953                         }
2954                         lun_req->args_nvl = nvlist_unpack(packed,
2955                             lun_req->args_len, 0);
2956                         free(packed, M_CTL);
2957
2958                         if (lun_req->args_nvl == NULL) {
2959                                 lun_req->status = CTL_LUN_ERROR;
2960                                 snprintf(lun_req->error_str, sizeof(lun_req->error_str),
2961                                     "Cannot unpack args nvlist.");
2962                                 break;
2963                         }
2964                 } else
2965                         lun_req->args_nvl = nvlist_create(0);
2966
2967                 retval = backend->ioctl(dev, cmd, addr, flag, td);
2968                 nvlist_destroy(lun_req->args_nvl);
2969                 lun_req->args_nvl = tmp_args_nvl;
2970
2971                 if (lun_req->result_nvl != NULL) {
2972                         if (lun_req->result != NULL) {
2973                                 packed = nvlist_pack(lun_req->result_nvl,
2974                                     &packed_len);
2975                                 if (packed == NULL) {
2976                                         lun_req->status = CTL_LUN_ERROR;
2977                                         snprintf(lun_req->error_str,
2978                                             sizeof(lun_req->error_str),
2979                                             "Cannot pack result nvlist.");
2980                                         break;
2981                                 }
2982
2983                                 if (packed_len > lun_req->result_len) {
2984                                         lun_req->status = CTL_LUN_ERROR;
2985                                         snprintf(lun_req->error_str,
2986                                             sizeof(lun_req->error_str),
2987                                             "Result nvlist too large.");
2988                                         free(packed, M_NVLIST);
2989                                         break;
2990                                 }
2991
2992                                 if (copyout(packed, lun_req->result, packed_len)) {
2993                                         lun_req->status = CTL_LUN_ERROR;
2994                                         snprintf(lun_req->error_str,
2995                                             sizeof(lun_req->error_str),
2996                                             "Cannot copyout() the result.");
2997                                         free(packed, M_NVLIST);
2998                                         break;
2999                                 }
3000
3001                                 lun_req->result_len = packed_len;
3002                                 free(packed, M_NVLIST);
3003                         }
3004
3005                         nvlist_destroy(lun_req->result_nvl);
3006                 }
3007                 break;
3008         }
3009         case CTL_LUN_LIST: {
3010                 struct sbuf *sb;
3011                 struct ctl_lun_list *list;
3012                 const char *name, *value;
3013                 void *cookie;
3014                 int type;
3015
3016                 list = (struct ctl_lun_list *)addr;
3017
3018                 /*
3019                  * Allocate a fixed length sbuf here, based on the length
3020                  * of the user's buffer.  We could allocate an auto-extending
3021                  * buffer, and then tell the user how much larger our
3022                  * amount of data is than his buffer, but that presents
3023                  * some problems:
3024                  *
3025                  * 1.  The sbuf(9) routines use a blocking malloc, and so
3026                  *     we can't hold a lock while calling them with an
3027                  *     auto-extending buffer.
3028                  *
3029                  * 2.  There is not currently a LUN reference counting
3030                  *     mechanism, outside of outstanding transactions on
3031                  *     the LUN's OOA queue.  So a LUN could go away on us
3032                  *     while we're getting the LUN number, backend-specific
3033                  *     information, etc.  Thus, given the way things
3034                  *     currently work, we need to hold the CTL lock while
3035                  *     grabbing LUN information.
3036                  *
3037                  * So, from the user's standpoint, the best thing to do is
3038                  * allocate what he thinks is a reasonable buffer length,
3039                  * and then if he gets a CTL_LUN_LIST_NEED_MORE_SPACE error,
3040                  * double the buffer length and try again.  (And repeat
3041                  * that until he succeeds.)
3042                  */
3043                 sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3044                 if (sb == NULL) {
3045                         list->status = CTL_LUN_LIST_ERROR;
3046                         snprintf(list->error_str, sizeof(list->error_str),
3047                                  "Unable to allocate %d bytes for LUN list",
3048                                  list->alloc_len);
3049                         break;
3050                 }
3051
3052                 sbuf_printf(sb, "<ctllunlist>\n");
3053
3054                 mtx_lock(&softc->ctl_lock);
3055                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
3056                         mtx_lock(&lun->lun_lock);
3057                         retval = sbuf_printf(sb, "<lun id=\"%ju\">\n",
3058                                              (uintmax_t)lun->lun);
3059
3060                         /*
3061                          * Bail out as soon as we see that we've overfilled
3062                          * the buffer.
3063                          */
3064                         if (retval != 0)
3065                                 break;
3066
3067                         retval = sbuf_printf(sb, "\t<backend_type>%s"
3068                                              "</backend_type>\n",
3069                                              (lun->backend == NULL) ?  "none" :
3070                                              lun->backend->name);
3071
3072                         if (retval != 0)
3073                                 break;
3074
3075                         retval = sbuf_printf(sb, "\t<lun_type>%d</lun_type>\n",
3076                                              lun->be_lun->lun_type);
3077
3078                         if (retval != 0)
3079                                 break;
3080
3081                         if (lun->backend == NULL) {
3082                                 retval = sbuf_printf(sb, "</lun>\n");
3083                                 if (retval != 0)
3084                                         break;
3085                                 continue;
3086                         }
3087
3088                         retval = sbuf_printf(sb, "\t<size>%ju</size>\n",
3089                                              (lun->be_lun->maxlba > 0) ?
3090                                              lun->be_lun->maxlba + 1 : 0);
3091
3092                         if (retval != 0)
3093                                 break;
3094
3095                         retval = sbuf_printf(sb, "\t<blocksize>%u</blocksize>\n",
3096                                              lun->be_lun->blocksize);
3097
3098                         if (retval != 0)
3099                                 break;
3100
3101                         retval = sbuf_printf(sb, "\t<serial_number>");
3102
3103                         if (retval != 0)
3104                                 break;
3105
3106                         retval = ctl_sbuf_printf_esc(sb,
3107                             lun->be_lun->serial_num,
3108                             sizeof(lun->be_lun->serial_num));
3109
3110                         if (retval != 0)
3111                                 break;
3112
3113                         retval = sbuf_printf(sb, "</serial_number>\n");
3114
3115                         if (retval != 0)
3116                                 break;
3117
3118                         retval = sbuf_printf(sb, "\t<device_id>");
3119
3120                         if (retval != 0)
3121                                 break;
3122
3123                         retval = ctl_sbuf_printf_esc(sb,
3124                             lun->be_lun->device_id,
3125                             sizeof(lun->be_lun->device_id));
3126
3127                         if (retval != 0)
3128                                 break;
3129
3130                         retval = sbuf_printf(sb, "</device_id>\n");
3131
3132                         if (retval != 0)
3133                                 break;
3134
3135                         if (lun->backend->lun_info != NULL) {
3136                                 retval = lun->backend->lun_info(lun->be_lun->be_lun, sb);
3137                                 if (retval != 0)
3138                                         break;
3139                         }
3140
3141                         cookie = NULL;
3142                         while ((name = nvlist_next(lun->be_lun->options, &type,
3143                             &cookie)) != NULL) {
3144                                 sbuf_printf(sb, "\t<%s>", name);
3145
3146                                 if (type == NV_TYPE_STRING) {
3147                                         value = dnvlist_get_string(
3148                                             lun->be_lun->options, name, NULL);
3149                                         if (value != NULL)
3150                                                 sbuf_printf(sb, "%s", value);
3151                                 }
3152
3153                                 sbuf_printf(sb, "</%s>\n", name);
3154                         }
3155
3156                         retval = sbuf_printf(sb, "</lun>\n");
3157
3158                         if (retval != 0)
3159                                 break;
3160                         mtx_unlock(&lun->lun_lock);
3161                 }
3162                 if (lun != NULL)
3163                         mtx_unlock(&lun->lun_lock);
3164                 mtx_unlock(&softc->ctl_lock);
3165
3166                 if ((retval != 0)
3167                  || ((retval = sbuf_printf(sb, "</ctllunlist>\n")) != 0)) {
3168                         retval = 0;
3169                         sbuf_delete(sb);
3170                         list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3171                         snprintf(list->error_str, sizeof(list->error_str),
3172                                  "Out of space, %d bytes is too small",
3173                                  list->alloc_len);
3174                         break;
3175                 }
3176
3177                 sbuf_finish(sb);
3178
3179                 retval = copyout(sbuf_data(sb), list->lun_xml,
3180                                  sbuf_len(sb) + 1);
3181
3182                 list->fill_len = sbuf_len(sb) + 1;
3183                 list->status = CTL_LUN_LIST_OK;
3184                 sbuf_delete(sb);
3185                 break;
3186         }
3187         case CTL_ISCSI: {
3188                 struct ctl_iscsi *ci;
3189                 struct ctl_frontend *fe;
3190
3191                 ci = (struct ctl_iscsi *)addr;
3192
3193                 fe = ctl_frontend_find("iscsi");
3194                 if (fe == NULL) {
3195                         ci->status = CTL_ISCSI_ERROR;
3196                         snprintf(ci->error_str, sizeof(ci->error_str),
3197                             "Frontend \"iscsi\" not found.");
3198                         break;
3199                 }
3200
3201                 retval = fe->ioctl(dev, cmd, addr, flag, td);
3202                 break;
3203         }
3204         case CTL_PORT_REQ: {
3205                 struct ctl_req *req;
3206                 struct ctl_frontend *fe;
3207                 void *packed;
3208                 nvlist_t *tmp_args_nvl;
3209                 size_t packed_len;
3210
3211                 req = (struct ctl_req *)addr;
3212                 tmp_args_nvl = req->args_nvl;
3213
3214                 fe = ctl_frontend_find(req->driver);
3215                 if (fe == NULL) {
3216                         req->status = CTL_LUN_ERROR;
3217                         snprintf(req->error_str, sizeof(req->error_str),
3218                             "Frontend \"%s\" not found.", req->driver);
3219                         break;
3220                 }
3221
3222                 if (req->args != NULL) {
3223                         packed = malloc(req->args_len, M_CTL, M_WAITOK);
3224                         if (copyin(req->args, packed, req->args_len) != 0) {
3225                                 free(packed, M_CTL);
3226                                 req->status = CTL_LUN_ERROR;
3227                                 snprintf(req->error_str, sizeof(req->error_str),
3228                                     "Cannot copyin args.");
3229                                 break;
3230                         }
3231                         req->args_nvl = nvlist_unpack(packed,
3232                             req->args_len, 0);
3233                         free(packed, M_CTL);
3234
3235                         if (req->args_nvl == NULL) {
3236                                 req->status = CTL_LUN_ERROR;
3237                                 snprintf(req->error_str, sizeof(req->error_str),
3238                                     "Cannot unpack args nvlist.");
3239                                 break;
3240                         }
3241                 } else
3242                         req->args_nvl = nvlist_create(0);
3243
3244                 if (fe->ioctl)
3245                         retval = fe->ioctl(dev, cmd, addr, flag, td);
3246                 else
3247                         retval = ENODEV;
3248
3249                 nvlist_destroy(req->args_nvl);
3250                 req->args_nvl = tmp_args_nvl;
3251
3252                 if (req->result_nvl != NULL) {
3253                         if (req->result != NULL) {
3254                                 packed = nvlist_pack(req->result_nvl,
3255                                     &packed_len);
3256                                 if (packed == NULL) {
3257                                         req->status = CTL_LUN_ERROR;
3258                                         snprintf(req->error_str,
3259                                             sizeof(req->error_str),
3260                                             "Cannot pack result nvlist.");
3261                                         break;
3262                                 }
3263
3264                                 if (packed_len > req->result_len) {
3265                                         req->status = CTL_LUN_ERROR;
3266                                         snprintf(req->error_str,
3267                                             sizeof(req->error_str),
3268                                             "Result nvlist too large.");
3269                                         free(packed, M_NVLIST);
3270                                         break;
3271                                 }
3272
3273                                 if (copyout(packed, req->result, packed_len)) {
3274                                         req->status = CTL_LUN_ERROR;
3275                                         snprintf(req->error_str,
3276                                             sizeof(req->error_str),
3277                                             "Cannot copyout() the result.");
3278                                         free(packed, M_NVLIST);
3279                                         break;
3280                                 }
3281
3282                                 req->result_len = packed_len;
3283                                 free(packed, M_NVLIST);
3284                         }
3285
3286                         nvlist_destroy(req->result_nvl);
3287                 }
3288                 break;
3289         }
3290         case CTL_PORT_LIST: {
3291                 struct sbuf *sb;
3292                 struct ctl_port *port;
3293                 struct ctl_lun_list *list;
3294                 const char *name, *value;
3295                 void *cookie;
3296                 int j, type;
3297                 uint32_t plun;
3298
3299                 list = (struct ctl_lun_list *)addr;
3300
3301                 sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3302                 if (sb == NULL) {
3303                         list->status = CTL_LUN_LIST_ERROR;
3304                         snprintf(list->error_str, sizeof(list->error_str),
3305                                  "Unable to allocate %d bytes for LUN list",
3306                                  list->alloc_len);
3307                         break;
3308                 }
3309
3310                 sbuf_printf(sb, "<ctlportlist>\n");
3311
3312                 mtx_lock(&softc->ctl_lock);
3313                 STAILQ_FOREACH(port, &softc->port_list, links) {
3314                         retval = sbuf_printf(sb, "<targ_port id=\"%ju\">\n",
3315                                              (uintmax_t)port->targ_port);
3316
3317                         /*
3318                          * Bail out as soon as we see that we've overfilled
3319                          * the buffer.
3320                          */
3321                         if (retval != 0)
3322                                 break;
3323
3324                         retval = sbuf_printf(sb, "\t<frontend_type>%s"
3325                             "</frontend_type>\n", port->frontend->name);
3326                         if (retval != 0)
3327                                 break;
3328
3329                         retval = sbuf_printf(sb, "\t<port_type>%d</port_type>\n",
3330                                              port->port_type);
3331                         if (retval != 0)
3332                                 break;
3333
3334                         retval = sbuf_printf(sb, "\t<online>%s</online>\n",
3335                             (port->status & CTL_PORT_STATUS_ONLINE) ? "YES" : "NO");
3336                         if (retval != 0)
3337                                 break;
3338
3339                         retval = sbuf_printf(sb, "\t<port_name>%s</port_name>\n",
3340                             port->port_name);
3341                         if (retval != 0)
3342                                 break;
3343
3344                         retval = sbuf_printf(sb, "\t<physical_port>%d</physical_port>\n",
3345                             port->physical_port);
3346                         if (retval != 0)
3347                                 break;
3348
3349                         retval = sbuf_printf(sb, "\t<virtual_port>%d</virtual_port>\n",
3350                             port->virtual_port);
3351                         if (retval != 0)
3352                                 break;
3353
3354                         if (port->target_devid != NULL) {
3355                                 sbuf_printf(sb, "\t<target>");
3356                                 ctl_id_sbuf(port->target_devid, sb);
3357                                 sbuf_printf(sb, "</target>\n");
3358                         }
3359
3360                         if (port->port_devid != NULL) {
3361                                 sbuf_printf(sb, "\t<port>");
3362                                 ctl_id_sbuf(port->port_devid, sb);
3363                                 sbuf_printf(sb, "</port>\n");
3364                         }
3365
3366                         if (port->port_info != NULL) {
3367                                 retval = port->port_info(port->onoff_arg, sb);
3368                                 if (retval != 0)
3369                                         break;
3370                         }
3371
3372                         cookie = NULL;
3373                         while ((name = nvlist_next(port->options, &type,
3374                             &cookie)) != NULL) {
3375                                 sbuf_printf(sb, "\t<%s>", name);
3376
3377                                 if (type == NV_TYPE_STRING) {
3378                                         value = dnvlist_get_string(port->options,
3379                                             name, NULL);
3380                                         if (value != NULL)
3381                                                 sbuf_printf(sb, "%s", value);
3382                                 }
3383
3384                                 sbuf_printf(sb, "</%s>\n", name);
3385                         }
3386
3387                         if (port->lun_map != NULL) {
3388                                 sbuf_printf(sb, "\t<lun_map>on</lun_map>\n");
3389                                 for (j = 0; j < port->lun_map_size; j++) {
3390                                         plun = ctl_lun_map_from_port(port, j);
3391                                         if (plun == UINT32_MAX)
3392                                                 continue;
3393                                         sbuf_printf(sb,
3394                                             "\t<lun id=\"%u\">%u</lun>\n",
3395                                             j, plun);
3396                                 }
3397                         }
3398
3399                         for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
3400                                 if (port->wwpn_iid[j].in_use == 0 ||
3401                                     (port->wwpn_iid[j].wwpn == 0 &&
3402                                      port->wwpn_iid[j].name == NULL))
3403                                         continue;
3404
3405                                 if (port->wwpn_iid[j].name != NULL)
3406                                         retval = sbuf_printf(sb,
3407                                             "\t<initiator id=\"%u\">%s</initiator>\n",
3408                                             j, port->wwpn_iid[j].name);
3409                                 else
3410                                         retval = sbuf_printf(sb,
3411                                             "\t<initiator id=\"%u\">naa.%08jx</initiator>\n",
3412                                             j, port->wwpn_iid[j].wwpn);
3413                                 if (retval != 0)
3414                                         break;
3415                         }
3416                         if (retval != 0)
3417                                 break;
3418
3419                         retval = sbuf_printf(sb, "</targ_port>\n");
3420                         if (retval != 0)
3421                                 break;
3422                 }
3423                 mtx_unlock(&softc->ctl_lock);
3424
3425                 if ((retval != 0)
3426                  || ((retval = sbuf_printf(sb, "</ctlportlist>\n")) != 0)) {
3427                         retval = 0;
3428                         sbuf_delete(sb);
3429                         list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3430                         snprintf(list->error_str, sizeof(list->error_str),
3431                                  "Out of space, %d bytes is too small",
3432                                  list->alloc_len);
3433                         break;
3434                 }
3435
3436                 sbuf_finish(sb);
3437
3438                 retval = copyout(sbuf_data(sb), list->lun_xml,
3439                                  sbuf_len(sb) + 1);
3440
3441                 list->fill_len = sbuf_len(sb) + 1;
3442                 list->status = CTL_LUN_LIST_OK;
3443                 sbuf_delete(sb);
3444                 break;
3445         }
3446         case CTL_LUN_MAP: {
3447                 struct ctl_lun_map *lm  = (struct ctl_lun_map *)addr;
3448                 struct ctl_port *port;
3449
3450                 mtx_lock(&softc->ctl_lock);
3451                 if (lm->port < softc->port_min ||
3452                     lm->port >= softc->port_max ||
3453                     (port = softc->ctl_ports[lm->port]) == NULL) {
3454                         mtx_unlock(&softc->ctl_lock);
3455                         return (ENXIO);
3456                 }
3457                 if (port->status & CTL_PORT_STATUS_ONLINE) {
3458                         STAILQ_FOREACH(lun, &softc->lun_list, links) {
3459                                 if (ctl_lun_map_to_port(port, lun->lun) ==
3460                                     UINT32_MAX)
3461                                         continue;
3462                                 mtx_lock(&lun->lun_lock);
3463                                 ctl_est_ua_port(lun, lm->port, -1,
3464                                     CTL_UA_LUN_CHANGE);
3465                                 mtx_unlock(&lun->lun_lock);
3466                         }
3467                 }
3468                 mtx_unlock(&softc->ctl_lock); // XXX: port_enable sleeps
3469                 if (lm->plun != UINT32_MAX) {
3470                         if (lm->lun == UINT32_MAX)
3471                                 retval = ctl_lun_map_unset(port, lm->plun);
3472                         else if (lm->lun < ctl_max_luns &&
3473                             softc->ctl_luns[lm->lun] != NULL)
3474                                 retval = ctl_lun_map_set(port, lm->plun, lm->lun);
3475                         else
3476                                 return (ENXIO);
3477                 } else {
3478                         if (lm->lun == UINT32_MAX)
3479                                 retval = ctl_lun_map_deinit(port);
3480                         else
3481                                 retval = ctl_lun_map_init(port);
3482                 }
3483                 if (port->status & CTL_PORT_STATUS_ONLINE)
3484                         ctl_isc_announce_port(port);
3485                 break;
3486         }
3487         case CTL_GET_LUN_STATS: {
3488                 struct ctl_get_io_stats *stats = (struct ctl_get_io_stats *)addr;
3489                 int i;
3490
3491                 /*
3492                  * XXX KDM no locking here.  If the LUN list changes,
3493                  * things can blow up.
3494                  */
3495                 i = 0;
3496                 stats->status = CTL_SS_OK;
3497                 stats->fill_len = 0;
3498                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
3499                         if (lun->lun < stats->first_item)
3500                                 continue;
3501                         if (stats->fill_len + sizeof(lun->stats) >
3502                             stats->alloc_len) {
3503                                 stats->status = CTL_SS_NEED_MORE_SPACE;
3504                                 break;
3505                         }
3506                         retval = copyout(&lun->stats, &stats->stats[i++],
3507                                          sizeof(lun->stats));
3508                         if (retval != 0)
3509                                 break;
3510                         stats->fill_len += sizeof(lun->stats);
3511                 }
3512                 stats->num_items = softc->num_luns;
3513                 stats->flags = CTL_STATS_FLAG_NONE;
3514 #ifdef CTL_TIME_IO
3515                 stats->flags |= CTL_STATS_FLAG_TIME_VALID;
3516 #endif
3517                 getnanouptime(&stats->timestamp);
3518                 break;
3519         }
3520         case CTL_GET_PORT_STATS: {
3521                 struct ctl_get_io_stats *stats = (struct ctl_get_io_stats *)addr;
3522                 int i;
3523
3524                 /*
3525                  * XXX KDM no locking here.  If the LUN list changes,
3526                  * things can blow up.
3527                  */
3528                 i = 0;
3529                 stats->status = CTL_SS_OK;
3530                 stats->fill_len = 0;
3531                 STAILQ_FOREACH(port, &softc->port_list, links) {
3532                         if (port->targ_port < stats->first_item)
3533                                 continue;
3534                         if (stats->fill_len + sizeof(port->stats) >
3535                             stats->alloc_len) {
3536                                 stats->status = CTL_SS_NEED_MORE_SPACE;
3537                                 break;
3538                         }
3539                         retval = copyout(&port->stats, &stats->stats[i++],
3540                                          sizeof(port->stats));
3541                         if (retval != 0)
3542                                 break;
3543                         stats->fill_len += sizeof(port->stats);
3544                 }
3545                 stats->num_items = softc->num_ports;
3546                 stats->flags = CTL_STATS_FLAG_NONE;
3547 #ifdef CTL_TIME_IO
3548                 stats->flags |= CTL_STATS_FLAG_TIME_VALID;
3549 #endif
3550                 getnanouptime(&stats->timestamp);
3551                 break;
3552         }
3553         default: {
3554                 /* XXX KDM should we fix this? */
3555 #if 0
3556                 struct ctl_backend_driver *backend;
3557                 unsigned int type;
3558                 int found;
3559
3560                 found = 0;
3561
3562                 /*
3563                  * We encode the backend type as the ioctl type for backend
3564                  * ioctls.  So parse it out here, and then search for a
3565                  * backend of this type.
3566                  */
3567                 type = _IOC_TYPE(cmd);
3568
3569                 STAILQ_FOREACH(backend, &softc->be_list, links) {
3570                         if (backend->type == type) {
3571                                 found = 1;
3572                                 break;
3573                         }
3574                 }
3575                 if (found == 0) {
3576                         printf("ctl: unknown ioctl command %#lx or backend "
3577                                "%d\n", cmd, type);
3578                         retval = EINVAL;
3579                         break;
3580                 }
3581                 retval = backend->ioctl(dev, cmd, addr, flag, td);
3582 #endif
3583                 retval = ENOTTY;
3584                 break;
3585         }
3586         }
3587         return (retval);
3588 }
3589
3590 uint32_t
3591 ctl_get_initindex(struct ctl_nexus *nexus)
3592 {
3593         return (nexus->initid + (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3594 }
3595
3596 int
3597 ctl_lun_map_init(struct ctl_port *port)
3598 {
3599         struct ctl_softc *softc = port->ctl_softc;
3600         struct ctl_lun *lun;
3601         int size = ctl_lun_map_size;
3602         uint32_t i;
3603
3604         if (port->lun_map == NULL || port->lun_map_size < size) {
3605                 port->lun_map_size = 0;
3606                 free(port->lun_map, M_CTL);
3607                 port->lun_map = malloc(size * sizeof(uint32_t),
3608                     M_CTL, M_NOWAIT);
3609         }
3610         if (port->lun_map == NULL)
3611                 return (ENOMEM);
3612         for (i = 0; i < size; i++)
3613                 port->lun_map[i] = UINT32_MAX;
3614         port->lun_map_size = size;
3615         if (port->status & CTL_PORT_STATUS_ONLINE) {
3616                 if (port->lun_disable != NULL) {
3617                         STAILQ_FOREACH(lun, &softc->lun_list, links)
3618                                 port->lun_disable(port->targ_lun_arg, lun->lun);
3619                 }
3620                 ctl_isc_announce_port(port);
3621         }
3622         return (0);
3623 }
3624
3625 int
3626 ctl_lun_map_deinit(struct ctl_port *port)
3627 {
3628         struct ctl_softc *softc = port->ctl_softc;
3629         struct ctl_lun *lun;
3630
3631         if (port->lun_map == NULL)
3632                 return (0);
3633         port->lun_map_size = 0;
3634         free(port->lun_map, M_CTL);
3635         port->lun_map = NULL;
3636         if (port->status & CTL_PORT_STATUS_ONLINE) {
3637                 if (port->lun_enable != NULL) {
3638                         STAILQ_FOREACH(lun, &softc->lun_list, links)
3639                                 port->lun_enable(port->targ_lun_arg, lun->lun);
3640                 }
3641                 ctl_isc_announce_port(port);
3642         }
3643         return (0);
3644 }
3645
3646 int
3647 ctl_lun_map_set(struct ctl_port *port, uint32_t plun, uint32_t glun)
3648 {
3649         int status;
3650         uint32_t old;
3651
3652         if (port->lun_map == NULL) {
3653                 status = ctl_lun_map_init(port);
3654                 if (status != 0)
3655                         return (status);
3656         }
3657         if (plun >= port->lun_map_size)
3658                 return (EINVAL);
3659         old = port->lun_map[plun];
3660         port->lun_map[plun] = glun;
3661         if ((port->status & CTL_PORT_STATUS_ONLINE) && old == UINT32_MAX) {
3662                 if (port->lun_enable != NULL)
3663                         port->lun_enable(port->targ_lun_arg, plun);
3664                 ctl_isc_announce_port(port);
3665         }
3666         return (0);
3667 }
3668
3669 int
3670 ctl_lun_map_unset(struct ctl_port *port, uint32_t plun)
3671 {
3672         uint32_t old;
3673
3674         if (port->lun_map == NULL || plun >= port->lun_map_size)
3675                 return (0);
3676         old = port->lun_map[plun];
3677         port->lun_map[plun] = UINT32_MAX;
3678         if ((port->status & CTL_PORT_STATUS_ONLINE) && old != UINT32_MAX) {
3679                 if (port->lun_disable != NULL)
3680                         port->lun_disable(port->targ_lun_arg, plun);
3681                 ctl_isc_announce_port(port);
3682         }
3683         return (0);
3684 }
3685
3686 uint32_t
3687 ctl_lun_map_from_port(struct ctl_port *port, uint32_t lun_id)
3688 {
3689
3690         if (port == NULL)
3691                 return (UINT32_MAX);
3692         if (port->lun_map == NULL)
3693                 return (lun_id);
3694         if (lun_id > port->lun_map_size)
3695                 return (UINT32_MAX);
3696         return (port->lun_map[lun_id]);
3697 }
3698
3699 uint32_t
3700 ctl_lun_map_to_port(struct ctl_port *port, uint32_t lun_id)
3701 {
3702         uint32_t i;
3703
3704         if (port == NULL)
3705                 return (UINT32_MAX);
3706         if (port->lun_map == NULL)
3707                 return (lun_id);
3708         for (i = 0; i < port->lun_map_size; i++) {
3709                 if (port->lun_map[i] == lun_id)
3710                         return (i);
3711         }
3712         return (UINT32_MAX);
3713 }
3714
3715 uint32_t
3716 ctl_decode_lun(uint64_t encoded)
3717 {
3718         uint8_t lun[8];
3719         uint32_t result = 0xffffffff;
3720
3721         be64enc(lun, encoded);
3722         switch (lun[0] & RPL_LUNDATA_ATYP_MASK) {
3723         case RPL_LUNDATA_ATYP_PERIPH:
3724                 if ((lun[0] & 0x3f) == 0 && lun[2] == 0 && lun[3] == 0 &&
3725                     lun[4] == 0 && lun[5] == 0 && lun[6] == 0 && lun[7] == 0)
3726                         result = lun[1];
3727                 break;
3728         case RPL_LUNDATA_ATYP_FLAT:
3729                 if (lun[2] == 0 && lun[3] == 0 && lun[4] == 0 && lun[5] == 0 &&
3730                     lun[6] == 0 && lun[7] == 0)
3731                         result = ((lun[0] & 0x3f) << 8) + lun[1];
3732                 break;
3733         case RPL_LUNDATA_ATYP_EXTLUN:
3734                 switch (lun[0] & RPL_LUNDATA_EXT_EAM_MASK) {
3735                 case 0x02:
3736                         switch (lun[0] & RPL_LUNDATA_EXT_LEN_MASK) {
3737                         case 0x00:
3738                                 result = lun[1];
3739                                 break;
3740                         case 0x10:
3741                                 result = (lun[1] << 16) + (lun[2] << 8) +
3742                                     lun[3];
3743                                 break;
3744                         case 0x20:
3745                                 if (lun[1] == 0 && lun[6] == 0 && lun[7] == 0)
3746                                         result = (lun[2] << 24) +
3747                                             (lun[3] << 16) + (lun[4] << 8) +
3748                                             lun[5];
3749                                 break;
3750                         }
3751                         break;
3752                 case RPL_LUNDATA_EXT_EAM_NOT_SPEC:
3753                         result = 0xffffffff;
3754                         break;
3755                 }
3756                 break;
3757         }
3758         return (result);
3759 }
3760
3761 uint64_t
3762 ctl_encode_lun(uint32_t decoded)
3763 {
3764         uint64_t l = decoded;
3765
3766         if (l <= 0xff)
3767                 return (((uint64_t)RPL_LUNDATA_ATYP_PERIPH << 56) | (l << 48));
3768         if (l <= 0x3fff)
3769                 return (((uint64_t)RPL_LUNDATA_ATYP_FLAT << 56) | (l << 48));
3770         if (l <= 0xffffff)
3771                 return (((uint64_t)(RPL_LUNDATA_ATYP_EXTLUN | 0x12) << 56) |
3772                     (l << 32));
3773         return ((((uint64_t)RPL_LUNDATA_ATYP_EXTLUN | 0x22) << 56) | (l << 16));
3774 }
3775
3776 int
3777 ctl_ffz(uint32_t *mask, uint32_t first, uint32_t last)
3778 {
3779         int i;
3780
3781         for (i = first; i < last; i++) {
3782                 if ((mask[i / 32] & (1 << (i % 32))) == 0)
3783                         return (i);
3784         }
3785         return (-1);
3786 }
3787
3788 int
3789 ctl_set_mask(uint32_t *mask, uint32_t bit)
3790 {
3791         uint32_t chunk, piece;
3792
3793         chunk = bit >> 5;
3794         piece = bit % (sizeof(uint32_t) * 8);
3795
3796         if ((mask[chunk] & (1 << piece)) != 0)
3797                 return (-1);
3798         else
3799                 mask[chunk] |= (1 << piece);
3800
3801         return (0);
3802 }
3803
3804 int
3805 ctl_clear_mask(uint32_t *mask, uint32_t bit)
3806 {
3807         uint32_t chunk, piece;
3808
3809         chunk = bit >> 5;
3810         piece = bit % (sizeof(uint32_t) * 8);
3811
3812         if ((mask[chunk] & (1 << piece)) == 0)
3813                 return (-1);
3814         else
3815                 mask[chunk] &= ~(1 << piece);
3816
3817         return (0);
3818 }
3819
3820 int
3821 ctl_is_set(uint32_t *mask, uint32_t bit)
3822 {
3823         uint32_t chunk, piece;
3824
3825         chunk = bit >> 5;
3826         piece = bit % (sizeof(uint32_t) * 8);
3827
3828         if ((mask[chunk] & (1 << piece)) == 0)
3829                 return (0);
3830         else
3831                 return (1);
3832 }
3833
3834 static uint64_t
3835 ctl_get_prkey(struct ctl_lun *lun, uint32_t residx)
3836 {
3837         uint64_t *t;
3838
3839         t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3840         if (t == NULL)
3841                 return (0);
3842         return (t[residx % CTL_MAX_INIT_PER_PORT]);
3843 }
3844
3845 static void
3846 ctl_clr_prkey(struct ctl_lun *lun, uint32_t residx)
3847 {
3848         uint64_t *t;
3849
3850         t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3851         if (t == NULL)
3852                 return;
3853         t[residx % CTL_MAX_INIT_PER_PORT] = 0;
3854 }
3855
3856 static void
3857 ctl_alloc_prkey(struct ctl_lun *lun, uint32_t residx)
3858 {
3859         uint64_t *p;
3860         u_int i;
3861
3862         i = residx/CTL_MAX_INIT_PER_PORT;
3863         if (lun->pr_keys[i] != NULL)
3864                 return;
3865         mtx_unlock(&lun->lun_lock);
3866         p = malloc(sizeof(uint64_t) * CTL_MAX_INIT_PER_PORT, M_CTL,
3867             M_WAITOK | M_ZERO);
3868         mtx_lock(&lun->lun_lock);
3869         if (lun->pr_keys[i] == NULL)
3870                 lun->pr_keys[i] = p;
3871         else
3872                 free(p, M_CTL);
3873 }
3874
3875 static void
3876 ctl_set_prkey(struct ctl_lun *lun, uint32_t residx, uint64_t key)
3877 {
3878         uint64_t *t;
3879
3880         t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3881         KASSERT(t != NULL, ("prkey %d is not allocated", residx));
3882         t[residx % CTL_MAX_INIT_PER_PORT] = key;
3883 }
3884
3885 /*
3886  * ctl_softc, pool_name, total_ctl_io are passed in.
3887  * npool is passed out.
3888  */
3889 int
3890 ctl_pool_create(struct ctl_softc *ctl_softc, const char *pool_name,
3891                 uint32_t total_ctl_io, void **npool)
3892 {
3893         struct ctl_io_pool *pool;
3894
3895         pool = (struct ctl_io_pool *)malloc(sizeof(*pool), M_CTL,
3896                                             M_NOWAIT | M_ZERO);
3897         if (pool == NULL)
3898                 return (ENOMEM);
3899
3900         snprintf(pool->name, sizeof(pool->name), "CTL IO %s", pool_name);
3901         pool->ctl_softc = ctl_softc;
3902 #ifdef IO_POOLS
3903         pool->zone = uma_zsecond_create(pool->name, NULL,
3904             NULL, NULL, NULL, ctl_softc->io_zone);
3905         /* uma_prealloc(pool->zone, total_ctl_io); */
3906 #else
3907         pool->zone = ctl_softc->io_zone;
3908 #endif
3909
3910         *npool = pool;
3911         return (0);
3912 }
3913
3914 void
3915 ctl_pool_free(struct ctl_io_pool *pool)
3916 {
3917
3918         if (pool == NULL)
3919                 return;
3920
3921 #ifdef IO_POOLS
3922         uma_zdestroy(pool->zone);
3923 #endif
3924         free(pool, M_CTL);
3925 }
3926
3927 union ctl_io *
3928 ctl_alloc_io(void *pool_ref)
3929 {
3930         struct ctl_io_pool *pool = (struct ctl_io_pool *)pool_ref;
3931         union ctl_io *io;
3932
3933         io = uma_zalloc(pool->zone, M_WAITOK);
3934         if (io != NULL) {
3935                 io->io_hdr.pool = pool_ref;
3936                 CTL_SOFTC(io) = pool->ctl_softc;
3937         }
3938         return (io);
3939 }
3940
3941 union ctl_io *
3942 ctl_alloc_io_nowait(void *pool_ref)
3943 {
3944         struct ctl_io_pool *pool = (struct ctl_io_pool *)pool_ref;
3945         union ctl_io *io;
3946
3947         io = uma_zalloc(pool->zone, M_NOWAIT);
3948         if (io != NULL) {
3949                 io->io_hdr.pool = pool_ref;
3950                 CTL_SOFTC(io) = pool->ctl_softc;
3951         }
3952         return (io);
3953 }
3954
3955 void
3956 ctl_free_io(union ctl_io *io)
3957 {
3958         struct ctl_io_pool *pool;
3959
3960         if (io == NULL)
3961                 return;
3962
3963         pool = (struct ctl_io_pool *)io->io_hdr.pool;
3964         uma_zfree(pool->zone, io);
3965 }
3966
3967 void
3968 ctl_zero_io(union ctl_io *io)
3969 {
3970         struct ctl_io_pool *pool;
3971
3972         if (io == NULL)
3973                 return;
3974
3975         /*
3976          * May need to preserve linked list pointers at some point too.
3977          */
3978         pool = io->io_hdr.pool;
3979         memset(io, 0, sizeof(*io));
3980         io->io_hdr.pool = pool;
3981         CTL_SOFTC(io) = pool->ctl_softc;
3982 }
3983
3984 int
3985 ctl_expand_number(const char *buf, uint64_t *num)
3986 {
3987         char *endptr;
3988         uint64_t number;
3989         unsigned shift;
3990
3991         number = strtoq(buf, &endptr, 0);
3992
3993         switch (tolower((unsigned char)*endptr)) {
3994         case 'e':
3995                 shift = 60;
3996                 break;
3997         case 'p':
3998                 shift = 50;
3999                 break;
4000         case 't':
4001                 shift = 40;
4002                 break;
4003         case 'g':
4004                 shift = 30;
4005                 break;
4006         case 'm':
4007                 shift = 20;
4008                 break;
4009         case 'k':
4010                 shift = 10;
4011                 break;
4012         case 'b':
4013         case '\0': /* No unit. */
4014                 *num = number;
4015                 return (0);
4016         default:
4017                 /* Unrecognized unit. */
4018                 return (-1);
4019         }
4020
4021         if ((number << shift) >> shift != number) {
4022                 /* Overflow */
4023                 return (-1);
4024         }
4025         *num = number << shift;
4026         return (0);
4027 }
4028
4029
4030 /*
4031  * This routine could be used in the future to load default and/or saved
4032  * mode page parameters for a particuar lun.
4033  */
4034 static int
4035 ctl_init_page_index(struct ctl_lun *lun)
4036 {
4037         int i, page_code;
4038         struct ctl_page_index *page_index;
4039         const char *value;
4040         uint64_t ival;
4041
4042         memcpy(&lun->mode_pages.index, page_index_template,
4043                sizeof(page_index_template));
4044
4045         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
4046
4047                 page_index = &lun->mode_pages.index[i];
4048                 if (lun->be_lun->lun_type == T_DIRECT &&
4049                     (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
4050                         continue;
4051                 if (lun->be_lun->lun_type == T_PROCESSOR &&
4052                     (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
4053                         continue;
4054                 if (lun->be_lun->lun_type == T_CDROM &&
4055                     (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
4056                         continue;
4057
4058                 page_code = page_index->page_code & SMPH_PC_MASK;
4059                 switch (page_code) {
4060                 case SMS_RW_ERROR_RECOVERY_PAGE: {
4061                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4062                             ("subpage %#x for page %#x is incorrect!",
4063                             page_index->subpage, page_code));
4064                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_CURRENT],
4065                                &rw_er_page_default,
4066                                sizeof(rw_er_page_default));
4067                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_CHANGEABLE],
4068                                &rw_er_page_changeable,
4069                                sizeof(rw_er_page_changeable));
4070                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_DEFAULT],
4071                                &rw_er_page_default,
4072                                sizeof(rw_er_page_default));
4073                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_SAVED],
4074                                &rw_er_page_default,
4075                                sizeof(rw_er_page_default));
4076                         page_index->page_data =
4077                                 (uint8_t *)lun->mode_pages.rw_er_page;
4078                         break;
4079                 }
4080                 case SMS_FORMAT_DEVICE_PAGE: {
4081                         struct scsi_format_page *format_page;
4082
4083                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4084                             ("subpage %#x for page %#x is incorrect!",
4085                             page_index->subpage, page_code));
4086
4087                         /*
4088                          * Sectors per track are set above.  Bytes per
4089                          * sector need to be set here on a per-LUN basis.
4090                          */
4091                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_CURRENT],
4092                                &format_page_default,
4093                                sizeof(format_page_default));
4094                         memcpy(&lun->mode_pages.format_page[
4095                                CTL_PAGE_CHANGEABLE], &format_page_changeable,
4096                                sizeof(format_page_changeable));
4097                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_DEFAULT],
4098                                &format_page_default,
4099                                sizeof(format_page_default));
4100                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_SAVED],
4101                                &format_page_default,
4102                                sizeof(format_page_default));
4103
4104                         format_page = &lun->mode_pages.format_page[
4105                                 CTL_PAGE_CURRENT];
4106                         scsi_ulto2b(lun->be_lun->blocksize,
4107                                     format_page->bytes_per_sector);
4108
4109                         format_page = &lun->mode_pages.format_page[
4110                                 CTL_PAGE_DEFAULT];
4111                         scsi_ulto2b(lun->be_lun->blocksize,
4112                                     format_page->bytes_per_sector);
4113
4114                         format_page = &lun->mode_pages.format_page[
4115                                 CTL_PAGE_SAVED];
4116                         scsi_ulto2b(lun->be_lun->blocksize,
4117                                     format_page->bytes_per_sector);
4118
4119                         page_index->page_data =
4120                                 (uint8_t *)lun->mode_pages.format_page;
4121                         break;
4122                 }
4123                 case SMS_RIGID_DISK_PAGE: {
4124                         struct scsi_rigid_disk_page *rigid_disk_page;
4125                         uint32_t sectors_per_cylinder;
4126                         uint64_t cylinders;
4127 #ifndef __XSCALE__
4128                         int shift;
4129 #endif /* !__XSCALE__ */
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                          * Rotation rate and sectors per track are set
4137                          * above.  We calculate the cylinders here based on
4138                          * capacity.  Due to the number of heads and
4139                          * sectors per track we're using, smaller arrays
4140                          * may turn out to have 0 cylinders.  Linux and
4141                          * FreeBSD don't pay attention to these mode pages
4142                          * to figure out capacity, but Solaris does.  It
4143                          * seems to deal with 0 cylinders just fine, and
4144                          * works out a fake geometry based on the capacity.
4145                          */
4146                         memcpy(&lun->mode_pages.rigid_disk_page[
4147                                CTL_PAGE_DEFAULT], &rigid_disk_page_default,
4148                                sizeof(rigid_disk_page_default));
4149                         memcpy(&lun->mode_pages.rigid_disk_page[
4150                                CTL_PAGE_CHANGEABLE],&rigid_disk_page_changeable,
4151                                sizeof(rigid_disk_page_changeable));
4152
4153                         sectors_per_cylinder = CTL_DEFAULT_SECTORS_PER_TRACK *
4154                                 CTL_DEFAULT_HEADS;
4155
4156                         /*
4157                          * The divide method here will be more accurate,
4158                          * probably, but results in floating point being
4159                          * used in the kernel on i386 (__udivdi3()).  On the
4160                          * XScale, though, __udivdi3() is implemented in
4161                          * software.
4162                          *
4163                          * The shift method for cylinder calculation is
4164                          * accurate if sectors_per_cylinder is a power of
4165                          * 2.  Otherwise it might be slightly off -- you
4166                          * might have a bit of a truncation problem.
4167                          */
4168 #ifdef  __XSCALE__
4169                         cylinders = (lun->be_lun->maxlba + 1) /
4170                                 sectors_per_cylinder;
4171 #else
4172                         for (shift = 31; shift > 0; shift--) {
4173                                 if (sectors_per_cylinder & (1 << shift))
4174                                         break;
4175                         }
4176                         cylinders = (lun->be_lun->maxlba + 1) >> shift;
4177 #endif
4178
4179                         /*
4180                          * We've basically got 3 bytes, or 24 bits for the
4181                          * cylinder size in the mode page.  If we're over,
4182                          * just round down to 2^24.
4183                          */
4184                         if (cylinders > 0xffffff)
4185                                 cylinders = 0xffffff;
4186
4187                         rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4188                                 CTL_PAGE_DEFAULT];
4189                         scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4190
4191                         if ((value = dnvlist_get_string(lun->be_lun->options,
4192                             "rpm", NULL)) != NULL) {
4193                                 scsi_ulto2b(strtol(value, NULL, 0),
4194                                      rigid_disk_page->rotation_rate);
4195                         }
4196
4197                         memcpy(&lun->mode_pages.rigid_disk_page[CTL_PAGE_CURRENT],
4198                                &lun->mode_pages.rigid_disk_page[CTL_PAGE_DEFAULT],
4199                                sizeof(rigid_disk_page_default));
4200                         memcpy(&lun->mode_pages.rigid_disk_page[CTL_PAGE_SAVED],
4201                                &lun->mode_pages.rigid_disk_page[CTL_PAGE_DEFAULT],
4202                                sizeof(rigid_disk_page_default));
4203
4204                         page_index->page_data =
4205                                 (uint8_t *)lun->mode_pages.rigid_disk_page;
4206                         break;
4207                 }
4208                 case SMS_VERIFY_ERROR_RECOVERY_PAGE: {
4209                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4210                             ("subpage %#x for page %#x is incorrect!",
4211                             page_index->subpage, page_code));
4212                         memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_CURRENT],
4213                                &verify_er_page_default,
4214                                sizeof(verify_er_page_default));
4215                         memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_CHANGEABLE],
4216                                &verify_er_page_changeable,
4217                                sizeof(verify_er_page_changeable));
4218                         memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_DEFAULT],
4219                                &verify_er_page_default,
4220                                sizeof(verify_er_page_default));
4221                         memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_SAVED],
4222                                &verify_er_page_default,
4223                                sizeof(verify_er_page_default));
4224                         page_index->page_data =
4225                                 (uint8_t *)lun->mode_pages.verify_er_page;
4226                         break;
4227                 }
4228                 case SMS_CACHING_PAGE: {
4229                         struct scsi_caching_page *caching_page;
4230
4231                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4232                             ("subpage %#x for page %#x is incorrect!",
4233                             page_index->subpage, page_code));
4234                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_DEFAULT],
4235                                &caching_page_default,
4236                                sizeof(caching_page_default));
4237                         memcpy(&lun->mode_pages.caching_page[
4238                                CTL_PAGE_CHANGEABLE], &caching_page_changeable,
4239                                sizeof(caching_page_changeable));
4240                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4241                                &caching_page_default,
4242                                sizeof(caching_page_default));
4243                         caching_page = &lun->mode_pages.caching_page[
4244                             CTL_PAGE_SAVED];
4245                         value = dnvlist_get_string(lun->be_lun->options,
4246                             "writecache", NULL);
4247                         if (value != NULL && strcmp(value, "off") == 0)
4248                                 caching_page->flags1 &= ~SCP_WCE;
4249                         value = dnvlist_get_string(lun->be_lun->options,
4250                             "readcache", NULL);
4251                         if (value != NULL && strcmp(value, "off") == 0)
4252                                 caching_page->flags1 |= SCP_RCD;
4253                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_CURRENT],
4254                                &lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4255                                sizeof(caching_page_default));
4256                         page_index->page_data =
4257                                 (uint8_t *)lun->mode_pages.caching_page;
4258                         break;
4259                 }
4260                 case SMS_CONTROL_MODE_PAGE: {
4261                         switch (page_index->subpage) {
4262                         case SMS_SUBPAGE_PAGE_0: {
4263                                 struct scsi_control_page *control_page;
4264
4265                                 memcpy(&lun->mode_pages.control_page[
4266                                     CTL_PAGE_DEFAULT],
4267                                        &control_page_default,
4268                                        sizeof(control_page_default));
4269                                 memcpy(&lun->mode_pages.control_page[
4270                                     CTL_PAGE_CHANGEABLE],
4271                                        &control_page_changeable,
4272                                        sizeof(control_page_changeable));
4273                                 memcpy(&lun->mode_pages.control_page[
4274                                     CTL_PAGE_SAVED],
4275                                        &control_page_default,
4276                                        sizeof(control_page_default));
4277                                 control_page = &lun->mode_pages.control_page[
4278                                     CTL_PAGE_SAVED];
4279                                 value = dnvlist_get_string(lun->be_lun->options,
4280                                     "reordering", NULL);
4281                                 if (value != NULL &&
4282                                     strcmp(value, "unrestricted") == 0) {
4283                                         control_page->queue_flags &=
4284                                             ~SCP_QUEUE_ALG_MASK;
4285                                         control_page->queue_flags |=
4286                                             SCP_QUEUE_ALG_UNRESTRICTED;
4287                                 }
4288                                 memcpy(&lun->mode_pages.control_page[
4289                                     CTL_PAGE_CURRENT],
4290                                        &lun->mode_pages.control_page[
4291                                     CTL_PAGE_SAVED],
4292                                        sizeof(control_page_default));
4293                                 page_index->page_data =
4294                                     (uint8_t *)lun->mode_pages.control_page;
4295                                 break;
4296                         }
4297                         case 0x01:
4298                                 memcpy(&lun->mode_pages.control_ext_page[
4299                                     CTL_PAGE_DEFAULT],
4300                                        &control_ext_page_default,
4301                                        sizeof(control_ext_page_default));
4302                                 memcpy(&lun->mode_pages.control_ext_page[
4303                                     CTL_PAGE_CHANGEABLE],
4304                                        &control_ext_page_changeable,
4305                                        sizeof(control_ext_page_changeable));
4306                                 memcpy(&lun->mode_pages.control_ext_page[
4307                                     CTL_PAGE_SAVED],
4308                                        &control_ext_page_default,
4309                                        sizeof(control_ext_page_default));
4310                                 memcpy(&lun->mode_pages.control_ext_page[
4311                                     CTL_PAGE_CURRENT],
4312                                        &lun->mode_pages.control_ext_page[
4313                                     CTL_PAGE_SAVED],
4314                                        sizeof(control_ext_page_default));
4315                                 page_index->page_data =
4316                                     (uint8_t *)lun->mode_pages.control_ext_page;
4317                                 break;
4318                         default:
4319                                 panic("subpage %#x for page %#x is incorrect!",
4320                                       page_index->subpage, page_code);
4321                         }
4322                         break;
4323                 }
4324                 case SMS_INFO_EXCEPTIONS_PAGE: {
4325                         switch (page_index->subpage) {
4326                         case SMS_SUBPAGE_PAGE_0:
4327                                 memcpy(&lun->mode_pages.ie_page[CTL_PAGE_CURRENT],
4328                                        &ie_page_default,
4329                                        sizeof(ie_page_default));
4330                                 memcpy(&lun->mode_pages.ie_page[
4331                                        CTL_PAGE_CHANGEABLE], &ie_page_changeable,
4332                                        sizeof(ie_page_changeable));
4333                                 memcpy(&lun->mode_pages.ie_page[CTL_PAGE_DEFAULT],
4334                                        &ie_page_default,
4335                                        sizeof(ie_page_default));
4336                                 memcpy(&lun->mode_pages.ie_page[CTL_PAGE_SAVED],
4337                                        &ie_page_default,
4338                                        sizeof(ie_page_default));
4339                                 page_index->page_data =
4340                                         (uint8_t *)lun->mode_pages.ie_page;
4341                                 break;
4342                         case 0x02: {
4343                                 struct ctl_logical_block_provisioning_page *page;
4344
4345                                 memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_DEFAULT],
4346                                        &lbp_page_default,
4347                                        sizeof(lbp_page_default));
4348                                 memcpy(&lun->mode_pages.lbp_page[
4349                                        CTL_PAGE_CHANGEABLE], &lbp_page_changeable,
4350                                        sizeof(lbp_page_changeable));
4351                                 memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_SAVED],
4352                                        &lbp_page_default,
4353                                        sizeof(lbp_page_default));
4354                                 page = &lun->mode_pages.lbp_page[CTL_PAGE_SAVED];
4355                                 value = dnvlist_get_string(lun->be_lun->options,
4356                                     "avail-threshold", NULL);
4357                                 if (value != NULL &&
4358                                     ctl_expand_number(value, &ival) == 0) {
4359                                         page->descr[0].flags |= SLBPPD_ENABLED |
4360                                             SLBPPD_ARMING_DEC;
4361                                         if (lun->be_lun->blocksize)
4362                                                 ival /= lun->be_lun->blocksize;
4363                                         else
4364                                                 ival /= 512;
4365                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4366                                             page->descr[0].count);
4367                                 }
4368                                 value = dnvlist_get_string(lun->be_lun->options,
4369                                     "used-threshold", NULL);
4370                                 if (value != NULL &&
4371                                     ctl_expand_number(value, &ival) == 0) {
4372                                         page->descr[1].flags |= SLBPPD_ENABLED |
4373                                             SLBPPD_ARMING_INC;
4374                                         if (lun->be_lun->blocksize)
4375                                                 ival /= lun->be_lun->blocksize;
4376                                         else
4377                                                 ival /= 512;
4378                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4379                                             page->descr[1].count);
4380                                 }
4381                                 value = dnvlist_get_string(lun->be_lun->options,
4382                                     "pool-avail-threshold", NULL);
4383                                 if (value != NULL &&
4384                                     ctl_expand_number(value, &ival) == 0) {
4385                                         page->descr[2].flags |= SLBPPD_ENABLED |
4386                                             SLBPPD_ARMING_DEC;
4387                                         if (lun->be_lun->blocksize)
4388                                                 ival /= lun->be_lun->blocksize;
4389                                         else
4390                                                 ival /= 512;
4391                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4392                                             page->descr[2].count);
4393                                 }
4394                                 value = dnvlist_get_string(lun->be_lun->options,
4395                                     "pool-used-threshold", NULL);
4396                                 if (value != NULL &&
4397                                     ctl_expand_number(value, &ival) == 0) {
4398                                         page->descr[3].flags |= SLBPPD_ENABLED |
4399                                             SLBPPD_ARMING_INC;
4400                                         if (lun->be_lun->blocksize)
4401                                                 ival /= lun->be_lun->blocksize;
4402                                         else
4403                                                 ival /= 512;
4404                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4405                                             page->descr[3].count);
4406                                 }
4407                                 memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_CURRENT],
4408                                        &lun->mode_pages.lbp_page[CTL_PAGE_SAVED],
4409                                        sizeof(lbp_page_default));
4410                                 page_index->page_data =
4411                                         (uint8_t *)lun->mode_pages.lbp_page;
4412                                 break;
4413                         }
4414                         default:
4415                                 panic("subpage %#x for page %#x is incorrect!",
4416                                       page_index->subpage, page_code);
4417                         }
4418                         break;
4419                 }
4420                 case SMS_CDDVD_CAPS_PAGE:{
4421                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4422                             ("subpage %#x for page %#x is incorrect!",
4423                             page_index->subpage, page_code));
4424                         memcpy(&lun->mode_pages.cddvd_page[CTL_PAGE_DEFAULT],
4425                                &cddvd_page_default,
4426                                sizeof(cddvd_page_default));
4427                         memcpy(&lun->mode_pages.cddvd_page[
4428                                CTL_PAGE_CHANGEABLE], &cddvd_page_changeable,
4429                                sizeof(cddvd_page_changeable));
4430                         memcpy(&lun->mode_pages.cddvd_page[CTL_PAGE_SAVED],
4431                                &cddvd_page_default,
4432                                sizeof(cddvd_page_default));
4433                         memcpy(&lun->mode_pages.cddvd_page[CTL_PAGE_CURRENT],
4434                                &lun->mode_pages.cddvd_page[CTL_PAGE_SAVED],
4435                                sizeof(cddvd_page_default));
4436                         page_index->page_data =
4437                                 (uint8_t *)lun->mode_pages.cddvd_page;
4438                         break;
4439                 }
4440                 default:
4441                         panic("invalid page code value %#x", page_code);
4442                 }
4443         }
4444
4445         return (CTL_RETVAL_COMPLETE);
4446 }
4447
4448 static int
4449 ctl_init_log_page_index(struct ctl_lun *lun)
4450 {
4451         struct ctl_page_index *page_index;
4452         int i, j, k, prev;
4453
4454         memcpy(&lun->log_pages.index, log_page_index_template,
4455                sizeof(log_page_index_template));
4456
4457         prev = -1;
4458         for (i = 0, j = 0, k = 0; i < CTL_NUM_LOG_PAGES; i++) {
4459
4460                 page_index = &lun->log_pages.index[i];
4461                 if (lun->be_lun->lun_type == T_DIRECT &&
4462                     (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
4463                         continue;
4464                 if (lun->be_lun->lun_type == T_PROCESSOR &&
4465                     (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
4466                         continue;
4467                 if (lun->be_lun->lun_type == T_CDROM &&
4468                     (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
4469                         continue;
4470
4471                 if (page_index->page_code == SLS_LOGICAL_BLOCK_PROVISIONING &&
4472                     lun->backend->lun_attr == NULL)
4473                         continue;
4474
4475                 if (page_index->page_code != prev) {
4476                         lun->log_pages.pages_page[j] = page_index->page_code;
4477                         prev = page_index->page_code;
4478                         j++;
4479                 }
4480                 lun->log_pages.subpages_page[k*2] = page_index->page_code;
4481                 lun->log_pages.subpages_page[k*2+1] = page_index->subpage;
4482                 k++;
4483         }
4484         lun->log_pages.index[0].page_data = &lun->log_pages.pages_page[0];
4485         lun->log_pages.index[0].page_len = j;
4486         lun->log_pages.index[1].page_data = &lun->log_pages.subpages_page[0];
4487         lun->log_pages.index[1].page_len = k * 2;
4488         lun->log_pages.index[2].page_data = &lun->log_pages.lbp_page[0];
4489         lun->log_pages.index[2].page_len = 12*CTL_NUM_LBP_PARAMS;
4490         lun->log_pages.index[3].page_data = (uint8_t *)&lun->log_pages.stat_page;
4491         lun->log_pages.index[3].page_len = sizeof(lun->log_pages.stat_page);
4492         lun->log_pages.index[4].page_data = (uint8_t *)&lun->log_pages.ie_page;
4493         lun->log_pages.index[4].page_len = sizeof(lun->log_pages.ie_page);
4494
4495         return (CTL_RETVAL_COMPLETE);
4496 }
4497
4498 static int
4499 hex2bin(const char *str, uint8_t *buf, int buf_size)
4500 {
4501         int i;
4502         u_char c;
4503
4504         memset(buf, 0, buf_size);
4505         while (isspace(str[0]))
4506                 str++;
4507         if (str[0] == '0' && (str[1] == 'x' || str[1] == 'X'))
4508                 str += 2;
4509         buf_size *= 2;
4510         for (i = 0; str[i] != 0 && i < buf_size; i++) {
4511                 while (str[i] == '-')   /* Skip dashes in UUIDs. */
4512                         str++;
4513                 c = str[i];
4514                 if (isdigit(c))
4515                         c -= '0';
4516                 else if (isalpha(c))
4517                         c -= isupper(c) ? 'A' - 10 : 'a' - 10;
4518                 else
4519                         break;
4520                 if (c >= 16)
4521                         break;
4522                 if ((i & 1) == 0)
4523                         buf[i / 2] |= (c << 4);
4524                 else
4525                         buf[i / 2] |= c;
4526         }
4527         return ((i + 1) / 2);
4528 }
4529
4530 /*
4531  * LUN allocation.
4532  *
4533  * Requirements:
4534  * - caller allocates and zeros LUN storage, or passes in a NULL LUN if he
4535  *   wants us to allocate the LUN and he can block.
4536  * - ctl_softc is always set
4537  * - be_lun is set if the LUN has a backend (needed for disk LUNs)
4538  *
4539  * Returns 0 for success, non-zero (errno) for failure.
4540  */
4541 static int
4542 ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *ctl_lun,
4543               struct ctl_be_lun *const be_lun)
4544 {
4545         struct ctl_lun *nlun, *lun;
4546         struct scsi_vpd_id_descriptor *desc;
4547         struct scsi_vpd_id_t10 *t10id;
4548         const char *eui, *naa, *scsiname, *uuid, *vendor, *value;
4549         int lun_number, lun_malloced;
4550         int devidlen, idlen1, idlen2 = 0, len;
4551
4552         if (be_lun == NULL)
4553                 return (EINVAL);
4554
4555         /*
4556          * We currently only support Direct Access or Processor LUN types.
4557          */
4558         switch (be_lun->lun_type) {
4559         case T_DIRECT:
4560         case T_PROCESSOR:
4561         case T_CDROM:
4562                 break;
4563         case T_SEQUENTIAL:
4564         case T_CHANGER:
4565         default:
4566                 be_lun->lun_config_status(be_lun->be_lun,
4567                                           CTL_LUN_CONFIG_FAILURE);
4568                 break;
4569         }
4570         if (ctl_lun == NULL) {
4571                 lun = malloc(sizeof(*lun), M_CTL, M_WAITOK);
4572                 lun_malloced = 1;
4573         } else {
4574                 lun_malloced = 0;
4575                 lun = ctl_lun;
4576         }
4577
4578         memset(lun, 0, sizeof(*lun));
4579         if (lun_malloced)
4580                 lun->flags = CTL_LUN_MALLOCED;
4581
4582         lun->pending_sense = malloc(sizeof(struct scsi_sense_data *) *
4583             ctl_max_ports, M_DEVBUF, M_WAITOK | M_ZERO);
4584         lun->pending_ua = malloc(sizeof(ctl_ua_type *) * ctl_max_ports,
4585             M_DEVBUF, M_WAITOK | M_ZERO);
4586         lun->pr_keys = malloc(sizeof(uint64_t *) * ctl_max_ports,
4587             M_DEVBUF, M_WAITOK | M_ZERO);
4588
4589         /* Generate LUN ID. */
4590         devidlen = max(CTL_DEVID_MIN_LEN,
4591             strnlen(be_lun->device_id, CTL_DEVID_LEN));
4592         idlen1 = sizeof(*t10id) + devidlen;
4593         len = sizeof(struct scsi_vpd_id_descriptor) + idlen1;
4594         scsiname = dnvlist_get_string(be_lun->options, "scsiname", NULL);
4595         if (scsiname != NULL) {
4596                 idlen2 = roundup2(strlen(scsiname) + 1, 4);
4597                 len += sizeof(struct scsi_vpd_id_descriptor) + idlen2;
4598         }
4599         eui = dnvlist_get_string(be_lun->options, "eui", NULL);
4600         if (eui != NULL) {
4601                 len += sizeof(struct scsi_vpd_id_descriptor) + 16;
4602         }
4603         naa = dnvlist_get_string(be_lun->options, "naa", NULL);
4604         if (naa != NULL) {
4605                 len += sizeof(struct scsi_vpd_id_descriptor) + 16;
4606         }
4607         uuid = dnvlist_get_string(be_lun->options, "uuid", NULL);
4608         if (uuid != NULL) {
4609                 len += sizeof(struct scsi_vpd_id_descriptor) + 18;
4610         }
4611         lun->lun_devid = malloc(sizeof(struct ctl_devid) + len,
4612             M_CTL, M_WAITOK | M_ZERO);
4613         desc = (struct scsi_vpd_id_descriptor *)lun->lun_devid->data;
4614         desc->proto_codeset = SVPD_ID_CODESET_ASCII;
4615         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_T10;
4616         desc->length = idlen1;
4617         t10id = (struct scsi_vpd_id_t10 *)&desc->identifier[0];
4618         memset(t10id->vendor, ' ', sizeof(t10id->vendor));
4619         if ((vendor = dnvlist_get_string(be_lun->options, "vendor", NULL)) == NULL) {
4620                 strncpy((char *)t10id->vendor, CTL_VENDOR, sizeof(t10id->vendor));
4621         } else {
4622                 strncpy(t10id->vendor, vendor,
4623                     min(sizeof(t10id->vendor), strlen(vendor)));
4624         }
4625         strncpy((char *)t10id->vendor_spec_id,
4626             (char *)be_lun->device_id, devidlen);
4627         if (scsiname != NULL) {
4628                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4629                     desc->length);
4630                 desc->proto_codeset = SVPD_ID_CODESET_UTF8;
4631                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4632                     SVPD_ID_TYPE_SCSI_NAME;
4633                 desc->length = idlen2;
4634                 strlcpy(desc->identifier, scsiname, idlen2);
4635         }
4636         if (eui != NULL) {
4637                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4638                     desc->length);
4639                 desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4640                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4641                     SVPD_ID_TYPE_EUI64;
4642                 desc->length = hex2bin(eui, desc->identifier, 16);
4643                 desc->length = desc->length > 12 ? 16 :
4644                     (desc->length > 8 ? 12 : 8);
4645                 len -= 16 - desc->length;
4646         }
4647         if (naa != NULL) {
4648                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4649                     desc->length);
4650                 desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4651                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4652                     SVPD_ID_TYPE_NAA;
4653                 desc->length = hex2bin(naa, desc->identifier, 16);
4654                 desc->length = desc->length > 8 ? 16 : 8;
4655                 len -= 16 - desc->length;
4656         }
4657         if (uuid != NULL) {
4658                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4659                     desc->length);
4660                 desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4661                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4662                     SVPD_ID_TYPE_UUID;
4663                 desc->identifier[0] = 0x10;
4664                 hex2bin(uuid, &desc->identifier[2], 16);
4665                 desc->length = 18;
4666         }
4667         lun->lun_devid->len = len;
4668
4669         mtx_lock(&ctl_softc->ctl_lock);
4670         /*
4671          * See if the caller requested a particular LUN number.  If so, see
4672          * if it is available.  Otherwise, allocate the first available LUN.
4673          */
4674         if (be_lun->flags & CTL_LUN_FLAG_ID_REQ) {
4675                 if ((be_lun->req_lun_id > (ctl_max_luns - 1))
4676                  || (ctl_is_set(ctl_softc->ctl_lun_mask, be_lun->req_lun_id))) {
4677                         mtx_unlock(&ctl_softc->ctl_lock);
4678                         if (be_lun->req_lun_id > (ctl_max_luns - 1)) {
4679                                 printf("ctl: requested LUN ID %d is higher "
4680                                        "than ctl_max_luns - 1 (%d)\n",
4681                                        be_lun->req_lun_id, ctl_max_luns - 1);
4682                         } else {
4683                                 /*
4684                                  * XXX KDM return an error, or just assign
4685                                  * another LUN ID in this case??
4686                                  */
4687                                 printf("ctl: requested LUN ID %d is already "
4688                                        "in use\n", be_lun->req_lun_id);
4689                         }
4690 fail:
4691                         free(lun->lun_devid, M_CTL);
4692                         if (lun->flags & CTL_LUN_MALLOCED)
4693                                 free(lun, M_CTL);
4694                         be_lun->lun_config_status(be_lun->be_lun,
4695                                                   CTL_LUN_CONFIG_FAILURE);
4696                         return (ENOSPC);
4697                 }
4698                 lun_number = be_lun->req_lun_id;
4699         } else {
4700                 lun_number = ctl_ffz(ctl_softc->ctl_lun_mask, 0, ctl_max_luns);
4701                 if (lun_number == -1) {
4702                         mtx_unlock(&ctl_softc->ctl_lock);
4703                         printf("ctl: can't allocate LUN, out of LUNs\n");
4704                         goto fail;
4705                 }
4706         }
4707         ctl_set_mask(ctl_softc->ctl_lun_mask, lun_number);
4708         mtx_unlock(&ctl_softc->ctl_lock);
4709
4710         mtx_init(&lun->lun_lock, "CTL LUN", NULL, MTX_DEF);
4711         lun->lun = lun_number;
4712         lun->be_lun = be_lun;
4713         /*
4714          * The processor LUN is always enabled.  Disk LUNs come on line
4715          * disabled, and must be enabled by the backend.
4716          */
4717         lun->flags |= CTL_LUN_DISABLED;
4718         lun->backend = be_lun->be;
4719         be_lun->ctl_lun = lun;
4720         be_lun->lun_id = lun_number;
4721         atomic_add_int(&be_lun->be->num_luns, 1);
4722         if (be_lun->flags & CTL_LUN_FLAG_EJECTED)
4723                 lun->flags |= CTL_LUN_EJECTED;
4724         if (be_lun->flags & CTL_LUN_FLAG_NO_MEDIA)
4725                 lun->flags |= CTL_LUN_NO_MEDIA;
4726         if (be_lun->flags & CTL_LUN_FLAG_STOPPED)
4727                 lun->flags |= CTL_LUN_STOPPED;
4728
4729         if (be_lun->flags & CTL_LUN_FLAG_PRIMARY)
4730                 lun->flags |= CTL_LUN_PRIMARY_SC;
4731
4732         value = dnvlist_get_string(be_lun->options, "removable", NULL);
4733         if (value != NULL) {
4734                 if (strcmp(value, "on") == 0)
4735                         lun->flags |= CTL_LUN_REMOVABLE;
4736         } else if (be_lun->lun_type == T_CDROM)
4737                 lun->flags |= CTL_LUN_REMOVABLE;
4738
4739         lun->ctl_softc = ctl_softc;
4740 #ifdef CTL_TIME_IO
4741         lun->last_busy = getsbinuptime();
4742 #endif
4743         TAILQ_INIT(&lun->ooa_queue);
4744         TAILQ_INIT(&lun->blocked_queue);
4745         STAILQ_INIT(&lun->error_list);
4746         lun->ie_reported = 1;
4747         callout_init_mtx(&lun->ie_callout, &lun->lun_lock, 0);
4748         ctl_tpc_lun_init(lun);
4749         if (lun->flags & CTL_LUN_REMOVABLE) {
4750                 lun->prevent = malloc((CTL_MAX_INITIATORS + 31) / 32 * 4,
4751                     M_CTL, M_WAITOK);
4752         }
4753
4754         /*
4755          * Initialize the mode and log page index.
4756          */
4757         ctl_init_page_index(lun);
4758         ctl_init_log_page_index(lun);
4759
4760         /* Setup statistics gathering */
4761 #ifdef CTL_LEGACY_STATS
4762         lun->legacy_stats.device_type = be_lun->lun_type;
4763         lun->legacy_stats.lun_number = lun_number;
4764         lun->legacy_stats.blocksize = be_lun->blocksize;
4765         if (be_lun->blocksize == 0)
4766                 lun->legacy_stats.flags = CTL_LUN_STATS_NO_BLOCKSIZE;
4767         lun->legacy_stats.ports = malloc(sizeof(struct ctl_lun_io_port_stats) *
4768             ctl_max_ports, M_DEVBUF, M_WAITOK | M_ZERO);
4769         for (len = 0; len < ctl_max_ports; len++)
4770                 lun->legacy_stats.ports[len].targ_port = len;
4771 #endif /* CTL_LEGACY_STATS */
4772         lun->stats.item = lun_number;
4773
4774         /*
4775          * Now, before we insert this lun on the lun list, set the lun
4776          * inventory changed UA for all other luns.
4777          */
4778         mtx_lock(&ctl_softc->ctl_lock);
4779         STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) {
4780                 mtx_lock(&nlun->lun_lock);
4781                 ctl_est_ua_all(nlun, -1, CTL_UA_LUN_CHANGE);
4782                 mtx_unlock(&nlun->lun_lock);
4783         }
4784         STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links);
4785         ctl_softc->ctl_luns[lun_number] = lun;
4786         ctl_softc->num_luns++;
4787         mtx_unlock(&ctl_softc->ctl_lock);
4788
4789         lun->be_lun->lun_config_status(lun->be_lun->be_lun, CTL_LUN_CONFIG_OK);
4790         return (0);
4791 }
4792
4793 /*
4794  * Delete a LUN.
4795  * Assumptions:
4796  * - LUN has already been marked invalid and any pending I/O has been taken
4797  *   care of.
4798  */
4799 static int
4800 ctl_free_lun(struct ctl_lun *lun)
4801 {
4802         struct ctl_softc *softc = lun->ctl_softc;
4803         struct ctl_lun *nlun;
4804         int i;
4805
4806         KASSERT(TAILQ_EMPTY(&lun->ooa_queue),
4807             ("Freeing a LUN %p with outstanding I/O!\n", lun));
4808
4809         mtx_lock(&softc->ctl_lock);
4810         STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links);
4811         ctl_clear_mask(softc->ctl_lun_mask, lun->lun);
4812         softc->ctl_luns[lun->lun] = NULL;
4813         softc->num_luns--;
4814         STAILQ_FOREACH(nlun, &softc->lun_list, links) {
4815                 mtx_lock(&nlun->lun_lock);
4816                 ctl_est_ua_all(nlun, -1, CTL_UA_LUN_CHANGE);
4817                 mtx_unlock(&nlun->lun_lock);
4818         }
4819         mtx_unlock(&softc->ctl_lock);
4820
4821         /*
4822          * Tell the backend to free resources, if this LUN has a backend.
4823          */
4824         atomic_subtract_int(&lun->be_lun->be->num_luns, 1);
4825         lun->be_lun->lun_shutdown(lun->be_lun->be_lun);
4826
4827         lun->ie_reportcnt = UINT32_MAX;
4828         callout_drain(&lun->ie_callout);
4829         ctl_tpc_lun_shutdown(lun);
4830         mtx_destroy(&lun->lun_lock);
4831         free(lun->lun_devid, M_CTL);
4832         for (i = 0; i < ctl_max_ports; i++)
4833                 free(lun->pending_ua[i], M_CTL);
4834         free(lun->pending_ua, M_DEVBUF);
4835         for (i = 0; i < ctl_max_ports; i++)
4836                 free(lun->pr_keys[i], M_CTL);
4837         free(lun->pr_keys, M_DEVBUF);
4838         free(lun->write_buffer, M_CTL);
4839         free(lun->prevent, M_CTL);
4840         if (lun->flags & CTL_LUN_MALLOCED)
4841                 free(lun, M_CTL);
4842
4843         return (0);
4844 }
4845
4846 static void
4847 ctl_create_lun(struct ctl_be_lun *be_lun)
4848 {
4849
4850         /*
4851          * ctl_alloc_lun() should handle all potential failure cases.
4852          */
4853         ctl_alloc_lun(control_softc, NULL, be_lun);
4854 }
4855
4856 int
4857 ctl_add_lun(struct ctl_be_lun *be_lun)
4858 {
4859         struct ctl_softc *softc = control_softc;
4860
4861         mtx_lock(&softc->ctl_lock);
4862         STAILQ_INSERT_TAIL(&softc->pending_lun_queue, be_lun, links);
4863         mtx_unlock(&softc->ctl_lock);
4864         wakeup(&softc->pending_lun_queue);
4865
4866         return (0);
4867 }
4868
4869 int
4870 ctl_enable_lun(struct ctl_be_lun *be_lun)
4871 {
4872         struct ctl_softc *softc;
4873         struct ctl_port *port, *nport;
4874         struct ctl_lun *lun;
4875         int retval;
4876
4877         lun = (struct ctl_lun *)be_lun->ctl_lun;
4878         softc = lun->ctl_softc;
4879
4880         mtx_lock(&softc->ctl_lock);
4881         mtx_lock(&lun->lun_lock);
4882         if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4883                 /*
4884                  * eh?  Why did we get called if the LUN is already
4885                  * enabled?
4886                  */
4887                 mtx_unlock(&lun->lun_lock);
4888                 mtx_unlock(&softc->ctl_lock);
4889                 return (0);
4890         }
4891         lun->flags &= ~CTL_LUN_DISABLED;
4892         mtx_unlock(&lun->lun_lock);
4893
4894         STAILQ_FOREACH_SAFE(port, &softc->port_list, links, nport) {
4895                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0 ||
4896                     port->lun_map != NULL || port->lun_enable == NULL)
4897                         continue;
4898
4899                 /*
4900                  * Drop the lock while we call the FETD's enable routine.
4901                  * This can lead to a callback into CTL (at least in the
4902                  * case of the internal initiator frontend.
4903                  */
4904                 mtx_unlock(&softc->ctl_lock);
4905                 retval = port->lun_enable(port->targ_lun_arg, lun->lun);
4906                 mtx_lock(&softc->ctl_lock);
4907                 if (retval != 0) {
4908                         printf("%s: FETD %s port %d returned error "
4909                                "%d for lun_enable on lun %jd\n",
4910                                __func__, port->port_name, port->targ_port,
4911                                retval, (intmax_t)lun->lun);
4912                 }
4913         }
4914
4915         mtx_unlock(&softc->ctl_lock);
4916         ctl_isc_announce_lun(lun);
4917
4918         return (0);
4919 }
4920
4921 int
4922 ctl_disable_lun(struct ctl_be_lun *be_lun)
4923 {
4924         struct ctl_softc *softc;
4925         struct ctl_port *port;
4926         struct ctl_lun *lun;
4927         int retval;
4928
4929         lun = (struct ctl_lun *)be_lun->ctl_lun;
4930         softc = lun->ctl_softc;
4931
4932         mtx_lock(&softc->ctl_lock);
4933         mtx_lock(&lun->lun_lock);
4934         if (lun->flags & CTL_LUN_DISABLED) {
4935                 mtx_unlock(&lun->lun_lock);
4936                 mtx_unlock(&softc->ctl_lock);
4937                 return (0);
4938         }
4939         lun->flags |= CTL_LUN_DISABLED;
4940         mtx_unlock(&lun->lun_lock);
4941
4942         STAILQ_FOREACH(port, &softc->port_list, links) {
4943                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0 ||
4944                     port->lun_map != NULL || port->lun_disable == NULL)
4945                         continue;
4946
4947                 /*
4948                  * Drop the lock before we call the frontend's disable
4949                  * routine, to avoid lock order reversals.
4950                  *
4951                  * XXX KDM what happens if the frontend list changes while
4952                  * we're traversing it?  It's unlikely, but should be handled.
4953                  */
4954                 mtx_unlock(&softc->ctl_lock);
4955                 retval = port->lun_disable(port->targ_lun_arg, lun->lun);
4956                 mtx_lock(&softc->ctl_lock);
4957                 if (retval != 0) {
4958                         printf("%s: FETD %s port %d returned error "
4959                                "%d for lun_disable on lun %jd\n",
4960                                __func__, port->port_name, port->targ_port,
4961                                retval, (intmax_t)lun->lun);
4962                 }
4963         }
4964
4965         mtx_unlock(&softc->ctl_lock);
4966         ctl_isc_announce_lun(lun);
4967
4968         return (0);
4969 }
4970
4971 int
4972 ctl_start_lun(struct ctl_be_lun *be_lun)
4973 {
4974         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4975
4976         mtx_lock(&lun->lun_lock);
4977         lun->flags &= ~CTL_LUN_STOPPED;
4978         mtx_unlock(&lun->lun_lock);
4979         return (0);
4980 }
4981
4982 int
4983 ctl_stop_lun(struct ctl_be_lun *be_lun)
4984 {
4985         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4986
4987         mtx_lock(&lun->lun_lock);
4988         lun->flags |= CTL_LUN_STOPPED;
4989         mtx_unlock(&lun->lun_lock);
4990         return (0);
4991 }
4992
4993 int
4994 ctl_lun_no_media(struct ctl_be_lun *be_lun)
4995 {
4996         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4997
4998         mtx_lock(&lun->lun_lock);
4999         lun->flags |= CTL_LUN_NO_MEDIA;
5000         mtx_unlock(&lun->lun_lock);
5001         return (0);
5002 }
5003
5004 int
5005 ctl_lun_has_media(struct ctl_be_lun *be_lun)
5006 {
5007         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
5008         union ctl_ha_msg msg;
5009
5010         mtx_lock(&lun->lun_lock);
5011         lun->flags &= ~(CTL_LUN_NO_MEDIA | CTL_LUN_EJECTED);
5012         if (lun->flags & CTL_LUN_REMOVABLE)
5013                 ctl_est_ua_all(lun, -1, CTL_UA_MEDIUM_CHANGE);
5014         mtx_unlock(&lun->lun_lock);
5015         if ((lun->flags & CTL_LUN_REMOVABLE) &&
5016             lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
5017                 bzero(&msg.ua, sizeof(msg.ua));
5018                 msg.hdr.msg_type = CTL_MSG_UA;
5019                 msg.hdr.nexus.initid = -1;
5020                 msg.hdr.nexus.targ_port = -1;
5021                 msg.hdr.nexus.targ_lun = lun->lun;
5022                 msg.hdr.nexus.targ_mapped_lun = lun->lun;
5023                 msg.ua.ua_all = 1;
5024                 msg.ua.ua_set = 1;
5025                 msg.ua.ua_type = CTL_UA_MEDIUM_CHANGE;
5026                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg.ua),
5027                     M_WAITOK);
5028         }
5029         return (0);
5030 }
5031
5032 int
5033 ctl_lun_ejected(struct ctl_be_lun *be_lun)
5034 {
5035         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
5036
5037         mtx_lock(&lun->lun_lock);
5038         lun->flags |= CTL_LUN_EJECTED;
5039         mtx_unlock(&lun->lun_lock);
5040         return (0);
5041 }
5042
5043 int
5044 ctl_lun_primary(struct ctl_be_lun *be_lun)
5045 {
5046         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
5047
5048         mtx_lock(&lun->lun_lock);
5049         lun->flags |= CTL_LUN_PRIMARY_SC;
5050         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
5051         mtx_unlock(&lun->lun_lock);
5052         ctl_isc_announce_lun(lun);
5053         return (0);
5054 }
5055
5056 int
5057 ctl_lun_secondary(struct ctl_be_lun *be_lun)
5058 {
5059         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
5060
5061         mtx_lock(&lun->lun_lock);
5062         lun->flags &= ~CTL_LUN_PRIMARY_SC;
5063         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
5064         mtx_unlock(&lun->lun_lock);
5065         ctl_isc_announce_lun(lun);
5066         return (0);
5067 }
5068
5069 int
5070 ctl_invalidate_lun(struct ctl_be_lun *be_lun)
5071 {
5072         struct ctl_lun *lun;
5073
5074         lun = (struct ctl_lun *)be_lun->ctl_lun;
5075
5076         mtx_lock(&lun->lun_lock);
5077
5078         /*
5079          * The LUN needs to be disabled before it can be marked invalid.
5080          */
5081         if ((lun->flags & CTL_LUN_DISABLED) == 0) {
5082                 mtx_unlock(&lun->lun_lock);
5083                 return (-1);
5084         }
5085         /*
5086          * Mark the LUN invalid.
5087          */
5088         lun->flags |= CTL_LUN_INVALID;
5089
5090         /*
5091          * If there is nothing in the OOA queue, go ahead and free the LUN.
5092          * If we have something in the OOA queue, we'll free it when the
5093          * last I/O completes.
5094          */
5095         if (TAILQ_EMPTY(&lun->ooa_queue)) {
5096                 mtx_unlock(&lun->lun_lock);
5097                 ctl_free_lun(lun);
5098         } else
5099                 mtx_unlock(&lun->lun_lock);
5100
5101         return (0);
5102 }
5103
5104 void
5105 ctl_lun_capacity_changed(struct ctl_be_lun *be_lun)
5106 {
5107         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
5108         union ctl_ha_msg msg;
5109
5110         mtx_lock(&lun->lun_lock);
5111         ctl_est_ua_all(lun, -1, CTL_UA_CAPACITY_CHANGE);
5112         mtx_unlock(&lun->lun_lock);
5113         if (lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
5114                 /* Send msg to other side. */
5115                 bzero(&msg.ua, sizeof(msg.ua));
5116                 msg.hdr.msg_type = CTL_MSG_UA;
5117                 msg.hdr.nexus.initid = -1;
5118                 msg.hdr.nexus.targ_port = -1;
5119                 msg.hdr.nexus.targ_lun = lun->lun;
5120                 msg.hdr.nexus.targ_mapped_lun = lun->lun;
5121                 msg.ua.ua_all = 1;
5122                 msg.ua.ua_set = 1;
5123                 msg.ua.ua_type = CTL_UA_CAPACITY_CHANGE;
5124                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg.ua),
5125                     M_WAITOK);
5126         }
5127 }
5128
5129 /*
5130  * Backend "memory move is complete" callback for requests that never
5131  * make it down to say RAIDCore's configuration code.
5132  */
5133 int
5134 ctl_config_move_done(union ctl_io *io)
5135 {
5136         int retval;
5137
5138         CTL_DEBUG_PRINT(("ctl_config_move_done\n"));
5139         KASSERT(io->io_hdr.io_type == CTL_IO_SCSI,
5140             ("Config I/O type isn't CTL_IO_SCSI (%d)!", io->io_hdr.io_type));
5141
5142         if ((io->io_hdr.port_status != 0) &&
5143             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5144              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5145                 ctl_set_internal_failure(&io->scsiio, /*sks_valid*/ 1,
5146                     /*retry_count*/ io->io_hdr.port_status);
5147         } else if (io->scsiio.kern_data_resid != 0 &&
5148             (io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT &&
5149             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5150              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5151                 ctl_set_invalid_field_ciu(&io->scsiio);
5152         }
5153
5154         if (ctl_debug & CTL_DEBUG_CDB_DATA)
5155                 ctl_data_print(io);
5156         if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN) ||
5157             ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
5158              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS) ||
5159             ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) {
5160                 /*
5161                  * XXX KDM just assuming a single pointer here, and not a
5162                  * S/G list.  If we start using S/G lists for config data,
5163                  * we'll need to know how to clean them up here as well.
5164                  */
5165                 if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5166                         free(io->scsiio.kern_data_ptr, M_CTL);
5167                 ctl_done(io);
5168                 retval = CTL_RETVAL_COMPLETE;
5169         } else {
5170                 /*
5171                  * XXX KDM now we need to continue data movement.  Some
5172                  * options:
5173                  * - call ctl_scsiio() again?  We don't do this for data
5174                  *   writes, because for those at least we know ahead of
5175                  *   time where the write will go and how long it is.  For
5176                  *   config writes, though, that information is largely
5177                  *   contained within the write itself, thus we need to
5178                  *   parse out the data again.
5179                  *
5180                  * - Call some other function once the data is in?
5181                  */
5182
5183                 /*
5184                  * XXX KDM call ctl_scsiio() again for now, and check flag
5185                  * bits to see whether we're allocated or not.
5186                  */
5187                 retval = ctl_scsiio(&io->scsiio);
5188         }
5189         return (retval);
5190 }
5191
5192 /*
5193  * This gets called by a backend driver when it is done with a
5194  * data_submit method.
5195  */
5196 void
5197 ctl_data_submit_done(union ctl_io *io)
5198 {
5199         /*
5200          * If the IO_CONT flag is set, we need to call the supplied
5201          * function to continue processing the I/O, instead of completing
5202          * the I/O just yet.
5203          *
5204          * If there is an error, though, we don't want to keep processing.
5205          * Instead, just send status back to the initiator.
5206          */
5207         if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5208             (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5209             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5210              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5211                 io->scsiio.io_cont(io);
5212                 return;
5213         }
5214         ctl_done(io);
5215 }
5216
5217 /*
5218  * This gets called by a backend driver when it is done with a
5219  * configuration write.
5220  */
5221 void
5222 ctl_config_write_done(union ctl_io *io)
5223 {
5224         uint8_t *buf;
5225
5226         /*
5227          * If the IO_CONT flag is set, we need to call the supplied
5228          * function to continue processing the I/O, instead of completing
5229          * the I/O just yet.
5230          *
5231          * If there is an error, though, we don't want to keep processing.
5232          * Instead, just send status back to the initiator.
5233          */
5234         if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5235             (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5236             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5237              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5238                 io->scsiio.io_cont(io);
5239                 return;
5240         }
5241         /*
5242          * Since a configuration write can be done for commands that actually
5243          * have data allocated, like write buffer, and commands that have
5244          * no data, like start/stop unit, we need to check here.
5245          */
5246         if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5247                 buf = io->scsiio.kern_data_ptr;
5248         else
5249                 buf = NULL;
5250         ctl_done(io);
5251         if (buf)
5252                 free(buf, M_CTL);
5253 }
5254
5255 void
5256 ctl_config_read_done(union ctl_io *io)
5257 {
5258         uint8_t *buf;
5259
5260         /*
5261          * If there is some error -- we are done, skip data transfer.
5262          */
5263         if ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0 ||
5264             ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
5265              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) {
5266                 if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5267                         buf = io->scsiio.kern_data_ptr;
5268                 else
5269                         buf = NULL;
5270                 ctl_done(io);
5271                 if (buf)
5272                         free(buf, M_CTL);
5273                 return;
5274         }
5275
5276         /*
5277          * If the IO_CONT flag is set, we need to call the supplied
5278          * function to continue processing the I/O, instead of completing
5279          * the I/O just yet.
5280          */
5281         if (io->io_hdr.flags & CTL_FLAG_IO_CONT) {
5282                 io->scsiio.io_cont(io);
5283                 return;
5284         }
5285
5286         ctl_datamove(io);
5287 }
5288
5289 /*
5290  * SCSI release command.
5291  */
5292 int
5293 ctl_scsi_release(struct ctl_scsiio *ctsio)
5294 {
5295         struct ctl_lun *lun = CTL_LUN(ctsio);
5296         uint32_t residx;
5297
5298         CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
5299
5300         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5301
5302         /*
5303          * XXX KDM right now, we only support LUN reservation.  We don't
5304          * support 3rd party reservations, or extent reservations, which
5305          * might actually need the parameter list.  If we've gotten this
5306          * far, we've got a LUN reservation.  Anything else got kicked out
5307          * above.  So, according to SPC, ignore the length.
5308          */
5309
5310         mtx_lock(&lun->lun_lock);
5311
5312         /*
5313          * According to SPC, it is not an error for an intiator to attempt
5314          * to release a reservation on a LUN that isn't reserved, or that
5315          * is reserved by another initiator.  The reservation can only be
5316          * released, though, by the initiator who made it or by one of
5317          * several reset type events.
5318          */
5319         if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == residx))
5320                         lun->flags &= ~CTL_LUN_RESERVED;
5321
5322         mtx_unlock(&lun->lun_lock);
5323
5324         ctl_set_success(ctsio);
5325         ctl_done((union ctl_io *)ctsio);
5326         return (CTL_RETVAL_COMPLETE);
5327 }
5328
5329 int
5330 ctl_scsi_reserve(struct ctl_scsiio *ctsio)
5331 {
5332         struct ctl_lun *lun = CTL_LUN(ctsio);
5333         uint32_t residx;
5334
5335         CTL_DEBUG_PRINT(("ctl_reserve\n"));
5336
5337         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5338
5339         /*
5340          * XXX KDM right now, we only support LUN reservation.  We don't
5341          * support 3rd party reservations, or extent reservations, which
5342          * might actually need the parameter list.  If we've gotten this
5343          * far, we've got a LUN reservation.  Anything else got kicked out
5344          * above.  So, according to SPC, ignore the length.
5345          */
5346
5347         mtx_lock(&lun->lun_lock);
5348         if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx != residx)) {
5349                 ctl_set_reservation_conflict(ctsio);
5350                 goto bailout;
5351         }
5352
5353         /* SPC-3 exceptions to SPC-2 RESERVE and RELEASE behavior. */
5354         if (lun->flags & CTL_LUN_PR_RESERVED) {
5355                 ctl_set_success(ctsio);
5356                 goto bailout;
5357         }
5358
5359         lun->flags |= CTL_LUN_RESERVED;
5360         lun->res_idx = residx;
5361         ctl_set_success(ctsio);
5362
5363 bailout:
5364         mtx_unlock(&lun->lun_lock);
5365         ctl_done((union ctl_io *)ctsio);
5366         return (CTL_RETVAL_COMPLETE);
5367 }
5368
5369 int
5370 ctl_start_stop(struct ctl_scsiio *ctsio)
5371 {
5372         struct ctl_lun *lun = CTL_LUN(ctsio);
5373         struct scsi_start_stop_unit *cdb;
5374         int retval;
5375
5376         CTL_DEBUG_PRINT(("ctl_start_stop\n"));
5377
5378         cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
5379
5380         if ((cdb->how & SSS_PC_MASK) == 0) {
5381                 if ((lun->flags & CTL_LUN_PR_RESERVED) &&
5382                     (cdb->how & SSS_START) == 0) {
5383                         uint32_t residx;
5384
5385                         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5386                         if (ctl_get_prkey(lun, residx) == 0 ||
5387                             (lun->pr_res_idx != residx && lun->pr_res_type < 4)) {
5388
5389                                 ctl_set_reservation_conflict(ctsio);
5390                                 ctl_done((union ctl_io *)ctsio);
5391                                 return (CTL_RETVAL_COMPLETE);
5392                         }
5393                 }
5394
5395                 if ((cdb->how & SSS_LOEJ) &&
5396                     (lun->flags & CTL_LUN_REMOVABLE) == 0) {
5397                         ctl_set_invalid_field(ctsio,
5398                                               /*sks_valid*/ 1,
5399                                               /*command*/ 1,
5400                                               /*field*/ 4,
5401                                               /*bit_valid*/ 1,
5402                                               /*bit*/ 1);
5403                         ctl_done((union ctl_io *)ctsio);
5404                         return (CTL_RETVAL_COMPLETE);
5405                 }
5406
5407                 if ((cdb->how & SSS_START) == 0 && (cdb->how & SSS_LOEJ) &&
5408                     lun->prevent_count > 0) {
5409                         /* "Medium removal prevented" */
5410                         ctl_set_sense(ctsio, /*current_error*/ 1,
5411                             /*sense_key*/(lun->flags & CTL_LUN_NO_MEDIA) ?
5412                              SSD_KEY_NOT_READY : SSD_KEY_ILLEGAL_REQUEST,
5413                             /*asc*/ 0x53, /*ascq*/ 0x02, SSD_ELEM_NONE);
5414                         ctl_done((union ctl_io *)ctsio);
5415                         return (CTL_RETVAL_COMPLETE);
5416                 }
5417         }
5418
5419         retval = lun->backend->config_write((union ctl_io *)ctsio);
5420         return (retval);
5421 }
5422
5423 int
5424 ctl_prevent_allow(struct ctl_scsiio *ctsio)
5425 {
5426         struct ctl_lun *lun = CTL_LUN(ctsio);
5427         struct scsi_prevent *cdb;
5428         int retval;
5429         uint32_t initidx;
5430
5431         CTL_DEBUG_PRINT(("ctl_prevent_allow\n"));
5432
5433         cdb = (struct scsi_prevent *)ctsio->cdb;
5434
5435         if ((lun->flags & CTL_LUN_REMOVABLE) == 0 || lun->prevent == NULL) {
5436                 ctl_set_invalid_opcode(ctsio);
5437                 ctl_done((union ctl_io *)ctsio);
5438                 return (CTL_RETVAL_COMPLETE);
5439         }
5440
5441         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5442         mtx_lock(&lun->lun_lock);
5443         if ((cdb->how & PR_PREVENT) &&
5444             ctl_is_set(lun->prevent, initidx) == 0) {
5445                 ctl_set_mask(lun->prevent, initidx);
5446                 lun->prevent_count++;
5447         } else if ((cdb->how & PR_PREVENT) == 0 &&
5448             ctl_is_set(lun->prevent, initidx)) {
5449                 ctl_clear_mask(lun->prevent, initidx);
5450                 lun->prevent_count--;
5451         }
5452         mtx_unlock(&lun->lun_lock);
5453         retval = lun->backend->config_write((union ctl_io *)ctsio);
5454         return (retval);
5455 }
5456
5457 /*
5458  * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
5459  * we don't really do anything with the LBA and length fields if the user
5460  * passes them in.  Instead we'll just flush out the cache for the entire
5461  * LUN.
5462  */
5463 int
5464 ctl_sync_cache(struct ctl_scsiio *ctsio)
5465 {
5466         struct ctl_lun *lun = CTL_LUN(ctsio);
5467         struct ctl_lba_len_flags *lbalen;
5468         uint64_t starting_lba;
5469         uint32_t block_count;
5470         int retval;
5471         uint8_t byte2;
5472
5473         CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
5474
5475         retval = 0;
5476
5477         switch (ctsio->cdb[0]) {
5478         case SYNCHRONIZE_CACHE: {
5479                 struct scsi_sync_cache *cdb;
5480                 cdb = (struct scsi_sync_cache *)ctsio->cdb;
5481
5482                 starting_lba = scsi_4btoul(cdb->begin_lba);
5483                 block_count = scsi_2btoul(cdb->lb_count);
5484                 byte2 = cdb->byte2;
5485                 break;
5486         }
5487         case SYNCHRONIZE_CACHE_16: {
5488                 struct scsi_sync_cache_16 *cdb;
5489                 cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5490
5491                 starting_lba = scsi_8btou64(cdb->begin_lba);
5492                 block_count = scsi_4btoul(cdb->lb_count);
5493                 byte2 = cdb->byte2;
5494                 break;
5495         }
5496         default:
5497                 ctl_set_invalid_opcode(ctsio);
5498                 ctl_done((union ctl_io *)ctsio);
5499                 goto bailout;
5500                 break; /* NOTREACHED */
5501         }
5502
5503         /*
5504          * We check the LBA and length, but don't do anything with them.
5505          * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5506          * get flushed.  This check will just help satisfy anyone who wants
5507          * to see an error for an out of range LBA.
5508          */
5509         if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5510                 ctl_set_lba_out_of_range(ctsio,
5511                     MAX(starting_lba, lun->be_lun->maxlba + 1));
5512                 ctl_done((union ctl_io *)ctsio);
5513                 goto bailout;
5514         }
5515
5516         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5517         lbalen->lba = starting_lba;
5518         lbalen->len = block_count;
5519         lbalen->flags = byte2;
5520         retval = lun->backend->config_write((union ctl_io *)ctsio);
5521
5522 bailout:
5523         return (retval);
5524 }
5525
5526 int
5527 ctl_format(struct ctl_scsiio *ctsio)
5528 {
5529         struct scsi_format *cdb;
5530         int length, defect_list_len;
5531
5532         CTL_DEBUG_PRINT(("ctl_format\n"));
5533
5534         cdb = (struct scsi_format *)ctsio->cdb;
5535
5536         length = 0;
5537         if (cdb->byte2 & SF_FMTDATA) {
5538                 if (cdb->byte2 & SF_LONGLIST)
5539                         length = sizeof(struct scsi_format_header_long);
5540                 else
5541                         length = sizeof(struct scsi_format_header_short);
5542         }
5543
5544         if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5545          && (length > 0)) {
5546                 ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5547                 ctsio->kern_data_len = length;
5548                 ctsio->kern_total_len = length;
5549                 ctsio->kern_rel_offset = 0;
5550                 ctsio->kern_sg_entries = 0;
5551                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5552                 ctsio->be_move_done = ctl_config_move_done;
5553                 ctl_datamove((union ctl_io *)ctsio);
5554
5555                 return (CTL_RETVAL_COMPLETE);
5556         }
5557
5558         defect_list_len = 0;
5559
5560         if (cdb->byte2 & SF_FMTDATA) {
5561                 if (cdb->byte2 & SF_LONGLIST) {
5562                         struct scsi_format_header_long *header;
5563
5564                         header = (struct scsi_format_header_long *)
5565                                 ctsio->kern_data_ptr;
5566
5567                         defect_list_len = scsi_4btoul(header->defect_list_len);
5568                         if (defect_list_len != 0) {
5569                                 ctl_set_invalid_field(ctsio,
5570                                                       /*sks_valid*/ 1,
5571                                                       /*command*/ 0,
5572                                                       /*field*/ 2,
5573                                                       /*bit_valid*/ 0,
5574                                                       /*bit*/ 0);
5575                                 goto bailout;
5576                         }
5577                 } else {
5578                         struct scsi_format_header_short *header;
5579
5580                         header = (struct scsi_format_header_short *)
5581                                 ctsio->kern_data_ptr;
5582
5583                         defect_list_len = scsi_2btoul(header->defect_list_len);
5584                         if (defect_list_len != 0) {
5585                                 ctl_set_invalid_field(ctsio,
5586                                                       /*sks_valid*/ 1,
5587                                                       /*command*/ 0,
5588                                                       /*field*/ 2,
5589                                                       /*bit_valid*/ 0,
5590                                                       /*bit*/ 0);
5591                                 goto bailout;
5592                         }
5593                 }
5594         }
5595
5596         ctl_set_success(ctsio);
5597 bailout:
5598
5599         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5600                 free(ctsio->kern_data_ptr, M_CTL);
5601                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5602         }
5603
5604         ctl_done((union ctl_io *)ctsio);
5605         return (CTL_RETVAL_COMPLETE);
5606 }
5607
5608 int
5609 ctl_read_buffer(struct ctl_scsiio *ctsio)
5610 {
5611         struct ctl_lun *lun = CTL_LUN(ctsio);
5612         uint64_t buffer_offset;
5613         uint32_t len;
5614         uint8_t byte2;
5615         static uint8_t descr[4];
5616         static uint8_t echo_descr[4] = { 0 };
5617
5618         CTL_DEBUG_PRINT(("ctl_read_buffer\n"));
5619
5620         switch (ctsio->cdb[0]) {
5621         case READ_BUFFER: {
5622                 struct scsi_read_buffer *cdb;
5623
5624                 cdb = (struct scsi_read_buffer *)ctsio->cdb;
5625                 buffer_offset = scsi_3btoul(cdb->offset);
5626                 len = scsi_3btoul(cdb->length);
5627                 byte2 = cdb->byte2;
5628                 break;
5629         }
5630         case READ_BUFFER_16: {
5631                 struct scsi_read_buffer_16 *cdb;
5632
5633                 cdb = (struct scsi_read_buffer_16 *)ctsio->cdb;
5634                 buffer_offset = scsi_8btou64(cdb->offset);
5635                 len = scsi_4btoul(cdb->length);
5636                 byte2 = cdb->byte2;
5637                 break;
5638         }
5639         default: /* This shouldn't happen. */
5640                 ctl_set_invalid_opcode(ctsio);
5641                 ctl_done((union ctl_io *)ctsio);
5642                 return (CTL_RETVAL_COMPLETE);
5643         }
5644
5645         if (buffer_offset > CTL_WRITE_BUFFER_SIZE ||
5646             buffer_offset + len > CTL_WRITE_BUFFER_SIZE) {
5647                 ctl_set_invalid_field(ctsio,
5648                                       /*sks_valid*/ 1,
5649                                       /*command*/ 1,
5650                                       /*field*/ 6,
5651                                       /*bit_valid*/ 0,
5652                                       /*bit*/ 0);
5653                 ctl_done((union ctl_io *)ctsio);
5654                 return (CTL_RETVAL_COMPLETE);
5655         }
5656
5657         if ((byte2 & RWB_MODE) == RWB_MODE_DESCR) {
5658                 descr[0] = 0;
5659                 scsi_ulto3b(CTL_WRITE_BUFFER_SIZE, &descr[1]);
5660                 ctsio->kern_data_ptr = descr;
5661                 len = min(len, sizeof(descr));
5662         } else if ((byte2 & RWB_MODE) == RWB_MODE_ECHO_DESCR) {
5663                 ctsio->kern_data_ptr = echo_descr;
5664                 len = min(len, sizeof(echo_descr));
5665         } else {
5666                 if (lun->write_buffer == NULL) {
5667                         lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE,
5668                             M_CTL, M_WAITOK);
5669                 }
5670                 ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5671         }
5672         ctsio->kern_data_len = len;
5673         ctsio->kern_total_len = len;
5674         ctsio->kern_rel_offset = 0;
5675         ctsio->kern_sg_entries = 0;
5676         ctl_set_success(ctsio);
5677         ctsio->be_move_done = ctl_config_move_done;
5678         ctl_datamove((union ctl_io *)ctsio);
5679         return (CTL_RETVAL_COMPLETE);
5680 }
5681
5682 int
5683 ctl_write_buffer(struct ctl_scsiio *ctsio)
5684 {
5685         struct ctl_lun *lun = CTL_LUN(ctsio);
5686         struct scsi_write_buffer *cdb;
5687         int buffer_offset, len;
5688
5689         CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5690
5691         cdb = (struct scsi_write_buffer *)ctsio->cdb;
5692
5693         len = scsi_3btoul(cdb->length);
5694         buffer_offset = scsi_3btoul(cdb->offset);
5695
5696         if (buffer_offset + len > CTL_WRITE_BUFFER_SIZE) {
5697                 ctl_set_invalid_field(ctsio,
5698                                       /*sks_valid*/ 1,
5699                                       /*command*/ 1,
5700                                       /*field*/ 6,
5701                                       /*bit_valid*/ 0,
5702                                       /*bit*/ 0);
5703                 ctl_done((union ctl_io *)ctsio);
5704                 return (CTL_RETVAL_COMPLETE);
5705         }
5706
5707         /*
5708          * If we've got a kernel request that hasn't been malloced yet,
5709          * malloc it and tell the caller the data buffer is here.
5710          */
5711         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5712                 if (lun->write_buffer == NULL) {
5713                         lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE,
5714                             M_CTL, M_WAITOK);
5715                 }
5716                 ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5717                 ctsio->kern_data_len = len;
5718                 ctsio->kern_total_len = len;
5719                 ctsio->kern_rel_offset = 0;
5720                 ctsio->kern_sg_entries = 0;
5721                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5722                 ctsio->be_move_done = ctl_config_move_done;
5723                 ctl_datamove((union ctl_io *)ctsio);
5724
5725                 return (CTL_RETVAL_COMPLETE);
5726         }
5727
5728         ctl_set_success(ctsio);
5729         ctl_done((union ctl_io *)ctsio);
5730         return (CTL_RETVAL_COMPLETE);
5731 }
5732
5733 int
5734 ctl_write_same(struct ctl_scsiio *ctsio)
5735 {
5736         struct ctl_lun *lun = CTL_LUN(ctsio);
5737         struct ctl_lba_len_flags *lbalen;
5738         uint64_t lba;
5739         uint32_t num_blocks;
5740         int len, retval;
5741         uint8_t byte2;
5742
5743         CTL_DEBUG_PRINT(("ctl_write_same\n"));
5744
5745         switch (ctsio->cdb[0]) {
5746         case WRITE_SAME_10: {
5747                 struct scsi_write_same_10 *cdb;
5748
5749                 cdb = (struct scsi_write_same_10 *)ctsio->cdb;
5750
5751                 lba = scsi_4btoul(cdb->addr);
5752                 num_blocks = scsi_2btoul(cdb->length);
5753                 byte2 = cdb->byte2;
5754                 break;
5755         }
5756         case WRITE_SAME_16: {
5757                 struct scsi_write_same_16 *cdb;
5758
5759                 cdb = (struct scsi_write_same_16 *)ctsio->cdb;
5760
5761                 lba = scsi_8btou64(cdb->addr);
5762                 num_blocks = scsi_4btoul(cdb->length);
5763                 byte2 = cdb->byte2;
5764                 break;
5765         }
5766         default:
5767                 /*
5768                  * We got a command we don't support.  This shouldn't
5769                  * happen, commands should be filtered out above us.
5770                  */
5771                 ctl_set_invalid_opcode(ctsio);
5772                 ctl_done((union ctl_io *)ctsio);
5773
5774                 return (CTL_RETVAL_COMPLETE);
5775                 break; /* NOTREACHED */
5776         }
5777
5778         /* ANCHOR flag can be used only together with UNMAP */
5779         if ((byte2 & SWS_UNMAP) == 0 && (byte2 & SWS_ANCHOR) != 0) {
5780                 ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
5781                     /*command*/ 1, /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 0);
5782                 ctl_done((union ctl_io *)ctsio);
5783                 return (CTL_RETVAL_COMPLETE);
5784         }
5785
5786         /*
5787          * The first check is to make sure we're in bounds, the second
5788          * check is to catch wrap-around problems.  If the lba + num blocks
5789          * is less than the lba, then we've wrapped around and the block
5790          * range is invalid anyway.
5791          */
5792         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5793          || ((lba + num_blocks) < lba)) {
5794                 ctl_set_lba_out_of_range(ctsio,
5795                     MAX(lba, lun->be_lun->maxlba + 1));
5796                 ctl_done((union ctl_io *)ctsio);
5797                 return (CTL_RETVAL_COMPLETE);
5798         }
5799
5800         /* Zero number of blocks means "to the last logical block" */
5801         if (num_blocks == 0) {
5802                 if ((lun->be_lun->maxlba + 1) - lba > UINT32_MAX) {
5803                         ctl_set_invalid_field(ctsio,
5804                                               /*sks_valid*/ 0,
5805                                               /*command*/ 1,
5806                                               /*field*/ 0,
5807                                               /*bit_valid*/ 0,
5808                                               /*bit*/ 0);
5809                         ctl_done((union ctl_io *)ctsio);
5810                         return (CTL_RETVAL_COMPLETE);
5811                 }
5812                 num_blocks = (lun->be_lun->maxlba + 1) - lba;
5813         }
5814
5815         len = lun->be_lun->blocksize;
5816
5817         /*
5818          * If we've got a kernel request that hasn't been malloced yet,
5819          * malloc it and tell the caller the data buffer is here.
5820          */
5821         if ((byte2 & SWS_NDOB) == 0 &&
5822             (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5823                 ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);
5824                 ctsio->kern_data_len = len;
5825                 ctsio->kern_total_len = len;
5826                 ctsio->kern_rel_offset = 0;
5827                 ctsio->kern_sg_entries = 0;
5828                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5829                 ctsio->be_move_done = ctl_config_move_done;
5830                 ctl_datamove((union ctl_io *)ctsio);
5831
5832                 return (CTL_RETVAL_COMPLETE);
5833         }
5834
5835         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5836         lbalen->lba = lba;
5837         lbalen->len = num_blocks;
5838         lbalen->flags = byte2;
5839         retval = lun->backend->config_write((union ctl_io *)ctsio);
5840
5841         return (retval);
5842 }
5843
5844 int
5845 ctl_unmap(struct ctl_scsiio *ctsio)
5846 {
5847         struct ctl_lun *lun = CTL_LUN(ctsio);
5848         struct scsi_unmap *cdb;
5849         struct ctl_ptr_len_flags *ptrlen;
5850         struct scsi_unmap_header *hdr;
5851         struct scsi_unmap_desc *buf, *end, *endnz, *range;
5852         uint64_t lba;
5853         uint32_t num_blocks;
5854         int len, retval;
5855         uint8_t byte2;
5856
5857         CTL_DEBUG_PRINT(("ctl_unmap\n"));
5858
5859         cdb = (struct scsi_unmap *)ctsio->cdb;
5860         len = scsi_2btoul(cdb->length);
5861         byte2 = cdb->byte2;
5862
5863         /*
5864          * If we've got a kernel request that hasn't been malloced yet,
5865          * malloc it and tell the caller the data buffer is here.
5866          */
5867         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5868                 ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);
5869                 ctsio->kern_data_len = len;
5870                 ctsio->kern_total_len = len;
5871                 ctsio->kern_rel_offset = 0;
5872                 ctsio->kern_sg_entries = 0;
5873                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5874                 ctsio->be_move_done = ctl_config_move_done;
5875                 ctl_datamove((union ctl_io *)ctsio);
5876
5877                 return (CTL_RETVAL_COMPLETE);
5878         }
5879
5880         len = ctsio->kern_total_len - ctsio->kern_data_resid;
5881         hdr = (struct scsi_unmap_header *)ctsio->kern_data_ptr;
5882         if (len < sizeof (*hdr) ||
5883             len < (scsi_2btoul(hdr->length) + sizeof(hdr->length)) ||
5884             len < (scsi_2btoul(hdr->desc_length) + sizeof (*hdr)) ||
5885             scsi_2btoul(hdr->desc_length) % sizeof(*buf) != 0) {
5886                 ctl_set_invalid_field(ctsio,
5887                                       /*sks_valid*/ 0,
5888                                       /*command*/ 0,
5889                                       /*field*/ 0,
5890                                       /*bit_valid*/ 0,
5891                                       /*bit*/ 0);
5892                 goto done;
5893         }
5894         len = scsi_2btoul(hdr->desc_length);
5895         buf = (struct scsi_unmap_desc *)(hdr + 1);
5896         end = buf + len / sizeof(*buf);
5897
5898         endnz = buf;
5899         for (range = buf; range < end; range++) {
5900                 lba = scsi_8btou64(range->lba);
5901                 num_blocks = scsi_4btoul(range->length);
5902                 if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5903                  || ((lba + num_blocks) < lba)) {
5904                         ctl_set_lba_out_of_range(ctsio,
5905                             MAX(lba, lun->be_lun->maxlba + 1));
5906                         ctl_done((union ctl_io *)ctsio);
5907                         return (CTL_RETVAL_COMPLETE);
5908                 }
5909                 if (num_blocks != 0)
5910                         endnz = range + 1;
5911         }
5912
5913         /*
5914          * Block backend can not handle zero last range.
5915          * Filter it out and return if there is nothing left.
5916          */
5917         len = (uint8_t *)endnz - (uint8_t *)buf;
5918         if (len == 0) {
5919                 ctl_set_success(ctsio);
5920                 goto done;
5921         }
5922
5923         mtx_lock(&lun->lun_lock);
5924         ptrlen = (struct ctl_ptr_len_flags *)
5925             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5926         ptrlen->ptr = (void *)buf;
5927         ptrlen->len = len;
5928         ptrlen->flags = byte2;
5929         ctl_check_blocked(lun);
5930         mtx_unlock(&lun->lun_lock);
5931
5932         retval = lun->backend->config_write((union ctl_io *)ctsio);
5933         return (retval);
5934
5935 done:
5936         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5937                 free(ctsio->kern_data_ptr, M_CTL);
5938                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5939         }
5940         ctl_done((union ctl_io *)ctsio);
5941         return (CTL_RETVAL_COMPLETE);
5942 }
5943
5944 int
5945 ctl_default_page_handler(struct ctl_scsiio *ctsio,
5946                          struct ctl_page_index *page_index, uint8_t *page_ptr)
5947 {
5948         struct ctl_lun *lun = CTL_LUN(ctsio);
5949         uint8_t *current_cp;
5950         int set_ua;
5951         uint32_t initidx;
5952
5953         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5954         set_ua = 0;
5955
5956         current_cp = (page_index->page_data + (page_index->page_len *
5957             CTL_PAGE_CURRENT));
5958
5959         mtx_lock(&lun->lun_lock);
5960         if (memcmp(current_cp, page_ptr, page_index->page_len)) {
5961                 memcpy(current_cp, page_ptr, page_index->page_len);
5962                 set_ua = 1;
5963         }
5964         if (set_ua != 0)
5965                 ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
5966         mtx_unlock(&lun->lun_lock);
5967         if (set_ua) {
5968                 ctl_isc_announce_mode(lun,
5969                     ctl_get_initindex(&ctsio->io_hdr.nexus),
5970                     page_index->page_code, page_index->subpage);
5971         }
5972         return (CTL_RETVAL_COMPLETE);
5973 }
5974
5975 static void
5976 ctl_ie_timer(void *arg)
5977 {
5978         struct ctl_lun *lun = arg;
5979         uint64_t t;
5980
5981         if (lun->ie_asc == 0)
5982                 return;
5983
5984         if (lun->MODE_IE.mrie == SIEP_MRIE_UA)
5985                 ctl_est_ua_all(lun, -1, CTL_UA_IE);
5986         else
5987                 lun->ie_reported = 0;
5988
5989         if (lun->ie_reportcnt < scsi_4btoul(lun->MODE_IE.report_count)) {
5990                 lun->ie_reportcnt++;
5991                 t = scsi_4btoul(lun->MODE_IE.interval_timer);
5992                 if (t == 0 || t == UINT32_MAX)
5993                         t = 3000;  /* 5 min */
5994                 callout_schedule(&lun->ie_callout, t * hz / 10);
5995         }
5996 }
5997
5998 int
5999 ctl_ie_page_handler(struct ctl_scsiio *ctsio,
6000                          struct ctl_page_index *page_index, uint8_t *page_ptr)
6001 {
6002         struct ctl_lun *lun = CTL_LUN(ctsio);
6003         struct scsi_info_exceptions_page *pg;
6004         uint64_t t;
6005
6006         (void)ctl_default_page_handler(ctsio, page_index, page_ptr);
6007
6008         pg = (struct scsi_info_exceptions_page *)page_ptr;
6009         mtx_lock(&lun->lun_lock);
6010         if (pg->info_flags & SIEP_FLAGS_TEST) {
6011                 lun->ie_asc = 0x5d;
6012                 lun->ie_ascq = 0xff;
6013                 if (pg->mrie == SIEP_MRIE_UA) {
6014                         ctl_est_ua_all(lun, -1, CTL_UA_IE);
6015                         lun->ie_reported = 1;
6016                 } else {
6017                         ctl_clr_ua_all(lun, -1, CTL_UA_IE);
6018                         lun->ie_reported = -1;
6019                 }
6020                 lun->ie_reportcnt = 1;
6021                 if (lun->ie_reportcnt < scsi_4btoul(pg->report_count)) {
6022                         lun->ie_reportcnt++;
6023                         t = scsi_4btoul(pg->interval_timer);
6024                         if (t == 0 || t == UINT32_MAX)
6025                                 t = 3000;  /* 5 min */
6026                         callout_reset(&lun->ie_callout, t * hz / 10,
6027                             ctl_ie_timer, lun);
6028                 }
6029         } else {
6030                 lun->ie_asc = 0;
6031                 lun->ie_ascq = 0;
6032                 lun->ie_reported = 1;
6033                 ctl_clr_ua_all(lun, -1, CTL_UA_IE);
6034                 lun->ie_reportcnt = UINT32_MAX;
6035                 callout_stop(&lun->ie_callout);
6036         }
6037         mtx_unlock(&lun->lun_lock);
6038         return (CTL_RETVAL_COMPLETE);
6039 }
6040
6041 static int
6042 ctl_do_mode_select(union ctl_io *io)
6043 {
6044         struct ctl_lun *lun = CTL_LUN(io);
6045         struct scsi_mode_page_header *page_header;
6046         struct ctl_page_index *page_index;
6047         struct ctl_scsiio *ctsio;
6048         int page_len, page_len_offset, page_len_size;
6049         union ctl_modepage_info *modepage_info;
6050         uint16_t *len_left, *len_used;
6051         int retval, i;
6052
6053         ctsio = &io->scsiio;
6054         page_index = NULL;
6055         page_len = 0;
6056
6057         modepage_info = (union ctl_modepage_info *)
6058                 ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6059         len_left = &modepage_info->header.len_left;
6060         len_used = &modepage_info->header.len_used;
6061
6062 do_next_page:
6063
6064         page_header = (struct scsi_mode_page_header *)
6065                 (ctsio->kern_data_ptr + *len_used);
6066
6067         if (*len_left == 0) {
6068                 free(ctsio->kern_data_ptr, M_CTL);
6069                 ctl_set_success(ctsio);
6070                 ctl_done((union ctl_io *)ctsio);
6071                 return (CTL_RETVAL_COMPLETE);
6072         } else if (*len_left < sizeof(struct scsi_mode_page_header)) {
6073
6074                 free(ctsio->kern_data_ptr, M_CTL);
6075                 ctl_set_param_len_error(ctsio);
6076                 ctl_done((union ctl_io *)ctsio);
6077                 return (CTL_RETVAL_COMPLETE);
6078
6079         } else if ((page_header->page_code & SMPH_SPF)
6080                 && (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
6081
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         /*
6090          * XXX KDM should we do something with the block descriptor?
6091          */
6092         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6093                 page_index = &lun->mode_pages.index[i];
6094                 if (lun->be_lun->lun_type == T_DIRECT &&
6095                     (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6096                         continue;
6097                 if (lun->be_lun->lun_type == T_PROCESSOR &&
6098                     (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6099                         continue;
6100                 if (lun->be_lun->lun_type == T_CDROM &&
6101                     (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6102                         continue;
6103
6104                 if ((page_index->page_code & SMPH_PC_MASK) !=
6105                     (page_header->page_code & SMPH_PC_MASK))
6106                         continue;
6107
6108                 /*
6109                  * If neither page has a subpage code, then we've got a
6110                  * match.
6111                  */
6112                 if (((page_index->page_code & SMPH_SPF) == 0)
6113                  && ((page_header->page_code & SMPH_SPF) == 0)) {
6114                         page_len = page_header->page_length;
6115                         break;
6116                 }
6117
6118                 /*
6119                  * If both pages have subpages, then the subpage numbers
6120                  * have to match.
6121                  */
6122                 if ((page_index->page_code & SMPH_SPF)
6123                   && (page_header->page_code & SMPH_SPF)) {
6124                         struct scsi_mode_page_header_sp *sph;
6125
6126                         sph = (struct scsi_mode_page_header_sp *)page_header;
6127                         if (page_index->subpage == sph->subpage) {
6128                                 page_len = scsi_2btoul(sph->page_length);
6129                                 break;
6130                         }
6131                 }
6132         }
6133
6134         /*
6135          * If we couldn't find the page, or if we don't have a mode select
6136          * handler for it, send back an error to the user.
6137          */
6138         if ((i >= CTL_NUM_MODE_PAGES)
6139          || (page_index->select_handler == NULL)) {
6140                 ctl_set_invalid_field(ctsio,
6141                                       /*sks_valid*/ 1,
6142                                       /*command*/ 0,
6143                                       /*field*/ *len_used,
6144                                       /*bit_valid*/ 0,
6145                                       /*bit*/ 0);
6146                 free(ctsio->kern_data_ptr, M_CTL);
6147                 ctl_done((union ctl_io *)ctsio);
6148                 return (CTL_RETVAL_COMPLETE);
6149         }
6150
6151         if (page_index->page_code & SMPH_SPF) {
6152                 page_len_offset = 2;
6153                 page_len_size = 2;
6154         } else {
6155                 page_len_size = 1;
6156                 page_len_offset = 1;
6157         }
6158
6159         /*
6160          * If the length the initiator gives us isn't the one we specify in
6161          * the mode page header, or if they didn't specify enough data in
6162          * the CDB to avoid truncating this page, kick out the request.
6163          */
6164         if (page_len != page_index->page_len - page_len_offset - page_len_size) {
6165                 ctl_set_invalid_field(ctsio,
6166                                       /*sks_valid*/ 1,
6167                                       /*command*/ 0,
6168                                       /*field*/ *len_used + page_len_offset,
6169                                       /*bit_valid*/ 0,
6170                                       /*bit*/ 0);
6171                 free(ctsio->kern_data_ptr, M_CTL);
6172                 ctl_done((union ctl_io *)ctsio);
6173                 return (CTL_RETVAL_COMPLETE);
6174         }
6175         if (*len_left < page_index->page_len) {
6176                 free(ctsio->kern_data_ptr, M_CTL);
6177                 ctl_set_param_len_error(ctsio);
6178                 ctl_done((union ctl_io *)ctsio);
6179                 return (CTL_RETVAL_COMPLETE);
6180         }
6181
6182         /*
6183          * Run through the mode page, checking to make sure that the bits
6184          * the user changed are actually legal for him to change.
6185          */
6186         for (i = 0; i < page_index->page_len; i++) {
6187                 uint8_t *user_byte, *change_mask, *current_byte;
6188                 int bad_bit;
6189                 int j;
6190
6191                 user_byte = (uint8_t *)page_header + i;
6192                 change_mask = page_index->page_data +
6193                               (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
6194                 current_byte = page_index->page_data +
6195                                (page_index->page_len * CTL_PAGE_CURRENT) + i;
6196
6197                 /*
6198                  * Check to see whether the user set any bits in this byte
6199                  * that he is not allowed to set.
6200                  */
6201                 if ((*user_byte & ~(*change_mask)) ==
6202                     (*current_byte & ~(*change_mask)))
6203                         continue;
6204
6205                 /*
6206                  * Go through bit by bit to determine which one is illegal.
6207                  */
6208                 bad_bit = 0;
6209                 for (j = 7; j >= 0; j--) {
6210                         if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
6211                             (((1 << i) & ~(*change_mask)) & *current_byte)) {
6212                                 bad_bit = i;
6213                                 break;
6214                         }
6215                 }
6216                 ctl_set_invalid_field(ctsio,
6217                                       /*sks_valid*/ 1,
6218                                       /*command*/ 0,
6219                                       /*field*/ *len_used + i,
6220                                       /*bit_valid*/ 1,
6221                                       /*bit*/ bad_bit);
6222                 free(ctsio->kern_data_ptr, M_CTL);
6223                 ctl_done((union ctl_io *)ctsio);
6224                 return (CTL_RETVAL_COMPLETE);
6225         }
6226
6227         /*
6228          * Decrement these before we call the page handler, since we may
6229          * end up getting called back one way or another before the handler
6230          * returns to this context.
6231          */
6232         *len_left -= page_index->page_len;
6233         *len_used += page_index->page_len;
6234
6235         retval = page_index->select_handler(ctsio, page_index,
6236                                             (uint8_t *)page_header);
6237
6238         /*
6239          * If the page handler returns CTL_RETVAL_QUEUED, then we need to
6240          * wait until this queued command completes to finish processing
6241          * the mode page.  If it returns anything other than
6242          * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
6243          * already set the sense information, freed the data pointer, and
6244          * completed the io for us.
6245          */
6246         if (retval != CTL_RETVAL_COMPLETE)
6247                 goto bailout_no_done;
6248
6249         /*
6250          * If the initiator sent us more than one page, parse the next one.
6251          */
6252         if (*len_left > 0)
6253                 goto do_next_page;
6254
6255         ctl_set_success(ctsio);
6256         free(ctsio->kern_data_ptr, M_CTL);
6257         ctl_done((union ctl_io *)ctsio);
6258
6259 bailout_no_done:
6260
6261         return (CTL_RETVAL_COMPLETE);
6262
6263 }
6264
6265 int
6266 ctl_mode_select(struct ctl_scsiio *ctsio)
6267 {
6268         struct ctl_lun *lun = CTL_LUN(ctsio);
6269         union ctl_modepage_info *modepage_info;
6270         int bd_len, i, header_size, param_len, rtd;
6271         uint32_t initidx;
6272
6273         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
6274         switch (ctsio->cdb[0]) {
6275         case MODE_SELECT_6: {
6276                 struct scsi_mode_select_6 *cdb;
6277
6278                 cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
6279
6280                 rtd = (cdb->byte2 & SMS_RTD) ? 1 : 0;
6281                 param_len = cdb->length;
6282                 header_size = sizeof(struct scsi_mode_header_6);
6283                 break;
6284         }
6285         case MODE_SELECT_10: {
6286                 struct scsi_mode_select_10 *cdb;
6287
6288                 cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
6289
6290                 rtd = (cdb->byte2 & SMS_RTD) ? 1 : 0;
6291                 param_len = scsi_2btoul(cdb->length);
6292                 header_size = sizeof(struct scsi_mode_header_10);
6293                 break;
6294         }
6295         default:
6296                 ctl_set_invalid_opcode(ctsio);
6297                 ctl_done((union ctl_io *)ctsio);
6298                 return (CTL_RETVAL_COMPLETE);
6299         }
6300
6301         if (rtd) {
6302                 if (param_len != 0) {
6303                         ctl_set_invalid_field(ctsio, /*sks_valid*/ 0,
6304                             /*command*/ 1, /*field*/ 0,
6305                             /*bit_valid*/ 0, /*bit*/ 0);
6306                         ctl_done((union ctl_io *)ctsio);
6307                         return (CTL_RETVAL_COMPLETE);
6308                 }
6309
6310                 /* Revert to defaults. */
6311                 ctl_init_page_index(lun);
6312                 mtx_lock(&lun->lun_lock);
6313                 ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
6314                 mtx_unlock(&lun->lun_lock);
6315                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6316                         ctl_isc_announce_mode(lun, -1,
6317                             lun->mode_pages.index[i].page_code & SMPH_PC_MASK,
6318                             lun->mode_pages.index[i].subpage);
6319                 }
6320                 ctl_set_success(ctsio);
6321                 ctl_done((union ctl_io *)ctsio);
6322                 return (CTL_RETVAL_COMPLETE);
6323         }
6324
6325         /*
6326          * From SPC-3:
6327          * "A parameter list length of zero indicates that the Data-Out Buffer
6328          * shall be empty. This condition shall not be considered as an error."
6329          */
6330         if (param_len == 0) {
6331                 ctl_set_success(ctsio);
6332                 ctl_done((union ctl_io *)ctsio);
6333                 return (CTL_RETVAL_COMPLETE);
6334         }
6335
6336         /*
6337          * Since we'll hit this the first time through, prior to
6338          * allocation, we don't need to free a data buffer here.
6339          */
6340         if (param_len < header_size) {
6341                 ctl_set_param_len_error(ctsio);
6342                 ctl_done((union ctl_io *)ctsio);
6343                 return (CTL_RETVAL_COMPLETE);
6344         }
6345
6346         /*
6347          * Allocate the data buffer and grab the user's data.  In theory,
6348          * we shouldn't have to sanity check the parameter list length here
6349          * because the maximum size is 64K.  We should be able to malloc
6350          * that much without too many problems.
6351          */
6352         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6353                 ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6354                 ctsio->kern_data_len = param_len;
6355                 ctsio->kern_total_len = param_len;
6356                 ctsio->kern_rel_offset = 0;
6357                 ctsio->kern_sg_entries = 0;
6358                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6359                 ctsio->be_move_done = ctl_config_move_done;
6360                 ctl_datamove((union ctl_io *)ctsio);
6361
6362                 return (CTL_RETVAL_COMPLETE);
6363         }
6364
6365         switch (ctsio->cdb[0]) {
6366         case MODE_SELECT_6: {
6367                 struct scsi_mode_header_6 *mh6;
6368
6369                 mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6370                 bd_len = mh6->blk_desc_len;
6371                 break;
6372         }
6373         case MODE_SELECT_10: {
6374                 struct scsi_mode_header_10 *mh10;
6375
6376                 mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6377                 bd_len = scsi_2btoul(mh10->blk_desc_len);
6378                 break;
6379         }
6380         default:
6381                 panic("%s: Invalid CDB type %#x", __func__, ctsio->cdb[0]);
6382         }
6383
6384         if (param_len < (header_size + bd_len)) {
6385                 free(ctsio->kern_data_ptr, M_CTL);
6386                 ctl_set_param_len_error(ctsio);
6387                 ctl_done((union ctl_io *)ctsio);
6388                 return (CTL_RETVAL_COMPLETE);
6389         }
6390
6391         /*
6392          * Set the IO_CONT flag, so that if this I/O gets passed to
6393          * ctl_config_write_done(), it'll get passed back to
6394          * ctl_do_mode_select() for further processing, or completion if
6395          * we're all done.
6396          */
6397         ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6398         ctsio->io_cont = ctl_do_mode_select;
6399
6400         modepage_info = (union ctl_modepage_info *)
6401                 ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6402         memset(modepage_info, 0, sizeof(*modepage_info));
6403         modepage_info->header.len_left = param_len - header_size - bd_len;
6404         modepage_info->header.len_used = header_size + bd_len;
6405
6406         return (ctl_do_mode_select((union ctl_io *)ctsio));
6407 }
6408
6409 int
6410 ctl_mode_sense(struct ctl_scsiio *ctsio)
6411 {
6412         struct ctl_lun *lun = CTL_LUN(ctsio);
6413         int pc, page_code, dbd, subpage;
6414         int alloc_len, page_len, header_len, total_len;
6415         struct scsi_mode_block_descr *block_desc;
6416         struct ctl_page_index *page_index;
6417
6418         dbd = 0;
6419         block_desc = NULL;
6420
6421         CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6422
6423         switch (ctsio->cdb[0]) {
6424         case MODE_SENSE_6: {
6425                 struct scsi_mode_sense_6 *cdb;
6426
6427                 cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6428
6429                 header_len = sizeof(struct scsi_mode_hdr_6);
6430                 if (cdb->byte2 & SMS_DBD)
6431                         dbd = 1;
6432                 else
6433                         header_len += sizeof(struct scsi_mode_block_descr);
6434
6435                 pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6436                 page_code = cdb->page & SMS_PAGE_CODE;
6437                 subpage = cdb->subpage;
6438                 alloc_len = cdb->length;
6439                 break;
6440         }
6441         case MODE_SENSE_10: {
6442                 struct scsi_mode_sense_10 *cdb;
6443
6444                 cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6445
6446                 header_len = sizeof(struct scsi_mode_hdr_10);
6447
6448                 if (cdb->byte2 & SMS_DBD)
6449                         dbd = 1;
6450                 else
6451                         header_len += sizeof(struct scsi_mode_block_descr);
6452                 pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6453                 page_code = cdb->page & SMS_PAGE_CODE;
6454                 subpage = cdb->subpage;
6455                 alloc_len = scsi_2btoul(cdb->length);
6456                 break;
6457         }
6458         default:
6459                 ctl_set_invalid_opcode(ctsio);
6460                 ctl_done((union ctl_io *)ctsio);
6461                 return (CTL_RETVAL_COMPLETE);
6462                 break; /* NOTREACHED */
6463         }
6464
6465         /*
6466          * We have to make a first pass through to calculate the size of
6467          * the pages that match the user's query.  Then we allocate enough
6468          * memory to hold it, and actually copy the data into the buffer.
6469          */
6470         switch (page_code) {
6471         case SMS_ALL_PAGES_PAGE: {
6472                 u_int i;
6473
6474                 page_len = 0;
6475
6476                 /*
6477                  * At the moment, values other than 0 and 0xff here are
6478                  * reserved according to SPC-3.
6479                  */
6480                 if ((subpage != SMS_SUBPAGE_PAGE_0)
6481                  && (subpage != SMS_SUBPAGE_ALL)) {
6482                         ctl_set_invalid_field(ctsio,
6483                                               /*sks_valid*/ 1,
6484                                               /*command*/ 1,
6485                                               /*field*/ 3,
6486                                               /*bit_valid*/ 0,
6487                                               /*bit*/ 0);
6488                         ctl_done((union ctl_io *)ctsio);
6489                         return (CTL_RETVAL_COMPLETE);
6490                 }
6491
6492                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6493                         page_index = &lun->mode_pages.index[i];
6494
6495                         /* Make sure the page is supported for this dev type */
6496                         if (lun->be_lun->lun_type == T_DIRECT &&
6497                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6498                                 continue;
6499                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6500                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6501                                 continue;
6502                         if (lun->be_lun->lun_type == T_CDROM &&
6503                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6504                                 continue;
6505
6506                         /*
6507                          * We don't use this subpage if the user didn't
6508                          * request all subpages.
6509                          */
6510                         if ((page_index->subpage != 0)
6511                          && (subpage == SMS_SUBPAGE_PAGE_0))
6512                                 continue;
6513
6514 #if 0
6515                         printf("found page %#x len %d\n",
6516                                page_index->page_code & SMPH_PC_MASK,
6517                                page_index->page_len);
6518 #endif
6519                         page_len += page_index->page_len;
6520                 }
6521                 break;
6522         }
6523         default: {
6524                 u_int i;
6525
6526                 page_len = 0;
6527
6528                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6529                         page_index = &lun->mode_pages.index[i];
6530
6531                         /* Make sure the page is supported for this dev type */
6532                         if (lun->be_lun->lun_type == T_DIRECT &&
6533                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6534                                 continue;
6535                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6536                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6537                                 continue;
6538                         if (lun->be_lun->lun_type == T_CDROM &&
6539                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6540                                 continue;
6541
6542                         /* Look for the right page code */
6543                         if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6544                                 continue;
6545
6546                         /* Look for the right subpage or the subpage wildcard*/
6547                         if ((page_index->subpage != subpage)
6548                          && (subpage != SMS_SUBPAGE_ALL))
6549                                 continue;
6550
6551 #if 0
6552                         printf("found page %#x len %d\n",
6553                                page_index->page_code & SMPH_PC_MASK,
6554                                page_index->page_len);
6555 #endif
6556
6557                         page_len += page_index->page_len;
6558                 }
6559
6560                 if (page_len == 0) {
6561                         ctl_set_invalid_field(ctsio,
6562                                               /*sks_valid*/ 1,
6563                                               /*command*/ 1,
6564                                               /*field*/ 2,
6565                                               /*bit_valid*/ 1,
6566                                               /*bit*/ 5);
6567                         ctl_done((union ctl_io *)ctsio);
6568                         return (CTL_RETVAL_COMPLETE);
6569                 }
6570                 break;
6571         }
6572         }
6573
6574         total_len = header_len + page_len;
6575 #if 0
6576         printf("header_len = %d, page_len = %d, total_len = %d\n",
6577                header_len, page_len, total_len);
6578 #endif
6579
6580         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6581         ctsio->kern_sg_entries = 0;
6582         ctsio->kern_rel_offset = 0;
6583         ctsio->kern_data_len = min(total_len, alloc_len);
6584         ctsio->kern_total_len = ctsio->kern_data_len;
6585
6586         switch (ctsio->cdb[0]) {
6587         case MODE_SENSE_6: {
6588                 struct scsi_mode_hdr_6 *header;
6589
6590                 header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
6591
6592                 header->datalen = MIN(total_len - 1, 254);
6593                 if (lun->be_lun->lun_type == T_DIRECT) {
6594                         header->dev_specific = 0x10; /* DPOFUA */
6595                         if ((lun->be_lun->flags & CTL_LUN_FLAG_READONLY) ||
6596                             (lun->MODE_CTRL.eca_and_aen & SCP_SWP) != 0)
6597                                 header->dev_specific |= 0x80; /* WP */
6598                 }
6599                 if (dbd)
6600                         header->block_descr_len = 0;
6601                 else
6602                         header->block_descr_len =
6603                                 sizeof(struct scsi_mode_block_descr);
6604                 block_desc = (struct scsi_mode_block_descr *)&header[1];
6605                 break;
6606         }
6607         case MODE_SENSE_10: {
6608                 struct scsi_mode_hdr_10 *header;
6609                 int datalen;
6610
6611                 header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
6612
6613                 datalen = MIN(total_len - 2, 65533);
6614                 scsi_ulto2b(datalen, header->datalen);
6615                 if (lun->be_lun->lun_type == T_DIRECT) {
6616                         header->dev_specific = 0x10; /* DPOFUA */
6617                         if ((lun->be_lun->flags & CTL_LUN_FLAG_READONLY) ||
6618                             (lun->MODE_CTRL.eca_and_aen & SCP_SWP) != 0)
6619                                 header->dev_specific |= 0x80; /* WP */
6620                 }
6621                 if (dbd)
6622                         scsi_ulto2b(0, header->block_descr_len);
6623                 else
6624                         scsi_ulto2b(sizeof(struct scsi_mode_block_descr),
6625                                     header->block_descr_len);
6626                 block_desc = (struct scsi_mode_block_descr *)&header[1];
6627                 break;
6628         }
6629         default:
6630                 panic("%s: Invalid CDB type %#x", __func__, ctsio->cdb[0]);
6631         }
6632
6633         /*
6634          * If we've got a disk, use its blocksize in the block
6635          * descriptor.  Otherwise, just set it to 0.
6636          */
6637         if (dbd == 0) {
6638                 if (lun->be_lun->lun_type == T_DIRECT)
6639                         scsi_ulto3b(lun->be_lun->blocksize,
6640                                     block_desc->block_len);
6641                 else
6642                         scsi_ulto3b(0, block_desc->block_len);
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_lbp_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_param_header *phdr;
6750         uint8_t *data;
6751         uint64_t val;
6752
6753         data = page_index->page_data;
6754
6755         if (lun->backend->lun_attr != NULL &&
6756             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "blocksavail"))
6757              != UINT64_MAX) {
6758                 phdr = (struct scsi_log_param_header *)data;
6759                 scsi_ulto2b(0x0001, phdr->param_code);
6760                 phdr->param_control = SLP_LBIN | SLP_LP;
6761                 phdr->param_len = 8;
6762                 data = (uint8_t *)(phdr + 1);
6763                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6764                 data[4] = 0x02; /* per-pool */
6765                 data += phdr->param_len;
6766         }
6767
6768         if (lun->backend->lun_attr != NULL &&
6769             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "blocksused"))
6770              != UINT64_MAX) {
6771                 phdr = (struct scsi_log_param_header *)data;
6772                 scsi_ulto2b(0x0002, phdr->param_code);
6773                 phdr->param_control = SLP_LBIN | SLP_LP;
6774                 phdr->param_len = 8;
6775                 data = (uint8_t *)(phdr + 1);
6776                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6777                 data[4] = 0x01; /* per-LUN */
6778                 data += phdr->param_len;
6779         }
6780
6781         if (lun->backend->lun_attr != NULL &&
6782             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "poolblocksavail"))
6783              != UINT64_MAX) {
6784                 phdr = (struct scsi_log_param_header *)data;
6785                 scsi_ulto2b(0x00f1, phdr->param_code);
6786                 phdr->param_control = SLP_LBIN | SLP_LP;
6787                 phdr->param_len = 8;
6788                 data = (uint8_t *)(phdr + 1);
6789                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6790                 data[4] = 0x02; /* per-pool */
6791                 data += phdr->param_len;
6792         }
6793
6794         if (lun->backend->lun_attr != NULL &&
6795             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "poolblocksused"))
6796              != UINT64_MAX) {
6797                 phdr = (struct scsi_log_param_header *)data;
6798                 scsi_ulto2b(0x00f2, phdr->param_code);
6799                 phdr->param_control = SLP_LBIN | SLP_LP;
6800                 phdr->param_len = 8;
6801                 data = (uint8_t *)(phdr + 1);
6802                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6803                 data[4] = 0x02; /* per-pool */
6804                 data += phdr->param_len;
6805         }
6806
6807         page_index->page_len = data - page_index->page_data;
6808         return (0);
6809 }
6810
6811 int
6812 ctl_sap_log_sense_handler(struct ctl_scsiio *ctsio,
6813                                struct ctl_page_index *page_index,
6814                                int pc)
6815 {
6816         struct ctl_lun *lun = CTL_LUN(ctsio);
6817         struct stat_page *data;
6818         struct bintime *t;
6819
6820         data = (struct stat_page *)page_index->page_data;
6821
6822         scsi_ulto2b(SLP_SAP, data->sap.hdr.param_code);
6823         data->sap.hdr.param_control = SLP_LBIN;
6824         data->sap.hdr.param_len = sizeof(struct scsi_log_stat_and_perf) -
6825             sizeof(struct scsi_log_param_header);
6826         scsi_u64to8b(lun->stats.operations[CTL_STATS_READ],
6827             data->sap.read_num);
6828         scsi_u64to8b(lun->stats.operations[CTL_STATS_WRITE],
6829             data->sap.write_num);
6830         if (lun->be_lun->blocksize > 0) {
6831                 scsi_u64to8b(lun->stats.bytes[CTL_STATS_WRITE] /
6832                     lun->be_lun->blocksize, data->sap.recvieved_lba);
6833                 scsi_u64to8b(lun->stats.bytes[CTL_STATS_READ] /
6834                     lun->be_lun->blocksize, data->sap.transmitted_lba);
6835         }
6836         t = &lun->stats.time[CTL_STATS_READ];
6837         scsi_u64to8b((uint64_t)t->sec * 1000 + t->frac / (UINT64_MAX / 1000),
6838             data->sap.read_int);
6839         t = &lun->stats.time[CTL_STATS_WRITE];
6840         scsi_u64to8b((uint64_t)t->sec * 1000 + t->frac / (UINT64_MAX / 1000),
6841             data->sap.write_int);
6842         scsi_u64to8b(0, data->sap.weighted_num);
6843         scsi_u64to8b(0, data->sap.weighted_int);
6844         scsi_ulto2b(SLP_IT, data->it.hdr.param_code);
6845         data->it.hdr.param_control = SLP_LBIN;
6846         data->it.hdr.param_len = sizeof(struct scsi_log_idle_time) -
6847             sizeof(struct scsi_log_param_header);
6848 #ifdef CTL_TIME_IO
6849         scsi_u64to8b(lun->idle_time / SBT_1MS, data->it.idle_int);
6850 #endif
6851         scsi_ulto2b(SLP_TI, data->ti.hdr.param_code);
6852         data->it.hdr.param_control = SLP_LBIN;
6853         data->ti.hdr.param_len = sizeof(struct scsi_log_time_interval) -
6854             sizeof(struct scsi_log_param_header);
6855         scsi_ulto4b(3, data->ti.exponent);
6856         scsi_ulto4b(1, data->ti.integer);
6857         return (0);
6858 }
6859
6860 int
6861 ctl_ie_log_sense_handler(struct ctl_scsiio *ctsio,
6862                                struct ctl_page_index *page_index,
6863                                int pc)
6864 {
6865         struct ctl_lun *lun = CTL_LUN(ctsio);
6866         struct scsi_log_informational_exceptions *data;
6867
6868         data = (struct scsi_log_informational_exceptions *)page_index->page_data;
6869
6870         scsi_ulto2b(SLP_IE_GEN, data->hdr.param_code);
6871         data->hdr.param_control = SLP_LBIN;
6872         data->hdr.param_len = sizeof(struct scsi_log_informational_exceptions) -
6873             sizeof(struct scsi_log_param_header);
6874         data->ie_asc = lun->ie_asc;
6875         data->ie_ascq = lun->ie_ascq;
6876         data->temperature = 0xff;
6877         return (0);
6878 }
6879
6880 int
6881 ctl_log_sense(struct ctl_scsiio *ctsio)
6882 {
6883         struct ctl_lun *lun = CTL_LUN(ctsio);
6884         int i, pc, page_code, subpage;
6885         int alloc_len, total_len;
6886         struct ctl_page_index *page_index;
6887         struct scsi_log_sense *cdb;
6888         struct scsi_log_header *header;
6889
6890         CTL_DEBUG_PRINT(("ctl_log_sense\n"));
6891
6892         cdb = (struct scsi_log_sense *)ctsio->cdb;
6893         pc = (cdb->page & SLS_PAGE_CTRL_MASK) >> 6;
6894         page_code = cdb->page & SLS_PAGE_CODE;
6895         subpage = cdb->subpage;
6896         alloc_len = scsi_2btoul(cdb->length);
6897
6898         page_index = NULL;
6899         for (i = 0; i < CTL_NUM_LOG_PAGES; i++) {
6900                 page_index = &lun->log_pages.index[i];
6901
6902                 /* Look for the right page code */
6903                 if ((page_index->page_code & SL_PAGE_CODE) != page_code)
6904                         continue;
6905
6906                 /* Look for the right subpage or the subpage wildcard*/
6907                 if (page_index->subpage != subpage)
6908                         continue;
6909
6910                 break;
6911         }
6912         if (i >= CTL_NUM_LOG_PAGES) {
6913                 ctl_set_invalid_field(ctsio,
6914                                       /*sks_valid*/ 1,
6915                                       /*command*/ 1,
6916                                       /*field*/ 2,
6917                                       /*bit_valid*/ 0,
6918                                       /*bit*/ 0);
6919                 ctl_done((union ctl_io *)ctsio);
6920                 return (CTL_RETVAL_COMPLETE);
6921         }
6922
6923         total_len = sizeof(struct scsi_log_header) + page_index->page_len;
6924
6925         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6926         ctsio->kern_sg_entries = 0;
6927         ctsio->kern_rel_offset = 0;
6928         ctsio->kern_data_len = min(total_len, alloc_len);
6929         ctsio->kern_total_len = ctsio->kern_data_len;
6930
6931         header = (struct scsi_log_header *)ctsio->kern_data_ptr;
6932         header->page = page_index->page_code;
6933         if (page_index->page_code == SLS_LOGICAL_BLOCK_PROVISIONING)
6934                 header->page |= SL_DS;
6935         if (page_index->subpage) {
6936                 header->page |= SL_SPF;
6937                 header->subpage = page_index->subpage;
6938         }
6939         scsi_ulto2b(page_index->page_len, header->datalen);
6940
6941         /*
6942          * Call the handler, if it exists, to update the
6943          * page to the latest values.
6944          */
6945         if (page_index->sense_handler != NULL)
6946                 page_index->sense_handler(ctsio, page_index, pc);
6947
6948         memcpy(header + 1, page_index->page_data, page_index->page_len);
6949
6950         ctl_set_success(ctsio);
6951         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6952         ctsio->be_move_done = ctl_config_move_done;
6953         ctl_datamove((union ctl_io *)ctsio);
6954         return (CTL_RETVAL_COMPLETE);
6955 }
6956
6957 int
6958 ctl_read_capacity(struct ctl_scsiio *ctsio)
6959 {
6960         struct ctl_lun *lun = CTL_LUN(ctsio);
6961         struct scsi_read_capacity *cdb;
6962         struct scsi_read_capacity_data *data;
6963         uint32_t lba;
6964
6965         CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
6966
6967         cdb = (struct scsi_read_capacity *)ctsio->cdb;
6968
6969         lba = scsi_4btoul(cdb->addr);
6970         if (((cdb->pmi & SRC_PMI) == 0)
6971          && (lba != 0)) {
6972                 ctl_set_invalid_field(/*ctsio*/ ctsio,
6973                                       /*sks_valid*/ 1,
6974                                       /*command*/ 1,
6975                                       /*field*/ 2,
6976                                       /*bit_valid*/ 0,
6977                                       /*bit*/ 0);
6978                 ctl_done((union ctl_io *)ctsio);
6979                 return (CTL_RETVAL_COMPLETE);
6980         }
6981
6982         ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
6983         data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
6984         ctsio->kern_data_len = sizeof(*data);
6985         ctsio->kern_total_len = sizeof(*data);
6986         ctsio->kern_rel_offset = 0;
6987         ctsio->kern_sg_entries = 0;
6988
6989         /*
6990          * If the maximum LBA is greater than 0xfffffffe, the user must
6991          * issue a SERVICE ACTION IN (16) command, with the read capacity
6992          * serivce action set.
6993          */
6994         if (lun->be_lun->maxlba > 0xfffffffe)
6995                 scsi_ulto4b(0xffffffff, data->addr);
6996         else
6997                 scsi_ulto4b(lun->be_lun->maxlba, data->addr);
6998
6999         /*
7000          * XXX KDM this may not be 512 bytes...
7001          */
7002         scsi_ulto4b(lun->be_lun->blocksize, data->length);
7003
7004         ctl_set_success(ctsio);
7005         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7006         ctsio->be_move_done = ctl_config_move_done;
7007         ctl_datamove((union ctl_io *)ctsio);
7008         return (CTL_RETVAL_COMPLETE);
7009 }
7010
7011 int
7012 ctl_read_capacity_16(struct ctl_scsiio *ctsio)
7013 {
7014         struct ctl_lun *lun = CTL_LUN(ctsio);
7015         struct scsi_read_capacity_16 *cdb;
7016         struct scsi_read_capacity_data_long *data;
7017         uint64_t lba;
7018         uint32_t alloc_len;
7019
7020         CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
7021
7022         cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
7023
7024         alloc_len = scsi_4btoul(cdb->alloc_len);
7025         lba = scsi_8btou64(cdb->addr);
7026
7027         if ((cdb->reladr & SRC16_PMI)
7028          && (lba != 0)) {
7029                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7030                                       /*sks_valid*/ 1,
7031                                       /*command*/ 1,
7032                                       /*field*/ 2,
7033                                       /*bit_valid*/ 0,
7034                                       /*bit*/ 0);
7035                 ctl_done((union ctl_io *)ctsio);
7036                 return (CTL_RETVAL_COMPLETE);
7037         }
7038
7039         ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7040         data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
7041         ctsio->kern_rel_offset = 0;
7042         ctsio->kern_sg_entries = 0;
7043         ctsio->kern_data_len = min(sizeof(*data), alloc_len);
7044         ctsio->kern_total_len = ctsio->kern_data_len;
7045
7046         scsi_u64to8b(lun->be_lun->maxlba, data->addr);
7047         /* XXX KDM this may not be 512 bytes... */
7048         scsi_ulto4b(lun->be_lun->blocksize, data->length);
7049         data->prot_lbppbe = lun->be_lun->pblockexp & SRC16_LBPPBE;
7050         scsi_ulto2b(lun->be_lun->pblockoff & SRC16_LALBA_A, data->lalba_lbp);
7051         if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
7052                 data->lalba_lbp[0] |= SRC16_LBPME | SRC16_LBPRZ;
7053
7054         ctl_set_success(ctsio);
7055         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7056         ctsio->be_move_done = ctl_config_move_done;
7057         ctl_datamove((union ctl_io *)ctsio);
7058         return (CTL_RETVAL_COMPLETE);
7059 }
7060
7061 int
7062 ctl_get_lba_status(struct ctl_scsiio *ctsio)
7063 {
7064         struct ctl_lun *lun = CTL_LUN(ctsio);
7065         struct scsi_get_lba_status *cdb;
7066         struct scsi_get_lba_status_data *data;
7067         struct ctl_lba_len_flags *lbalen;
7068         uint64_t lba;
7069         uint32_t alloc_len, total_len;
7070         int retval;
7071
7072         CTL_DEBUG_PRINT(("ctl_get_lba_status\n"));
7073
7074         cdb = (struct scsi_get_lba_status *)ctsio->cdb;
7075         lba = scsi_8btou64(cdb->addr);
7076         alloc_len = scsi_4btoul(cdb->alloc_len);
7077
7078         if (lba > lun->be_lun->maxlba) {
7079                 ctl_set_lba_out_of_range(ctsio, lba);
7080                 ctl_done((union ctl_io *)ctsio);
7081                 return (CTL_RETVAL_COMPLETE);
7082         }
7083
7084         total_len = sizeof(*data) + sizeof(data->descr[0]);
7085         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7086         data = (struct scsi_get_lba_status_data *)ctsio->kern_data_ptr;
7087         ctsio->kern_rel_offset = 0;
7088         ctsio->kern_sg_entries = 0;
7089         ctsio->kern_data_len = min(total_len, alloc_len);
7090         ctsio->kern_total_len = ctsio->kern_data_len;
7091
7092         /* Fill dummy data in case backend can't tell anything. */
7093         scsi_ulto4b(4 + sizeof(data->descr[0]), data->length);
7094         scsi_u64to8b(lba, data->descr[0].addr);
7095         scsi_ulto4b(MIN(UINT32_MAX, lun->be_lun->maxlba + 1 - lba),
7096             data->descr[0].length);
7097         data->descr[0].status = 0; /* Mapped or unknown. */
7098
7099         ctl_set_success(ctsio);
7100         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7101         ctsio->be_move_done = ctl_config_move_done;
7102
7103         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
7104         lbalen->lba = lba;
7105         lbalen->len = total_len;
7106         lbalen->flags = 0;
7107         retval = lun->backend->config_read((union ctl_io *)ctsio);
7108         return (retval);
7109 }
7110
7111 int
7112 ctl_read_defect(struct ctl_scsiio *ctsio)
7113 {
7114         struct scsi_read_defect_data_10 *ccb10;
7115         struct scsi_read_defect_data_12 *ccb12;
7116         struct scsi_read_defect_data_hdr_10 *data10;
7117         struct scsi_read_defect_data_hdr_12 *data12;
7118         uint32_t alloc_len, data_len;
7119         uint8_t format;
7120
7121         CTL_DEBUG_PRINT(("ctl_read_defect\n"));
7122
7123         if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
7124                 ccb10 = (struct scsi_read_defect_data_10 *)&ctsio->cdb;
7125                 format = ccb10->format;
7126                 alloc_len = scsi_2btoul(ccb10->alloc_length);
7127                 data_len = sizeof(*data10);
7128         } else {
7129                 ccb12 = (struct scsi_read_defect_data_12 *)&ctsio->cdb;
7130                 format = ccb12->format;
7131                 alloc_len = scsi_4btoul(ccb12->alloc_length);
7132                 data_len = sizeof(*data12);
7133         }
7134         if (alloc_len == 0) {
7135                 ctl_set_success(ctsio);
7136                 ctl_done((union ctl_io *)ctsio);
7137                 return (CTL_RETVAL_COMPLETE);
7138         }
7139
7140         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
7141         ctsio->kern_rel_offset = 0;
7142         ctsio->kern_sg_entries = 0;
7143         ctsio->kern_data_len = min(data_len, alloc_len);
7144         ctsio->kern_total_len = ctsio->kern_data_len;
7145
7146         if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
7147                 data10 = (struct scsi_read_defect_data_hdr_10 *)
7148                     ctsio->kern_data_ptr;
7149                 data10->format = format;
7150                 scsi_ulto2b(0, data10->length);
7151         } else {
7152                 data12 = (struct scsi_read_defect_data_hdr_12 *)
7153                     ctsio->kern_data_ptr;
7154                 data12->format = format;
7155                 scsi_ulto2b(0, data12->generation);
7156                 scsi_ulto4b(0, data12->length);
7157         }
7158
7159         ctl_set_success(ctsio);
7160         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7161         ctsio->be_move_done = ctl_config_move_done;
7162         ctl_datamove((union ctl_io *)ctsio);
7163         return (CTL_RETVAL_COMPLETE);
7164 }
7165
7166 int
7167 ctl_report_tagret_port_groups(struct ctl_scsiio *ctsio)
7168 {
7169         struct ctl_softc *softc = CTL_SOFTC(ctsio);
7170         struct ctl_lun *lun = CTL_LUN(ctsio);
7171         struct scsi_maintenance_in *cdb;
7172         int retval;
7173         int alloc_len, ext, total_len = 0, g, pc, pg, ts, os;
7174         int num_ha_groups, num_target_ports, shared_group;
7175         struct ctl_port *port;
7176         struct scsi_target_group_data *rtg_ptr;
7177         struct scsi_target_group_data_extended *rtg_ext_ptr;
7178         struct scsi_target_port_group_descriptor *tpg_desc;
7179
7180         CTL_DEBUG_PRINT(("ctl_report_tagret_port_groups\n"));
7181
7182         cdb = (struct scsi_maintenance_in *)ctsio->cdb;
7183         retval = CTL_RETVAL_COMPLETE;
7184
7185         switch (cdb->byte2 & STG_PDF_MASK) {
7186         case STG_PDF_LENGTH:
7187                 ext = 0;
7188                 break;
7189         case STG_PDF_EXTENDED:
7190                 ext = 1;
7191                 break;
7192         default:
7193                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7194                                       /*sks_valid*/ 1,
7195                                       /*command*/ 1,
7196                                       /*field*/ 2,
7197                                       /*bit_valid*/ 1,
7198                                       /*bit*/ 5);
7199                 ctl_done((union ctl_io *)ctsio);
7200                 return(retval);
7201         }
7202
7203         num_target_ports = 0;
7204         shared_group = (softc->is_single != 0);
7205         mtx_lock(&softc->ctl_lock);
7206         STAILQ_FOREACH(port, &softc->port_list, links) {
7207                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7208                         continue;
7209                 if (ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
7210                         continue;
7211                 num_target_ports++;
7212                 if (port->status & CTL_PORT_STATUS_HA_SHARED)
7213                         shared_group = 1;
7214         }
7215         mtx_unlock(&softc->ctl_lock);
7216         num_ha_groups = (softc->is_single) ? 0 : NUM_HA_SHELVES;
7217
7218         if (ext)
7219                 total_len = sizeof(struct scsi_target_group_data_extended);
7220         else
7221                 total_len = sizeof(struct scsi_target_group_data);
7222         total_len += sizeof(struct scsi_target_port_group_descriptor) *
7223                 (shared_group + num_ha_groups) +
7224             sizeof(struct scsi_target_port_descriptor) * num_target_ports;
7225
7226         alloc_len = scsi_4btoul(cdb->length);
7227
7228         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7229         ctsio->kern_sg_entries = 0;
7230         ctsio->kern_rel_offset = 0;
7231         ctsio->kern_data_len = min(total_len, alloc_len);
7232         ctsio->kern_total_len = ctsio->kern_data_len;
7233
7234         if (ext) {
7235                 rtg_ext_ptr = (struct scsi_target_group_data_extended *)
7236                     ctsio->kern_data_ptr;
7237                 scsi_ulto4b(total_len - 4, rtg_ext_ptr->length);
7238                 rtg_ext_ptr->format_type = 0x10;
7239                 rtg_ext_ptr->implicit_transition_time = 0;
7240                 tpg_desc = &rtg_ext_ptr->groups[0];
7241         } else {
7242                 rtg_ptr = (struct scsi_target_group_data *)
7243                     ctsio->kern_data_ptr;
7244                 scsi_ulto4b(total_len - 4, rtg_ptr->length);
7245                 tpg_desc = &rtg_ptr->groups[0];
7246         }
7247
7248         mtx_lock(&softc->ctl_lock);
7249         pg = softc->port_min / softc->port_cnt;
7250         if (lun->flags & (CTL_LUN_PRIMARY_SC | CTL_LUN_PEER_SC_PRIMARY)) {
7251                 /* Some shelf is known to be primary. */
7252                 if (softc->ha_link == CTL_HA_LINK_OFFLINE)
7253                         os = TPG_ASYMMETRIC_ACCESS_UNAVAILABLE;
7254                 else if (softc->ha_link == CTL_HA_LINK_UNKNOWN)
7255                         os = TPG_ASYMMETRIC_ACCESS_TRANSITIONING;
7256                 else if (softc->ha_mode == CTL_HA_MODE_ACT_STBY)
7257                         os = TPG_ASYMMETRIC_ACCESS_STANDBY;
7258                 else
7259                         os = TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7260                 if (lun->flags & CTL_LUN_PRIMARY_SC) {
7261                         ts = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7262                 } else {
7263                         ts = os;
7264                         os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7265                 }
7266         } else {
7267                 /* No known primary shelf. */
7268                 if (softc->ha_link == CTL_HA_LINK_OFFLINE) {
7269                         ts = TPG_ASYMMETRIC_ACCESS_UNAVAILABLE;
7270                         os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7271                 } else if (softc->ha_link == CTL_HA_LINK_UNKNOWN) {
7272                         ts = TPG_ASYMMETRIC_ACCESS_TRANSITIONING;
7273                         os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7274                 } else {
7275                         ts = os = TPG_ASYMMETRIC_ACCESS_TRANSITIONING;
7276                 }
7277         }
7278         if (shared_group) {
7279                 tpg_desc->pref_state = ts;
7280                 tpg_desc->support = TPG_AO_SUP | TPG_AN_SUP | TPG_S_SUP |
7281                     TPG_U_SUP | TPG_T_SUP;
7282                 scsi_ulto2b(1, tpg_desc->target_port_group);
7283                 tpg_desc->status = TPG_IMPLICIT;
7284                 pc = 0;
7285                 STAILQ_FOREACH(port, &softc->port_list, links) {
7286                         if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7287                                 continue;
7288                         if (!softc->is_single &&
7289                             (port->status & CTL_PORT_STATUS_HA_SHARED) == 0)
7290                                 continue;
7291                         if (ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
7292                                 continue;
7293                         scsi_ulto2b(port->targ_port, tpg_desc->descriptors[pc].
7294                             relative_target_port_identifier);
7295                         pc++;
7296                 }
7297                 tpg_desc->target_port_count = pc;
7298                 tpg_desc = (struct scsi_target_port_group_descriptor *)
7299                     &tpg_desc->descriptors[pc];
7300         }
7301         for (g = 0; g < num_ha_groups; g++) {
7302                 tpg_desc->pref_state = (g == pg) ? ts : os;
7303                 tpg_desc->support = TPG_AO_SUP | TPG_AN_SUP | TPG_S_SUP |
7304                     TPG_U_SUP | TPG_T_SUP;
7305                 scsi_ulto2b(2 + g, tpg_desc->target_port_group);
7306                 tpg_desc->status = TPG_IMPLICIT;
7307                 pc = 0;
7308                 STAILQ_FOREACH(port, &softc->port_list, links) {
7309                         if (port->targ_port < g * softc->port_cnt ||
7310                             port->targ_port >= (g + 1) * softc->port_cnt)
7311                                 continue;
7312                         if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7313                                 continue;
7314                         if (port->status & CTL_PORT_STATUS_HA_SHARED)
7315                                 continue;
7316                         if (ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
7317                                 continue;
7318                         scsi_ulto2b(port->targ_port, tpg_desc->descriptors[pc].
7319                             relative_target_port_identifier);
7320                         pc++;
7321                 }
7322                 tpg_desc->target_port_count = pc;
7323                 tpg_desc = (struct scsi_target_port_group_descriptor *)
7324                     &tpg_desc->descriptors[pc];
7325         }
7326         mtx_unlock(&softc->ctl_lock);
7327
7328         ctl_set_success(ctsio);
7329         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7330         ctsio->be_move_done = ctl_config_move_done;
7331         ctl_datamove((union ctl_io *)ctsio);
7332         return(retval);
7333 }
7334
7335 int
7336 ctl_report_supported_opcodes(struct ctl_scsiio *ctsio)
7337 {
7338         struct ctl_lun *lun = CTL_LUN(ctsio);
7339         struct scsi_report_supported_opcodes *cdb;
7340         const struct ctl_cmd_entry *entry, *sentry;
7341         struct scsi_report_supported_opcodes_all *all;
7342         struct scsi_report_supported_opcodes_descr *descr;
7343         struct scsi_report_supported_opcodes_one *one;
7344         int retval;
7345         int alloc_len, total_len;
7346         int opcode, service_action, i, j, num;
7347
7348         CTL_DEBUG_PRINT(("ctl_report_supported_opcodes\n"));
7349
7350         cdb = (struct scsi_report_supported_opcodes *)ctsio->cdb;
7351         retval = CTL_RETVAL_COMPLETE;
7352
7353         opcode = cdb->requested_opcode;
7354         service_action = scsi_2btoul(cdb->requested_service_action);
7355         switch (cdb->options & RSO_OPTIONS_MASK) {
7356         case RSO_OPTIONS_ALL:
7357                 num = 0;
7358                 for (i = 0; i < 256; i++) {
7359                         entry = &ctl_cmd_table[i];
7360                         if (entry->flags & CTL_CMD_FLAG_SA5) {
7361                                 for (j = 0; j < 32; j++) {
7362                                         sentry = &((const struct ctl_cmd_entry *)
7363                                             entry->execute)[j];
7364                                         if (ctl_cmd_applicable(
7365                                             lun->be_lun->lun_type, sentry))
7366                                                 num++;
7367                                 }
7368                         } else {
7369                                 if (ctl_cmd_applicable(lun->be_lun->lun_type,
7370                                     entry))
7371                                         num++;
7372                         }
7373                 }
7374                 total_len = sizeof(struct scsi_report_supported_opcodes_all) +
7375                     num * sizeof(struct scsi_report_supported_opcodes_descr);
7376                 break;
7377         case RSO_OPTIONS_OC:
7378                 if (ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) {
7379                         ctl_set_invalid_field(/*ctsio*/ ctsio,
7380                                               /*sks_valid*/ 1,
7381                                               /*command*/ 1,
7382                                               /*field*/ 2,
7383                                               /*bit_valid*/ 1,
7384                                               /*bit*/ 2);
7385                         ctl_done((union ctl_io *)ctsio);
7386                         return (CTL_RETVAL_COMPLETE);
7387                 }
7388                 total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7389                 break;
7390         case RSO_OPTIONS_OC_SA:
7391                 if ((ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) == 0 ||
7392                     service_action >= 32) {
7393                         ctl_set_invalid_field(/*ctsio*/ ctsio,
7394                                               /*sks_valid*/ 1,
7395                                               /*command*/ 1,
7396                                               /*field*/ 2,
7397                                               /*bit_valid*/ 1,
7398                                               /*bit*/ 2);
7399                         ctl_done((union ctl_io *)ctsio);
7400                         return (CTL_RETVAL_COMPLETE);
7401                 }
7402                 /* FALLTHROUGH */
7403         case RSO_OPTIONS_OC_ASA:
7404                 total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7405                 break;
7406         default:
7407                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7408                                       /*sks_valid*/ 1,
7409                                       /*command*/ 1,
7410                                       /*field*/ 2,
7411                                       /*bit_valid*/ 1,
7412                                       /*bit*/ 2);
7413                 ctl_done((union ctl_io *)ctsio);
7414                 return (CTL_RETVAL_COMPLETE);
7415         }
7416
7417         alloc_len = scsi_4btoul(cdb->length);
7418
7419         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7420         ctsio->kern_sg_entries = 0;
7421         ctsio->kern_rel_offset = 0;
7422         ctsio->kern_data_len = min(total_len, alloc_len);
7423         ctsio->kern_total_len = ctsio->kern_data_len;
7424
7425         switch (cdb->options & RSO_OPTIONS_MASK) {
7426         case RSO_OPTIONS_ALL:
7427                 all = (struct scsi_report_supported_opcodes_all *)
7428                     ctsio->kern_data_ptr;
7429                 num = 0;
7430                 for (i = 0; i < 256; i++) {
7431                         entry = &ctl_cmd_table[i];
7432                         if (entry->flags & CTL_CMD_FLAG_SA5) {
7433                                 for (j = 0; j < 32; j++) {
7434                                         sentry = &((const struct ctl_cmd_entry *)
7435                                             entry->execute)[j];
7436                                         if (!ctl_cmd_applicable(
7437                                             lun->be_lun->lun_type, sentry))
7438                                                 continue;
7439                                         descr = &all->descr[num++];
7440                                         descr->opcode = i;
7441                                         scsi_ulto2b(j, descr->service_action);
7442                                         descr->flags = RSO_SERVACTV;
7443                                         scsi_ulto2b(sentry->length,
7444                                             descr->cdb_length);
7445                                 }
7446                         } else {
7447                                 if (!ctl_cmd_applicable(lun->be_lun->lun_type,
7448                                     entry))
7449                                         continue;
7450                                 descr = &all->descr[num++];
7451                                 descr->opcode = i;
7452                                 scsi_ulto2b(0, descr->service_action);
7453                                 descr->flags = 0;
7454                                 scsi_ulto2b(entry->length, descr->cdb_length);
7455                         }
7456                 }
7457                 scsi_ulto4b(
7458                     num * sizeof(struct scsi_report_supported_opcodes_descr),
7459                     all->length);
7460                 break;
7461         case RSO_OPTIONS_OC:
7462                 one = (struct scsi_report_supported_opcodes_one *)
7463                     ctsio->kern_data_ptr;
7464                 entry = &ctl_cmd_table[opcode];
7465                 goto fill_one;
7466         case RSO_OPTIONS_OC_SA:
7467                 one = (struct scsi_report_supported_opcodes_one *)
7468                     ctsio->kern_data_ptr;
7469                 entry = &ctl_cmd_table[opcode];
7470                 entry = &((const struct ctl_cmd_entry *)
7471                     entry->execute)[service_action];
7472 fill_one:
7473                 if (ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
7474                         one->support = 3;
7475                         scsi_ulto2b(entry->length, one->cdb_length);
7476                         one->cdb_usage[0] = opcode;
7477                         memcpy(&one->cdb_usage[1], entry->usage,
7478                             entry->length - 1);
7479                 } else
7480                         one->support = 1;
7481                 break;
7482         case RSO_OPTIONS_OC_ASA:
7483                 one = (struct scsi_report_supported_opcodes_one *)
7484                     ctsio->kern_data_ptr;
7485                 entry = &ctl_cmd_table[opcode];
7486                 if (entry->flags & CTL_CMD_FLAG_SA5) {
7487                         entry = &((const struct ctl_cmd_entry *)
7488                             entry->execute)[service_action];
7489                 } else if (service_action != 0) {
7490                         one->support = 1;
7491                         break;
7492                 }
7493                 goto fill_one;
7494         }
7495
7496         ctl_set_success(ctsio);
7497         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7498         ctsio->be_move_done = ctl_config_move_done;
7499         ctl_datamove((union ctl_io *)ctsio);
7500         return(retval);
7501 }
7502
7503 int
7504 ctl_report_supported_tmf(struct ctl_scsiio *ctsio)
7505 {
7506         struct scsi_report_supported_tmf *cdb;
7507         struct scsi_report_supported_tmf_ext_data *data;
7508         int retval;
7509         int alloc_len, total_len;
7510
7511         CTL_DEBUG_PRINT(("ctl_report_supported_tmf\n"));
7512
7513         cdb = (struct scsi_report_supported_tmf *)ctsio->cdb;
7514
7515         retval = CTL_RETVAL_COMPLETE;
7516
7517         if (cdb->options & RST_REPD)
7518                 total_len = sizeof(struct scsi_report_supported_tmf_ext_data);
7519         else
7520                 total_len = sizeof(struct scsi_report_supported_tmf_data);
7521         alloc_len = scsi_4btoul(cdb->length);
7522
7523         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7524         ctsio->kern_sg_entries = 0;
7525         ctsio->kern_rel_offset = 0;
7526         ctsio->kern_data_len = min(total_len, alloc_len);
7527         ctsio->kern_total_len = ctsio->kern_data_len;
7528
7529         data = (struct scsi_report_supported_tmf_ext_data *)ctsio->kern_data_ptr;
7530         data->byte1 |= RST_ATS | RST_ATSS | RST_CTSS | RST_LURS | RST_QTS |
7531             RST_TRS;
7532         data->byte2 |= RST_QAES | RST_QTSS | RST_ITNRS;
7533         data->length = total_len - 4;
7534
7535         ctl_set_success(ctsio);
7536         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7537         ctsio->be_move_done = ctl_config_move_done;
7538         ctl_datamove((union ctl_io *)ctsio);
7539         return (retval);
7540 }
7541
7542 int
7543 ctl_report_timestamp(struct ctl_scsiio *ctsio)
7544 {
7545         struct scsi_report_timestamp *cdb;
7546         struct scsi_report_timestamp_data *data;
7547         struct timeval tv;
7548         int64_t timestamp;
7549         int retval;
7550         int alloc_len, total_len;
7551
7552         CTL_DEBUG_PRINT(("ctl_report_timestamp\n"));
7553
7554         cdb = (struct scsi_report_timestamp *)ctsio->cdb;
7555
7556         retval = CTL_RETVAL_COMPLETE;
7557
7558         total_len = sizeof(struct scsi_report_timestamp_data);
7559         alloc_len = scsi_4btoul(cdb->length);
7560
7561         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7562         ctsio->kern_sg_entries = 0;
7563         ctsio->kern_rel_offset = 0;
7564         ctsio->kern_data_len = min(total_len, alloc_len);
7565         ctsio->kern_total_len = ctsio->kern_data_len;
7566
7567         data = (struct scsi_report_timestamp_data *)ctsio->kern_data_ptr;
7568         scsi_ulto2b(sizeof(*data) - 2, data->length);
7569         data->origin = RTS_ORIG_OUTSIDE;
7570         getmicrotime(&tv);
7571         timestamp = (int64_t)tv.tv_sec * 1000 + tv.tv_usec / 1000;
7572         scsi_ulto4b(timestamp >> 16, data->timestamp);
7573         scsi_ulto2b(timestamp & 0xffff, &data->timestamp[4]);
7574
7575         ctl_set_success(ctsio);
7576         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7577         ctsio->be_move_done = ctl_config_move_done;
7578         ctl_datamove((union ctl_io *)ctsio);
7579         return (retval);
7580 }
7581
7582 int
7583 ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7584 {
7585         struct ctl_softc *softc = CTL_SOFTC(ctsio);
7586         struct ctl_lun *lun = CTL_LUN(ctsio);
7587         struct scsi_per_res_in *cdb;
7588         int alloc_len, total_len = 0;
7589         /* struct scsi_per_res_in_rsrv in_data; */
7590         uint64_t key;
7591
7592         CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7593
7594         cdb = (struct scsi_per_res_in *)ctsio->cdb;
7595
7596         alloc_len = scsi_2btoul(cdb->length);
7597
7598 retry:
7599         mtx_lock(&lun->lun_lock);
7600         switch (cdb->action) {
7601         case SPRI_RK: /* read keys */
7602                 total_len = sizeof(struct scsi_per_res_in_keys) +
7603                         lun->pr_key_count *
7604                         sizeof(struct scsi_per_res_key);
7605                 break;
7606         case SPRI_RR: /* read reservation */
7607                 if (lun->flags & CTL_LUN_PR_RESERVED)
7608                         total_len = sizeof(struct scsi_per_res_in_rsrv);
7609                 else
7610                         total_len = sizeof(struct scsi_per_res_in_header);
7611                 break;
7612         case SPRI_RC: /* report capabilities */
7613                 total_len = sizeof(struct scsi_per_res_cap);
7614                 break;
7615         case SPRI_RS: /* read full status */
7616                 total_len = sizeof(struct scsi_per_res_in_header) +
7617                     (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7618                     lun->pr_key_count;
7619                 break;
7620         default:
7621                 panic("%s: Invalid PR type %#x", __func__, cdb->action);
7622         }
7623         mtx_unlock(&lun->lun_lock);
7624
7625         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7626         ctsio->kern_rel_offset = 0;
7627         ctsio->kern_sg_entries = 0;
7628         ctsio->kern_data_len = min(total_len, alloc_len);
7629         ctsio->kern_total_len = ctsio->kern_data_len;
7630
7631         mtx_lock(&lun->lun_lock);
7632         switch (cdb->action) {
7633         case SPRI_RK: { // read keys
7634         struct scsi_per_res_in_keys *res_keys;
7635                 int i, key_count;
7636
7637                 res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7638
7639                 /*
7640                  * We had to drop the lock to allocate our buffer, which
7641                  * leaves time for someone to come in with another
7642                  * persistent reservation.  (That is unlikely, though,
7643                  * since this should be the only persistent reservation
7644                  * command active right now.)
7645                  */
7646                 if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7647                     (lun->pr_key_count *
7648                      sizeof(struct scsi_per_res_key)))){
7649                         mtx_unlock(&lun->lun_lock);
7650                         free(ctsio->kern_data_ptr, M_CTL);
7651                         printf("%s: reservation length changed, retrying\n",
7652                                __func__);
7653                         goto retry;
7654                 }
7655
7656                 scsi_ulto4b(lun->pr_generation, res_keys->header.generation);
7657
7658                 scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7659                              lun->pr_key_count, res_keys->header.length);
7660
7661                 for (i = 0, key_count = 0; i < CTL_MAX_INITIATORS; i++) {
7662                         if ((key = ctl_get_prkey(lun, i)) == 0)
7663                                 continue;
7664
7665                         /*
7666                          * We used lun->pr_key_count to calculate the
7667                          * size to allocate.  If it turns out the number of
7668                          * initiators with the registered flag set is
7669                          * larger than that (i.e. they haven't been kept in
7670                          * sync), we've got a problem.
7671                          */
7672                         if (key_count >= lun->pr_key_count) {
7673                                 key_count++;
7674                                 continue;
7675                         }
7676                         scsi_u64to8b(key, res_keys->keys[key_count].key);
7677                         key_count++;
7678                 }
7679                 break;
7680         }
7681         case SPRI_RR: { // read reservation
7682                 struct scsi_per_res_in_rsrv *res;
7683                 int tmp_len, header_only;
7684
7685                 res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7686
7687                 scsi_ulto4b(lun->pr_generation, res->header.generation);
7688
7689                 if (lun->flags & CTL_LUN_PR_RESERVED)
7690                 {
7691                         tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7692                         scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7693                                     res->header.length);
7694                         header_only = 0;
7695                 } else {
7696                         tmp_len = sizeof(struct scsi_per_res_in_header);
7697                         scsi_ulto4b(0, res->header.length);
7698                         header_only = 1;
7699                 }
7700
7701                 /*
7702                  * We had to drop the lock to allocate our buffer, which
7703                  * leaves time for someone to come in with another
7704                  * persistent reservation.  (That is unlikely, though,
7705                  * since this should be the only persistent reservation
7706                  * command active right now.)
7707                  */
7708                 if (tmp_len != total_len) {
7709                         mtx_unlock(&lun->lun_lock);
7710                         free(ctsio->kern_data_ptr, M_CTL);
7711                         printf("%s: reservation status changed, retrying\n",
7712                                __func__);
7713                         goto retry;
7714                 }
7715
7716                 /*
7717                  * No reservation held, so we're done.
7718                  */
7719                 if (header_only != 0)
7720                         break;
7721
7722                 /*
7723                  * If the registration is an All Registrants type, the key
7724                  * is 0, since it doesn't really matter.
7725                  */
7726                 if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7727                         scsi_u64to8b(ctl_get_prkey(lun, lun->pr_res_idx),
7728                             res->data.reservation);
7729                 }
7730                 res->data.scopetype = lun->pr_res_type;
7731                 break;
7732         }
7733         case SPRI_RC:     //report capabilities
7734         {
7735                 struct scsi_per_res_cap *res_cap;
7736                 uint16_t type_mask;
7737
7738                 res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7739                 scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7740                 res_cap->flags1 = SPRI_CRH;
7741                 res_cap->flags2 = SPRI_TMV | SPRI_ALLOW_5;
7742                 type_mask = SPRI_TM_WR_EX_AR |
7743                             SPRI_TM_EX_AC_RO |
7744                             SPRI_TM_WR_EX_RO |
7745                             SPRI_TM_EX_AC |
7746                             SPRI_TM_WR_EX |
7747                             SPRI_TM_EX_AC_AR;
7748                 scsi_ulto2b(type_mask, res_cap->type_mask);
7749                 break;
7750         }
7751         case SPRI_RS: { // read full status
7752                 struct scsi_per_res_in_full *res_status;
7753                 struct scsi_per_res_in_full_desc *res_desc;
7754                 struct ctl_port *port;
7755                 int i, len;
7756
7757                 res_status = (struct scsi_per_res_in_full*)ctsio->kern_data_ptr;
7758
7759                 /*
7760                  * We had to drop the lock to allocate our buffer, which
7761                  * leaves time for someone to come in with another
7762                  * persistent reservation.  (That is unlikely, though,
7763                  * since this should be the only persistent reservation
7764                  * command active right now.)
7765                  */
7766                 if (total_len < (sizeof(struct scsi_per_res_in_header) +
7767                     (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7768                      lun->pr_key_count)){
7769                         mtx_unlock(&lun->lun_lock);
7770                         free(ctsio->kern_data_ptr, M_CTL);
7771                         printf("%s: reservation length changed, retrying\n",
7772                                __func__);
7773                         goto retry;
7774                 }
7775
7776                 scsi_ulto4b(lun->pr_generation, res_status->header.generation);
7777
7778                 res_desc = &res_status->desc[0];
7779                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
7780                         if ((key = ctl_get_prkey(lun, i)) == 0)
7781                                 continue;
7782
7783                         scsi_u64to8b(key, res_desc->res_key.key);
7784                         if ((lun->flags & CTL_LUN_PR_RESERVED) &&
7785                             (lun->pr_res_idx == i ||
7786                              lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS)) {
7787                                 res_desc->flags = SPRI_FULL_R_HOLDER;
7788                                 res_desc->scopetype = lun->pr_res_type;
7789                         }
7790                         scsi_ulto2b(i / CTL_MAX_INIT_PER_PORT,
7791                             res_desc->rel_trgt_port_id);
7792                         len = 0;
7793                         port = softc->ctl_ports[i / CTL_MAX_INIT_PER_PORT];
7794                         if (port != NULL)
7795                                 len = ctl_create_iid(port,
7796                                     i % CTL_MAX_INIT_PER_PORT,
7797                                     res_desc->transport_id);
7798                         scsi_ulto4b(len, res_desc->additional_length);
7799                         res_desc = (struct scsi_per_res_in_full_desc *)
7800                             &res_desc->transport_id[len];
7801                 }
7802                 scsi_ulto4b((uint8_t *)res_desc - (uint8_t *)&res_status->desc[0],
7803                     res_status->header.length);
7804                 break;
7805         }
7806         default:
7807                 panic("%s: Invalid PR type %#x", __func__, cdb->action);
7808         }
7809         mtx_unlock(&lun->lun_lock);
7810
7811         ctl_set_success(ctsio);
7812         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7813         ctsio->be_move_done = ctl_config_move_done;
7814         ctl_datamove((union ctl_io *)ctsio);
7815         return (CTL_RETVAL_COMPLETE);
7816 }
7817
7818 /*
7819  * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
7820  * it should return.
7821  */
7822 static int
7823 ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
7824                 uint64_t sa_res_key, uint8_t type, uint32_t residx,
7825                 struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
7826                 struct scsi_per_res_out_parms* param)
7827 {
7828         union ctl_ha_msg persis_io;
7829         int i;
7830
7831         mtx_lock(&lun->lun_lock);
7832         if (sa_res_key == 0) {
7833                 if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
7834                         /* validate scope and type */
7835                         if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7836                              SPR_LU_SCOPE) {
7837                                 mtx_unlock(&lun->lun_lock);
7838                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7839                                                       /*sks_valid*/ 1,
7840                                                       /*command*/ 1,
7841                                                       /*field*/ 2,
7842                                                       /*bit_valid*/ 1,
7843                                                       /*bit*/ 4);
7844                                 ctl_done((union ctl_io *)ctsio);
7845                                 return (1);
7846                         }
7847
7848                         if (type>8 || type==2 || type==4 || type==0) {
7849                                 mtx_unlock(&lun->lun_lock);
7850                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7851                                                       /*sks_valid*/ 1,
7852                                                       /*command*/ 1,
7853                                                       /*field*/ 2,
7854                                                       /*bit_valid*/ 1,
7855                                                       /*bit*/ 0);
7856                                 ctl_done((union ctl_io *)ctsio);
7857                                 return (1);
7858                         }
7859
7860                         /*
7861                          * Unregister everybody else and build UA for
7862                          * them
7863                          */
7864                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
7865                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
7866                                         continue;
7867
7868                                 ctl_clr_prkey(lun, i);
7869                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7870                         }
7871                         lun->pr_key_count = 1;
7872                         lun->pr_res_type = type;
7873                         if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
7874                             lun->pr_res_type != SPR_TYPE_EX_AC_AR)
7875                                 lun->pr_res_idx = residx;
7876                         lun->pr_generation++;
7877                         mtx_unlock(&lun->lun_lock);
7878
7879                         /* send msg to other side */
7880                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7881                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7882                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7883                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
7884                         persis_io.pr.pr_info.res_type = type;
7885                         memcpy(persis_io.pr.pr_info.sa_res_key,
7886                                param->serv_act_res_key,
7887                                sizeof(param->serv_act_res_key));
7888                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
7889                             sizeof(persis_io.pr), M_WAITOK);
7890                 } else {
7891                         /* not all registrants */
7892                         mtx_unlock(&lun->lun_lock);
7893                         free(ctsio->kern_data_ptr, M_CTL);
7894                         ctl_set_invalid_field(ctsio,
7895                                               /*sks_valid*/ 1,
7896                                               /*command*/ 0,
7897                                               /*field*/ 8,
7898                                               /*bit_valid*/ 0,
7899                                               /*bit*/ 0);
7900                         ctl_done((union ctl_io *)ctsio);
7901                         return (1);
7902                 }
7903         } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
7904                 || !(lun->flags & CTL_LUN_PR_RESERVED)) {
7905                 int found = 0;
7906
7907                 if (res_key == sa_res_key) {
7908                         /* special case */
7909                         /*
7910                          * The spec implies this is not good but doesn't
7911                          * say what to do. There are two choices either
7912                          * generate a res conflict or check condition
7913                          * with illegal field in parameter data. Since
7914                          * that is what is done when the sa_res_key is
7915                          * zero I'll take that approach since this has
7916                          * to do with the sa_res_key.
7917                          */
7918                         mtx_unlock(&lun->lun_lock);
7919                         free(ctsio->kern_data_ptr, M_CTL);
7920                         ctl_set_invalid_field(ctsio,
7921                                               /*sks_valid*/ 1,
7922                                               /*command*/ 0,
7923                                               /*field*/ 8,
7924                                               /*bit_valid*/ 0,
7925                                               /*bit*/ 0);
7926                         ctl_done((union ctl_io *)ctsio);
7927                         return (1);
7928                 }
7929
7930                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
7931                         if (ctl_get_prkey(lun, i) != sa_res_key)
7932                                 continue;
7933
7934                         found = 1;
7935                         ctl_clr_prkey(lun, i);
7936                         lun->pr_key_count--;
7937                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7938                 }
7939                 if (!found) {
7940                         mtx_unlock(&lun->lun_lock);
7941                         free(ctsio->kern_data_ptr, M_CTL);
7942                         ctl_set_reservation_conflict(ctsio);
7943                         ctl_done((union ctl_io *)ctsio);
7944                         return (CTL_RETVAL_COMPLETE);
7945                 }
7946                 lun->pr_generation++;
7947                 mtx_unlock(&lun->lun_lock);
7948
7949                 /* send msg to other side */
7950                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7951                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7952                 persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7953                 persis_io.pr.pr_info.residx = lun->pr_res_idx;
7954                 persis_io.pr.pr_info.res_type = type;
7955                 memcpy(persis_io.pr.pr_info.sa_res_key,
7956                        param->serv_act_res_key,
7957                        sizeof(param->serv_act_res_key));
7958                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
7959                     sizeof(persis_io.pr), M_WAITOK);
7960         } else {
7961                 /* Reserved but not all registrants */
7962                 /* sa_res_key is res holder */
7963                 if (sa_res_key == ctl_get_prkey(lun, lun->pr_res_idx)) {
7964                         /* validate scope and type */
7965                         if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7966                              SPR_LU_SCOPE) {
7967                                 mtx_unlock(&lun->lun_lock);
7968                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7969                                                       /*sks_valid*/ 1,
7970                                                       /*command*/ 1,
7971                                                       /*field*/ 2,
7972                                                       /*bit_valid*/ 1,
7973                                                       /*bit*/ 4);
7974                                 ctl_done((union ctl_io *)ctsio);
7975                                 return (1);
7976                         }
7977
7978                         if (type>8 || type==2 || type==4 || type==0) {
7979                                 mtx_unlock(&lun->lun_lock);
7980                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7981                                                       /*sks_valid*/ 1,
7982                                                       /*command*/ 1,
7983                                                       /*field*/ 2,
7984                                                       /*bit_valid*/ 1,
7985                                                       /*bit*/ 0);
7986                                 ctl_done((union ctl_io *)ctsio);
7987                                 return (1);
7988                         }
7989
7990                         /*
7991                          * Do the following:
7992                          * if sa_res_key != res_key remove all
7993                          * registrants w/sa_res_key and generate UA
7994                          * for these registrants(Registrations
7995                          * Preempted) if it wasn't an exclusive
7996                          * reservation generate UA(Reservations
7997                          * Preempted) for all other registered nexuses
7998                          * if the type has changed. Establish the new
7999                          * reservation and holder. If res_key and
8000                          * sa_res_key are the same do the above
8001                          * except don't unregister the res holder.
8002                          */
8003
8004                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
8005                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
8006                                         continue;
8007
8008                                 if (sa_res_key == ctl_get_prkey(lun, i)) {
8009                                         ctl_clr_prkey(lun, i);
8010                                         lun->pr_key_count--;
8011                                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8012                                 } else if (type != lun->pr_res_type &&
8013                                     (lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8014                                      lun->pr_res_type == SPR_TYPE_EX_AC_RO)) {
8015                                         ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8016                                 }
8017                         }
8018                         lun->pr_res_type = type;
8019                         if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
8020                             lun->pr_res_type != SPR_TYPE_EX_AC_AR)
8021                                 lun->pr_res_idx = residx;
8022                         else
8023                                 lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8024                         lun->pr_generation++;
8025                         mtx_unlock(&lun->lun_lock);
8026
8027                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8028                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8029                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8030                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8031                         persis_io.pr.pr_info.res_type = type;
8032                         memcpy(persis_io.pr.pr_info.sa_res_key,
8033                                param->serv_act_res_key,
8034                                sizeof(param->serv_act_res_key));
8035                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8036                             sizeof(persis_io.pr), M_WAITOK);
8037                 } else {
8038                         /*
8039                          * sa_res_key is not the res holder just
8040                          * remove registrants
8041                          */
8042                         int found=0;
8043
8044                         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8045                                 if (sa_res_key != ctl_get_prkey(lun, i))
8046                                         continue;
8047
8048                                 found = 1;
8049                                 ctl_clr_prkey(lun, i);
8050                                 lun->pr_key_count--;
8051                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8052                         }
8053
8054                         if (!found) {
8055                                 mtx_unlock(&lun->lun_lock);
8056                                 free(ctsio->kern_data_ptr, M_CTL);
8057                                 ctl_set_reservation_conflict(ctsio);
8058                                 ctl_done((union ctl_io *)ctsio);
8059                                 return (1);
8060                         }
8061                         lun->pr_generation++;
8062                         mtx_unlock(&lun->lun_lock);
8063
8064                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8065                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8066                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8067                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8068                         persis_io.pr.pr_info.res_type = type;
8069                         memcpy(persis_io.pr.pr_info.sa_res_key,
8070                                param->serv_act_res_key,
8071                                sizeof(param->serv_act_res_key));
8072                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8073                             sizeof(persis_io.pr), M_WAITOK);
8074                 }
8075         }
8076         return (0);
8077 }
8078
8079 static void
8080 ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
8081 {
8082         uint64_t sa_res_key;
8083         int i;
8084
8085         sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
8086
8087         if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8088          || lun->pr_res_idx == CTL_PR_NO_RESERVATION
8089          || sa_res_key != ctl_get_prkey(lun, lun->pr_res_idx)) {
8090                 if (sa_res_key == 0) {
8091                         /*
8092                          * Unregister everybody else and build UA for
8093                          * them
8094                          */
8095                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
8096                                 if (i == msg->pr.pr_info.residx ||
8097                                     ctl_get_prkey(lun, i) == 0)
8098                                         continue;
8099
8100                                 ctl_clr_prkey(lun, i);
8101                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8102                         }
8103
8104                         lun->pr_key_count = 1;
8105                         lun->pr_res_type = msg->pr.pr_info.res_type;
8106                         if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
8107                             lun->pr_res_type != SPR_TYPE_EX_AC_AR)
8108                                 lun->pr_res_idx = msg->pr.pr_info.residx;
8109                 } else {
8110                         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8111                                 if (sa_res_key == ctl_get_prkey(lun, i))
8112                                         continue;
8113
8114                                 ctl_clr_prkey(lun, i);
8115                                 lun->pr_key_count--;
8116                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8117                         }
8118                 }
8119         } else {
8120                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8121                         if (i == msg->pr.pr_info.residx ||
8122                             ctl_get_prkey(lun, i) == 0)
8123                                 continue;
8124
8125                         if (sa_res_key == ctl_get_prkey(lun, i)) {
8126                                 ctl_clr_prkey(lun, i);
8127                                 lun->pr_key_count--;
8128                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8129                         } else if (msg->pr.pr_info.res_type != lun->pr_res_type
8130                             && (lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8131                              lun->pr_res_type == SPR_TYPE_EX_AC_RO)) {
8132                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8133                         }
8134                 }
8135                 lun->pr_res_type = msg->pr.pr_info.res_type;
8136                 if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
8137                     lun->pr_res_type != SPR_TYPE_EX_AC_AR)
8138                         lun->pr_res_idx = msg->pr.pr_info.residx;
8139                 else
8140                         lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8141         }
8142         lun->pr_generation++;
8143
8144 }
8145
8146
8147 int
8148 ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
8149 {
8150         struct ctl_softc *softc = CTL_SOFTC(ctsio);
8151         struct ctl_lun *lun = CTL_LUN(ctsio);
8152         int retval;
8153         u_int32_t param_len;
8154         struct scsi_per_res_out *cdb;
8155         struct scsi_per_res_out_parms* param;
8156         uint32_t residx;
8157         uint64_t res_key, sa_res_key, key;
8158         uint8_t type;
8159         union ctl_ha_msg persis_io;
8160         int    i;
8161
8162         CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n"));
8163
8164         cdb = (struct scsi_per_res_out *)ctsio->cdb;
8165         retval = CTL_RETVAL_COMPLETE;
8166
8167         /*
8168          * We only support whole-LUN scope.  The scope & type are ignored for
8169          * register, register and ignore existing key and clear.
8170          * We sometimes ignore scope and type on preempts too!!
8171          * Verify reservation type here as well.
8172          */
8173         type = cdb->scope_type & SPR_TYPE_MASK;
8174         if ((cdb->action == SPRO_RESERVE)
8175          || (cdb->action == SPRO_RELEASE)) {
8176                 if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
8177                         ctl_set_invalid_field(/*ctsio*/ ctsio,
8178                                               /*sks_valid*/ 1,
8179                                               /*command*/ 1,
8180                                               /*field*/ 2,
8181                                               /*bit_valid*/ 1,
8182                                               /*bit*/ 4);
8183                         ctl_done((union ctl_io *)ctsio);
8184                         return (CTL_RETVAL_COMPLETE);
8185                 }
8186
8187                 if (type>8 || type==2 || type==4 || type==0) {
8188                         ctl_set_invalid_field(/*ctsio*/ ctsio,
8189                                               /*sks_valid*/ 1,
8190                                               /*command*/ 1,
8191                                               /*field*/ 2,
8192                                               /*bit_valid*/ 1,
8193                                               /*bit*/ 0);
8194                         ctl_done((union ctl_io *)ctsio);
8195                         return (CTL_RETVAL_COMPLETE);
8196                 }
8197         }
8198
8199         param_len = scsi_4btoul(cdb->length);
8200
8201         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
8202                 ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
8203                 ctsio->kern_data_len = param_len;
8204                 ctsio->kern_total_len = param_len;
8205                 ctsio->kern_rel_offset = 0;
8206                 ctsio->kern_sg_entries = 0;
8207                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
8208                 ctsio->be_move_done = ctl_config_move_done;
8209                 ctl_datamove((union ctl_io *)ctsio);
8210
8211                 return (CTL_RETVAL_COMPLETE);
8212         }
8213
8214         param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
8215
8216         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
8217         res_key = scsi_8btou64(param->res_key.key);
8218         sa_res_key = scsi_8btou64(param->serv_act_res_key);
8219
8220         /*
8221          * Validate the reservation key here except for SPRO_REG_IGNO
8222          * This must be done for all other service actions
8223          */
8224         if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
8225                 mtx_lock(&lun->lun_lock);
8226                 if ((key = ctl_get_prkey(lun, residx)) != 0) {
8227                         if (res_key != key) {
8228                                 /*
8229                                  * The current key passed in doesn't match
8230                                  * the one the initiator previously
8231                                  * registered.
8232                                  */
8233                                 mtx_unlock(&lun->lun_lock);
8234                                 free(ctsio->kern_data_ptr, M_CTL);
8235                                 ctl_set_reservation_conflict(ctsio);
8236                                 ctl_done((union ctl_io *)ctsio);
8237                                 return (CTL_RETVAL_COMPLETE);
8238                         }
8239                 } else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
8240                         /*
8241                          * We are not registered
8242                          */
8243                         mtx_unlock(&lun->lun_lock);
8244                         free(ctsio->kern_data_ptr, M_CTL);
8245                         ctl_set_reservation_conflict(ctsio);
8246                         ctl_done((union ctl_io *)ctsio);
8247                         return (CTL_RETVAL_COMPLETE);
8248                 } else if (res_key != 0) {
8249                         /*
8250                          * We are not registered and trying to register but
8251                          * the register key isn't zero.
8252                          */
8253                         mtx_unlock(&lun->lun_lock);
8254                         free(ctsio->kern_data_ptr, M_CTL);
8255                         ctl_set_reservation_conflict(ctsio);
8256                         ctl_done((union ctl_io *)ctsio);
8257                         return (CTL_RETVAL_COMPLETE);
8258                 }
8259                 mtx_unlock(&lun->lun_lock);
8260         }
8261
8262         switch (cdb->action & SPRO_ACTION_MASK) {
8263         case SPRO_REGISTER:
8264         case SPRO_REG_IGNO: {
8265
8266 #if 0
8267                 printf("Registration received\n");
8268 #endif
8269
8270                 /*
8271                  * We don't support any of these options, as we report in
8272                  * the read capabilities request (see
8273                  * ctl_persistent_reserve_in(), above).
8274                  */
8275                 if ((param->flags & SPR_SPEC_I_PT)
8276                  || (param->flags & SPR_ALL_TG_PT)
8277                  || (param->flags & SPR_APTPL)) {
8278                         int bit_ptr;
8279
8280                         if (param->flags & SPR_APTPL)
8281                                 bit_ptr = 0;
8282                         else if (param->flags & SPR_ALL_TG_PT)
8283                                 bit_ptr = 2;
8284                         else /* SPR_SPEC_I_PT */
8285                                 bit_ptr = 3;
8286
8287                         free(ctsio->kern_data_ptr, M_CTL);
8288                         ctl_set_invalid_field(ctsio,
8289                                               /*sks_valid*/ 1,
8290                                               /*command*/ 0,
8291                                               /*field*/ 20,
8292                                               /*bit_valid*/ 1,
8293                                               /*bit*/ bit_ptr);
8294                         ctl_done((union ctl_io *)ctsio);
8295                         return (CTL_RETVAL_COMPLETE);
8296                 }
8297
8298                 mtx_lock(&lun->lun_lock);
8299
8300                 /*
8301                  * The initiator wants to clear the
8302                  * key/unregister.
8303                  */
8304                 if (sa_res_key == 0) {
8305                         if ((res_key == 0
8306                           && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
8307                          || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
8308                           && ctl_get_prkey(lun, residx) == 0)) {
8309                                 mtx_unlock(&lun->lun_lock);
8310                                 goto done;
8311                         }
8312
8313                         ctl_clr_prkey(lun, residx);
8314                         lun->pr_key_count--;
8315
8316                         if (residx == lun->pr_res_idx) {
8317                                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8318                                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8319
8320                                 if ((lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8321                                      lun->pr_res_type == SPR_TYPE_EX_AC_RO) &&
8322                                     lun->pr_key_count) {
8323                                         /*
8324                                          * If the reservation is a registrants
8325                                          * only type we need to generate a UA
8326                                          * for other registered inits.  The
8327                                          * sense code should be RESERVATIONS
8328                                          * RELEASED
8329                                          */
8330
8331                                         for (i = softc->init_min; i < softc->init_max; i++){
8332                                                 if (ctl_get_prkey(lun, i) == 0)
8333                                                         continue;
8334                                                 ctl_est_ua(lun, i,
8335                                                     CTL_UA_RES_RELEASE);
8336                                         }
8337                                 }
8338                                 lun->pr_res_type = 0;
8339                         } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8340                                 if (lun->pr_key_count==0) {
8341                                         lun->flags &= ~CTL_LUN_PR_RESERVED;
8342                                         lun->pr_res_type = 0;
8343                                         lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8344                                 }
8345                         }
8346                         lun->pr_generation++;
8347                         mtx_unlock(&lun->lun_lock);
8348
8349                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8350                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8351                         persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
8352                         persis_io.pr.pr_info.residx = residx;
8353                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8354                             sizeof(persis_io.pr), M_WAITOK);
8355                 } else /* sa_res_key != 0 */ {
8356
8357                         /*
8358                          * If we aren't registered currently then increment
8359                          * the key count and set the registered flag.
8360                          */
8361                         ctl_alloc_prkey(lun, residx);
8362                         if (ctl_get_prkey(lun, residx) == 0)
8363                                 lun->pr_key_count++;
8364                         ctl_set_prkey(lun, residx, sa_res_key);
8365                         lun->pr_generation++;
8366                         mtx_unlock(&lun->lun_lock);
8367
8368                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8369                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8370                         persis_io.pr.pr_info.action = CTL_PR_REG_KEY;
8371                         persis_io.pr.pr_info.residx = residx;
8372                         memcpy(persis_io.pr.pr_info.sa_res_key,
8373                                param->serv_act_res_key,
8374                                sizeof(param->serv_act_res_key));
8375                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8376                             sizeof(persis_io.pr), M_WAITOK);
8377                 }
8378
8379                 break;
8380         }
8381         case SPRO_RESERVE:
8382 #if 0
8383                 printf("Reserve executed type %d\n", type);
8384 #endif
8385                 mtx_lock(&lun->lun_lock);
8386                 if (lun->flags & CTL_LUN_PR_RESERVED) {
8387                         /*
8388                          * if this isn't the reservation holder and it's
8389                          * not a "all registrants" type or if the type is
8390                          * different then we have a conflict
8391                          */
8392                         if ((lun->pr_res_idx != residx
8393                           && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8394                          || lun->pr_res_type != type) {
8395                                 mtx_unlock(&lun->lun_lock);
8396                                 free(ctsio->kern_data_ptr, M_CTL);
8397                                 ctl_set_reservation_conflict(ctsio);
8398                                 ctl_done((union ctl_io *)ctsio);
8399                                 return (CTL_RETVAL_COMPLETE);
8400                         }
8401                         mtx_unlock(&lun->lun_lock);
8402                 } else /* create a reservation */ {
8403                         /*
8404                          * If it's not an "all registrants" type record
8405                          * reservation holder
8406                          */
8407                         if (type != SPR_TYPE_WR_EX_AR
8408                          && type != SPR_TYPE_EX_AC_AR)
8409                                 lun->pr_res_idx = residx; /* Res holder */
8410                         else
8411                                 lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8412
8413                         lun->flags |= CTL_LUN_PR_RESERVED;
8414                         lun->pr_res_type = type;
8415
8416                         mtx_unlock(&lun->lun_lock);
8417
8418                         /* send msg to other side */
8419                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8420                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8421                         persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8422                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8423                         persis_io.pr.pr_info.res_type = type;
8424                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8425                             sizeof(persis_io.pr), M_WAITOK);
8426                 }
8427                 break;
8428
8429         case SPRO_RELEASE:
8430                 mtx_lock(&lun->lun_lock);
8431                 if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8432                         /* No reservation exists return good status */
8433                         mtx_unlock(&lun->lun_lock);
8434                         goto done;
8435                 }
8436                 /*
8437                  * Is this nexus a reservation holder?
8438                  */
8439                 if (lun->pr_res_idx != residx
8440                  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8441                         /*
8442                          * not a res holder return good status but
8443                          * do nothing
8444                          */
8445                         mtx_unlock(&lun->lun_lock);
8446                         goto done;
8447                 }
8448
8449                 if (lun->pr_res_type != type) {
8450                         mtx_unlock(&lun->lun_lock);
8451                         free(ctsio->kern_data_ptr, M_CTL);
8452                         ctl_set_illegal_pr_release(ctsio);
8453                         ctl_done((union ctl_io *)ctsio);
8454                         return (CTL_RETVAL_COMPLETE);
8455                 }
8456
8457                 /* okay to release */
8458                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8459                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8460                 lun->pr_res_type = 0;
8461
8462                 /*
8463                  * If this isn't an exclusive access reservation and NUAR
8464                  * is not set, generate UA for all other registrants.
8465                  */
8466                 if (type != SPR_TYPE_EX_AC && type != SPR_TYPE_WR_EX &&
8467                     (lun->MODE_CTRL.queue_flags & SCP_NUAR) == 0) {
8468                         for (i = softc->init_min; i < softc->init_max; i++) {
8469                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
8470                                         continue;
8471                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8472                         }
8473                 }
8474                 mtx_unlock(&lun->lun_lock);
8475
8476                 /* Send msg to other side */
8477                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8478                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8479                 persis_io.pr.pr_info.action = CTL_PR_RELEASE;
8480                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8481                      sizeof(persis_io.pr), M_WAITOK);
8482                 break;
8483
8484         case SPRO_CLEAR:
8485                 /* send msg to other side */
8486
8487                 mtx_lock(&lun->lun_lock);
8488                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8489                 lun->pr_res_type = 0;
8490                 lun->pr_key_count = 0;
8491                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8492
8493                 ctl_clr_prkey(lun, residx);
8494                 for (i = 0; i < CTL_MAX_INITIATORS; i++)
8495                         if (ctl_get_prkey(lun, i) != 0) {
8496                                 ctl_clr_prkey(lun, i);
8497                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8498                         }
8499                 lun->pr_generation++;
8500                 mtx_unlock(&lun->lun_lock);
8501
8502                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8503                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8504                 persis_io.pr.pr_info.action = CTL_PR_CLEAR;
8505                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8506                      sizeof(persis_io.pr), M_WAITOK);
8507                 break;
8508
8509         case SPRO_PREEMPT:
8510         case SPRO_PRE_ABO: {
8511                 int nretval;
8512
8513                 nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8514                                           residx, ctsio, cdb, param);
8515                 if (nretval != 0)
8516                         return (CTL_RETVAL_COMPLETE);
8517                 break;
8518         }
8519         default:
8520                 panic("%s: Invalid PR type %#x", __func__, cdb->action);
8521         }
8522
8523 done:
8524         free(ctsio->kern_data_ptr, M_CTL);
8525         ctl_set_success(ctsio);
8526         ctl_done((union ctl_io *)ctsio);
8527
8528         return (retval);
8529 }
8530
8531 /*
8532  * This routine is for handling a message from the other SC pertaining to
8533  * persistent reserve out. All the error checking will have been done
8534  * so only perorming the action need be done here to keep the two
8535  * in sync.
8536  */
8537 static void
8538 ctl_hndl_per_res_out_on_other_sc(union ctl_io *io)
8539 {
8540         struct ctl_softc *softc = CTL_SOFTC(io);
8541         union ctl_ha_msg *msg = (union ctl_ha_msg *)&io->presio.pr_msg;
8542         struct ctl_lun *lun;
8543         int i;
8544         uint32_t residx, targ_lun;
8545
8546         targ_lun = msg->hdr.nexus.targ_mapped_lun;
8547         mtx_lock(&softc->ctl_lock);
8548         if (targ_lun >= ctl_max_luns ||
8549             (lun = softc->ctl_luns[targ_lun]) == NULL) {
8550                 mtx_unlock(&softc->ctl_lock);
8551                 return;
8552         }
8553         mtx_lock(&lun->lun_lock);
8554         mtx_unlock(&softc->ctl_lock);
8555         if (lun->flags & CTL_LUN_DISABLED) {
8556                 mtx_unlock(&lun->lun_lock);
8557                 return;
8558         }
8559         residx = ctl_get_initindex(&msg->hdr.nexus);
8560         switch(msg->pr.pr_info.action) {
8561         case CTL_PR_REG_KEY:
8562                 ctl_alloc_prkey(lun, msg->pr.pr_info.residx);
8563                 if (ctl_get_prkey(lun, msg->pr.pr_info.residx) == 0)
8564                         lun->pr_key_count++;
8565                 ctl_set_prkey(lun, msg->pr.pr_info.residx,
8566                     scsi_8btou64(msg->pr.pr_info.sa_res_key));
8567                 lun->pr_generation++;
8568                 break;
8569
8570         case CTL_PR_UNREG_KEY:
8571                 ctl_clr_prkey(lun, msg->pr.pr_info.residx);
8572                 lun->pr_key_count--;
8573
8574                 /* XXX Need to see if the reservation has been released */
8575                 /* if so do we need to generate UA? */
8576                 if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8577                         lun->flags &= ~CTL_LUN_PR_RESERVED;
8578                         lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8579
8580                         if ((lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8581                              lun->pr_res_type == SPR_TYPE_EX_AC_RO) &&
8582                             lun->pr_key_count) {
8583                                 /*
8584                                  * If the reservation is a registrants
8585                                  * only type we need to generate a UA
8586                                  * for other registered inits.  The
8587                                  * sense code should be RESERVATIONS
8588                                  * RELEASED
8589                                  */
8590
8591                                 for (i = softc->init_min; i < softc->init_max; i++) {
8592                                         if (ctl_get_prkey(lun, i) == 0)
8593                                                 continue;
8594
8595                                         ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8596                                 }
8597                         }
8598                         lun->pr_res_type = 0;
8599                 } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8600                         if (lun->pr_key_count==0) {
8601                                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8602                                 lun->pr_res_type = 0;
8603                                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8604                         }
8605                 }
8606                 lun->pr_generation++;
8607                 break;
8608
8609         case CTL_PR_RESERVE:
8610                 lun->flags |= CTL_LUN_PR_RESERVED;
8611                 lun->pr_res_type = msg->pr.pr_info.res_type;
8612                 lun->pr_res_idx = msg->pr.pr_info.residx;
8613
8614                 break;
8615
8616         case CTL_PR_RELEASE:
8617                 /*
8618                  * If this isn't an exclusive access reservation and NUAR
8619                  * is not set, generate UA for all other registrants.
8620                  */
8621                 if (lun->pr_res_type != SPR_TYPE_EX_AC &&
8622                     lun->pr_res_type != SPR_TYPE_WR_EX &&
8623                     (lun->MODE_CTRL.queue_flags & SCP_NUAR) == 0) {
8624                         for (i = softc->init_min; i < softc->init_max; i++) {
8625                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
8626                                         continue;
8627                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8628                         }
8629                 }
8630
8631                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8632                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8633                 lun->pr_res_type = 0;
8634                 break;
8635
8636         case CTL_PR_PREEMPT:
8637                 ctl_pro_preempt_other(lun, msg);
8638                 break;
8639         case CTL_PR_CLEAR:
8640                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8641                 lun->pr_res_type = 0;
8642                 lun->pr_key_count = 0;
8643                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8644
8645                 for (i=0; i < CTL_MAX_INITIATORS; i++) {
8646                         if (ctl_get_prkey(lun, i) == 0)
8647                                 continue;
8648                         ctl_clr_prkey(lun, i);
8649                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8650                 }
8651                 lun->pr_generation++;
8652                 break;
8653         }
8654
8655         mtx_unlock(&lun->lun_lock);
8656 }
8657
8658 int
8659 ctl_read_write(struct ctl_scsiio *ctsio)
8660 {
8661         struct ctl_lun *lun = CTL_LUN(ctsio);
8662         struct ctl_lba_len_flags *lbalen;
8663         uint64_t lba;
8664         uint32_t num_blocks;
8665         int flags, retval;
8666         int isread;
8667
8668         CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
8669
8670         flags = 0;
8671         isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
8672               || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
8673         switch (ctsio->cdb[0]) {
8674         case READ_6:
8675         case WRITE_6: {
8676                 struct scsi_rw_6 *cdb;
8677
8678                 cdb = (struct scsi_rw_6 *)ctsio->cdb;
8679
8680                 lba = scsi_3btoul(cdb->addr);
8681                 /* only 5 bits are valid in the most significant address byte */
8682                 lba &= 0x1fffff;
8683                 num_blocks = cdb->length;
8684                 /*
8685                  * This is correct according to SBC-2.
8686                  */
8687                 if (num_blocks == 0)
8688                         num_blocks = 256;
8689                 break;
8690         }
8691         case READ_10:
8692         case WRITE_10: {
8693                 struct scsi_rw_10 *cdb;
8694
8695                 cdb = (struct scsi_rw_10 *)ctsio->cdb;
8696                 if (cdb->byte2 & SRW10_FUA)
8697                         flags |= CTL_LLF_FUA;
8698                 if (cdb->byte2 & SRW10_DPO)
8699                         flags |= CTL_LLF_DPO;
8700                 lba = scsi_4btoul(cdb->addr);
8701                 num_blocks = scsi_2btoul(cdb->length);
8702                 break;
8703         }
8704         case WRITE_VERIFY_10: {
8705                 struct scsi_write_verify_10 *cdb;
8706
8707                 cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
8708                 flags |= CTL_LLF_FUA;
8709                 if (cdb->byte2 & SWV_DPO)
8710                         flags |= CTL_LLF_DPO;
8711                 lba = scsi_4btoul(cdb->addr);
8712                 num_blocks = scsi_2btoul(cdb->length);
8713                 break;
8714         }
8715         case READ_12:
8716         case WRITE_12: {
8717                 struct scsi_rw_12 *cdb;
8718
8719                 cdb = (struct scsi_rw_12 *)ctsio->cdb;
8720                 if (cdb->byte2 & SRW12_FUA)
8721                         flags |= CTL_LLF_FUA;
8722                 if (cdb->byte2 & SRW12_DPO)
8723                         flags |= CTL_LLF_DPO;
8724                 lba = scsi_4btoul(cdb->addr);
8725                 num_blocks = scsi_4btoul(cdb->length);
8726                 break;
8727         }
8728         case WRITE_VERIFY_12: {
8729                 struct scsi_write_verify_12 *cdb;
8730
8731                 cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
8732                 flags |= CTL_LLF_FUA;
8733                 if (cdb->byte2 & SWV_DPO)
8734                         flags |= CTL_LLF_DPO;
8735                 lba = scsi_4btoul(cdb->addr);
8736                 num_blocks = scsi_4btoul(cdb->length);
8737                 break;
8738         }
8739         case READ_16:
8740         case WRITE_16: {
8741                 struct scsi_rw_16 *cdb;
8742
8743                 cdb = (struct scsi_rw_16 *)ctsio->cdb;
8744                 if (cdb->byte2 & SRW12_FUA)
8745                         flags |= CTL_LLF_FUA;
8746                 if (cdb->byte2 & SRW12_DPO)
8747                         flags |= CTL_LLF_DPO;
8748                 lba = scsi_8btou64(cdb->addr);
8749                 num_blocks = scsi_4btoul(cdb->length);
8750                 break;
8751         }
8752         case WRITE_ATOMIC_16: {
8753                 struct scsi_write_atomic_16 *cdb;
8754
8755                 if (lun->be_lun->atomicblock == 0) {
8756                         ctl_set_invalid_opcode(ctsio);
8757                         ctl_done((union ctl_io *)ctsio);
8758                         return (CTL_RETVAL_COMPLETE);
8759                 }
8760
8761                 cdb = (struct scsi_write_atomic_16 *)ctsio->cdb;
8762                 if (cdb->byte2 & SRW12_FUA)
8763                         flags |= CTL_LLF_FUA;
8764                 if (cdb->byte2 & SRW12_DPO)
8765                         flags |= CTL_LLF_DPO;
8766                 lba = scsi_8btou64(cdb->addr);
8767                 num_blocks = scsi_2btoul(cdb->length);
8768                 if (num_blocks > lun->be_lun->atomicblock) {
8769                         ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
8770                             /*command*/ 1, /*field*/ 12, /*bit_valid*/ 0,
8771                             /*bit*/ 0);
8772                         ctl_done((union ctl_io *)ctsio);
8773                         return (CTL_RETVAL_COMPLETE);
8774                 }
8775                 break;
8776         }
8777         case WRITE_VERIFY_16: {
8778                 struct scsi_write_verify_16 *cdb;
8779
8780                 cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
8781                 flags |= CTL_LLF_FUA;
8782                 if (cdb->byte2 & SWV_DPO)
8783                         flags |= CTL_LLF_DPO;
8784                 lba = scsi_8btou64(cdb->addr);
8785                 num_blocks = scsi_4btoul(cdb->length);
8786                 break;
8787         }
8788         default:
8789                 /*
8790                  * We got a command we don't support.  This shouldn't
8791                  * happen, commands should be filtered out above us.
8792                  */
8793                 ctl_set_invalid_opcode(ctsio);
8794                 ctl_done((union ctl_io *)ctsio);
8795
8796                 return (CTL_RETVAL_COMPLETE);
8797                 break; /* NOTREACHED */
8798         }
8799
8800         /*
8801          * The first check is to make sure we're in bounds, the second
8802          * check is to catch wrap-around problems.  If the lba + num blocks
8803          * is less than the lba, then we've wrapped around and the block
8804          * range is invalid anyway.
8805          */
8806         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8807          || ((lba + num_blocks) < lba)) {
8808                 ctl_set_lba_out_of_range(ctsio,
8809                     MAX(lba, lun->be_lun->maxlba + 1));
8810                 ctl_done((union ctl_io *)ctsio);
8811                 return (CTL_RETVAL_COMPLETE);
8812         }
8813
8814         /*
8815          * According to SBC-3, a transfer length of 0 is not an error.
8816          * Note that this cannot happen with WRITE(6) or READ(6), since 0
8817          * translates to 256 blocks for those commands.
8818          */
8819         if (num_blocks == 0) {
8820                 ctl_set_success(ctsio);
8821                 ctl_done((union ctl_io *)ctsio);
8822                 return (CTL_RETVAL_COMPLETE);
8823         }
8824
8825         /* Set FUA and/or DPO if caches are disabled. */
8826         if (isread) {
8827                 if ((lun->MODE_CACHING.flags1 & SCP_RCD) != 0)
8828                         flags |= CTL_LLF_FUA | CTL_LLF_DPO;
8829         } else {
8830                 if ((lun->MODE_CACHING.flags1 & SCP_WCE) == 0)
8831                         flags |= CTL_LLF_FUA;
8832         }
8833
8834         lbalen = (struct ctl_lba_len_flags *)
8835             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8836         lbalen->lba = lba;
8837         lbalen->len = num_blocks;
8838         lbalen->flags = (isread ? CTL_LLF_READ : CTL_LLF_WRITE) | flags;
8839
8840         ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
8841         ctsio->kern_rel_offset = 0;
8842
8843         CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
8844
8845         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8846         return (retval);
8847 }
8848
8849 static int
8850 ctl_cnw_cont(union ctl_io *io)
8851 {
8852         struct ctl_lun *lun = CTL_LUN(io);
8853         struct ctl_scsiio *ctsio;
8854         struct ctl_lba_len_flags *lbalen;
8855         int retval;
8856
8857         ctsio = &io->scsiio;
8858         ctsio->io_hdr.status = CTL_STATUS_NONE;
8859         ctsio->io_hdr.flags &= ~CTL_FLAG_IO_CONT;
8860         lbalen = (struct ctl_lba_len_flags *)
8861             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8862         lbalen->flags &= ~CTL_LLF_COMPARE;
8863         lbalen->flags |= CTL_LLF_WRITE;
8864
8865         CTL_DEBUG_PRINT(("ctl_cnw_cont: calling data_submit()\n"));
8866         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8867         return (retval);
8868 }
8869
8870 int
8871 ctl_cnw(struct ctl_scsiio *ctsio)
8872 {
8873         struct ctl_lun *lun = CTL_LUN(ctsio);
8874         struct ctl_lba_len_flags *lbalen;
8875         uint64_t lba;
8876         uint32_t num_blocks;
8877         int flags, retval;
8878
8879         CTL_DEBUG_PRINT(("ctl_cnw: command: %#x\n", ctsio->cdb[0]));
8880
8881         flags = 0;
8882         switch (ctsio->cdb[0]) {
8883         case COMPARE_AND_WRITE: {
8884                 struct scsi_compare_and_write *cdb;
8885
8886                 cdb = (struct scsi_compare_and_write *)ctsio->cdb;
8887                 if (cdb->byte2 & SRW10_FUA)
8888                         flags |= CTL_LLF_FUA;
8889                 if (cdb->byte2 & SRW10_DPO)
8890                         flags |= CTL_LLF_DPO;
8891                 lba = scsi_8btou64(cdb->addr);
8892                 num_blocks = cdb->length;
8893                 break;
8894         }
8895         default:
8896                 /*
8897                  * We got a command we don't support.  This shouldn't
8898                  * happen, commands should be filtered out above us.
8899                  */
8900                 ctl_set_invalid_opcode(ctsio);
8901                 ctl_done((union ctl_io *)ctsio);
8902
8903                 return (CTL_RETVAL_COMPLETE);
8904                 break; /* NOTREACHED */
8905         }
8906
8907         /*
8908          * The first check is to make sure we're in bounds, the second
8909          * check is to catch wrap-around problems.  If the lba + num blocks
8910          * is less than the lba, then we've wrapped around and the block
8911          * range is invalid anyway.
8912          */
8913         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8914          || ((lba + num_blocks) < lba)) {
8915                 ctl_set_lba_out_of_range(ctsio,
8916                     MAX(lba, lun->be_lun->maxlba + 1));
8917                 ctl_done((union ctl_io *)ctsio);
8918                 return (CTL_RETVAL_COMPLETE);
8919         }
8920
8921         /*
8922          * According to SBC-3, a transfer length of 0 is not an error.
8923          */
8924         if (num_blocks == 0) {
8925                 ctl_set_success(ctsio);
8926                 ctl_done((union ctl_io *)ctsio);
8927                 return (CTL_RETVAL_COMPLETE);
8928         }
8929
8930         /* Set FUA if write cache is disabled. */
8931         if ((lun->MODE_CACHING.flags1 & SCP_WCE) == 0)
8932                 flags |= CTL_LLF_FUA;
8933
8934         ctsio->kern_total_len = 2 * num_blocks * lun->be_lun->blocksize;
8935         ctsio->kern_rel_offset = 0;
8936
8937         /*
8938          * Set the IO_CONT flag, so that if this I/O gets passed to
8939          * ctl_data_submit_done(), it'll get passed back to
8940          * ctl_ctl_cnw_cont() for further processing.
8941          */
8942         ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
8943         ctsio->io_cont = ctl_cnw_cont;
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 = CTL_LLF_COMPARE | flags;
8950
8951         CTL_DEBUG_PRINT(("ctl_cnw: calling data_submit()\n"));
8952         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8953         return (retval);
8954 }
8955
8956 int
8957 ctl_verify(struct ctl_scsiio *ctsio)
8958 {
8959         struct ctl_lun *lun = CTL_LUN(ctsio);
8960         struct ctl_lba_len_flags *lbalen;
8961         uint64_t lba;
8962         uint32_t num_blocks;
8963         int bytchk, flags;
8964         int retval;
8965
8966         CTL_DEBUG_PRINT(("ctl_verify: command: %#x\n", ctsio->cdb[0]));
8967
8968         bytchk = 0;
8969         flags = CTL_LLF_FUA;
8970         switch (ctsio->cdb[0]) {
8971         case VERIFY_10: {
8972                 struct scsi_verify_10 *cdb;
8973
8974                 cdb = (struct scsi_verify_10 *)ctsio->cdb;
8975                 if (cdb->byte2 & SVFY_BYTCHK)
8976                         bytchk = 1;
8977                 if (cdb->byte2 & SVFY_DPO)
8978                         flags |= CTL_LLF_DPO;
8979                 lba = scsi_4btoul(cdb->addr);
8980                 num_blocks = scsi_2btoul(cdb->length);
8981                 break;
8982         }
8983         case VERIFY_12: {
8984                 struct scsi_verify_12 *cdb;
8985
8986                 cdb = (struct scsi_verify_12 *)ctsio->cdb;
8987                 if (cdb->byte2 & SVFY_BYTCHK)
8988                         bytchk = 1;
8989                 if (cdb->byte2 & SVFY_DPO)
8990                         flags |= CTL_LLF_DPO;
8991                 lba = scsi_4btoul(cdb->addr);
8992                 num_blocks = scsi_4btoul(cdb->length);
8993                 break;
8994         }
8995         case VERIFY_16: {
8996                 struct scsi_rw_16 *cdb;
8997
8998                 cdb = (struct scsi_rw_16 *)ctsio->cdb;
8999                 if (cdb->byte2 & SVFY_BYTCHK)
9000                         bytchk = 1;
9001                 if (cdb->byte2 & SVFY_DPO)
9002                         flags |= CTL_LLF_DPO;
9003                 lba = scsi_8btou64(cdb->addr);
9004                 num_blocks = scsi_4btoul(cdb->length);
9005                 break;
9006         }
9007         default:
9008                 /*
9009                  * We got a command we don't support.  This shouldn't
9010                  * happen, commands should be filtered out above us.
9011                  */
9012                 ctl_set_invalid_opcode(ctsio);
9013                 ctl_done((union ctl_io *)ctsio);
9014                 return (CTL_RETVAL_COMPLETE);
9015         }
9016
9017         /*
9018          * The first check is to make sure we're in bounds, the second
9019          * check is to catch wrap-around problems.  If the lba + num blocks
9020          * is less than the lba, then we've wrapped around and the block
9021          * range is invalid anyway.
9022          */
9023         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9024          || ((lba + num_blocks) < lba)) {
9025                 ctl_set_lba_out_of_range(ctsio,
9026                     MAX(lba, lun->be_lun->maxlba + 1));
9027                 ctl_done((union ctl_io *)ctsio);
9028                 return (CTL_RETVAL_COMPLETE);
9029         }
9030
9031         /*
9032          * According to SBC-3, a transfer length of 0 is not an error.
9033          */
9034         if (num_blocks == 0) {
9035                 ctl_set_success(ctsio);
9036                 ctl_done((union ctl_io *)ctsio);
9037                 return (CTL_RETVAL_COMPLETE);
9038         }
9039
9040         lbalen = (struct ctl_lba_len_flags *)
9041             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9042         lbalen->lba = lba;
9043         lbalen->len = num_blocks;
9044         if (bytchk) {
9045                 lbalen->flags = CTL_LLF_COMPARE | flags;
9046                 ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9047         } else {
9048                 lbalen->flags = CTL_LLF_VERIFY | flags;
9049                 ctsio->kern_total_len = 0;
9050         }
9051         ctsio->kern_rel_offset = 0;
9052
9053         CTL_DEBUG_PRINT(("ctl_verify: calling data_submit()\n"));
9054         retval = lun->backend->data_submit((union ctl_io *)ctsio);
9055         return (retval);
9056 }
9057
9058 int
9059 ctl_report_luns(struct ctl_scsiio *ctsio)
9060 {
9061         struct ctl_softc *softc = CTL_SOFTC(ctsio);
9062         struct ctl_port *port = CTL_PORT(ctsio);
9063         struct ctl_lun *lun, *request_lun = CTL_LUN(ctsio);
9064         struct scsi_report_luns *cdb;
9065         struct scsi_report_luns_data *lun_data;
9066         int num_filled, num_luns, num_port_luns, retval;
9067         uint32_t alloc_len, lun_datalen;
9068         uint32_t initidx, targ_lun_id, lun_id;
9069
9070         retval = CTL_RETVAL_COMPLETE;
9071         cdb = (struct scsi_report_luns *)ctsio->cdb;
9072
9073         CTL_DEBUG_PRINT(("ctl_report_luns\n"));
9074
9075         num_luns = 0;
9076         num_port_luns = port->lun_map ? port->lun_map_size : ctl_max_luns;
9077         mtx_lock(&softc->ctl_lock);
9078         for (targ_lun_id = 0; targ_lun_id < num_port_luns; targ_lun_id++) {
9079                 if (ctl_lun_map_from_port(port, targ_lun_id) != UINT32_MAX)
9080                         num_luns++;
9081         }
9082         mtx_unlock(&softc->ctl_lock);
9083
9084         switch (cdb->select_report) {
9085         case RPL_REPORT_DEFAULT:
9086         case RPL_REPORT_ALL:
9087         case RPL_REPORT_NONSUBSID:
9088                 break;
9089         case RPL_REPORT_WELLKNOWN:
9090         case RPL_REPORT_ADMIN:
9091         case RPL_REPORT_CONGLOM:
9092                 num_luns = 0;
9093                 break;
9094         default:
9095                 ctl_set_invalid_field(ctsio,
9096                                       /*sks_valid*/ 1,
9097                                       /*command*/ 1,
9098                                       /*field*/ 2,
9099                                       /*bit_valid*/ 0,
9100                                       /*bit*/ 0);
9101                 ctl_done((union ctl_io *)ctsio);
9102                 return (retval);
9103                 break; /* NOTREACHED */
9104         }
9105
9106         alloc_len = scsi_4btoul(cdb->length);
9107         /*
9108          * The initiator has to allocate at least 16 bytes for this request,
9109          * so he can at least get the header and the first LUN.  Otherwise
9110          * we reject the request (per SPC-3 rev 14, section 6.21).
9111          */
9112         if (alloc_len < (sizeof(struct scsi_report_luns_data) +
9113             sizeof(struct scsi_report_luns_lundata))) {
9114                 ctl_set_invalid_field(ctsio,
9115                                       /*sks_valid*/ 1,
9116                                       /*command*/ 1,
9117                                       /*field*/ 6,
9118                                       /*bit_valid*/ 0,
9119                                       /*bit*/ 0);
9120                 ctl_done((union ctl_io *)ctsio);
9121                 return (retval);
9122         }
9123
9124         lun_datalen = sizeof(*lun_data) +
9125                 (num_luns * sizeof(struct scsi_report_luns_lundata));
9126
9127         ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO);
9128         lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
9129         ctsio->kern_sg_entries = 0;
9130
9131         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9132
9133         mtx_lock(&softc->ctl_lock);
9134         for (targ_lun_id = 0, num_filled = 0;
9135             targ_lun_id < num_port_luns && num_filled < num_luns;
9136             targ_lun_id++) {
9137                 lun_id = ctl_lun_map_from_port(port, targ_lun_id);
9138                 if (lun_id == UINT32_MAX)
9139                         continue;
9140                 lun = softc->ctl_luns[lun_id];
9141                 if (lun == NULL)
9142                         continue;
9143
9144                 be64enc(lun_data->luns[num_filled++].lundata,
9145                     ctl_encode_lun(targ_lun_id));
9146
9147                 /*
9148                  * According to SPC-3, rev 14 section 6.21:
9149                  *
9150                  * "The execution of a REPORT LUNS command to any valid and
9151                  * installed logical unit shall clear the REPORTED LUNS DATA
9152                  * HAS CHANGED unit attention condition for all logical
9153                  * units of that target with respect to the requesting
9154                  * initiator. A valid and installed logical unit is one
9155                  * having a PERIPHERAL QUALIFIER of 000b in the standard
9156                  * INQUIRY data (see 6.4.2)."
9157                  *
9158                  * If request_lun is NULL, the LUN this report luns command
9159                  * was issued to is either disabled or doesn't exist. In that
9160                  * case, we shouldn't clear any pending lun change unit
9161                  * attention.
9162                  */
9163                 if (request_lun != NULL) {
9164                         mtx_lock(&lun->lun_lock);
9165                         ctl_clr_ua(lun, initidx, CTL_UA_LUN_CHANGE);
9166                         mtx_unlock(&lun->lun_lock);
9167                 }
9168         }
9169         mtx_unlock(&softc->ctl_lock);
9170
9171         /*
9172          * It's quite possible that we've returned fewer LUNs than we allocated
9173          * space for.  Trim it.
9174          */
9175         lun_datalen = sizeof(*lun_data) +
9176                 (num_filled * sizeof(struct scsi_report_luns_lundata));
9177         ctsio->kern_rel_offset = 0;
9178         ctsio->kern_sg_entries = 0;
9179         ctsio->kern_data_len = min(lun_datalen, alloc_len);
9180         ctsio->kern_total_len = ctsio->kern_data_len;
9181
9182         /*
9183          * We set this to the actual data length, regardless of how much
9184          * space we actually have to return results.  If the user looks at
9185          * this value, he'll know whether or not he allocated enough space
9186          * and reissue the command if necessary.  We don't support well
9187          * known logical units, so if the user asks for that, return none.
9188          */
9189         scsi_ulto4b(lun_datalen - 8, lun_data->length);
9190
9191         /*
9192          * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
9193          * this request.
9194          */
9195         ctl_set_success(ctsio);
9196         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9197         ctsio->be_move_done = ctl_config_move_done;
9198         ctl_datamove((union ctl_io *)ctsio);
9199         return (retval);
9200 }
9201
9202 int
9203 ctl_request_sense(struct ctl_scsiio *ctsio)
9204 {
9205         struct ctl_softc *softc = CTL_SOFTC(ctsio);
9206         struct ctl_lun *lun = CTL_LUN(ctsio);
9207         struct scsi_request_sense *cdb;
9208         struct scsi_sense_data *sense_ptr, *ps;
9209         uint32_t initidx;
9210         int have_error;
9211         u_int sense_len = SSD_FULL_SIZE;
9212         scsi_sense_data_type sense_format;
9213         ctl_ua_type ua_type;
9214         uint8_t asc = 0, ascq = 0;
9215
9216         cdb = (struct scsi_request_sense *)ctsio->cdb;
9217
9218         CTL_DEBUG_PRINT(("ctl_request_sense\n"));
9219
9220         /*
9221          * Determine which sense format the user wants.
9222          */
9223         if (cdb->byte2 & SRS_DESC)
9224                 sense_format = SSD_TYPE_DESC;
9225         else
9226                 sense_format = SSD_TYPE_FIXED;
9227
9228         ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
9229         sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
9230         ctsio->kern_sg_entries = 0;
9231         ctsio->kern_rel_offset = 0;
9232
9233         /*
9234          * struct scsi_sense_data, which is currently set to 256 bytes, is
9235          * larger than the largest allowed value for the length field in the
9236          * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
9237          */
9238         ctsio->kern_data_len = cdb->length;
9239         ctsio->kern_total_len = cdb->length;
9240
9241         /*
9242          * If we don't have a LUN, we don't have any pending sense.
9243          */
9244         if (lun == NULL ||
9245             ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
9246              softc->ha_link < CTL_HA_LINK_UNKNOWN)) {
9247                 /* "Logical unit not supported" */
9248                 ctl_set_sense_data(sense_ptr, &sense_len, NULL, sense_format,
9249                     /*current_error*/ 1,
9250                     /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
9251                     /*asc*/ 0x25,
9252                     /*ascq*/ 0x00,
9253                     SSD_ELEM_NONE);
9254                 goto send;
9255         }
9256
9257         have_error = 0;
9258         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9259         /*
9260          * Check for pending sense, and then for pending unit attentions.
9261          * Pending sense gets returned first, then pending unit attentions.
9262          */
9263         mtx_lock(&lun->lun_lock);
9264         ps = lun->pending_sense[initidx / CTL_MAX_INIT_PER_PORT];
9265         if (ps != NULL)
9266                 ps += initidx % CTL_MAX_INIT_PER_PORT;
9267         if (ps != NULL && ps->error_code != 0) {
9268                 scsi_sense_data_type stored_format;
9269
9270                 /*
9271                  * Check to see which sense format was used for the stored
9272                  * sense data.
9273                  */
9274                 stored_format = scsi_sense_type(ps);
9275
9276                 /*
9277                  * If the user requested a different sense format than the
9278                  * one we stored, then we need to convert it to the other
9279                  * format.  If we're going from descriptor to fixed format
9280                  * sense data, we may lose things in translation, depending
9281                  * on what options were used.
9282                  *
9283                  * If the stored format is SSD_TYPE_NONE (i.e. invalid),
9284                  * for some reason we'll just copy it out as-is.
9285                  */
9286                 if ((stored_format == SSD_TYPE_FIXED)
9287                  && (sense_format == SSD_TYPE_DESC))
9288                         ctl_sense_to_desc((struct scsi_sense_data_fixed *)
9289                             ps, (struct scsi_sense_data_desc *)sense_ptr);
9290                 else if ((stored_format == SSD_TYPE_DESC)
9291                       && (sense_format == SSD_TYPE_FIXED))
9292                         ctl_sense_to_fixed((struct scsi_sense_data_desc *)
9293                             ps, (struct scsi_sense_data_fixed *)sense_ptr);
9294                 else
9295                         memcpy(sense_ptr, ps, sizeof(*sense_ptr));
9296
9297                 ps->error_code = 0;
9298                 have_error = 1;
9299         } else {
9300                 ua_type = ctl_build_ua(lun, initidx, sense_ptr, &sense_len,
9301                     sense_format);
9302                 if (ua_type != CTL_UA_NONE)
9303                         have_error = 1;
9304         }
9305         if (have_error == 0) {
9306                 /*
9307                  * Report informational exception if have one and allowed.
9308                  */
9309                 if (lun->MODE_IE.mrie != SIEP_MRIE_NO) {
9310                         asc = lun->ie_asc;
9311                         ascq = lun->ie_ascq;
9312                 }
9313                 ctl_set_sense_data(sense_ptr, &sense_len, lun, sense_format,
9314                     /*current_error*/ 1,
9315                     /*sense_key*/ SSD_KEY_NO_SENSE,
9316                     /*asc*/ asc,
9317                     /*ascq*/ ascq,
9318                     SSD_ELEM_NONE);
9319         }
9320         mtx_unlock(&lun->lun_lock);
9321
9322 send:
9323         /*
9324          * We report the SCSI status as OK, since the status of the command
9325          * itself is OK.  We're reporting sense as parameter data.
9326          */
9327         ctl_set_success(ctsio);
9328         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9329         ctsio->be_move_done = ctl_config_move_done;
9330         ctl_datamove((union ctl_io *)ctsio);
9331         return (CTL_RETVAL_COMPLETE);
9332 }
9333
9334 int
9335 ctl_tur(struct ctl_scsiio *ctsio)
9336 {
9337
9338         CTL_DEBUG_PRINT(("ctl_tur\n"));
9339
9340         ctl_set_success(ctsio);
9341         ctl_done((union ctl_io *)ctsio);
9342
9343         return (CTL_RETVAL_COMPLETE);
9344 }
9345
9346 /*
9347  * SCSI VPD page 0x00, the Supported VPD Pages page.
9348  */
9349 static int
9350 ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
9351 {
9352         struct ctl_lun *lun = CTL_LUN(ctsio);
9353         struct scsi_vpd_supported_pages *pages;
9354         int sup_page_size;
9355         int p;
9356
9357         sup_page_size = sizeof(struct scsi_vpd_supported_pages) *
9358             SCSI_EVPD_NUM_SUPPORTED_PAGES;
9359         ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO);
9360         pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
9361         ctsio->kern_rel_offset = 0;
9362         ctsio->kern_sg_entries = 0;
9363         ctsio->kern_data_len = min(sup_page_size, alloc_len);
9364         ctsio->kern_total_len = ctsio->kern_data_len;
9365
9366         /*
9367          * The control device is always connected.  The disk device, on the
9368          * other hand, may not be online all the time.  Need to change this
9369          * to figure out whether the disk device is actually online or not.
9370          */
9371         if (lun != NULL)
9372                 pages->device = (SID_QUAL_LU_CONNECTED << 5) |
9373                                 lun->be_lun->lun_type;
9374         else
9375                 pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9376
9377         p = 0;
9378         /* Supported VPD pages */
9379         pages->page_list[p++] = SVPD_SUPPORTED_PAGES;
9380         /* Serial Number */
9381         pages->page_list[p++] = SVPD_UNIT_SERIAL_NUMBER;
9382         /* Device Identification */
9383         pages->page_list[p++] = SVPD_DEVICE_ID;
9384         /* Extended INQUIRY Data */
9385         pages->page_list[p++] = SVPD_EXTENDED_INQUIRY_DATA;
9386         /* Mode Page Policy */
9387         pages->page_list[p++] = SVPD_MODE_PAGE_POLICY;
9388         /* SCSI Ports */
9389         pages->page_list[p++] = SVPD_SCSI_PORTS;
9390         /* Third-party Copy */
9391         pages->page_list[p++] = SVPD_SCSI_TPC;
9392         if (lun != NULL && lun->be_lun->lun_type == T_DIRECT) {
9393                 /* Block limits */
9394                 pages->page_list[p++] = SVPD_BLOCK_LIMITS;
9395                 /* Block Device Characteristics */
9396                 pages->page_list[p++] = SVPD_BDC;
9397                 /* Logical Block Provisioning */
9398                 pages->page_list[p++] = SVPD_LBP;
9399         }
9400         pages->length = p;
9401
9402         ctl_set_success(ctsio);
9403         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9404         ctsio->be_move_done = ctl_config_move_done;
9405         ctl_datamove((union ctl_io *)ctsio);
9406         return (CTL_RETVAL_COMPLETE);
9407 }
9408
9409 /*
9410  * SCSI VPD page 0x80, the Unit Serial Number page.
9411  */
9412 static int
9413 ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
9414 {
9415         struct ctl_lun *lun = CTL_LUN(ctsio);
9416         struct scsi_vpd_unit_serial_number *sn_ptr;
9417         int data_len;
9418
9419         data_len = 4 + CTL_SN_LEN;
9420         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9421         sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
9422         ctsio->kern_rel_offset = 0;
9423         ctsio->kern_sg_entries = 0;
9424         ctsio->kern_data_len = min(data_len, alloc_len);
9425         ctsio->kern_total_len = ctsio->kern_data_len;
9426
9427         /*
9428          * The control device is always connected.  The disk device, on the
9429          * other hand, may not be online all the time.  Need to change this
9430          * to figure out whether the disk device is actually online or not.
9431          */
9432         if (lun != NULL)
9433                 sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9434                                   lun->be_lun->lun_type;
9435         else
9436                 sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9437
9438         sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9439         sn_ptr->length = CTL_SN_LEN;
9440         /*
9441          * If we don't have a LUN, we just leave the serial number as
9442          * all spaces.
9443          */
9444         if (lun != NULL) {
9445                 strncpy((char *)sn_ptr->serial_num,
9446                         (char *)lun->be_lun->serial_num, CTL_SN_LEN);
9447         } else
9448                 memset(sn_ptr->serial_num, 0x20, CTL_SN_LEN);
9449
9450         ctl_set_success(ctsio);
9451         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9452         ctsio->be_move_done = ctl_config_move_done;
9453         ctl_datamove((union ctl_io *)ctsio);
9454         return (CTL_RETVAL_COMPLETE);
9455 }
9456
9457
9458 /*
9459  * SCSI VPD page 0x86, the Extended INQUIRY Data page.
9460  */
9461 static int
9462 ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len)
9463 {
9464         struct ctl_lun *lun = CTL_LUN(ctsio);
9465         struct scsi_vpd_extended_inquiry_data *eid_ptr;
9466         int data_len;
9467
9468         data_len = sizeof(struct scsi_vpd_extended_inquiry_data);
9469         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9470         eid_ptr = (struct scsi_vpd_extended_inquiry_data *)ctsio->kern_data_ptr;
9471         ctsio->kern_sg_entries = 0;
9472         ctsio->kern_rel_offset = 0;
9473         ctsio->kern_data_len = min(data_len, alloc_len);
9474         ctsio->kern_total_len = ctsio->kern_data_len;
9475
9476         /*
9477          * The control device is always connected.  The disk device, on the
9478          * other hand, may not be online all the time.
9479          */
9480         if (lun != NULL)
9481                 eid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9482                                      lun->be_lun->lun_type;
9483         else
9484                 eid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9485         eid_ptr->page_code = SVPD_EXTENDED_INQUIRY_DATA;
9486         scsi_ulto2b(data_len - 4, eid_ptr->page_length);
9487         /*
9488          * We support head of queue, ordered and simple tags.
9489          */
9490         eid_ptr->flags2 = SVPD_EID_HEADSUP | SVPD_EID_ORDSUP | SVPD_EID_SIMPSUP;
9491         /*
9492          * Volatile cache supported.
9493          */
9494         eid_ptr->flags3 = SVPD_EID_V_SUP;
9495
9496         /*
9497          * This means that we clear the REPORTED LUNS DATA HAS CHANGED unit
9498          * attention for a particular IT nexus on all LUNs once we report
9499          * it to that nexus once.  This bit is required as of SPC-4.
9500          */
9501         eid_ptr->flags4 = SVPD_EID_LUICLR;
9502
9503         /*
9504          * We support revert to defaults (RTD) bit in MODE SELECT.
9505          */
9506         eid_ptr->flags5 = SVPD_EID_RTD_SUP;
9507
9508         /*
9509          * XXX KDM in order to correctly answer this, we would need
9510          * information from the SIM to determine how much sense data it
9511          * can send.  So this would really be a path inquiry field, most
9512          * likely.  This can be set to a maximum of 252 according to SPC-4,
9513          * but the hardware may or may not be able to support that much.
9514          * 0 just means that the maximum sense data length is not reported.
9515          */
9516         eid_ptr->max_sense_length = 0;
9517
9518         ctl_set_success(ctsio);
9519         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9520         ctsio->be_move_done = ctl_config_move_done;
9521         ctl_datamove((union ctl_io *)ctsio);
9522         return (CTL_RETVAL_COMPLETE);
9523 }
9524
9525 static int
9526 ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len)
9527 {
9528         struct ctl_lun *lun = CTL_LUN(ctsio);
9529         struct scsi_vpd_mode_page_policy *mpp_ptr;
9530         int data_len;
9531
9532         data_len = sizeof(struct scsi_vpd_mode_page_policy) +
9533             sizeof(struct scsi_vpd_mode_page_policy_descr);
9534
9535         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9536         mpp_ptr = (struct scsi_vpd_mode_page_policy *)ctsio->kern_data_ptr;
9537         ctsio->kern_rel_offset = 0;
9538         ctsio->kern_sg_entries = 0;
9539         ctsio->kern_data_len = min(data_len, alloc_len);
9540         ctsio->kern_total_len = ctsio->kern_data_len;
9541
9542         /*
9543          * The control device is always connected.  The disk device, on the
9544          * other hand, may not be online all the time.
9545          */
9546         if (lun != NULL)
9547                 mpp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9548                                      lun->be_lun->lun_type;
9549         else
9550                 mpp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9551         mpp_ptr->page_code = SVPD_MODE_PAGE_POLICY;
9552         scsi_ulto2b(data_len - 4, mpp_ptr->page_length);
9553         mpp_ptr->descr[0].page_code = 0x3f;
9554         mpp_ptr->descr[0].subpage_code = 0xff;
9555         mpp_ptr->descr[0].policy = SVPD_MPP_SHARED;
9556
9557         ctl_set_success(ctsio);
9558         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9559         ctsio->be_move_done = ctl_config_move_done;
9560         ctl_datamove((union ctl_io *)ctsio);
9561         return (CTL_RETVAL_COMPLETE);
9562 }
9563
9564 /*
9565  * SCSI VPD page 0x83, the Device Identification page.
9566  */
9567 static int
9568 ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
9569 {
9570         struct ctl_softc *softc = CTL_SOFTC(ctsio);
9571         struct ctl_port *port = CTL_PORT(ctsio);
9572         struct ctl_lun *lun = CTL_LUN(ctsio);
9573         struct scsi_vpd_device_id *devid_ptr;
9574         struct scsi_vpd_id_descriptor *desc;
9575         int data_len, g;
9576         uint8_t proto;
9577
9578         data_len = sizeof(struct scsi_vpd_device_id) +
9579             sizeof(struct scsi_vpd_id_descriptor) +
9580                 sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
9581             sizeof(struct scsi_vpd_id_descriptor) +
9582                 sizeof(struct scsi_vpd_id_trgt_port_grp_id);
9583         if (lun && lun->lun_devid)
9584                 data_len += lun->lun_devid->len;
9585         if (port && port->port_devid)
9586                 data_len += port->port_devid->len;
9587         if (port && port->target_devid)
9588                 data_len += port->target_devid->len;
9589
9590         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9591         devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
9592         ctsio->kern_sg_entries = 0;
9593         ctsio->kern_rel_offset = 0;
9594         ctsio->kern_sg_entries = 0;
9595         ctsio->kern_data_len = min(data_len, alloc_len);
9596         ctsio->kern_total_len = ctsio->kern_data_len;
9597
9598         /*
9599          * The control device is always connected.  The disk device, on the
9600          * other hand, may not be online all the time.
9601          */
9602         if (lun != NULL)
9603                 devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9604                                      lun->be_lun->lun_type;
9605         else
9606                 devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9607         devid_ptr->page_code = SVPD_DEVICE_ID;
9608         scsi_ulto2b(data_len - 4, devid_ptr->length);
9609
9610         if (port && port->port_type == CTL_PORT_FC)
9611                 proto = SCSI_PROTO_FC << 4;
9612         else if (port && port->port_type == CTL_PORT_SAS)
9613                 proto = SCSI_PROTO_SAS << 4;
9614         else if (port && port->port_type == CTL_PORT_ISCSI)
9615                 proto = SCSI_PROTO_ISCSI << 4;
9616         else
9617                 proto = SCSI_PROTO_SPI << 4;
9618         desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
9619
9620         /*
9621          * We're using a LUN association here.  i.e., this device ID is a
9622          * per-LUN identifier.
9623          */
9624         if (lun && lun->lun_devid) {
9625                 memcpy(desc, lun->lun_devid->data, lun->lun_devid->len);
9626                 desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9627                     lun->lun_devid->len);
9628         }
9629
9630         /*
9631          * This is for the WWPN which is a port association.
9632          */
9633         if (port && port->port_devid) {
9634                 memcpy(desc, port->port_devid->data, port->port_devid->len);
9635                 desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9636                     port->port_devid->len);
9637         }
9638
9639         /*
9640          * This is for the Relative Target Port(type 4h) identifier
9641          */
9642         desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9643         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9644             SVPD_ID_TYPE_RELTARG;
9645         desc->length = 4;
9646         scsi_ulto2b(ctsio->io_hdr.nexus.targ_port, &desc->identifier[2]);
9647         desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9648             sizeof(struct scsi_vpd_id_rel_trgt_port_id));
9649
9650         /*
9651          * This is for the Target Port Group(type 5h) identifier
9652          */
9653         desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9654         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9655             SVPD_ID_TYPE_TPORTGRP;
9656         desc->length = 4;
9657         if (softc->is_single ||
9658             (port && port->status & CTL_PORT_STATUS_HA_SHARED))
9659                 g = 1;
9660         else
9661                 g = 2 + ctsio->io_hdr.nexus.targ_port / softc->port_cnt;
9662         scsi_ulto2b(g, &desc->identifier[2]);
9663         desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9664             sizeof(struct scsi_vpd_id_trgt_port_grp_id));
9665
9666         /*
9667          * This is for the Target identifier
9668          */
9669         if (port && port->target_devid) {
9670                 memcpy(desc, port->target_devid->data, port->target_devid->len);
9671         }
9672
9673         ctl_set_success(ctsio);
9674         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9675         ctsio->be_move_done = ctl_config_move_done;
9676         ctl_datamove((union ctl_io *)ctsio);
9677         return (CTL_RETVAL_COMPLETE);
9678 }
9679
9680 static int
9681 ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio, int alloc_len)
9682 {
9683         struct ctl_softc *softc = CTL_SOFTC(ctsio);
9684         struct ctl_lun *lun = CTL_LUN(ctsio);
9685         struct scsi_vpd_scsi_ports *sp;
9686         struct scsi_vpd_port_designation *pd;
9687         struct scsi_vpd_port_designation_cont *pdc;
9688         struct ctl_port *port;
9689         int data_len, num_target_ports, iid_len, id_len;
9690
9691         num_target_ports = 0;
9692         iid_len = 0;
9693         id_len = 0;
9694         mtx_lock(&softc->ctl_lock);
9695         STAILQ_FOREACH(port, &softc->port_list, links) {
9696                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
9697                         continue;
9698                 if (lun != NULL &&
9699                     ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
9700                         continue;
9701                 num_target_ports++;
9702                 if (port->init_devid)
9703                         iid_len += port->init_devid->len;
9704                 if (port->port_devid)
9705                         id_len += port->port_devid->len;
9706         }
9707         mtx_unlock(&softc->ctl_lock);
9708
9709         data_len = sizeof(struct scsi_vpd_scsi_ports) +
9710             num_target_ports * (sizeof(struct scsi_vpd_port_designation) +
9711              sizeof(struct scsi_vpd_port_designation_cont)) + iid_len + id_len;
9712         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9713         sp = (struct scsi_vpd_scsi_ports *)ctsio->kern_data_ptr;
9714         ctsio->kern_sg_entries = 0;
9715         ctsio->kern_rel_offset = 0;
9716         ctsio->kern_sg_entries = 0;
9717         ctsio->kern_data_len = min(data_len, alloc_len);
9718         ctsio->kern_total_len = ctsio->kern_data_len;
9719
9720         /*
9721          * The control device is always connected.  The disk device, on the
9722          * other hand, may not be online all the time.  Need to change this
9723          * to figure out whether the disk device is actually online or not.
9724          */
9725         if (lun != NULL)
9726                 sp->device = (SID_QUAL_LU_CONNECTED << 5) |
9727                                   lun->be_lun->lun_type;
9728         else
9729                 sp->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9730
9731         sp->page_code = SVPD_SCSI_PORTS;
9732         scsi_ulto2b(data_len - sizeof(struct scsi_vpd_scsi_ports),
9733             sp->page_length);
9734         pd = &sp->design[0];
9735
9736         mtx_lock(&softc->ctl_lock);
9737         STAILQ_FOREACH(port, &softc->port_list, links) {
9738                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
9739                         continue;
9740                 if (lun != NULL &&
9741                     ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
9742                         continue;
9743                 scsi_ulto2b(port->targ_port, pd->relative_port_id);
9744                 if (port->init_devid) {
9745                         iid_len = port->init_devid->len;
9746                         memcpy(pd->initiator_transportid,
9747                             port->init_devid->data, port->init_devid->len);
9748                 } else
9749                         iid_len = 0;
9750                 scsi_ulto2b(iid_len, pd->initiator_transportid_length);
9751                 pdc = (struct scsi_vpd_port_designation_cont *)
9752                     (&pd->initiator_transportid[iid_len]);
9753                 if (port->port_devid) {
9754                         id_len = port->port_devid->len;
9755                         memcpy(pdc->target_port_descriptors,
9756                             port->port_devid->data, port->port_devid->len);
9757                 } else
9758                         id_len = 0;
9759                 scsi_ulto2b(id_len, pdc->target_port_descriptors_length);
9760                 pd = (struct scsi_vpd_port_designation *)
9761                     ((uint8_t *)pdc->target_port_descriptors + id_len);
9762         }
9763         mtx_unlock(&softc->ctl_lock);
9764
9765         ctl_set_success(ctsio);
9766         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9767         ctsio->be_move_done = ctl_config_move_done;
9768         ctl_datamove((union ctl_io *)ctsio);
9769         return (CTL_RETVAL_COMPLETE);
9770 }
9771
9772 static int
9773 ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio, int alloc_len)
9774 {
9775         struct ctl_lun *lun = CTL_LUN(ctsio);
9776         struct scsi_vpd_block_limits *bl_ptr;
9777         const char *val;
9778         uint64_t ival;
9779
9780         ctsio->kern_data_ptr = malloc(sizeof(*bl_ptr), M_CTL, M_WAITOK | M_ZERO);
9781         bl_ptr = (struct scsi_vpd_block_limits *)ctsio->kern_data_ptr;
9782         ctsio->kern_sg_entries = 0;
9783         ctsio->kern_rel_offset = 0;
9784         ctsio->kern_sg_entries = 0;
9785         ctsio->kern_data_len = min(sizeof(*bl_ptr), alloc_len);
9786         ctsio->kern_total_len = ctsio->kern_data_len;
9787
9788         /*
9789          * The control device is always connected.  The disk device, on the
9790          * other hand, may not be online all the time.  Need to change this
9791          * to figure out whether the disk device is actually online or not.
9792          */
9793         if (lun != NULL)
9794                 bl_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9795                                   lun->be_lun->lun_type;
9796         else
9797                 bl_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9798
9799         bl_ptr->page_code = SVPD_BLOCK_LIMITS;
9800         scsi_ulto2b(sizeof(*bl_ptr) - 4, bl_ptr->page_length);
9801         bl_ptr->max_cmp_write_len = 0xff;
9802         scsi_ulto4b(0xffffffff, bl_ptr->max_txfer_len);
9803         if (lun != NULL) {
9804                 scsi_ulto4b(lun->be_lun->opttxferlen, bl_ptr->opt_txfer_len);
9805                 if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
9806                         ival = 0xffffffff;
9807                         val = dnvlist_get_string(lun->be_lun->options,
9808                             "unmap_max_lba", NULL);
9809                         if (val != NULL)
9810                                 ctl_expand_number(val, &ival);
9811                         scsi_ulto4b(ival, bl_ptr->max_unmap_lba_cnt);
9812                         ival = 0xffffffff;
9813                         val = dnvlist_get_string(lun->be_lun->options,
9814                             "unmap_max_descr", NULL);
9815                         if (val != NULL)
9816                                 ctl_expand_number(val, &ival);
9817                         scsi_ulto4b(ival, bl_ptr->max_unmap_blk_cnt);
9818                         if (lun->be_lun->ublockexp != 0) {
9819                                 scsi_ulto4b((1 << lun->be_lun->ublockexp),
9820                                     bl_ptr->opt_unmap_grain);
9821                                 scsi_ulto4b(0x80000000 | lun->be_lun->ublockoff,
9822                                     bl_ptr->unmap_grain_align);
9823                         }
9824                 }
9825                 scsi_ulto4b(lun->be_lun->atomicblock,
9826                     bl_ptr->max_atomic_transfer_length);
9827                 scsi_ulto4b(0, bl_ptr->atomic_alignment);
9828                 scsi_ulto4b(0, bl_ptr->atomic_transfer_length_granularity);
9829                 scsi_ulto4b(0, bl_ptr->max_atomic_transfer_length_with_atomic_boundary);
9830                 scsi_ulto4b(0, bl_ptr->max_atomic_boundary_size);
9831                 ival = UINT64_MAX;
9832                 val = dnvlist_get_string(lun->be_lun->options,
9833                     "write_same_max_lba", NULL);
9834                 if (val != NULL)
9835                         ctl_expand_number(val, &ival);
9836                 scsi_u64to8b(ival, bl_ptr->max_write_same_length);
9837         }
9838
9839         ctl_set_success(ctsio);
9840         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9841         ctsio->be_move_done = ctl_config_move_done;
9842         ctl_datamove((union ctl_io *)ctsio);
9843         return (CTL_RETVAL_COMPLETE);
9844 }
9845
9846 static int
9847 ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len)
9848 {
9849         struct ctl_lun *lun = CTL_LUN(ctsio);
9850         struct scsi_vpd_block_device_characteristics *bdc_ptr;
9851         const char *value;
9852         u_int i;
9853
9854         ctsio->kern_data_ptr = malloc(sizeof(*bdc_ptr), M_CTL, M_WAITOK | M_ZERO);
9855         bdc_ptr = (struct scsi_vpd_block_device_characteristics *)ctsio->kern_data_ptr;
9856         ctsio->kern_sg_entries = 0;
9857         ctsio->kern_rel_offset = 0;
9858         ctsio->kern_data_len = min(sizeof(*bdc_ptr), alloc_len);
9859         ctsio->kern_total_len = ctsio->kern_data_len;
9860
9861         /*
9862          * The control device is always connected.  The disk device, on the
9863          * other hand, may not be online all the time.  Need to change this
9864          * to figure out whether the disk device is actually online or not.
9865          */
9866         if (lun != NULL)
9867                 bdc_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9868                                   lun->be_lun->lun_type;
9869         else
9870                 bdc_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9871         bdc_ptr->page_code = SVPD_BDC;
9872         scsi_ulto2b(sizeof(*bdc_ptr) - 4, bdc_ptr->page_length);
9873         if (lun != NULL &&
9874             (value = dnvlist_get_string(lun->be_lun->options, "rpm", NULL)) != NULL)
9875                 i = strtol(value, NULL, 0);
9876         else
9877                 i = CTL_DEFAULT_ROTATION_RATE;
9878         scsi_ulto2b(i, bdc_ptr->medium_rotation_rate);
9879         if (lun != NULL &&
9880             (value = dnvlist_get_string(lun->be_lun->options, "formfactor", NULL)) != NULL)
9881                 i = strtol(value, NULL, 0);
9882         else
9883                 i = 0;
9884         bdc_ptr->wab_wac_ff = (i & 0x0f);
9885         bdc_ptr->flags = SVPD_FUAB | SVPD_VBULS;
9886
9887         ctl_set_success(ctsio);
9888         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9889         ctsio->be_move_done = ctl_config_move_done;
9890         ctl_datamove((union ctl_io *)ctsio);
9891         return (CTL_RETVAL_COMPLETE);
9892 }
9893
9894 static int
9895 ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len)
9896 {
9897         struct ctl_lun *lun = CTL_LUN(ctsio);
9898         struct scsi_vpd_logical_block_prov *lbp_ptr;
9899         const char *value;
9900
9901         ctsio->kern_data_ptr = malloc(sizeof(*lbp_ptr), M_CTL, M_WAITOK | M_ZERO);
9902         lbp_ptr = (struct scsi_vpd_logical_block_prov *)ctsio->kern_data_ptr;
9903         ctsio->kern_sg_entries = 0;
9904         ctsio->kern_rel_offset = 0;
9905         ctsio->kern_data_len = min(sizeof(*lbp_ptr), alloc_len);
9906         ctsio->kern_total_len = ctsio->kern_data_len;
9907
9908         /*
9909          * The control device is always connected.  The disk device, on the
9910          * other hand, may not be online all the time.  Need to change this
9911          * to figure out whether the disk device is actually online or not.
9912          */
9913         if (lun != NULL)
9914                 lbp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9915                                   lun->be_lun->lun_type;
9916         else
9917                 lbp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9918
9919         lbp_ptr->page_code = SVPD_LBP;
9920         scsi_ulto2b(sizeof(*lbp_ptr) - 4, lbp_ptr->page_length);
9921         lbp_ptr->threshold_exponent = CTL_LBP_EXPONENT;
9922         if (lun != NULL && lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
9923                 lbp_ptr->flags = SVPD_LBP_UNMAP | SVPD_LBP_WS16 |
9924                     SVPD_LBP_WS10 | SVPD_LBP_RZ | SVPD_LBP_ANC_SUP;
9925                 value = dnvlist_get_string(lun->be_lun->options,
9926                     "provisioning_type", NULL);
9927                 if (value != NULL) {
9928                         if (strcmp(value, "resource") == 0)
9929                                 lbp_ptr->prov_type = SVPD_LBP_RESOURCE;
9930                         else if (strcmp(value, "thin") == 0)
9931                                 lbp_ptr->prov_type = SVPD_LBP_THIN;
9932                 } else
9933                         lbp_ptr->prov_type = SVPD_LBP_THIN;
9934         }
9935
9936         ctl_set_success(ctsio);
9937         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9938         ctsio->be_move_done = ctl_config_move_done;
9939         ctl_datamove((union ctl_io *)ctsio);
9940         return (CTL_RETVAL_COMPLETE);
9941 }
9942
9943 /*
9944  * INQUIRY with the EVPD bit set.
9945  */
9946 static int
9947 ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
9948 {
9949         struct ctl_lun *lun = CTL_LUN(ctsio);
9950         struct scsi_inquiry *cdb;
9951         int alloc_len, retval;
9952
9953         cdb = (struct scsi_inquiry *)ctsio->cdb;
9954         alloc_len = scsi_2btoul(cdb->length);
9955
9956         switch (cdb->page_code) {
9957         case SVPD_SUPPORTED_PAGES:
9958                 retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
9959                 break;
9960         case SVPD_UNIT_SERIAL_NUMBER:
9961                 retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
9962                 break;
9963         case SVPD_DEVICE_ID:
9964                 retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
9965                 break;
9966         case SVPD_EXTENDED_INQUIRY_DATA:
9967                 retval = ctl_inquiry_evpd_eid(ctsio, alloc_len);
9968                 break;
9969         case SVPD_MODE_PAGE_POLICY:
9970                 retval = ctl_inquiry_evpd_mpp(ctsio, alloc_len);
9971                 break;
9972         case SVPD_SCSI_PORTS:
9973                 retval = ctl_inquiry_evpd_scsi_ports(ctsio, alloc_len);
9974                 break;
9975         case SVPD_SCSI_TPC:
9976                 retval = ctl_inquiry_evpd_tpc(ctsio, alloc_len);
9977                 break;
9978         case SVPD_BLOCK_LIMITS:
9979                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
9980                         goto err;
9981                 retval = ctl_inquiry_evpd_block_limits(ctsio, alloc_len);
9982                 break;
9983         case SVPD_BDC:
9984                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
9985                         goto err;
9986                 retval = ctl_inquiry_evpd_bdc(ctsio, alloc_len);
9987                 break;
9988         case SVPD_LBP:
9989                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
9990                         goto err;
9991                 retval = ctl_inquiry_evpd_lbp(ctsio, alloc_len);
9992                 break;
9993         default:
9994 err:
9995                 ctl_set_invalid_field(ctsio,
9996                                       /*sks_valid*/ 1,
9997                                       /*command*/ 1,
9998                                       /*field*/ 2,
9999                                       /*bit_valid*/ 0,
10000                                       /*bit*/ 0);
10001                 ctl_done((union ctl_io *)ctsio);
10002                 retval = CTL_RETVAL_COMPLETE;
10003                 break;
10004         }
10005
10006         return (retval);
10007 }
10008
10009 /*
10010  * Standard INQUIRY data.
10011  */
10012 static int
10013 ctl_inquiry_std(struct ctl_scsiio *ctsio)
10014 {
10015         struct ctl_softc *softc = CTL_SOFTC(ctsio);
10016         struct ctl_port *port = CTL_PORT(ctsio);
10017         struct ctl_lun *lun = CTL_LUN(ctsio);
10018         struct scsi_inquiry_data *inq_ptr;
10019         struct scsi_inquiry *cdb;
10020         const char *val;
10021         uint32_t alloc_len, data_len;
10022         ctl_port_type port_type;
10023
10024         port_type = port->port_type;
10025         if (port_type == CTL_PORT_IOCTL || port_type == CTL_PORT_INTERNAL)
10026                 port_type = CTL_PORT_SCSI;
10027
10028         cdb = (struct scsi_inquiry *)ctsio->cdb;
10029         alloc_len = scsi_2btoul(cdb->length);
10030
10031         /*
10032          * We malloc the full inquiry data size here and fill it
10033          * in.  If the user only asks for less, we'll give him
10034          * that much.
10035          */
10036         data_len = offsetof(struct scsi_inquiry_data, vendor_specific1);
10037         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10038         inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
10039         ctsio->kern_sg_entries = 0;
10040         ctsio->kern_rel_offset = 0;
10041         ctsio->kern_data_len = min(data_len, alloc_len);
10042         ctsio->kern_total_len = ctsio->kern_data_len;
10043
10044         if (lun != NULL) {
10045                 if ((lun->flags & CTL_LUN_PRIMARY_SC) ||
10046                     softc->ha_link >= CTL_HA_LINK_UNKNOWN) {
10047                         inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10048                             lun->be_lun->lun_type;
10049                 } else {
10050                         inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) |
10051                             lun->be_lun->lun_type;
10052                 }
10053                 if (lun->flags & CTL_LUN_REMOVABLE)
10054                         inq_ptr->dev_qual2 |= SID_RMB;
10055         } else
10056                 inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE;
10057
10058         /* RMB in byte 2 is 0 */
10059         inq_ptr->version = SCSI_REV_SPC5;
10060
10061         /*
10062          * According to SAM-3, even if a device only supports a single
10063          * level of LUN addressing, it should still set the HISUP bit:
10064          *
10065          * 4.9.1 Logical unit numbers overview
10066          *
10067          * All logical unit number formats described in this standard are
10068          * hierarchical in structure even when only a single level in that
10069          * hierarchy is used. The HISUP bit shall be set to one in the
10070          * standard INQUIRY data (see SPC-2) when any logical unit number
10071          * format described in this standard is used.  Non-hierarchical
10072          * formats are outside the scope of this standard.
10073          *
10074          * Therefore we set the HiSup bit here.
10075          *
10076          * The response format is 2, per SPC-3.
10077          */
10078         inq_ptr->response_format = SID_HiSup | 2;
10079
10080         inq_ptr->additional_length = data_len -
10081             (offsetof(struct scsi_inquiry_data, additional_length) + 1);
10082         CTL_DEBUG_PRINT(("additional_length = %d\n",
10083                          inq_ptr->additional_length));
10084
10085         inq_ptr->spc3_flags = SPC3_SID_3PC | SPC3_SID_TPGS_IMPLICIT;
10086         if (port_type == CTL_PORT_SCSI)
10087                 inq_ptr->spc2_flags = SPC2_SID_ADDR16;
10088         inq_ptr->spc2_flags |= SPC2_SID_MultiP;
10089         inq_ptr->flags = SID_CmdQue;
10090         if (port_type == CTL_PORT_SCSI)
10091                 inq_ptr->flags |= SID_WBus16 | SID_Sync;
10092
10093         /*
10094          * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
10095          * We have 8 bytes for the vendor name, and 16 bytes for the device
10096          * name and 4 bytes for the revision.
10097          */
10098         if (lun == NULL || (val = dnvlist_get_string(lun->be_lun->options,
10099             "vendor", NULL)) == NULL) {
10100                 strncpy(inq_ptr->vendor, CTL_VENDOR, sizeof(inq_ptr->vendor));
10101         } else {
10102                 memset(inq_ptr->vendor, ' ', sizeof(inq_ptr->vendor));
10103                 strncpy(inq_ptr->vendor, val,
10104                     min(sizeof(inq_ptr->vendor), strlen(val)));
10105         }
10106         if (lun == NULL) {
10107                 strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10108                     sizeof(inq_ptr->product));
10109         } else if ((val = dnvlist_get_string(lun->be_lun->options, "product",
10110             NULL)) == NULL) {
10111                 switch (lun->be_lun->lun_type) {
10112                 case T_DIRECT:
10113                         strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10114                             sizeof(inq_ptr->product));
10115                         break;
10116                 case T_PROCESSOR:
10117                         strncpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT,
10118                             sizeof(inq_ptr->product));
10119                         break;
10120                 case T_CDROM:
10121                         strncpy(inq_ptr->product, CTL_CDROM_PRODUCT,
10122                             sizeof(inq_ptr->product));
10123                         break;
10124                 default:
10125                         strncpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT,
10126                             sizeof(inq_ptr->product));
10127                         break;
10128                 }
10129         } else {
10130                 memset(inq_ptr->product, ' ', sizeof(inq_ptr->product));
10131                 strncpy(inq_ptr->product, val,
10132                     min(sizeof(inq_ptr->product), strlen(val)));
10133         }
10134
10135         /*
10136          * XXX make this a macro somewhere so it automatically gets
10137          * incremented when we make changes.
10138          */
10139         if (lun == NULL || (val = dnvlist_get_string(lun->be_lun->options,
10140             "revision", NULL)) == NULL) {
10141                 strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
10142         } else {
10143                 memset(inq_ptr->revision, ' ', sizeof(inq_ptr->revision));
10144                 strncpy(inq_ptr->revision, val,
10145                     min(sizeof(inq_ptr->revision), strlen(val)));
10146         }
10147
10148         /*
10149          * For parallel SCSI, we support double transition and single
10150          * transition clocking.  We also support QAS (Quick Arbitration
10151          * and Selection) and Information Unit transfers on both the
10152          * control and array devices.
10153          */
10154         if (port_type == CTL_PORT_SCSI)
10155                 inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
10156                                     SID_SPI_IUS;
10157
10158         /* SAM-6 (no version claimed) */
10159         scsi_ulto2b(0x00C0, inq_ptr->version1);
10160         /* SPC-5 (no version claimed) */
10161         scsi_ulto2b(0x05C0, inq_ptr->version2);
10162         if (port_type == CTL_PORT_FC) {
10163                 /* FCP-2 ANSI INCITS.350:2003 */
10164                 scsi_ulto2b(0x0917, inq_ptr->version3);
10165         } else if (port_type == CTL_PORT_SCSI) {
10166                 /* SPI-4 ANSI INCITS.362:200x */
10167                 scsi_ulto2b(0x0B56, inq_ptr->version3);
10168         } else if (port_type == CTL_PORT_ISCSI) {
10169                 /* iSCSI (no version claimed) */
10170                 scsi_ulto2b(0x0960, inq_ptr->version3);
10171         } else if (port_type == CTL_PORT_SAS) {
10172                 /* SAS (no version claimed) */
10173                 scsi_ulto2b(0x0BE0, inq_ptr->version3);
10174         } else if (port_type == CTL_PORT_UMASS) {
10175                 /* USB Mass Storage Class Bulk-Only Transport, Revision 1.0 */
10176                 scsi_ulto2b(0x1730, inq_ptr->version3);
10177         }
10178
10179         if (lun == NULL) {
10180                 /* SBC-4 (no version claimed) */
10181                 scsi_ulto2b(0x0600, inq_ptr->version4);
10182         } else {
10183                 switch (lun->be_lun->lun_type) {
10184                 case T_DIRECT:
10185                         /* SBC-4 (no version claimed) */
10186                         scsi_ulto2b(0x0600, inq_ptr->version4);
10187                         break;
10188                 case T_PROCESSOR:
10189                         break;
10190                 case T_CDROM:
10191                         /* MMC-6 (no version claimed) */
10192                         scsi_ulto2b(0x04E0, inq_ptr->version4);
10193                         break;
10194                 default:
10195                         break;
10196                 }
10197         }
10198
10199         ctl_set_success(ctsio);
10200         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10201         ctsio->be_move_done = ctl_config_move_done;
10202         ctl_datamove((union ctl_io *)ctsio);
10203         return (CTL_RETVAL_COMPLETE);
10204 }
10205
10206 int
10207 ctl_inquiry(struct ctl_scsiio *ctsio)
10208 {
10209         struct scsi_inquiry *cdb;
10210         int retval;
10211
10212         CTL_DEBUG_PRINT(("ctl_inquiry\n"));
10213
10214         cdb = (struct scsi_inquiry *)ctsio->cdb;
10215         if (cdb->byte2 & SI_EVPD)
10216                 retval = ctl_inquiry_evpd(ctsio);
10217         else if (cdb->page_code == 0)
10218                 retval = ctl_inquiry_std(ctsio);
10219         else {
10220                 ctl_set_invalid_field(ctsio,
10221                                       /*sks_valid*/ 1,
10222                                       /*command*/ 1,
10223                                       /*field*/ 2,
10224                                       /*bit_valid*/ 0,
10225                                       /*bit*/ 0);
10226                 ctl_done((union ctl_io *)ctsio);
10227                 return (CTL_RETVAL_COMPLETE);
10228         }
10229
10230         return (retval);
10231 }
10232
10233 int
10234 ctl_get_config(struct ctl_scsiio *ctsio)
10235 {
10236         struct ctl_lun *lun = CTL_LUN(ctsio);
10237         struct scsi_get_config_header *hdr;
10238         struct scsi_get_config_feature *feature;
10239         struct scsi_get_config *cdb;
10240         uint32_t alloc_len, data_len;
10241         int rt, starting;
10242
10243         cdb = (struct scsi_get_config *)ctsio->cdb;
10244         rt = (cdb->rt & SGC_RT_MASK);
10245         starting = scsi_2btoul(cdb->starting_feature);
10246         alloc_len = scsi_2btoul(cdb->length);
10247
10248         data_len = sizeof(struct scsi_get_config_header) +
10249             sizeof(struct scsi_get_config_feature) + 8 +
10250             sizeof(struct scsi_get_config_feature) + 8 +
10251             sizeof(struct scsi_get_config_feature) + 4 +
10252             sizeof(struct scsi_get_config_feature) + 4 +
10253             sizeof(struct scsi_get_config_feature) + 8 +
10254             sizeof(struct scsi_get_config_feature) +
10255             sizeof(struct scsi_get_config_feature) + 4 +
10256             sizeof(struct scsi_get_config_feature) + 4 +
10257             sizeof(struct scsi_get_config_feature) + 4 +
10258             sizeof(struct scsi_get_config_feature) + 4 +
10259             sizeof(struct scsi_get_config_feature) + 4 +
10260             sizeof(struct scsi_get_config_feature) + 4;
10261         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10262         ctsio->kern_sg_entries = 0;
10263         ctsio->kern_rel_offset = 0;
10264
10265         hdr = (struct scsi_get_config_header *)ctsio->kern_data_ptr;
10266         if (lun->flags & CTL_LUN_NO_MEDIA)
10267                 scsi_ulto2b(0x0000, hdr->current_profile);
10268         else
10269                 scsi_ulto2b(0x0010, hdr->current_profile);
10270         feature = (struct scsi_get_config_feature *)(hdr + 1);
10271
10272         if (starting > 0x003b)
10273                 goto done;
10274         if (starting > 0x003a)
10275                 goto f3b;
10276         if (starting > 0x002b)
10277                 goto f3a;
10278         if (starting > 0x002a)
10279                 goto f2b;
10280         if (starting > 0x001f)
10281                 goto f2a;
10282         if (starting > 0x001e)
10283                 goto f1f;
10284         if (starting > 0x001d)
10285                 goto f1e;
10286         if (starting > 0x0010)
10287                 goto f1d;
10288         if (starting > 0x0003)
10289                 goto f10;
10290         if (starting > 0x0002)
10291                 goto f3;
10292         if (starting > 0x0001)
10293                 goto f2;
10294         if (starting > 0x0000)
10295                 goto f1;
10296
10297         /* Profile List */
10298         scsi_ulto2b(0x0000, feature->feature_code);
10299         feature->flags = SGC_F_PERSISTENT | SGC_F_CURRENT;
10300         feature->add_length = 8;
10301         scsi_ulto2b(0x0008, &feature->feature_data[0]); /* CD-ROM */
10302         feature->feature_data[2] = 0x00;
10303         scsi_ulto2b(0x0010, &feature->feature_data[4]); /* DVD-ROM */
10304         feature->feature_data[6] = 0x01;
10305         feature = (struct scsi_get_config_feature *)
10306             &feature->feature_data[feature->add_length];
10307
10308 f1:     /* Core */
10309         scsi_ulto2b(0x0001, feature->feature_code);
10310         feature->flags = 0x08 | SGC_F_PERSISTENT | SGC_F_CURRENT;
10311         feature->add_length = 8;
10312         scsi_ulto4b(0x00000000, &feature->feature_data[0]);
10313         feature->feature_data[4] = 0x03;
10314         feature = (struct scsi_get_config_feature *)
10315             &feature->feature_data[feature->add_length];
10316
10317 f2:     /* Morphing */
10318         scsi_ulto2b(0x0002, feature->feature_code);
10319         feature->flags = 0x04 | SGC_F_PERSISTENT | SGC_F_CURRENT;
10320         feature->add_length = 4;
10321         feature->feature_data[0] = 0x02;
10322         feature = (struct scsi_get_config_feature *)
10323             &feature->feature_data[feature->add_length];
10324
10325 f3:     /* Removable Medium */
10326         scsi_ulto2b(0x0003, feature->feature_code);
10327         feature->flags = 0x04 | SGC_F_PERSISTENT | SGC_F_CURRENT;
10328         feature->add_length = 4;
10329         feature->feature_data[0] = 0x39;
10330         feature = (struct scsi_get_config_feature *)
10331             &feature->feature_data[feature->add_length];
10332
10333         if (rt == SGC_RT_CURRENT && (lun->flags & CTL_LUN_NO_MEDIA))
10334                 goto done;
10335
10336 f10:    /* Random Read */
10337         scsi_ulto2b(0x0010, feature->feature_code);
10338         feature->flags = 0x00;
10339         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10340                 feature->flags |= SGC_F_CURRENT;
10341         feature->add_length = 8;
10342         scsi_ulto4b(lun->be_lun->blocksize, &feature->feature_data[0]);
10343         scsi_ulto2b(1, &feature->feature_data[4]);
10344         feature->feature_data[6] = 0x00;
10345         feature = (struct scsi_get_config_feature *)
10346             &feature->feature_data[feature->add_length];
10347
10348 f1d:    /* Multi-Read */
10349         scsi_ulto2b(0x001D, feature->feature_code);
10350         feature->flags = 0x00;
10351         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10352                 feature->flags |= SGC_F_CURRENT;
10353         feature->add_length = 0;
10354         feature = (struct scsi_get_config_feature *)
10355             &feature->feature_data[feature->add_length];
10356
10357 f1e:    /* CD Read */
10358         scsi_ulto2b(0x001E, feature->feature_code);
10359         feature->flags = 0x00;
10360         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10361                 feature->flags |= SGC_F_CURRENT;
10362         feature->add_length = 4;
10363         feature->feature_data[0] = 0x00;
10364         feature = (struct scsi_get_config_feature *)
10365             &feature->feature_data[feature->add_length];
10366
10367 f1f:    /* DVD Read */
10368         scsi_ulto2b(0x001F, feature->feature_code);
10369         feature->flags = 0x08;
10370         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10371                 feature->flags |= SGC_F_CURRENT;
10372         feature->add_length = 4;
10373         feature->feature_data[0] = 0x01;
10374         feature->feature_data[2] = 0x03;
10375         feature = (struct scsi_get_config_feature *)
10376             &feature->feature_data[feature->add_length];
10377
10378 f2a:    /* DVD+RW */
10379         scsi_ulto2b(0x002A, feature->feature_code);
10380         feature->flags = 0x04;
10381         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10382                 feature->flags |= SGC_F_CURRENT;
10383         feature->add_length = 4;
10384         feature->feature_data[0] = 0x00;
10385         feature->feature_data[1] = 0x00;
10386         feature = (struct scsi_get_config_feature *)
10387             &feature->feature_data[feature->add_length];
10388
10389 f2b:    /* DVD+R */
10390         scsi_ulto2b(0x002B, feature->feature_code);
10391         feature->flags = 0x00;
10392         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10393                 feature->flags |= SGC_F_CURRENT;
10394         feature->add_length = 4;
10395         feature->feature_data[0] = 0x00;
10396         feature = (struct scsi_get_config_feature *)
10397             &feature->feature_data[feature->add_length];
10398
10399 f3a:    /* DVD+RW Dual Layer */
10400         scsi_ulto2b(0x003A, feature->feature_code);
10401         feature->flags = 0x00;
10402         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10403                 feature->flags |= SGC_F_CURRENT;
10404         feature->add_length = 4;
10405         feature->feature_data[0] = 0x00;
10406         feature->feature_data[1] = 0x00;
10407         feature = (struct scsi_get_config_feature *)
10408             &feature->feature_data[feature->add_length];
10409
10410 f3b:    /* DVD+R Dual Layer */
10411         scsi_ulto2b(0x003B, feature->feature_code);
10412         feature->flags = 0x00;
10413         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10414                 feature->flags |= SGC_F_CURRENT;
10415         feature->add_length = 4;
10416         feature->feature_data[0] = 0x00;
10417         feature = (struct scsi_get_config_feature *)
10418             &feature->feature_data[feature->add_length];
10419
10420 done:
10421         data_len = (uint8_t *)feature - (uint8_t *)hdr;
10422         if (rt == SGC_RT_SPECIFIC && data_len > 4) {
10423                 feature = (struct scsi_get_config_feature *)(hdr + 1);
10424                 if (scsi_2btoul(feature->feature_code) == starting)
10425                         feature = (struct scsi_get_config_feature *)
10426                             &feature->feature_data[feature->add_length];
10427                 data_len = (uint8_t *)feature - (uint8_t *)hdr;
10428         }
10429         scsi_ulto4b(data_len - 4, hdr->data_length);
10430         ctsio->kern_data_len = min(data_len, alloc_len);
10431         ctsio->kern_total_len = ctsio->kern_data_len;
10432
10433         ctl_set_success(ctsio);
10434         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10435         ctsio->be_move_done = ctl_config_move_done;
10436         ctl_datamove((union ctl_io *)ctsio);
10437         return (CTL_RETVAL_COMPLETE);
10438 }
10439
10440 int
10441 ctl_get_event_status(struct ctl_scsiio *ctsio)
10442 {
10443         struct scsi_get_event_status_header *hdr;
10444         struct scsi_get_event_status *cdb;
10445         uint32_t alloc_len, data_len;
10446
10447         cdb = (struct scsi_get_event_status *)ctsio->cdb;
10448         if ((cdb->byte2 & SGESN_POLLED) == 0) {
10449                 ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 1,
10450                     /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 0);
10451                 ctl_done((union ctl_io *)ctsio);
10452                 return (CTL_RETVAL_COMPLETE);
10453         }
10454         alloc_len = scsi_2btoul(cdb->length);
10455
10456         data_len = sizeof(struct scsi_get_event_status_header);
10457         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10458         ctsio->kern_sg_entries = 0;
10459         ctsio->kern_rel_offset = 0;
10460         ctsio->kern_data_len = min(data_len, alloc_len);
10461         ctsio->kern_total_len = ctsio->kern_data_len;
10462
10463         hdr = (struct scsi_get_event_status_header *)ctsio->kern_data_ptr;
10464         scsi_ulto2b(0, hdr->descr_length);
10465         hdr->nea_class = SGESN_NEA;
10466         hdr->supported_class = 0;
10467
10468         ctl_set_success(ctsio);
10469         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10470         ctsio->be_move_done = ctl_config_move_done;
10471         ctl_datamove((union ctl_io *)ctsio);
10472         return (CTL_RETVAL_COMPLETE);
10473 }
10474
10475 int
10476 ctl_mechanism_status(struct ctl_scsiio *ctsio)
10477 {
10478         struct scsi_mechanism_status_header *hdr;
10479         struct scsi_mechanism_status *cdb;
10480         uint32_t alloc_len, data_len;
10481
10482         cdb = (struct scsi_mechanism_status *)ctsio->cdb;
10483         alloc_len = scsi_2btoul(cdb->length);
10484
10485         data_len = sizeof(struct scsi_mechanism_status_header);
10486         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10487         ctsio->kern_sg_entries = 0;
10488         ctsio->kern_rel_offset = 0;
10489         ctsio->kern_data_len = min(data_len, alloc_len);
10490         ctsio->kern_total_len = ctsio->kern_data_len;
10491
10492         hdr = (struct scsi_mechanism_status_header *)ctsio->kern_data_ptr;
10493         hdr->state1 = 0x00;
10494         hdr->state2 = 0xe0;
10495         scsi_ulto3b(0, hdr->lba);
10496         hdr->slots_num = 0;
10497         scsi_ulto2b(0, hdr->slots_length);
10498
10499         ctl_set_success(ctsio);
10500         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10501         ctsio->be_move_done = ctl_config_move_done;
10502         ctl_datamove((union ctl_io *)ctsio);
10503         return (CTL_RETVAL_COMPLETE);
10504 }
10505
10506 static void
10507 ctl_ultomsf(uint32_t lba, uint8_t *buf)
10508 {
10509
10510         lba += 150;
10511         buf[0] = 0;
10512         buf[1] = bin2bcd((lba / 75) / 60);
10513         buf[2] = bin2bcd((lba / 75) % 60);
10514         buf[3] = bin2bcd(lba % 75);
10515 }
10516
10517 int
10518 ctl_read_toc(struct ctl_scsiio *ctsio)
10519 {
10520         struct ctl_lun *lun = CTL_LUN(ctsio);
10521         struct scsi_read_toc_hdr *hdr;
10522         struct scsi_read_toc_type01_descr *descr;
10523         struct scsi_read_toc *cdb;
10524         uint32_t alloc_len, data_len;
10525         int format, msf;
10526
10527         cdb = (struct scsi_read_toc *)ctsio->cdb;
10528         msf = (cdb->byte2 & CD_MSF) != 0;
10529         format = cdb->format;
10530         alloc_len = scsi_2btoul(cdb->data_len);
10531
10532         data_len = sizeof(struct scsi_read_toc_hdr);
10533         if (format == 0)
10534                 data_len += 2 * sizeof(struct scsi_read_toc_type01_descr);
10535         else
10536                 data_len += sizeof(struct scsi_read_toc_type01_descr);
10537         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10538         ctsio->kern_sg_entries = 0;
10539         ctsio->kern_rel_offset = 0;
10540         ctsio->kern_data_len = min(data_len, alloc_len);
10541         ctsio->kern_total_len = ctsio->kern_data_len;
10542
10543         hdr = (struct scsi_read_toc_hdr *)ctsio->kern_data_ptr;
10544         if (format == 0) {
10545                 scsi_ulto2b(0x12, hdr->data_length);
10546                 hdr->first = 1;
10547                 hdr->last = 1;
10548                 descr = (struct scsi_read_toc_type01_descr *)(hdr + 1);
10549                 descr->addr_ctl = 0x14;
10550                 descr->track_number = 1;
10551                 if (msf)
10552                         ctl_ultomsf(0, descr->track_start);
10553                 else
10554                         scsi_ulto4b(0, descr->track_start);
10555                 descr++;
10556                 descr->addr_ctl = 0x14;
10557                 descr->track_number = 0xaa;
10558                 if (msf)
10559                         ctl_ultomsf(lun->be_lun->maxlba+1, descr->track_start);
10560                 else
10561                         scsi_ulto4b(lun->be_lun->maxlba+1, descr->track_start);
10562         } else {
10563                 scsi_ulto2b(0x0a, hdr->data_length);
10564                 hdr->first = 1;
10565                 hdr->last = 1;
10566                 descr = (struct scsi_read_toc_type01_descr *)(hdr + 1);
10567                 descr->addr_ctl = 0x14;
10568                 descr->track_number = 1;
10569                 if (msf)
10570                         ctl_ultomsf(0, descr->track_start);
10571                 else
10572                         scsi_ulto4b(0, descr->track_start);
10573         }
10574
10575         ctl_set_success(ctsio);
10576         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10577         ctsio->be_move_done = ctl_config_move_done;
10578         ctl_datamove((union ctl_io *)ctsio);
10579         return (CTL_RETVAL_COMPLETE);
10580 }
10581
10582 /*
10583  * For known CDB types, parse the LBA and length.
10584  */
10585 static int
10586 ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len)
10587 {
10588         if (io->io_hdr.io_type != CTL_IO_SCSI)
10589                 return (1);
10590
10591         switch (io->scsiio.cdb[0]) {
10592         case COMPARE_AND_WRITE: {
10593                 struct scsi_compare_and_write *cdb;
10594
10595                 cdb = (struct scsi_compare_and_write *)io->scsiio.cdb;
10596
10597                 *lba = scsi_8btou64(cdb->addr);
10598                 *len = cdb->length;
10599                 break;
10600         }
10601         case READ_6:
10602         case WRITE_6: {
10603                 struct scsi_rw_6 *cdb;
10604
10605                 cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
10606
10607                 *lba = scsi_3btoul(cdb->addr);
10608                 /* only 5 bits are valid in the most significant address byte */
10609                 *lba &= 0x1fffff;
10610                 *len = cdb->length;
10611                 break;
10612         }
10613         case READ_10:
10614         case WRITE_10: {
10615                 struct scsi_rw_10 *cdb;
10616
10617                 cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
10618
10619                 *lba = scsi_4btoul(cdb->addr);
10620                 *len = scsi_2btoul(cdb->length);
10621                 break;
10622         }
10623         case WRITE_VERIFY_10: {
10624                 struct scsi_write_verify_10 *cdb;
10625
10626                 cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
10627
10628                 *lba = scsi_4btoul(cdb->addr);
10629                 *len = scsi_2btoul(cdb->length);
10630                 break;
10631         }
10632         case READ_12:
10633         case WRITE_12: {
10634                 struct scsi_rw_12 *cdb;
10635
10636                 cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
10637
10638                 *lba = scsi_4btoul(cdb->addr);
10639                 *len = scsi_4btoul(cdb->length);
10640                 break;
10641         }
10642         case WRITE_VERIFY_12: {
10643                 struct scsi_write_verify_12 *cdb;
10644
10645                 cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
10646
10647                 *lba = scsi_4btoul(cdb->addr);
10648                 *len = scsi_4btoul(cdb->length);
10649                 break;
10650         }
10651         case READ_16:
10652         case WRITE_16: {
10653                 struct scsi_rw_16 *cdb;
10654
10655                 cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
10656
10657                 *lba = scsi_8btou64(cdb->addr);
10658                 *len = scsi_4btoul(cdb->length);
10659                 break;
10660         }
10661         case WRITE_ATOMIC_16: {
10662                 struct scsi_write_atomic_16 *cdb;
10663
10664                 cdb = (struct scsi_write_atomic_16 *)io->scsiio.cdb;
10665
10666                 *lba = scsi_8btou64(cdb->addr);
10667                 *len = scsi_2btoul(cdb->length);
10668                 break;
10669         }
10670         case WRITE_VERIFY_16: {
10671                 struct scsi_write_verify_16 *cdb;
10672
10673                 cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
10674
10675                 *lba = scsi_8btou64(cdb->addr);
10676                 *len = scsi_4btoul(cdb->length);
10677                 break;
10678         }
10679         case WRITE_SAME_10: {
10680                 struct scsi_write_same_10 *cdb;
10681
10682                 cdb = (struct scsi_write_same_10 *)io->scsiio.cdb;
10683
10684                 *lba = scsi_4btoul(cdb->addr);
10685                 *len = scsi_2btoul(cdb->length);
10686                 break;
10687         }
10688         case WRITE_SAME_16: {
10689                 struct scsi_write_same_16 *cdb;
10690
10691                 cdb = (struct scsi_write_same_16 *)io->scsiio.cdb;
10692
10693                 *lba = scsi_8btou64(cdb->addr);
10694                 *len = scsi_4btoul(cdb->length);
10695                 break;
10696         }
10697         case VERIFY_10: {
10698                 struct scsi_verify_10 *cdb;
10699
10700                 cdb = (struct scsi_verify_10 *)io->scsiio.cdb;
10701
10702                 *lba = scsi_4btoul(cdb->addr);
10703                 *len = scsi_2btoul(cdb->length);
10704                 break;
10705         }
10706         case VERIFY_12: {
10707                 struct scsi_verify_12 *cdb;
10708
10709                 cdb = (struct scsi_verify_12 *)io->scsiio.cdb;
10710
10711                 *lba = scsi_4btoul(cdb->addr);
10712                 *len = scsi_4btoul(cdb->length);
10713                 break;
10714         }
10715         case VERIFY_16: {
10716                 struct scsi_verify_16 *cdb;
10717
10718                 cdb = (struct scsi_verify_16 *)io->scsiio.cdb;
10719
10720                 *lba = scsi_8btou64(cdb->addr);
10721                 *len = scsi_4btoul(cdb->length);
10722                 break;
10723         }
10724         case UNMAP: {
10725                 *lba = 0;
10726                 *len = UINT64_MAX;
10727                 break;
10728         }
10729         case SERVICE_ACTION_IN: {       /* GET LBA STATUS */
10730                 struct scsi_get_lba_status *cdb;
10731
10732                 cdb = (struct scsi_get_lba_status *)io->scsiio.cdb;
10733                 *lba = scsi_8btou64(cdb->addr);
10734                 *len = UINT32_MAX;
10735                 break;
10736         }
10737         default:
10738                 return (1);
10739                 break; /* NOTREACHED */
10740         }
10741
10742         return (0);
10743 }
10744
10745 static ctl_action
10746 ctl_extent_check_lba(uint64_t lba1, uint64_t len1, uint64_t lba2, uint64_t len2,
10747     bool seq)
10748 {
10749         uint64_t endlba1, endlba2;
10750
10751         endlba1 = lba1 + len1 - (seq ? 0 : 1);
10752         endlba2 = lba2 + len2 - 1;
10753
10754         if ((endlba1 < lba2) || (endlba2 < lba1))
10755                 return (CTL_ACTION_PASS);
10756         else
10757                 return (CTL_ACTION_BLOCK);
10758 }
10759
10760 static int
10761 ctl_extent_check_unmap(union ctl_io *io, uint64_t lba2, uint64_t len2)
10762 {
10763         struct ctl_ptr_len_flags *ptrlen;
10764         struct scsi_unmap_desc *buf, *end, *range;
10765         uint64_t lba;
10766         uint32_t len;
10767
10768         /* If not UNMAP -- go other way. */
10769         if (io->io_hdr.io_type != CTL_IO_SCSI ||
10770             io->scsiio.cdb[0] != UNMAP)
10771                 return (CTL_ACTION_ERROR);
10772
10773         /* If UNMAP without data -- block and wait for data. */
10774         ptrlen = (struct ctl_ptr_len_flags *)
10775             &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
10776         if ((io->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0 ||
10777             ptrlen->ptr == NULL)
10778                 return (CTL_ACTION_BLOCK);
10779
10780         /* UNMAP with data -- check for collision. */
10781         buf = (struct scsi_unmap_desc *)ptrlen->ptr;
10782         end = buf + ptrlen->len / sizeof(*buf);
10783         for (range = buf; range < end; range++) {
10784                 lba = scsi_8btou64(range->lba);
10785                 len = scsi_4btoul(range->length);
10786                 if ((lba < lba2 + len2) && (lba + len > lba2))
10787                         return (CTL_ACTION_BLOCK);
10788         }
10789         return (CTL_ACTION_PASS);
10790 }
10791
10792 static ctl_action
10793 ctl_extent_check(union ctl_io *io1, union ctl_io *io2, bool seq)
10794 {
10795         uint64_t lba1, lba2;
10796         uint64_t len1, len2;
10797         int retval;
10798
10799         if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
10800                 return (CTL_ACTION_ERROR);
10801
10802         retval = ctl_extent_check_unmap(io1, lba2, len2);
10803         if (retval != CTL_ACTION_ERROR)
10804                 return (retval);
10805
10806         if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
10807                 return (CTL_ACTION_ERROR);
10808
10809         if (io1->io_hdr.flags & CTL_FLAG_SERSEQ_DONE)
10810                 seq = FALSE;
10811         return (ctl_extent_check_lba(lba1, len1, lba2, len2, seq));
10812 }
10813
10814 static ctl_action
10815 ctl_extent_check_seq(union ctl_io *io1, union ctl_io *io2)
10816 {
10817         uint64_t lba1, lba2;
10818         uint64_t len1, len2;
10819
10820         if (io1->io_hdr.flags & CTL_FLAG_SERSEQ_DONE)
10821                 return (CTL_ACTION_PASS);
10822         if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
10823                 return (CTL_ACTION_ERROR);
10824         if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
10825                 return (CTL_ACTION_ERROR);
10826
10827         if (lba1 + len1 == lba2)
10828                 return (CTL_ACTION_BLOCK);
10829         return (CTL_ACTION_PASS);
10830 }
10831
10832 static ctl_action
10833 ctl_check_for_blockage(struct ctl_lun *lun, union ctl_io *pending_io,
10834     union ctl_io *ooa_io)
10835 {
10836         const struct ctl_cmd_entry *pending_entry, *ooa_entry;
10837         const ctl_serialize_action *serialize_row;
10838
10839         /*
10840          * The initiator attempted multiple untagged commands at the same
10841          * time.  Can't do that.
10842          */
10843         if ((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10844          && (ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10845          && ((pending_io->io_hdr.nexus.targ_port ==
10846               ooa_io->io_hdr.nexus.targ_port)
10847           && (pending_io->io_hdr.nexus.initid ==
10848               ooa_io->io_hdr.nexus.initid))
10849          && ((ooa_io->io_hdr.flags & (CTL_FLAG_ABORT |
10850               CTL_FLAG_STATUS_SENT)) == 0))
10851                 return (CTL_ACTION_OVERLAP);
10852
10853         /*
10854          * The initiator attempted to send multiple tagged commands with
10855          * the same ID.  (It's fine if different initiators have the same
10856          * tag ID.)
10857          *
10858          * Even if all of those conditions are true, we don't kill the I/O
10859          * if the command ahead of us has been aborted.  We won't end up
10860          * sending it to the FETD, and it's perfectly legal to resend a
10861          * command with the same tag number as long as the previous
10862          * instance of this tag number has been aborted somehow.
10863          */
10864         if ((pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10865          && (ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10866          && (pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
10867          && ((pending_io->io_hdr.nexus.targ_port ==
10868               ooa_io->io_hdr.nexus.targ_port)
10869           && (pending_io->io_hdr.nexus.initid ==
10870               ooa_io->io_hdr.nexus.initid))
10871          && ((ooa_io->io_hdr.flags & (CTL_FLAG_ABORT |
10872               CTL_FLAG_STATUS_SENT)) == 0))
10873                 return (CTL_ACTION_OVERLAP_TAG);
10874
10875         /*
10876          * If we get a head of queue tag, SAM-3 says that we should
10877          * immediately execute it.
10878          *
10879          * What happens if this command would normally block for some other
10880          * reason?  e.g. a request sense with a head of queue tag
10881          * immediately after a write.  Normally that would block, but this
10882          * will result in its getting executed immediately...
10883          *
10884          * We currently return "pass" instead of "skip", so we'll end up
10885          * going through the rest of the queue to check for overlapped tags.
10886          *
10887          * XXX KDM check for other types of blockage first??
10888          */
10889         if (pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10890                 return (CTL_ACTION_PASS);
10891
10892         /*
10893          * Ordered tags have to block until all items ahead of them
10894          * have completed.  If we get called with an ordered tag, we always
10895          * block, if something else is ahead of us in the queue.
10896          */
10897         if (pending_io->scsiio.tag_type == CTL_TAG_ORDERED)
10898                 return (CTL_ACTION_BLOCK);
10899
10900         /*
10901          * Simple tags get blocked until all head of queue and ordered tags
10902          * ahead of them have completed.  I'm lumping untagged commands in
10903          * with simple tags here.  XXX KDM is that the right thing to do?
10904          */
10905         if (((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10906           || (pending_io->scsiio.tag_type == CTL_TAG_SIMPLE))
10907          && ((ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10908           || (ooa_io->scsiio.tag_type == CTL_TAG_ORDERED)))
10909                 return (CTL_ACTION_BLOCK);
10910
10911         pending_entry = ctl_get_cmd_entry(&pending_io->scsiio, NULL);
10912         KASSERT(pending_entry->seridx < CTL_SERIDX_COUNT,
10913             ("%s: Invalid seridx %d for pending CDB %02x %02x @ %p",
10914              __func__, pending_entry->seridx, pending_io->scsiio.cdb[0],
10915              pending_io->scsiio.cdb[1], pending_io));
10916         ooa_entry = ctl_get_cmd_entry(&ooa_io->scsiio, NULL);
10917         if (ooa_entry->seridx == CTL_SERIDX_INVLD)
10918                 return (CTL_ACTION_PASS); /* Unsupported command in OOA queue */
10919         KASSERT(ooa_entry->seridx < CTL_SERIDX_COUNT,
10920             ("%s: Invalid seridx %d for ooa CDB %02x %02x @ %p",
10921              __func__, ooa_entry->seridx, ooa_io->scsiio.cdb[0],
10922              ooa_io->scsiio.cdb[1], ooa_io));
10923
10924         serialize_row = ctl_serialize_table[ooa_entry->seridx];
10925
10926         switch (serialize_row[pending_entry->seridx]) {
10927         case CTL_SER_BLOCK:
10928                 return (CTL_ACTION_BLOCK);
10929         case CTL_SER_EXTENT:
10930                 return (ctl_extent_check(ooa_io, pending_io,
10931                     (lun->be_lun && lun->be_lun->serseq == CTL_LUN_SERSEQ_ON)));
10932         case CTL_SER_EXTENTOPT:
10933                 if ((lun->MODE_CTRL.queue_flags & SCP_QUEUE_ALG_MASK) !=
10934                     SCP_QUEUE_ALG_UNRESTRICTED)
10935                         return (ctl_extent_check(ooa_io, pending_io,
10936                             (lun->be_lun &&
10937                              lun->be_lun->serseq == CTL_LUN_SERSEQ_ON)));
10938                 return (CTL_ACTION_PASS);
10939         case CTL_SER_EXTENTSEQ:
10940                 if (lun->be_lun && lun->be_lun->serseq != CTL_LUN_SERSEQ_OFF)
10941                         return (ctl_extent_check_seq(ooa_io, pending_io));
10942                 return (CTL_ACTION_PASS);
10943         case CTL_SER_PASS:
10944                 return (CTL_ACTION_PASS);
10945         case CTL_SER_BLOCKOPT:
10946                 if ((lun->MODE_CTRL.queue_flags & SCP_QUEUE_ALG_MASK) !=
10947                     SCP_QUEUE_ALG_UNRESTRICTED)
10948                         return (CTL_ACTION_BLOCK);
10949                 return (CTL_ACTION_PASS);
10950         case CTL_SER_SKIP:
10951                 return (CTL_ACTION_SKIP);
10952         default:
10953                 panic("%s: Invalid serialization value %d for %d => %d",
10954                     __func__, serialize_row[pending_entry->seridx],
10955                     pending_entry->seridx, ooa_entry->seridx);
10956         }
10957
10958         return (CTL_ACTION_ERROR);
10959 }
10960
10961 /*
10962  * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
10963  * Assumptions:
10964  * - pending_io is generally either incoming, or on the blocked queue
10965  * - starting I/O is the I/O we want to start the check with.
10966  */
10967 static ctl_action
10968 ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
10969               union ctl_io *starting_io)
10970 {
10971         union ctl_io *ooa_io;
10972         ctl_action action;
10973
10974         mtx_assert(&lun->lun_lock, MA_OWNED);
10975
10976         /*
10977          * Run back along the OOA queue, starting with the current
10978          * blocked I/O and going through every I/O before it on the
10979          * queue.  If starting_io is NULL, we'll just end up returning
10980          * CTL_ACTION_PASS.
10981          */
10982         for (ooa_io = starting_io; ooa_io != NULL;
10983              ooa_io = (union ctl_io *)TAILQ_PREV(&ooa_io->io_hdr, ctl_ooaq,
10984              ooa_links)){
10985
10986                 /*
10987                  * This routine just checks to see whether
10988                  * cur_blocked is blocked by ooa_io, which is ahead
10989                  * of it in the queue.  It doesn't queue/dequeue
10990                  * cur_blocked.
10991                  */
10992                 action = ctl_check_for_blockage(lun, pending_io, ooa_io);
10993                 switch (action) {
10994                 case CTL_ACTION_BLOCK:
10995                 case CTL_ACTION_OVERLAP:
10996                 case CTL_ACTION_OVERLAP_TAG:
10997                 case CTL_ACTION_SKIP:
10998                 case CTL_ACTION_ERROR:
10999                         return (action);
11000                         break; /* NOTREACHED */
11001                 case CTL_ACTION_PASS:
11002                         break;
11003                 default:
11004                         panic("%s: Invalid action %d\n", __func__, action);
11005                 }
11006         }
11007
11008         return (CTL_ACTION_PASS);
11009 }
11010
11011 /*
11012  * Assumptions:
11013  * - An I/O has just completed, and has been removed from the per-LUN OOA
11014  *   queue, so some items on the blocked queue may now be unblocked.
11015  */
11016 static int
11017 ctl_check_blocked(struct ctl_lun *lun)
11018 {
11019         struct ctl_softc *softc = lun->ctl_softc;
11020         union ctl_io *cur_blocked, *next_blocked;
11021
11022         mtx_assert(&lun->lun_lock, MA_OWNED);
11023
11024         /*
11025          * Run forward from the head of the blocked queue, checking each
11026          * entry against the I/Os prior to it on the OOA queue to see if
11027          * there is still any blockage.
11028          *
11029          * We cannot use the TAILQ_FOREACH() macro, because it can't deal
11030          * with our removing a variable on it while it is traversing the
11031          * list.
11032          */
11033         for (cur_blocked = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue);
11034              cur_blocked != NULL; cur_blocked = next_blocked) {
11035                 union ctl_io *prev_ooa;
11036                 ctl_action action;
11037
11038                 next_blocked = (union ctl_io *)TAILQ_NEXT(&cur_blocked->io_hdr,
11039                                                           blocked_links);
11040
11041                 prev_ooa = (union ctl_io *)TAILQ_PREV(&cur_blocked->io_hdr,
11042                                                       ctl_ooaq, ooa_links);
11043
11044                 /*
11045                  * If cur_blocked happens to be the first item in the OOA
11046                  * queue now, prev_ooa will be NULL, and the action
11047                  * returned will just be CTL_ACTION_PASS.
11048                  */
11049                 action = ctl_check_ooa(lun, cur_blocked, prev_ooa);
11050
11051                 switch (action) {
11052                 case CTL_ACTION_BLOCK:
11053                         /* Nothing to do here, still blocked */
11054                         break;
11055                 case CTL_ACTION_OVERLAP:
11056                 case CTL_ACTION_OVERLAP_TAG:
11057                         /*
11058                          * This shouldn't happen!  In theory we've already
11059                          * checked this command for overlap...
11060                          */
11061                         break;
11062                 case CTL_ACTION_PASS:
11063                 case CTL_ACTION_SKIP: {
11064                         const struct ctl_cmd_entry *entry;
11065
11066                         /*
11067                          * The skip case shouldn't happen, this transaction
11068                          * should have never made it onto the blocked queue.
11069                          */
11070                         /*
11071                          * This I/O is no longer blocked, we can remove it
11072                          * from the blocked queue.  Since this is a TAILQ
11073                          * (doubly linked list), we can do O(1) removals
11074                          * from any place on the list.
11075                          */
11076                         TAILQ_REMOVE(&lun->blocked_queue, &cur_blocked->io_hdr,
11077                                      blocked_links);
11078                         cur_blocked->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
11079
11080                         if ((softc->ha_mode != CTL_HA_MODE_XFER) &&
11081                             (cur_blocked->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)){
11082                                 /*
11083                                  * Need to send IO back to original side to
11084                                  * run
11085                                  */
11086                                 union ctl_ha_msg msg_info;
11087
11088                                 cur_blocked->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11089                                 msg_info.hdr.original_sc =
11090                                         cur_blocked->io_hdr.original_sc;
11091                                 msg_info.hdr.serializing_sc = cur_blocked;
11092                                 msg_info.hdr.msg_type = CTL_MSG_R2R;
11093                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11094                                     sizeof(msg_info.hdr), M_NOWAIT);
11095                                 break;
11096                         }
11097                         entry = ctl_get_cmd_entry(&cur_blocked->scsiio, NULL);
11098
11099                         /*
11100                          * Check this I/O for LUN state changes that may
11101                          * have happened while this command was blocked.
11102                          * The LUN state may have been changed by a command
11103                          * ahead of us in the queue, so we need to re-check
11104                          * for any states that can be caused by SCSI
11105                          * commands.
11106                          */
11107                         if (ctl_scsiio_lun_check(lun, entry,
11108                                                  &cur_blocked->scsiio) == 0) {
11109                                 cur_blocked->io_hdr.flags |=
11110                                                       CTL_FLAG_IS_WAS_ON_RTR;
11111                                 ctl_enqueue_rtr(cur_blocked);
11112                         } else
11113                                 ctl_done(cur_blocked);
11114                         break;
11115                 }
11116                 default:
11117                         /*
11118                          * This probably shouldn't happen -- we shouldn't
11119                          * get CTL_ACTION_ERROR, or anything else.
11120                          */
11121                         break;
11122                 }
11123         }
11124
11125         return (CTL_RETVAL_COMPLETE);
11126 }
11127
11128 /*
11129  * This routine (with one exception) checks LUN flags that can be set by
11130  * commands ahead of us in the OOA queue.  These flags have to be checked
11131  * when a command initially comes in, and when we pull a command off the
11132  * blocked queue and are preparing to execute it.  The reason we have to
11133  * check these flags for commands on the blocked queue is that the LUN
11134  * state may have been changed by a command ahead of us while we're on the
11135  * blocked queue.
11136  *
11137  * Ordering is somewhat important with these checks, so please pay
11138  * careful attention to the placement of any new checks.
11139  */
11140 static int
11141 ctl_scsiio_lun_check(struct ctl_lun *lun,
11142     const struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
11143 {
11144         struct ctl_softc *softc = lun->ctl_softc;
11145         int retval;
11146         uint32_t residx;
11147
11148         retval = 0;
11149
11150         mtx_assert(&lun->lun_lock, MA_OWNED);
11151
11152         /*
11153          * If this shelf is a secondary shelf controller, we may have to
11154          * reject some commands disallowed by HA mode and link state.
11155          */
11156         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
11157                 if (softc->ha_link == CTL_HA_LINK_OFFLINE &&
11158                     (entry->flags & CTL_CMD_FLAG_OK_ON_UNAVAIL) == 0) {
11159                         ctl_set_lun_unavail(ctsio);
11160                         retval = 1;
11161                         goto bailout;
11162                 }
11163                 if ((lun->flags & CTL_LUN_PEER_SC_PRIMARY) == 0 &&
11164                     (entry->flags & CTL_CMD_FLAG_OK_ON_UNAVAIL) == 0) {
11165                         ctl_set_lun_transit(ctsio);
11166                         retval = 1;
11167                         goto bailout;
11168                 }
11169                 if (softc->ha_mode == CTL_HA_MODE_ACT_STBY &&
11170                     (entry->flags & CTL_CMD_FLAG_OK_ON_STANDBY) == 0) {
11171                         ctl_set_lun_standby(ctsio);
11172                         retval = 1;
11173                         goto bailout;
11174                 }
11175
11176                 /* The rest of checks are only done on executing side */
11177                 if (softc->ha_mode == CTL_HA_MODE_XFER)
11178                         goto bailout;
11179         }
11180
11181         if (entry->pattern & CTL_LUN_PAT_WRITE) {
11182                 if (lun->be_lun &&
11183                     lun->be_lun->flags & CTL_LUN_FLAG_READONLY) {
11184                         ctl_set_hw_write_protected(ctsio);
11185                         retval = 1;
11186                         goto bailout;
11187                 }
11188                 if ((lun->MODE_CTRL.eca_and_aen & SCP_SWP) != 0) {
11189                         ctl_set_sense(ctsio, /*current_error*/ 1,
11190                             /*sense_key*/ SSD_KEY_DATA_PROTECT,
11191                             /*asc*/ 0x27, /*ascq*/ 0x02, SSD_ELEM_NONE);
11192                         retval = 1;
11193                         goto bailout;
11194                 }
11195         }
11196
11197         /*
11198          * Check for a reservation conflict.  If this command isn't allowed
11199          * even on reserved LUNs, and if this initiator isn't the one who
11200          * reserved us, reject the command with a reservation conflict.
11201          */
11202         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11203         if ((lun->flags & CTL_LUN_RESERVED)
11204          && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
11205                 if (lun->res_idx != residx) {
11206                         ctl_set_reservation_conflict(ctsio);
11207                         retval = 1;
11208                         goto bailout;
11209                 }
11210         }
11211
11212         if ((lun->flags & CTL_LUN_PR_RESERVED) == 0 ||
11213             (entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV)) {
11214                 /* No reservation or command is allowed. */;
11215         } else if ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_WRESV) &&
11216             (lun->pr_res_type == SPR_TYPE_WR_EX ||
11217              lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
11218              lun->pr_res_type == SPR_TYPE_WR_EX_AR)) {
11219                 /* The command is allowed for Write Exclusive resv. */;
11220         } else {
11221                 /*
11222                  * if we aren't registered or it's a res holder type
11223                  * reservation and this isn't the res holder then set a
11224                  * conflict.
11225                  */
11226                 if (ctl_get_prkey(lun, residx) == 0 ||
11227                     (residx != lun->pr_res_idx && lun->pr_res_type < 4)) {
11228                         ctl_set_reservation_conflict(ctsio);
11229                         retval = 1;
11230                         goto bailout;
11231                 }
11232         }
11233
11234         if ((entry->flags & CTL_CMD_FLAG_OK_ON_NO_MEDIA) == 0) {
11235                 if (lun->flags & CTL_LUN_EJECTED)
11236                         ctl_set_lun_ejected(ctsio);
11237                 else if (lun->flags & CTL_LUN_NO_MEDIA) {
11238                         if (lun->flags & CTL_LUN_REMOVABLE)
11239                                 ctl_set_lun_no_media(ctsio);
11240                         else
11241                                 ctl_set_lun_int_reqd(ctsio);
11242                 } else if (lun->flags & CTL_LUN_STOPPED)
11243                         ctl_set_lun_stopped(ctsio);
11244                 else
11245                         goto bailout;
11246                 retval = 1;
11247                 goto bailout;
11248         }
11249
11250 bailout:
11251         return (retval);
11252 }
11253
11254 static void
11255 ctl_failover_io(union ctl_io *io, int have_lock)
11256 {
11257         ctl_set_busy(&io->scsiio);
11258         ctl_done(io);
11259 }
11260
11261 static void
11262 ctl_failover_lun(union ctl_io *rio)
11263 {
11264         struct ctl_softc *softc = CTL_SOFTC(rio);
11265         struct ctl_lun *lun;
11266         struct ctl_io_hdr *io, *next_io;
11267         uint32_t targ_lun;
11268
11269         targ_lun = rio->io_hdr.nexus.targ_mapped_lun;
11270         CTL_DEBUG_PRINT(("FAILOVER for lun %ju\n", targ_lun));
11271
11272         /* Find and lock the LUN. */
11273         mtx_lock(&softc->ctl_lock);
11274         if (targ_lun > ctl_max_luns ||
11275             (lun = softc->ctl_luns[targ_lun]) == NULL) {
11276                 mtx_unlock(&softc->ctl_lock);
11277                 return;
11278         }
11279         mtx_lock(&lun->lun_lock);
11280         mtx_unlock(&softc->ctl_lock);
11281         if (lun->flags & CTL_LUN_DISABLED) {
11282                 mtx_unlock(&lun->lun_lock);
11283                 return;
11284         }
11285
11286         if (softc->ha_mode == CTL_HA_MODE_XFER) {
11287                 TAILQ_FOREACH_SAFE(io, &lun->ooa_queue, ooa_links, next_io) {
11288                         /* We are master */
11289                         if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
11290                                 if (io->flags & CTL_FLAG_IO_ACTIVE) {
11291                                         io->flags |= CTL_FLAG_ABORT;
11292                                         io->flags |= CTL_FLAG_FAILOVER;
11293                                 } else { /* This can be only due to DATAMOVE */
11294                                         io->msg_type = CTL_MSG_DATAMOVE_DONE;
11295                                         io->flags &= ~CTL_FLAG_DMA_INPROG;
11296                                         io->flags |= CTL_FLAG_IO_ACTIVE;
11297                                         io->port_status = 31340;
11298                                         ctl_enqueue_isc((union ctl_io *)io);
11299                                 }
11300                         }
11301                         /* We are slave */
11302                         if (io->flags & CTL_FLAG_SENT_2OTHER_SC) {
11303                                 io->flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11304                                 if (io->flags & CTL_FLAG_IO_ACTIVE) {
11305                                         io->flags |= CTL_FLAG_FAILOVER;
11306                                 } else {
11307                                         ctl_set_busy(&((union ctl_io *)io)->
11308                                             scsiio);
11309                                         ctl_done((union ctl_io *)io);
11310                                 }
11311                         }
11312                 }
11313         } else { /* SERIALIZE modes */
11314                 TAILQ_FOREACH_SAFE(io, &lun->blocked_queue, blocked_links,
11315                     next_io) {
11316                         /* We are master */
11317                         if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
11318                                 TAILQ_REMOVE(&lun->blocked_queue, io,
11319                                     blocked_links);
11320                                 io->flags &= ~CTL_FLAG_BLOCKED;
11321                                 TAILQ_REMOVE(&lun->ooa_queue, io, ooa_links);
11322                                 ctl_free_io((union ctl_io *)io);
11323                         }
11324                 }
11325                 TAILQ_FOREACH_SAFE(io, &lun->ooa_queue, ooa_links, next_io) {
11326                         /* We are master */
11327                         if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
11328                                 TAILQ_REMOVE(&lun->ooa_queue, io, ooa_links);
11329                                 ctl_free_io((union ctl_io *)io);
11330                         }
11331                         /* We are slave */
11332                         if (io->flags & CTL_FLAG_SENT_2OTHER_SC) {
11333                                 io->flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11334                                 if (!(io->flags & CTL_FLAG_IO_ACTIVE)) {
11335                                         ctl_set_busy(&((union ctl_io *)io)->
11336                                             scsiio);
11337                                         ctl_done((union ctl_io *)io);
11338                                 }
11339                         }
11340                 }
11341                 ctl_check_blocked(lun);
11342         }
11343         mtx_unlock(&lun->lun_lock);
11344 }
11345
11346 static int
11347 ctl_scsiio_precheck(struct ctl_softc *softc, struct ctl_scsiio *ctsio)
11348 {
11349         struct ctl_lun *lun;
11350         const struct ctl_cmd_entry *entry;
11351         uint32_t initidx, targ_lun;
11352         int retval = 0;
11353
11354         lun = NULL;
11355         targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
11356         if (targ_lun < ctl_max_luns)
11357                 lun = softc->ctl_luns[targ_lun];
11358         if (lun) {
11359                 /*
11360                  * If the LUN is invalid, pretend that it doesn't exist.
11361                  * It will go away as soon as all pending I/O has been
11362                  * completed.
11363                  */
11364                 mtx_lock(&lun->lun_lock);
11365                 if (lun->flags & CTL_LUN_DISABLED) {
11366                         mtx_unlock(&lun->lun_lock);
11367                         lun = NULL;
11368                 }
11369         }
11370         CTL_LUN(ctsio) = lun;
11371         if (lun) {
11372                 CTL_BACKEND_LUN(ctsio) = lun->be_lun;
11373
11374                 /*
11375                  * Every I/O goes into the OOA queue for a particular LUN,
11376                  * and stays there until completion.
11377                  */
11378 #ifdef CTL_TIME_IO
11379                 if (TAILQ_EMPTY(&lun->ooa_queue))
11380                         lun->idle_time += getsbinuptime() - lun->last_busy;
11381 #endif
11382                 TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
11383         }
11384
11385         /* Get command entry and return error if it is unsuppotyed. */
11386         entry = ctl_validate_command(ctsio);
11387         if (entry == NULL) {
11388                 if (lun)
11389                         mtx_unlock(&lun->lun_lock);
11390                 return (retval);
11391         }
11392
11393         ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
11394         ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
11395
11396         /*
11397          * Check to see whether we can send this command to LUNs that don't
11398          * exist.  This should pretty much only be the case for inquiry
11399          * and request sense.  Further checks, below, really require having
11400          * a LUN, so we can't really check the command anymore.  Just put
11401          * it on the rtr queue.
11402          */
11403         if (lun == NULL) {
11404                 if (entry->flags & CTL_CMD_FLAG_OK_ON_NO_LUN) {
11405                         ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11406                         ctl_enqueue_rtr((union ctl_io *)ctsio);
11407                         return (retval);
11408                 }
11409
11410                 ctl_set_unsupported_lun(ctsio);
11411                 ctl_done((union ctl_io *)ctsio);
11412                 CTL_DEBUG_PRINT(("ctl_scsiio_precheck: bailing out due to invalid LUN\n"));
11413                 return (retval);
11414         } else {
11415                 /*
11416                  * Make sure we support this particular command on this LUN.
11417                  * e.g., we don't support writes to the control LUN.
11418                  */
11419                 if (!ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
11420                         mtx_unlock(&lun->lun_lock);
11421                         ctl_set_invalid_opcode(ctsio);
11422                         ctl_done((union ctl_io *)ctsio);
11423                         return (retval);
11424                 }
11425         }
11426
11427         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11428
11429         /*
11430          * If we've got a request sense, it'll clear the contingent
11431          * allegiance condition.  Otherwise, if we have a CA condition for
11432          * this initiator, clear it, because it sent down a command other
11433          * than request sense.
11434          */
11435         if (ctsio->cdb[0] != REQUEST_SENSE) {
11436                 struct scsi_sense_data *ps;
11437
11438                 ps = lun->pending_sense[initidx / CTL_MAX_INIT_PER_PORT];
11439                 if (ps != NULL)
11440                         ps[initidx % CTL_MAX_INIT_PER_PORT].error_code = 0;
11441         }
11442
11443         /*
11444          * If the command has this flag set, it handles its own unit
11445          * attention reporting, we shouldn't do anything.  Otherwise we
11446          * check for any pending unit attentions, and send them back to the
11447          * initiator.  We only do this when a command initially comes in,
11448          * not when we pull it off the blocked queue.
11449          *
11450          * According to SAM-3, section 5.3.2, the order that things get
11451          * presented back to the host is basically unit attentions caused
11452          * by some sort of reset event, busy status, reservation conflicts
11453          * or task set full, and finally any other status.
11454          *
11455          * One issue here is that some of the unit attentions we report
11456          * don't fall into the "reset" category (e.g. "reported luns data
11457          * has changed").  So reporting it here, before the reservation
11458          * check, may be technically wrong.  I guess the only thing to do
11459          * would be to check for and report the reset events here, and then
11460          * check for the other unit attention types after we check for a
11461          * reservation conflict.
11462          *
11463          * XXX KDM need to fix this
11464          */
11465         if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
11466                 ctl_ua_type ua_type;
11467                 u_int sense_len = 0;
11468
11469                 ua_type = ctl_build_ua(lun, initidx, &ctsio->sense_data,
11470                     &sense_len, SSD_TYPE_NONE);
11471                 if (ua_type != CTL_UA_NONE) {
11472                         mtx_unlock(&lun->lun_lock);
11473                         ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
11474                         ctsio->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
11475                         ctsio->sense_len = sense_len;
11476                         ctl_done((union ctl_io *)ctsio);
11477                         return (retval);
11478                 }
11479         }
11480
11481
11482         if (ctl_scsiio_lun_check(lun, entry, ctsio) != 0) {
11483                 mtx_unlock(&lun->lun_lock);
11484                 ctl_done((union ctl_io *)ctsio);
11485                 return (retval);
11486         }
11487
11488         /*
11489          * XXX CHD this is where we want to send IO to other side if
11490          * this LUN is secondary on this SC. We will need to make a copy
11491          * of the IO and flag the IO on this side as SENT_2OTHER and the flag
11492          * the copy we send as FROM_OTHER.
11493          * We also need to stuff the address of the original IO so we can
11494          * find it easily. Something similar will need be done on the other
11495          * side so when we are done we can find the copy.
11496          */
11497         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
11498             (lun->flags & CTL_LUN_PEER_SC_PRIMARY) != 0 &&
11499             (entry->flags & CTL_CMD_FLAG_RUN_HERE) == 0) {
11500                 union ctl_ha_msg msg_info;
11501                 int isc_retval;
11502
11503                 ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11504                 ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11505                 mtx_unlock(&lun->lun_lock);
11506
11507                 msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
11508                 msg_info.hdr.original_sc = (union ctl_io *)ctsio;
11509                 msg_info.hdr.serializing_sc = NULL;
11510                 msg_info.hdr.nexus = ctsio->io_hdr.nexus;
11511                 msg_info.scsi.tag_num = ctsio->tag_num;
11512                 msg_info.scsi.tag_type = ctsio->tag_type;
11513                 msg_info.scsi.cdb_len = ctsio->cdb_len;
11514                 memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
11515
11516                 if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11517                     sizeof(msg_info.scsi) - sizeof(msg_info.scsi.sense_data),
11518                     M_WAITOK)) > CTL_HA_STATUS_SUCCESS) {
11519                         ctl_set_busy(ctsio);
11520                         ctl_done((union ctl_io *)ctsio);
11521                         return (retval);
11522                 }
11523                 return (retval);
11524         }
11525
11526         switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
11527                               (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr,
11528                               ctl_ooaq, ooa_links))) {
11529         case CTL_ACTION_BLOCK:
11530                 ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
11531                 TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
11532                                   blocked_links);
11533                 mtx_unlock(&lun->lun_lock);
11534                 return (retval);
11535         case CTL_ACTION_PASS:
11536         case CTL_ACTION_SKIP:
11537                 ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11538                 mtx_unlock(&lun->lun_lock);
11539                 ctl_enqueue_rtr((union ctl_io *)ctsio);
11540                 break;
11541         case CTL_ACTION_OVERLAP:
11542                 mtx_unlock(&lun->lun_lock);
11543                 ctl_set_overlapped_cmd(ctsio);
11544                 ctl_done((union ctl_io *)ctsio);
11545                 break;
11546         case CTL_ACTION_OVERLAP_TAG:
11547                 mtx_unlock(&lun->lun_lock);
11548                 ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
11549                 ctl_done((union ctl_io *)ctsio);
11550                 break;
11551         case CTL_ACTION_ERROR:
11552         default:
11553                 mtx_unlock(&lun->lun_lock);
11554                 ctl_set_internal_failure(ctsio,
11555                                          /*sks_valid*/ 0,
11556                                          /*retry_count*/ 0);
11557                 ctl_done((union ctl_io *)ctsio);
11558                 break;
11559         }
11560         return (retval);
11561 }
11562
11563 const struct ctl_cmd_entry *
11564 ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa)
11565 {
11566         const struct ctl_cmd_entry *entry;
11567         int service_action;
11568
11569         entry = &ctl_cmd_table[ctsio->cdb[0]];
11570         if (sa)
11571                 *sa = ((entry->flags & CTL_CMD_FLAG_SA5) != 0);
11572         if (entry->flags & CTL_CMD_FLAG_SA5) {
11573                 service_action = ctsio->cdb[1] & SERVICE_ACTION_MASK;
11574                 entry = &((const struct ctl_cmd_entry *)
11575                     entry->execute)[service_action];
11576         }
11577         return (entry);
11578 }
11579
11580 const struct ctl_cmd_entry *
11581 ctl_validate_command(struct ctl_scsiio *ctsio)
11582 {
11583         const struct ctl_cmd_entry *entry;
11584         int i, sa;
11585         uint8_t diff;
11586
11587         entry = ctl_get_cmd_entry(ctsio, &sa);
11588         if (entry->execute == NULL) {
11589                 if (sa)
11590                         ctl_set_invalid_field(ctsio,
11591                                               /*sks_valid*/ 1,
11592                                               /*command*/ 1,
11593                                               /*field*/ 1,
11594                                               /*bit_valid*/ 1,
11595                                               /*bit*/ 4);
11596                 else
11597                         ctl_set_invalid_opcode(ctsio);
11598                 ctl_done((union ctl_io *)ctsio);
11599                 return (NULL);
11600         }
11601         KASSERT(entry->length > 0,
11602             ("Not defined length for command 0x%02x/0x%02x",
11603              ctsio->cdb[0], ctsio->cdb[1]));
11604         for (i = 1; i < entry->length; i++) {
11605                 diff = ctsio->cdb[i] & ~entry->usage[i - 1];
11606                 if (diff == 0)
11607                         continue;
11608                 ctl_set_invalid_field(ctsio,
11609                                       /*sks_valid*/ 1,
11610                                       /*command*/ 1,
11611                                       /*field*/ i,
11612                                       /*bit_valid*/ 1,
11613                                       /*bit*/ fls(diff) - 1);
11614                 ctl_done((union ctl_io *)ctsio);
11615                 return (NULL);
11616         }
11617         return (entry);
11618 }
11619
11620 static int
11621 ctl_cmd_applicable(uint8_t lun_type, const struct ctl_cmd_entry *entry)
11622 {
11623
11624         switch (lun_type) {
11625         case T_DIRECT:
11626                 if ((entry->flags & CTL_CMD_FLAG_OK_ON_DIRECT) == 0)
11627                         return (0);
11628                 break;
11629         case T_PROCESSOR:
11630                 if ((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0)
11631                         return (0);
11632                 break;
11633         case T_CDROM:
11634                 if ((entry->flags & CTL_CMD_FLAG_OK_ON_CDROM) == 0)
11635                         return (0);
11636                 break;
11637         default:
11638                 return (0);
11639         }
11640         return (1);
11641 }
11642
11643 static int
11644 ctl_scsiio(struct ctl_scsiio *ctsio)
11645 {
11646         int retval;
11647         const struct ctl_cmd_entry *entry;
11648
11649         retval = CTL_RETVAL_COMPLETE;
11650
11651         CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
11652
11653         entry = ctl_get_cmd_entry(ctsio, NULL);
11654
11655         /*
11656          * If this I/O has been aborted, just send it straight to
11657          * ctl_done() without executing it.
11658          */
11659         if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
11660                 ctl_done((union ctl_io *)ctsio);
11661                 goto bailout;
11662         }
11663
11664         /*
11665          * All the checks should have been handled by ctl_scsiio_precheck().
11666          * We should be clear now to just execute the I/O.
11667          */
11668         retval = entry->execute(ctsio);
11669
11670 bailout:
11671         return (retval);
11672 }
11673
11674 static int
11675 ctl_target_reset(union ctl_io *io)
11676 {
11677         struct ctl_softc *softc = CTL_SOFTC(io);
11678         struct ctl_port *port = CTL_PORT(io);
11679         struct ctl_lun *lun;
11680         uint32_t initidx;
11681         ctl_ua_type ua_type;
11682
11683         if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11684                 union ctl_ha_msg msg_info;
11685
11686                 msg_info.hdr.nexus = io->io_hdr.nexus;
11687                 msg_info.task.task_action = io->taskio.task_action;
11688                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11689                 msg_info.hdr.original_sc = NULL;
11690                 msg_info.hdr.serializing_sc = NULL;
11691                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11692                     sizeof(msg_info.task), M_WAITOK);
11693         }
11694
11695         initidx = ctl_get_initindex(&io->io_hdr.nexus);
11696         if (io->taskio.task_action == CTL_TASK_TARGET_RESET)
11697                 ua_type = CTL_UA_TARG_RESET;
11698         else
11699                 ua_type = CTL_UA_BUS_RESET;
11700         mtx_lock(&softc->ctl_lock);
11701         STAILQ_FOREACH(lun, &softc->lun_list, links) {
11702                 if (port != NULL &&
11703                     ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
11704                         continue;
11705                 ctl_do_lun_reset(lun, initidx, ua_type);
11706         }
11707         mtx_unlock(&softc->ctl_lock);
11708         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11709         return (0);
11710 }
11711
11712 /*
11713  * The LUN should always be set.  The I/O is optional, and is used to
11714  * distinguish between I/Os sent by this initiator, and by other
11715  * initiators.  We set unit attention for initiators other than this one.
11716  * SAM-3 is vague on this point.  It does say that a unit attention should
11717  * be established for other initiators when a LUN is reset (see section
11718  * 5.7.3), but it doesn't specifically say that the unit attention should
11719  * be established for this particular initiator when a LUN is reset.  Here
11720  * is the relevant text, from SAM-3 rev 8:
11721  *
11722  * 5.7.2 When a SCSI initiator port aborts its own tasks
11723  *
11724  * When a SCSI initiator port causes its own task(s) to be aborted, no
11725  * notification that the task(s) have been aborted shall be returned to
11726  * the SCSI initiator port other than the completion response for the
11727  * command or task management function action that caused the task(s) to
11728  * be aborted and notification(s) associated with related effects of the
11729  * action (e.g., a reset unit attention condition).
11730  *
11731  * XXX KDM for now, we're setting unit attention for all initiators.
11732  */
11733 static void
11734 ctl_do_lun_reset(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua_type)
11735 {
11736         union ctl_io *xio;
11737         int i;
11738
11739         mtx_lock(&lun->lun_lock);
11740         /* Abort tasks. */
11741         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11742              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11743                 xio->io_hdr.flags |= CTL_FLAG_ABORT | CTL_FLAG_ABORT_STATUS;
11744         }
11745         /* Clear CA. */
11746         for (i = 0; i < ctl_max_ports; i++) {
11747                 free(lun->pending_sense[i], M_CTL);
11748                 lun->pending_sense[i] = NULL;
11749         }
11750         /* Clear reservation. */
11751         lun->flags &= ~CTL_LUN_RESERVED;
11752         /* Clear prevent media removal. */
11753         if (lun->prevent) {
11754                 for (i = 0; i < CTL_MAX_INITIATORS; i++)
11755                         ctl_clear_mask(lun->prevent, i);
11756                 lun->prevent_count = 0;
11757         }
11758         /* Clear TPC status */
11759         ctl_tpc_lun_clear(lun, -1);
11760         /* Establish UA. */
11761 #if 0
11762         ctl_est_ua_all(lun, initidx, ua_type);
11763 #else
11764         ctl_est_ua_all(lun, -1, ua_type);
11765 #endif
11766         mtx_unlock(&lun->lun_lock);
11767 }
11768
11769 static int
11770 ctl_lun_reset(union ctl_io *io)
11771 {
11772         struct ctl_softc *softc = CTL_SOFTC(io);
11773         struct ctl_lun *lun;
11774         uint32_t targ_lun, initidx;
11775
11776         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11777         initidx = ctl_get_initindex(&io->io_hdr.nexus);
11778         mtx_lock(&softc->ctl_lock);
11779         if (targ_lun >= ctl_max_luns ||
11780             (lun = softc->ctl_luns[targ_lun]) == NULL) {
11781                 mtx_unlock(&softc->ctl_lock);
11782                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
11783                 return (1);
11784         }
11785         ctl_do_lun_reset(lun, initidx, CTL_UA_LUN_RESET);
11786         mtx_unlock(&softc->ctl_lock);
11787         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11788
11789         if ((io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) == 0) {
11790                 union ctl_ha_msg msg_info;
11791
11792                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11793                 msg_info.hdr.nexus = io->io_hdr.nexus;
11794                 msg_info.task.task_action = CTL_TASK_LUN_RESET;
11795                 msg_info.hdr.original_sc = NULL;
11796                 msg_info.hdr.serializing_sc = NULL;
11797                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11798                     sizeof(msg_info.task), M_WAITOK);
11799         }
11800         return (0);
11801 }
11802
11803 static void
11804 ctl_abort_tasks_lun(struct ctl_lun *lun, uint32_t targ_port, uint32_t init_id,
11805     int other_sc)
11806 {
11807         union ctl_io *xio;
11808
11809         mtx_assert(&lun->lun_lock, MA_OWNED);
11810
11811         /*
11812          * Run through the OOA queue and attempt to find the given I/O.
11813          * The target port, initiator ID, tag type and tag number have to
11814          * match the values that we got from the initiator.  If we have an
11815          * untagged command to abort, simply abort the first untagged command
11816          * we come to.  We only allow one untagged command at a time of course.
11817          */
11818         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11819              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11820
11821                 if ((targ_port == UINT32_MAX ||
11822                      targ_port == xio->io_hdr.nexus.targ_port) &&
11823                     (init_id == UINT32_MAX ||
11824                      init_id == xio->io_hdr.nexus.initid)) {
11825                         if (targ_port != xio->io_hdr.nexus.targ_port ||
11826                             init_id != xio->io_hdr.nexus.initid)
11827                                 xio->io_hdr.flags |= CTL_FLAG_ABORT_STATUS;
11828                         xio->io_hdr.flags |= CTL_FLAG_ABORT;
11829                         if (!other_sc && !(lun->flags & CTL_LUN_PRIMARY_SC)) {
11830                                 union ctl_ha_msg msg_info;
11831
11832                                 msg_info.hdr.nexus = xio->io_hdr.nexus;
11833                                 msg_info.task.task_action = CTL_TASK_ABORT_TASK;
11834                                 msg_info.task.tag_num = xio->scsiio.tag_num;
11835                                 msg_info.task.tag_type = xio->scsiio.tag_type;
11836                                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11837                                 msg_info.hdr.original_sc = NULL;
11838                                 msg_info.hdr.serializing_sc = NULL;
11839                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11840                                     sizeof(msg_info.task), M_NOWAIT);
11841                         }
11842                 }
11843         }
11844 }
11845
11846 static int
11847 ctl_abort_task_set(union ctl_io *io)
11848 {
11849         struct ctl_softc *softc = CTL_SOFTC(io);
11850         struct ctl_lun *lun;
11851         uint32_t targ_lun;
11852
11853         /*
11854          * Look up the LUN.
11855          */
11856         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11857         mtx_lock(&softc->ctl_lock);
11858         if (targ_lun >= ctl_max_luns ||
11859             (lun = softc->ctl_luns[targ_lun]) == NULL) {
11860                 mtx_unlock(&softc->ctl_lock);
11861                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
11862                 return (1);
11863         }
11864
11865         mtx_lock(&lun->lun_lock);
11866         mtx_unlock(&softc->ctl_lock);
11867         if (io->taskio.task_action == CTL_TASK_ABORT_TASK_SET) {
11868                 ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
11869                     io->io_hdr.nexus.initid,
11870                     (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
11871         } else { /* CTL_TASK_CLEAR_TASK_SET */
11872                 ctl_abort_tasks_lun(lun, UINT32_MAX, UINT32_MAX,
11873                     (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
11874         }
11875         mtx_unlock(&lun->lun_lock);
11876         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11877         return (0);
11878 }
11879
11880 static void
11881 ctl_i_t_nexus_loss(struct ctl_softc *softc, uint32_t initidx,
11882     ctl_ua_type ua_type)
11883 {
11884         struct ctl_lun *lun;
11885         struct scsi_sense_data *ps;
11886         uint32_t p, i;
11887
11888         p = initidx / CTL_MAX_INIT_PER_PORT;
11889         i = initidx % CTL_MAX_INIT_PER_PORT;
11890         mtx_lock(&softc->ctl_lock);
11891         STAILQ_FOREACH(lun, &softc->lun_list, links) {
11892                 mtx_lock(&lun->lun_lock);
11893                 /* Abort tasks. */
11894                 ctl_abort_tasks_lun(lun, p, i, 1);
11895                 /* Clear CA. */
11896                 ps = lun->pending_sense[p];
11897                 if (ps != NULL)
11898                         ps[i].error_code = 0;
11899                 /* Clear reservation. */
11900                 if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == initidx))
11901                         lun->flags &= ~CTL_LUN_RESERVED;
11902                 /* Clear prevent media removal. */
11903                 if (lun->prevent && ctl_is_set(lun->prevent, initidx)) {
11904                         ctl_clear_mask(lun->prevent, initidx);
11905                         lun->prevent_count--;
11906                 }
11907                 /* Clear TPC status */
11908                 ctl_tpc_lun_clear(lun, initidx);
11909                 /* Establish UA. */
11910                 ctl_est_ua(lun, initidx, ua_type);
11911                 mtx_unlock(&lun->lun_lock);
11912         }
11913         mtx_unlock(&softc->ctl_lock);
11914 }
11915
11916 static int
11917 ctl_i_t_nexus_reset(union ctl_io *io)
11918 {
11919         struct ctl_softc *softc = CTL_SOFTC(io);
11920         uint32_t initidx;
11921
11922         if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11923                 union ctl_ha_msg msg_info;
11924
11925                 msg_info.hdr.nexus = io->io_hdr.nexus;
11926                 msg_info.task.task_action = CTL_TASK_I_T_NEXUS_RESET;
11927                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11928                 msg_info.hdr.original_sc = NULL;
11929                 msg_info.hdr.serializing_sc = NULL;
11930                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11931                     sizeof(msg_info.task), M_WAITOK);
11932         }
11933
11934         initidx = ctl_get_initindex(&io->io_hdr.nexus);
11935         ctl_i_t_nexus_loss(softc, initidx, CTL_UA_I_T_NEXUS_LOSS);
11936         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11937         return (0);
11938 }
11939
11940 static int
11941 ctl_abort_task(union ctl_io *io)
11942 {
11943         struct ctl_softc *softc = CTL_SOFTC(io);
11944         union ctl_io *xio;
11945         struct ctl_lun *lun;
11946 #if 0
11947         struct sbuf sb;
11948         char printbuf[128];
11949 #endif
11950         int found;
11951         uint32_t targ_lun;
11952
11953         found = 0;
11954
11955         /*
11956          * Look up the LUN.
11957          */
11958         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11959         mtx_lock(&softc->ctl_lock);
11960         if (targ_lun >= ctl_max_luns ||
11961             (lun = softc->ctl_luns[targ_lun]) == NULL) {
11962                 mtx_unlock(&softc->ctl_lock);
11963                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
11964                 return (1);
11965         }
11966
11967 #if 0
11968         printf("ctl_abort_task: called for lun %lld, tag %d type %d\n",
11969                lun->lun, io->taskio.tag_num, io->taskio.tag_type);
11970 #endif
11971
11972         mtx_lock(&lun->lun_lock);
11973         mtx_unlock(&softc->ctl_lock);
11974         /*
11975          * Run through the OOA queue and attempt to find the given I/O.
11976          * The target port, initiator ID, tag type and tag number have to
11977          * match the values that we got from the initiator.  If we have an
11978          * untagged command to abort, simply abort the first untagged command
11979          * we come to.  We only allow one untagged command at a time of course.
11980          */
11981         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11982              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11983 #if 0
11984                 sbuf_new(&sb, printbuf, sizeof(printbuf), SBUF_FIXEDLEN);
11985
11986                 sbuf_printf(&sb, "LUN %lld tag %d type %d%s%s%s%s: ",
11987                             lun->lun, xio->scsiio.tag_num,
11988                             xio->scsiio.tag_type,
11989                             (xio->io_hdr.blocked_links.tqe_prev
11990                             == NULL) ? "" : " BLOCKED",
11991                             (xio->io_hdr.flags &
11992                             CTL_FLAG_DMA_INPROG) ? " DMA" : "",
11993                             (xio->io_hdr.flags &
11994                             CTL_FLAG_ABORT) ? " ABORT" : "",
11995                             (xio->io_hdr.flags &
11996                             CTL_FLAG_IS_WAS_ON_RTR ? " RTR" : ""));
11997                 ctl_scsi_command_string(&xio->scsiio, NULL, &sb);
11998                 sbuf_finish(&sb);
11999                 printf("%s\n", sbuf_data(&sb));
12000 #endif
12001
12002                 if ((xio->io_hdr.nexus.targ_port != io->io_hdr.nexus.targ_port)
12003                  || (xio->io_hdr.nexus.initid != io->io_hdr.nexus.initid)
12004                  || (xio->io_hdr.flags & CTL_FLAG_ABORT))
12005                         continue;
12006
12007                 /*
12008                  * If the abort says that the task is untagged, the
12009                  * task in the queue must be untagged.  Otherwise,
12010                  * we just check to see whether the tag numbers
12011                  * match.  This is because the QLogic firmware
12012                  * doesn't pass back the tag type in an abort
12013                  * request.
12014                  */
12015 #if 0
12016                 if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
12017                   && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
12018                  || (xio->scsiio.tag_num == io->taskio.tag_num))
12019 #endif
12020                 /*
12021                  * XXX KDM we've got problems with FC, because it
12022                  * doesn't send down a tag type with aborts.  So we
12023                  * can only really go by the tag number...
12024                  * This may cause problems with parallel SCSI.
12025                  * Need to figure that out!!
12026                  */
12027                 if (xio->scsiio.tag_num == io->taskio.tag_num) {
12028                         xio->io_hdr.flags |= CTL_FLAG_ABORT;
12029                         found = 1;
12030                         if ((io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) == 0 &&
12031                             !(lun->flags & CTL_LUN_PRIMARY_SC)) {
12032                                 union ctl_ha_msg msg_info;
12033
12034                                 msg_info.hdr.nexus = io->io_hdr.nexus;
12035                                 msg_info.task.task_action = CTL_TASK_ABORT_TASK;
12036                                 msg_info.task.tag_num = io->taskio.tag_num;
12037                                 msg_info.task.tag_type = io->taskio.tag_type;
12038                                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
12039                                 msg_info.hdr.original_sc = NULL;
12040                                 msg_info.hdr.serializing_sc = NULL;
12041 #if 0
12042                                 printf("Sent Abort to other side\n");
12043 #endif
12044                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
12045                                     sizeof(msg_info.task), M_NOWAIT);
12046                         }
12047 #if 0
12048                         printf("ctl_abort_task: found I/O to abort\n");
12049 #endif
12050                 }
12051         }
12052         mtx_unlock(&lun->lun_lock);
12053
12054         if (found == 0) {
12055                 /*
12056                  * This isn't really an error.  It's entirely possible for
12057                  * the abort and command completion to cross on the wire.
12058                  * This is more of an informative/diagnostic error.
12059                  */
12060 #if 0
12061                 printf("ctl_abort_task: ABORT sent for nonexistent I/O: "
12062                        "%u:%u:%u tag %d type %d\n",
12063                        io->io_hdr.nexus.initid,
12064                        io->io_hdr.nexus.targ_port,
12065                        io->io_hdr.nexus.targ_lun, io->taskio.tag_num,
12066                        io->taskio.tag_type);
12067 #endif
12068         }
12069         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12070         return (0);
12071 }
12072
12073 static int
12074 ctl_query_task(union ctl_io *io, int task_set)
12075 {
12076         struct ctl_softc *softc = CTL_SOFTC(io);
12077         union ctl_io *xio;
12078         struct ctl_lun *lun;
12079         int found = 0;
12080         uint32_t targ_lun;
12081
12082         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12083         mtx_lock(&softc->ctl_lock);
12084         if (targ_lun >= ctl_max_luns ||
12085             (lun = softc->ctl_luns[targ_lun]) == NULL) {
12086                 mtx_unlock(&softc->ctl_lock);
12087                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
12088                 return (1);
12089         }
12090         mtx_lock(&lun->lun_lock);
12091         mtx_unlock(&softc->ctl_lock);
12092         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12093              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12094
12095                 if ((xio->io_hdr.nexus.targ_port != io->io_hdr.nexus.targ_port)
12096                  || (xio->io_hdr.nexus.initid != io->io_hdr.nexus.initid)
12097                  || (xio->io_hdr.flags & CTL_FLAG_ABORT))
12098                         continue;
12099
12100                 if (task_set || xio->scsiio.tag_num == io->taskio.tag_num) {
12101                         found = 1;
12102                         break;
12103                 }
12104         }
12105         mtx_unlock(&lun->lun_lock);
12106         if (found)
12107                 io->taskio.task_status = CTL_TASK_FUNCTION_SUCCEEDED;
12108         else
12109                 io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12110         return (0);
12111 }
12112
12113 static int
12114 ctl_query_async_event(union ctl_io *io)
12115 {
12116         struct ctl_softc *softc = CTL_SOFTC(io);
12117         struct ctl_lun *lun;
12118         ctl_ua_type ua;
12119         uint32_t targ_lun, initidx;
12120
12121         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12122         mtx_lock(&softc->ctl_lock);
12123         if (targ_lun >= ctl_max_luns ||
12124             (lun = softc->ctl_luns[targ_lun]) == NULL) {
12125                 mtx_unlock(&softc->ctl_lock);
12126                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
12127                 return (1);
12128         }
12129         mtx_lock(&lun->lun_lock);
12130         mtx_unlock(&softc->ctl_lock);
12131         initidx = ctl_get_initindex(&io->io_hdr.nexus);
12132         ua = ctl_build_qae(lun, initidx, io->taskio.task_resp);
12133         mtx_unlock(&lun->lun_lock);
12134         if (ua != CTL_UA_NONE)
12135                 io->taskio.task_status = CTL_TASK_FUNCTION_SUCCEEDED;
12136         else
12137                 io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12138         return (0);
12139 }
12140
12141 static void
12142 ctl_run_task(union ctl_io *io)
12143 {
12144         int retval = 1;
12145
12146         CTL_DEBUG_PRINT(("ctl_run_task\n"));
12147         KASSERT(io->io_hdr.io_type == CTL_IO_TASK,
12148             ("ctl_run_task: Unextected io_type %d\n", io->io_hdr.io_type));
12149         io->taskio.task_status = CTL_TASK_FUNCTION_NOT_SUPPORTED;
12150         bzero(io->taskio.task_resp, sizeof(io->taskio.task_resp));
12151         switch (io->taskio.task_action) {
12152         case CTL_TASK_ABORT_TASK:
12153                 retval = ctl_abort_task(io);
12154                 break;
12155         case CTL_TASK_ABORT_TASK_SET:
12156         case CTL_TASK_CLEAR_TASK_SET:
12157                 retval = ctl_abort_task_set(io);
12158                 break;
12159         case CTL_TASK_CLEAR_ACA:
12160                 break;
12161         case CTL_TASK_I_T_NEXUS_RESET:
12162                 retval = ctl_i_t_nexus_reset(io);
12163                 break;
12164         case CTL_TASK_LUN_RESET:
12165                 retval = ctl_lun_reset(io);
12166                 break;
12167         case CTL_TASK_TARGET_RESET:
12168         case CTL_TASK_BUS_RESET:
12169                 retval = ctl_target_reset(io);
12170                 break;
12171         case CTL_TASK_PORT_LOGIN:
12172                 break;
12173         case CTL_TASK_PORT_LOGOUT:
12174                 break;
12175         case CTL_TASK_QUERY_TASK:
12176                 retval = ctl_query_task(io, 0);
12177                 break;
12178         case CTL_TASK_QUERY_TASK_SET:
12179                 retval = ctl_query_task(io, 1);
12180                 break;
12181         case CTL_TASK_QUERY_ASYNC_EVENT:
12182                 retval = ctl_query_async_event(io);
12183                 break;
12184         default:
12185                 printf("%s: got unknown task management event %d\n",
12186                        __func__, io->taskio.task_action);
12187                 break;
12188         }
12189         if (retval == 0)
12190                 io->io_hdr.status = CTL_SUCCESS;
12191         else
12192                 io->io_hdr.status = CTL_ERROR;
12193         ctl_done(io);
12194 }
12195
12196 /*
12197  * For HA operation.  Handle commands that come in from the other
12198  * controller.
12199  */
12200 static void
12201 ctl_handle_isc(union ctl_io *io)
12202 {
12203         struct ctl_softc *softc = CTL_SOFTC(io);
12204         struct ctl_lun *lun;
12205         const struct ctl_cmd_entry *entry;
12206         uint32_t targ_lun;
12207
12208         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12209         switch (io->io_hdr.msg_type) {
12210         case CTL_MSG_SERIALIZE:
12211                 ctl_serialize_other_sc_cmd(&io->scsiio);
12212                 break;
12213         case CTL_MSG_R2R:               /* Only used in SER_ONLY mode. */
12214                 entry = ctl_get_cmd_entry(&io->scsiio, NULL);
12215                 if (targ_lun >= ctl_max_luns ||
12216                     (lun = softc->ctl_luns[targ_lun]) == NULL) {
12217                         ctl_done(io);
12218                         break;
12219                 }
12220                 mtx_lock(&lun->lun_lock);
12221                 if (ctl_scsiio_lun_check(lun, entry, &io->scsiio) != 0) {
12222                         mtx_unlock(&lun->lun_lock);
12223                         ctl_done(io);
12224                         break;
12225                 }
12226                 io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
12227                 mtx_unlock(&lun->lun_lock);
12228                 ctl_enqueue_rtr(io);
12229                 break;
12230         case CTL_MSG_FINISH_IO:
12231                 if (softc->ha_mode == CTL_HA_MODE_XFER) {
12232                         ctl_done(io);
12233                         break;
12234                 }
12235                 if (targ_lun >= ctl_max_luns ||
12236                     (lun = softc->ctl_luns[targ_lun]) == NULL) {
12237                         ctl_free_io(io);
12238                         break;
12239                 }
12240                 mtx_lock(&lun->lun_lock);
12241                 TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
12242                 ctl_check_blocked(lun);
12243                 mtx_unlock(&lun->lun_lock);
12244                 ctl_free_io(io);
12245                 break;
12246         case CTL_MSG_PERS_ACTION:
12247                 ctl_hndl_per_res_out_on_other_sc(io);
12248                 ctl_free_io(io);
12249                 break;
12250         case CTL_MSG_BAD_JUJU:
12251                 ctl_done(io);
12252                 break;
12253         case CTL_MSG_DATAMOVE:          /* Only used in XFER mode */
12254                 ctl_datamove_remote(io);
12255                 break;
12256         case CTL_MSG_DATAMOVE_DONE:     /* Only used in XFER mode */
12257                 io->scsiio.be_move_done(io);
12258                 break;
12259         case CTL_MSG_FAILOVER:
12260                 ctl_failover_lun(io);
12261                 ctl_free_io(io);
12262                 break;
12263         default:
12264                 printf("%s: Invalid message type %d\n",
12265                        __func__, io->io_hdr.msg_type);
12266                 ctl_free_io(io);
12267                 break;
12268         }
12269
12270 }
12271
12272
12273 /*
12274  * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
12275  * there is no match.
12276  */
12277 static ctl_lun_error_pattern
12278 ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
12279 {
12280         const struct ctl_cmd_entry *entry;
12281         ctl_lun_error_pattern filtered_pattern, pattern;
12282
12283         pattern = desc->error_pattern;
12284
12285         /*
12286          * XXX KDM we need more data passed into this function to match a
12287          * custom pattern, and we actually need to implement custom pattern
12288          * matching.
12289          */
12290         if (pattern & CTL_LUN_PAT_CMD)
12291                 return (CTL_LUN_PAT_CMD);
12292
12293         if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
12294                 return (CTL_LUN_PAT_ANY);
12295
12296         entry = ctl_get_cmd_entry(ctsio, NULL);
12297
12298         filtered_pattern = entry->pattern & pattern;
12299
12300         /*
12301          * If the user requested specific flags in the pattern (e.g.
12302          * CTL_LUN_PAT_RANGE), make sure the command supports all of those
12303          * flags.
12304          *
12305          * If the user did not specify any flags, it doesn't matter whether
12306          * or not the command supports the flags.
12307          */
12308         if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
12309              (pattern & ~CTL_LUN_PAT_MASK))
12310                 return (CTL_LUN_PAT_NONE);
12311
12312         /*
12313          * If the user asked for a range check, see if the requested LBA
12314          * range overlaps with this command's LBA range.
12315          */
12316         if (filtered_pattern & CTL_LUN_PAT_RANGE) {
12317                 uint64_t lba1;
12318                 uint64_t len1;
12319                 ctl_action action;
12320                 int retval;
12321
12322                 retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
12323                 if (retval != 0)
12324                         return (CTL_LUN_PAT_NONE);
12325
12326                 action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
12327                                               desc->lba_range.len, FALSE);
12328                 /*
12329                  * A "pass" means that the LBA ranges don't overlap, so
12330                  * this doesn't match the user's range criteria.
12331                  */
12332                 if (action == CTL_ACTION_PASS)
12333                         return (CTL_LUN_PAT_NONE);
12334         }
12335
12336         return (filtered_pattern);
12337 }
12338
12339 static void
12340 ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
12341 {
12342         struct ctl_error_desc *desc, *desc2;
12343
12344         mtx_assert(&lun->lun_lock, MA_OWNED);
12345
12346         STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
12347                 ctl_lun_error_pattern pattern;
12348                 /*
12349                  * Check to see whether this particular command matches
12350                  * the pattern in the descriptor.
12351                  */
12352                 pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
12353                 if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
12354                         continue;
12355
12356                 switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
12357                 case CTL_LUN_INJ_ABORTED:
12358                         ctl_set_aborted(&io->scsiio);
12359                         break;
12360                 case CTL_LUN_INJ_MEDIUM_ERR:
12361                         ctl_set_medium_error(&io->scsiio,
12362                             (io->io_hdr.flags & CTL_FLAG_DATA_MASK) !=
12363                              CTL_FLAG_DATA_OUT);
12364                         break;
12365                 case CTL_LUN_INJ_UA:
12366                         /* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
12367                          * OCCURRED */
12368                         ctl_set_ua(&io->scsiio, 0x29, 0x00);
12369                         break;
12370                 case CTL_LUN_INJ_CUSTOM:
12371                         /*
12372                          * We're assuming the user knows what he is doing.
12373                          * Just copy the sense information without doing
12374                          * checks.
12375                          */
12376                         bcopy(&desc->custom_sense, &io->scsiio.sense_data,
12377                               MIN(sizeof(desc->custom_sense),
12378                                   sizeof(io->scsiio.sense_data)));
12379                         io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
12380                         io->scsiio.sense_len = SSD_FULL_SIZE;
12381                         io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
12382                         break;
12383                 case CTL_LUN_INJ_NONE:
12384                 default:
12385                         /*
12386                          * If this is an error injection type we don't know
12387                          * about, clear the continuous flag (if it is set)
12388                          * so it will get deleted below.
12389                          */
12390                         desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
12391                         break;
12392                 }
12393                 /*
12394                  * By default, each error injection action is a one-shot
12395                  */
12396                 if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
12397                         continue;
12398
12399                 STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
12400
12401                 free(desc, M_CTL);
12402         }
12403 }
12404
12405 #ifdef CTL_IO_DELAY
12406 static void
12407 ctl_datamove_timer_wakeup(void *arg)
12408 {
12409         union ctl_io *io;
12410
12411         io = (union ctl_io *)arg;
12412
12413         ctl_datamove(io);
12414 }
12415 #endif /* CTL_IO_DELAY */
12416
12417 void
12418 ctl_datamove(union ctl_io *io)
12419 {
12420         void (*fe_datamove)(union ctl_io *io);
12421
12422         mtx_assert(&((struct ctl_softc *)CTL_SOFTC(io))->ctl_lock, MA_NOTOWNED);
12423
12424         CTL_DEBUG_PRINT(("ctl_datamove\n"));
12425
12426         /* No data transferred yet.  Frontend must update this when done. */
12427         io->scsiio.kern_data_resid = io->scsiio.kern_data_len;
12428
12429 #ifdef CTL_TIME_IO
12430         if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12431                 char str[256];
12432                 char path_str[64];
12433                 struct sbuf sb;
12434
12435                 ctl_scsi_path_string(io, path_str, sizeof(path_str));
12436                 sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12437
12438                 sbuf_cat(&sb, path_str);
12439                 switch (io->io_hdr.io_type) {
12440                 case CTL_IO_SCSI:
12441                         ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12442                         sbuf_printf(&sb, "\n");
12443                         sbuf_cat(&sb, path_str);
12444                         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12445                                     io->scsiio.tag_num, io->scsiio.tag_type);
12446                         break;
12447                 case CTL_IO_TASK:
12448                         sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12449                                     "Tag Type: %d\n", io->taskio.task_action,
12450                                     io->taskio.tag_num, io->taskio.tag_type);
12451                         break;
12452                 default:
12453                         panic("%s: Invalid CTL I/O type %d\n",
12454                             __func__, io->io_hdr.io_type);
12455                 }
12456                 sbuf_cat(&sb, path_str);
12457                 sbuf_printf(&sb, "ctl_datamove: %jd seconds\n",
12458                             (intmax_t)time_uptime - io->io_hdr.start_time);
12459                 sbuf_finish(&sb);
12460                 printf("%s", sbuf_data(&sb));
12461         }
12462 #endif /* CTL_TIME_IO */
12463
12464 #ifdef CTL_IO_DELAY
12465         if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
12466                 io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
12467         } else {
12468                 struct ctl_lun *lun;
12469
12470                 lun = CTL_LUN(io);
12471                 if ((lun != NULL)
12472                  && (lun->delay_info.datamove_delay > 0)) {
12473
12474                         callout_init(&io->io_hdr.delay_callout, /*mpsafe*/ 1);
12475                         io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
12476                         callout_reset(&io->io_hdr.delay_callout,
12477                                       lun->delay_info.datamove_delay * hz,
12478                                       ctl_datamove_timer_wakeup, io);
12479                         if (lun->delay_info.datamove_type ==
12480                             CTL_DELAY_TYPE_ONESHOT)
12481                                 lun->delay_info.datamove_delay = 0;
12482                         return;
12483                 }
12484         }
12485 #endif
12486
12487         /*
12488          * This command has been aborted.  Set the port status, so we fail
12489          * the data move.
12490          */
12491         if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12492                 printf("ctl_datamove: tag 0x%04x on (%u:%u:%u) aborted\n",
12493                        io->scsiio.tag_num, io->io_hdr.nexus.initid,
12494                        io->io_hdr.nexus.targ_port,
12495                        io->io_hdr.nexus.targ_lun);
12496                 io->io_hdr.port_status = 31337;
12497                 /*
12498                  * Note that the backend, in this case, will get the
12499                  * callback in its context.  In other cases it may get
12500                  * called in the frontend's interrupt thread context.
12501                  */
12502                 io->scsiio.be_move_done(io);
12503                 return;
12504         }
12505
12506         /* Don't confuse frontend with zero length data move. */
12507         if (io->scsiio.kern_data_len == 0) {
12508                 io->scsiio.be_move_done(io);
12509                 return;
12510         }
12511
12512         fe_datamove = CTL_PORT(io)->fe_datamove;
12513         fe_datamove(io);
12514 }
12515
12516 static void
12517 ctl_send_datamove_done(union ctl_io *io, int have_lock)
12518 {
12519         union ctl_ha_msg msg;
12520 #ifdef CTL_TIME_IO
12521         struct bintime cur_bt;
12522 #endif
12523
12524         memset(&msg, 0, sizeof(msg));
12525         msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
12526         msg.hdr.original_sc = io;
12527         msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
12528         msg.hdr.nexus = io->io_hdr.nexus;
12529         msg.hdr.status = io->io_hdr.status;
12530         msg.scsi.kern_data_resid = io->scsiio.kern_data_resid;
12531         msg.scsi.tag_num = io->scsiio.tag_num;
12532         msg.scsi.tag_type = io->scsiio.tag_type;
12533         msg.scsi.scsi_status = io->scsiio.scsi_status;
12534         memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
12535                io->scsiio.sense_len);
12536         msg.scsi.sense_len = io->scsiio.sense_len;
12537         msg.scsi.port_status = io->io_hdr.port_status;
12538         io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12539         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12540                 ctl_failover_io(io, /*have_lock*/ have_lock);
12541                 return;
12542         }
12543         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12544             sizeof(msg.scsi) - sizeof(msg.scsi.sense_data) +
12545             msg.scsi.sense_len, M_WAITOK);
12546
12547 #ifdef CTL_TIME_IO
12548         getbinuptime(&cur_bt);
12549         bintime_sub(&cur_bt, &io->io_hdr.dma_start_bt);
12550         bintime_add(&io->io_hdr.dma_bt, &cur_bt);
12551 #endif
12552         io->io_hdr.num_dmas++;
12553 }
12554
12555 /*
12556  * The DMA to the remote side is done, now we need to tell the other side
12557  * we're done so it can continue with its data movement.
12558  */
12559 static void
12560 ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
12561 {
12562         union ctl_io *io;
12563         uint32_t i;
12564
12565         io = rq->context;
12566
12567         if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12568                 printf("%s: ISC DMA write failed with error %d", __func__,
12569                        rq->ret);
12570                 ctl_set_internal_failure(&io->scsiio,
12571                                          /*sks_valid*/ 1,
12572                                          /*retry_count*/ rq->ret);
12573         }
12574
12575         ctl_dt_req_free(rq);
12576
12577         for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12578                 free(io->io_hdr.local_sglist[i].addr, M_CTL);
12579         free(io->io_hdr.remote_sglist, M_CTL);
12580         io->io_hdr.remote_sglist = NULL;
12581         io->io_hdr.local_sglist = NULL;
12582
12583         /*
12584          * The data is in local and remote memory, so now we need to send
12585          * status (good or back) back to the other side.
12586          */
12587         ctl_send_datamove_done(io, /*have_lock*/ 0);
12588 }
12589
12590 /*
12591  * We've moved the data from the host/controller into local memory.  Now we
12592  * need to push it over to the remote controller's memory.
12593  */
12594 static int
12595 ctl_datamove_remote_dm_write_cb(union ctl_io *io)
12596 {
12597         int retval;
12598
12599         retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
12600                                           ctl_datamove_remote_write_cb);
12601         return (retval);
12602 }
12603
12604 static void
12605 ctl_datamove_remote_write(union ctl_io *io)
12606 {
12607         int retval;
12608         void (*fe_datamove)(union ctl_io *io);
12609
12610         /*
12611          * - Get the data from the host/HBA into local memory.
12612          * - DMA memory from the local controller to the remote controller.
12613          * - Send status back to the remote controller.
12614          */
12615
12616         retval = ctl_datamove_remote_sgl_setup(io);
12617         if (retval != 0)
12618                 return;
12619
12620         /* Switch the pointer over so the FETD knows what to do */
12621         io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12622
12623         /*
12624          * Use a custom move done callback, since we need to send completion
12625          * back to the other controller, not to the backend on this side.
12626          */
12627         io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
12628
12629         fe_datamove = CTL_PORT(io)->fe_datamove;
12630         fe_datamove(io);
12631 }
12632
12633 static int
12634 ctl_datamove_remote_dm_read_cb(union ctl_io *io)
12635 {
12636 #if 0
12637         char str[256];
12638         char path_str[64];
12639         struct sbuf sb;
12640 #endif
12641         uint32_t i;
12642
12643         for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12644                 free(io->io_hdr.local_sglist[i].addr, M_CTL);
12645         free(io->io_hdr.remote_sglist, M_CTL);
12646         io->io_hdr.remote_sglist = NULL;
12647         io->io_hdr.local_sglist = NULL;
12648
12649 #if 0
12650         scsi_path_string(io, path_str, sizeof(path_str));
12651         sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12652         sbuf_cat(&sb, path_str);
12653         scsi_command_string(&io->scsiio, NULL, &sb);
12654         sbuf_printf(&sb, "\n");
12655         sbuf_cat(&sb, path_str);
12656         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12657                     io->scsiio.tag_num, io->scsiio.tag_type);
12658         sbuf_cat(&sb, path_str);
12659         sbuf_printf(&sb, "%s: flags %#x, status %#x\n", __func__,
12660                     io->io_hdr.flags, io->io_hdr.status);
12661         sbuf_finish(&sb);
12662         printk("%s", sbuf_data(&sb));
12663 #endif
12664
12665
12666         /*
12667          * The read is done, now we need to send status (good or bad) back
12668          * to the other side.
12669          */
12670         ctl_send_datamove_done(io, /*have_lock*/ 0);
12671
12672         return (0);
12673 }
12674
12675 static void
12676 ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
12677 {
12678         union ctl_io *io;
12679         void (*fe_datamove)(union ctl_io *io);
12680
12681         io = rq->context;
12682
12683         if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12684                 printf("%s: ISC DMA read failed with error %d\n", __func__,
12685                        rq->ret);
12686                 ctl_set_internal_failure(&io->scsiio,
12687                                          /*sks_valid*/ 1,
12688                                          /*retry_count*/ rq->ret);
12689         }
12690
12691         ctl_dt_req_free(rq);
12692
12693         /* Switch the pointer over so the FETD knows what to do */
12694         io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12695
12696         /*
12697          * Use a custom move done callback, since we need to send completion
12698          * back to the other controller, not to the backend on this side.
12699          */
12700         io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
12701
12702         /* XXX KDM add checks like the ones in ctl_datamove? */
12703
12704         fe_datamove = CTL_PORT(io)->fe_datamove;
12705         fe_datamove(io);
12706 }
12707
12708 static int
12709 ctl_datamove_remote_sgl_setup(union ctl_io *io)
12710 {
12711         struct ctl_sg_entry *local_sglist;
12712         uint32_t len_to_go;
12713         int retval;
12714         int i;
12715
12716         retval = 0;
12717         local_sglist = io->io_hdr.local_sglist;
12718         len_to_go = io->scsiio.kern_data_len;
12719
12720         /*
12721          * The difficult thing here is that the size of the various
12722          * S/G segments may be different than the size from the
12723          * remote controller.  That'll make it harder when DMAing
12724          * the data back to the other side.
12725          */
12726         for (i = 0; len_to_go > 0; i++) {
12727                 local_sglist[i].len = MIN(len_to_go, CTL_HA_DATAMOVE_SEGMENT);
12728                 local_sglist[i].addr =
12729                     malloc(local_sglist[i].len, M_CTL, M_WAITOK);
12730
12731                 len_to_go -= local_sglist[i].len;
12732         }
12733         /*
12734          * Reset the number of S/G entries accordingly.  The original
12735          * number of S/G entries is available in rem_sg_entries.
12736          */
12737         io->scsiio.kern_sg_entries = i;
12738
12739 #if 0
12740         printf("%s: kern_sg_entries = %d\n", __func__,
12741                io->scsiio.kern_sg_entries);
12742         for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12743                 printf("%s: sg[%d] = %p, %lu\n", __func__, i,
12744                        local_sglist[i].addr, local_sglist[i].len);
12745 #endif
12746
12747         return (retval);
12748 }
12749
12750 static int
12751 ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
12752                          ctl_ha_dt_cb callback)
12753 {
12754         struct ctl_ha_dt_req *rq;
12755         struct ctl_sg_entry *remote_sglist, *local_sglist;
12756         uint32_t local_used, remote_used, total_used;
12757         int i, j, isc_ret;
12758
12759         rq = ctl_dt_req_alloc();
12760
12761         /*
12762          * If we failed to allocate the request, and if the DMA didn't fail
12763          * anyway, set busy status.  This is just a resource allocation
12764          * failure.
12765          */
12766         if ((rq == NULL)
12767          && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
12768              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS))
12769                 ctl_set_busy(&io->scsiio);
12770
12771         if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
12772             (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS) {
12773
12774                 if (rq != NULL)
12775                         ctl_dt_req_free(rq);
12776
12777                 /*
12778                  * The data move failed.  We need to return status back
12779                  * to the other controller.  No point in trying to DMA
12780                  * data to the remote controller.
12781                  */
12782
12783                 ctl_send_datamove_done(io, /*have_lock*/ 0);
12784
12785                 return (1);
12786         }
12787
12788         local_sglist = io->io_hdr.local_sglist;
12789         remote_sglist = io->io_hdr.remote_sglist;
12790         local_used = 0;
12791         remote_used = 0;
12792         total_used = 0;
12793
12794         /*
12795          * Pull/push the data over the wire from/to the other controller.
12796          * This takes into account the possibility that the local and
12797          * remote sglists may not be identical in terms of the size of
12798          * the elements and the number of elements.
12799          *
12800          * One fundamental assumption here is that the length allocated for
12801          * both the local and remote sglists is identical.  Otherwise, we've
12802          * essentially got a coding error of some sort.
12803          */
12804         isc_ret = CTL_HA_STATUS_SUCCESS;
12805         for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
12806                 uint32_t cur_len;
12807                 uint8_t *tmp_ptr;
12808
12809                 rq->command = command;
12810                 rq->context = io;
12811
12812                 /*
12813                  * Both pointers should be aligned.  But it is possible
12814                  * that the allocation length is not.  They should both
12815                  * also have enough slack left over at the end, though,
12816                  * to round up to the next 8 byte boundary.
12817                  */
12818                 cur_len = MIN(local_sglist[i].len - local_used,
12819                               remote_sglist[j].len - remote_used);
12820                 rq->size = cur_len;
12821
12822                 tmp_ptr = (uint8_t *)local_sglist[i].addr;
12823                 tmp_ptr += local_used;
12824
12825 #if 0
12826                 /* Use physical addresses when talking to ISC hardware */
12827                 if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
12828                         /* XXX KDM use busdma */
12829                         rq->local = vtophys(tmp_ptr);
12830                 } else
12831                         rq->local = tmp_ptr;
12832 #else
12833                 KASSERT((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0,
12834                     ("HA does not support BUS_ADDR"));
12835                 rq->local = tmp_ptr;
12836 #endif
12837
12838                 tmp_ptr = (uint8_t *)remote_sglist[j].addr;
12839                 tmp_ptr += remote_used;
12840                 rq->remote = tmp_ptr;
12841
12842                 rq->callback = NULL;
12843
12844                 local_used += cur_len;
12845                 if (local_used >= local_sglist[i].len) {
12846                         i++;
12847                         local_used = 0;
12848                 }
12849
12850                 remote_used += cur_len;
12851                 if (remote_used >= remote_sglist[j].len) {
12852                         j++;
12853                         remote_used = 0;
12854                 }
12855                 total_used += cur_len;
12856
12857                 if (total_used >= io->scsiio.kern_data_len)
12858                         rq->callback = callback;
12859
12860 #if 0
12861                 printf("%s: %s: local %p remote %p size %d\n", __func__,
12862                        (command == CTL_HA_DT_CMD_WRITE) ? "WRITE" : "READ",
12863                        rq->local, rq->remote, rq->size);
12864 #endif
12865
12866                 isc_ret = ctl_dt_single(rq);
12867                 if (isc_ret > CTL_HA_STATUS_SUCCESS)
12868                         break;
12869         }
12870         if (isc_ret != CTL_HA_STATUS_WAIT) {
12871                 rq->ret = isc_ret;
12872                 callback(rq);
12873         }
12874
12875         return (0);
12876 }
12877
12878 static void
12879 ctl_datamove_remote_read(union ctl_io *io)
12880 {
12881         int retval;
12882         uint32_t i;
12883
12884         /*
12885          * This will send an error to the other controller in the case of a
12886          * failure.
12887          */
12888         retval = ctl_datamove_remote_sgl_setup(io);
12889         if (retval != 0)
12890                 return;
12891
12892         retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
12893                                           ctl_datamove_remote_read_cb);
12894         if (retval != 0) {
12895                 /*
12896                  * Make sure we free memory if there was an error..  The
12897                  * ctl_datamove_remote_xfer() function will send the
12898                  * datamove done message, or call the callback with an
12899                  * error if there is a problem.
12900                  */
12901                 for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12902                         free(io->io_hdr.local_sglist[i].addr, M_CTL);
12903                 free(io->io_hdr.remote_sglist, M_CTL);
12904                 io->io_hdr.remote_sglist = NULL;
12905                 io->io_hdr.local_sglist = NULL;
12906         }
12907 }
12908
12909 /*
12910  * Process a datamove request from the other controller.  This is used for
12911  * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
12912  * first.  Once that is complete, the data gets DMAed into the remote
12913  * controller's memory.  For reads, we DMA from the remote controller's
12914  * memory into our memory first, and then move it out to the FETD.
12915  */
12916 static void
12917 ctl_datamove_remote(union ctl_io *io)
12918 {
12919
12920         mtx_assert(&((struct ctl_softc *)CTL_SOFTC(io))->ctl_lock, MA_NOTOWNED);
12921
12922         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12923                 ctl_failover_io(io, /*have_lock*/ 0);
12924                 return;
12925         }
12926
12927         /*
12928          * Note that we look for an aborted I/O here, but don't do some of
12929          * the other checks that ctl_datamove() normally does.
12930          * We don't need to run the datamove delay code, since that should
12931          * have been done if need be on the other controller.
12932          */
12933         if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12934                 printf("%s: tag 0x%04x on (%u:%u:%u) aborted\n", __func__,
12935                        io->scsiio.tag_num, io->io_hdr.nexus.initid,
12936                        io->io_hdr.nexus.targ_port,
12937                        io->io_hdr.nexus.targ_lun);
12938                 io->io_hdr.port_status = 31338;
12939                 ctl_send_datamove_done(io, /*have_lock*/ 0);
12940                 return;
12941         }
12942
12943         if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT)
12944                 ctl_datamove_remote_write(io);
12945         else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN)
12946                 ctl_datamove_remote_read(io);
12947         else {
12948                 io->io_hdr.port_status = 31339;
12949                 ctl_send_datamove_done(io, /*have_lock*/ 0);
12950         }
12951 }
12952
12953 static void
12954 ctl_process_done(union ctl_io *io)
12955 {
12956         struct ctl_softc *softc = CTL_SOFTC(io);
12957         struct ctl_port *port = CTL_PORT(io);
12958         struct ctl_lun *lun = CTL_LUN(io);
12959         void (*fe_done)(union ctl_io *io);
12960         union ctl_ha_msg msg;
12961
12962         CTL_DEBUG_PRINT(("ctl_process_done\n"));
12963         fe_done = port->fe_done;
12964
12965 #ifdef CTL_TIME_IO
12966         if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12967                 char str[256];
12968                 char path_str[64];
12969                 struct sbuf sb;
12970
12971                 ctl_scsi_path_string(io, path_str, sizeof(path_str));
12972                 sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12973
12974                 sbuf_cat(&sb, path_str);
12975                 switch (io->io_hdr.io_type) {
12976                 case CTL_IO_SCSI:
12977                         ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12978                         sbuf_printf(&sb, "\n");
12979                         sbuf_cat(&sb, path_str);
12980                         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12981                                     io->scsiio.tag_num, io->scsiio.tag_type);
12982                         break;
12983                 case CTL_IO_TASK:
12984                         sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12985                                     "Tag Type: %d\n", io->taskio.task_action,
12986                                     io->taskio.tag_num, io->taskio.tag_type);
12987                         break;
12988                 default:
12989                         panic("%s: Invalid CTL I/O type %d\n",
12990                             __func__, io->io_hdr.io_type);
12991                 }
12992                 sbuf_cat(&sb, path_str);
12993                 sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
12994                             (intmax_t)time_uptime - io->io_hdr.start_time);
12995                 sbuf_finish(&sb);
12996                 printf("%s", sbuf_data(&sb));
12997         }
12998 #endif /* CTL_TIME_IO */
12999
13000         switch (io->io_hdr.io_type) {
13001         case CTL_IO_SCSI:
13002                 break;
13003         case CTL_IO_TASK:
13004                 if (ctl_debug & CTL_DEBUG_INFO)
13005                         ctl_io_error_print(io, NULL);
13006                 fe_done(io);
13007                 return;
13008         default:
13009                 panic("%s: Invalid CTL I/O type %d\n",
13010                     __func__, io->io_hdr.io_type);
13011         }
13012
13013         if (lun == NULL) {
13014                 CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
13015                                  io->io_hdr.nexus.targ_mapped_lun));
13016                 goto bailout;
13017         }
13018
13019         mtx_lock(&lun->lun_lock);
13020
13021         /*
13022          * Check to see if we have any informational exception and status
13023          * of this command can be modified to report it in form of either
13024          * RECOVERED ERROR or NO SENSE, depending on MRIE mode page field.
13025          */
13026         if (lun->ie_reported == 0 && lun->ie_asc != 0 &&
13027             io->io_hdr.status == CTL_SUCCESS &&
13028             (io->io_hdr.flags & CTL_FLAG_STATUS_SENT) == 0) {
13029                 uint8_t mrie = lun->MODE_IE.mrie;
13030                 uint8_t per = ((lun->MODE_RWER.byte3 & SMS_RWER_PER) ||
13031                     (lun->MODE_VER.byte3 & SMS_VER_PER));
13032                 if (((mrie == SIEP_MRIE_REC_COND && per) ||
13033                      mrie == SIEP_MRIE_REC_UNCOND ||
13034                      mrie == SIEP_MRIE_NO_SENSE) &&
13035                     (ctl_get_cmd_entry(&io->scsiio, NULL)->flags &
13036                      CTL_CMD_FLAG_NO_SENSE) == 0) {
13037                         ctl_set_sense(&io->scsiio,
13038                               /*current_error*/ 1,
13039                               /*sense_key*/ (mrie == SIEP_MRIE_NO_SENSE) ?
13040                                 SSD_KEY_NO_SENSE : SSD_KEY_RECOVERED_ERROR,
13041                               /*asc*/ lun->ie_asc,
13042                               /*ascq*/ lun->ie_ascq,
13043                               SSD_ELEM_NONE);
13044                         lun->ie_reported = 1;
13045                 }
13046         } else if (lun->ie_reported < 0)
13047                 lun->ie_reported = 0;
13048
13049         /*
13050          * Check to see if we have any errors to inject here.  We only
13051          * inject errors for commands that don't already have errors set.
13052          */
13053         if (!STAILQ_EMPTY(&lun->error_list) &&
13054             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS) &&
13055             ((io->io_hdr.flags & CTL_FLAG_STATUS_SENT) == 0))
13056                 ctl_inject_error(lun, io);
13057
13058         /*
13059          * XXX KDM how do we treat commands that aren't completed
13060          * successfully?
13061          *
13062          * XXX KDM should we also track I/O latency?
13063          */
13064         if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS &&
13065             io->io_hdr.io_type == CTL_IO_SCSI) {
13066                 int type;
13067 #ifdef CTL_TIME_IO
13068                 struct bintime bt;
13069
13070                 getbinuptime(&bt);
13071                 bintime_sub(&bt, &io->io_hdr.start_bt);
13072 #endif
13073                 if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13074                     CTL_FLAG_DATA_IN)
13075                         type = CTL_STATS_READ;
13076                 else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13077                     CTL_FLAG_DATA_OUT)
13078                         type = CTL_STATS_WRITE;
13079                 else
13080                         type = CTL_STATS_NO_IO;
13081
13082 #ifdef CTL_LEGACY_STATS
13083                 uint32_t targ_port = port->targ_port;
13084                 lun->legacy_stats.ports[targ_port].bytes[type] +=
13085                     io->scsiio.kern_total_len;
13086                 lun->legacy_stats.ports[targ_port].operations[type] ++;
13087                 lun->legacy_stats.ports[targ_port].num_dmas[type] +=
13088                     io->io_hdr.num_dmas;
13089 #ifdef CTL_TIME_IO
13090                 bintime_add(&lun->legacy_stats.ports[targ_port].dma_time[type],
13091                    &io->io_hdr.dma_bt);
13092                 bintime_add(&lun->legacy_stats.ports[targ_port].time[type],
13093                     &bt);
13094 #endif
13095 #endif /* CTL_LEGACY_STATS */
13096
13097                 lun->stats.bytes[type] += io->scsiio.kern_total_len;
13098                 lun->stats.operations[type] ++;
13099                 lun->stats.dmas[type] += io->io_hdr.num_dmas;
13100 #ifdef CTL_TIME_IO
13101                 bintime_add(&lun->stats.dma_time[type], &io->io_hdr.dma_bt);
13102                 bintime_add(&lun->stats.time[type], &bt);
13103 #endif
13104
13105                 mtx_lock(&port->port_lock);
13106                 port->stats.bytes[type] += io->scsiio.kern_total_len;
13107                 port->stats.operations[type] ++;
13108                 port->stats.dmas[type] += io->io_hdr.num_dmas;
13109 #ifdef CTL_TIME_IO
13110                 bintime_add(&port->stats.dma_time[type], &io->io_hdr.dma_bt);
13111                 bintime_add(&port->stats.time[type], &bt);
13112 #endif
13113                 mtx_unlock(&port->port_lock);
13114         }
13115
13116         /*
13117          * Remove this from the OOA queue.
13118          */
13119         TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
13120 #ifdef CTL_TIME_IO
13121         if (TAILQ_EMPTY(&lun->ooa_queue))
13122                 lun->last_busy = getsbinuptime();
13123 #endif
13124
13125         /*
13126          * Run through the blocked queue on this LUN and see if anything
13127          * has become unblocked, now that this transaction is done.
13128          */
13129         ctl_check_blocked(lun);
13130
13131         /*
13132          * If the LUN has been invalidated, free it if there is nothing
13133          * left on its OOA queue.
13134          */
13135         if ((lun->flags & CTL_LUN_INVALID)
13136          && TAILQ_EMPTY(&lun->ooa_queue)) {
13137                 mtx_unlock(&lun->lun_lock);
13138                 ctl_free_lun(lun);
13139         } else
13140                 mtx_unlock(&lun->lun_lock);
13141
13142 bailout:
13143
13144         /*
13145          * If this command has been aborted, make sure we set the status
13146          * properly.  The FETD is responsible for freeing the I/O and doing
13147          * whatever it needs to do to clean up its state.
13148          */
13149         if (io->io_hdr.flags & CTL_FLAG_ABORT)
13150                 ctl_set_task_aborted(&io->scsiio);
13151
13152         /*
13153          * If enabled, print command error status.
13154          */
13155         if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS &&
13156             (ctl_debug & CTL_DEBUG_INFO) != 0)
13157                 ctl_io_error_print(io, NULL);
13158
13159         /*
13160          * Tell the FETD or the other shelf controller we're done with this
13161          * command.  Note that only SCSI commands get to this point.  Task
13162          * management commands are completed above.
13163          */
13164         if ((softc->ha_mode != CTL_HA_MODE_XFER) &&
13165             (io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)) {
13166                 memset(&msg, 0, sizeof(msg));
13167                 msg.hdr.msg_type = CTL_MSG_FINISH_IO;
13168                 msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
13169                 msg.hdr.nexus = io->io_hdr.nexus;
13170                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13171                     sizeof(msg.scsi) - sizeof(msg.scsi.sense_data),
13172                     M_WAITOK);
13173         }
13174
13175         fe_done(io);
13176 }
13177
13178 /*
13179  * Front end should call this if it doesn't do autosense.  When the request
13180  * sense comes back in from the initiator, we'll dequeue this and send it.
13181  */
13182 int
13183 ctl_queue_sense(union ctl_io *io)
13184 {
13185         struct ctl_softc *softc = CTL_SOFTC(io);
13186         struct ctl_port *port = CTL_PORT(io);
13187         struct ctl_lun *lun;
13188         struct scsi_sense_data *ps;
13189         uint32_t initidx, p, targ_lun;
13190
13191         CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
13192
13193         targ_lun = ctl_lun_map_from_port(port, io->io_hdr.nexus.targ_lun);
13194
13195         /*
13196          * LUN lookup will likely move to the ctl_work_thread() once we
13197          * have our new queueing infrastructure (that doesn't put things on
13198          * a per-LUN queue initially).  That is so that we can handle
13199          * things like an INQUIRY to a LUN that we don't have enabled.  We
13200          * can't deal with that right now.
13201          * If we don't have a LUN for this, just toss the sense information.
13202          */
13203         mtx_lock(&softc->ctl_lock);
13204         if (targ_lun >= ctl_max_luns ||
13205             (lun = softc->ctl_luns[targ_lun]) == NULL) {
13206                 mtx_unlock(&softc->ctl_lock);
13207                 goto bailout;
13208         }
13209         mtx_lock(&lun->lun_lock);
13210         mtx_unlock(&softc->ctl_lock);
13211
13212         initidx = ctl_get_initindex(&io->io_hdr.nexus);
13213         p = initidx / CTL_MAX_INIT_PER_PORT;
13214         if (lun->pending_sense[p] == NULL) {
13215                 lun->pending_sense[p] = malloc(sizeof(*ps) * CTL_MAX_INIT_PER_PORT,
13216                     M_CTL, M_NOWAIT | M_ZERO);
13217         }
13218         if ((ps = lun->pending_sense[p]) != NULL) {
13219                 ps += initidx % CTL_MAX_INIT_PER_PORT;
13220                 memset(ps, 0, sizeof(*ps));
13221                 memcpy(ps, &io->scsiio.sense_data, io->scsiio.sense_len);
13222         }
13223         mtx_unlock(&lun->lun_lock);
13224
13225 bailout:
13226         ctl_free_io(io);
13227         return (CTL_RETVAL_COMPLETE);
13228 }
13229
13230 /*
13231  * Primary command inlet from frontend ports.  All SCSI and task I/O
13232  * requests must go through this function.
13233  */
13234 int
13235 ctl_queue(union ctl_io *io)
13236 {
13237         struct ctl_port *port = CTL_PORT(io);
13238
13239         CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
13240
13241 #ifdef CTL_TIME_IO
13242         io->io_hdr.start_time = time_uptime;
13243         getbinuptime(&io->io_hdr.start_bt);
13244 #endif /* CTL_TIME_IO */
13245
13246         /* Map FE-specific LUN ID into global one. */
13247         io->io_hdr.nexus.targ_mapped_lun =
13248             ctl_lun_map_from_port(port, io->io_hdr.nexus.targ_lun);
13249
13250         switch (io->io_hdr.io_type) {
13251         case CTL_IO_SCSI:
13252         case CTL_IO_TASK:
13253                 if (ctl_debug & CTL_DEBUG_CDB)
13254                         ctl_io_print(io);
13255                 ctl_enqueue_incoming(io);
13256                 break;
13257         default:
13258                 printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
13259                 return (EINVAL);
13260         }
13261
13262         return (CTL_RETVAL_COMPLETE);
13263 }
13264
13265 #ifdef CTL_IO_DELAY
13266 static void
13267 ctl_done_timer_wakeup(void *arg)
13268 {
13269         union ctl_io *io;
13270
13271         io = (union ctl_io *)arg;
13272         ctl_done(io);
13273 }
13274 #endif /* CTL_IO_DELAY */
13275
13276 void
13277 ctl_serseq_done(union ctl_io *io)
13278 {
13279         struct ctl_lun *lun = CTL_LUN(io);;
13280
13281         if (lun->be_lun == NULL ||
13282             lun->be_lun->serseq == CTL_LUN_SERSEQ_OFF)
13283                 return;
13284         mtx_lock(&lun->lun_lock);
13285         io->io_hdr.flags |= CTL_FLAG_SERSEQ_DONE;
13286         ctl_check_blocked(lun);
13287         mtx_unlock(&lun->lun_lock);
13288 }
13289
13290 void
13291 ctl_done(union ctl_io *io)
13292 {
13293
13294         /*
13295          * Enable this to catch duplicate completion issues.
13296          */
13297 #if 0
13298         if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
13299                 printf("%s: type %d msg %d cdb %x iptl: "
13300                        "%u:%u:%u tag 0x%04x "
13301                        "flag %#x status %x\n",
13302                         __func__,
13303                         io->io_hdr.io_type,
13304                         io->io_hdr.msg_type,
13305                         io->scsiio.cdb[0],
13306                         io->io_hdr.nexus.initid,
13307                         io->io_hdr.nexus.targ_port,
13308                         io->io_hdr.nexus.targ_lun,
13309                         (io->io_hdr.io_type ==
13310                         CTL_IO_TASK) ?
13311                         io->taskio.tag_num :
13312                         io->scsiio.tag_num,
13313                         io->io_hdr.flags,
13314                         io->io_hdr.status);
13315         } else
13316                 io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
13317 #endif
13318
13319         /*
13320          * This is an internal copy of an I/O, and should not go through
13321          * the normal done processing logic.
13322          */
13323         if (io->io_hdr.flags & CTL_FLAG_INT_COPY)
13324                 return;
13325
13326 #ifdef CTL_IO_DELAY
13327         if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
13328                 io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
13329         } else {
13330                 struct ctl_lun *lun = CTL_LUN(io);
13331
13332                 if ((lun != NULL)
13333                  && (lun->delay_info.done_delay > 0)) {
13334
13335                         callout_init(&io->io_hdr.delay_callout, /*mpsafe*/ 1);
13336                         io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
13337                         callout_reset(&io->io_hdr.delay_callout,
13338                                       lun->delay_info.done_delay * hz,
13339                                       ctl_done_timer_wakeup, io);
13340                         if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
13341                                 lun->delay_info.done_delay = 0;
13342                         return;
13343                 }
13344         }
13345 #endif /* CTL_IO_DELAY */
13346
13347         ctl_enqueue_done(io);
13348 }
13349
13350 static void
13351 ctl_work_thread(void *arg)
13352 {
13353         struct ctl_thread *thr = (struct ctl_thread *)arg;
13354         struct ctl_softc *softc = thr->ctl_softc;
13355         union ctl_io *io;
13356         int retval;
13357
13358         CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
13359
13360         while (!softc->shutdown) {
13361                 /*
13362                  * We handle the queues in this order:
13363                  * - ISC
13364                  * - done queue (to free up resources, unblock other commands)
13365                  * - incoming queue
13366                  * - RtR queue
13367                  *
13368                  * If those queues are empty, we break out of the loop and
13369                  * go to sleep.
13370                  */
13371                 mtx_lock(&thr->queue_lock);
13372                 io = (union ctl_io *)STAILQ_FIRST(&thr->isc_queue);
13373                 if (io != NULL) {
13374                         STAILQ_REMOVE_HEAD(&thr->isc_queue, links);
13375                         mtx_unlock(&thr->queue_lock);
13376                         ctl_handle_isc(io);
13377                         continue;
13378                 }
13379                 io = (union ctl_io *)STAILQ_FIRST(&thr->done_queue);
13380                 if (io != NULL) {
13381                         STAILQ_REMOVE_HEAD(&thr->done_queue, links);
13382                         /* clear any blocked commands, call fe_done */
13383                         mtx_unlock(&thr->queue_lock);
13384                         ctl_process_done(io);
13385                         continue;
13386                 }
13387                 io = (union ctl_io *)STAILQ_FIRST(&thr->incoming_queue);
13388                 if (io != NULL) {
13389                         STAILQ_REMOVE_HEAD(&thr->incoming_queue, links);
13390                         mtx_unlock(&thr->queue_lock);
13391                         if (io->io_hdr.io_type == CTL_IO_TASK)
13392                                 ctl_run_task(io);
13393                         else
13394                                 ctl_scsiio_precheck(softc, &io->scsiio);
13395                         continue;
13396                 }
13397                 io = (union ctl_io *)STAILQ_FIRST(&thr->rtr_queue);
13398                 if (io != NULL) {
13399                         STAILQ_REMOVE_HEAD(&thr->rtr_queue, links);
13400                         mtx_unlock(&thr->queue_lock);
13401                         retval = ctl_scsiio(&io->scsiio);
13402                         if (retval != CTL_RETVAL_COMPLETE)
13403                                 CTL_DEBUG_PRINT(("ctl_scsiio failed\n"));
13404                         continue;
13405                 }
13406
13407                 /* Sleep until we have something to do. */
13408                 mtx_sleep(thr, &thr->queue_lock, PDROP | PRIBIO, "-", 0);
13409         }
13410         thr->thread = NULL;
13411         kthread_exit();
13412 }
13413
13414 static void
13415 ctl_lun_thread(void *arg)
13416 {
13417         struct ctl_softc *softc = (struct ctl_softc *)arg;
13418         struct ctl_be_lun *be_lun;
13419
13420         CTL_DEBUG_PRINT(("ctl_lun_thread starting\n"));
13421
13422         while (!softc->shutdown) {
13423                 mtx_lock(&softc->ctl_lock);
13424                 be_lun = STAILQ_FIRST(&softc->pending_lun_queue);
13425                 if (be_lun != NULL) {
13426                         STAILQ_REMOVE_HEAD(&softc->pending_lun_queue, links);
13427                         mtx_unlock(&softc->ctl_lock);
13428                         ctl_create_lun(be_lun);
13429                         continue;
13430                 }
13431
13432                 /* Sleep until we have something to do. */
13433                 mtx_sleep(&softc->pending_lun_queue, &softc->ctl_lock,
13434                     PDROP | PRIBIO, "-", 0);
13435         }
13436         softc->lun_thread = NULL;
13437         kthread_exit();
13438 }
13439
13440 static void
13441 ctl_thresh_thread(void *arg)
13442 {
13443         struct ctl_softc *softc = (struct ctl_softc *)arg;
13444         struct ctl_lun *lun;
13445         struct ctl_logical_block_provisioning_page *page;
13446         const char *attr;
13447         union ctl_ha_msg msg;
13448         uint64_t thres, val;
13449         int i, e, set;
13450
13451         CTL_DEBUG_PRINT(("ctl_thresh_thread starting\n"));
13452
13453         while (!softc->shutdown) {
13454                 mtx_lock(&softc->ctl_lock);
13455                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
13456                         if ((lun->flags & CTL_LUN_DISABLED) ||
13457                             (lun->flags & CTL_LUN_NO_MEDIA) ||
13458                             lun->backend->lun_attr == NULL)
13459                                 continue;
13460                         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
13461                             softc->ha_mode == CTL_HA_MODE_XFER)
13462                                 continue;
13463                         if ((lun->MODE_RWER.byte8 & SMS_RWER_LBPERE) == 0)
13464                                 continue;
13465                         e = 0;
13466                         page = &lun->MODE_LBP;
13467                         for (i = 0; i < CTL_NUM_LBP_THRESH; i++) {
13468                                 if ((page->descr[i].flags & SLBPPD_ENABLED) == 0)
13469                                         continue;
13470                                 thres = scsi_4btoul(page->descr[i].count);
13471                                 thres <<= CTL_LBP_EXPONENT;
13472                                 switch (page->descr[i].resource) {
13473                                 case 0x01:
13474                                         attr = "blocksavail";
13475                                         break;
13476                                 case 0x02:
13477                                         attr = "blocksused";
13478                                         break;
13479                                 case 0xf1:
13480                                         attr = "poolblocksavail";
13481                                         break;
13482                                 case 0xf2:
13483                                         attr = "poolblocksused";
13484                                         break;
13485                                 default:
13486                                         continue;
13487                                 }
13488                                 mtx_unlock(&softc->ctl_lock); // XXX
13489                                 val = lun->backend->lun_attr(
13490                                     lun->be_lun->be_lun, attr);
13491                                 mtx_lock(&softc->ctl_lock);
13492                                 if (val == UINT64_MAX)
13493                                         continue;
13494                                 if ((page->descr[i].flags & SLBPPD_ARMING_MASK)
13495                                     == SLBPPD_ARMING_INC)
13496                                         e = (val >= thres);
13497                                 else
13498                                         e = (val <= thres);
13499                                 if (e)
13500                                         break;
13501                         }
13502                         mtx_lock(&lun->lun_lock);
13503                         if (e) {
13504                                 scsi_u64to8b((uint8_t *)&page->descr[i] -
13505                                     (uint8_t *)page, lun->ua_tpt_info);
13506                                 if (lun->lasttpt == 0 ||
13507                                     time_uptime - lun->lasttpt >= CTL_LBP_UA_PERIOD) {
13508                                         lun->lasttpt = time_uptime;
13509                                         ctl_est_ua_all(lun, -1, CTL_UA_THIN_PROV_THRES);
13510                                         set = 1;
13511                                 } else
13512                                         set = 0;
13513                         } else {
13514                                 lun->lasttpt = 0;
13515                                 ctl_clr_ua_all(lun, -1, CTL_UA_THIN_PROV_THRES);
13516                                 set = -1;
13517                         }
13518                         mtx_unlock(&lun->lun_lock);
13519                         if (set != 0 &&
13520                             lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
13521                                 /* Send msg to other side. */
13522                                 bzero(&msg.ua, sizeof(msg.ua));
13523                                 msg.hdr.msg_type = CTL_MSG_UA;
13524                                 msg.hdr.nexus.initid = -1;
13525                                 msg.hdr.nexus.targ_port = -1;
13526                                 msg.hdr.nexus.targ_lun = lun->lun;
13527                                 msg.hdr.nexus.targ_mapped_lun = lun->lun;
13528                                 msg.ua.ua_all = 1;
13529                                 msg.ua.ua_set = (set > 0);
13530                                 msg.ua.ua_type = CTL_UA_THIN_PROV_THRES;
13531                                 memcpy(msg.ua.ua_info, lun->ua_tpt_info, 8);
13532                                 mtx_unlock(&softc->ctl_lock); // XXX
13533                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13534                                     sizeof(msg.ua), M_WAITOK);
13535                                 mtx_lock(&softc->ctl_lock);
13536                         }
13537                 }
13538                 mtx_sleep(&softc->thresh_thread, &softc->ctl_lock,
13539                     PDROP | PRIBIO, "-", CTL_LBP_PERIOD * hz);
13540         }
13541         softc->thresh_thread = NULL;
13542         kthread_exit();
13543 }
13544
13545 static void
13546 ctl_enqueue_incoming(union ctl_io *io)
13547 {
13548         struct ctl_softc *softc = CTL_SOFTC(io);
13549         struct ctl_thread *thr;
13550         u_int idx;
13551
13552         idx = (io->io_hdr.nexus.targ_port * 127 +
13553                io->io_hdr.nexus.initid) % worker_threads;
13554         thr = &softc->threads[idx];
13555         mtx_lock(&thr->queue_lock);
13556         STAILQ_INSERT_TAIL(&thr->incoming_queue, &io->io_hdr, links);
13557         mtx_unlock(&thr->queue_lock);
13558         wakeup(thr);
13559 }
13560
13561 static void
13562 ctl_enqueue_rtr(union ctl_io *io)
13563 {
13564         struct ctl_softc *softc = CTL_SOFTC(io);
13565         struct ctl_thread *thr;
13566
13567         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13568         mtx_lock(&thr->queue_lock);
13569         STAILQ_INSERT_TAIL(&thr->rtr_queue, &io->io_hdr, links);
13570         mtx_unlock(&thr->queue_lock);
13571         wakeup(thr);
13572 }
13573
13574 static void
13575 ctl_enqueue_done(union ctl_io *io)
13576 {
13577         struct ctl_softc *softc = CTL_SOFTC(io);
13578         struct ctl_thread *thr;
13579
13580         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13581         mtx_lock(&thr->queue_lock);
13582         STAILQ_INSERT_TAIL(&thr->done_queue, &io->io_hdr, links);
13583         mtx_unlock(&thr->queue_lock);
13584         wakeup(thr);
13585 }
13586
13587 static void
13588 ctl_enqueue_isc(union ctl_io *io)
13589 {
13590         struct ctl_softc *softc = CTL_SOFTC(io);
13591         struct ctl_thread *thr;
13592
13593         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13594         mtx_lock(&thr->queue_lock);
13595         STAILQ_INSERT_TAIL(&thr->isc_queue, &io->io_hdr, links);
13596         mtx_unlock(&thr->queue_lock);
13597         wakeup(thr);
13598 }
13599
13600 /*
13601  *  vim: ts=8
13602  */