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  * All rights reserved.
   8  *
   9  * Portions of this software were developed by Edward Tomasz Napierala
  10  * under sponsorship from the FreeBSD Foundation.
  11  *
  12  * Redistribution and use in source and binary forms, with or without
  13  * modification, are permitted provided that the following conditions
  14  * are met:
  15  * 1. Redistributions of source code must retain the above copyright
  16  *    notice, this list of conditions, and the following disclaimer,
  17  *    without modification.
  18  * 2. Redistributions in binary form must reproduce at minimum a disclaimer
  19  *    substantially similar to the "NO WARRANTY" disclaimer below
  20  *    ("Disclaimer") and any redistribution must be conditioned upon
  21  *    including a substantially similar Disclaimer requirement for further
  22  *    binary redistribution.
  23  *
  24  * NO WARRANTY
  25  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  26  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  27  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR
  28  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  29  * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  30  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  31  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  32  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
  33  * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
  34  * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
  35  * POSSIBILITY OF SUCH DAMAGES.
  36  *
  37  * $Id$
  38  */
  39 /*
  40  * CAM Target Layer, a SCSI device emulation subsystem.
  41  *
  42  * Author: Ken Merry <ken@FreeBSD.org>
  43  */
  44
  45 #include <sys/cdefs.h>
  46 __FBSDID("$FreeBSD$");
  47
  48 #include <sys/param.h>
  49 #include <sys/systm.h>
  50 #include <sys/ctype.h>
  51 #include <sys/kernel.h>
  52 #include <sys/types.h>
  53 #include <sys/kthread.h>
  54 #include <sys/bio.h>
  55 #include <sys/fcntl.h>
  56 #include <sys/lock.h>
  57 #include <sys/module.h>
  58 #include <sys/mutex.h>
  59 #include <sys/condvar.h>
  60 #include <sys/malloc.h>
  61 #include <sys/conf.h>
  62 #include <sys/ioccom.h>
  63 #include <sys/queue.h>
  64 #include <sys/sbuf.h>
  65 #include <sys/smp.h>
  66 #include <sys/endian.h>
  67 #include <sys/sysctl.h>
  68 #include <vm/uma.h>
  69
  70 #include <cam/cam.h>
  71 #include <cam/scsi/scsi_all.h>
  72 #include <cam/scsi/scsi_cd.h>
  73 #include <cam/scsi/scsi_da.h>
  74 #include <cam/ctl/ctl_io.h>
  75 #include <cam/ctl/ctl.h>
  76 #include <cam/ctl/ctl_frontend.h>
  77 #include <cam/ctl/ctl_util.h>
  78 #include <cam/ctl/ctl_backend.h>
  79 #include <cam/ctl/ctl_ioctl.h>
  80 #include <cam/ctl/ctl_ha.h>
  81 #include <cam/ctl/ctl_private.h>
  82 #include <cam/ctl/ctl_debug.h>
  83 #include <cam/ctl/ctl_scsi_all.h>
  84 #include <cam/ctl/ctl_error.h>
  85
  86 struct ctl_softc *control_softc = NULL;
  87
  88 /*
  89  * Template mode pages.
  90  */
  91
  92 /*
  93  * Note that these are default values only.  The actual values will be
  94  * filled in when the user does a mode sense.
  95  */
  96 const static struct scsi_da_rw_recovery_page rw_er_page_default = {
  97         /*page_code*/SMS_RW_ERROR_RECOVERY_PAGE,
  98         /*page_length*/sizeof(struct scsi_da_rw_recovery_page) - 2,
  99         /*byte3*/SMS_RWER_AWRE|SMS_RWER_ARRE,
 100         /*read_retry_count*/0,
 101         /*correction_span*/0,
 102         /*head_offset_count*/0,
 103         /*data_strobe_offset_cnt*/0,
 104         /*byte8*/SMS_RWER_LBPERE,
 105         /*write_retry_count*/0,
 106         /*reserved2*/0,
 107         /*recovery_time_limit*/{0, 0},
 108 };
 109
 110 const static struct scsi_da_rw_recovery_page rw_er_page_changeable = {
 111         /*page_code*/SMS_RW_ERROR_RECOVERY_PAGE,
 112         /*page_length*/sizeof(struct scsi_da_rw_recovery_page) - 2,
 113         /*byte3*/SMS_RWER_PER,
 114         /*read_retry_count*/0,
 115         /*correction_span*/0,
 116         /*head_offset_count*/0,
 117         /*data_strobe_offset_cnt*/0,
 118         /*byte8*/SMS_RWER_LBPERE,
 119         /*write_retry_count*/0,
 120         /*reserved2*/0,
 121         /*recovery_time_limit*/{0, 0},
 122 };
 123
 124 const static struct scsi_format_page format_page_default = {
 125         /*page_code*/SMS_FORMAT_DEVICE_PAGE,
 126         /*page_length*/sizeof(struct scsi_format_page) - 2,
 127         /*tracks_per_zone*/ {0, 0},
 128         /*alt_sectors_per_zone*/ {0, 0},
 129         /*alt_tracks_per_zone*/ {0, 0},
 130         /*alt_tracks_per_lun*/ {0, 0},
 131         /*sectors_per_track*/ {(CTL_DEFAULT_SECTORS_PER_TRACK >> 8) & 0xff,
 132                                 CTL_DEFAULT_SECTORS_PER_TRACK & 0xff},
 133         /*bytes_per_sector*/ {0, 0},
 134         /*interleave*/ {0, 0},
 135         /*track_skew*/ {0, 0},
 136         /*cylinder_skew*/ {0, 0},
 137         /*flags*/ SFP_HSEC,
 138         /*reserved*/ {0, 0, 0}
 139 };
 140
 141 const static struct scsi_format_page format_page_changeable = {
 142         /*page_code*/SMS_FORMAT_DEVICE_PAGE,
 143         /*page_length*/sizeof(struct scsi_format_page) - 2,
 144         /*tracks_per_zone*/ {0, 0},
 145         /*alt_sectors_per_zone*/ {0, 0},
 146         /*alt_tracks_per_zone*/ {0, 0},
 147         /*alt_tracks_per_lun*/ {0, 0},
 148         /*sectors_per_track*/ {0, 0},
 149         /*bytes_per_sector*/ {0, 0},
 150         /*interleave*/ {0, 0},
 151         /*track_skew*/ {0, 0},
 152         /*cylinder_skew*/ {0, 0},
 153         /*flags*/ 0,
 154         /*reserved*/ {0, 0, 0}
 155 };
 156
 157 const static struct scsi_rigid_disk_page rigid_disk_page_default = {
 158         /*page_code*/SMS_RIGID_DISK_PAGE,
 159         /*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
 160         /*cylinders*/ {0, 0, 0},
 161         /*heads*/ CTL_DEFAULT_HEADS,
 162         /*start_write_precomp*/ {0, 0, 0},
 163         /*start_reduced_current*/ {0, 0, 0},
 164         /*step_rate*/ {0, 0},
 165         /*landing_zone_cylinder*/ {0, 0, 0},
 166         /*rpl*/ SRDP_RPL_DISABLED,
 167         /*rotational_offset*/ 0,
 168         /*reserved1*/ 0,
 169         /*rotation_rate*/ {(CTL_DEFAULT_ROTATION_RATE >> 8) & 0xff,
 170                            CTL_DEFAULT_ROTATION_RATE & 0xff},
 171         /*reserved2*/ {0, 0}
 172 };
 173
 174 const static struct scsi_rigid_disk_page rigid_disk_page_changeable = {
 175         /*page_code*/SMS_RIGID_DISK_PAGE,
 176         /*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
 177         /*cylinders*/ {0, 0, 0},
 178         /*heads*/ 0,
 179         /*start_write_precomp*/ {0, 0, 0},
 180         /*start_reduced_current*/ {0, 0, 0},
 181         /*step_rate*/ {0, 0},
 182         /*landing_zone_cylinder*/ {0, 0, 0},
 183         /*rpl*/ 0,
 184         /*rotational_offset*/ 0,
 185         /*reserved1*/ 0,
 186         /*rotation_rate*/ {0, 0},
 187         /*reserved2*/ {0, 0}
 188 };
 189
 190 const static struct scsi_da_verify_recovery_page verify_er_page_default = {
 191         /*page_code*/SMS_VERIFY_ERROR_RECOVERY_PAGE,
 192         /*page_length*/sizeof(struct scsi_da_verify_recovery_page) - 2,
 193         /*byte3*/0,
 194         /*read_retry_count*/0,
 195         /*reserved*/{ 0, 0, 0, 0, 0, 0 },
 196         /*recovery_time_limit*/{0, 0},
 197 };
 198
 199 const static struct scsi_da_verify_recovery_page verify_er_page_changeable = {
 200         /*page_code*/SMS_VERIFY_ERROR_RECOVERY_PAGE,
 201         /*page_length*/sizeof(struct scsi_da_verify_recovery_page) - 2,
 202         /*byte3*/SMS_VER_PER,
 203         /*read_retry_count*/0,
 204         /*reserved*/{ 0, 0, 0, 0, 0, 0 },
 205         /*recovery_time_limit*/{0, 0},
 206 };
 207
 208 const static struct scsi_caching_page caching_page_default = {
 209         /*page_code*/SMS_CACHING_PAGE,
 210         /*page_length*/sizeof(struct scsi_caching_page) - 2,
 211         /*flags1*/ SCP_DISC | SCP_WCE,
 212         /*ret_priority*/ 0,
 213         /*disable_pf_transfer_len*/ {0xff, 0xff},
 214         /*min_prefetch*/ {0, 0},
 215         /*max_prefetch*/ {0xff, 0xff},
 216         /*max_pf_ceiling*/ {0xff, 0xff},
 217         /*flags2*/ 0,
 218         /*cache_segments*/ 0,
 219         /*cache_seg_size*/ {0, 0},
 220         /*reserved*/ 0,
 221         /*non_cache_seg_size*/ {0, 0, 0}
 222 };
 223
 224 const static struct scsi_caching_page caching_page_changeable = {
 225         /*page_code*/SMS_CACHING_PAGE,
 226         /*page_length*/sizeof(struct scsi_caching_page) - 2,
 227         /*flags1*/ SCP_WCE | SCP_RCD,
 228         /*ret_priority*/ 0,
 229         /*disable_pf_transfer_len*/ {0, 0},
 230         /*min_prefetch*/ {0, 0},
 231         /*max_prefetch*/ {0, 0},
 232         /*max_pf_ceiling*/ {0, 0},
 233         /*flags2*/ 0,
 234         /*cache_segments*/ 0,
 235         /*cache_seg_size*/ {0, 0},
 236         /*reserved*/ 0,
 237         /*non_cache_seg_size*/ {0, 0, 0}
 238 };
 239
 240 const static struct scsi_control_page control_page_default = {
 241         /*page_code*/SMS_CONTROL_MODE_PAGE,
 242         /*page_length*/sizeof(struct scsi_control_page) - 2,
 243         /*rlec*/0,
 244         /*queue_flags*/SCP_QUEUE_ALG_RESTRICTED,
 245         /*eca_and_aen*/0,
 246         /*flags4*/SCP_TAS,
 247         /*aen_holdoff_period*/{0, 0},
 248         /*busy_timeout_period*/{0, 0},
 249         /*extended_selftest_completion_time*/{0, 0}
 250 };
 251
 252 const static struct scsi_control_page control_page_changeable = {
 253         /*page_code*/SMS_CONTROL_MODE_PAGE,
 254         /*page_length*/sizeof(struct scsi_control_page) - 2,
 255         /*rlec*/SCP_DSENSE,
 256         /*queue_flags*/SCP_QUEUE_ALG_MASK | SCP_NUAR,
 257         /*eca_and_aen*/SCP_SWP,
 258         /*flags4*/0,
 259         /*aen_holdoff_period*/{0, 0},
 260         /*busy_timeout_period*/{0, 0},
 261         /*extended_selftest_completion_time*/{0, 0}
 262 };
 263
 264 #define CTL_CEM_LEN     (sizeof(struct scsi_control_ext_page) - 4)
 265
 266 const static struct scsi_control_ext_page control_ext_page_default = {
 267         /*page_code*/SMS_CONTROL_MODE_PAGE | SMPH_SPF,
 268         /*subpage_code*/0x01,
 269         /*page_length*/{CTL_CEM_LEN >> 8, CTL_CEM_LEN},
 270         /*flags*/0,
 271         /*prio*/0,
 272         /*max_sense*/0
 273 };
 274
 275 const static struct scsi_control_ext_page control_ext_page_changeable = {
 276         /*page_code*/SMS_CONTROL_MODE_PAGE | SMPH_SPF,
 277         /*subpage_code*/0x01,
 278         /*page_length*/{CTL_CEM_LEN >> 8, CTL_CEM_LEN},
 279         /*flags*/0,
 280         /*prio*/0,
 281         /*max_sense*/0xff
 282 };
 283
 284 const static struct scsi_info_exceptions_page ie_page_default = {
 285         /*page_code*/SMS_INFO_EXCEPTIONS_PAGE,
 286         /*page_length*/sizeof(struct scsi_info_exceptions_page) - 2,
 287         /*info_flags*/SIEP_FLAGS_EWASC,
 288         /*mrie*/SIEP_MRIE_NO,
 289         /*interval_timer*/{0, 0, 0, 0},
 290         /*report_count*/{0, 0, 0, 1}
 291 };
 292
 293 const static struct scsi_info_exceptions_page ie_page_changeable = {
 294         /*page_code*/SMS_INFO_EXCEPTIONS_PAGE,
 295         /*page_length*/sizeof(struct scsi_info_exceptions_page) - 2,
 296         /*info_flags*/SIEP_FLAGS_EWASC | SIEP_FLAGS_DEXCPT | SIEP_FLAGS_TEST |
 297             SIEP_FLAGS_LOGERR,
 298         /*mrie*/0x0f,
 299         /*interval_timer*/{0xff, 0xff, 0xff, 0xff},
 300         /*report_count*/{0xff, 0xff, 0xff, 0xff}
 301 };
 302
 303 #define CTL_LBPM_LEN    (sizeof(struct ctl_logical_block_provisioning_page) - 4)
 304
 305 const static struct ctl_logical_block_provisioning_page lbp_page_default = {{
 306         /*page_code*/SMS_INFO_EXCEPTIONS_PAGE | SMPH_SPF,
 307         /*subpage_code*/0x02,
 308         /*page_length*/{CTL_LBPM_LEN >> 8, CTL_LBPM_LEN},
 309         /*flags*/0,
 310         /*reserved*/{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
 311         /*descr*/{}},
 312         {{/*flags*/0,
 313           /*resource*/0x01,
 314           /*reserved*/{0, 0},
 315           /*count*/{0, 0, 0, 0}},
 316          {/*flags*/0,
 317           /*resource*/0x02,
 318           /*reserved*/{0, 0},
 319           /*count*/{0, 0, 0, 0}},
 320          {/*flags*/0,
 321           /*resource*/0xf1,
 322           /*reserved*/{0, 0},
 323           /*count*/{0, 0, 0, 0}},
 324          {/*flags*/0,
 325           /*resource*/0xf2,
 326           /*reserved*/{0, 0},
 327           /*count*/{0, 0, 0, 0}}
 328         }
 329 };
 330
 331 const static struct ctl_logical_block_provisioning_page lbp_page_changeable = {{
 332         /*page_code*/SMS_INFO_EXCEPTIONS_PAGE | SMPH_SPF,
 333         /*subpage_code*/0x02,
 334         /*page_length*/{CTL_LBPM_LEN >> 8, CTL_LBPM_LEN},
 335         /*flags*/SLBPP_SITUA,
 336         /*reserved*/{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
 337         /*descr*/{}},
 338         {{/*flags*/0,
 339           /*resource*/0,
 340           /*reserved*/{0, 0},
 341           /*count*/{0, 0, 0, 0}},
 342          {/*flags*/0,
 343           /*resource*/0,
 344           /*reserved*/{0, 0},
 345           /*count*/{0, 0, 0, 0}},
 346          {/*flags*/0,
 347           /*resource*/0,
 348           /*reserved*/{0, 0},
 349           /*count*/{0, 0, 0, 0}},
 350          {/*flags*/0,
 351           /*resource*/0,
 352           /*reserved*/{0, 0},
 353           /*count*/{0, 0, 0, 0}}
 354         }
 355 };
 356
 357 const static struct scsi_cddvd_capabilities_page cddvd_page_default = {
 358         /*page_code*/SMS_CDDVD_CAPS_PAGE,
 359         /*page_length*/sizeof(struct scsi_cddvd_capabilities_page) - 2,
 360         /*caps1*/0x3f,
 361         /*caps2*/0x00,
 362         /*caps3*/0xf0,
 363         /*caps4*/0x00,
 364         /*caps5*/0x29,
 365         /*caps6*/0x00,
 366         /*obsolete*/{0, 0},
 367         /*nvol_levels*/{0, 0},
 368         /*buffer_size*/{8, 0},
 369         /*obsolete2*/{0, 0},
 370         /*reserved*/0,
 371         /*digital*/0,
 372         /*obsolete3*/0,
 373         /*copy_management*/0,
 374         /*reserved2*/0,
 375         /*rotation_control*/0,
 376         /*cur_write_speed*/0,
 377         /*num_speed_descr*/0,
 378 };
 379
 380 const static struct scsi_cddvd_capabilities_page cddvd_page_changeable = {
 381         /*page_code*/SMS_CDDVD_CAPS_PAGE,
 382         /*page_length*/sizeof(struct scsi_cddvd_capabilities_page) - 2,
 383         /*caps1*/0,
 384         /*caps2*/0,
 385         /*caps3*/0,
 386         /*caps4*/0,
 387         /*caps5*/0,
 388         /*caps6*/0,
 389         /*obsolete*/{0, 0},
 390         /*nvol_levels*/{0, 0},
 391         /*buffer_size*/{0, 0},
 392         /*obsolete2*/{0, 0},
 393         /*reserved*/0,
 394         /*digital*/0,
 395         /*obsolete3*/0,
 396         /*copy_management*/0,
 397         /*reserved2*/0,
 398         /*rotation_control*/0,
 399         /*cur_write_speed*/0,
 400         /*num_speed_descr*/0,
 401 };
 402
 403 SYSCTL_NODE(_kern_cam, OID_AUTO, ctl, CTLFLAG_RD, 0, "CAM Target Layer");
 404 static int worker_threads = -1;
 405 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, worker_threads, CTLFLAG_RDTUN,
 406     &worker_threads, 1, "Number of worker threads");
 407 static int ctl_debug = CTL_DEBUG_NONE;
 408 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, debug, CTLFLAG_RWTUN,
 409     &ctl_debug, 0, "Enabled debug flags");
 410 static int ctl_lun_map_size = 1024;
 411 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, lun_map_size, CTLFLAG_RWTUN,
 412     &ctl_lun_map_size, 0, "Size of per-port LUN map (max LUN + 1)");
 413 #ifdef  CTL_TIME_IO
 414 static int ctl_time_io_secs = CTL_TIME_IO_DEFAULT_SECS;
 415 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, time_io_secs, CTLFLAG_RWTUN,
 416     &ctl_time_io_secs, 0, "Log requests taking more seconds");
 417 #endif
 418
 419 /*
 420  * Maximum number of LUNs we support.  MUST be a power of 2.
 421  */
 422 #define CTL_DEFAULT_MAX_LUNS    1024
 423 static int ctl_max_luns = CTL_DEFAULT_MAX_LUNS;
 424 TUNABLE_INT("kern.cam.ctl.max_luns", &ctl_max_luns);
 425 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, max_luns, CTLFLAG_RDTUN,
 426     &ctl_max_luns, CTL_DEFAULT_MAX_LUNS, "Maximum number of LUNs");
 427
 428 /*
 429  * Maximum number of ports registered at one time.
 430  */
 431 #define CTL_DEFAULT_MAX_PORTS           256
 432 static int ctl_max_ports = CTL_DEFAULT_MAX_PORTS;
 433 TUNABLE_INT("kern.cam.ctl.max_ports", &ctl_max_ports);
 434 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, max_ports, CTLFLAG_RDTUN,
 435     &ctl_max_ports, CTL_DEFAULT_MAX_LUNS, "Maximum number of ports");
 436
 437 /*
 438  * Maximum number of initiators we support.
 439  */
 440 #define CTL_MAX_INITIATORS      (CTL_MAX_INIT_PER_PORT * ctl_max_ports)
 441
 442 /*
 443  * Supported pages (0x00), Serial number (0x80), Device ID (0x83),
 444  * Extended INQUIRY Data (0x86), Mode Page Policy (0x87),
 445  * SCSI Ports (0x88), Third-party Copy (0x8F), Block limits (0xB0),
 446  * Block Device Characteristics (0xB1) and Logical Block Provisioning (0xB2)
 447  */
 448 #define SCSI_EVPD_NUM_SUPPORTED_PAGES   10
 449
 450 static void ctl_isc_event_handler(ctl_ha_channel chanel, ctl_ha_event event,
 451                                   int param);
 452 static void ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest);
 453 static void ctl_copy_sense_data_back(union ctl_io *src, union ctl_ha_msg *dest);
 454 static int ctl_init(void);
 455 static int ctl_shutdown(void);
 456 static int ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td);
 457 static int ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td);
 458 static void ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio);
 459 static void ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
 460                               struct ctl_ooa *ooa_hdr,
 461                               struct ctl_ooa_entry *kern_entries);
 462 static int ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
 463                      struct thread *td);
 464 static int ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
 465                          struct ctl_be_lun *be_lun);
 466 static int ctl_free_lun(struct ctl_lun *lun);
 467 static void ctl_create_lun(struct ctl_be_lun *be_lun);
 468
 469 static int ctl_do_mode_select(union ctl_io *io);
 470 static int ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun,
 471                            uint64_t res_key, uint64_t sa_res_key,
 472                            uint8_t type, uint32_t residx,
 473                            struct ctl_scsiio *ctsio,
 474                            struct scsi_per_res_out *cdb,
 475                            struct scsi_per_res_out_parms* param);
 476 static void ctl_pro_preempt_other(struct ctl_lun *lun,
 477                                   union ctl_ha_msg *msg);
 478 static void ctl_hndl_per_res_out_on_other_sc(union ctl_io *io);
 479 static int ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len);
 480 static int ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len);
 481 static int ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len);
 482 static int ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len);
 483 static int ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len);
 484 static int ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio,
 485                                          int alloc_len);
 486 static int ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio,
 487                                          int alloc_len);
 488 static int ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len);
 489 static int ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len);
 490 static int ctl_inquiry_evpd(struct ctl_scsiio *ctsio);
 491 static int ctl_inquiry_std(struct ctl_scsiio *ctsio);
 492 static int ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len);
 493 static ctl_action ctl_extent_check(union ctl_io *io1, union ctl_io *io2,
 494     bool seq);
 495 static ctl_action ctl_extent_check_seq(union ctl_io *io1, union ctl_io *io2);
 496 static ctl_action ctl_check_for_blockage(struct ctl_lun *lun,
 497     union ctl_io *pending_io, union ctl_io *ooa_io);
 498 static ctl_action ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
 499                                 union ctl_io *starting_io);
 500 static int ctl_check_blocked(struct ctl_lun *lun);
 501 static int ctl_scsiio_lun_check(struct ctl_lun *lun,
 502                                 const struct ctl_cmd_entry *entry,
 503                                 struct ctl_scsiio *ctsio);
 504 static void ctl_failover_lun(union ctl_io *io);
 505 static int ctl_scsiio_precheck(struct ctl_softc *ctl_softc,
 506                                struct ctl_scsiio *ctsio);
 507 static int ctl_scsiio(struct ctl_scsiio *ctsio);
 508
 509 static int ctl_target_reset(union ctl_io *io);
 510 static void ctl_do_lun_reset(struct ctl_lun *lun, uint32_t initidx,
 511                          ctl_ua_type ua_type);
 512 static int ctl_lun_reset(union ctl_io *io);
 513 static int ctl_abort_task(union ctl_io *io);
 514 static int ctl_abort_task_set(union ctl_io *io);
 515 static int ctl_query_task(union ctl_io *io, int task_set);
 516 static void ctl_i_t_nexus_loss(struct ctl_softc *softc, uint32_t initidx,
 517                               ctl_ua_type ua_type);
 518 static int ctl_i_t_nexus_reset(union ctl_io *io);
 519 static int ctl_query_async_event(union ctl_io *io);
 520 static void ctl_run_task(union ctl_io *io);
 521 #ifdef CTL_IO_DELAY
 522 static void ctl_datamove_timer_wakeup(void *arg);
 523 static void ctl_done_timer_wakeup(void *arg);
 524 #endif /* CTL_IO_DELAY */
 525
 526 static void ctl_send_datamove_done(union ctl_io *io, int have_lock);
 527 static void ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq);
 528 static int ctl_datamove_remote_dm_write_cb(union ctl_io *io);
 529 static void ctl_datamove_remote_write(union ctl_io *io);
 530 static int ctl_datamove_remote_dm_read_cb(union ctl_io *io);
 531 static void ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq);
 532 static int ctl_datamove_remote_sgl_setup(union ctl_io *io);
 533 static int ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
 534                                     ctl_ha_dt_cb callback);
 535 static void ctl_datamove_remote_read(union ctl_io *io);
 536 static void ctl_datamove_remote(union ctl_io *io);
 537 static void ctl_process_done(union ctl_io *io);
 538 static void ctl_lun_thread(void *arg);
 539 static void ctl_thresh_thread(void *arg);
 540 static void ctl_work_thread(void *arg);
 541 static void ctl_enqueue_incoming(union ctl_io *io);
 542 static void ctl_enqueue_rtr(union ctl_io *io);
 543 static void ctl_enqueue_done(union ctl_io *io);
 544 static void ctl_enqueue_isc(union ctl_io *io);
 545 static const struct ctl_cmd_entry *
 546     ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa);
 547 static const struct ctl_cmd_entry *
 548     ctl_validate_command(struct ctl_scsiio *ctsio);
 549 static int ctl_cmd_applicable(uint8_t lun_type,
 550     const struct ctl_cmd_entry *entry);
 551 static int ctl_ha_init(void);
 552 static int ctl_ha_shutdown(void);
 553
 554 static uint64_t ctl_get_prkey(struct ctl_lun *lun, uint32_t residx);
 555 static void ctl_clr_prkey(struct ctl_lun *lun, uint32_t residx);
 556 static void ctl_alloc_prkey(struct ctl_lun *lun, uint32_t residx);
 557 static void ctl_set_prkey(struct ctl_lun *lun, uint32_t residx, uint64_t key);
 558
 559 /*
 560  * Load the serialization table.  This isn't very pretty, but is probably
 561  * the easiest way to do it.
 562  */
 563 #include "ctl_ser_table.c"
 564
 565 /*
 566  * We only need to define open, close and ioctl routines for this driver.
 567  */
 568 static struct cdevsw ctl_cdevsw = {
 569         .d_version =    D_VERSION,
 570         .d_flags =      0,
 571         .d_open =       ctl_open,
 572         .d_close =      ctl_close,
 573         .d_ioctl =      ctl_ioctl,
 574         .d_name =       "ctl",
 575 };
 576
 577
 578 MALLOC_DEFINE(M_CTL, "ctlmem", "Memory used for CTL");
 579
 580 static int ctl_module_event_handler(module_t, int /*modeventtype_t*/, void *);
 581
 582 static moduledata_t ctl_moduledata = {
 583         "ctl",
 584         ctl_module_event_handler,
 585         NULL
 586 };
 587
 588 DECLARE_MODULE(ctl, ctl_moduledata, SI_SUB_CONFIGURE, SI_ORDER_THIRD);
 589 MODULE_VERSION(ctl, 1);
 590
 591 static struct ctl_frontend ha_frontend =
 592 {
 593         .name = "ha",
 594         .init = ctl_ha_init,
 595         .shutdown = ctl_ha_shutdown,
 596 };
 597
 598 static int
 599 ctl_ha_init(void)
 600 {
 601         struct ctl_softc *softc = control_softc;
 602
 603         if (ctl_pool_create(softc, "othersc", CTL_POOL_ENTRIES_OTHER_SC,
 604                             &softc->othersc_pool) != 0)
 605                 return (ENOMEM);
 606         if (ctl_ha_msg_init(softc) != CTL_HA_STATUS_SUCCESS) {
 607                 ctl_pool_free(softc->othersc_pool);
 608                 return (EIO);
 609         }
 610         if (ctl_ha_msg_register(CTL_HA_CHAN_CTL, ctl_isc_event_handler)
 611             != CTL_HA_STATUS_SUCCESS) {
 612                 ctl_ha_msg_destroy(softc);
 613                 ctl_pool_free(softc->othersc_pool);
 614                 return (EIO);
 615         }
 616         return (0);
 617 };
 618
 619 static int
 620 ctl_ha_shutdown(void)
 621 {
 622         struct ctl_softc *softc = control_softc;
 623         struct ctl_port *port;
 624
 625         ctl_ha_msg_shutdown(softc);
 626         if (ctl_ha_msg_deregister(CTL_HA_CHAN_CTL) != CTL_HA_STATUS_SUCCESS)
 627                 return (EIO);
 628         if (ctl_ha_msg_destroy(softc) != CTL_HA_STATUS_SUCCESS)
 629                 return (EIO);
 630         ctl_pool_free(softc->othersc_pool);
 631         while ((port = STAILQ_FIRST(&ha_frontend.port_list)) != NULL) {
 632                 ctl_port_deregister(port);
 633                 free(port->port_name, M_CTL);
 634                 free(port, M_CTL);
 635         }
 636         return (0);
 637 };
 638
 639 static void
 640 ctl_ha_datamove(union ctl_io *io)
 641 {
 642         struct ctl_lun *lun = CTL_LUN(io);
 643         struct ctl_sg_entry *sgl;
 644         union ctl_ha_msg msg;
 645         uint32_t sg_entries_sent;
 646         int do_sg_copy, i, j;
 647
 648         memset(&msg.dt, 0, sizeof(msg.dt));
 649         msg.hdr.msg_type = CTL_MSG_DATAMOVE;
 650         msg.hdr.original_sc = io->io_hdr.original_sc;
 651         msg.hdr.serializing_sc = io;
 652         msg.hdr.nexus = io->io_hdr.nexus;
 653         msg.hdr.status = io->io_hdr.status;
 654         msg.dt.flags = io->io_hdr.flags;
 655
 656         /*
 657          * We convert everything into a S/G list here.  We can't
 658          * pass by reference, only by value between controllers.
 659          * So we can't pass a pointer to the S/G list, only as many
 660          * S/G entries as we can fit in here.  If it's possible for
 661          * us to get more than CTL_HA_MAX_SG_ENTRIES S/G entries,
 662          * then we need to break this up into multiple transfers.
 663          */
 664         if (io->scsiio.kern_sg_entries == 0) {
 665                 msg.dt.kern_sg_entries = 1;
 666 #if 0
 667                 if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
 668                         msg.dt.sg_list[0].addr = io->scsiio.kern_data_ptr;
 669                 } else {
 670                         /* XXX KDM use busdma here! */
 671                         msg.dt.sg_list[0].addr =
 672                             (void *)vtophys(io->scsiio.kern_data_ptr);
 673                 }
 674 #else
 675                 KASSERT((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0,
 676                     ("HA does not support BUS_ADDR"));
 677                 msg.dt.sg_list[0].addr = io->scsiio.kern_data_ptr;
 678 #endif
 679                 msg.dt.sg_list[0].len = io->scsiio.kern_data_len;
 680                 do_sg_copy = 0;
 681         } else {
 682                 msg.dt.kern_sg_entries = io->scsiio.kern_sg_entries;
 683                 do_sg_copy = 1;
 684         }
 685
 686         msg.dt.kern_data_len = io->scsiio.kern_data_len;
 687         msg.dt.kern_total_len = io->scsiio.kern_total_len;
 688         msg.dt.kern_data_resid = io->scsiio.kern_data_resid;
 689         msg.dt.kern_rel_offset = io->scsiio.kern_rel_offset;
 690         msg.dt.sg_sequence = 0;
 691
 692         /*
 693          * Loop until we've sent all of the S/G entries.  On the
 694          * other end, we'll recompose these S/G entries into one
 695          * contiguous list before processing.
 696          */
 697         for (sg_entries_sent = 0; sg_entries_sent < msg.dt.kern_sg_entries;
 698             msg.dt.sg_sequence++) {
 699                 msg.dt.cur_sg_entries = MIN((sizeof(msg.dt.sg_list) /
 700                     sizeof(msg.dt.sg_list[0])),
 701                     msg.dt.kern_sg_entries - sg_entries_sent);
 702                 if (do_sg_copy != 0) {
 703                         sgl = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr;
 704                         for (i = sg_entries_sent, j = 0;
 705                              i < msg.dt.cur_sg_entries; i++, j++) {
 706 #if 0
 707                                 if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
 708                                         msg.dt.sg_list[j].addr = sgl[i].addr;
 709                                 } else {
 710                                         /* XXX KDM use busdma here! */
 711                                         msg.dt.sg_list[j].addr =
 712                                             (void *)vtophys(sgl[i].addr);
 713                                 }
 714 #else
 715                                 KASSERT((io->io_hdr.flags &
 716                                     CTL_FLAG_BUS_ADDR) == 0,
 717                                     ("HA does not support BUS_ADDR"));
 718                                 msg.dt.sg_list[j].addr = sgl[i].addr;
 719 #endif
 720                                 msg.dt.sg_list[j].len = sgl[i].len;
 721                         }
 722                 }
 723
 724                 sg_entries_sent += msg.dt.cur_sg_entries;
 725                 msg.dt.sg_last = (sg_entries_sent >= msg.dt.kern_sg_entries);
 726                 if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
 727                     sizeof(msg.dt) - sizeof(msg.dt.sg_list) +
 728                     sizeof(struct ctl_sg_entry) * msg.dt.cur_sg_entries,
 729                     M_WAITOK) > CTL_HA_STATUS_SUCCESS) {
 730                         io->io_hdr.port_status = 31341;
 731                         io->scsiio.be_move_done(io);
 732                         return;
 733                 }
 734                 msg.dt.sent_sg_entries = sg_entries_sent;
 735         }
 736
 737         /*
 738          * Officially handover the request from us to peer.
 739          * If failover has just happened, then we must return error.
 740          * If failover happen just after, then it is not our problem.
 741          */
 742         if (lun)
 743                 mtx_lock(&lun->lun_lock);
 744         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
 745                 if (lun)
 746                         mtx_unlock(&lun->lun_lock);
 747                 io->io_hdr.port_status = 31342;
 748                 io->scsiio.be_move_done(io);
 749                 return;
 750         }
 751         io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
 752         io->io_hdr.flags |= CTL_FLAG_DMA_INPROG;
 753         if (lun)
 754                 mtx_unlock(&lun->lun_lock);
 755 }
 756
 757 static void
 758 ctl_ha_done(union ctl_io *io)
 759 {
 760         union ctl_ha_msg msg;
 761
 762         if (io->io_hdr.io_type == CTL_IO_SCSI) {
 763                 memset(&msg, 0, sizeof(msg));
 764                 msg.hdr.msg_type = CTL_MSG_FINISH_IO;
 765                 msg.hdr.original_sc = io->io_hdr.original_sc;
 766                 msg.hdr.nexus = io->io_hdr.nexus;
 767                 msg.hdr.status = io->io_hdr.status;
 768                 msg.scsi.scsi_status = io->scsiio.scsi_status;
 769                 msg.scsi.tag_num = io->scsiio.tag_num;
 770                 msg.scsi.tag_type = io->scsiio.tag_type;
 771                 msg.scsi.sense_len = io->scsiio.sense_len;
 772                 memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
 773                     io->scsiio.sense_len);
 774                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
 775                     sizeof(msg.scsi) - sizeof(msg.scsi.sense_data) +
 776                     msg.scsi.sense_len, M_WAITOK);
 777         }
 778         ctl_free_io(io);
 779 }
 780
 781 static void
 782 ctl_isc_handler_finish_xfer(struct ctl_softc *ctl_softc,
 783                             union ctl_ha_msg *msg_info)
 784 {
 785         struct ctl_scsiio *ctsio;
 786
 787         if (msg_info->hdr.original_sc == NULL) {
 788                 printf("%s: original_sc == NULL!\n", __func__);
 789                 /* XXX KDM now what? */
 790                 return;
 791         }
 792
 793         ctsio = &msg_info->hdr.original_sc->scsiio;
 794         ctsio->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
 795         ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
 796         ctsio->io_hdr.status = msg_info->hdr.status;
 797         ctsio->scsi_status = msg_info->scsi.scsi_status;
 798         ctsio->sense_len = msg_info->scsi.sense_len;
 799         memcpy(&ctsio->sense_data, &msg_info->scsi.sense_data,
 800                msg_info->scsi.sense_len);
 801         ctl_enqueue_isc((union ctl_io *)ctsio);
 802 }
 803
 804 static void
 805 ctl_isc_handler_finish_ser_only(struct ctl_softc *ctl_softc,
 806                                 union ctl_ha_msg *msg_info)
 807 {
 808         struct ctl_scsiio *ctsio;
 809
 810         if (msg_info->hdr.serializing_sc == NULL) {
 811                 printf("%s: serializing_sc == NULL!\n", __func__);
 812                 /* XXX KDM now what? */
 813                 return;
 814         }
 815
 816         ctsio = &msg_info->hdr.serializing_sc->scsiio;
 817         ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
 818         ctl_enqueue_isc((union ctl_io *)ctsio);
 819 }
 820
 821 void
 822 ctl_isc_announce_lun(struct ctl_lun *lun)
 823 {
 824         struct ctl_softc *softc = lun->ctl_softc;
 825         union ctl_ha_msg *msg;
 826         struct ctl_ha_msg_lun_pr_key pr_key;
 827         int i, k;
 828
 829         if (softc->ha_link != CTL_HA_LINK_ONLINE)
 830                 return;
 831         mtx_lock(&lun->lun_lock);
 832         i = sizeof(msg->lun);
 833         if (lun->lun_devid)
 834                 i += lun->lun_devid->len;
 835         i += sizeof(pr_key) * lun->pr_key_count;
 836 alloc:
 837         mtx_unlock(&lun->lun_lock);
 838         msg = malloc(i, M_CTL, M_WAITOK);
 839         mtx_lock(&lun->lun_lock);
 840         k = sizeof(msg->lun);
 841         if (lun->lun_devid)
 842                 k += lun->lun_devid->len;
 843         k += sizeof(pr_key) * lun->pr_key_count;
 844         if (i < k) {
 845                 free(msg, M_CTL);
 846                 i = k;
 847                 goto alloc;
 848         }
 849         bzero(&msg->lun, sizeof(msg->lun));
 850         msg->hdr.msg_type = CTL_MSG_LUN_SYNC;
 851         msg->hdr.nexus.targ_lun = lun->lun;
 852         msg->hdr.nexus.targ_mapped_lun = lun->lun;
 853         msg->lun.flags = lun->flags;
 854         msg->lun.pr_generation = lun->pr_generation;
 855         msg->lun.pr_res_idx = lun->pr_res_idx;
 856         msg->lun.pr_res_type = lun->pr_res_type;
 857         msg->lun.pr_key_count = lun->pr_key_count;
 858         i = 0;
 859         if (lun->lun_devid) {
 860                 msg->lun.lun_devid_len = lun->lun_devid->len;
 861                 memcpy(&msg->lun.data[i], lun->lun_devid->data,
 862                     msg->lun.lun_devid_len);
 863                 i += msg->lun.lun_devid_len;
 864         }
 865         for (k = 0; k < CTL_MAX_INITIATORS; k++) {
 866                 if ((pr_key.pr_key = ctl_get_prkey(lun, k)) == 0)
 867                         continue;
 868                 pr_key.pr_iid = k;
 869                 memcpy(&msg->lun.data[i], &pr_key, sizeof(pr_key));
 870                 i += sizeof(pr_key);
 871         }
 872         mtx_unlock(&lun->lun_lock);
 873         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->port, sizeof(msg->port) + i,
 874             M_WAITOK);
 875         free(msg, M_CTL);
 876
 877         if (lun->flags & CTL_LUN_PRIMARY_SC) {
 878                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
 879                         ctl_isc_announce_mode(lun, -1,
 880                             lun->mode_pages.index[i].page_code & SMPH_PC_MASK,
 881                             lun->mode_pages.index[i].subpage);
 882                 }
 883         }
 884 }
 885
 886 void
 887 ctl_isc_announce_port(struct ctl_port *port)
 888 {
 889         struct ctl_softc *softc = port->ctl_softc;
 890         union ctl_ha_msg *msg;
 891         int i;
 892
 893         if (port->targ_port < softc->port_min ||
 894             port->targ_port >= softc->port_max ||
 895             softc->ha_link != CTL_HA_LINK_ONLINE)
 896                 return;
 897         i = sizeof(msg->port) + strlen(port->port_name) + 1;
 898         if (port->lun_map)
 899                 i += port->lun_map_size * sizeof(uint32_t);
 900         if (port->port_devid)
 901                 i += port->port_devid->len;
 902         if (port->target_devid)
 903                 i += port->target_devid->len;
 904         if (port->init_devid)
 905                 i += port->init_devid->len;
 906         msg = malloc(i, M_CTL, M_WAITOK);
 907         bzero(&msg->port, sizeof(msg->port));
 908         msg->hdr.msg_type = CTL_MSG_PORT_SYNC;
 909         msg->hdr.nexus.targ_port = port->targ_port;
 910         msg->port.port_type = port->port_type;
 911         msg->port.physical_port = port->physical_port;
 912         msg->port.virtual_port = port->virtual_port;
 913         msg->port.status = port->status;
 914         i = 0;
 915         msg->port.name_len = sprintf(&msg->port.data[i],
 916             "%d:%s", softc->ha_id, port->port_name) + 1;
 917         i += msg->port.name_len;
 918         if (port->lun_map) {
 919                 msg->port.lun_map_len = port->lun_map_size * sizeof(uint32_t);
 920                 memcpy(&msg->port.data[i], port->lun_map,
 921                     msg->port.lun_map_len);
 922                 i += msg->port.lun_map_len;
 923         }
 924         if (port->port_devid) {
 925                 msg->port.port_devid_len = port->port_devid->len;
 926                 memcpy(&msg->port.data[i], port->port_devid->data,
 927                     msg->port.port_devid_len);
 928                 i += msg->port.port_devid_len;
 929         }
 930         if (port->target_devid) {
 931                 msg->port.target_devid_len = port->target_devid->len;
 932                 memcpy(&msg->port.data[i], port->target_devid->data,
 933                     msg->port.target_devid_len);
 934                 i += msg->port.target_devid_len;
 935         }
 936         if (port->init_devid) {
 937                 msg->port.init_devid_len = port->init_devid->len;
 938                 memcpy(&msg->port.data[i], port->init_devid->data,
 939                     msg->port.init_devid_len);
 940                 i += msg->port.init_devid_len;
 941         }
 942         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->port, sizeof(msg->port) + i,
 943             M_WAITOK);
 944         free(msg, M_CTL);
 945 }
 946
 947 void
 948 ctl_isc_announce_iid(struct ctl_port *port, int iid)
 949 {
 950         struct ctl_softc *softc = port->ctl_softc;
 951         union ctl_ha_msg *msg;
 952         int i, l;
 953
 954         if (port->targ_port < softc->port_min ||
 955             port->targ_port >= softc->port_max ||
 956             softc->ha_link != CTL_HA_LINK_ONLINE)
 957                 return;
 958         mtx_lock(&softc->ctl_lock);
 959         i = sizeof(msg->iid);
 960         l = 0;
 961         if (port->wwpn_iid[iid].name)
 962                 l = strlen(port->wwpn_iid[iid].name) + 1;
 963         i += l;
 964         msg = malloc(i, M_CTL, M_NOWAIT);
 965         if (msg == NULL) {
 966                 mtx_unlock(&softc->ctl_lock);
 967                 return;
 968         }
 969         bzero(&msg->iid, sizeof(msg->iid));
 970         msg->hdr.msg_type = CTL_MSG_IID_SYNC;
 971         msg->hdr.nexus.targ_port = port->targ_port;
 972         msg->hdr.nexus.initid = iid;
 973         msg->iid.in_use = port->wwpn_iid[iid].in_use;
 974         msg->iid.name_len = l;
 975         msg->iid.wwpn = port->wwpn_iid[iid].wwpn;
 976         if (port->wwpn_iid[iid].name)
 977                 strlcpy(msg->iid.data, port->wwpn_iid[iid].name, l);
 978         mtx_unlock(&softc->ctl_lock);
 979         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->iid, i, M_NOWAIT);
 980         free(msg, M_CTL);
 981 }
 982
 983 void
 984 ctl_isc_announce_mode(struct ctl_lun *lun, uint32_t initidx,
 985     uint8_t page, uint8_t subpage)
 986 {
 987         struct ctl_softc *softc = lun->ctl_softc;
 988         union ctl_ha_msg msg;
 989         u_int i;
 990
 991         if (softc->ha_link != CTL_HA_LINK_ONLINE)
 992                 return;
 993         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
 994                 if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) ==
 995                     page && lun->mode_pages.index[i].subpage == subpage)
 996                         break;
 997         }
 998         if (i == CTL_NUM_MODE_PAGES)
 999                 return;
1000
1001         /* Don't try to replicate pages not present on this device. */
1002         if (lun->mode_pages.index[i].page_data == NULL)
1003                 return;
1004
1005         bzero(&msg.mode, sizeof(msg.mode));
1006         msg.hdr.msg_type = CTL_MSG_MODE_SYNC;
1007         msg.hdr.nexus.targ_port = initidx / CTL_MAX_INIT_PER_PORT;
1008         msg.hdr.nexus.initid = initidx % CTL_MAX_INIT_PER_PORT;
1009         msg.hdr.nexus.targ_lun = lun->lun;
1010         msg.hdr.nexus.targ_mapped_lun = lun->lun;
1011         msg.mode.page_code = page;
1012         msg.mode.subpage = subpage;
1013         msg.mode.page_len = lun->mode_pages.index[i].page_len;
1014         memcpy(msg.mode.data, lun->mode_pages.index[i].page_data,
1015             msg.mode.page_len);
1016         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg.mode, sizeof(msg.mode),
1017             M_WAITOK);
1018 }
1019
1020 static void
1021 ctl_isc_ha_link_up(struct ctl_softc *softc)
1022 {
1023         struct ctl_port *port;
1024         struct ctl_lun *lun;
1025         union ctl_ha_msg msg;
1026         int i;
1027
1028         /* Announce this node parameters to peer for validation. */
1029         msg.login.msg_type = CTL_MSG_LOGIN;
1030         msg.login.version = CTL_HA_VERSION;
1031         msg.login.ha_mode = softc->ha_mode;
1032         msg.login.ha_id = softc->ha_id;
1033         msg.login.max_luns = ctl_max_luns;
1034         msg.login.max_ports = ctl_max_ports;
1035         msg.login.max_init_per_port = CTL_MAX_INIT_PER_PORT;
1036         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg.login, sizeof(msg.login),
1037             M_WAITOK);
1038
1039         STAILQ_FOREACH(port, &softc->port_list, links) {
1040                 ctl_isc_announce_port(port);
1041                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1042                         if (port->wwpn_iid[i].in_use)
1043                                 ctl_isc_announce_iid(port, i);
1044                 }
1045         }
1046         STAILQ_FOREACH(lun, &softc->lun_list, links)
1047                 ctl_isc_announce_lun(lun);
1048 }
1049
1050 static void
1051 ctl_isc_ha_link_down(struct ctl_softc *softc)
1052 {
1053         struct ctl_port *port;
1054         struct ctl_lun *lun;
1055         union ctl_io *io;
1056         int i;
1057
1058         mtx_lock(&softc->ctl_lock);
1059         STAILQ_FOREACH(lun, &softc->lun_list, links) {
1060                 mtx_lock(&lun->lun_lock);
1061                 if (lun->flags & CTL_LUN_PEER_SC_PRIMARY) {
1062                         lun->flags &= ~CTL_LUN_PEER_SC_PRIMARY;
1063                         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
1064                 }
1065                 mtx_unlock(&lun->lun_lock);
1066
1067                 mtx_unlock(&softc->ctl_lock);
1068                 io = ctl_alloc_io(softc->othersc_pool);
1069                 mtx_lock(&softc->ctl_lock);
1070                 ctl_zero_io(io);
1071                 io->io_hdr.msg_type = CTL_MSG_FAILOVER;
1072                 io->io_hdr.nexus.targ_mapped_lun = lun->lun;
1073                 ctl_enqueue_isc(io);
1074         }
1075
1076         STAILQ_FOREACH(port, &softc->port_list, links) {
1077                 if (port->targ_port >= softc->port_min &&
1078                     port->targ_port < softc->port_max)
1079                         continue;
1080                 port->status &= ~CTL_PORT_STATUS_ONLINE;
1081                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1082                         port->wwpn_iid[i].in_use = 0;
1083                         free(port->wwpn_iid[i].name, M_CTL);
1084                         port->wwpn_iid[i].name = NULL;
1085                 }
1086         }
1087         mtx_unlock(&softc->ctl_lock);
1088 }
1089
1090 static void
1091 ctl_isc_ua(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1092 {
1093         struct ctl_lun *lun;
1094         uint32_t iid = ctl_get_initindex(&msg->hdr.nexus);
1095
1096         mtx_lock(&softc->ctl_lock);
1097         if (msg->hdr.nexus.targ_mapped_lun >= ctl_max_luns ||
1098             (lun = softc->ctl_luns[msg->hdr.nexus.targ_mapped_lun]) == NULL) {
1099                 mtx_unlock(&softc->ctl_lock);
1100                 return;
1101         }
1102         mtx_lock(&lun->lun_lock);
1103         mtx_unlock(&softc->ctl_lock);
1104         if (msg->ua.ua_type == CTL_UA_THIN_PROV_THRES && msg->ua.ua_set)
1105                 memcpy(lun->ua_tpt_info, msg->ua.ua_info, 8);
1106         if (msg->ua.ua_all) {
1107                 if (msg->ua.ua_set)
1108                         ctl_est_ua_all(lun, iid, msg->ua.ua_type);
1109                 else
1110                         ctl_clr_ua_all(lun, iid, msg->ua.ua_type);
1111         } else {
1112                 if (msg->ua.ua_set)
1113                         ctl_est_ua(lun, iid, msg->ua.ua_type);
1114                 else
1115                         ctl_clr_ua(lun, iid, msg->ua.ua_type);
1116         }
1117         mtx_unlock(&lun->lun_lock);
1118 }
1119
1120 static void
1121 ctl_isc_lun_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1122 {
1123         struct ctl_lun *lun;
1124         struct ctl_ha_msg_lun_pr_key pr_key;
1125         int i, k;
1126         ctl_lun_flags oflags;
1127         uint32_t targ_lun;
1128
1129         targ_lun = msg->hdr.nexus.targ_mapped_lun;
1130         mtx_lock(&softc->ctl_lock);
1131         if (targ_lun >= ctl_max_luns ||
1132             (lun = softc->ctl_luns[targ_lun]) == NULL) {
1133                 mtx_unlock(&softc->ctl_lock);
1134                 return;
1135         }
1136         mtx_lock(&lun->lun_lock);
1137         mtx_unlock(&softc->ctl_lock);
1138         if (lun->flags & CTL_LUN_DISABLED) {
1139                 mtx_unlock(&lun->lun_lock);
1140                 return;
1141         }
1142         i = (lun->lun_devid != NULL) ? lun->lun_devid->len : 0;
1143         if (msg->lun.lun_devid_len != i || (i > 0 &&
1144             memcmp(&msg->lun.data[0], lun->lun_devid->data, i) != 0)) {
1145                 mtx_unlock(&lun->lun_lock);
1146                 printf("%s: Received conflicting HA LUN %d\n",
1147                     __func__, targ_lun);
1148                 return;
1149         } else {
1150                 /* Record whether peer is primary. */
1151                 oflags = lun->flags;
1152                 if ((msg->lun.flags & CTL_LUN_PRIMARY_SC) &&
1153                     (msg->lun.flags & CTL_LUN_DISABLED) == 0)
1154                         lun->flags |= CTL_LUN_PEER_SC_PRIMARY;
1155                 else
1156                         lun->flags &= ~CTL_LUN_PEER_SC_PRIMARY;
1157                 if (oflags != lun->flags)
1158                         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
1159
1160                 /* If peer is primary and we are not -- use data */
1161                 if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
1162                     (lun->flags & CTL_LUN_PEER_SC_PRIMARY)) {
1163                         lun->pr_generation = msg->lun.pr_generation;
1164                         lun->pr_res_idx = msg->lun.pr_res_idx;
1165                         lun->pr_res_type = msg->lun.pr_res_type;
1166                         lun->pr_key_count = msg->lun.pr_key_count;
1167                         for (k = 0; k < CTL_MAX_INITIATORS; k++)
1168                                 ctl_clr_prkey(lun, k);
1169                         for (k = 0; k < msg->lun.pr_key_count; k++) {
1170                                 memcpy(&pr_key, &msg->lun.data[i],
1171                                     sizeof(pr_key));
1172                                 ctl_alloc_prkey(lun, pr_key.pr_iid);
1173                                 ctl_set_prkey(lun, pr_key.pr_iid,
1174                                     pr_key.pr_key);
1175                                 i += sizeof(pr_key);
1176                         }
1177                 }
1178
1179                 mtx_unlock(&lun->lun_lock);
1180                 CTL_DEBUG_PRINT(("%s: Known LUN %d, peer is %s\n",
1181                     __func__, targ_lun,
1182                     (msg->lun.flags & CTL_LUN_PRIMARY_SC) ?
1183                     "primary" : "secondary"));
1184
1185                 /* If we are primary but peer doesn't know -- notify */
1186                 if ((lun->flags & CTL_LUN_PRIMARY_SC) &&
1187                     (msg->lun.flags & CTL_LUN_PEER_SC_PRIMARY) == 0)
1188                         ctl_isc_announce_lun(lun);
1189         }
1190 }
1191
1192 static void
1193 ctl_isc_port_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1194 {
1195         struct ctl_port *port;
1196         struct ctl_lun *lun;
1197         int i, new;
1198
1199         port = softc->ctl_ports[msg->hdr.nexus.targ_port];
1200         if (port == NULL) {
1201                 CTL_DEBUG_PRINT(("%s: New port %d\n", __func__,
1202                     msg->hdr.nexus.targ_port));
1203                 new = 1;
1204                 port = malloc(sizeof(*port), M_CTL, M_WAITOK | M_ZERO);
1205                 port->frontend = &ha_frontend;
1206                 port->targ_port = msg->hdr.nexus.targ_port;
1207                 port->fe_datamove = ctl_ha_datamove;
1208                 port->fe_done = ctl_ha_done;
1209         } else if (port->frontend == &ha_frontend) {
1210                 CTL_DEBUG_PRINT(("%s: Updated port %d\n", __func__,
1211                     msg->hdr.nexus.targ_port));
1212                 new = 0;
1213         } else {
1214                 printf("%s: Received conflicting HA port %d\n",
1215                     __func__, msg->hdr.nexus.targ_port);
1216                 return;
1217         }
1218         port->port_type = msg->port.port_type;
1219         port->physical_port = msg->port.physical_port;
1220         port->virtual_port = msg->port.virtual_port;
1221         port->status = msg->port.status;
1222         i = 0;
1223         free(port->port_name, M_CTL);
1224         port->port_name = strndup(&msg->port.data[i], msg->port.name_len,
1225             M_CTL);
1226         i += msg->port.name_len;
1227         if (msg->port.lun_map_len != 0) {
1228                 if (port->lun_map == NULL ||
1229                     port->lun_map_size * sizeof(uint32_t) <
1230                     msg->port.lun_map_len) {
1231                         port->lun_map_size = 0;
1232                         free(port->lun_map, M_CTL);
1233                         port->lun_map = malloc(msg->port.lun_map_len,
1234                             M_CTL, M_WAITOK);
1235                 }
1236                 memcpy(port->lun_map, &msg->port.data[i], msg->port.lun_map_len);
1237                 port->lun_map_size = msg->port.lun_map_len / sizeof(uint32_t);
1238                 i += msg->port.lun_map_len;
1239         } else {
1240                 port->lun_map_size = 0;
1241                 free(port->lun_map, M_CTL);
1242                 port->lun_map = NULL;
1243         }
1244         if (msg->port.port_devid_len != 0) {
1245                 if (port->port_devid == NULL ||
1246                     port->port_devid->len < msg->port.port_devid_len) {
1247                         free(port->port_devid, M_CTL);
1248                         port->port_devid = malloc(sizeof(struct ctl_devid) +
1249                             msg->port.port_devid_len, M_CTL, M_WAITOK);
1250                 }
1251                 memcpy(port->port_devid->data, &msg->port.data[i],
1252                     msg->port.port_devid_len);
1253                 port->port_devid->len = msg->port.port_devid_len;
1254                 i += msg->port.port_devid_len;
1255         } else {
1256                 free(port->port_devid, M_CTL);
1257                 port->port_devid = NULL;
1258         }
1259         if (msg->port.target_devid_len != 0) {
1260                 if (port->target_devid == NULL ||
1261                     port->target_devid->len < msg->port.target_devid_len) {
1262                         free(port->target_devid, M_CTL);
1263                         port->target_devid = malloc(sizeof(struct ctl_devid) +
1264                             msg->port.target_devid_len, M_CTL, M_WAITOK);
1265                 }
1266                 memcpy(port->target_devid->data, &msg->port.data[i],
1267                     msg->port.target_devid_len);
1268                 port->target_devid->len = msg->port.target_devid_len;
1269                 i += msg->port.target_devid_len;
1270         } else {
1271                 free(port->target_devid, M_CTL);
1272                 port->target_devid = NULL;
1273         }
1274         if (msg->port.init_devid_len != 0) {
1275                 if (port->init_devid == NULL ||
1276                     port->init_devid->len < msg->port.init_devid_len) {
1277                         free(port->init_devid, M_CTL);
1278                         port->init_devid = malloc(sizeof(struct ctl_devid) +
1279                             msg->port.init_devid_len, M_CTL, M_WAITOK);
1280                 }
1281                 memcpy(port->init_devid->data, &msg->port.data[i],
1282                     msg->port.init_devid_len);
1283                 port->init_devid->len = msg->port.init_devid_len;
1284                 i += msg->port.init_devid_len;
1285         } else {
1286                 free(port->init_devid, M_CTL);
1287                 port->init_devid = NULL;
1288         }
1289         if (new) {
1290                 if (ctl_port_register(port) != 0) {
1291                         printf("%s: ctl_port_register() failed with error\n",
1292                             __func__);
1293                 }
1294         }
1295         mtx_lock(&softc->ctl_lock);
1296         STAILQ_FOREACH(lun, &softc->lun_list, links) {
1297                 if (ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
1298                         continue;
1299                 mtx_lock(&lun->lun_lock);
1300                 ctl_est_ua_all(lun, -1, CTL_UA_INQ_CHANGE);
1301                 mtx_unlock(&lun->lun_lock);
1302         }
1303         mtx_unlock(&softc->ctl_lock);
1304 }
1305
1306 static void
1307 ctl_isc_iid_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1308 {
1309         struct ctl_port *port;
1310         int iid;
1311
1312         port = softc->ctl_ports[msg->hdr.nexus.targ_port];
1313         if (port == NULL) {
1314                 printf("%s: Received IID for unknown port %d\n",
1315                     __func__, msg->hdr.nexus.targ_port);
1316                 return;
1317         }
1318         iid = msg->hdr.nexus.initid;
1319         if (port->wwpn_iid[iid].in_use != 0 &&
1320             msg->iid.in_use == 0)
1321                 ctl_i_t_nexus_loss(softc, iid, CTL_UA_POWERON);
1322         port->wwpn_iid[iid].in_use = msg->iid.in_use;
1323         port->wwpn_iid[iid].wwpn = msg->iid.wwpn;
1324         free(port->wwpn_iid[iid].name, M_CTL);
1325         if (msg->iid.name_len) {
1326                 port->wwpn_iid[iid].name = strndup(&msg->iid.data[0],
1327                     msg->iid.name_len, M_CTL);
1328         } else
1329                 port->wwpn_iid[iid].name = NULL;
1330 }
1331
1332 static void
1333 ctl_isc_login(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1334 {
1335
1336         if (msg->login.version != CTL_HA_VERSION) {
1337                 printf("CTL HA peers have different versions %d != %d\n",
1338                     msg->login.version, CTL_HA_VERSION);
1339                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1340                 return;
1341         }
1342         if (msg->login.ha_mode != softc->ha_mode) {
1343                 printf("CTL HA peers have different ha_mode %d != %d\n",
1344                     msg->login.ha_mode, softc->ha_mode);
1345                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1346                 return;
1347         }
1348         if (msg->login.ha_id == softc->ha_id) {
1349                 printf("CTL HA peers have same ha_id %d\n", msg->login.ha_id);
1350                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1351                 return;
1352         }
1353         if (msg->login.max_luns != ctl_max_luns ||
1354             msg->login.max_ports != ctl_max_ports ||
1355             msg->login.max_init_per_port != CTL_MAX_INIT_PER_PORT) {
1356                 printf("CTL HA peers have different limits\n");
1357                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1358                 return;
1359         }
1360 }
1361
1362 static void
1363 ctl_isc_mode_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1364 {
1365         struct ctl_lun *lun;
1366         u_int i;
1367         uint32_t initidx, targ_lun;
1368
1369         targ_lun = msg->hdr.nexus.targ_mapped_lun;
1370         mtx_lock(&softc->ctl_lock);
1371         if (targ_lun >= ctl_max_luns ||
1372             (lun = softc->ctl_luns[targ_lun]) == NULL) {
1373                 mtx_unlock(&softc->ctl_lock);
1374                 return;
1375         }
1376         mtx_lock(&lun->lun_lock);
1377         mtx_unlock(&softc->ctl_lock);
1378         if (lun->flags & CTL_LUN_DISABLED) {
1379                 mtx_unlock(&lun->lun_lock);
1380                 return;
1381         }
1382         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
1383                 if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) ==
1384                     msg->mode.page_code &&
1385                     lun->mode_pages.index[i].subpage == msg->mode.subpage)
1386                         break;
1387         }
1388         if (i == CTL_NUM_MODE_PAGES) {
1389                 mtx_unlock(&lun->lun_lock);
1390                 return;
1391         }
1392         memcpy(lun->mode_pages.index[i].page_data, msg->mode.data,
1393             lun->mode_pages.index[i].page_len);
1394         initidx = ctl_get_initindex(&msg->hdr.nexus);
1395         if (initidx != -1)
1396                 ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
1397         mtx_unlock(&lun->lun_lock);
1398 }
1399
1400 /*
1401  * ISC (Inter Shelf Communication) event handler.  Events from the HA
1402  * subsystem come in here.
1403  */
1404 static void
1405 ctl_isc_event_handler(ctl_ha_channel channel, ctl_ha_event event, int param)
1406 {
1407         struct ctl_softc *softc = control_softc;
1408         union ctl_io *io;
1409         struct ctl_prio *presio;
1410         ctl_ha_status isc_status;
1411
1412         CTL_DEBUG_PRINT(("CTL: Isc Msg event %d\n", event));
1413         if (event == CTL_HA_EVT_MSG_RECV) {
1414                 union ctl_ha_msg *msg, msgbuf;
1415
1416                 if (param > sizeof(msgbuf))
1417                         msg = malloc(param, M_CTL, M_WAITOK);
1418                 else
1419                         msg = &msgbuf;
1420                 isc_status = ctl_ha_msg_recv(CTL_HA_CHAN_CTL, msg, param,
1421                     M_WAITOK);
1422                 if (isc_status != CTL_HA_STATUS_SUCCESS) {
1423                         printf("%s: Error receiving message: %d\n",
1424                             __func__, isc_status);
1425                         if (msg != &msgbuf)
1426                                 free(msg, M_CTL);
1427                         return;
1428                 }
1429
1430                 CTL_DEBUG_PRINT(("CTL: msg_type %d\n", msg->msg_type));
1431                 switch (msg->hdr.msg_type) {
1432                 case CTL_MSG_SERIALIZE:
1433                         io = ctl_alloc_io(softc->othersc_pool);
1434                         ctl_zero_io(io);
1435                         // populate ctsio from msg
1436                         io->io_hdr.io_type = CTL_IO_SCSI;
1437                         io->io_hdr.msg_type = CTL_MSG_SERIALIZE;
1438                         io->io_hdr.original_sc = msg->hdr.original_sc;
1439                         io->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC |
1440                                             CTL_FLAG_IO_ACTIVE;
1441                         /*
1442                          * If we're in serialization-only mode, we don't
1443                          * want to go through full done processing.  Thus
1444                          * the COPY flag.
1445                          *
1446                          * XXX KDM add another flag that is more specific.
1447                          */
1448                         if (softc->ha_mode != CTL_HA_MODE_XFER)
1449                                 io->io_hdr.flags |= CTL_FLAG_INT_COPY;
1450                         io->io_hdr.nexus = msg->hdr.nexus;
1451 #if 0
1452                         printf("port %u, iid %u, lun %u\n",
1453                                io->io_hdr.nexus.targ_port,
1454                                io->io_hdr.nexus.initid,
1455                                io->io_hdr.nexus.targ_lun);
1456 #endif
1457                         io->scsiio.tag_num = msg->scsi.tag_num;
1458                         io->scsiio.tag_type = msg->scsi.tag_type;
1459 #ifdef CTL_TIME_IO
1460                         io->io_hdr.start_time = time_uptime;
1461                         getbinuptime(&io->io_hdr.start_bt);
1462 #endif /* CTL_TIME_IO */
1463                         io->scsiio.cdb_len = msg->scsi.cdb_len;
1464                         memcpy(io->scsiio.cdb, msg->scsi.cdb,
1465                                CTL_MAX_CDBLEN);
1466                         if (softc->ha_mode == CTL_HA_MODE_XFER) {
1467                                 const struct ctl_cmd_entry *entry;
1468
1469                                 entry = ctl_get_cmd_entry(&io->scsiio, NULL);
1470                                 io->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
1471                                 io->io_hdr.flags |=
1472                                         entry->flags & CTL_FLAG_DATA_MASK;
1473                         }
1474                         ctl_enqueue_isc(io);
1475                         break;
1476
1477                 /* Performed on the Originating SC, XFER mode only */
1478                 case CTL_MSG_DATAMOVE: {
1479                         struct ctl_sg_entry *sgl;
1480                         int i, j;
1481
1482                         io = msg->hdr.original_sc;
1483                         if (io == NULL) {
1484                                 printf("%s: original_sc == NULL!\n", __func__);
1485                                 /* XXX KDM do something here */
1486                                 break;
1487                         }
1488                         io->io_hdr.msg_type = CTL_MSG_DATAMOVE;
1489                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1490                         /*
1491                          * Keep track of this, we need to send it back over
1492                          * when the datamove is complete.
1493                          */
1494                         io->io_hdr.serializing_sc = msg->hdr.serializing_sc;
1495                         if (msg->hdr.status == CTL_SUCCESS)
1496                                 io->io_hdr.status = msg->hdr.status;
1497
1498                         if (msg->dt.sg_sequence == 0) {
1499 #ifdef CTL_TIME_IO
1500                                 getbinuptime(&io->io_hdr.dma_start_bt);
1501 #endif
1502                                 i = msg->dt.kern_sg_entries +
1503                                     msg->dt.kern_data_len /
1504                                     CTL_HA_DATAMOVE_SEGMENT + 1;
1505                                 sgl = malloc(sizeof(*sgl) * i, M_CTL,
1506                                     M_WAITOK | M_ZERO);
1507                                 io->io_hdr.remote_sglist = sgl;
1508                                 io->io_hdr.local_sglist =
1509                                     &sgl[msg->dt.kern_sg_entries];
1510
1511                                 io->scsiio.kern_data_ptr = (uint8_t *)sgl;
1512
1513                                 io->scsiio.kern_sg_entries =
1514                                         msg->dt.kern_sg_entries;
1515                                 io->scsiio.rem_sg_entries =
1516                                         msg->dt.kern_sg_entries;
1517                                 io->scsiio.kern_data_len =
1518                                         msg->dt.kern_data_len;
1519                                 io->scsiio.kern_total_len =
1520                                         msg->dt.kern_total_len;
1521                                 io->scsiio.kern_data_resid =
1522                                         msg->dt.kern_data_resid;
1523                                 io->scsiio.kern_rel_offset =
1524                                         msg->dt.kern_rel_offset;
1525                                 io->io_hdr.flags &= ~CTL_FLAG_BUS_ADDR;
1526                                 io->io_hdr.flags |= msg->dt.flags &
1527                                     CTL_FLAG_BUS_ADDR;
1528                         } else
1529                                 sgl = (struct ctl_sg_entry *)
1530                                         io->scsiio.kern_data_ptr;
1531
1532                         for (i = msg->dt.sent_sg_entries, j = 0;
1533                              i < (msg->dt.sent_sg_entries +
1534                              msg->dt.cur_sg_entries); i++, j++) {
1535                                 sgl[i].addr = msg->dt.sg_list[j].addr;
1536                                 sgl[i].len = msg->dt.sg_list[j].len;
1537
1538 #if 0
1539                                 printf("%s: DATAMOVE: %p,%lu j=%d, i=%d\n",
1540                                     __func__, sgl[i].addr, sgl[i].len, j, i);
1541 #endif
1542                         }
1543
1544                         /*
1545                          * If this is the last piece of the I/O, we've got
1546                          * the full S/G list.  Queue processing in the thread.
1547                          * Otherwise wait for the next piece.
1548                          */
1549                         if (msg->dt.sg_last != 0)
1550                                 ctl_enqueue_isc(io);
1551                         break;
1552                 }
1553                 /* Performed on the Serializing (primary) SC, XFER mode only */
1554                 case CTL_MSG_DATAMOVE_DONE: {
1555                         if (msg->hdr.serializing_sc == NULL) {
1556                                 printf("%s: serializing_sc == NULL!\n",
1557                                        __func__);
1558                                 /* XXX KDM now what? */
1559                                 break;
1560                         }
1561                         /*
1562                          * We grab the sense information here in case
1563                          * there was a failure, so we can return status
1564                          * back to the initiator.
1565                          */
1566                         io = msg->hdr.serializing_sc;
1567                         io->io_hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
1568                         io->io_hdr.flags &= ~CTL_FLAG_DMA_INPROG;
1569                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1570                         io->io_hdr.port_status = msg->scsi.port_status;
1571                         io->scsiio.kern_data_resid = msg->scsi.kern_data_resid;
1572                         if (msg->hdr.status != CTL_STATUS_NONE) {
1573                                 io->io_hdr.status = msg->hdr.status;
1574                                 io->scsiio.scsi_status = msg->scsi.scsi_status;
1575                                 io->scsiio.sense_len = msg->scsi.sense_len;
1576                                 memcpy(&io->scsiio.sense_data,
1577                                     &msg->scsi.sense_data,
1578                                     msg->scsi.sense_len);
1579                                 if (msg->hdr.status == CTL_SUCCESS)
1580                                         io->io_hdr.flags |= CTL_FLAG_STATUS_SENT;
1581                         }
1582                         ctl_enqueue_isc(io);
1583                         break;
1584                 }
1585
1586                 /* Preformed on Originating SC, SER_ONLY mode */
1587                 case CTL_MSG_R2R:
1588                         io = msg->hdr.original_sc;
1589                         if (io == NULL) {
1590                                 printf("%s: original_sc == NULL!\n",
1591                                     __func__);
1592                                 break;
1593                         }
1594                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1595                         io->io_hdr.msg_type = CTL_MSG_R2R;
1596                         io->io_hdr.serializing_sc = msg->hdr.serializing_sc;
1597                         ctl_enqueue_isc(io);
1598                         break;
1599
1600                 /*
1601                  * Performed on Serializing(i.e. primary SC) SC in SER_ONLY
1602                  * mode.
1603                  * Performed on the Originating (i.e. secondary) SC in XFER
1604                  * mode
1605                  */
1606                 case CTL_MSG_FINISH_IO:
1607                         if (softc->ha_mode == CTL_HA_MODE_XFER)
1608                                 ctl_isc_handler_finish_xfer(softc, msg);
1609                         else
1610                                 ctl_isc_handler_finish_ser_only(softc, msg);
1611                         break;
1612
1613                 /* Preformed on Originating SC */
1614                 case CTL_MSG_BAD_JUJU:
1615                         io = msg->hdr.original_sc;
1616                         if (io == NULL) {
1617                                 printf("%s: Bad JUJU!, original_sc is NULL!\n",
1618                                        __func__);
1619                                 break;
1620                         }
1621                         ctl_copy_sense_data(msg, io);
1622                         /*
1623                          * IO should have already been cleaned up on other
1624                          * SC so clear this flag so we won't send a message
1625                          * back to finish the IO there.
1626                          */
1627                         io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
1628                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1629
1630                         /* io = msg->hdr.serializing_sc; */
1631                         io->io_hdr.msg_type = CTL_MSG_BAD_JUJU;
1632                         ctl_enqueue_isc(io);
1633                         break;
1634
1635                 /* Handle resets sent from the other side */
1636                 case CTL_MSG_MANAGE_TASKS: {
1637                         struct ctl_taskio *taskio;
1638                         taskio = (struct ctl_taskio *)ctl_alloc_io(
1639                             softc->othersc_pool);
1640                         ctl_zero_io((union ctl_io *)taskio);
1641                         taskio->io_hdr.io_type = CTL_IO_TASK;
1642                         taskio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
1643                         taskio->io_hdr.nexus = msg->hdr.nexus;
1644                         taskio->task_action = msg->task.task_action;
1645                         taskio->tag_num = msg->task.tag_num;
1646                         taskio->tag_type = msg->task.tag_type;
1647 #ifdef CTL_TIME_IO
1648                         taskio->io_hdr.start_time = time_uptime;
1649                         getbinuptime(&taskio->io_hdr.start_bt);
1650 #endif /* CTL_TIME_IO */
1651                         ctl_run_task((union ctl_io *)taskio);
1652                         break;
1653                 }
1654                 /* Persistent Reserve action which needs attention */
1655                 case CTL_MSG_PERS_ACTION:
1656                         presio = (struct ctl_prio *)ctl_alloc_io(
1657                             softc->othersc_pool);
1658                         ctl_zero_io((union ctl_io *)presio);
1659                         presio->io_hdr.msg_type = CTL_MSG_PERS_ACTION;
1660                         presio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
1661                         presio->io_hdr.nexus = msg->hdr.nexus;
1662                         presio->pr_msg = msg->pr;
1663                         ctl_enqueue_isc((union ctl_io *)presio);
1664                         break;
1665                 case CTL_MSG_UA:
1666                         ctl_isc_ua(softc, msg, param);
1667                         break;
1668                 case CTL_MSG_PORT_SYNC:
1669                         ctl_isc_port_sync(softc, msg, param);
1670                         break;
1671                 case CTL_MSG_LUN_SYNC:
1672                         ctl_isc_lun_sync(softc, msg, param);
1673                         break;
1674                 case CTL_MSG_IID_SYNC:
1675                         ctl_isc_iid_sync(softc, msg, param);
1676                         break;
1677                 case CTL_MSG_LOGIN:
1678                         ctl_isc_login(softc, msg, param);
1679                         break;
1680                 case CTL_MSG_MODE_SYNC:
1681                         ctl_isc_mode_sync(softc, msg, param);
1682                         break;
1683                 default:
1684                         printf("Received HA message of unknown type %d\n",
1685                             msg->hdr.msg_type);
1686                         ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1687                         break;
1688                 }
1689                 if (msg != &msgbuf)
1690                         free(msg, M_CTL);
1691         } else if (event == CTL_HA_EVT_LINK_CHANGE) {
1692                 printf("CTL: HA link status changed from %d to %d\n",
1693                     softc->ha_link, param);
1694                 if (param == softc->ha_link)
1695                         return;
1696                 if (softc->ha_link == CTL_HA_LINK_ONLINE) {
1697                         softc->ha_link = param;
1698                         ctl_isc_ha_link_down(softc);
1699                 } else {
1700                         softc->ha_link = param;
1701                         if (softc->ha_link == CTL_HA_LINK_ONLINE)
1702                                 ctl_isc_ha_link_up(softc);
1703                 }
1704                 return;
1705         } else {
1706                 printf("ctl_isc_event_handler: Unknown event %d\n", event);
1707                 return;
1708         }
1709 }
1710
1711 static void
1712 ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest)
1713 {
1714
1715         memcpy(&dest->scsiio.sense_data, &src->scsi.sense_data,
1716             src->scsi.sense_len);
1717         dest->scsiio.scsi_status = src->scsi.scsi_status;
1718         dest->scsiio.sense_len = src->scsi.sense_len;
1719         dest->io_hdr.status = src->hdr.status;
1720 }
1721
1722 static void
1723 ctl_copy_sense_data_back(union ctl_io *src, union ctl_ha_msg *dest)
1724 {
1725
1726         memcpy(&dest->scsi.sense_data, &src->scsiio.sense_data,
1727             src->scsiio.sense_len);
1728         dest->scsi.scsi_status = src->scsiio.scsi_status;
1729         dest->scsi.sense_len = src->scsiio.sense_len;
1730         dest->hdr.status = src->io_hdr.status;
1731 }
1732
1733 void
1734 ctl_est_ua(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua)
1735 {
1736         struct ctl_softc *softc = lun->ctl_softc;
1737         ctl_ua_type *pu;
1738
1739         if (initidx < softc->init_min || initidx >= softc->init_max)
1740                 return;
1741         mtx_assert(&lun->lun_lock, MA_OWNED);
1742         pu = lun->pending_ua[initidx / CTL_MAX_INIT_PER_PORT];
1743         if (pu == NULL)
1744                 return;
1745         pu[initidx % CTL_MAX_INIT_PER_PORT] |= ua;
1746 }
1747
1748 void
1749 ctl_est_ua_port(struct ctl_lun *lun, int port, uint32_t except, ctl_ua_type ua)
1750 {
1751         int i;
1752
1753         mtx_assert(&lun->lun_lock, MA_OWNED);
1754         if (lun->pending_ua[port] == NULL)
1755                 return;
1756         for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1757                 if (port * CTL_MAX_INIT_PER_PORT + i == except)
1758                         continue;
1759                 lun->pending_ua[port][i] |= ua;
1760         }
1761 }
1762
1763 void
1764 ctl_est_ua_all(struct ctl_lun *lun, uint32_t except, ctl_ua_type ua)
1765 {
1766         struct ctl_softc *softc = lun->ctl_softc;
1767         int i;
1768
1769         mtx_assert(&lun->lun_lock, MA_OWNED);
1770         for (i = softc->port_min; i < softc->port_max; i++)
1771                 ctl_est_ua_port(lun, i, except, ua);
1772 }
1773
1774 void
1775 ctl_clr_ua(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua)
1776 {
1777         struct ctl_softc *softc = lun->ctl_softc;
1778         ctl_ua_type *pu;
1779
1780         if (initidx < softc->init_min || initidx >= softc->init_max)
1781                 return;
1782         mtx_assert(&lun->lun_lock, MA_OWNED);
1783         pu = lun->pending_ua[initidx / CTL_MAX_INIT_PER_PORT];
1784         if (pu == NULL)
1785                 return;
1786         pu[initidx % CTL_MAX_INIT_PER_PORT] &= ~ua;
1787 }
1788
1789 void
1790 ctl_clr_ua_all(struct ctl_lun *lun, uint32_t except, ctl_ua_type ua)
1791 {
1792         struct ctl_softc *softc = lun->ctl_softc;
1793         int i, j;
1794
1795         mtx_assert(&lun->lun_lock, MA_OWNED);
1796         for (i = softc->port_min; i < softc->port_max; i++) {
1797                 if (lun->pending_ua[i] == NULL)
1798                         continue;
1799                 for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
1800                         if (i * CTL_MAX_INIT_PER_PORT + j == except)
1801                                 continue;
1802                         lun->pending_ua[i][j] &= ~ua;
1803                 }
1804         }
1805 }
1806
1807 void
1808 ctl_clr_ua_allluns(struct ctl_softc *ctl_softc, uint32_t initidx,
1809     ctl_ua_type ua_type)
1810 {
1811         struct ctl_lun *lun;
1812
1813         mtx_assert(&ctl_softc->ctl_lock, MA_OWNED);
1814         STAILQ_FOREACH(lun, &ctl_softc->lun_list, links) {
1815                 mtx_lock(&lun->lun_lock);
1816                 ctl_clr_ua(lun, initidx, ua_type);
1817                 mtx_unlock(&lun->lun_lock);
1818         }
1819 }
1820
1821 static int
1822 ctl_ha_role_sysctl(SYSCTL_HANDLER_ARGS)
1823 {
1824         struct ctl_softc *softc = (struct ctl_softc *)arg1;
1825         struct ctl_lun *lun;
1826         struct ctl_lun_req ireq;
1827         int error, value;
1828
1829         value = (softc->flags & CTL_FLAG_ACTIVE_SHELF) ? 0 : 1;
1830         error = sysctl_handle_int(oidp, &value, 0, req);
1831         if ((error != 0) || (req->newptr == NULL))
1832                 return (error);
1833
1834         mtx_lock(&softc->ctl_lock);
1835         if (value == 0)
1836                 softc->flags |= CTL_FLAG_ACTIVE_SHELF;
1837         else
1838                 softc->flags &= ~CTL_FLAG_ACTIVE_SHELF;
1839         STAILQ_FOREACH(lun, &softc->lun_list, links) {
1840                 mtx_unlock(&softc->ctl_lock);
1841                 bzero(&ireq, sizeof(ireq));
1842                 ireq.reqtype = CTL_LUNREQ_MODIFY;
1843                 ireq.reqdata.modify.lun_id = lun->lun;
1844                 lun->backend->ioctl(NULL, CTL_LUN_REQ, (caddr_t)&ireq, 0,
1845                     curthread);
1846                 if (ireq.status != CTL_LUN_OK) {
1847                         printf("%s: CTL_LUNREQ_MODIFY returned %d '%s'\n",
1848                             __func__, ireq.status, ireq.error_str);
1849                 }
1850                 mtx_lock(&softc->ctl_lock);
1851         }
1852         mtx_unlock(&softc->ctl_lock);
1853         return (0);
1854 }
1855
1856 static int
1857 ctl_init(void)
1858 {
1859         struct make_dev_args args;
1860         struct ctl_softc *softc;
1861         int i, error;
1862
1863         softc = control_softc = malloc(sizeof(*control_softc), M_DEVBUF,
1864                                M_WAITOK | M_ZERO);
1865
1866         make_dev_args_init(&args);
1867         args.mda_devsw = &ctl_cdevsw;
1868         args.mda_uid = UID_ROOT;
1869         args.mda_gid = GID_OPERATOR;
1870         args.mda_mode = 0600;
1871         args.mda_si_drv1 = softc;
1872         error = make_dev_s(&args, &softc->dev, "cam/ctl");
1873         if (error != 0) {
1874                 free(softc, M_DEVBUF);
1875                 control_softc = NULL;
1876                 return (error);
1877         }
1878
1879         sysctl_ctx_init(&softc->sysctl_ctx);
1880         softc->sysctl_tree = SYSCTL_ADD_NODE(&softc->sysctl_ctx,
1881                 SYSCTL_STATIC_CHILDREN(_kern_cam), OID_AUTO, "ctl",
1882                 CTLFLAG_RD, 0, "CAM Target Layer");
1883
1884         if (softc->sysctl_tree == NULL) {
1885                 printf("%s: unable to allocate sysctl tree\n", __func__);
1886                 destroy_dev(softc->dev);
1887                 free(softc, M_DEVBUF);
1888                 control_softc = NULL;
1889                 return (ENOMEM);
1890         }
1891
1892         mtx_init(&softc->ctl_lock, "CTL mutex", NULL, MTX_DEF);
1893         softc->io_zone = uma_zcreate("CTL IO", sizeof(union ctl_io),
1894             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1895         softc->flags = 0;
1896
1897         SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1898             OID_AUTO, "ha_mode", CTLFLAG_RDTUN, (int *)&softc->ha_mode, 0,
1899             "HA mode (0 - act/stby, 1 - serialize only, 2 - xfer)");
1900
1901         if (ctl_max_luns <= 0 || powerof2(ctl_max_luns) == 0) {
1902                 printf("Bad value %d for kern.cam.ctl.max_luns, must be a power of two, using %d\n",
1903                     ctl_max_luns, CTL_DEFAULT_MAX_LUNS);
1904                 ctl_max_luns = CTL_DEFAULT_MAX_LUNS;
1905         }
1906         softc->ctl_luns = malloc(sizeof(struct ctl_lun *) * ctl_max_luns,
1907             M_DEVBUF, M_WAITOK | M_ZERO);
1908         softc->ctl_lun_mask = malloc(sizeof(uint32_t) *
1909             ((ctl_max_luns + 31) / 32), M_DEVBUF, M_WAITOK | M_ZERO);
1910         if (ctl_max_ports <= 0 || powerof2(ctl_max_ports) == 0) {
1911                 printf("Bad value %d for kern.cam.ctl.max_ports, must be a power of two, using %d\n",
1912                     ctl_max_ports, CTL_DEFAULT_MAX_PORTS);
1913                 ctl_max_ports = CTL_DEFAULT_MAX_PORTS;
1914         }
1915         softc->ctl_port_mask = malloc(sizeof(uint32_t) *
1916           ((ctl_max_ports + 31) / 32), M_DEVBUF, M_WAITOK | M_ZERO);
1917         softc->ctl_ports = malloc(sizeof(struct ctl_port *) * ctl_max_ports,
1918              M_DEVBUF, M_WAITOK | M_ZERO);
1919
1920
1921         /*
1922          * In Copan's HA scheme, the "master" and "slave" roles are
1923          * figured out through the slot the controller is in.  Although it
1924          * is an active/active system, someone has to be in charge.
1925          */
1926         SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1927             OID_AUTO, "ha_id", CTLFLAG_RDTUN, &softc->ha_id, 0,
1928             "HA head ID (0 - no HA)");
1929         if (softc->ha_id == 0 || softc->ha_id > NUM_HA_SHELVES) {
1930                 softc->flags |= CTL_FLAG_ACTIVE_SHELF;
1931                 softc->is_single = 1;
1932                 softc->port_cnt = ctl_max_ports;
1933                 softc->port_min = 0;
1934         } else {
1935                 softc->port_cnt = ctl_max_ports / NUM_HA_SHELVES;
1936                 softc->port_min = (softc->ha_id - 1) * softc->port_cnt;
1937         }
1938         softc->port_max = softc->port_min + softc->port_cnt;
1939         softc->init_min = softc->port_min * CTL_MAX_INIT_PER_PORT;
1940         softc->init_max = softc->port_max * CTL_MAX_INIT_PER_PORT;
1941
1942         SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1943             OID_AUTO, "ha_link", CTLFLAG_RD, (int *)&softc->ha_link, 0,
1944             "HA link state (0 - offline, 1 - unknown, 2 - online)");
1945
1946         STAILQ_INIT(&softc->lun_list);
1947         STAILQ_INIT(&softc->pending_lun_queue);
1948         STAILQ_INIT(&softc->fe_list);
1949         STAILQ_INIT(&softc->port_list);
1950         STAILQ_INIT(&softc->be_list);
1951         ctl_tpc_init(softc);
1952
1953         if (worker_threads <= 0)
1954                 worker_threads = max(1, mp_ncpus / 4);
1955         if (worker_threads > CTL_MAX_THREADS)
1956                 worker_threads = CTL_MAX_THREADS;
1957
1958         for (i = 0; i < worker_threads; i++) {
1959                 struct ctl_thread *thr = &softc->threads[i];
1960
1961                 mtx_init(&thr->queue_lock, "CTL queue mutex", NULL, MTX_DEF);
1962                 thr->ctl_softc = softc;
1963                 STAILQ_INIT(&thr->incoming_queue);
1964                 STAILQ_INIT(&thr->rtr_queue);
1965                 STAILQ_INIT(&thr->done_queue);
1966                 STAILQ_INIT(&thr->isc_queue);
1967
1968                 error = kproc_kthread_add(ctl_work_thread, thr,
1969                     &softc->ctl_proc, &thr->thread, 0, 0, "ctl", "work%d", i);
1970                 if (error != 0) {
1971                         printf("error creating CTL work thread!\n");
1972                         return (error);
1973                 }
1974         }
1975         error = kproc_kthread_add(ctl_lun_thread, softc,
1976             &softc->ctl_proc, &softc->lun_thread, 0, 0, "ctl", "lun");
1977         if (error != 0) {
1978                 printf("error creating CTL lun thread!\n");
1979                 return (error);
1980         }
1981         error = kproc_kthread_add(ctl_thresh_thread, softc,
1982             &softc->ctl_proc, &softc->thresh_thread, 0, 0, "ctl", "thresh");
1983         if (error != 0) {
1984                 printf("error creating CTL threshold thread!\n");
1985                 return (error);
1986         }
1987
1988         SYSCTL_ADD_PROC(&softc->sysctl_ctx,SYSCTL_CHILDREN(softc->sysctl_tree),
1989             OID_AUTO, "ha_role", CTLTYPE_INT | CTLFLAG_RWTUN,
1990             softc, 0, ctl_ha_role_sysctl, "I", "HA role for this head");
1991
1992         if (softc->is_single == 0) {
1993                 if (ctl_frontend_register(&ha_frontend) != 0)
1994                         softc->is_single = 1;
1995         }
1996         return (0);
1997 }
1998
1999 static int
2000 ctl_shutdown(void)
2001 {
2002         struct ctl_softc *softc = control_softc;
2003         int i;
2004
2005         if (softc->is_single == 0)
2006                 ctl_frontend_deregister(&ha_frontend);
2007
2008         destroy_dev(softc->dev);
2009
2010         /* Shutdown CTL threads. */
2011         softc->shutdown = 1;
2012         for (i = 0; i < worker_threads; i++) {
2013                 struct ctl_thread *thr = &softc->threads[i];
2014                 while (thr->thread != NULL) {
2015                         wakeup(thr);
2016                         if (thr->thread != NULL)
2017                                 pause("CTL thr shutdown", 1);
2018                 }
2019                 mtx_destroy(&thr->queue_lock);
2020         }
2021         while (softc->lun_thread != NULL) {
2022                 wakeup(&softc->pending_lun_queue);
2023                 if (softc->lun_thread != NULL)
2024                         pause("CTL thr shutdown", 1);
2025         }
2026         while (softc->thresh_thread != NULL) {
2027                 wakeup(softc->thresh_thread);
2028                 if (softc->thresh_thread != NULL)
2029                         pause("CTL thr shutdown", 1);
2030         }
2031
2032         ctl_tpc_shutdown(softc);
2033         uma_zdestroy(softc->io_zone);
2034         mtx_destroy(&softc->ctl_lock);
2035
2036         free(softc->ctl_luns, M_DEVBUF);
2037         free(softc->ctl_lun_mask, M_DEVBUF);
2038         free(softc->ctl_port_mask, M_DEVBUF);
2039         free(softc->ctl_ports, M_DEVBUF);
2040
2041         sysctl_ctx_free(&softc->sysctl_ctx);
2042
2043         free(softc, M_DEVBUF);
2044         control_softc = NULL;
2045         return (0);
2046 }
2047
2048 static int
2049 ctl_module_event_handler(module_t mod, int what, void *arg)
2050 {
2051
2052         switch (what) {
2053         case MOD_LOAD:
2054                 return (ctl_init());
2055         case MOD_UNLOAD:
2056                 return (ctl_shutdown());
2057         default:
2058                 return (EOPNOTSUPP);
2059         }
2060 }
2061
2062 /*
2063  * XXX KDM should we do some access checks here?  Bump a reference count to
2064  * prevent a CTL module from being unloaded while someone has it open?
2065  */
2066 static int
2067 ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td)
2068 {
2069         return (0);
2070 }
2071
2072 static int
2073 ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td)
2074 {
2075         return (0);
2076 }
2077
2078 /*
2079  * Remove an initiator by port number and initiator ID.
2080  * Returns 0 for success, -1 for failure.
2081  */
2082 int
2083 ctl_remove_initiator(struct ctl_port *port, int iid)
2084 {
2085         struct ctl_softc *softc = port->ctl_softc;
2086         int last;
2087
2088         mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
2089
2090         if (iid > CTL_MAX_INIT_PER_PORT) {
2091                 printf("%s: initiator ID %u > maximun %u!\n",
2092                        __func__, iid, CTL_MAX_INIT_PER_PORT);
2093                 return (-1);
2094         }
2095
2096         mtx_lock(&softc->ctl_lock);
2097         last = (--port->wwpn_iid[iid].in_use == 0);
2098         port->wwpn_iid[iid].last_use = time_uptime;
2099         mtx_unlock(&softc->ctl_lock);
2100         if (last)
2101                 ctl_i_t_nexus_loss(softc, iid, CTL_UA_POWERON);
2102         ctl_isc_announce_iid(port, iid);
2103
2104         return (0);
2105 }
2106
2107 /*
2108  * Add an initiator to the initiator map.
2109  * Returns iid for success, < 0 for failure.
2110  */
2111 int
2112 ctl_add_initiator(struct ctl_port *port, int iid, uint64_t wwpn, char *name)
2113 {
2114         struct ctl_softc *softc = port->ctl_softc;
2115         time_t best_time;
2116         int i, best;
2117
2118         mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
2119
2120         if (iid >= CTL_MAX_INIT_PER_PORT) {
2121                 printf("%s: WWPN %#jx initiator ID %u > maximum %u!\n",
2122                        __func__, wwpn, iid, CTL_MAX_INIT_PER_PORT);
2123                 free(name, M_CTL);
2124                 return (-1);
2125         }
2126
2127         mtx_lock(&softc->ctl_lock);
2128
2129         if (iid < 0 && (wwpn != 0 || name != NULL)) {
2130                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
2131                         if (wwpn != 0 && wwpn == port->wwpn_iid[i].wwpn) {
2132                                 iid = i;
2133                                 break;
2134                         }
2135                         if (name != NULL && port->wwpn_iid[i].name != NULL &&
2136                             strcmp(name, port->wwpn_iid[i].name) == 0) {
2137                                 iid = i;
2138                                 break;
2139                         }
2140                 }
2141         }
2142
2143         if (iid < 0) {
2144                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
2145                         if (port->wwpn_iid[i].in_use == 0 &&
2146                             port->wwpn_iid[i].wwpn == 0 &&
2147                             port->wwpn_iid[i].name == NULL) {
2148                                 iid = i;
2149                                 break;
2150                         }
2151                 }
2152         }
2153
2154         if (iid < 0) {
2155                 best = -1;
2156                 best_time = INT32_MAX;
2157                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
2158                         if (port->wwpn_iid[i].in_use == 0) {
2159                                 if (port->wwpn_iid[i].last_use < best_time) {
2160                                         best = i;
2161                                         best_time = port->wwpn_iid[i].last_use;
2162                                 }
2163                         }
2164                 }
2165                 iid = best;
2166         }
2167
2168         if (iid < 0) {
2169                 mtx_unlock(&softc->ctl_lock);
2170                 free(name, M_CTL);
2171                 return (-2);
2172         }
2173
2174         if (port->wwpn_iid[iid].in_use > 0 && (wwpn != 0 || name != NULL)) {
2175                 /*
2176                  * This is not an error yet.
2177                  */
2178                 if (wwpn != 0 && wwpn == port->wwpn_iid[iid].wwpn) {
2179 #if 0
2180                         printf("%s: port %d iid %u WWPN %#jx arrived"
2181                             " again\n", __func__, port->targ_port,
2182                             iid, (uintmax_t)wwpn);
2183 #endif
2184                         goto take;
2185                 }
2186                 if (name != NULL && port->wwpn_iid[iid].name != NULL &&
2187                     strcmp(name, port->wwpn_iid[iid].name) == 0) {
2188 #if 0
2189                         printf("%s: port %d iid %u name '%s' arrived"
2190                             " again\n", __func__, port->targ_port,
2191                             iid, name);
2192 #endif
2193                         goto take;
2194                 }
2195
2196                 /*
2197                  * This is an error, but what do we do about it?  The
2198                  * driver is telling us we have a new WWPN for this
2199                  * initiator ID, so we pretty much need to use it.
2200                  */
2201                 printf("%s: port %d iid %u WWPN %#jx '%s' arrived,"
2202                     " but WWPN %#jx '%s' is still at that address\n",
2203                     __func__, port->targ_port, iid, wwpn, name,
2204                     (uintmax_t)port->wwpn_iid[iid].wwpn,
2205                     port->wwpn_iid[iid].name);
2206         }
2207 take:
2208         free(port->wwpn_iid[iid].name, M_CTL);
2209         port->wwpn_iid[iid].name = name;
2210         port->wwpn_iid[iid].wwpn = wwpn;
2211         port->wwpn_iid[iid].in_use++;
2212         mtx_unlock(&softc->ctl_lock);
2213         ctl_isc_announce_iid(port, iid);
2214
2215         return (iid);
2216 }
2217
2218 static int
2219 ctl_create_iid(struct ctl_port *port, int iid, uint8_t *buf)
2220 {
2221         int len;
2222
2223         switch (port->port_type) {
2224         case CTL_PORT_FC:
2225         {
2226                 struct scsi_transportid_fcp *id =
2227                     (struct scsi_transportid_fcp *)buf;
2228                 if (port->wwpn_iid[iid].wwpn == 0)
2229                         return (0);
2230                 memset(id, 0, sizeof(*id));
2231                 id->format_protocol = SCSI_PROTO_FC;
2232                 scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->n_port_name);
2233                 return (sizeof(*id));
2234         }
2235         case CTL_PORT_ISCSI:
2236         {
2237                 struct scsi_transportid_iscsi_port *id =
2238                     (struct scsi_transportid_iscsi_port *)buf;
2239                 if (port->wwpn_iid[iid].name == NULL)
2240                         return (0);
2241                 memset(id, 0, 256);
2242                 id->format_protocol = SCSI_TRN_ISCSI_FORMAT_PORT |
2243                     SCSI_PROTO_ISCSI;
2244                 len = strlcpy(id->iscsi_name, port->wwpn_iid[iid].name, 252) + 1;
2245                 len = roundup2(min(len, 252), 4);
2246                 scsi_ulto2b(len, id->additional_length);
2247                 return (sizeof(*id) + len);
2248         }
2249         case CTL_PORT_SAS:
2250         {
2251                 struct scsi_transportid_sas *id =
2252                     (struct scsi_transportid_sas *)buf;
2253                 if (port->wwpn_iid[iid].wwpn == 0)
2254                         return (0);
2255                 memset(id, 0, sizeof(*id));
2256                 id->format_protocol = SCSI_PROTO_SAS;
2257                 scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->sas_address);
2258                 return (sizeof(*id));
2259         }
2260         default:
2261         {
2262                 struct scsi_transportid_spi *id =
2263                     (struct scsi_transportid_spi *)buf;
2264                 memset(id, 0, sizeof(*id));
2265                 id->format_protocol = SCSI_PROTO_SPI;
2266                 scsi_ulto2b(iid, id->scsi_addr);
2267                 scsi_ulto2b(port->targ_port, id->rel_trgt_port_id);
2268                 return (sizeof(*id));
2269         }
2270         }
2271 }
2272
2273 /*
2274  * Serialize a command that went down the "wrong" side, and so was sent to
2275  * this controller for execution.  The logic is a little different than the
2276  * standard case in ctl_scsiio_precheck().  Errors in this case need to get
2277  * sent back to the other side, but in the success case, we execute the
2278  * command on this side (XFER mode) or tell the other side to execute it
2279  * (SER_ONLY mode).
2280  */
2281 static void
2282 ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio)
2283 {
2284         struct ctl_softc *softc = CTL_SOFTC(ctsio);
2285         struct ctl_port *port = CTL_PORT(ctsio);
2286         union ctl_ha_msg msg_info;
2287         struct ctl_lun *lun;
2288         const struct ctl_cmd_entry *entry;
2289         uint32_t targ_lun;
2290
2291         targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
2292
2293         /* Make sure that we know about this port. */
2294         if (port == NULL || (port->status & CTL_PORT_STATUS_ONLINE) == 0) {
2295                 ctl_set_internal_failure(ctsio, /*sks_valid*/ 0,
2296                                          /*retry_count*/ 1);
2297                 goto badjuju;
2298         }
2299
2300         /* Make sure that we know about this LUN. */
2301         mtx_lock(&softc->ctl_lock);
2302         if (targ_lun >= ctl_max_luns ||
2303             (lun = softc->ctl_luns[targ_lun]) == NULL) {
2304                 mtx_unlock(&softc->ctl_lock);
2305
2306                 /*
2307                  * The other node would not send this request to us unless
2308                  * received announce that we are primary node for this LUN.
2309                  * If this LUN does not exist now, it is probably result of
2310                  * a race, so respond to initiator in the most opaque way.
2311                  */
2312                 ctl_set_busy(ctsio);
2313                 goto badjuju;
2314         }
2315         mtx_lock(&lun->lun_lock);
2316         mtx_unlock(&softc->ctl_lock);
2317
2318         /*
2319          * If the LUN is invalid, pretend that it doesn't exist.
2320          * It will go away as soon as all pending I/Os completed.
2321          */
2322         if (lun->flags & CTL_LUN_DISABLED) {
2323                 mtx_unlock(&lun->lun_lock);
2324                 ctl_set_busy(ctsio);
2325                 goto badjuju;
2326         }
2327
2328         entry = ctl_get_cmd_entry(ctsio, NULL);
2329         if (ctl_scsiio_lun_check(lun, entry, ctsio) != 0) {
2330                 mtx_unlock(&lun->lun_lock);
2331                 goto badjuju;
2332         }
2333
2334         CTL_LUN(ctsio) = lun;
2335         CTL_BACKEND_LUN(ctsio) = lun->be_lun;
2336
2337         /*
2338          * Every I/O goes into the OOA queue for a
2339          * particular LUN, and stays there until completion.
2340          */
2341 #ifdef CTL_TIME_IO
2342         if (TAILQ_EMPTY(&lun->ooa_queue))
2343                 lun->idle_time += getsbinuptime() - lun->last_busy;
2344 #endif
2345         TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2346
2347         switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
2348                 (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr, ctl_ooaq,
2349                  ooa_links))) {
2350         case CTL_ACTION_BLOCK:
2351                 ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
2352                 TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
2353                                   blocked_links);
2354                 mtx_unlock(&lun->lun_lock);
2355                 break;
2356         case CTL_ACTION_PASS:
2357         case CTL_ACTION_SKIP:
2358                 if (softc->ha_mode == CTL_HA_MODE_XFER) {
2359                         ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
2360                         ctl_enqueue_rtr((union ctl_io *)ctsio);
2361                         mtx_unlock(&lun->lun_lock);
2362                 } else {
2363                         ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
2364                         mtx_unlock(&lun->lun_lock);
2365
2366                         /* send msg back to other side */
2367                         msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
2368                         msg_info.hdr.serializing_sc = (union ctl_io *)ctsio;
2369                         msg_info.hdr.msg_type = CTL_MSG_R2R;
2370                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
2371                             sizeof(msg_info.hdr), M_WAITOK);
2372                 }
2373                 break;
2374         case CTL_ACTION_OVERLAP:
2375                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2376                 mtx_unlock(&lun->lun_lock);
2377                 ctl_set_overlapped_cmd(ctsio);
2378                 goto badjuju;
2379         case CTL_ACTION_OVERLAP_TAG:
2380                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2381                 mtx_unlock(&lun->lun_lock);
2382                 ctl_set_overlapped_tag(ctsio, ctsio->tag_num);
2383                 goto badjuju;
2384         case CTL_ACTION_ERROR:
2385         default:
2386                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2387                 mtx_unlock(&lun->lun_lock);
2388
2389                 ctl_set_internal_failure(ctsio, /*sks_valid*/ 0,
2390                                          /*retry_count*/ 0);
2391 badjuju:
2392                 ctl_copy_sense_data_back((union ctl_io *)ctsio, &msg_info);
2393                 msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
2394                 msg_info.hdr.serializing_sc = NULL;
2395                 msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
2396                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
2397                     sizeof(msg_info.scsi), M_WAITOK);
2398                 ctl_free_io((union ctl_io *)ctsio);
2399                 break;
2400         }
2401 }
2402
2403 /*
2404  * Returns 0 for success, errno for failure.
2405  */
2406 static void
2407 ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
2408                    struct ctl_ooa *ooa_hdr, struct ctl_ooa_entry *kern_entries)
2409 {
2410         union ctl_io *io;
2411
2412         mtx_lock(&lun->lun_lock);
2413         for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); (io != NULL);
2414              (*cur_fill_num)++, io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2415              ooa_links)) {
2416                 struct ctl_ooa_entry *entry;
2417
2418                 /*
2419                  * If we've got more than we can fit, just count the
2420                  * remaining entries.
2421                  */
2422                 if (*cur_fill_num >= ooa_hdr->alloc_num)
2423                         continue;
2424
2425                 entry = &kern_entries[*cur_fill_num];
2426
2427                 entry->tag_num = io->scsiio.tag_num;
2428                 entry->lun_num = lun->lun;
2429 #ifdef CTL_TIME_IO
2430                 entry->start_bt = io->io_hdr.start_bt;
2431 #endif
2432                 bcopy(io->scsiio.cdb, entry->cdb, io->scsiio.cdb_len);
2433                 entry->cdb_len = io->scsiio.cdb_len;
2434                 if (io->io_hdr.flags & CTL_FLAG_BLOCKED)
2435                         entry->cmd_flags |= CTL_OOACMD_FLAG_BLOCKED;
2436
2437                 if (io->io_hdr.flags & CTL_FLAG_DMA_INPROG)
2438                         entry->cmd_flags |= CTL_OOACMD_FLAG_DMA;
2439
2440                 if (io->io_hdr.flags & CTL_FLAG_ABORT)
2441                         entry->cmd_flags |= CTL_OOACMD_FLAG_ABORT;
2442
2443                 if (io->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR)
2444                         entry->cmd_flags |= CTL_OOACMD_FLAG_RTR;
2445
2446                 if (io->io_hdr.flags & CTL_FLAG_DMA_QUEUED)
2447                         entry->cmd_flags |= CTL_OOACMD_FLAG_DMA_QUEUED;
2448         }
2449         mtx_unlock(&lun->lun_lock);
2450 }
2451
2452 static void *
2453 ctl_copyin_alloc(void *user_addr, unsigned int len, char *error_str,
2454                  size_t error_str_len)
2455 {
2456         void *kptr;
2457
2458         kptr = malloc(len, M_CTL, M_WAITOK | M_ZERO);
2459
2460         if (copyin(user_addr, kptr, len) != 0) {
2461                 snprintf(error_str, error_str_len, "Error copying %d bytes "
2462                          "from user address %p to kernel address %p", len,
2463                          user_addr, kptr);
2464                 free(kptr, M_CTL);
2465                 return (NULL);
2466         }
2467
2468         return (kptr);
2469 }
2470
2471 static void
2472 ctl_free_args(int num_args, struct ctl_be_arg *args)
2473 {
2474         int i;
2475
2476         if (args == NULL)
2477                 return;
2478
2479         for (i = 0; i < num_args; i++) {
2480                 free(args[i].kname, M_CTL);
2481                 free(args[i].kvalue, M_CTL);
2482         }
2483
2484         free(args, M_CTL);
2485 }
2486
2487 static struct ctl_be_arg *
2488 ctl_copyin_args(int num_args, struct ctl_be_arg *uargs,
2489                 char *error_str, size_t error_str_len)
2490 {
2491         struct ctl_be_arg *args;
2492         int i;
2493
2494         args = ctl_copyin_alloc(uargs, num_args * sizeof(*args),
2495                                 error_str, error_str_len);
2496
2497         if (args == NULL)
2498                 goto bailout;
2499
2500         for (i = 0; i < num_args; i++) {
2501                 args[i].kname = NULL;
2502                 args[i].kvalue = NULL;
2503         }
2504
2505         for (i = 0; i < num_args; i++) {
2506                 uint8_t *tmpptr;
2507
2508                 if (args[i].namelen == 0) {
2509                         snprintf(error_str, error_str_len, "Argument %d "
2510                                  "name length is zero", i);
2511                         goto bailout;
2512                 }
2513
2514                 args[i].kname = ctl_copyin_alloc(args[i].name,
2515                         args[i].namelen, error_str, error_str_len);
2516                 if (args[i].kname == NULL)
2517                         goto bailout;
2518
2519                 if (args[i].kname[args[i].namelen - 1] != '\0') {
2520                         snprintf(error_str, error_str_len, "Argument %d "
2521                                  "name is not NUL-terminated", i);
2522                         goto bailout;
2523                 }
2524
2525                 if (args[i].flags & CTL_BEARG_RD) {
2526                         if (args[i].vallen == 0) {
2527                                 snprintf(error_str, error_str_len, "Argument %d "
2528                                          "value length is zero", i);
2529                                 goto bailout;
2530                         }
2531
2532                         tmpptr = ctl_copyin_alloc(args[i].value,
2533                                 args[i].vallen, error_str, error_str_len);
2534                         if (tmpptr == NULL)
2535                                 goto bailout;
2536
2537                         if ((args[i].flags & CTL_BEARG_ASCII)
2538                          && (tmpptr[args[i].vallen - 1] != '\0')) {
2539                                 snprintf(error_str, error_str_len, "Argument "
2540                                     "%d value is not NUL-terminated", i);
2541                                 free(tmpptr, M_CTL);
2542                                 goto bailout;
2543                         }
2544                         args[i].kvalue = tmpptr;
2545                 } else {
2546                         args[i].kvalue = malloc(args[i].vallen,
2547                             M_CTL, M_WAITOK | M_ZERO);
2548                 }
2549         }
2550
2551         return (args);
2552 bailout:
2553
2554         ctl_free_args(num_args, args);
2555
2556         return (NULL);
2557 }
2558
2559 static void
2560 ctl_copyout_args(int num_args, struct ctl_be_arg *args)
2561 {
2562         int i;
2563
2564         for (i = 0; i < num_args; i++) {
2565                 if (args[i].flags & CTL_BEARG_WR)
2566                         copyout(args[i].kvalue, args[i].value, args[i].vallen);
2567         }
2568 }
2569
2570 /*
2571  * Escape characters that are illegal or not recommended in XML.
2572  */
2573 int
2574 ctl_sbuf_printf_esc(struct sbuf *sb, char *str, int size)
2575 {
2576         char *end = str + size;
2577         int retval;
2578
2579         retval = 0;
2580
2581         for (; *str && str < end; str++) {
2582                 switch (*str) {
2583                 case '&':
2584                         retval = sbuf_printf(sb, "&amp;");
2585                         break;
2586                 case '>':
2587                         retval = sbuf_printf(sb, "&gt;");
2588                         break;
2589                 case '<':
2590                         retval = sbuf_printf(sb, "&lt;");
2591                         break;
2592                 default:
2593                         retval = sbuf_putc(sb, *str);
2594                         break;
2595                 }
2596
2597                 if (retval != 0)
2598                         break;
2599
2600         }
2601
2602         return (retval);
2603 }
2604
2605 static void
2606 ctl_id_sbuf(struct ctl_devid *id, struct sbuf *sb)
2607 {
2608         struct scsi_vpd_id_descriptor *desc;
2609         int i;
2610
2611         if (id == NULL || id->len < 4)
2612                 return;
2613         desc = (struct scsi_vpd_id_descriptor *)id->data;
2614         switch (desc->id_type & SVPD_ID_TYPE_MASK) {
2615         case SVPD_ID_TYPE_T10:
2616                 sbuf_printf(sb, "t10.");
2617                 break;
2618         case SVPD_ID_TYPE_EUI64:
2619                 sbuf_printf(sb, "eui.");
2620                 break;
2621         case SVPD_ID_TYPE_NAA:
2622                 sbuf_printf(sb, "naa.");
2623                 break;
2624         case SVPD_ID_TYPE_SCSI_NAME:
2625                 break;
2626         }
2627         switch (desc->proto_codeset & SVPD_ID_CODESET_MASK) {
2628         case SVPD_ID_CODESET_BINARY:
2629                 for (i = 0; i < desc->length; i++)
2630                         sbuf_printf(sb, "%02x", desc->identifier[i]);
2631                 break;
2632         case SVPD_ID_CODESET_ASCII:
2633                 sbuf_printf(sb, "%.*s", (int)desc->length,
2634                     (char *)desc->identifier);
2635                 break;
2636         case SVPD_ID_CODESET_UTF8:
2637                 sbuf_printf(sb, "%s", (char *)desc->identifier);
2638                 break;
2639         }
2640 }
2641
2642 static int
2643 ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
2644           struct thread *td)
2645 {
2646         struct ctl_softc *softc = dev->si_drv1;
2647         struct ctl_port *port;
2648         struct ctl_lun *lun;
2649         int retval;
2650
2651         retval = 0;
2652
2653         switch (cmd) {
2654         case CTL_IO:
2655                 retval = ctl_ioctl_io(dev, cmd, addr, flag, td);
2656                 break;
2657         case CTL_ENABLE_PORT:
2658         case CTL_DISABLE_PORT:
2659         case CTL_SET_PORT_WWNS: {
2660                 struct ctl_port *port;
2661                 struct ctl_port_entry *entry;
2662
2663                 entry = (struct ctl_port_entry *)addr;
2664
2665                 mtx_lock(&softc->ctl_lock);
2666                 STAILQ_FOREACH(port, &softc->port_list, links) {
2667                         int action, done;
2668
2669                         if (port->targ_port < softc->port_min ||
2670                             port->targ_port >= softc->port_max)
2671                                 continue;
2672
2673                         action = 0;
2674                         done = 0;
2675                         if ((entry->port_type == CTL_PORT_NONE)
2676                          && (entry->targ_port == port->targ_port)) {
2677                                 /*
2678                                  * If the user only wants to enable or
2679                                  * disable or set WWNs on a specific port,
2680                                  * do the operation and we're done.
2681                                  */
2682                                 action = 1;
2683                                 done = 1;
2684                         } else if (entry->port_type & port->port_type) {
2685                                 /*
2686                                  * Compare the user's type mask with the
2687                                  * particular frontend type to see if we
2688                                  * have a match.
2689                                  */
2690                                 action = 1;
2691                                 done = 0;
2692
2693                                 /*
2694                                  * Make sure the user isn't trying to set
2695                                  * WWNs on multiple ports at the same time.
2696                                  */
2697                                 if (cmd == CTL_SET_PORT_WWNS) {
2698                                         printf("%s: Can't set WWNs on "
2699                                                "multiple ports\n", __func__);
2700                                         retval = EINVAL;
2701                                         break;
2702                                 }
2703                         }
2704                         if (action == 0)
2705                                 continue;
2706
2707                         /*
2708                          * XXX KDM we have to drop the lock here, because
2709                          * the online/offline operations can potentially
2710                          * block.  We need to reference count the frontends
2711                          * so they can't go away,
2712                          */
2713                         if (cmd == CTL_ENABLE_PORT) {
2714                                 mtx_unlock(&softc->ctl_lock);
2715                                 ctl_port_online(port);
2716                                 mtx_lock(&softc->ctl_lock);
2717                         } else if (cmd == CTL_DISABLE_PORT) {
2718                                 mtx_unlock(&softc->ctl_lock);
2719                                 ctl_port_offline(port);
2720                                 mtx_lock(&softc->ctl_lock);
2721                         } else if (cmd == CTL_SET_PORT_WWNS) {
2722                                 ctl_port_set_wwns(port,
2723                                     (entry->flags & CTL_PORT_WWNN_VALID) ?
2724                                     1 : 0, entry->wwnn,
2725                                     (entry->flags & CTL_PORT_WWPN_VALID) ?
2726                                     1 : 0, entry->wwpn);
2727                         }
2728                         if (done != 0)
2729                                 break;
2730                 }
2731                 mtx_unlock(&softc->ctl_lock);
2732                 break;
2733         }
2734         case CTL_GET_OOA: {
2735                 struct ctl_ooa *ooa_hdr;
2736                 struct ctl_ooa_entry *entries;
2737                 uint32_t cur_fill_num;
2738
2739                 ooa_hdr = (struct ctl_ooa *)addr;
2740
2741                 if ((ooa_hdr->alloc_len == 0)
2742                  || (ooa_hdr->alloc_num == 0)) {
2743                         printf("%s: CTL_GET_OOA: alloc len %u and alloc num %u "
2744                                "must be non-zero\n", __func__,
2745                                ooa_hdr->alloc_len, ooa_hdr->alloc_num);
2746                         retval = EINVAL;
2747                         break;
2748                 }
2749
2750                 if (ooa_hdr->alloc_len != (ooa_hdr->alloc_num *
2751                     sizeof(struct ctl_ooa_entry))) {
2752                         printf("%s: CTL_GET_OOA: alloc len %u must be alloc "
2753                                "num %d * sizeof(struct ctl_ooa_entry) %zd\n",
2754                                __func__, ooa_hdr->alloc_len,
2755                                ooa_hdr->alloc_num,sizeof(struct ctl_ooa_entry));
2756                         retval = EINVAL;
2757                         break;
2758                 }
2759
2760                 entries = malloc(ooa_hdr->alloc_len, M_CTL, M_WAITOK | M_ZERO);
2761                 if (entries == NULL) {
2762                         printf("%s: could not allocate %d bytes for OOA "
2763                                "dump\n", __func__, ooa_hdr->alloc_len);
2764                         retval = ENOMEM;
2765                         break;
2766                 }
2767
2768                 mtx_lock(&softc->ctl_lock);
2769                 if ((ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) == 0 &&
2770                     (ooa_hdr->lun_num >= ctl_max_luns ||
2771                      softc->ctl_luns[ooa_hdr->lun_num] == NULL)) {
2772                         mtx_unlock(&softc->ctl_lock);
2773                         free(entries, M_CTL);
2774                         printf("%s: CTL_GET_OOA: invalid LUN %ju\n",
2775                                __func__, (uintmax_t)ooa_hdr->lun_num);
2776                         retval = EINVAL;
2777                         break;
2778                 }
2779
2780                 cur_fill_num = 0;
2781
2782                 if (ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) {
2783                         STAILQ_FOREACH(lun, &softc->lun_list, links) {
2784                                 ctl_ioctl_fill_ooa(lun, &cur_fill_num,
2785                                     ooa_hdr, entries);
2786                         }
2787                 } else {
2788                         lun = softc->ctl_luns[ooa_hdr->lun_num];
2789                         ctl_ioctl_fill_ooa(lun, &cur_fill_num, ooa_hdr,
2790                             entries);
2791                 }
2792                 mtx_unlock(&softc->ctl_lock);
2793
2794                 ooa_hdr->fill_num = min(cur_fill_num, ooa_hdr->alloc_num);
2795                 ooa_hdr->fill_len = ooa_hdr->fill_num *
2796                         sizeof(struct ctl_ooa_entry);
2797                 retval = copyout(entries, ooa_hdr->entries, ooa_hdr->fill_len);
2798                 if (retval != 0) {
2799                         printf("%s: error copying out %d bytes for OOA dump\n",
2800                                __func__, ooa_hdr->fill_len);
2801                 }
2802
2803                 getbinuptime(&ooa_hdr->cur_bt);
2804
2805                 if (cur_fill_num > ooa_hdr->alloc_num) {
2806                         ooa_hdr->dropped_num = cur_fill_num -ooa_hdr->alloc_num;
2807                         ooa_hdr->status = CTL_OOA_NEED_MORE_SPACE;
2808                 } else {
2809                         ooa_hdr->dropped_num = 0;
2810                         ooa_hdr->status = CTL_OOA_OK;
2811                 }
2812
2813                 free(entries, M_CTL);
2814                 break;
2815         }
2816         case CTL_DELAY_IO: {
2817                 struct ctl_io_delay_info *delay_info;
2818
2819                 delay_info = (struct ctl_io_delay_info *)addr;
2820
2821 #ifdef CTL_IO_DELAY
2822                 mtx_lock(&softc->ctl_lock);
2823                 if (delay_info->lun_id >= ctl_max_luns ||
2824                     (lun = softc->ctl_luns[delay_info->lun_id]) == NULL) {
2825                         mtx_unlock(&softc->ctl_lock);
2826                         delay_info->status = CTL_DELAY_STATUS_INVALID_LUN;
2827                         break;
2828                 }
2829                 mtx_lock(&lun->lun_lock);
2830                 mtx_unlock(&softc->ctl_lock);
2831                 delay_info->status = CTL_DELAY_STATUS_OK;
2832                 switch (delay_info->delay_type) {
2833                 case CTL_DELAY_TYPE_CONT:
2834                 case CTL_DELAY_TYPE_ONESHOT:
2835                         break;
2836                 default:
2837                         delay_info->status = CTL_DELAY_STATUS_INVALID_TYPE;
2838                         break;
2839                 }
2840                 switch (delay_info->delay_loc) {
2841                 case CTL_DELAY_LOC_DATAMOVE:
2842                         lun->delay_info.datamove_type = delay_info->delay_type;
2843                         lun->delay_info.datamove_delay = delay_info->delay_secs;
2844                         break;
2845                 case CTL_DELAY_LOC_DONE:
2846                         lun->delay_info.done_type = delay_info->delay_type;
2847                         lun->delay_info.done_delay = delay_info->delay_secs;
2848                         break;
2849                 default:
2850                         delay_info->status = CTL_DELAY_STATUS_INVALID_LOC;
2851                         break;
2852                 }
2853                 mtx_unlock(&lun->lun_lock);
2854 #else
2855                 delay_info->status = CTL_DELAY_STATUS_NOT_IMPLEMENTED;
2856 #endif /* CTL_IO_DELAY */
2857                 break;
2858         }
2859 #ifdef CTL_LEGACY_STATS
2860         case CTL_GETSTATS: {
2861                 struct ctl_stats *stats = (struct ctl_stats *)addr;
2862                 int i;
2863
2864                 /*
2865                  * XXX KDM no locking here.  If the LUN list changes,
2866                  * things can blow up.
2867                  */
2868                 i = 0;
2869                 stats->status = CTL_SS_OK;
2870                 stats->fill_len = 0;
2871                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
2872                         if (stats->fill_len + sizeof(lun->legacy_stats) >
2873                             stats->alloc_len) {
2874                                 stats->status = CTL_SS_NEED_MORE_SPACE;
2875                                 break;
2876                         }
2877                         retval = copyout(&lun->legacy_stats, &stats->lun_stats[i++],
2878                                          sizeof(lun->legacy_stats));
2879                         if (retval != 0)
2880                                 break;
2881                         stats->fill_len += sizeof(lun->legacy_stats);
2882                 }
2883                 stats->num_luns = softc->num_luns;
2884                 stats->flags = CTL_STATS_FLAG_NONE;
2885 #ifdef CTL_TIME_IO
2886                 stats->flags |= CTL_STATS_FLAG_TIME_VALID;
2887 #endif
2888                 getnanouptime(&stats->timestamp);
2889                 break;
2890         }
2891 #endif /* CTL_LEGACY_STATS */
2892         case CTL_ERROR_INJECT: {
2893                 struct ctl_error_desc *err_desc, *new_err_desc;
2894
2895                 err_desc = (struct ctl_error_desc *)addr;
2896
2897                 new_err_desc = malloc(sizeof(*new_err_desc), M_CTL,
2898                                       M_WAITOK | M_ZERO);
2899                 bcopy(err_desc, new_err_desc, sizeof(*new_err_desc));
2900
2901                 mtx_lock(&softc->ctl_lock);
2902                 if (err_desc->lun_id >= ctl_max_luns ||
2903                     (lun = softc->ctl_luns[err_desc->lun_id]) == NULL) {
2904                         mtx_unlock(&softc->ctl_lock);
2905                         free(new_err_desc, M_CTL);
2906                         printf("%s: CTL_ERROR_INJECT: invalid LUN %ju\n",
2907                                __func__, (uintmax_t)err_desc->lun_id);
2908                         retval = EINVAL;
2909                         break;
2910                 }
2911                 mtx_lock(&lun->lun_lock);
2912                 mtx_unlock(&softc->ctl_lock);
2913
2914                 /*
2915                  * We could do some checking here to verify the validity
2916                  * of the request, but given the complexity of error
2917                  * injection requests, the checking logic would be fairly
2918                  * complex.
2919                  *
2920                  * For now, if the request is invalid, it just won't get
2921                  * executed and might get deleted.
2922                  */
2923                 STAILQ_INSERT_TAIL(&lun->error_list, new_err_desc, links);
2924
2925                 /*
2926                  * XXX KDM check to make sure the serial number is unique,
2927                  * in case we somehow manage to wrap.  That shouldn't
2928                  * happen for a very long time, but it's the right thing to
2929                  * do.
2930                  */
2931                 new_err_desc->serial = lun->error_serial;
2932                 err_desc->serial = lun->error_serial;
2933                 lun->error_serial++;
2934
2935                 mtx_unlock(&lun->lun_lock);
2936                 break;
2937         }
2938         case CTL_ERROR_INJECT_DELETE: {
2939                 struct ctl_error_desc *delete_desc, *desc, *desc2;
2940                 int delete_done;
2941
2942                 delete_desc = (struct ctl_error_desc *)addr;
2943                 delete_done = 0;
2944
2945                 mtx_lock(&softc->ctl_lock);
2946                 if (delete_desc->lun_id >= ctl_max_luns ||
2947                     (lun = softc->ctl_luns[delete_desc->lun_id]) == NULL) {
2948                         mtx_unlock(&softc->ctl_lock);
2949                         printf("%s: CTL_ERROR_INJECT_DELETE: invalid LUN %ju\n",
2950                                __func__, (uintmax_t)delete_desc->lun_id);
2951                         retval = EINVAL;
2952                         break;
2953                 }
2954                 mtx_lock(&lun->lun_lock);
2955                 mtx_unlock(&softc->ctl_lock);
2956                 STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
2957                         if (desc->serial != delete_desc->serial)
2958                                 continue;
2959
2960                         STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc,
2961                                       links);
2962                         free(desc, M_CTL);
2963                         delete_done = 1;
2964                 }
2965                 mtx_unlock(&lun->lun_lock);
2966                 if (delete_done == 0) {
2967                         printf("%s: CTL_ERROR_INJECT_DELETE: can't find "
2968                                "error serial %ju on LUN %u\n", __func__,
2969                                delete_desc->serial, delete_desc->lun_id);
2970                         retval = EINVAL;
2971                         break;
2972                 }
2973                 break;
2974         }
2975         case CTL_DUMP_STRUCTS: {
2976                 int j, k;
2977                 struct ctl_port *port;
2978                 struct ctl_frontend *fe;
2979
2980                 mtx_lock(&softc->ctl_lock);
2981                 printf("CTL Persistent Reservation information start:\n");
2982                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
2983                         mtx_lock(&lun->lun_lock);
2984                         if ((lun->flags & CTL_LUN_DISABLED) != 0) {
2985                                 mtx_unlock(&lun->lun_lock);
2986                                 continue;
2987                         }
2988
2989                         for (j = 0; j < ctl_max_ports; j++) {
2990                                 if (lun->pr_keys[j] == NULL)
2991                                         continue;
2992                                 for (k = 0; k < CTL_MAX_INIT_PER_PORT; k++){
2993                                         if (lun->pr_keys[j][k] == 0)
2994                                                 continue;
2995                                         printf("  LUN %ju port %d iid %d key "
2996                                                "%#jx\n", lun->lun, j, k,
2997                                                (uintmax_t)lun->pr_keys[j][k]);
2998                                 }
2999                         }
3000                         mtx_unlock(&lun->lun_lock);
3001                 }
3002                 printf("CTL Persistent Reservation information end\n");
3003                 printf("CTL Ports:\n");
3004                 STAILQ_FOREACH(port, &softc->port_list, links) {
3005                         printf("  Port %d '%s' Frontend '%s' Type %u pp %d vp %d WWNN "
3006                                "%#jx WWPN %#jx\n", port->targ_port, port->port_name,
3007                                port->frontend->name, port->port_type,
3008                                port->physical_port, port->virtual_port,
3009                                (uintmax_t)port->wwnn, (uintmax_t)port->wwpn);
3010                         for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
3011                                 if (port->wwpn_iid[j].in_use == 0 &&
3012                                     port->wwpn_iid[j].wwpn == 0 &&
3013                                     port->wwpn_iid[j].name == NULL)
3014                                         continue;
3015
3016                                 printf("    iid %u use %d WWPN %#jx '%s'\n",
3017                                     j, port->wwpn_iid[j].in_use,
3018                                     (uintmax_t)port->wwpn_iid[j].wwpn,
3019                                     port->wwpn_iid[j].name);
3020                         }
3021                 }
3022                 printf("CTL Port information end\n");
3023                 mtx_unlock(&softc->ctl_lock);
3024                 /*
3025                  * XXX KDM calling this without a lock.  We'd likely want
3026                  * to drop the lock before calling the frontend's dump
3027                  * routine anyway.
3028                  */
3029                 printf("CTL Frontends:\n");
3030                 STAILQ_FOREACH(fe, &softc->fe_list, links) {
3031                         printf("  Frontend '%s'\n", fe->name);
3032                         if (fe->fe_dump != NULL)
3033                                 fe->fe_dump();
3034                 }
3035                 printf("CTL Frontend information end\n");
3036                 break;
3037         }
3038         case CTL_LUN_REQ: {
3039                 struct ctl_lun_req *lun_req;
3040                 struct ctl_backend_driver *backend;
3041
3042                 lun_req = (struct ctl_lun_req *)addr;
3043
3044                 backend = ctl_backend_find(lun_req->backend);
3045                 if (backend == NULL) {
3046                         lun_req->status = CTL_LUN_ERROR;
3047                         snprintf(lun_req->error_str,
3048                                  sizeof(lun_req->error_str),
3049                                  "Backend \"%s\" not found.",
3050                                  lun_req->backend);
3051                         break;
3052                 }
3053                 if (lun_req->num_be_args > 0) {
3054                         lun_req->kern_be_args = ctl_copyin_args(
3055                                 lun_req->num_be_args,
3056                                 lun_req->be_args,
3057                                 lun_req->error_str,
3058                                 sizeof(lun_req->error_str));
3059                         if (lun_req->kern_be_args == NULL) {
3060                                 lun_req->status = CTL_LUN_ERROR;
3061                                 break;
3062                         }
3063                 }
3064
3065                 retval = backend->ioctl(dev, cmd, addr, flag, td);
3066
3067                 if (lun_req->num_be_args > 0) {
3068                         ctl_copyout_args(lun_req->num_be_args,
3069                                       lun_req->kern_be_args);
3070                         ctl_free_args(lun_req->num_be_args,
3071                                       lun_req->kern_be_args);
3072                 }
3073                 break;
3074         }
3075         case CTL_LUN_LIST: {
3076                 struct sbuf *sb;
3077                 struct ctl_lun_list *list;
3078                 struct ctl_option *opt;
3079
3080                 list = (struct ctl_lun_list *)addr;
3081
3082                 /*
3083                  * Allocate a fixed length sbuf here, based on the length
3084                  * of the user's buffer.  We could allocate an auto-extending
3085                  * buffer, and then tell the user how much larger our
3086                  * amount of data is than his buffer, but that presents
3087                  * some problems:
3088                  *
3089                  * 1.  The sbuf(9) routines use a blocking malloc, and so
3090                  *     we can't hold a lock while calling them with an
3091                  *     auto-extending buffer.
3092                  *
3093                  * 2.  There is not currently a LUN reference counting
3094                  *     mechanism, outside of outstanding transactions on
3095                  *     the LUN's OOA queue.  So a LUN could go away on us
3096                  *     while we're getting the LUN number, backend-specific
3097                  *     information, etc.  Thus, given the way things
3098                  *     currently work, we need to hold the CTL lock while
3099                  *     grabbing LUN information.
3100                  *
3101                  * So, from the user's standpoint, the best thing to do is
3102                  * allocate what he thinks is a reasonable buffer length,
3103                  * and then if he gets a CTL_LUN_LIST_NEED_MORE_SPACE error,
3104                  * double the buffer length and try again.  (And repeat
3105                  * that until he succeeds.)
3106                  */
3107                 sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3108                 if (sb == NULL) {
3109                         list->status = CTL_LUN_LIST_ERROR;
3110                         snprintf(list->error_str, sizeof(list->error_str),
3111                                  "Unable to allocate %d bytes for LUN list",
3112                                  list->alloc_len);
3113                         break;
3114                 }
3115
3116                 sbuf_printf(sb, "<ctllunlist>\n");
3117
3118                 mtx_lock(&softc->ctl_lock);
3119                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
3120                         mtx_lock(&lun->lun_lock);
3121                         retval = sbuf_printf(sb, "<lun id=\"%ju\">\n",
3122                                              (uintmax_t)lun->lun);
3123
3124                         /*
3125                          * Bail out as soon as we see that we've overfilled
3126                          * the buffer.
3127                          */
3128                         if (retval != 0)
3129                                 break;
3130
3131                         retval = sbuf_printf(sb, "\t<backend_type>%s"
3132                                              "</backend_type>\n",
3133                                              (lun->backend == NULL) ?  "none" :
3134                                              lun->backend->name);
3135
3136                         if (retval != 0)
3137                                 break;
3138
3139                         retval = sbuf_printf(sb, "\t<lun_type>%d</lun_type>\n",
3140                                              lun->be_lun->lun_type);
3141
3142                         if (retval != 0)
3143                                 break;
3144
3145                         if (lun->backend == NULL) {
3146                                 retval = sbuf_printf(sb, "</lun>\n");
3147                                 if (retval != 0)
3148                                         break;
3149                                 continue;
3150                         }
3151
3152                         retval = sbuf_printf(sb, "\t<size>%ju</size>\n",
3153                                              (lun->be_lun->maxlba > 0) ?
3154                                              lun->be_lun->maxlba + 1 : 0);
3155
3156                         if (retval != 0)
3157                                 break;
3158
3159                         retval = sbuf_printf(sb, "\t<blocksize>%u</blocksize>\n",
3160                                              lun->be_lun->blocksize);
3161
3162                         if (retval != 0)
3163                                 break;
3164
3165                         retval = sbuf_printf(sb, "\t<serial_number>");
3166
3167                         if (retval != 0)
3168                                 break;
3169
3170                         retval = ctl_sbuf_printf_esc(sb,
3171                             lun->be_lun->serial_num,
3172                             sizeof(lun->be_lun->serial_num));
3173
3174                         if (retval != 0)
3175                                 break;
3176
3177                         retval = sbuf_printf(sb, "</serial_number>\n");
3178
3179                         if (retval != 0)
3180                                 break;
3181
3182                         retval = sbuf_printf(sb, "\t<device_id>");
3183
3184                         if (retval != 0)
3185                                 break;
3186
3187                         retval = ctl_sbuf_printf_esc(sb,
3188                             lun->be_lun->device_id,
3189                             sizeof(lun->be_lun->device_id));
3190
3191                         if (retval != 0)
3192                                 break;
3193
3194                         retval = sbuf_printf(sb, "</device_id>\n");
3195
3196                         if (retval != 0)
3197                                 break;
3198
3199                         if (lun->backend->lun_info != NULL) {
3200                                 retval = lun->backend->lun_info(lun->be_lun->be_lun, sb);
3201                                 if (retval != 0)
3202                                         break;
3203                         }
3204                         STAILQ_FOREACH(opt, &lun->be_lun->options, links) {
3205                                 retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3206                                     opt->name, opt->value, opt->name);
3207                                 if (retval != 0)
3208                                         break;
3209                         }
3210
3211                         retval = sbuf_printf(sb, "</lun>\n");
3212
3213                         if (retval != 0)
3214                                 break;
3215                         mtx_unlock(&lun->lun_lock);
3216                 }
3217                 if (lun != NULL)
3218                         mtx_unlock(&lun->lun_lock);
3219                 mtx_unlock(&softc->ctl_lock);
3220
3221                 if ((retval != 0)
3222                  || ((retval = sbuf_printf(sb, "</ctllunlist>\n")) != 0)) {
3223                         retval = 0;
3224                         sbuf_delete(sb);
3225                         list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3226                         snprintf(list->error_str, sizeof(list->error_str),
3227                                  "Out of space, %d bytes is too small",
3228                                  list->alloc_len);
3229                         break;
3230                 }
3231
3232                 sbuf_finish(sb);
3233
3234                 retval = copyout(sbuf_data(sb), list->lun_xml,
3235                                  sbuf_len(sb) + 1);
3236
3237                 list->fill_len = sbuf_len(sb) + 1;
3238                 list->status = CTL_LUN_LIST_OK;
3239                 sbuf_delete(sb);
3240                 break;
3241         }
3242         case CTL_ISCSI: {
3243                 struct ctl_iscsi *ci;
3244                 struct ctl_frontend *fe;
3245
3246                 ci = (struct ctl_iscsi *)addr;
3247
3248                 fe = ctl_frontend_find("iscsi");
3249                 if (fe == NULL) {
3250                         ci->status = CTL_ISCSI_ERROR;
3251                         snprintf(ci->error_str, sizeof(ci->error_str),
3252                             "Frontend \"iscsi\" not found.");
3253                         break;
3254                 }
3255
3256                 retval = fe->ioctl(dev, cmd, addr, flag, td);
3257                 break;
3258         }
3259         case CTL_PORT_REQ: {
3260                 struct ctl_req *req;
3261                 struct ctl_frontend *fe;
3262
3263                 req = (struct ctl_req *)addr;
3264
3265                 fe = ctl_frontend_find(req->driver);
3266                 if (fe == NULL) {
3267                         req->status = CTL_LUN_ERROR;
3268                         snprintf(req->error_str, sizeof(req->error_str),
3269                             "Frontend \"%s\" not found.", req->driver);
3270                         break;
3271                 }
3272                 if (req->num_args > 0) {
3273                         req->kern_args = ctl_copyin_args(req->num_args,
3274                             req->args, req->error_str, sizeof(req->error_str));
3275                         if (req->kern_args == NULL) {
3276                                 req->status = CTL_LUN_ERROR;
3277                                 break;
3278                         }
3279                 }
3280
3281                 if (fe->ioctl)
3282                         retval = fe->ioctl(dev, cmd, addr, flag, td);
3283                 else
3284                         retval = ENODEV;
3285
3286                 if (req->num_args > 0) {
3287                         ctl_copyout_args(req->num_args, req->kern_args);
3288                         ctl_free_args(req->num_args, req->kern_args);
3289                 }
3290                 break;
3291         }
3292         case CTL_PORT_LIST: {
3293                 struct sbuf *sb;
3294                 struct ctl_port *port;
3295                 struct ctl_lun_list *list;
3296                 struct ctl_option *opt;
3297                 int j;
3298                 uint32_t plun;
3299
3300                 list = (struct ctl_lun_list *)addr;
3301
3302                 sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3303                 if (sb == NULL) {
3304                         list->status = CTL_LUN_LIST_ERROR;
3305                         snprintf(list->error_str, sizeof(list->error_str),
3306                                  "Unable to allocate %d bytes for LUN list",
3307                                  list->alloc_len);
3308                         break;
3309                 }
3310
3311                 sbuf_printf(sb, "<ctlportlist>\n");
3312
3313                 mtx_lock(&softc->ctl_lock);
3314                 STAILQ_FOREACH(port, &softc->port_list, links) {
3315                         retval = sbuf_printf(sb, "<targ_port id=\"%ju\">\n",
3316                                              (uintmax_t)port->targ_port);
3317
3318                         /*
3319                          * Bail out as soon as we see that we've overfilled
3320                          * the buffer.
3321                          */
3322                         if (retval != 0)
3323                                 break;
3324
3325                         retval = sbuf_printf(sb, "\t<frontend_type>%s"
3326                             "</frontend_type>\n", port->frontend->name);
3327                         if (retval != 0)
3328                                 break;
3329
3330                         retval = sbuf_printf(sb, "\t<port_type>%d</port_type>\n",
3331                                              port->port_type);
3332                         if (retval != 0)
3333                                 break;
3334
3335                         retval = sbuf_printf(sb, "\t<online>%s</online>\n",
3336                             (port->status & CTL_PORT_STATUS_ONLINE) ? "YES" : "NO");
3337                         if (retval != 0)
3338                                 break;
3339
3340                         retval = sbuf_printf(sb, "\t<port_name>%s</port_name>\n",
3341                             port->port_name);
3342                         if (retval != 0)
3343                                 break;
3344
3345                         retval = sbuf_printf(sb, "\t<physical_port>%d</physical_port>\n",
3346                             port->physical_port);
3347                         if (retval != 0)
3348                                 break;
3349
3350                         retval = sbuf_printf(sb, "\t<virtual_port>%d</virtual_port>\n",
3351                             port->virtual_port);
3352                         if (retval != 0)
3353                                 break;
3354
3355                         if (port->target_devid != NULL) {
3356                                 sbuf_printf(sb, "\t<target>");
3357                                 ctl_id_sbuf(port->target_devid, sb);
3358                                 sbuf_printf(sb, "</target>\n");
3359                         }
3360
3361                         if (port->port_devid != NULL) {
3362                                 sbuf_printf(sb, "\t<port>");
3363                                 ctl_id_sbuf(port->port_devid, sb);
3364                                 sbuf_printf(sb, "</port>\n");
3365                         }
3366
3367                         if (port->port_info != NULL) {
3368                                 retval = port->port_info(port->onoff_arg, sb);
3369                                 if (retval != 0)
3370                                         break;
3371                         }
3372                         STAILQ_FOREACH(opt, &port->options, links) {
3373                                 retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3374                                     opt->name, opt->value, opt->name);
3375                                 if (retval != 0)
3376                                         break;
3377                         }
3378
3379                         if (port->lun_map != NULL) {
3380                                 sbuf_printf(sb, "\t<lun_map>on</lun_map>\n");
3381                                 for (j = 0; j < port->lun_map_size; j++) {
3382                                         plun = ctl_lun_map_from_port(port, j);
3383                                         if (plun == UINT32_MAX)
3384                                                 continue;
3385                                         sbuf_printf(sb,
3386                                             "\t<lun id=\"%u\">%u</lun>\n",
3387                                             j, plun);
3388                                 }
3389                         }
3390
3391                         for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
3392                                 if (port->wwpn_iid[j].in_use == 0 ||
3393                                     (port->wwpn_iid[j].wwpn == 0 &&
3394                                      port->wwpn_iid[j].name == NULL))
3395                                         continue;
3396
3397                                 if (port->wwpn_iid[j].name != NULL)
3398                                         retval = sbuf_printf(sb,
3399                                             "\t<initiator id=\"%u\">%s</initiator>\n",
3400                                             j, port->wwpn_iid[j].name);
3401                                 else
3402                                         retval = sbuf_printf(sb,
3403                                             "\t<initiator id=\"%u\">naa.%08jx</initiator>\n",
3404                                             j, port->wwpn_iid[j].wwpn);
3405                                 if (retval != 0)
3406                                         break;
3407                         }
3408                         if (retval != 0)
3409                                 break;
3410
3411                         retval = sbuf_printf(sb, "</targ_port>\n");
3412                         if (retval != 0)
3413                                 break;
3414                 }
3415                 mtx_unlock(&softc->ctl_lock);
3416
3417                 if ((retval != 0)
3418                  || ((retval = sbuf_printf(sb, "</ctlportlist>\n")) != 0)) {
3419                         retval = 0;
3420                         sbuf_delete(sb);
3421                         list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3422                         snprintf(list->error_str, sizeof(list->error_str),
3423                                  "Out of space, %d bytes is too small",
3424                                  list->alloc_len);
3425                         break;
3426                 }
3427
3428                 sbuf_finish(sb);
3429
3430                 retval = copyout(sbuf_data(sb), list->lun_xml,
3431                                  sbuf_len(sb) + 1);
3432
3433                 list->fill_len = sbuf_len(sb) + 1;
3434                 list->status = CTL_LUN_LIST_OK;
3435                 sbuf_delete(sb);
3436                 break;
3437         }
3438         case CTL_LUN_MAP: {
3439                 struct ctl_lun_map *lm  = (struct ctl_lun_map *)addr;
3440                 struct ctl_port *port;
3441
3442                 mtx_lock(&softc->ctl_lock);
3443                 if (lm->port < softc->port_min ||
3444                     lm->port >= softc->port_max ||
3445                     (port = softc->ctl_ports[lm->port]) == NULL) {
3446                         mtx_unlock(&softc->ctl_lock);
3447                         return (ENXIO);
3448                 }
3449                 if (port->status & CTL_PORT_STATUS_ONLINE) {
3450                         STAILQ_FOREACH(lun, &softc->lun_list, links) {
3451                                 if (ctl_lun_map_to_port(port, lun->lun) ==
3452                                     UINT32_MAX)
3453                                         continue;
3454                                 mtx_lock(&lun->lun_lock);
3455                                 ctl_est_ua_port(lun, lm->port, -1,
3456                                     CTL_UA_LUN_CHANGE);
3457                                 mtx_unlock(&lun->lun_lock);
3458                         }
3459                 }
3460                 mtx_unlock(&softc->ctl_lock); // XXX: port_enable sleeps
3461                 if (lm->plun != UINT32_MAX) {
3462                         if (lm->lun == UINT32_MAX)
3463                                 retval = ctl_lun_map_unset(port, lm->plun);
3464                         else if (lm->lun < ctl_max_luns &&
3465                             softc->ctl_luns[lm->lun] != NULL)
3466                                 retval = ctl_lun_map_set(port, lm->plun, lm->lun);
3467                         else
3468                                 return (ENXIO);
3469                 } else {
3470                         if (lm->lun == UINT32_MAX)
3471                                 retval = ctl_lun_map_deinit(port);
3472                         else
3473                                 retval = ctl_lun_map_init(port);
3474                 }
3475                 if (port->status & CTL_PORT_STATUS_ONLINE)
3476                         ctl_isc_announce_port(port);
3477                 break;
3478         }
3479         case CTL_GET_LUN_STATS: {
3480                 struct ctl_get_io_stats *stats = (struct ctl_get_io_stats *)addr;
3481                 int i;
3482
3483                 /*
3484                  * XXX KDM no locking here.  If the LUN list changes,
3485                  * things can blow up.
3486                  */
3487                 i = 0;
3488                 stats->status = CTL_SS_OK;
3489                 stats->fill_len = 0;
3490                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
3491                         if (lun->lun < stats->first_item)
3492                                 continue;
3493                         if (stats->fill_len + sizeof(lun->stats) >
3494                             stats->alloc_len) {
3495                                 stats->status = CTL_SS_NEED_MORE_SPACE;
3496                                 break;
3497                         }
3498                         retval = copyout(&lun->stats, &stats->stats[i++],
3499                                          sizeof(lun->stats));
3500                         if (retval != 0)
3501                                 break;
3502                         stats->fill_len += sizeof(lun->stats);
3503                 }
3504                 stats->num_items = softc->num_luns;
3505                 stats->flags = CTL_STATS_FLAG_NONE;
3506 #ifdef CTL_TIME_IO
3507                 stats->flags |= CTL_STATS_FLAG_TIME_VALID;
3508 #endif
3509                 getnanouptime(&stats->timestamp);
3510                 break;
3511         }
3512         case CTL_GET_PORT_STATS: {
3513                 struct ctl_get_io_stats *stats = (struct ctl_get_io_stats *)addr;
3514                 int i;
3515
3516                 /*
3517                  * XXX KDM no locking here.  If the LUN list changes,
3518                  * things can blow up.
3519                  */
3520                 i = 0;
3521                 stats->status = CTL_SS_OK;
3522                 stats->fill_len = 0;
3523                 STAILQ_FOREACH(port, &softc->port_list, links) {
3524                         if (port->targ_port < stats->first_item)
3525                                 continue;
3526                         if (stats->fill_len + sizeof(port->stats) >
3527                             stats->alloc_len) {
3528                                 stats->status = CTL_SS_NEED_MORE_SPACE;
3529                                 break;
3530                         }
3531                         retval = copyout(&port->stats, &stats->stats[i++],
3532                                          sizeof(port->stats));
3533                         if (retval != 0)
3534                                 break;
3535                         stats->fill_len += sizeof(port->stats);
3536                 }
3537                 stats->num_items = softc->num_ports;
3538                 stats->flags = CTL_STATS_FLAG_NONE;
3539 #ifdef CTL_TIME_IO
3540                 stats->flags |= CTL_STATS_FLAG_TIME_VALID;
3541 #endif
3542                 getnanouptime(&stats->timestamp);
3543                 break;
3544         }
3545         default: {
3546                 /* XXX KDM should we fix this? */
3547 #if 0
3548                 struct ctl_backend_driver *backend;
3549                 unsigned int type;
3550                 int found;
3551
3552                 found = 0;
3553
3554                 /*
3555                  * We encode the backend type as the ioctl type for backend
3556                  * ioctls.  So parse it out here, and then search for a
3557                  * backend of this type.
3558                  */
3559                 type = _IOC_TYPE(cmd);
3560
3561                 STAILQ_FOREACH(backend, &softc->be_list, links) {
3562                         if (backend->type == type) {
3563                                 found = 1;
3564                                 break;
3565                         }
3566                 }
3567                 if (found == 0) {
3568                         printf("ctl: unknown ioctl command %#lx or backend "
3569                                "%d\n", cmd, type);
3570                         retval = EINVAL;
3571                         break;
3572                 }
3573                 retval = backend->ioctl(dev, cmd, addr, flag, td);
3574 #endif
3575                 retval = ENOTTY;
3576                 break;
3577         }
3578         }
3579         return (retval);
3580 }
3581
3582 uint32_t
3583 ctl_get_initindex(struct ctl_nexus *nexus)
3584 {
3585         return (nexus->initid + (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3586 }
3587
3588 int
3589 ctl_lun_map_init(struct ctl_port *port)
3590 {
3591         struct ctl_softc *softc = port->ctl_softc;
3592         struct ctl_lun *lun;
3593         int size = ctl_lun_map_size;
3594         uint32_t i;
3595
3596         if (port->lun_map == NULL || port->lun_map_size < size) {
3597                 port->lun_map_size = 0;
3598                 free(port->lun_map, M_CTL);
3599                 port->lun_map = malloc(size * sizeof(uint32_t),
3600                     M_CTL, M_NOWAIT);
3601         }
3602         if (port->lun_map == NULL)
3603                 return (ENOMEM);
3604         for (i = 0; i < size; i++)
3605                 port->lun_map[i] = UINT32_MAX;
3606         port->lun_map_size = size;
3607         if (port->status & CTL_PORT_STATUS_ONLINE) {
3608                 if (port->lun_disable != NULL) {
3609                         STAILQ_FOREACH(lun, &softc->lun_list, links)
3610                                 port->lun_disable(port->targ_lun_arg, lun->lun);
3611                 }
3612                 ctl_isc_announce_port(port);
3613         }
3614         return (0);
3615 }
3616
3617 int
3618 ctl_lun_map_deinit(struct ctl_port *port)
3619 {
3620         struct ctl_softc *softc = port->ctl_softc;
3621         struct ctl_lun *lun;
3622
3623         if (port->lun_map == NULL)
3624                 return (0);
3625         port->lun_map_size = 0;
3626         free(port->lun_map, M_CTL);
3627         port->lun_map = NULL;
3628         if (port->status & CTL_PORT_STATUS_ONLINE) {
3629                 if (port->lun_enable != NULL) {
3630                         STAILQ_FOREACH(lun, &softc->lun_list, links)
3631                                 port->lun_enable(port->targ_lun_arg, lun->lun);
3632                 }
3633                 ctl_isc_announce_port(port);
3634         }
3635         return (0);
3636 }
3637
3638 int
3639 ctl_lun_map_set(struct ctl_port *port, uint32_t plun, uint32_t glun)
3640 {
3641         int status;
3642         uint32_t old;
3643
3644         if (port->lun_map == NULL) {
3645                 status = ctl_lun_map_init(port);
3646                 if (status != 0)
3647                         return (status);
3648         }
3649         if (plun >= port->lun_map_size)
3650                 return (EINVAL);
3651         old = port->lun_map[plun];
3652         port->lun_map[plun] = glun;
3653         if ((port->status & CTL_PORT_STATUS_ONLINE) && old == UINT32_MAX) {
3654                 if (port->lun_enable != NULL)
3655                         port->lun_enable(port->targ_lun_arg, plun);
3656                 ctl_isc_announce_port(port);
3657         }
3658         return (0);
3659 }
3660
3661 int
3662 ctl_lun_map_unset(struct ctl_port *port, uint32_t plun)
3663 {
3664         uint32_t old;
3665
3666         if (port->lun_map == NULL || plun >= port->lun_map_size)
3667                 return (0);
3668         old = port->lun_map[plun];
3669         port->lun_map[plun] = UINT32_MAX;
3670         if ((port->status & CTL_PORT_STATUS_ONLINE) && old != UINT32_MAX) {
3671                 if (port->lun_disable != NULL)
3672                         port->lun_disable(port->targ_lun_arg, plun);
3673                 ctl_isc_announce_port(port);
3674         }
3675         return (0);
3676 }
3677
3678 uint32_t
3679 ctl_lun_map_from_port(struct ctl_port *port, uint32_t lun_id)
3680 {
3681
3682         if (port == NULL)
3683                 return (UINT32_MAX);
3684         if (port->lun_map == NULL)
3685                 return (lun_id);
3686         if (lun_id > port->lun_map_size)
3687                 return (UINT32_MAX);
3688         return (port->lun_map[lun_id]);
3689 }
3690
3691 uint32_t
3692 ctl_lun_map_to_port(struct ctl_port *port, uint32_t lun_id)
3693 {
3694         uint32_t i;
3695
3696         if (port == NULL)
3697                 return (UINT32_MAX);
3698         if (port->lun_map == NULL)
3699                 return (lun_id);
3700         for (i = 0; i < port->lun_map_size; i++) {
3701                 if (port->lun_map[i] == lun_id)
3702                         return (i);
3703         }
3704         return (UINT32_MAX);
3705 }
3706
3707 uint32_t
3708 ctl_decode_lun(uint64_t encoded)
3709 {
3710         uint8_t lun[8];
3711         uint32_t result = 0xffffffff;
3712
3713         be64enc(lun, encoded);
3714         switch (lun[0] & RPL_LUNDATA_ATYP_MASK) {
3715         case RPL_LUNDATA_ATYP_PERIPH:
3716                 if ((lun[0] & 0x3f) == 0 && lun[2] == 0 && lun[3] == 0 &&
3717                     lun[4] == 0 && lun[5] == 0 && lun[6] == 0 && lun[7] == 0)
3718                         result = lun[1];
3719                 break;
3720         case RPL_LUNDATA_ATYP_FLAT:
3721                 if (lun[2] == 0 && lun[3] == 0 && lun[4] == 0 && lun[5] == 0 &&
3722                     lun[6] == 0 && lun[7] == 0)
3723                         result = ((lun[0] & 0x3f) << 8) + lun[1];
3724                 break;
3725         case RPL_LUNDATA_ATYP_EXTLUN:
3726                 switch (lun[0] & RPL_LUNDATA_EXT_EAM_MASK) {
3727                 case 0x02:
3728                         switch (lun[0] & RPL_LUNDATA_EXT_LEN_MASK) {
3729                         case 0x00:
3730                                 result = lun[1];
3731                                 break;
3732                         case 0x10:
3733                                 result = (lun[1] << 16) + (lun[2] << 8) +
3734                                     lun[3];
3735                                 break;
3736                         case 0x20:
3737                                 if (lun[1] == 0 && lun[6] == 0 && lun[7] == 0)
3738                                         result = (lun[2] << 24) +
3739                                             (lun[3] << 16) + (lun[4] << 8) +
3740                                             lun[5];
3741                                 break;
3742                         }
3743                         break;
3744                 case RPL_LUNDATA_EXT_EAM_NOT_SPEC:
3745                         result = 0xffffffff;
3746                         break;
3747                 }
3748                 break;
3749         }
3750         return (result);
3751 }
3752
3753 uint64_t
3754 ctl_encode_lun(uint32_t decoded)
3755 {
3756         uint64_t l = decoded;
3757
3758         if (l <= 0xff)
3759                 return (((uint64_t)RPL_LUNDATA_ATYP_PERIPH << 56) | (l << 48));
3760         if (l <= 0x3fff)
3761                 return (((uint64_t)RPL_LUNDATA_ATYP_FLAT << 56) | (l << 48));
3762         if (l <= 0xffffff)
3763                 return (((uint64_t)(RPL_LUNDATA_ATYP_EXTLUN | 0x12) << 56) |
3764                     (l << 32));
3765         return ((((uint64_t)RPL_LUNDATA_ATYP_EXTLUN | 0x22) << 56) | (l << 16));
3766 }
3767
3768 int
3769 ctl_ffz(uint32_t *mask, uint32_t first, uint32_t last)
3770 {
3771         int i;
3772
3773         for (i = first; i < last; i++) {
3774                 if ((mask[i / 32] & (1 << (i % 32))) == 0)
3775                         return (i);
3776         }
3777         return (-1);
3778 }
3779
3780 int
3781 ctl_set_mask(uint32_t *mask, uint32_t bit)
3782 {
3783         uint32_t chunk, piece;
3784
3785         chunk = bit >> 5;
3786         piece = bit % (sizeof(uint32_t) * 8);
3787
3788         if ((mask[chunk] & (1 << piece)) != 0)
3789                 return (-1);
3790         else
3791                 mask[chunk] |= (1 << piece);
3792
3793         return (0);
3794 }
3795
3796 int
3797 ctl_clear_mask(uint32_t *mask, uint32_t bit)
3798 {
3799         uint32_t chunk, piece;
3800
3801         chunk = bit >> 5;
3802         piece = bit % (sizeof(uint32_t) * 8);
3803
3804         if ((mask[chunk] & (1 << piece)) == 0)
3805                 return (-1);
3806         else
3807                 mask[chunk] &= ~(1 << piece);
3808
3809         return (0);
3810 }
3811
3812 int
3813 ctl_is_set(uint32_t *mask, uint32_t bit)
3814 {
3815         uint32_t chunk, piece;
3816
3817         chunk = bit >> 5;
3818         piece = bit % (sizeof(uint32_t) * 8);
3819
3820         if ((mask[chunk] & (1 << piece)) == 0)
3821                 return (0);
3822         else
3823                 return (1);
3824 }
3825
3826 static uint64_t
3827 ctl_get_prkey(struct ctl_lun *lun, uint32_t residx)
3828 {
3829         uint64_t *t;
3830
3831         t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3832         if (t == NULL)
3833                 return (0);
3834         return (t[residx % CTL_MAX_INIT_PER_PORT]);
3835 }
3836
3837 static void
3838 ctl_clr_prkey(struct ctl_lun *lun, uint32_t residx)
3839 {
3840         uint64_t *t;
3841
3842         t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3843         if (t == NULL)
3844                 return;
3845         t[residx % CTL_MAX_INIT_PER_PORT] = 0;
3846 }
3847
3848 static void
3849 ctl_alloc_prkey(struct ctl_lun *lun, uint32_t residx)
3850 {
3851         uint64_t *p;
3852         u_int i;
3853
3854         i = residx/CTL_MAX_INIT_PER_PORT;
3855         if (lun->pr_keys[i] != NULL)
3856                 return;
3857         mtx_unlock(&lun->lun_lock);
3858         p = malloc(sizeof(uint64_t) * CTL_MAX_INIT_PER_PORT, M_CTL,
3859             M_WAITOK | M_ZERO);
3860         mtx_lock(&lun->lun_lock);
3861         if (lun->pr_keys[i] == NULL)
3862                 lun->pr_keys[i] = p;
3863         else
3864                 free(p, M_CTL);
3865 }
3866
3867 static void
3868 ctl_set_prkey(struct ctl_lun *lun, uint32_t residx, uint64_t key)
3869 {
3870         uint64_t *t;
3871
3872         t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3873         KASSERT(t != NULL, ("prkey %d is not allocated", residx));
3874         t[residx % CTL_MAX_INIT_PER_PORT] = key;
3875 }
3876
3877 /*
3878  * ctl_softc, pool_name, total_ctl_io are passed in.
3879  * npool is passed out.
3880  */
3881 int
3882 ctl_pool_create(struct ctl_softc *ctl_softc, const char *pool_name,
3883                 uint32_t total_ctl_io, void **npool)
3884 {
3885         struct ctl_io_pool *pool;
3886
3887         pool = (struct ctl_io_pool *)malloc(sizeof(*pool), M_CTL,
3888                                             M_NOWAIT | M_ZERO);
3889         if (pool == NULL)
3890                 return (ENOMEM);
3891
3892         snprintf(pool->name, sizeof(pool->name), "CTL IO %s", pool_name);
3893         pool->ctl_softc = ctl_softc;
3894 #ifdef IO_POOLS
3895         pool->zone = uma_zsecond_create(pool->name, NULL,
3896             NULL, NULL, NULL, ctl_softc->io_zone);
3897         /* uma_prealloc(pool->zone, total_ctl_io); */
3898 #else
3899         pool->zone = ctl_softc->io_zone;
3900 #endif
3901
3902         *npool = pool;
3903         return (0);
3904 }
3905
3906 void
3907 ctl_pool_free(struct ctl_io_pool *pool)
3908 {
3909
3910         if (pool == NULL)
3911                 return;
3912
3913 #ifdef IO_POOLS
3914         uma_zdestroy(pool->zone);
3915 #endif
3916         free(pool, M_CTL);
3917 }
3918
3919 union ctl_io *
3920 ctl_alloc_io(void *pool_ref)
3921 {
3922         struct ctl_io_pool *pool = (struct ctl_io_pool *)pool_ref;
3923         union ctl_io *io;
3924
3925         io = uma_zalloc(pool->zone, M_WAITOK);
3926         if (io != NULL) {
3927                 io->io_hdr.pool = pool_ref;
3928                 CTL_SOFTC(io) = pool->ctl_softc;
3929         }
3930         return (io);
3931 }
3932
3933 union ctl_io *
3934 ctl_alloc_io_nowait(void *pool_ref)
3935 {
3936         struct ctl_io_pool *pool = (struct ctl_io_pool *)pool_ref;
3937         union ctl_io *io;
3938
3939         io = uma_zalloc(pool->zone, M_NOWAIT);
3940         if (io != NULL) {
3941                 io->io_hdr.pool = pool_ref;
3942                 CTL_SOFTC(io) = pool->ctl_softc;
3943         }
3944         return (io);
3945 }
3946
3947 void
3948 ctl_free_io(union ctl_io *io)
3949 {
3950         struct ctl_io_pool *pool;
3951
3952         if (io == NULL)
3953                 return;
3954
3955         pool = (struct ctl_io_pool *)io->io_hdr.pool;
3956         uma_zfree(pool->zone, io);
3957 }
3958
3959 void
3960 ctl_zero_io(union ctl_io *io)
3961 {
3962         struct ctl_io_pool *pool;
3963
3964         if (io == NULL)
3965                 return;
3966
3967         /*
3968          * May need to preserve linked list pointers at some point too.
3969          */
3970         pool = io->io_hdr.pool;
3971         memset(io, 0, sizeof(*io));
3972         io->io_hdr.pool = pool;
3973         CTL_SOFTC(io) = pool->ctl_softc;
3974 }
3975
3976 int
3977 ctl_expand_number(const char *buf, uint64_t *num)
3978 {
3979         char *endptr;
3980         uint64_t number;
3981         unsigned shift;
3982
3983         number = strtoq(buf, &endptr, 0);
3984
3985         switch (tolower((unsigned char)*endptr)) {
3986         case 'e':
3987                 shift = 60;
3988                 break;
3989         case 'p':
3990                 shift = 50;
3991                 break;
3992         case 't':
3993                 shift = 40;
3994                 break;
3995         case 'g':
3996                 shift = 30;
3997                 break;
3998         case 'm':
3999                 shift = 20;
4000                 break;
4001         case 'k':
4002                 shift = 10;
4003                 break;
4004         case 'b':
4005         case '\0': /* No unit. */
4006                 *num = number;
4007                 return (0);
4008         default:
4009                 /* Unrecognized unit. */
4010                 return (-1);
4011         }
4012
4013         if ((number << shift) >> shift != number) {
4014                 /* Overflow */
4015                 return (-1);
4016         }
4017         *num = number << shift;
4018         return (0);
4019 }
4020
4021
4022 /*
4023  * This routine could be used in the future to load default and/or saved
4024  * mode page parameters for a particuar lun.
4025  */
4026 static int
4027 ctl_init_page_index(struct ctl_lun *lun)
4028 {
4029         int i, page_code;
4030         struct ctl_page_index *page_index;
4031         const char *value;
4032         uint64_t ival;
4033
4034         memcpy(&lun->mode_pages.index, page_index_template,
4035                sizeof(page_index_template));
4036
4037         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
4038
4039                 page_index = &lun->mode_pages.index[i];
4040                 if (lun->be_lun->lun_type == T_DIRECT &&
4041                     (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
4042                         continue;
4043                 if (lun->be_lun->lun_type == T_PROCESSOR &&
4044                     (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
4045                         continue;
4046                 if (lun->be_lun->lun_type == T_CDROM &&
4047                     (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
4048                         continue;
4049
4050                 page_code = page_index->page_code & SMPH_PC_MASK;
4051                 switch (page_code) {
4052                 case SMS_RW_ERROR_RECOVERY_PAGE: {
4053                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4054                             ("subpage %#x for page %#x is incorrect!",
4055                             page_index->subpage, page_code));
4056                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_CURRENT],
4057                                &rw_er_page_default,
4058                                sizeof(rw_er_page_default));
4059                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_CHANGEABLE],
4060                                &rw_er_page_changeable,
4061                                sizeof(rw_er_page_changeable));
4062                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_DEFAULT],
4063                                &rw_er_page_default,
4064                                sizeof(rw_er_page_default));
4065                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_SAVED],
4066                                &rw_er_page_default,
4067                                sizeof(rw_er_page_default));
4068                         page_index->page_data =
4069                                 (uint8_t *)lun->mode_pages.rw_er_page;
4070                         break;
4071                 }
4072                 case SMS_FORMAT_DEVICE_PAGE: {
4073                         struct scsi_format_page *format_page;
4074
4075                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4076                             ("subpage %#x for page %#x is incorrect!",
4077                             page_index->subpage, page_code));
4078
4079                         /*
4080                          * Sectors per track are set above.  Bytes per
4081                          * sector need to be set here on a per-LUN basis.
4082                          */
4083                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_CURRENT],
4084                                &format_page_default,
4085                                sizeof(format_page_default));
4086                         memcpy(&lun->mode_pages.format_page[
4087                                CTL_PAGE_CHANGEABLE], &format_page_changeable,
4088                                sizeof(format_page_changeable));
4089                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_DEFAULT],
4090                                &format_page_default,
4091                                sizeof(format_page_default));
4092                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_SAVED],
4093                                &format_page_default,
4094                                sizeof(format_page_default));
4095
4096                         format_page = &lun->mode_pages.format_page[
4097                                 CTL_PAGE_CURRENT];
4098                         scsi_ulto2b(lun->be_lun->blocksize,
4099                                     format_page->bytes_per_sector);
4100
4101                         format_page = &lun->mode_pages.format_page[
4102                                 CTL_PAGE_DEFAULT];
4103                         scsi_ulto2b(lun->be_lun->blocksize,
4104                                     format_page->bytes_per_sector);
4105
4106                         format_page = &lun->mode_pages.format_page[
4107                                 CTL_PAGE_SAVED];
4108                         scsi_ulto2b(lun->be_lun->blocksize,
4109                                     format_page->bytes_per_sector);
4110
4111                         page_index->page_data =
4112                                 (uint8_t *)lun->mode_pages.format_page;
4113                         break;
4114                 }
4115                 case SMS_RIGID_DISK_PAGE: {
4116                         struct scsi_rigid_disk_page *rigid_disk_page;
4117                         uint32_t sectors_per_cylinder;
4118                         uint64_t cylinders;
4119 #ifndef __XSCALE__
4120                         int shift;
4121 #endif /* !__XSCALE__ */
4122
4123                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4124                             ("subpage %#x for page %#x is incorrect!",
4125                             page_index->subpage, page_code));
4126
4127                         /*
4128                          * Rotation rate and sectors per track are set
4129                          * above.  We calculate the cylinders here based on
4130                          * capacity.  Due to the number of heads and
4131                          * sectors per track we're using, smaller arrays
4132                          * may turn out to have 0 cylinders.  Linux and
4133                          * FreeBSD don't pay attention to these mode pages
4134                          * to figure out capacity, but Solaris does.  It
4135                          * seems to deal with 0 cylinders just fine, and
4136                          * works out a fake geometry based on the capacity.
4137                          */
4138                         memcpy(&lun->mode_pages.rigid_disk_page[
4139                                CTL_PAGE_DEFAULT], &rigid_disk_page_default,
4140                                sizeof(rigid_disk_page_default));
4141                         memcpy(&lun->mode_pages.rigid_disk_page[
4142                                CTL_PAGE_CHANGEABLE],&rigid_disk_page_changeable,
4143                                sizeof(rigid_disk_page_changeable));
4144
4145                         sectors_per_cylinder = CTL_DEFAULT_SECTORS_PER_TRACK *
4146                                 CTL_DEFAULT_HEADS;
4147
4148                         /*
4149                          * The divide method here will be more accurate,
4150                          * probably, but results in floating point being
4151                          * used in the kernel on i386 (__udivdi3()).  On the
4152                          * XScale, though, __udivdi3() is implemented in
4153                          * software.
4154                          *
4155                          * The shift method for cylinder calculation is
4156                          * accurate if sectors_per_cylinder is a power of
4157                          * 2.  Otherwise it might be slightly off -- you
4158                          * might have a bit of a truncation problem.
4159                          */
4160 #ifdef  __XSCALE__
4161                         cylinders = (lun->be_lun->maxlba + 1) /
4162                                 sectors_per_cylinder;
4163 #else
4164                         for (shift = 31; shift > 0; shift--) {
4165                                 if (sectors_per_cylinder & (1 << shift))
4166                                         break;
4167                         }
4168                         cylinders = (lun->be_lun->maxlba + 1) >> shift;
4169 #endif
4170
4171                         /*
4172                          * We've basically got 3 bytes, or 24 bits for the
4173                          * cylinder size in the mode page.  If we're over,
4174                          * just round down to 2^24.
4175                          */
4176                         if (cylinders > 0xffffff)
4177                                 cylinders = 0xffffff;
4178
4179                         rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4180                                 CTL_PAGE_DEFAULT];
4181                         scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4182
4183                         if ((value = ctl_get_opt(&lun->be_lun->options,
4184                             "rpm")) != NULL) {
4185                                 scsi_ulto2b(strtol(value, NULL, 0),
4186                                      rigid_disk_page->rotation_rate);
4187                         }
4188
4189                         memcpy(&lun->mode_pages.rigid_disk_page[CTL_PAGE_CURRENT],
4190                                &lun->mode_pages.rigid_disk_page[CTL_PAGE_DEFAULT],
4191                                sizeof(rigid_disk_page_default));
4192                         memcpy(&lun->mode_pages.rigid_disk_page[CTL_PAGE_SAVED],
4193                                &lun->mode_pages.rigid_disk_page[CTL_PAGE_DEFAULT],
4194                                sizeof(rigid_disk_page_default));
4195
4196                         page_index->page_data =
4197                                 (uint8_t *)lun->mode_pages.rigid_disk_page;
4198                         break;
4199                 }
4200                 case SMS_VERIFY_ERROR_RECOVERY_PAGE: {
4201                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4202                             ("subpage %#x for page %#x is incorrect!",
4203                             page_index->subpage, page_code));
4204                         memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_CURRENT],
4205                                &verify_er_page_default,
4206                                sizeof(verify_er_page_default));
4207                         memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_CHANGEABLE],
4208                                &verify_er_page_changeable,
4209                                sizeof(verify_er_page_changeable));
4210                         memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_DEFAULT],
4211                                &verify_er_page_default,
4212                                sizeof(verify_er_page_default));
4213                         memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_SAVED],
4214                                &verify_er_page_default,
4215                                sizeof(verify_er_page_default));
4216                         page_index->page_data =
4217                                 (uint8_t *)lun->mode_pages.verify_er_page;
4218                         break;
4219                 }
4220                 case SMS_CACHING_PAGE: {
4221                         struct scsi_caching_page *caching_page;
4222
4223                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4224                             ("subpage %#x for page %#x is incorrect!",
4225                             page_index->subpage, page_code));
4226                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_DEFAULT],
4227                                &caching_page_default,
4228                                sizeof(caching_page_default));
4229                         memcpy(&lun->mode_pages.caching_page[
4230                                CTL_PAGE_CHANGEABLE], &caching_page_changeable,
4231                                sizeof(caching_page_changeable));
4232                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4233                                &caching_page_default,
4234                                sizeof(caching_page_default));
4235                         caching_page = &lun->mode_pages.caching_page[
4236                             CTL_PAGE_SAVED];
4237                         value = ctl_get_opt(&lun->be_lun->options, "writecache");
4238                         if (value != NULL && strcmp(value, "off") == 0)
4239                                 caching_page->flags1 &= ~SCP_WCE;
4240                         value = ctl_get_opt(&lun->be_lun->options, "readcache");
4241                         if (value != NULL && strcmp(value, "off") == 0)
4242                                 caching_page->flags1 |= SCP_RCD;
4243                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_CURRENT],
4244                                &lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4245                                sizeof(caching_page_default));
4246                         page_index->page_data =
4247                                 (uint8_t *)lun->mode_pages.caching_page;
4248                         break;
4249                 }
4250                 case SMS_CONTROL_MODE_PAGE: {
4251                         switch (page_index->subpage) {
4252                         case SMS_SUBPAGE_PAGE_0: {
4253                                 struct scsi_control_page *control_page;
4254
4255                                 memcpy(&lun->mode_pages.control_page[
4256                                     CTL_PAGE_DEFAULT],
4257                                        &control_page_default,
4258                                        sizeof(control_page_default));
4259                                 memcpy(&lun->mode_pages.control_page[
4260                                     CTL_PAGE_CHANGEABLE],
4261                                        &control_page_changeable,
4262                                        sizeof(control_page_changeable));
4263                                 memcpy(&lun->mode_pages.control_page[
4264                                     CTL_PAGE_SAVED],
4265                                        &control_page_default,
4266                                        sizeof(control_page_default));
4267                                 control_page = &lun->mode_pages.control_page[
4268                                     CTL_PAGE_SAVED];
4269                                 value = ctl_get_opt(&lun->be_lun->options,
4270                                     "reordering");
4271                                 if (value != NULL &&
4272                                     strcmp(value, "unrestricted") == 0) {
4273                                         control_page->queue_flags &=
4274                                             ~SCP_QUEUE_ALG_MASK;
4275                                         control_page->queue_flags |=
4276                                             SCP_QUEUE_ALG_UNRESTRICTED;
4277                                 }
4278                                 memcpy(&lun->mode_pages.control_page[
4279                                     CTL_PAGE_CURRENT],
4280                                        &lun->mode_pages.control_page[
4281                                     CTL_PAGE_SAVED],
4282                                        sizeof(control_page_default));
4283                                 page_index->page_data =
4284                                     (uint8_t *)lun->mode_pages.control_page;
4285                                 break;
4286                         }
4287                         case 0x01:
4288                                 memcpy(&lun->mode_pages.control_ext_page[
4289                                     CTL_PAGE_DEFAULT],
4290                                        &control_ext_page_default,
4291                                        sizeof(control_ext_page_default));
4292                                 memcpy(&lun->mode_pages.control_ext_page[
4293                                     CTL_PAGE_CHANGEABLE],
4294                                        &control_ext_page_changeable,
4295                                        sizeof(control_ext_page_changeable));
4296                                 memcpy(&lun->mode_pages.control_ext_page[
4297                                     CTL_PAGE_SAVED],
4298                                        &control_ext_page_default,
4299                                        sizeof(control_ext_page_default));
4300                                 memcpy(&lun->mode_pages.control_ext_page[
4301                                     CTL_PAGE_CURRENT],
4302                                        &lun->mode_pages.control_ext_page[
4303                                     CTL_PAGE_SAVED],
4304                                        sizeof(control_ext_page_default));
4305                                 page_index->page_data =
4306                                     (uint8_t *)lun->mode_pages.control_ext_page;
4307                                 break;
4308                         default:
4309                                 panic("subpage %#x for page %#x is incorrect!",
4310                                       page_index->subpage, page_code);
4311                         }
4312                         break;
4313                 }
4314                 case SMS_INFO_EXCEPTIONS_PAGE: {
4315                         switch (page_index->subpage) {
4316                         case SMS_SUBPAGE_PAGE_0:
4317                                 memcpy(&lun->mode_pages.ie_page[CTL_PAGE_CURRENT],
4318                                        &ie_page_default,
4319                                        sizeof(ie_page_default));
4320                                 memcpy(&lun->mode_pages.ie_page[
4321                                        CTL_PAGE_CHANGEABLE], &ie_page_changeable,
4322                                        sizeof(ie_page_changeable));
4323                                 memcpy(&lun->mode_pages.ie_page[CTL_PAGE_DEFAULT],
4324                                        &ie_page_default,
4325                                        sizeof(ie_page_default));
4326                                 memcpy(&lun->mode_pages.ie_page[CTL_PAGE_SAVED],
4327                                        &ie_page_default,
4328                                        sizeof(ie_page_default));
4329                                 page_index->page_data =
4330                                         (uint8_t *)lun->mode_pages.ie_page;
4331                                 break;
4332                         case 0x02: {
4333                                 struct ctl_logical_block_provisioning_page *page;
4334
4335                                 memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_DEFAULT],
4336                                        &lbp_page_default,
4337                                        sizeof(lbp_page_default));
4338                                 memcpy(&lun->mode_pages.lbp_page[
4339                                        CTL_PAGE_CHANGEABLE], &lbp_page_changeable,
4340                                        sizeof(lbp_page_changeable));
4341                                 memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_SAVED],
4342                                        &lbp_page_default,
4343                                        sizeof(lbp_page_default));
4344                                 page = &lun->mode_pages.lbp_page[CTL_PAGE_SAVED];
4345                                 value = ctl_get_opt(&lun->be_lun->options,
4346                                     "avail-threshold");
4347                                 if (value != NULL &&
4348                                     ctl_expand_number(value, &ival) == 0) {
4349                                         page->descr[0].flags |= SLBPPD_ENABLED |
4350                                             SLBPPD_ARMING_DEC;
4351                                         if (lun->be_lun->blocksize)
4352                                                 ival /= lun->be_lun->blocksize;
4353                                         else
4354                                                 ival /= 512;
4355                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4356                                             page->descr[0].count);
4357                                 }
4358                                 value = ctl_get_opt(&lun->be_lun->options,
4359                                     "used-threshold");
4360                                 if (value != NULL &&
4361                                     ctl_expand_number(value, &ival) == 0) {
4362                                         page->descr[1].flags |= SLBPPD_ENABLED |
4363                                             SLBPPD_ARMING_INC;
4364                                         if (lun->be_lun->blocksize)
4365                                                 ival /= lun->be_lun->blocksize;
4366                                         else
4367                                                 ival /= 512;
4368                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4369                                             page->descr[1].count);
4370                                 }
4371                                 value = ctl_get_opt(&lun->be_lun->options,
4372                                     "pool-avail-threshold");
4373                                 if (value != NULL &&
4374                                     ctl_expand_number(value, &ival) == 0) {
4375                                         page->descr[2].flags |= SLBPPD_ENABLED |
4376                                             SLBPPD_ARMING_DEC;
4377                                         if (lun->be_lun->blocksize)
4378                                                 ival /= lun->be_lun->blocksize;
4379                                         else
4380                                                 ival /= 512;
4381                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4382                                             page->descr[2].count);
4383                                 }
4384                                 value = ctl_get_opt(&lun->be_lun->options,
4385                                     "pool-used-threshold");
4386                                 if (value != NULL &&
4387                                     ctl_expand_number(value, &ival) == 0) {
4388                                         page->descr[3].flags |= SLBPPD_ENABLED |
4389                                             SLBPPD_ARMING_INC;
4390                                         if (lun->be_lun->blocksize)
4391                                                 ival /= lun->be_lun->blocksize;
4392                                         else
4393                                                 ival /= 512;
4394                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4395                                             page->descr[3].count);
4396                                 }
4397                                 memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_CURRENT],
4398                                        &lun->mode_pages.lbp_page[CTL_PAGE_SAVED],
4399                                        sizeof(lbp_page_default));
4400                                 page_index->page_data =
4401                                         (uint8_t *)lun->mode_pages.lbp_page;
4402                                 break;
4403                         }
4404                         default:
4405                                 panic("subpage %#x for page %#x is incorrect!",
4406                                       page_index->subpage, page_code);
4407                         }
4408                         break;
4409                 }
4410                 case SMS_CDDVD_CAPS_PAGE:{
4411                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4412                             ("subpage %#x for page %#x is incorrect!",
4413                             page_index->subpage, page_code));
4414                         memcpy(&lun->mode_pages.cddvd_page[CTL_PAGE_DEFAULT],
4415                                &cddvd_page_default,
4416                                sizeof(cddvd_page_default));
4417                         memcpy(&lun->mode_pages.cddvd_page[
4418                                CTL_PAGE_CHANGEABLE], &cddvd_page_changeable,
4419                                sizeof(cddvd_page_changeable));
4420                         memcpy(&lun->mode_pages.cddvd_page[CTL_PAGE_SAVED],
4421                                &cddvd_page_default,
4422                                sizeof(cddvd_page_default));
4423                         memcpy(&lun->mode_pages.cddvd_page[CTL_PAGE_CURRENT],
4424                                &lun->mode_pages.cddvd_page[CTL_PAGE_SAVED],
4425                                sizeof(cddvd_page_default));
4426                         page_index->page_data =
4427                                 (uint8_t *)lun->mode_pages.cddvd_page;
4428                         break;
4429                 }
4430                 default:
4431                         panic("invalid page code value %#x", page_code);
4432                 }
4433         }
4434
4435         return (CTL_RETVAL_COMPLETE);
4436 }
4437
4438 static int
4439 ctl_init_log_page_index(struct ctl_lun *lun)
4440 {
4441         struct ctl_page_index *page_index;
4442         int i, j, k, prev;
4443
4444         memcpy(&lun->log_pages.index, log_page_index_template,
4445                sizeof(log_page_index_template));
4446
4447         prev = -1;
4448         for (i = 0, j = 0, k = 0; i < CTL_NUM_LOG_PAGES; i++) {
4449
4450                 page_index = &lun->log_pages.index[i];
4451                 if (lun->be_lun->lun_type == T_DIRECT &&
4452                     (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
4453                         continue;
4454                 if (lun->be_lun->lun_type == T_PROCESSOR &&
4455                     (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
4456                         continue;
4457                 if (lun->be_lun->lun_type == T_CDROM &&
4458                     (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
4459                         continue;
4460
4461                 if (page_index->page_code == SLS_LOGICAL_BLOCK_PROVISIONING &&
4462                     lun->backend->lun_attr == NULL)
4463                         continue;
4464
4465                 if (page_index->page_code != prev) {
4466                         lun->log_pages.pages_page[j] = page_index->page_code;
4467                         prev = page_index->page_code;
4468                         j++;
4469                 }
4470                 lun->log_pages.subpages_page[k*2] = page_index->page_code;
4471                 lun->log_pages.subpages_page[k*2+1] = page_index->subpage;
4472                 k++;
4473         }
4474         lun->log_pages.index[0].page_data = &lun->log_pages.pages_page[0];
4475         lun->log_pages.index[0].page_len = j;
4476         lun->log_pages.index[1].page_data = &lun->log_pages.subpages_page[0];
4477         lun->log_pages.index[1].page_len = k * 2;
4478         lun->log_pages.index[2].page_data = &lun->log_pages.lbp_page[0];
4479         lun->log_pages.index[2].page_len = 12*CTL_NUM_LBP_PARAMS;
4480         lun->log_pages.index[3].page_data = (uint8_t *)&lun->log_pages.stat_page;
4481         lun->log_pages.index[3].page_len = sizeof(lun->log_pages.stat_page);
4482         lun->log_pages.index[4].page_data = (uint8_t *)&lun->log_pages.ie_page;
4483         lun->log_pages.index[4].page_len = sizeof(lun->log_pages.ie_page);
4484
4485         return (CTL_RETVAL_COMPLETE);
4486 }
4487
4488 static int
4489 hex2bin(const char *str, uint8_t *buf, int buf_size)
4490 {
4491         int i;
4492         u_char c;
4493
4494         memset(buf, 0, buf_size);
4495         while (isspace(str[0]))
4496                 str++;
4497         if (str[0] == '0' && (str[1] == 'x' || str[1] == 'X'))
4498                 str += 2;
4499         buf_size *= 2;
4500         for (i = 0; str[i] != 0 && i < buf_size; i++) {
4501                 while (str[i] == '-')   /* Skip dashes in UUIDs. */
4502                         str++;
4503                 c = str[i];
4504                 if (isdigit(c))
4505                         c -= '0';
4506                 else if (isalpha(c))
4507                         c -= isupper(c) ? 'A' - 10 : 'a' - 10;
4508                 else
4509                         break;
4510                 if (c >= 16)
4511                         break;
4512                 if ((i & 1) == 0)
4513                         buf[i / 2] |= (c << 4);
4514                 else
4515                         buf[i / 2] |= c;
4516         }
4517         return ((i + 1) / 2);
4518 }
4519
4520 /*
4521  * LUN allocation.
4522  *
4523  * Requirements:
4524  * - caller allocates and zeros LUN storage, or passes in a NULL LUN if he
4525  *   wants us to allocate the LUN and he can block.
4526  * - ctl_softc is always set
4527  * - be_lun is set if the LUN has a backend (needed for disk LUNs)
4528  *
4529  * Returns 0 for success, non-zero (errno) for failure.
4530  */
4531 static int
4532 ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *ctl_lun,
4533               struct ctl_be_lun *const be_lun)
4534 {
4535         struct ctl_lun *nlun, *lun;
4536         struct scsi_vpd_id_descriptor *desc;
4537         struct scsi_vpd_id_t10 *t10id;
4538         const char *eui, *naa, *scsiname, *uuid, *vendor, *value;
4539         int lun_number, lun_malloced;
4540         int devidlen, idlen1, idlen2 = 0, len;
4541
4542         if (be_lun == NULL)
4543                 return (EINVAL);
4544
4545         /*
4546          * We currently only support Direct Access or Processor LUN types.
4547          */
4548         switch (be_lun->lun_type) {
4549         case T_DIRECT:
4550         case T_PROCESSOR:
4551         case T_CDROM:
4552                 break;
4553         case T_SEQUENTIAL:
4554         case T_CHANGER:
4555         default:
4556                 be_lun->lun_config_status(be_lun->be_lun,
4557                                           CTL_LUN_CONFIG_FAILURE);
4558                 break;
4559         }
4560         if (ctl_lun == NULL) {
4561                 lun = malloc(sizeof(*lun), M_CTL, M_WAITOK);
4562                 lun_malloced = 1;
4563         } else {
4564                 lun_malloced = 0;
4565                 lun = ctl_lun;
4566         }
4567
4568         memset(lun, 0, sizeof(*lun));
4569         if (lun_malloced)
4570                 lun->flags = CTL_LUN_MALLOCED;
4571
4572         lun->pending_sense = malloc(sizeof(struct scsi_sense_data *) *
4573             ctl_max_ports, M_DEVBUF, M_WAITOK | M_ZERO);
4574         lun->pending_ua = malloc(sizeof(ctl_ua_type *) * ctl_max_ports,
4575             M_DEVBUF, M_WAITOK | M_ZERO);
4576         lun->pr_keys = malloc(sizeof(uint64_t *) * ctl_max_ports,
4577             M_DEVBUF, M_WAITOK | M_ZERO);
4578
4579         /* Generate LUN ID. */
4580         devidlen = max(CTL_DEVID_MIN_LEN,
4581             strnlen(be_lun->device_id, CTL_DEVID_LEN));
4582         idlen1 = sizeof(*t10id) + devidlen;
4583         len = sizeof(struct scsi_vpd_id_descriptor) + idlen1;
4584         scsiname = ctl_get_opt(&be_lun->options, "scsiname");
4585         if (scsiname != NULL) {
4586                 idlen2 = roundup2(strlen(scsiname) + 1, 4);
4587                 len += sizeof(struct scsi_vpd_id_descriptor) + idlen2;
4588         }
4589         eui = ctl_get_opt(&be_lun->options, "eui");
4590         if (eui != NULL) {
4591                 len += sizeof(struct scsi_vpd_id_descriptor) + 16;
4592         }
4593         naa = ctl_get_opt(&be_lun->options, "naa");
4594         if (naa != NULL) {
4595                 len += sizeof(struct scsi_vpd_id_descriptor) + 16;
4596         }
4597         uuid = ctl_get_opt(&be_lun->options, "uuid");
4598         if (uuid != NULL) {
4599                 len += sizeof(struct scsi_vpd_id_descriptor) + 18;
4600         }
4601         lun->lun_devid = malloc(sizeof(struct ctl_devid) + len,
4602             M_CTL, M_WAITOK | M_ZERO);
4603         desc = (struct scsi_vpd_id_descriptor *)lun->lun_devid->data;
4604         desc->proto_codeset = SVPD_ID_CODESET_ASCII;
4605         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_T10;
4606         desc->length = idlen1;
4607         t10id = (struct scsi_vpd_id_t10 *)&desc->identifier[0];
4608         memset(t10id->vendor, ' ', sizeof(t10id->vendor));
4609         if ((vendor = ctl_get_opt(&be_lun->options, "vendor")) == NULL) {
4610                 strncpy((char *)t10id->vendor, CTL_VENDOR, sizeof(t10id->vendor));
4611         } else {
4612                 strncpy(t10id->vendor, vendor,
4613                     min(sizeof(t10id->vendor), strlen(vendor)));
4614         }
4615         strncpy((char *)t10id->vendor_spec_id,
4616             (char *)be_lun->device_id, devidlen);
4617         if (scsiname != NULL) {
4618                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4619                     desc->length);
4620                 desc->proto_codeset = SVPD_ID_CODESET_UTF8;
4621                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4622                     SVPD_ID_TYPE_SCSI_NAME;
4623                 desc->length = idlen2;
4624                 strlcpy(desc->identifier, scsiname, idlen2);
4625         }
4626         if (eui != NULL) {
4627                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4628                     desc->length);
4629                 desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4630                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4631                     SVPD_ID_TYPE_EUI64;
4632                 desc->length = hex2bin(eui, desc->identifier, 16);
4633                 desc->length = desc->length > 12 ? 16 :
4634                     (desc->length > 8 ? 12 : 8);
4635                 len -= 16 - desc->length;
4636         }
4637         if (naa != NULL) {
4638                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4639                     desc->length);
4640                 desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4641                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4642                     SVPD_ID_TYPE_NAA;
4643                 desc->length = hex2bin(naa, desc->identifier, 16);
4644                 desc->length = desc->length > 8 ? 16 : 8;
4645                 len -= 16 - desc->length;
4646         }
4647         if (uuid != 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_UUID;
4653                 desc->identifier[0] = 0x10;
4654                 hex2bin(uuid, &desc->identifier[2], 16);
4655                 desc->length = 18;
4656         }
4657         lun->lun_devid->len = len;
4658
4659         mtx_lock(&ctl_softc->ctl_lock);
4660         /*
4661          * See if the caller requested a particular LUN number.  If so, see
4662          * if it is available.  Otherwise, allocate the first available LUN.
4663          */
4664         if (be_lun->flags & CTL_LUN_FLAG_ID_REQ) {
4665                 if ((be_lun->req_lun_id > (ctl_max_luns - 1))
4666                  || (ctl_is_set(ctl_softc->ctl_lun_mask, be_lun->req_lun_id))) {
4667                         mtx_unlock(&ctl_softc->ctl_lock);
4668                         if (be_lun->req_lun_id > (ctl_max_luns - 1)) {
4669                                 printf("ctl: requested LUN ID %d is higher "
4670                                        "than ctl_max_luns - 1 (%d)\n",
4671                                        be_lun->req_lun_id, ctl_max_luns - 1);
4672                         } else {
4673                                 /*
4674                                  * XXX KDM return an error, or just assign
4675                                  * another LUN ID in this case??
4676                                  */
4677                                 printf("ctl: requested LUN ID %d is already "
4678                                        "in use\n", be_lun->req_lun_id);
4679                         }
4680 fail:
4681                         free(lun->lun_devid, M_CTL);
4682                         if (lun->flags & CTL_LUN_MALLOCED)
4683                                 free(lun, M_CTL);
4684                         be_lun->lun_config_status(be_lun->be_lun,
4685                                                   CTL_LUN_CONFIG_FAILURE);
4686                         return (ENOSPC);
4687                 }
4688                 lun_number = be_lun->req_lun_id;
4689         } else {
4690                 lun_number = ctl_ffz(ctl_softc->ctl_lun_mask, 0, ctl_max_luns);
4691                 if (lun_number == -1) {
4692                         mtx_unlock(&ctl_softc->ctl_lock);
4693                         printf("ctl: can't allocate LUN, out of LUNs\n");
4694                         goto fail;
4695                 }
4696         }
4697         ctl_set_mask(ctl_softc->ctl_lun_mask, lun_number);
4698         mtx_unlock(&ctl_softc->ctl_lock);
4699
4700         mtx_init(&lun->lun_lock, "CTL LUN", NULL, MTX_DEF);
4701         lun->lun = lun_number;
4702         lun->be_lun = be_lun;
4703         /*
4704          * The processor LUN is always enabled.  Disk LUNs come on line
4705          * disabled, and must be enabled by the backend.
4706          */
4707         lun->flags |= CTL_LUN_DISABLED;
4708         lun->backend = be_lun->be;
4709         be_lun->ctl_lun = lun;
4710         be_lun->lun_id = lun_number;
4711         atomic_add_int(&be_lun->be->num_luns, 1);
4712         if (be_lun->flags & CTL_LUN_FLAG_EJECTED)
4713                 lun->flags |= CTL_LUN_EJECTED;
4714         if (be_lun->flags & CTL_LUN_FLAG_NO_MEDIA)
4715                 lun->flags |= CTL_LUN_NO_MEDIA;
4716         if (be_lun->flags & CTL_LUN_FLAG_STOPPED)
4717                 lun->flags |= CTL_LUN_STOPPED;
4718
4719         if (be_lun->flags & CTL_LUN_FLAG_PRIMARY)
4720                 lun->flags |= CTL_LUN_PRIMARY_SC;
4721
4722         value = ctl_get_opt(&be_lun->options, "removable");
4723         if (value != NULL) {
4724                 if (strcmp(value, "on") == 0)
4725                         lun->flags |= CTL_LUN_REMOVABLE;
4726         } else if (be_lun->lun_type == T_CDROM)
4727                 lun->flags |= CTL_LUN_REMOVABLE;
4728
4729         lun->ctl_softc = ctl_softc;
4730 #ifdef CTL_TIME_IO
4731         lun->last_busy = getsbinuptime();
4732 #endif
4733         TAILQ_INIT(&lun->ooa_queue);
4734         TAILQ_INIT(&lun->blocked_queue);
4735         STAILQ_INIT(&lun->error_list);
4736         lun->ie_reported = 1;
4737         callout_init_mtx(&lun->ie_callout, &lun->lun_lock, 0);
4738         ctl_tpc_lun_init(lun);
4739         if (lun->flags & CTL_LUN_REMOVABLE) {
4740                 lun->prevent = malloc((CTL_MAX_INITIATORS + 31) / 32 * 4,
4741                     M_CTL, M_WAITOK);
4742         }
4743
4744         /*
4745          * Initialize the mode and log page index.
4746          */
4747         ctl_init_page_index(lun);
4748         ctl_init_log_page_index(lun);
4749
4750         /* Setup statistics gathering */
4751 #ifdef CTL_LEGACY_STATS
4752         lun->legacy_stats.device_type = be_lun->lun_type;
4753         lun->legacy_stats.lun_number = lun_number;
4754         lun->legacy_stats.blocksize = be_lun->blocksize;
4755         if (be_lun->blocksize == 0)
4756                 lun->legacy_stats.flags = CTL_LUN_STATS_NO_BLOCKSIZE;
4757         lun->legacy_stats.ports = malloc(sizeof(struct ctl_lun_io_port_stats) *
4758             ctl_max_ports, M_DEVBUF, M_WAITOK | M_ZERO);
4759         for (len = 0; len < ctl_max_ports; len++)
4760                 lun->legacy_stats.ports[len].targ_port = len;
4761 #endif /* CTL_LEGACY_STATS */
4762         lun->stats.item = lun_number;
4763
4764         /*
4765          * Now, before we insert this lun on the lun list, set the lun
4766          * inventory changed UA for all other luns.
4767          */
4768         mtx_lock(&ctl_softc->ctl_lock);
4769         STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) {
4770                 mtx_lock(&nlun->lun_lock);
4771                 ctl_est_ua_all(nlun, -1, CTL_UA_LUN_CHANGE);
4772                 mtx_unlock(&nlun->lun_lock);
4773         }
4774         STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links);
4775         ctl_softc->ctl_luns[lun_number] = lun;
4776         ctl_softc->num_luns++;
4777         mtx_unlock(&ctl_softc->ctl_lock);
4778
4779         lun->be_lun->lun_config_status(lun->be_lun->be_lun, CTL_LUN_CONFIG_OK);
4780         return (0);
4781 }
4782
4783 /*
4784  * Delete a LUN.
4785  * Assumptions:
4786  * - LUN has already been marked invalid and any pending I/O has been taken
4787  *   care of.
4788  */
4789 static int
4790 ctl_free_lun(struct ctl_lun *lun)
4791 {
4792         struct ctl_softc *softc = lun->ctl_softc;
4793         struct ctl_lun *nlun;
4794         int i;
4795
4796         KASSERT(TAILQ_EMPTY(&lun->ooa_queue),
4797             ("Freeing a LUN %p with outstanding I/O!\n", lun));
4798
4799         mtx_lock(&softc->ctl_lock);
4800         STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links);
4801         ctl_clear_mask(softc->ctl_lun_mask, lun->lun);
4802         softc->ctl_luns[lun->lun] = NULL;
4803         softc->num_luns--;
4804         STAILQ_FOREACH(nlun, &softc->lun_list, links) {
4805                 mtx_lock(&nlun->lun_lock);
4806                 ctl_est_ua_all(nlun, -1, CTL_UA_LUN_CHANGE);
4807                 mtx_unlock(&nlun->lun_lock);
4808         }
4809         mtx_unlock(&softc->ctl_lock);
4810
4811         /*
4812          * Tell the backend to free resources, if this LUN has a backend.
4813          */
4814         atomic_subtract_int(&lun->be_lun->be->num_luns, 1);
4815         lun->be_lun->lun_shutdown(lun->be_lun->be_lun);
4816
4817         lun->ie_reportcnt = UINT32_MAX;
4818         callout_drain(&lun->ie_callout);
4819         ctl_tpc_lun_shutdown(lun);
4820         mtx_destroy(&lun->lun_lock);
4821         free(lun->lun_devid, M_CTL);
4822         for (i = 0; i < ctl_max_ports; i++)
4823                 free(lun->pending_ua[i], M_CTL);
4824         free(lun->pending_ua, M_DEVBUF);
4825         for (i = 0; i < ctl_max_ports; i++)
4826                 free(lun->pr_keys[i], M_CTL);
4827         free(lun->pr_keys, M_DEVBUF);
4828         free(lun->write_buffer, M_CTL);
4829         free(lun->prevent, M_CTL);
4830         if (lun->flags & CTL_LUN_MALLOCED)
4831                 free(lun, M_CTL);
4832
4833         return (0);
4834 }
4835
4836 static void
4837 ctl_create_lun(struct ctl_be_lun *be_lun)
4838 {
4839
4840         /*
4841          * ctl_alloc_lun() should handle all potential failure cases.
4842          */
4843         ctl_alloc_lun(control_softc, NULL, be_lun);
4844 }
4845
4846 int
4847 ctl_add_lun(struct ctl_be_lun *be_lun)
4848 {
4849         struct ctl_softc *softc = control_softc;
4850
4851         mtx_lock(&softc->ctl_lock);
4852         STAILQ_INSERT_TAIL(&softc->pending_lun_queue, be_lun, links);
4853         mtx_unlock(&softc->ctl_lock);
4854         wakeup(&softc->pending_lun_queue);
4855
4856         return (0);
4857 }
4858
4859 int
4860 ctl_enable_lun(struct ctl_be_lun *be_lun)
4861 {
4862         struct ctl_softc *softc;
4863         struct ctl_port *port, *nport;
4864         struct ctl_lun *lun;
4865         int retval;
4866
4867         lun = (struct ctl_lun *)be_lun->ctl_lun;
4868         softc = lun->ctl_softc;
4869
4870         mtx_lock(&softc->ctl_lock);
4871         mtx_lock(&lun->lun_lock);
4872         if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4873                 /*
4874                  * eh?  Why did we get called if the LUN is already
4875                  * enabled?
4876                  */
4877                 mtx_unlock(&lun->lun_lock);
4878                 mtx_unlock(&softc->ctl_lock);
4879                 return (0);
4880         }
4881         lun->flags &= ~CTL_LUN_DISABLED;
4882         mtx_unlock(&lun->lun_lock);
4883
4884         STAILQ_FOREACH_SAFE(port, &softc->port_list, links, nport) {
4885                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0 ||
4886                     port->lun_map != NULL || port->lun_enable == NULL)
4887                         continue;
4888
4889                 /*
4890                  * Drop the lock while we call the FETD's enable routine.
4891                  * This can lead to a callback into CTL (at least in the
4892                  * case of the internal initiator frontend.
4893                  */
4894                 mtx_unlock(&softc->ctl_lock);
4895                 retval = port->lun_enable(port->targ_lun_arg, lun->lun);
4896                 mtx_lock(&softc->ctl_lock);
4897                 if (retval != 0) {
4898                         printf("%s: FETD %s port %d returned error "
4899                                "%d for lun_enable on lun %jd\n",
4900                                __func__, port->port_name, port->targ_port,
4901                                retval, (intmax_t)lun->lun);
4902                 }
4903         }
4904
4905         mtx_unlock(&softc->ctl_lock);
4906         ctl_isc_announce_lun(lun);
4907
4908         return (0);
4909 }
4910
4911 int
4912 ctl_disable_lun(struct ctl_be_lun *be_lun)
4913 {
4914         struct ctl_softc *softc;
4915         struct ctl_port *port;
4916         struct ctl_lun *lun;
4917         int retval;
4918
4919         lun = (struct ctl_lun *)be_lun->ctl_lun;
4920         softc = lun->ctl_softc;
4921
4922         mtx_lock(&softc->ctl_lock);
4923         mtx_lock(&lun->lun_lock);
4924         if (lun->flags & CTL_LUN_DISABLED) {
4925                 mtx_unlock(&lun->lun_lock);
4926                 mtx_unlock(&softc->ctl_lock);
4927                 return (0);
4928         }
4929         lun->flags |= CTL_LUN_DISABLED;
4930         mtx_unlock(&lun->lun_lock);
4931
4932         STAILQ_FOREACH(port, &softc->port_list, links) {
4933                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0 ||
4934                     port->lun_map != NULL || port->lun_disable == NULL)
4935                         continue;
4936
4937                 /*
4938                  * Drop the lock before we call the frontend's disable
4939                  * routine, to avoid lock order reversals.
4940                  *
4941                  * XXX KDM what happens if the frontend list changes while
4942                  * we're traversing it?  It's unlikely, but should be handled.
4943                  */
4944                 mtx_unlock(&softc->ctl_lock);
4945                 retval = port->lun_disable(port->targ_lun_arg, lun->lun);
4946                 mtx_lock(&softc->ctl_lock);
4947                 if (retval != 0) {
4948                         printf("%s: FETD %s port %d returned error "
4949                                "%d for lun_disable on lun %jd\n",
4950                                __func__, port->port_name, port->targ_port,
4951                                retval, (intmax_t)lun->lun);
4952                 }
4953         }
4954
4955         mtx_unlock(&softc->ctl_lock);
4956         ctl_isc_announce_lun(lun);
4957
4958         return (0);
4959 }
4960
4961 int
4962 ctl_start_lun(struct ctl_be_lun *be_lun)
4963 {
4964         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4965
4966         mtx_lock(&lun->lun_lock);
4967         lun->flags &= ~CTL_LUN_STOPPED;
4968         mtx_unlock(&lun->lun_lock);
4969         return (0);
4970 }
4971
4972 int
4973 ctl_stop_lun(struct ctl_be_lun *be_lun)
4974 {
4975         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4976
4977         mtx_lock(&lun->lun_lock);
4978         lun->flags |= CTL_LUN_STOPPED;
4979         mtx_unlock(&lun->lun_lock);
4980         return (0);
4981 }
4982
4983 int
4984 ctl_lun_no_media(struct ctl_be_lun *be_lun)
4985 {
4986         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4987
4988         mtx_lock(&lun->lun_lock);
4989         lun->flags |= CTL_LUN_NO_MEDIA;
4990         mtx_unlock(&lun->lun_lock);
4991         return (0);
4992 }
4993
4994 int
4995 ctl_lun_has_media(struct ctl_be_lun *be_lun)
4996 {
4997         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4998         union ctl_ha_msg msg;
4999
5000         mtx_lock(&lun->lun_lock);
5001         lun->flags &= ~(CTL_LUN_NO_MEDIA | CTL_LUN_EJECTED);
5002         if (lun->flags & CTL_LUN_REMOVABLE)
5003                 ctl_est_ua_all(lun, -1, CTL_UA_MEDIUM_CHANGE);
5004         mtx_unlock(&lun->lun_lock);
5005         if ((lun->flags & CTL_LUN_REMOVABLE) &&
5006             lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
5007                 bzero(&msg.ua, sizeof(msg.ua));
5008                 msg.hdr.msg_type = CTL_MSG_UA;
5009                 msg.hdr.nexus.initid = -1;
5010                 msg.hdr.nexus.targ_port = -1;
5011                 msg.hdr.nexus.targ_lun = lun->lun;
5012                 msg.hdr.nexus.targ_mapped_lun = lun->lun;
5013                 msg.ua.ua_all = 1;
5014                 msg.ua.ua_set = 1;
5015                 msg.ua.ua_type = CTL_UA_MEDIUM_CHANGE;
5016                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg.ua),
5017                     M_WAITOK);
5018         }
5019         return (0);
5020 }
5021
5022 int
5023 ctl_lun_ejected(struct ctl_be_lun *be_lun)
5024 {
5025         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
5026
5027         mtx_lock(&lun->lun_lock);
5028         lun->flags |= CTL_LUN_EJECTED;
5029         mtx_unlock(&lun->lun_lock);
5030         return (0);
5031 }
5032
5033 int
5034 ctl_lun_primary(struct ctl_be_lun *be_lun)
5035 {
5036         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
5037
5038         mtx_lock(&lun->lun_lock);
5039         lun->flags |= CTL_LUN_PRIMARY_SC;
5040         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
5041         mtx_unlock(&lun->lun_lock);
5042         ctl_isc_announce_lun(lun);
5043         return (0);
5044 }
5045
5046 int
5047 ctl_lun_secondary(struct ctl_be_lun *be_lun)
5048 {
5049         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
5050
5051         mtx_lock(&lun->lun_lock);
5052         lun->flags &= ~CTL_LUN_PRIMARY_SC;
5053         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
5054         mtx_unlock(&lun->lun_lock);
5055         ctl_isc_announce_lun(lun);
5056         return (0);
5057 }
5058
5059 int
5060 ctl_invalidate_lun(struct ctl_be_lun *be_lun)
5061 {
5062         struct ctl_softc *softc;
5063         struct ctl_lun *lun;
5064
5065         lun = (struct ctl_lun *)be_lun->ctl_lun;
5066         softc = lun->ctl_softc;
5067
5068         mtx_lock(&lun->lun_lock);
5069
5070         /*
5071          * The LUN needs to be disabled before it can be marked invalid.
5072          */
5073         if ((lun->flags & CTL_LUN_DISABLED) == 0) {
5074                 mtx_unlock(&lun->lun_lock);
5075                 return (-1);
5076         }
5077         /*
5078          * Mark the LUN invalid.
5079          */
5080         lun->flags |= CTL_LUN_INVALID;
5081
5082         /*
5083          * If there is nothing in the OOA queue, go ahead and free the LUN.
5084          * If we have something in the OOA queue, we'll free it when the
5085          * last I/O completes.
5086          */
5087         if (TAILQ_EMPTY(&lun->ooa_queue)) {
5088                 mtx_unlock(&lun->lun_lock);
5089                 ctl_free_lun(lun);
5090         } else
5091                 mtx_unlock(&lun->lun_lock);
5092
5093         return (0);
5094 }
5095
5096 void
5097 ctl_lun_capacity_changed(struct ctl_be_lun *be_lun)
5098 {
5099         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
5100         union ctl_ha_msg msg;
5101
5102         mtx_lock(&lun->lun_lock);
5103         ctl_est_ua_all(lun, -1, CTL_UA_CAPACITY_CHANGE);
5104         mtx_unlock(&lun->lun_lock);
5105         if (lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
5106                 /* Send msg to other side. */
5107                 bzero(&msg.ua, sizeof(msg.ua));
5108                 msg.hdr.msg_type = CTL_MSG_UA;
5109                 msg.hdr.nexus.initid = -1;
5110                 msg.hdr.nexus.targ_port = -1;
5111                 msg.hdr.nexus.targ_lun = lun->lun;
5112                 msg.hdr.nexus.targ_mapped_lun = lun->lun;
5113                 msg.ua.ua_all = 1;
5114                 msg.ua.ua_set = 1;
5115                 msg.ua.ua_type = CTL_UA_CAPACITY_CHANGE;
5116                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg.ua),
5117                     M_WAITOK);
5118         }
5119 }
5120
5121 /*
5122  * Backend "memory move is complete" callback for requests that never
5123  * make it down to say RAIDCore's configuration code.
5124  */
5125 int
5126 ctl_config_move_done(union ctl_io *io)
5127 {
5128         int retval;
5129
5130         CTL_DEBUG_PRINT(("ctl_config_move_done\n"));
5131         KASSERT(io->io_hdr.io_type == CTL_IO_SCSI,
5132             ("Config I/O type isn't CTL_IO_SCSI (%d)!", io->io_hdr.io_type));
5133
5134         if ((io->io_hdr.port_status != 0) &&
5135             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5136              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5137                 ctl_set_internal_failure(&io->scsiio, /*sks_valid*/ 1,
5138                     /*retry_count*/ io->io_hdr.port_status);
5139         } else if (io->scsiio.kern_data_resid != 0 &&
5140             (io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT &&
5141             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5142              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5143                 ctl_set_invalid_field_ciu(&io->scsiio);
5144         }
5145
5146         if (ctl_debug & CTL_DEBUG_CDB_DATA)
5147                 ctl_data_print(io);
5148         if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN) ||
5149             ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
5150              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS) ||
5151             ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) {
5152                 /*
5153                  * XXX KDM just assuming a single pointer here, and not a
5154                  * S/G list.  If we start using S/G lists for config data,
5155                  * we'll need to know how to clean them up here as well.
5156                  */
5157                 if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5158                         free(io->scsiio.kern_data_ptr, M_CTL);
5159                 ctl_done(io);
5160                 retval = CTL_RETVAL_COMPLETE;
5161         } else {
5162                 /*
5163                  * XXX KDM now we need to continue data movement.  Some
5164                  * options:
5165                  * - call ctl_scsiio() again?  We don't do this for data
5166                  *   writes, because for those at least we know ahead of
5167                  *   time where the write will go and how long it is.  For
5168                  *   config writes, though, that information is largely
5169                  *   contained within the write itself, thus we need to
5170                  *   parse out the data again.
5171                  *
5172                  * - Call some other function once the data is in?
5173                  */
5174
5175                 /*
5176                  * XXX KDM call ctl_scsiio() again for now, and check flag
5177                  * bits to see whether we're allocated or not.
5178                  */
5179                 retval = ctl_scsiio(&io->scsiio);
5180         }
5181         return (retval);
5182 }
5183
5184 /*
5185  * This gets called by a backend driver when it is done with a
5186  * data_submit method.
5187  */
5188 void
5189 ctl_data_submit_done(union ctl_io *io)
5190 {
5191         /*
5192          * If the IO_CONT flag is set, we need to call the supplied
5193          * function to continue processing the I/O, instead of completing
5194          * the I/O just yet.
5195          *
5196          * If there is an error, though, we don't want to keep processing.
5197          * Instead, just send status back to the initiator.
5198          */
5199         if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5200             (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5201             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5202              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5203                 io->scsiio.io_cont(io);
5204                 return;
5205         }
5206         ctl_done(io);
5207 }
5208
5209 /*
5210  * This gets called by a backend driver when it is done with a
5211  * configuration write.
5212  */
5213 void
5214 ctl_config_write_done(union ctl_io *io)
5215 {
5216         uint8_t *buf;
5217
5218         /*
5219          * If the IO_CONT flag is set, we need to call the supplied
5220          * function to continue processing the I/O, instead of completing
5221          * the I/O just yet.
5222          *
5223          * If there is an error, though, we don't want to keep processing.
5224          * Instead, just send status back to the initiator.
5225          */
5226         if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5227             (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5228             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5229              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5230                 io->scsiio.io_cont(io);
5231                 return;
5232         }
5233         /*
5234          * Since a configuration write can be done for commands that actually
5235          * have data allocated, like write buffer, and commands that have
5236          * no data, like start/stop unit, we need to check here.
5237          */
5238         if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5239                 buf = io->scsiio.kern_data_ptr;
5240         else
5241                 buf = NULL;
5242         ctl_done(io);
5243         if (buf)
5244                 free(buf, M_CTL);
5245 }
5246
5247 void
5248 ctl_config_read_done(union ctl_io *io)
5249 {
5250         uint8_t *buf;
5251
5252         /*
5253          * If there is some error -- we are done, skip data transfer.
5254          */
5255         if ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0 ||
5256             ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
5257              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) {
5258                 if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5259                         buf = io->scsiio.kern_data_ptr;
5260                 else
5261                         buf = NULL;
5262                 ctl_done(io);
5263                 if (buf)
5264                         free(buf, M_CTL);
5265                 return;
5266         }
5267
5268         /*
5269          * If the IO_CONT flag is set, we need to call the supplied
5270          * function to continue processing the I/O, instead of completing
5271          * the I/O just yet.
5272          */
5273         if (io->io_hdr.flags & CTL_FLAG_IO_CONT) {
5274                 io->scsiio.io_cont(io);
5275                 return;
5276         }
5277
5278         ctl_datamove(io);
5279 }
5280
5281 /*
5282  * SCSI release command.
5283  */
5284 int
5285 ctl_scsi_release(struct ctl_scsiio *ctsio)
5286 {
5287         struct ctl_lun *lun = CTL_LUN(ctsio);
5288         uint32_t residx;
5289
5290         CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
5291
5292         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5293
5294         /*
5295          * XXX KDM right now, we only support LUN reservation.  We don't
5296          * support 3rd party reservations, or extent reservations, which
5297          * might actually need the parameter list.  If we've gotten this
5298          * far, we've got a LUN reservation.  Anything else got kicked out
5299          * above.  So, according to SPC, ignore the length.
5300          */
5301
5302         mtx_lock(&lun->lun_lock);
5303
5304         /*
5305          * According to SPC, it is not an error for an intiator to attempt
5306          * to release a reservation on a LUN that isn't reserved, or that
5307          * is reserved by another initiator.  The reservation can only be
5308          * released, though, by the initiator who made it or by one of
5309          * several reset type events.
5310          */
5311         if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == residx))
5312                         lun->flags &= ~CTL_LUN_RESERVED;
5313
5314         mtx_unlock(&lun->lun_lock);
5315
5316         ctl_set_success(ctsio);
5317         ctl_done((union ctl_io *)ctsio);
5318         return (CTL_RETVAL_COMPLETE);
5319 }
5320
5321 int
5322 ctl_scsi_reserve(struct ctl_scsiio *ctsio)
5323 {
5324         struct ctl_lun *lun = CTL_LUN(ctsio);
5325         uint32_t residx;
5326
5327         CTL_DEBUG_PRINT(("ctl_reserve\n"));
5328
5329         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5330
5331         /*
5332          * XXX KDM right now, we only support LUN reservation.  We don't
5333          * support 3rd party reservations, or extent reservations, which
5334          * might actually need the parameter list.  If we've gotten this
5335          * far, we've got a LUN reservation.  Anything else got kicked out
5336          * above.  So, according to SPC, ignore the length.
5337          */
5338
5339         mtx_lock(&lun->lun_lock);
5340         if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx != residx)) {
5341                 ctl_set_reservation_conflict(ctsio);
5342                 goto bailout;
5343         }
5344
5345         /* SPC-3 exceptions to SPC-2 RESERVE and RELEASE behavior. */
5346         if (lun->flags & CTL_LUN_PR_RESERVED) {
5347                 ctl_set_success(ctsio);
5348                 goto bailout;
5349         }
5350
5351         lun->flags |= CTL_LUN_RESERVED;
5352         lun->res_idx = residx;
5353         ctl_set_success(ctsio);
5354
5355 bailout:
5356         mtx_unlock(&lun->lun_lock);
5357         ctl_done((union ctl_io *)ctsio);
5358         return (CTL_RETVAL_COMPLETE);
5359 }
5360
5361 int
5362 ctl_start_stop(struct ctl_scsiio *ctsio)
5363 {
5364         struct ctl_lun *lun = CTL_LUN(ctsio);
5365         struct scsi_start_stop_unit *cdb;
5366         int retval;
5367
5368         CTL_DEBUG_PRINT(("ctl_start_stop\n"));
5369
5370         cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
5371
5372         if ((cdb->how & SSS_PC_MASK) == 0) {
5373                 if ((lun->flags & CTL_LUN_PR_RESERVED) &&
5374                     (cdb->how & SSS_START) == 0) {
5375                         uint32_t residx;
5376
5377                         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5378                         if (ctl_get_prkey(lun, residx) == 0 ||
5379                             (lun->pr_res_idx != residx && lun->pr_res_type < 4)) {
5380
5381                                 ctl_set_reservation_conflict(ctsio);
5382                                 ctl_done((union ctl_io *)ctsio);
5383                                 return (CTL_RETVAL_COMPLETE);
5384                         }
5385                 }
5386
5387                 if ((cdb->how & SSS_LOEJ) &&
5388                     (lun->flags & CTL_LUN_REMOVABLE) == 0) {
5389                         ctl_set_invalid_field(ctsio,
5390                                               /*sks_valid*/ 1,
5391                                               /*command*/ 1,
5392                                               /*field*/ 4,
5393                                               /*bit_valid*/ 1,
5394                                               /*bit*/ 1);
5395                         ctl_done((union ctl_io *)ctsio);
5396                         return (CTL_RETVAL_COMPLETE);
5397                 }
5398
5399                 if ((cdb->how & SSS_START) == 0 && (cdb->how & SSS_LOEJ) &&
5400                     lun->prevent_count > 0) {
5401                         /* "Medium removal prevented" */
5402                         ctl_set_sense(ctsio, /*current_error*/ 1,
5403                             /*sense_key*/(lun->flags & CTL_LUN_NO_MEDIA) ?
5404                              SSD_KEY_NOT_READY : SSD_KEY_ILLEGAL_REQUEST,
5405                             /*asc*/ 0x53, /*ascq*/ 0x02, SSD_ELEM_NONE);
5406                         ctl_done((union ctl_io *)ctsio);
5407                         return (CTL_RETVAL_COMPLETE);
5408                 }
5409         }
5410
5411         retval = lun->backend->config_write((union ctl_io *)ctsio);
5412         return (retval);
5413 }
5414
5415 int
5416 ctl_prevent_allow(struct ctl_scsiio *ctsio)
5417 {
5418         struct ctl_lun *lun = CTL_LUN(ctsio);
5419         struct scsi_prevent *cdb;
5420         int retval;
5421         uint32_t initidx;
5422
5423         CTL_DEBUG_PRINT(("ctl_prevent_allow\n"));
5424
5425         cdb = (struct scsi_prevent *)ctsio->cdb;
5426
5427         if ((lun->flags & CTL_LUN_REMOVABLE) == 0 || lun->prevent == NULL) {
5428                 ctl_set_invalid_opcode(ctsio);
5429                 ctl_done((union ctl_io *)ctsio);
5430                 return (CTL_RETVAL_COMPLETE);
5431         }
5432
5433         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5434         mtx_lock(&lun->lun_lock);
5435         if ((cdb->how & PR_PREVENT) &&
5436             ctl_is_set(lun->prevent, initidx) == 0) {
5437                 ctl_set_mask(lun->prevent, initidx);
5438                 lun->prevent_count++;
5439         } else if ((cdb->how & PR_PREVENT) == 0 &&
5440             ctl_is_set(lun->prevent, initidx)) {
5441                 ctl_clear_mask(lun->prevent, initidx);
5442                 lun->prevent_count--;
5443         }
5444         mtx_unlock(&lun->lun_lock);
5445         retval = lun->backend->config_write((union ctl_io *)ctsio);
5446         return (retval);
5447 }
5448
5449 /*
5450  * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
5451  * we don't really do anything with the LBA and length fields if the user
5452  * passes them in.  Instead we'll just flush out the cache for the entire
5453  * LUN.
5454  */
5455 int
5456 ctl_sync_cache(struct ctl_scsiio *ctsio)
5457 {
5458         struct ctl_lun *lun = CTL_LUN(ctsio);
5459         struct ctl_lba_len_flags *lbalen;
5460         uint64_t starting_lba;
5461         uint32_t block_count;
5462         int retval;
5463         uint8_t byte2;
5464
5465         CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
5466
5467         retval = 0;
5468
5469         switch (ctsio->cdb[0]) {
5470         case SYNCHRONIZE_CACHE: {
5471                 struct scsi_sync_cache *cdb;
5472                 cdb = (struct scsi_sync_cache *)ctsio->cdb;
5473
5474                 starting_lba = scsi_4btoul(cdb->begin_lba);
5475                 block_count = scsi_2btoul(cdb->lb_count);
5476                 byte2 = cdb->byte2;
5477                 break;
5478         }
5479         case SYNCHRONIZE_CACHE_16: {
5480                 struct scsi_sync_cache_16 *cdb;
5481                 cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5482
5483                 starting_lba = scsi_8btou64(cdb->begin_lba);
5484                 block_count = scsi_4btoul(cdb->lb_count);
5485                 byte2 = cdb->byte2;
5486                 break;
5487         }
5488         default:
5489                 ctl_set_invalid_opcode(ctsio);
5490                 ctl_done((union ctl_io *)ctsio);
5491                 goto bailout;
5492                 break; /* NOTREACHED */
5493         }
5494
5495         /*
5496          * We check the LBA and length, but don't do anything with them.
5497          * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5498          * get flushed.  This check will just help satisfy anyone who wants
5499          * to see an error for an out of range LBA.
5500          */
5501         if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5502                 ctl_set_lba_out_of_range(ctsio,
5503                     MAX(starting_lba, lun->be_lun->maxlba + 1));
5504                 ctl_done((union ctl_io *)ctsio);
5505                 goto bailout;
5506         }
5507
5508         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5509         lbalen->lba = starting_lba;
5510         lbalen->len = block_count;
5511         lbalen->flags = byte2;
5512         retval = lun->backend->config_write((union ctl_io *)ctsio);
5513
5514 bailout:
5515         return (retval);
5516 }
5517
5518 int
5519 ctl_format(struct ctl_scsiio *ctsio)
5520 {
5521         struct scsi_format *cdb;
5522         int length, defect_list_len;
5523
5524         CTL_DEBUG_PRINT(("ctl_format\n"));
5525
5526         cdb = (struct scsi_format *)ctsio->cdb;
5527
5528         length = 0;
5529         if (cdb->byte2 & SF_FMTDATA) {
5530                 if (cdb->byte2 & SF_LONGLIST)
5531                         length = sizeof(struct scsi_format_header_long);
5532                 else
5533                         length = sizeof(struct scsi_format_header_short);
5534         }
5535
5536         if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5537          && (length > 0)) {
5538                 ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5539                 ctsio->kern_data_len = length;
5540                 ctsio->kern_total_len = length;
5541                 ctsio->kern_rel_offset = 0;
5542                 ctsio->kern_sg_entries = 0;
5543                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5544                 ctsio->be_move_done = ctl_config_move_done;
5545                 ctl_datamove((union ctl_io *)ctsio);
5546
5547                 return (CTL_RETVAL_COMPLETE);
5548         }
5549
5550         defect_list_len = 0;
5551
5552         if (cdb->byte2 & SF_FMTDATA) {
5553                 if (cdb->byte2 & SF_LONGLIST) {
5554                         struct scsi_format_header_long *header;
5555
5556                         header = (struct scsi_format_header_long *)
5557                                 ctsio->kern_data_ptr;
5558
5559                         defect_list_len = scsi_4btoul(header->defect_list_len);
5560                         if (defect_list_len != 0) {
5561                                 ctl_set_invalid_field(ctsio,
5562                                                       /*sks_valid*/ 1,
5563                                                       /*command*/ 0,
5564                                                       /*field*/ 2,
5565                                                       /*bit_valid*/ 0,
5566                                                       /*bit*/ 0);
5567                                 goto bailout;
5568                         }
5569                 } else {
5570                         struct scsi_format_header_short *header;
5571
5572                         header = (struct scsi_format_header_short *)
5573                                 ctsio->kern_data_ptr;
5574
5575                         defect_list_len = scsi_2btoul(header->defect_list_len);
5576                         if (defect_list_len != 0) {
5577                                 ctl_set_invalid_field(ctsio,
5578                                                       /*sks_valid*/ 1,
5579                                                       /*command*/ 0,
5580                                                       /*field*/ 2,
5581                                                       /*bit_valid*/ 0,
5582                                                       /*bit*/ 0);
5583                                 goto bailout;
5584                         }
5585                 }
5586         }
5587
5588         ctl_set_success(ctsio);
5589 bailout:
5590
5591         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5592                 free(ctsio->kern_data_ptr, M_CTL);
5593                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5594         }
5595
5596         ctl_done((union ctl_io *)ctsio);
5597         return (CTL_RETVAL_COMPLETE);
5598 }
5599
5600 int
5601 ctl_read_buffer(struct ctl_scsiio *ctsio)
5602 {
5603         struct ctl_lun *lun = CTL_LUN(ctsio);
5604         uint64_t buffer_offset;
5605         uint32_t len;
5606         uint8_t byte2;
5607         static uint8_t descr[4];
5608         static uint8_t echo_descr[4] = { 0 };
5609
5610         CTL_DEBUG_PRINT(("ctl_read_buffer\n"));
5611
5612         switch (ctsio->cdb[0]) {
5613         case READ_BUFFER: {
5614                 struct scsi_read_buffer *cdb;
5615
5616                 cdb = (struct scsi_read_buffer *)ctsio->cdb;
5617                 buffer_offset = scsi_3btoul(cdb->offset);
5618                 len = scsi_3btoul(cdb->length);
5619                 byte2 = cdb->byte2;
5620                 break;
5621         }
5622         case READ_BUFFER_16: {
5623                 struct scsi_read_buffer_16 *cdb;
5624
5625                 cdb = (struct scsi_read_buffer_16 *)ctsio->cdb;
5626                 buffer_offset = scsi_8btou64(cdb->offset);
5627                 len = scsi_4btoul(cdb->length);
5628                 byte2 = cdb->byte2;
5629                 break;
5630         }
5631         default: /* This shouldn't happen. */
5632                 ctl_set_invalid_opcode(ctsio);
5633                 ctl_done((union ctl_io *)ctsio);
5634                 return (CTL_RETVAL_COMPLETE);
5635         }
5636
5637         if (buffer_offset > CTL_WRITE_BUFFER_SIZE ||
5638             buffer_offset + len > CTL_WRITE_BUFFER_SIZE) {
5639                 ctl_set_invalid_field(ctsio,
5640                                       /*sks_valid*/ 1,
5641                                       /*command*/ 1,
5642                                       /*field*/ 6,
5643                                       /*bit_valid*/ 0,
5644                                       /*bit*/ 0);
5645                 ctl_done((union ctl_io *)ctsio);
5646                 return (CTL_RETVAL_COMPLETE);
5647         }
5648
5649         if ((byte2 & RWB_MODE) == RWB_MODE_DESCR) {
5650                 descr[0] = 0;
5651                 scsi_ulto3b(CTL_WRITE_BUFFER_SIZE, &descr[1]);
5652                 ctsio->kern_data_ptr = descr;
5653                 len = min(len, sizeof(descr));
5654         } else if ((byte2 & RWB_MODE) == RWB_MODE_ECHO_DESCR) {
5655                 ctsio->kern_data_ptr = echo_descr;
5656                 len = min(len, sizeof(echo_descr));
5657         } else {
5658                 if (lun->write_buffer == NULL) {
5659                         lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE,
5660                             M_CTL, M_WAITOK);
5661                 }
5662                 ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5663         }
5664         ctsio->kern_data_len = len;
5665         ctsio->kern_total_len = len;
5666         ctsio->kern_rel_offset = 0;
5667         ctsio->kern_sg_entries = 0;
5668         ctl_set_success(ctsio);
5669         ctsio->be_move_done = ctl_config_move_done;
5670         ctl_datamove((union ctl_io *)ctsio);
5671         return (CTL_RETVAL_COMPLETE);
5672 }
5673
5674 int
5675 ctl_write_buffer(struct ctl_scsiio *ctsio)
5676 {
5677         struct ctl_lun *lun = CTL_LUN(ctsio);
5678         struct scsi_write_buffer *cdb;
5679         int buffer_offset, len;
5680
5681         CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5682
5683         cdb = (struct scsi_write_buffer *)ctsio->cdb;
5684
5685         len = scsi_3btoul(cdb->length);
5686         buffer_offset = scsi_3btoul(cdb->offset);
5687
5688         if (buffer_offset + len > CTL_WRITE_BUFFER_SIZE) {
5689                 ctl_set_invalid_field(ctsio,
5690                                       /*sks_valid*/ 1,
5691                                       /*command*/ 1,
5692                                       /*field*/ 6,
5693                                       /*bit_valid*/ 0,
5694                                       /*bit*/ 0);
5695                 ctl_done((union ctl_io *)ctsio);
5696                 return (CTL_RETVAL_COMPLETE);
5697         }
5698
5699         /*
5700          * If we've got a kernel request that hasn't been malloced yet,
5701          * malloc it and tell the caller the data buffer is here.
5702          */
5703         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5704                 if (lun->write_buffer == NULL) {
5705                         lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE,
5706                             M_CTL, M_WAITOK);
5707                 }
5708                 ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5709                 ctsio->kern_data_len = len;
5710                 ctsio->kern_total_len = len;
5711                 ctsio->kern_rel_offset = 0;
5712                 ctsio->kern_sg_entries = 0;
5713                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5714                 ctsio->be_move_done = ctl_config_move_done;
5715                 ctl_datamove((union ctl_io *)ctsio);
5716
5717                 return (CTL_RETVAL_COMPLETE);
5718         }
5719
5720         ctl_set_success(ctsio);
5721         ctl_done((union ctl_io *)ctsio);
5722         return (CTL_RETVAL_COMPLETE);
5723 }
5724
5725 int
5726 ctl_write_same(struct ctl_scsiio *ctsio)
5727 {
5728         struct ctl_lun *lun = CTL_LUN(ctsio);
5729         struct ctl_lba_len_flags *lbalen;
5730         uint64_t lba;
5731         uint32_t num_blocks;
5732         int len, retval;
5733         uint8_t byte2;
5734
5735         CTL_DEBUG_PRINT(("ctl_write_same\n"));
5736
5737         switch (ctsio->cdb[0]) {
5738         case WRITE_SAME_10: {
5739                 struct scsi_write_same_10 *cdb;
5740
5741                 cdb = (struct scsi_write_same_10 *)ctsio->cdb;
5742
5743                 lba = scsi_4btoul(cdb->addr);
5744                 num_blocks = scsi_2btoul(cdb->length);
5745                 byte2 = cdb->byte2;
5746                 break;
5747         }
5748         case WRITE_SAME_16: {
5749                 struct scsi_write_same_16 *cdb;
5750
5751                 cdb = (struct scsi_write_same_16 *)ctsio->cdb;
5752
5753                 lba = scsi_8btou64(cdb->addr);
5754                 num_blocks = scsi_4btoul(cdb->length);
5755                 byte2 = cdb->byte2;
5756                 break;
5757         }
5758         default:
5759                 /*
5760                  * We got a command we don't support.  This shouldn't
5761                  * happen, commands should be filtered out above us.
5762                  */
5763                 ctl_set_invalid_opcode(ctsio);
5764                 ctl_done((union ctl_io *)ctsio);
5765
5766                 return (CTL_RETVAL_COMPLETE);
5767                 break; /* NOTREACHED */
5768         }
5769
5770         /* ANCHOR flag can be used only together with UNMAP */
5771         if ((byte2 & SWS_UNMAP) == 0 && (byte2 & SWS_ANCHOR) != 0) {
5772                 ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
5773                     /*command*/ 1, /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 0);
5774                 ctl_done((union ctl_io *)ctsio);
5775                 return (CTL_RETVAL_COMPLETE);
5776         }
5777
5778         /*
5779          * The first check is to make sure we're in bounds, the second
5780          * check is to catch wrap-around problems.  If the lba + num blocks
5781          * is less than the lba, then we've wrapped around and the block
5782          * range is invalid anyway.
5783          */
5784         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5785          || ((lba + num_blocks) < lba)) {
5786                 ctl_set_lba_out_of_range(ctsio,
5787                     MAX(lba, lun->be_lun->maxlba + 1));
5788                 ctl_done((union ctl_io *)ctsio);
5789                 return (CTL_RETVAL_COMPLETE);
5790         }
5791
5792         /* Zero number of blocks means "to the last logical block" */
5793         if (num_blocks == 0) {
5794                 if ((lun->be_lun->maxlba + 1) - lba > UINT32_MAX) {
5795                         ctl_set_invalid_field(ctsio,
5796                                               /*sks_valid*/ 0,
5797                                               /*command*/ 1,
5798                                               /*field*/ 0,
5799                                               /*bit_valid*/ 0,
5800                                               /*bit*/ 0);
5801                         ctl_done((union ctl_io *)ctsio);
5802                         return (CTL_RETVAL_COMPLETE);
5803                 }
5804                 num_blocks = (lun->be_lun->maxlba + 1) - lba;
5805         }
5806
5807         len = lun->be_lun->blocksize;
5808
5809         /*
5810          * If we've got a kernel request that hasn't been malloced yet,
5811          * malloc it and tell the caller the data buffer is here.
5812          */
5813         if ((byte2 & SWS_NDOB) == 0 &&
5814             (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5815                 ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);
5816                 ctsio->kern_data_len = len;
5817                 ctsio->kern_total_len = len;
5818                 ctsio->kern_rel_offset = 0;
5819                 ctsio->kern_sg_entries = 0;
5820                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5821                 ctsio->be_move_done = ctl_config_move_done;
5822                 ctl_datamove((union ctl_io *)ctsio);
5823
5824                 return (CTL_RETVAL_COMPLETE);
5825         }
5826
5827         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5828         lbalen->lba = lba;
5829         lbalen->len = num_blocks;
5830         lbalen->flags = byte2;
5831         retval = lun->backend->config_write((union ctl_io *)ctsio);
5832
5833         return (retval);
5834 }
5835
5836 int
5837 ctl_unmap(struct ctl_scsiio *ctsio)
5838 {
5839         struct ctl_lun *lun = CTL_LUN(ctsio);
5840         struct scsi_unmap *cdb;
5841         struct ctl_ptr_len_flags *ptrlen;
5842         struct scsi_unmap_header *hdr;
5843         struct scsi_unmap_desc *buf, *end, *endnz, *range;
5844         uint64_t lba;
5845         uint32_t num_blocks;
5846         int len, retval;
5847         uint8_t byte2;
5848
5849         CTL_DEBUG_PRINT(("ctl_unmap\n"));
5850
5851         cdb = (struct scsi_unmap *)ctsio->cdb;
5852         len = scsi_2btoul(cdb->length);
5853         byte2 = cdb->byte2;
5854
5855         /*
5856          * If we've got a kernel request that hasn't been malloced yet,
5857          * malloc it and tell the caller the data buffer is here.
5858          */
5859         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5860                 ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);
5861                 ctsio->kern_data_len = len;
5862                 ctsio->kern_total_len = len;
5863                 ctsio->kern_rel_offset = 0;
5864                 ctsio->kern_sg_entries = 0;
5865                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5866                 ctsio->be_move_done = ctl_config_move_done;
5867                 ctl_datamove((union ctl_io *)ctsio);
5868
5869                 return (CTL_RETVAL_COMPLETE);
5870         }
5871
5872         len = ctsio->kern_total_len - ctsio->kern_data_resid;
5873         hdr = (struct scsi_unmap_header *)ctsio->kern_data_ptr;
5874         if (len < sizeof (*hdr) ||
5875             len < (scsi_2btoul(hdr->length) + sizeof(hdr->length)) ||
5876             len < (scsi_2btoul(hdr->desc_length) + sizeof (*hdr)) ||
5877             scsi_2btoul(hdr->desc_length) % sizeof(*buf) != 0) {
5878                 ctl_set_invalid_field(ctsio,
5879                                       /*sks_valid*/ 0,
5880                                       /*command*/ 0,
5881                                       /*field*/ 0,
5882                                       /*bit_valid*/ 0,
5883                                       /*bit*/ 0);
5884                 goto done;
5885         }
5886         len = scsi_2btoul(hdr->desc_length);
5887         buf = (struct scsi_unmap_desc *)(hdr + 1);
5888         end = buf + len / sizeof(*buf);
5889
5890         endnz = buf;
5891         for (range = buf; range < end; range++) {
5892                 lba = scsi_8btou64(range->lba);
5893                 num_blocks = scsi_4btoul(range->length);
5894                 if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5895                  || ((lba + num_blocks) < lba)) {
5896                         ctl_set_lba_out_of_range(ctsio,
5897                             MAX(lba, lun->be_lun->maxlba + 1));
5898                         ctl_done((union ctl_io *)ctsio);
5899                         return (CTL_RETVAL_COMPLETE);
5900                 }
5901                 if (num_blocks != 0)
5902                         endnz = range + 1;
5903         }
5904
5905         /*
5906          * Block backend can not handle zero last range.
5907          * Filter it out and return if there is nothing left.
5908          */
5909         len = (uint8_t *)endnz - (uint8_t *)buf;
5910         if (len == 0) {
5911                 ctl_set_success(ctsio);
5912                 goto done;
5913         }
5914
5915         mtx_lock(&lun->lun_lock);
5916         ptrlen = (struct ctl_ptr_len_flags *)
5917             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5918         ptrlen->ptr = (void *)buf;
5919         ptrlen->len = len;
5920         ptrlen->flags = byte2;
5921         ctl_check_blocked(lun);
5922         mtx_unlock(&lun->lun_lock);
5923
5924         retval = lun->backend->config_write((union ctl_io *)ctsio);
5925         return (retval);
5926
5927 done:
5928         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5929                 free(ctsio->kern_data_ptr, M_CTL);
5930                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5931         }
5932         ctl_done((union ctl_io *)ctsio);
5933         return (CTL_RETVAL_COMPLETE);
5934 }
5935
5936 int
5937 ctl_default_page_handler(struct ctl_scsiio *ctsio,
5938                          struct ctl_page_index *page_index, uint8_t *page_ptr)
5939 {
5940         struct ctl_lun *lun = CTL_LUN(ctsio);
5941         uint8_t *current_cp;
5942         int set_ua;
5943         uint32_t initidx;
5944
5945         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5946         set_ua = 0;
5947
5948         current_cp = (page_index->page_data + (page_index->page_len *
5949             CTL_PAGE_CURRENT));
5950
5951         mtx_lock(&lun->lun_lock);
5952         if (memcmp(current_cp, page_ptr, page_index->page_len)) {
5953                 memcpy(current_cp, page_ptr, page_index->page_len);
5954                 set_ua = 1;
5955         }
5956         if (set_ua != 0)
5957                 ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
5958         mtx_unlock(&lun->lun_lock);
5959         if (set_ua) {
5960                 ctl_isc_announce_mode(lun,
5961                     ctl_get_initindex(&ctsio->io_hdr.nexus),
5962                     page_index->page_code, page_index->subpage);
5963         }
5964         return (CTL_RETVAL_COMPLETE);
5965 }
5966
5967 static void
5968 ctl_ie_timer(void *arg)
5969 {
5970         struct ctl_lun *lun = arg;
5971         uint64_t t;
5972
5973         if (lun->ie_asc == 0)
5974                 return;
5975
5976         if (lun->MODE_IE.mrie == SIEP_MRIE_UA)
5977                 ctl_est_ua_all(lun, -1, CTL_UA_IE);
5978         else
5979                 lun->ie_reported = 0;
5980
5981         if (lun->ie_reportcnt < scsi_4btoul(lun->MODE_IE.report_count)) {
5982                 lun->ie_reportcnt++;
5983                 t = scsi_4btoul(lun->MODE_IE.interval_timer);
5984                 if (t == 0 || t == UINT32_MAX)
5985                         t = 3000;  /* 5 min */
5986                 callout_schedule(&lun->ie_callout, t * hz / 10);
5987         }
5988 }
5989
5990 int
5991 ctl_ie_page_handler(struct ctl_scsiio *ctsio,
5992                          struct ctl_page_index *page_index, uint8_t *page_ptr)
5993 {
5994         struct ctl_lun *lun = CTL_LUN(ctsio);
5995         struct scsi_info_exceptions_page *pg;
5996         uint64_t t;
5997
5998         (void)ctl_default_page_handler(ctsio, page_index, page_ptr);
5999
6000         pg = (struct scsi_info_exceptions_page *)page_ptr;
6001         mtx_lock(&lun->lun_lock);
6002         if (pg->info_flags & SIEP_FLAGS_TEST) {
6003                 lun->ie_asc = 0x5d;
6004                 lun->ie_ascq = 0xff;
6005                 if (pg->mrie == SIEP_MRIE_UA) {
6006                         ctl_est_ua_all(lun, -1, CTL_UA_IE);
6007                         lun->ie_reported = 1;
6008                 } else {
6009                         ctl_clr_ua_all(lun, -1, CTL_UA_IE);
6010                         lun->ie_reported = -1;
6011                 }
6012                 lun->ie_reportcnt = 1;
6013                 if (lun->ie_reportcnt < scsi_4btoul(pg->report_count)) {
6014                         lun->ie_reportcnt++;
6015                         t = scsi_4btoul(pg->interval_timer);
6016                         if (t == 0 || t == UINT32_MAX)
6017                                 t = 3000;  /* 5 min */
6018                         callout_reset(&lun->ie_callout, t * hz / 10,
6019                             ctl_ie_timer, lun);
6020                 }
6021         } else {
6022                 lun->ie_asc = 0;
6023                 lun->ie_ascq = 0;
6024                 lun->ie_reported = 1;
6025                 ctl_clr_ua_all(lun, -1, CTL_UA_IE);
6026                 lun->ie_reportcnt = UINT32_MAX;
6027                 callout_stop(&lun->ie_callout);
6028         }
6029         mtx_unlock(&lun->lun_lock);
6030         return (CTL_RETVAL_COMPLETE);
6031 }
6032
6033 static int
6034 ctl_do_mode_select(union ctl_io *io)
6035 {
6036         struct ctl_lun *lun = CTL_LUN(io);
6037         struct scsi_mode_page_header *page_header;
6038         struct ctl_page_index *page_index;
6039         struct ctl_scsiio *ctsio;
6040         int page_len, page_len_offset, page_len_size;
6041         union ctl_modepage_info *modepage_info;
6042         uint16_t *len_left, *len_used;
6043         int retval, i;
6044
6045         ctsio = &io->scsiio;
6046         page_index = NULL;
6047         page_len = 0;
6048
6049         modepage_info = (union ctl_modepage_info *)
6050                 ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6051         len_left = &modepage_info->header.len_left;
6052         len_used = &modepage_info->header.len_used;
6053
6054 do_next_page:
6055
6056         page_header = (struct scsi_mode_page_header *)
6057                 (ctsio->kern_data_ptr + *len_used);
6058
6059         if (*len_left == 0) {
6060                 free(ctsio->kern_data_ptr, M_CTL);
6061                 ctl_set_success(ctsio);
6062                 ctl_done((union ctl_io *)ctsio);
6063                 return (CTL_RETVAL_COMPLETE);
6064         } else if (*len_left < sizeof(struct scsi_mode_page_header)) {
6065
6066                 free(ctsio->kern_data_ptr, M_CTL);
6067                 ctl_set_param_len_error(ctsio);
6068                 ctl_done((union ctl_io *)ctsio);
6069                 return (CTL_RETVAL_COMPLETE);
6070
6071         } else if ((page_header->page_code & SMPH_SPF)
6072                 && (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
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
6080
6081         /*
6082          * XXX KDM should we do something with the block descriptor?
6083          */
6084         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6085                 page_index = &lun->mode_pages.index[i];
6086                 if (lun->be_lun->lun_type == T_DIRECT &&
6087                     (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6088                         continue;
6089                 if (lun->be_lun->lun_type == T_PROCESSOR &&
6090                     (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6091                         continue;
6092                 if (lun->be_lun->lun_type == T_CDROM &&
6093                     (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6094                         continue;
6095
6096                 if ((page_index->page_code & SMPH_PC_MASK) !=
6097                     (page_header->page_code & SMPH_PC_MASK))
6098                         continue;
6099
6100                 /*
6101                  * If neither page has a subpage code, then we've got a
6102                  * match.
6103                  */
6104                 if (((page_index->page_code & SMPH_SPF) == 0)
6105                  && ((page_header->page_code & SMPH_SPF) == 0)) {
6106                         page_len = page_header->page_length;
6107                         break;
6108                 }
6109
6110                 /*
6111                  * If both pages have subpages, then the subpage numbers
6112                  * have to match.
6113                  */
6114                 if ((page_index->page_code & SMPH_SPF)
6115                   && (page_header->page_code & SMPH_SPF)) {
6116                         struct scsi_mode_page_header_sp *sph;
6117
6118                         sph = (struct scsi_mode_page_header_sp *)page_header;
6119                         if (page_index->subpage == sph->subpage) {
6120                                 page_len = scsi_2btoul(sph->page_length);
6121                                 break;
6122                         }
6123                 }
6124         }
6125
6126         /*
6127          * If we couldn't find the page, or if we don't have a mode select
6128          * handler for it, send back an error to the user.
6129          */
6130         if ((i >= CTL_NUM_MODE_PAGES)
6131          || (page_index->select_handler == NULL)) {
6132                 ctl_set_invalid_field(ctsio,
6133                                       /*sks_valid*/ 1,
6134                                       /*command*/ 0,
6135                                       /*field*/ *len_used,
6136                                       /*bit_valid*/ 0,
6137                                       /*bit*/ 0);
6138                 free(ctsio->kern_data_ptr, M_CTL);
6139                 ctl_done((union ctl_io *)ctsio);
6140                 return (CTL_RETVAL_COMPLETE);
6141         }
6142
6143         if (page_index->page_code & SMPH_SPF) {
6144                 page_len_offset = 2;
6145                 page_len_size = 2;
6146         } else {
6147                 page_len_size = 1;
6148                 page_len_offset = 1;
6149         }
6150
6151         /*
6152          * If the length the initiator gives us isn't the one we specify in
6153          * the mode page header, or if they didn't specify enough data in
6154          * the CDB to avoid truncating this page, kick out the request.
6155          */
6156         if (page_len != page_index->page_len - page_len_offset - page_len_size) {
6157                 ctl_set_invalid_field(ctsio,
6158                                       /*sks_valid*/ 1,
6159                                       /*command*/ 0,
6160                                       /*field*/ *len_used + page_len_offset,
6161                                       /*bit_valid*/ 0,
6162                                       /*bit*/ 0);
6163                 free(ctsio->kern_data_ptr, M_CTL);
6164                 ctl_done((union ctl_io *)ctsio);
6165                 return (CTL_RETVAL_COMPLETE);
6166         }
6167         if (*len_left < page_index->page_len) {
6168                 free(ctsio->kern_data_ptr, M_CTL);
6169                 ctl_set_param_len_error(ctsio);
6170                 ctl_done((union ctl_io *)ctsio);
6171                 return (CTL_RETVAL_COMPLETE);
6172         }
6173
6174         /*
6175          * Run through the mode page, checking to make sure that the bits
6176          * the user changed are actually legal for him to change.
6177          */
6178         for (i = 0; i < page_index->page_len; i++) {
6179                 uint8_t *user_byte, *change_mask, *current_byte;
6180                 int bad_bit;
6181                 int j;
6182
6183                 user_byte = (uint8_t *)page_header + i;
6184                 change_mask = page_index->page_data +
6185                               (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
6186                 current_byte = page_index->page_data +
6187                                (page_index->page_len * CTL_PAGE_CURRENT) + i;
6188
6189                 /*
6190                  * Check to see whether the user set any bits in this byte
6191                  * that he is not allowed to set.
6192                  */
6193                 if ((*user_byte & ~(*change_mask)) ==
6194                     (*current_byte & ~(*change_mask)))
6195                         continue;
6196
6197                 /*
6198                  * Go through bit by bit to determine which one is illegal.
6199                  */
6200                 bad_bit = 0;
6201                 for (j = 7; j >= 0; j--) {
6202                         if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
6203                             (((1 << i) & ~(*change_mask)) & *current_byte)) {
6204                                 bad_bit = i;
6205                                 break;
6206                         }
6207                 }
6208                 ctl_set_invalid_field(ctsio,
6209                                       /*sks_valid*/ 1,
6210                                       /*command*/ 0,
6211                                       /*field*/ *len_used + i,
6212                                       /*bit_valid*/ 1,
6213                                       /*bit*/ bad_bit);
6214                 free(ctsio->kern_data_ptr, M_CTL);
6215                 ctl_done((union ctl_io *)ctsio);
6216                 return (CTL_RETVAL_COMPLETE);
6217         }
6218
6219         /*
6220          * Decrement these before we call the page handler, since we may
6221          * end up getting called back one way or another before the handler
6222          * returns to this context.
6223          */
6224         *len_left -= page_index->page_len;
6225         *len_used += page_index->page_len;
6226
6227         retval = page_index->select_handler(ctsio, page_index,
6228                                             (uint8_t *)page_header);
6229
6230         /*
6231          * If the page handler returns CTL_RETVAL_QUEUED, then we need to
6232          * wait until this queued command completes to finish processing
6233          * the mode page.  If it returns anything other than
6234          * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
6235          * already set the sense information, freed the data pointer, and
6236          * completed the io for us.
6237          */
6238         if (retval != CTL_RETVAL_COMPLETE)
6239                 goto bailout_no_done;
6240
6241         /*
6242          * If the initiator sent us more than one page, parse the next one.
6243          */
6244         if (*len_left > 0)
6245                 goto do_next_page;
6246
6247         ctl_set_success(ctsio);
6248         free(ctsio->kern_data_ptr, M_CTL);
6249         ctl_done((union ctl_io *)ctsio);
6250
6251 bailout_no_done:
6252
6253         return (CTL_RETVAL_COMPLETE);
6254
6255 }
6256
6257 int
6258 ctl_mode_select(struct ctl_scsiio *ctsio)
6259 {
6260         struct ctl_lun *lun = CTL_LUN(ctsio);
6261         union ctl_modepage_info *modepage_info;
6262         int bd_len, i, header_size, param_len, pf, rtd, sp;
6263         uint32_t initidx;
6264
6265         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
6266         switch (ctsio->cdb[0]) {
6267         case MODE_SELECT_6: {
6268                 struct scsi_mode_select_6 *cdb;
6269
6270                 cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
6271
6272                 pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6273                 rtd = (cdb->byte2 & SMS_RTD) ? 1 : 0;
6274                 sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6275                 param_len = cdb->length;
6276                 header_size = sizeof(struct scsi_mode_header_6);
6277                 break;
6278         }
6279         case MODE_SELECT_10: {
6280                 struct scsi_mode_select_10 *cdb;
6281
6282                 cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
6283
6284                 pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6285                 rtd = (cdb->byte2 & SMS_RTD) ? 1 : 0;
6286                 sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6287                 param_len = scsi_2btoul(cdb->length);
6288                 header_size = sizeof(struct scsi_mode_header_10);
6289                 break;
6290         }
6291         default:
6292                 ctl_set_invalid_opcode(ctsio);
6293                 ctl_done((union ctl_io *)ctsio);
6294                 return (CTL_RETVAL_COMPLETE);
6295         }
6296
6297         if (rtd) {
6298                 if (param_len != 0) {
6299                         ctl_set_invalid_field(ctsio, /*sks_valid*/ 0,
6300                             /*command*/ 1, /*field*/ 0,
6301                             /*bit_valid*/ 0, /*bit*/ 0);
6302                         ctl_done((union ctl_io *)ctsio);
6303                         return (CTL_RETVAL_COMPLETE);
6304                 }
6305
6306                 /* Revert to defaults. */
6307                 ctl_init_page_index(lun);
6308                 mtx_lock(&lun->lun_lock);
6309                 ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
6310                 mtx_unlock(&lun->lun_lock);
6311                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6312                         ctl_isc_announce_mode(lun, -1,
6313                             lun->mode_pages.index[i].page_code & SMPH_PC_MASK,
6314                             lun->mode_pages.index[i].subpage);
6315                 }
6316                 ctl_set_success(ctsio);
6317                 ctl_done((union ctl_io *)ctsio);
6318                 return (CTL_RETVAL_COMPLETE);
6319         }
6320
6321         /*
6322          * From SPC-3:
6323          * "A parameter list length of zero indicates that the Data-Out Buffer
6324          * shall be empty. This condition shall not be considered as an error."
6325          */
6326         if (param_len == 0) {
6327                 ctl_set_success(ctsio);
6328                 ctl_done((union ctl_io *)ctsio);
6329                 return (CTL_RETVAL_COMPLETE);
6330         }
6331
6332         /*
6333          * Since we'll hit this the first time through, prior to
6334          * allocation, we don't need to free a data buffer here.
6335          */
6336         if (param_len < header_size) {
6337                 ctl_set_param_len_error(ctsio);
6338                 ctl_done((union ctl_io *)ctsio);
6339                 return (CTL_RETVAL_COMPLETE);
6340         }
6341
6342         /*
6343          * Allocate the data buffer and grab the user's data.  In theory,
6344          * we shouldn't have to sanity check the parameter list length here
6345          * because the maximum size is 64K.  We should be able to malloc
6346          * that much without too many problems.
6347          */
6348         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6349                 ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6350                 ctsio->kern_data_len = param_len;
6351                 ctsio->kern_total_len = param_len;
6352                 ctsio->kern_rel_offset = 0;
6353                 ctsio->kern_sg_entries = 0;
6354                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6355                 ctsio->be_move_done = ctl_config_move_done;
6356                 ctl_datamove((union ctl_io *)ctsio);
6357
6358                 return (CTL_RETVAL_COMPLETE);
6359         }
6360
6361         switch (ctsio->cdb[0]) {
6362         case MODE_SELECT_6: {
6363                 struct scsi_mode_header_6 *mh6;
6364
6365                 mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6366                 bd_len = mh6->blk_desc_len;
6367                 break;
6368         }
6369         case MODE_SELECT_10: {
6370                 struct scsi_mode_header_10 *mh10;
6371
6372                 mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6373                 bd_len = scsi_2btoul(mh10->blk_desc_len);
6374                 break;
6375         }
6376         default:
6377                 panic("%s: Invalid CDB type %#x", __func__, ctsio->cdb[0]);
6378         }
6379
6380         if (param_len < (header_size + bd_len)) {
6381                 free(ctsio->kern_data_ptr, M_CTL);
6382                 ctl_set_param_len_error(ctsio);
6383                 ctl_done((union ctl_io *)ctsio);
6384                 return (CTL_RETVAL_COMPLETE);
6385         }
6386
6387         /*
6388          * Set the IO_CONT flag, so that if this I/O gets passed to
6389          * ctl_config_write_done(), it'll get passed back to
6390          * ctl_do_mode_select() for further processing, or completion if
6391          * we're all done.
6392          */
6393         ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6394         ctsio->io_cont = ctl_do_mode_select;
6395
6396         modepage_info = (union ctl_modepage_info *)
6397                 ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6398         memset(modepage_info, 0, sizeof(*modepage_info));
6399         modepage_info->header.len_left = param_len - header_size - bd_len;
6400         modepage_info->header.len_used = header_size + bd_len;
6401
6402         return (ctl_do_mode_select((union ctl_io *)ctsio));
6403 }
6404
6405 int
6406 ctl_mode_sense(struct ctl_scsiio *ctsio)
6407 {
6408         struct ctl_lun *lun = CTL_LUN(ctsio);
6409         int pc, page_code, dbd, llba, subpage;
6410         int alloc_len, page_len, header_len, total_len;
6411         struct scsi_mode_block_descr *block_desc;
6412         struct ctl_page_index *page_index;
6413
6414         dbd = 0;
6415         llba = 0;
6416         block_desc = NULL;
6417
6418         CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6419
6420         switch (ctsio->cdb[0]) {
6421         case MODE_SENSE_6: {
6422                 struct scsi_mode_sense_6 *cdb;
6423
6424                 cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6425
6426                 header_len = sizeof(struct scsi_mode_hdr_6);
6427                 if (cdb->byte2 & SMS_DBD)
6428                         dbd = 1;
6429                 else
6430                         header_len += sizeof(struct scsi_mode_block_descr);
6431
6432                 pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6433                 page_code = cdb->page & SMS_PAGE_CODE;
6434                 subpage = cdb->subpage;
6435                 alloc_len = cdb->length;
6436                 break;
6437         }
6438         case MODE_SENSE_10: {
6439                 struct scsi_mode_sense_10 *cdb;
6440
6441                 cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6442
6443                 header_len = sizeof(struct scsi_mode_hdr_10);
6444
6445                 if (cdb->byte2 & SMS_DBD)
6446                         dbd = 1;
6447                 else
6448                         header_len += sizeof(struct scsi_mode_block_descr);
6449                 if (cdb->byte2 & SMS10_LLBAA)
6450                         llba = 1;
6451                 pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6452                 page_code = cdb->page & SMS_PAGE_CODE;
6453                 subpage = cdb->subpage;
6454                 alloc_len = scsi_2btoul(cdb->length);
6455                 break;
6456         }
6457         default:
6458                 ctl_set_invalid_opcode(ctsio);
6459                 ctl_done((union ctl_io *)ctsio);
6460                 return (CTL_RETVAL_COMPLETE);
6461                 break; /* NOTREACHED */
6462         }
6463
6464         /*
6465          * We have to make a first pass through to calculate the size of
6466          * the pages that match the user's query.  Then we allocate enough
6467          * memory to hold it, and actually copy the data into the buffer.
6468          */
6469         switch (page_code) {
6470         case SMS_ALL_PAGES_PAGE: {
6471                 u_int i;
6472
6473                 page_len = 0;
6474
6475                 /*
6476                  * At the moment, values other than 0 and 0xff here are
6477                  * reserved according to SPC-3.
6478                  */
6479                 if ((subpage != SMS_SUBPAGE_PAGE_0)
6480                  && (subpage != SMS_SUBPAGE_ALL)) {
6481                         ctl_set_invalid_field(ctsio,
6482                                               /*sks_valid*/ 1,
6483                                               /*command*/ 1,
6484                                               /*field*/ 3,
6485                                               /*bit_valid*/ 0,
6486                                               /*bit*/ 0);
6487                         ctl_done((union ctl_io *)ctsio);
6488                         return (CTL_RETVAL_COMPLETE);
6489                 }
6490
6491                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6492                         page_index = &lun->mode_pages.index[i];
6493
6494                         /* Make sure the page is supported for this dev type */
6495                         if (lun->be_lun->lun_type == T_DIRECT &&
6496                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6497                                 continue;
6498                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6499                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6500                                 continue;
6501                         if (lun->be_lun->lun_type == T_CDROM &&
6502                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6503                                 continue;
6504
6505                         /*
6506                          * We don't use this subpage if the user didn't
6507                          * request all subpages.
6508                          */
6509                         if ((page_index->subpage != 0)
6510                          && (subpage == SMS_SUBPAGE_PAGE_0))
6511                                 continue;
6512
6513 #if 0
6514                         printf("found page %#x len %d\n",
6515                                page_index->page_code & SMPH_PC_MASK,
6516                                page_index->page_len);
6517 #endif
6518                         page_len += page_index->page_len;
6519                 }
6520                 break;
6521         }
6522         default: {
6523                 u_int i;
6524
6525                 page_len = 0;
6526
6527                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6528                         page_index = &lun->mode_pages.index[i];
6529
6530                         /* Make sure the page is supported for this dev type */
6531                         if (lun->be_lun->lun_type == T_DIRECT &&
6532                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6533                                 continue;
6534                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6535                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6536                                 continue;
6537                         if (lun->be_lun->lun_type == T_CDROM &&
6538                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6539                                 continue;
6540
6541                         /* Look for the right page code */
6542                         if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6543                                 continue;
6544
6545                         /* Look for the right subpage or the subpage wildcard*/
6546                         if ((page_index->subpage != subpage)
6547                          && (subpage != SMS_SUBPAGE_ALL))
6548                                 continue;
6549
6550 #if 0
6551                         printf("found page %#x len %d\n",
6552                                page_index->page_code & SMPH_PC_MASK,
6553                                page_index->page_len);
6554 #endif
6555
6556                         page_len += page_index->page_len;
6557                 }
6558
6559                 if (page_len == 0) {
6560                         ctl_set_invalid_field(ctsio,
6561                                               /*sks_valid*/ 1,
6562                                               /*command*/ 1,
6563                                               /*field*/ 2,
6564                                               /*bit_valid*/ 1,
6565                                               /*bit*/ 5);
6566                         ctl_done((union ctl_io *)ctsio);
6567                         return (CTL_RETVAL_COMPLETE);
6568                 }
6569                 break;
6570         }
6571         }
6572
6573         total_len = header_len + page_len;
6574 #if 0
6575         printf("header_len = %d, page_len = %d, total_len = %d\n",
6576                header_len, page_len, total_len);
6577 #endif
6578
6579         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6580         ctsio->kern_sg_entries = 0;
6581         ctsio->kern_rel_offset = 0;
6582         ctsio->kern_data_len = min(total_len, alloc_len);
6583         ctsio->kern_total_len = ctsio->kern_data_len;
6584
6585         switch (ctsio->cdb[0]) {
6586         case MODE_SENSE_6: {
6587                 struct scsi_mode_hdr_6 *header;
6588
6589                 header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
6590
6591                 header->datalen = MIN(total_len - 1, 254);
6592                 if (lun->be_lun->lun_type == T_DIRECT) {
6593                         header->dev_specific = 0x10; /* DPOFUA */
6594                         if ((lun->be_lun->flags & CTL_LUN_FLAG_READONLY) ||
6595                             (lun->MODE_CTRL.eca_and_aen & SCP_SWP) != 0)
6596                                 header->dev_specific |= 0x80; /* WP */
6597                 }
6598                 if (dbd)
6599                         header->block_descr_len = 0;
6600                 else
6601                         header->block_descr_len =
6602                                 sizeof(struct scsi_mode_block_descr);
6603                 block_desc = (struct scsi_mode_block_descr *)&header[1];
6604                 break;
6605         }
6606         case MODE_SENSE_10: {
6607                 struct scsi_mode_hdr_10 *header;
6608                 int datalen;
6609
6610                 header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
6611
6612                 datalen = MIN(total_len - 2, 65533);
6613                 scsi_ulto2b(datalen, header->datalen);
6614                 if (lun->be_lun->lun_type == T_DIRECT) {
6615                         header->dev_specific = 0x10; /* DPOFUA */
6616                         if ((lun->be_lun->flags & CTL_LUN_FLAG_READONLY) ||
6617                             (lun->MODE_CTRL.eca_and_aen & SCP_SWP) != 0)
6618                                 header->dev_specific |= 0x80; /* WP */
6619                 }
6620                 if (dbd)
6621                         scsi_ulto2b(0, header->block_descr_len);
6622                 else
6623                         scsi_ulto2b(sizeof(struct scsi_mode_block_descr),
6624                                     header->block_descr_len);
6625                 block_desc = (struct scsi_mode_block_descr *)&header[1];
6626                 break;
6627         }
6628         default:
6629                 panic("%s: Invalid CDB type %#x", __func__, ctsio->cdb[0]);
6630         }
6631
6632         /*
6633          * If we've got a disk, use its blocksize in the block
6634          * descriptor.  Otherwise, just set it to 0.
6635          */
6636         if (dbd == 0) {
6637                 if (lun->be_lun->lun_type == T_DIRECT)
6638                         scsi_ulto3b(lun->be_lun->blocksize,
6639                                     block_desc->block_len);
6640                 else
6641                         scsi_ulto3b(0, block_desc->block_len);
6642         }
6643
6644         switch (page_code) {
6645         case SMS_ALL_PAGES_PAGE: {
6646                 int i, data_used;
6647
6648                 data_used = header_len;
6649                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6650                         struct ctl_page_index *page_index;
6651
6652                         page_index = &lun->mode_pages.index[i];
6653                         if (lun->be_lun->lun_type == T_DIRECT &&
6654                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6655                                 continue;
6656                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6657                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6658                                 continue;
6659                         if (lun->be_lun->lun_type == T_CDROM &&
6660                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6661                                 continue;
6662
6663                         /*
6664                          * We don't use this subpage if the user didn't
6665                          * request all subpages.  We already checked (above)
6666                          * to make sure the user only specified a subpage
6667                          * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case.
6668                          */
6669                         if ((page_index->subpage != 0)
6670                          && (subpage == SMS_SUBPAGE_PAGE_0))
6671                                 continue;
6672
6673                         /*
6674                          * Call the handler, if it exists, to update the
6675                          * page to the latest values.
6676                          */
6677                         if (page_index->sense_handler != NULL)
6678                                 page_index->sense_handler(ctsio, page_index,pc);
6679
6680                         memcpy(ctsio->kern_data_ptr + data_used,
6681                                page_index->page_data +
6682                                (page_index->page_len * pc),
6683                                page_index->page_len);
6684                         data_used += page_index->page_len;
6685                 }
6686                 break;
6687         }
6688         default: {
6689                 int i, data_used;
6690
6691                 data_used = header_len;
6692
6693                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6694                         struct ctl_page_index *page_index;
6695
6696                         page_index = &lun->mode_pages.index[i];
6697
6698                         /* Look for the right page code */
6699                         if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6700                                 continue;
6701
6702                         /* Look for the right subpage or the subpage wildcard*/
6703                         if ((page_index->subpage != subpage)
6704                          && (subpage != SMS_SUBPAGE_ALL))
6705                                 continue;
6706
6707                         /* Make sure the page is supported for this dev type */
6708                         if (lun->be_lun->lun_type == T_DIRECT &&
6709                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6710                                 continue;
6711                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6712                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6713                                 continue;
6714                         if (lun->be_lun->lun_type == T_CDROM &&
6715                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6716                                 continue;
6717
6718                         /*
6719                          * Call the handler, if it exists, to update the
6720                          * page to the latest values.
6721                          */
6722                         if (page_index->sense_handler != NULL)
6723                                 page_index->sense_handler(ctsio, page_index,pc);
6724
6725                         memcpy(ctsio->kern_data_ptr + data_used,
6726                                page_index->page_data +
6727                                (page_index->page_len * pc),
6728                                page_index->page_len);
6729                         data_used += page_index->page_len;
6730                 }
6731                 break;
6732         }
6733         }
6734
6735         ctl_set_success(ctsio);
6736         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6737         ctsio->be_move_done = ctl_config_move_done;
6738         ctl_datamove((union ctl_io *)ctsio);
6739         return (CTL_RETVAL_COMPLETE);
6740 }
6741
6742 int
6743 ctl_lbp_log_sense_handler(struct ctl_scsiio *ctsio,
6744                                struct ctl_page_index *page_index,
6745                                int pc)
6746 {
6747         struct ctl_lun *lun = CTL_LUN(ctsio);
6748         struct scsi_log_param_header *phdr;
6749         uint8_t *data;
6750         uint64_t val;
6751
6752         data = page_index->page_data;
6753
6754         if (lun->backend->lun_attr != NULL &&
6755             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "blocksavail"))
6756              != UINT64_MAX) {
6757                 phdr = (struct scsi_log_param_header *)data;
6758                 scsi_ulto2b(0x0001, phdr->param_code);
6759                 phdr->param_control = SLP_LBIN | SLP_LP;
6760                 phdr->param_len = 8;
6761                 data = (uint8_t *)(phdr + 1);
6762                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6763                 data[4] = 0x02; /* per-pool */
6764                 data += phdr->param_len;
6765         }
6766
6767         if (lun->backend->lun_attr != NULL &&
6768             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "blocksused"))
6769              != UINT64_MAX) {
6770                 phdr = (struct scsi_log_param_header *)data;
6771                 scsi_ulto2b(0x0002, phdr->param_code);
6772                 phdr->param_control = SLP_LBIN | SLP_LP;
6773                 phdr->param_len = 8;
6774                 data = (uint8_t *)(phdr + 1);
6775                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6776                 data[4] = 0x01; /* per-LUN */
6777                 data += phdr->param_len;
6778         }
6779
6780         if (lun->backend->lun_attr != NULL &&
6781             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "poolblocksavail"))
6782              != UINT64_MAX) {
6783                 phdr = (struct scsi_log_param_header *)data;
6784                 scsi_ulto2b(0x00f1, phdr->param_code);
6785                 phdr->param_control = SLP_LBIN | SLP_LP;
6786                 phdr->param_len = 8;
6787                 data = (uint8_t *)(phdr + 1);
6788                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6789                 data[4] = 0x02; /* per-pool */
6790                 data += phdr->param_len;
6791         }
6792
6793         if (lun->backend->lun_attr != NULL &&
6794             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "poolblocksused"))
6795              != UINT64_MAX) {
6796                 phdr = (struct scsi_log_param_header *)data;
6797                 scsi_ulto2b(0x00f2, phdr->param_code);
6798                 phdr->param_control = SLP_LBIN | SLP_LP;
6799                 phdr->param_len = 8;
6800                 data = (uint8_t *)(phdr + 1);
6801                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6802                 data[4] = 0x02; /* per-pool */
6803                 data += phdr->param_len;
6804         }
6805
6806         page_index->page_len = data - page_index->page_data;
6807         return (0);
6808 }
6809
6810 int
6811 ctl_sap_log_sense_handler(struct ctl_scsiio *ctsio,
6812                                struct ctl_page_index *page_index,
6813                                int pc)
6814 {
6815         struct ctl_lun *lun = CTL_LUN(ctsio);
6816         struct stat_page *data;
6817         struct bintime *t;
6818
6819         data = (struct stat_page *)page_index->page_data;
6820
6821         scsi_ulto2b(SLP_SAP, data->sap.hdr.param_code);
6822         data->sap.hdr.param_control = SLP_LBIN;
6823         data->sap.hdr.param_len = sizeof(struct scsi_log_stat_and_perf) -
6824             sizeof(struct scsi_log_param_header);
6825         scsi_u64to8b(lun->stats.operations[CTL_STATS_READ],
6826             data->sap.read_num);
6827         scsi_u64to8b(lun->stats.operations[CTL_STATS_WRITE],
6828             data->sap.write_num);
6829         if (lun->be_lun->blocksize > 0) {
6830                 scsi_u64to8b(lun->stats.bytes[CTL_STATS_WRITE] /
6831                     lun->be_lun->blocksize, data->sap.recvieved_lba);
6832                 scsi_u64to8b(lun->stats.bytes[CTL_STATS_READ] /
6833                     lun->be_lun->blocksize, data->sap.transmitted_lba);
6834         }
6835         t = &lun->stats.time[CTL_STATS_READ];
6836         scsi_u64to8b((uint64_t)t->sec * 1000 + t->frac / (UINT64_MAX / 1000),
6837             data->sap.read_int);
6838         t = &lun->stats.time[CTL_STATS_WRITE];
6839         scsi_u64to8b((uint64_t)t->sec * 1000 + t->frac / (UINT64_MAX / 1000),
6840             data->sap.write_int);
6841         scsi_u64to8b(0, data->sap.weighted_num);
6842         scsi_u64to8b(0, data->sap.weighted_int);
6843         scsi_ulto2b(SLP_IT, data->it.hdr.param_code);
6844         data->it.hdr.param_control = SLP_LBIN;
6845         data->it.hdr.param_len = sizeof(struct scsi_log_idle_time) -
6846             sizeof(struct scsi_log_param_header);
6847 #ifdef CTL_TIME_IO
6848         scsi_u64to8b(lun->idle_time / SBT_1MS, data->it.idle_int);
6849 #endif
6850         scsi_ulto2b(SLP_TI, data->ti.hdr.param_code);
6851         data->it.hdr.param_control = SLP_LBIN;
6852         data->ti.hdr.param_len = sizeof(struct scsi_log_time_interval) -
6853             sizeof(struct scsi_log_param_header);
6854         scsi_ulto4b(3, data->ti.exponent);
6855         scsi_ulto4b(1, data->ti.integer);
6856         return (0);
6857 }
6858
6859 int
6860 ctl_ie_log_sense_handler(struct ctl_scsiio *ctsio,
6861                                struct ctl_page_index *page_index,
6862                                int pc)
6863 {
6864         struct ctl_lun *lun = CTL_LUN(ctsio);
6865         struct scsi_log_informational_exceptions *data;
6866
6867         data = (struct scsi_log_informational_exceptions *)page_index->page_data;
6868
6869         scsi_ulto2b(SLP_IE_GEN, data->hdr.param_code);
6870         data->hdr.param_control = SLP_LBIN;
6871         data->hdr.param_len = sizeof(struct scsi_log_informational_exceptions) -
6872             sizeof(struct scsi_log_param_header);
6873         data->ie_asc = lun->ie_asc;
6874         data->ie_ascq = lun->ie_ascq;
6875         data->temperature = 0xff;
6876         return (0);
6877 }
6878
6879 int
6880 ctl_log_sense(struct ctl_scsiio *ctsio)
6881 {
6882         struct ctl_lun *lun = CTL_LUN(ctsio);
6883         int i, pc, page_code, subpage;
6884         int alloc_len, total_len;
6885         struct ctl_page_index *page_index;
6886         struct scsi_log_sense *cdb;
6887         struct scsi_log_header *header;
6888
6889         CTL_DEBUG_PRINT(("ctl_log_sense\n"));
6890
6891         cdb = (struct scsi_log_sense *)ctsio->cdb;
6892         pc = (cdb->page & SLS_PAGE_CTRL_MASK) >> 6;
6893         page_code = cdb->page & SLS_PAGE_CODE;
6894         subpage = cdb->subpage;
6895         alloc_len = scsi_2btoul(cdb->length);
6896
6897         page_index = NULL;
6898         for (i = 0; i < CTL_NUM_LOG_PAGES; i++) {
6899                 page_index = &lun->log_pages.index[i];
6900
6901                 /* Look for the right page code */
6902                 if ((page_index->page_code & SL_PAGE_CODE) != page_code)
6903                         continue;
6904
6905                 /* Look for the right subpage or the subpage wildcard*/
6906                 if (page_index->subpage != subpage)
6907                         continue;
6908
6909                 break;
6910         }
6911         if (i >= CTL_NUM_LOG_PAGES) {
6912                 ctl_set_invalid_field(ctsio,
6913                                       /*sks_valid*/ 1,
6914                                       /*command*/ 1,
6915                                       /*field*/ 2,
6916                                       /*bit_valid*/ 0,
6917                                       /*bit*/ 0);
6918                 ctl_done((union ctl_io *)ctsio);
6919                 return (CTL_RETVAL_COMPLETE);
6920         }
6921
6922         total_len = sizeof(struct scsi_log_header) + page_index->page_len;
6923
6924         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6925         ctsio->kern_sg_entries = 0;
6926         ctsio->kern_rel_offset = 0;
6927         ctsio->kern_data_len = min(total_len, alloc_len);
6928         ctsio->kern_total_len = ctsio->kern_data_len;
6929
6930         header = (struct scsi_log_header *)ctsio->kern_data_ptr;
6931         header->page = page_index->page_code;
6932         if (page_index->page_code == SLS_LOGICAL_BLOCK_PROVISIONING)
6933                 header->page |= SL_DS;
6934         if (page_index->subpage) {
6935                 header->page |= SL_SPF;
6936                 header->subpage = page_index->subpage;
6937         }
6938         scsi_ulto2b(page_index->page_len, header->datalen);
6939
6940         /*
6941          * Call the handler, if it exists, to update the
6942          * page to the latest values.
6943          */
6944         if (page_index->sense_handler != NULL)
6945                 page_index->sense_handler(ctsio, page_index, pc);
6946
6947         memcpy(header + 1, page_index->page_data, page_index->page_len);
6948
6949         ctl_set_success(ctsio);
6950         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6951         ctsio->be_move_done = ctl_config_move_done;
6952         ctl_datamove((union ctl_io *)ctsio);
6953         return (CTL_RETVAL_COMPLETE);
6954 }
6955
6956 int
6957 ctl_read_capacity(struct ctl_scsiio *ctsio)
6958 {
6959         struct ctl_lun *lun = CTL_LUN(ctsio);
6960         struct scsi_read_capacity *cdb;
6961         struct scsi_read_capacity_data *data;
6962         uint32_t lba;
6963
6964         CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
6965
6966         cdb = (struct scsi_read_capacity *)ctsio->cdb;
6967
6968         lba = scsi_4btoul(cdb->addr);
6969         if (((cdb->pmi & SRC_PMI) == 0)
6970          && (lba != 0)) {
6971                 ctl_set_invalid_field(/*ctsio*/ ctsio,
6972                                       /*sks_valid*/ 1,
6973                                       /*command*/ 1,
6974                                       /*field*/ 2,
6975                                       /*bit_valid*/ 0,
6976                                       /*bit*/ 0);
6977                 ctl_done((union ctl_io *)ctsio);
6978                 return (CTL_RETVAL_COMPLETE);
6979         }
6980
6981         ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
6982         data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
6983         ctsio->kern_data_len = sizeof(*data);
6984         ctsio->kern_total_len = sizeof(*data);
6985         ctsio->kern_rel_offset = 0;
6986         ctsio->kern_sg_entries = 0;
6987
6988         /*
6989          * If the maximum LBA is greater than 0xfffffffe, the user must
6990          * issue a SERVICE ACTION IN (16) command, with the read capacity
6991          * serivce action set.
6992          */
6993         if (lun->be_lun->maxlba > 0xfffffffe)
6994                 scsi_ulto4b(0xffffffff, data->addr);
6995         else
6996                 scsi_ulto4b(lun->be_lun->maxlba, data->addr);
6997
6998         /*
6999          * XXX KDM this may not be 512 bytes...
7000          */
7001         scsi_ulto4b(lun->be_lun->blocksize, data->length);
7002
7003         ctl_set_success(ctsio);
7004         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7005         ctsio->be_move_done = ctl_config_move_done;
7006         ctl_datamove((union ctl_io *)ctsio);
7007         return (CTL_RETVAL_COMPLETE);
7008 }
7009
7010 int
7011 ctl_read_capacity_16(struct ctl_scsiio *ctsio)
7012 {
7013         struct ctl_lun *lun = CTL_LUN(ctsio);
7014         struct scsi_read_capacity_16 *cdb;
7015         struct scsi_read_capacity_data_long *data;
7016         uint64_t lba;
7017         uint32_t alloc_len;
7018
7019         CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
7020
7021         cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
7022
7023         alloc_len = scsi_4btoul(cdb->alloc_len);
7024         lba = scsi_8btou64(cdb->addr);
7025
7026         if ((cdb->reladr & SRC16_PMI)
7027          && (lba != 0)) {
7028                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7029                                       /*sks_valid*/ 1,
7030                                       /*command*/ 1,
7031                                       /*field*/ 2,
7032                                       /*bit_valid*/ 0,
7033                                       /*bit*/ 0);
7034                 ctl_done((union ctl_io *)ctsio);
7035                 return (CTL_RETVAL_COMPLETE);
7036         }
7037
7038         ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7039         data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
7040         ctsio->kern_rel_offset = 0;
7041         ctsio->kern_sg_entries = 0;
7042         ctsio->kern_data_len = min(sizeof(*data), alloc_len);
7043         ctsio->kern_total_len = ctsio->kern_data_len;
7044
7045         scsi_u64to8b(lun->be_lun->maxlba, data->addr);
7046         /* XXX KDM this may not be 512 bytes... */
7047         scsi_ulto4b(lun->be_lun->blocksize, data->length);
7048         data->prot_lbppbe = lun->be_lun->pblockexp & SRC16_LBPPBE;
7049         scsi_ulto2b(lun->be_lun->pblockoff & SRC16_LALBA_A, data->lalba_lbp);
7050         if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
7051                 data->lalba_lbp[0] |= SRC16_LBPME | SRC16_LBPRZ;
7052
7053         ctl_set_success(ctsio);
7054         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7055         ctsio->be_move_done = ctl_config_move_done;
7056         ctl_datamove((union ctl_io *)ctsio);
7057         return (CTL_RETVAL_COMPLETE);
7058 }
7059
7060 int
7061 ctl_get_lba_status(struct ctl_scsiio *ctsio)
7062 {
7063         struct ctl_lun *lun = CTL_LUN(ctsio);
7064         struct scsi_get_lba_status *cdb;
7065         struct scsi_get_lba_status_data *data;
7066         struct ctl_lba_len_flags *lbalen;
7067         uint64_t lba;
7068         uint32_t alloc_len, total_len;
7069         int retval;
7070
7071         CTL_DEBUG_PRINT(("ctl_get_lba_status\n"));
7072
7073         cdb = (struct scsi_get_lba_status *)ctsio->cdb;
7074         lba = scsi_8btou64(cdb->addr);
7075         alloc_len = scsi_4btoul(cdb->alloc_len);
7076
7077         if (lba > lun->be_lun->maxlba) {
7078                 ctl_set_lba_out_of_range(ctsio, lba);
7079                 ctl_done((union ctl_io *)ctsio);
7080                 return (CTL_RETVAL_COMPLETE);
7081         }
7082
7083         total_len = sizeof(*data) + sizeof(data->descr[0]);
7084         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7085         data = (struct scsi_get_lba_status_data *)ctsio->kern_data_ptr;
7086         ctsio->kern_rel_offset = 0;
7087         ctsio->kern_sg_entries = 0;
7088         ctsio->kern_data_len = min(total_len, alloc_len);
7089         ctsio->kern_total_len = ctsio->kern_data_len;
7090
7091         /* Fill dummy data in case backend can't tell anything. */
7092         scsi_ulto4b(4 + sizeof(data->descr[0]), data->length);
7093         scsi_u64to8b(lba, data->descr[0].addr);
7094         scsi_ulto4b(MIN(UINT32_MAX, lun->be_lun->maxlba + 1 - lba),
7095             data->descr[0].length);
7096         data->descr[0].status = 0; /* Mapped or unknown. */
7097
7098         ctl_set_success(ctsio);
7099         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7100         ctsio->be_move_done = ctl_config_move_done;
7101
7102         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
7103         lbalen->lba = lba;
7104         lbalen->len = total_len;
7105         lbalen->flags = 0;
7106         retval = lun->backend->config_read((union ctl_io *)ctsio);
7107         return (retval);
7108 }
7109
7110 int
7111 ctl_read_defect(struct ctl_scsiio *ctsio)
7112 {
7113         struct scsi_read_defect_data_10 *ccb10;
7114         struct scsi_read_defect_data_12 *ccb12;
7115         struct scsi_read_defect_data_hdr_10 *data10;
7116         struct scsi_read_defect_data_hdr_12 *data12;
7117         uint32_t alloc_len, data_len;
7118         uint8_t format;
7119
7120         CTL_DEBUG_PRINT(("ctl_read_defect\n"));
7121
7122         if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
7123                 ccb10 = (struct scsi_read_defect_data_10 *)&ctsio->cdb;
7124                 format = ccb10->format;
7125                 alloc_len = scsi_2btoul(ccb10->alloc_length);
7126                 data_len = sizeof(*data10);
7127         } else {
7128                 ccb12 = (struct scsi_read_defect_data_12 *)&ctsio->cdb;
7129                 format = ccb12->format;
7130                 alloc_len = scsi_4btoul(ccb12->alloc_length);
7131                 data_len = sizeof(*data12);
7132         }
7133         if (alloc_len == 0) {
7134                 ctl_set_success(ctsio);
7135                 ctl_done((union ctl_io *)ctsio);
7136                 return (CTL_RETVAL_COMPLETE);
7137         }
7138
7139         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
7140         ctsio->kern_rel_offset = 0;
7141         ctsio->kern_sg_entries = 0;
7142         ctsio->kern_data_len = min(data_len, alloc_len);
7143         ctsio->kern_total_len = ctsio->kern_data_len;
7144
7145         if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
7146                 data10 = (struct scsi_read_defect_data_hdr_10 *)
7147                     ctsio->kern_data_ptr;
7148                 data10->format = format;
7149                 scsi_ulto2b(0, data10->length);
7150         } else {
7151                 data12 = (struct scsi_read_defect_data_hdr_12 *)
7152                     ctsio->kern_data_ptr;
7153                 data12->format = format;
7154                 scsi_ulto2b(0, data12->generation);
7155                 scsi_ulto4b(0, data12->length);
7156         }
7157
7158         ctl_set_success(ctsio);
7159         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7160         ctsio->be_move_done = ctl_config_move_done;
7161         ctl_datamove((union ctl_io *)ctsio);
7162         return (CTL_RETVAL_COMPLETE);
7163 }
7164
7165 int
7166 ctl_report_tagret_port_groups(struct ctl_scsiio *ctsio)
7167 {
7168         struct ctl_softc *softc = CTL_SOFTC(ctsio);
7169         struct ctl_lun *lun = CTL_LUN(ctsio);
7170         struct scsi_maintenance_in *cdb;
7171         int retval;
7172         int alloc_len, ext, total_len = 0, g, pc, pg, ts, os;
7173         int num_ha_groups, num_target_ports, shared_group;
7174         struct ctl_port *port;
7175         struct scsi_target_group_data *rtg_ptr;
7176         struct scsi_target_group_data_extended *rtg_ext_ptr;
7177         struct scsi_target_port_group_descriptor *tpg_desc;
7178
7179         CTL_DEBUG_PRINT(("ctl_report_tagret_port_groups\n"));
7180
7181         cdb = (struct scsi_maintenance_in *)ctsio->cdb;
7182         retval = CTL_RETVAL_COMPLETE;
7183
7184         switch (cdb->byte2 & STG_PDF_MASK) {
7185         case STG_PDF_LENGTH:
7186                 ext = 0;
7187                 break;
7188         case STG_PDF_EXTENDED:
7189                 ext = 1;
7190                 break;
7191         default:
7192                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7193                                       /*sks_valid*/ 1,
7194                                       /*command*/ 1,
7195                                       /*field*/ 2,
7196                                       /*bit_valid*/ 1,
7197                                       /*bit*/ 5);
7198                 ctl_done((union ctl_io *)ctsio);
7199                 return(retval);
7200         }
7201
7202         num_target_ports = 0;
7203         shared_group = (softc->is_single != 0);
7204         mtx_lock(&softc->ctl_lock);
7205         STAILQ_FOREACH(port, &softc->port_list, links) {
7206                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7207                         continue;
7208                 if (ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
7209                         continue;
7210                 num_target_ports++;
7211                 if (port->status & CTL_PORT_STATUS_HA_SHARED)
7212                         shared_group = 1;
7213         }
7214         mtx_unlock(&softc->ctl_lock);
7215         num_ha_groups = (softc->is_single) ? 0 : NUM_HA_SHELVES;
7216
7217         if (ext)
7218                 total_len = sizeof(struct scsi_target_group_data_extended);
7219         else
7220                 total_len = sizeof(struct scsi_target_group_data);
7221         total_len += sizeof(struct scsi_target_port_group_descriptor) *
7222                 (shared_group + num_ha_groups) +
7223             sizeof(struct scsi_target_port_descriptor) * num_target_ports;
7224
7225         alloc_len = scsi_4btoul(cdb->length);
7226
7227         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7228         ctsio->kern_sg_entries = 0;
7229         ctsio->kern_rel_offset = 0;
7230         ctsio->kern_data_len = min(total_len, alloc_len);
7231         ctsio->kern_total_len = ctsio->kern_data_len;
7232
7233         if (ext) {
7234                 rtg_ext_ptr = (struct scsi_target_group_data_extended *)
7235                     ctsio->kern_data_ptr;
7236                 scsi_ulto4b(total_len - 4, rtg_ext_ptr->length);
7237                 rtg_ext_ptr->format_type = 0x10;
7238                 rtg_ext_ptr->implicit_transition_time = 0;
7239                 tpg_desc = &rtg_ext_ptr->groups[0];
7240         } else {
7241                 rtg_ptr = (struct scsi_target_group_data *)
7242                     ctsio->kern_data_ptr;
7243                 scsi_ulto4b(total_len - 4, rtg_ptr->length);
7244                 tpg_desc = &rtg_ptr->groups[0];
7245         }
7246
7247         mtx_lock(&softc->ctl_lock);
7248         pg = softc->port_min / softc->port_cnt;
7249         if (lun->flags & (CTL_LUN_PRIMARY_SC | CTL_LUN_PEER_SC_PRIMARY)) {
7250                 /* Some shelf is known to be primary. */
7251                 if (softc->ha_link == CTL_HA_LINK_OFFLINE)
7252                         os = TPG_ASYMMETRIC_ACCESS_UNAVAILABLE;
7253                 else if (softc->ha_link == CTL_HA_LINK_UNKNOWN)
7254                         os = TPG_ASYMMETRIC_ACCESS_TRANSITIONING;
7255                 else if (softc->ha_mode == CTL_HA_MODE_ACT_STBY)
7256                         os = TPG_ASYMMETRIC_ACCESS_STANDBY;
7257                 else
7258                         os = TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7259                 if (lun->flags & CTL_LUN_PRIMARY_SC) {
7260                         ts = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7261                 } else {
7262                         ts = os;
7263                         os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7264                 }
7265         } else {
7266                 /* No known primary shelf. */
7267                 if (softc->ha_link == CTL_HA_LINK_OFFLINE) {
7268                         ts = TPG_ASYMMETRIC_ACCESS_UNAVAILABLE;
7269                         os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7270                 } else if (softc->ha_link == CTL_HA_LINK_UNKNOWN) {
7271                         ts = TPG_ASYMMETRIC_ACCESS_TRANSITIONING;
7272                         os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7273                 } else {
7274                         ts = os = TPG_ASYMMETRIC_ACCESS_TRANSITIONING;
7275                 }
7276         }
7277         if (shared_group) {
7278                 tpg_desc->pref_state = ts;
7279                 tpg_desc->support = TPG_AO_SUP | TPG_AN_SUP | TPG_S_SUP |
7280                     TPG_U_SUP | TPG_T_SUP;
7281                 scsi_ulto2b(1, tpg_desc->target_port_group);
7282                 tpg_desc->status = TPG_IMPLICIT;
7283                 pc = 0;
7284                 STAILQ_FOREACH(port, &softc->port_list, links) {
7285                         if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7286                                 continue;
7287                         if (!softc->is_single &&
7288                             (port->status & CTL_PORT_STATUS_HA_SHARED) == 0)
7289                                 continue;
7290                         if (ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
7291                                 continue;
7292                         scsi_ulto2b(port->targ_port, tpg_desc->descriptors[pc].
7293                             relative_target_port_identifier);
7294                         pc++;
7295                 }
7296                 tpg_desc->target_port_count = pc;
7297                 tpg_desc = (struct scsi_target_port_group_descriptor *)
7298                     &tpg_desc->descriptors[pc];
7299         }
7300         for (g = 0; g < num_ha_groups; g++) {
7301                 tpg_desc->pref_state = (g == pg) ? ts : os;
7302                 tpg_desc->support = TPG_AO_SUP | TPG_AN_SUP | TPG_S_SUP |
7303                     TPG_U_SUP | TPG_T_SUP;
7304                 scsi_ulto2b(2 + g, tpg_desc->target_port_group);
7305                 tpg_desc->status = TPG_IMPLICIT;
7306                 pc = 0;
7307                 STAILQ_FOREACH(port, &softc->port_list, links) {
7308                         if (port->targ_port < g * softc->port_cnt ||
7309                             port->targ_port >= (g + 1) * softc->port_cnt)
7310                                 continue;
7311                         if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7312                                 continue;
7313                         if (port->status & CTL_PORT_STATUS_HA_SHARED)
7314                                 continue;
7315                         if (ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
7316                                 continue;
7317                         scsi_ulto2b(port->targ_port, tpg_desc->descriptors[pc].
7318                             relative_target_port_identifier);
7319                         pc++;
7320                 }
7321                 tpg_desc->target_port_count = pc;
7322                 tpg_desc = (struct scsi_target_port_group_descriptor *)
7323                     &tpg_desc->descriptors[pc];
7324         }
7325         mtx_unlock(&softc->ctl_lock);
7326
7327         ctl_set_success(ctsio);
7328         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7329         ctsio->be_move_done = ctl_config_move_done;
7330         ctl_datamove((union ctl_io *)ctsio);
7331         return(retval);
7332 }
7333
7334 int
7335 ctl_report_supported_opcodes(struct ctl_scsiio *ctsio)
7336 {
7337         struct ctl_lun *lun = CTL_LUN(ctsio);
7338         struct scsi_report_supported_opcodes *cdb;
7339         const struct ctl_cmd_entry *entry, *sentry;
7340         struct scsi_report_supported_opcodes_all *all;
7341         struct scsi_report_supported_opcodes_descr *descr;
7342         struct scsi_report_supported_opcodes_one *one;
7343         int retval;
7344         int alloc_len, total_len;
7345         int opcode, service_action, i, j, num;
7346
7347         CTL_DEBUG_PRINT(("ctl_report_supported_opcodes\n"));
7348
7349         cdb = (struct scsi_report_supported_opcodes *)ctsio->cdb;
7350         retval = CTL_RETVAL_COMPLETE;
7351
7352         opcode = cdb->requested_opcode;
7353         service_action = scsi_2btoul(cdb->requested_service_action);
7354         switch (cdb->options & RSO_OPTIONS_MASK) {
7355         case RSO_OPTIONS_ALL:
7356                 num = 0;
7357                 for (i = 0; i < 256; i++) {
7358                         entry = &ctl_cmd_table[i];
7359                         if (entry->flags & CTL_CMD_FLAG_SA5) {
7360                                 for (j = 0; j < 32; j++) {
7361                                         sentry = &((const struct ctl_cmd_entry *)
7362                                             entry->execute)[j];
7363                                         if (ctl_cmd_applicable(
7364                                             lun->be_lun->lun_type, sentry))
7365                                                 num++;
7366                                 }
7367                         } else {
7368                                 if (ctl_cmd_applicable(lun->be_lun->lun_type,
7369                                     entry))
7370                                         num++;
7371                         }
7372                 }
7373                 total_len = sizeof(struct scsi_report_supported_opcodes_all) +
7374                     num * sizeof(struct scsi_report_supported_opcodes_descr);
7375                 break;
7376         case RSO_OPTIONS_OC:
7377                 if (ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) {
7378                         ctl_set_invalid_field(/*ctsio*/ ctsio,
7379                                               /*sks_valid*/ 1,
7380                                               /*command*/ 1,
7381                                               /*field*/ 2,
7382                                               /*bit_valid*/ 1,
7383                                               /*bit*/ 2);
7384                         ctl_done((union ctl_io *)ctsio);
7385                         return (CTL_RETVAL_COMPLETE);
7386                 }
7387                 total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7388                 break;
7389         case RSO_OPTIONS_OC_SA:
7390                 if ((ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) == 0 ||
7391                     service_action >= 32) {
7392                         ctl_set_invalid_field(/*ctsio*/ ctsio,
7393                                               /*sks_valid*/ 1,
7394                                               /*command*/ 1,
7395                                               /*field*/ 2,
7396                                               /*bit_valid*/ 1,
7397                                               /*bit*/ 2);
7398                         ctl_done((union ctl_io *)ctsio);
7399                         return (CTL_RETVAL_COMPLETE);
7400                 }
7401                 /* FALLTHROUGH */
7402         case RSO_OPTIONS_OC_ASA:
7403                 total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7404                 break;
7405         default:
7406                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7407                                       /*sks_valid*/ 1,
7408                                       /*command*/ 1,
7409                                       /*field*/ 2,
7410                                       /*bit_valid*/ 1,
7411                                       /*bit*/ 2);
7412                 ctl_done((union ctl_io *)ctsio);
7413                 return (CTL_RETVAL_COMPLETE);
7414         }
7415
7416         alloc_len = scsi_4btoul(cdb->length);
7417
7418         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7419         ctsio->kern_sg_entries = 0;
7420         ctsio->kern_rel_offset = 0;
7421         ctsio->kern_data_len = min(total_len, alloc_len);
7422         ctsio->kern_total_len = ctsio->kern_data_len;
7423
7424         switch (cdb->options & RSO_OPTIONS_MASK) {
7425         case RSO_OPTIONS_ALL:
7426                 all = (struct scsi_report_supported_opcodes_all *)
7427                     ctsio->kern_data_ptr;
7428                 num = 0;
7429                 for (i = 0; i < 256; i++) {
7430                         entry = &ctl_cmd_table[i];
7431                         if (entry->flags & CTL_CMD_FLAG_SA5) {
7432                                 for (j = 0; j < 32; j++) {
7433                                         sentry = &((const struct ctl_cmd_entry *)
7434                                             entry->execute)[j];
7435                                         if (!ctl_cmd_applicable(
7436                                             lun->be_lun->lun_type, sentry))
7437                                                 continue;
7438                                         descr = &all->descr[num++];
7439                                         descr->opcode = i;
7440                                         scsi_ulto2b(j, descr->service_action);
7441                                         descr->flags = RSO_SERVACTV;
7442                                         scsi_ulto2b(sentry->length,
7443                                             descr->cdb_length);
7444                                 }
7445                         } else {
7446                                 if (!ctl_cmd_applicable(lun->be_lun->lun_type,
7447                                     entry))
7448                                         continue;
7449                                 descr = &all->descr[num++];
7450                                 descr->opcode = i;
7451                                 scsi_ulto2b(0, descr->service_action);
7452                                 descr->flags = 0;
7453                                 scsi_ulto2b(entry->length, descr->cdb_length);
7454                         }
7455                 }
7456                 scsi_ulto4b(
7457                     num * sizeof(struct scsi_report_supported_opcodes_descr),
7458                     all->length);
7459                 break;
7460         case RSO_OPTIONS_OC:
7461                 one = (struct scsi_report_supported_opcodes_one *)
7462                     ctsio->kern_data_ptr;
7463                 entry = &ctl_cmd_table[opcode];
7464                 goto fill_one;
7465         case RSO_OPTIONS_OC_SA:
7466                 one = (struct scsi_report_supported_opcodes_one *)
7467                     ctsio->kern_data_ptr;
7468                 entry = &ctl_cmd_table[opcode];
7469                 entry = &((const struct ctl_cmd_entry *)
7470                     entry->execute)[service_action];
7471 fill_one:
7472                 if (ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
7473                         one->support = 3;
7474                         scsi_ulto2b(entry->length, one->cdb_length);
7475                         one->cdb_usage[0] = opcode;
7476                         memcpy(&one->cdb_usage[1], entry->usage,
7477                             entry->length - 1);
7478                 } else
7479                         one->support = 1;
7480                 break;
7481         case RSO_OPTIONS_OC_ASA:
7482                 one = (struct scsi_report_supported_opcodes_one *)
7483                     ctsio->kern_data_ptr;
7484                 entry = &ctl_cmd_table[opcode];
7485                 if (entry->flags & CTL_CMD_FLAG_SA5) {
7486                         entry = &((const struct ctl_cmd_entry *)
7487                             entry->execute)[service_action];
7488                 } else if (service_action != 0) {
7489                         one->support = 1;
7490                         break;
7491                 }
7492                 goto fill_one;
7493         }
7494
7495         ctl_set_success(ctsio);
7496         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7497         ctsio->be_move_done = ctl_config_move_done;
7498         ctl_datamove((union ctl_io *)ctsio);
7499         return(retval);
7500 }
7501
7502 int
7503 ctl_report_supported_tmf(struct ctl_scsiio *ctsio)
7504 {
7505         struct scsi_report_supported_tmf *cdb;
7506         struct scsi_report_supported_tmf_ext_data *data;
7507         int retval;
7508         int alloc_len, total_len;
7509
7510         CTL_DEBUG_PRINT(("ctl_report_supported_tmf\n"));
7511
7512         cdb = (struct scsi_report_supported_tmf *)ctsio->cdb;
7513
7514         retval = CTL_RETVAL_COMPLETE;
7515
7516         if (cdb->options & RST_REPD)
7517                 total_len = sizeof(struct scsi_report_supported_tmf_ext_data);
7518         else
7519                 total_len = sizeof(struct scsi_report_supported_tmf_data);
7520         alloc_len = scsi_4btoul(cdb->length);
7521
7522         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7523         ctsio->kern_sg_entries = 0;
7524         ctsio->kern_rel_offset = 0;
7525         ctsio->kern_data_len = min(total_len, alloc_len);
7526         ctsio->kern_total_len = ctsio->kern_data_len;
7527
7528         data = (struct scsi_report_supported_tmf_ext_data *)ctsio->kern_data_ptr;
7529         data->byte1 |= RST_ATS | RST_ATSS | RST_CTSS | RST_LURS | RST_QTS |
7530             RST_TRS;
7531         data->byte2 |= RST_QAES | RST_QTSS | RST_ITNRS;
7532         data->length = total_len - 4;
7533
7534         ctl_set_success(ctsio);
7535         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7536         ctsio->be_move_done = ctl_config_move_done;
7537         ctl_datamove((union ctl_io *)ctsio);
7538         return (retval);
7539 }
7540
7541 int
7542 ctl_report_timestamp(struct ctl_scsiio *ctsio)
7543 {
7544         struct scsi_report_timestamp *cdb;
7545         struct scsi_report_timestamp_data *data;
7546         struct timeval tv;
7547         int64_t timestamp;
7548         int retval;
7549         int alloc_len, total_len;
7550
7551         CTL_DEBUG_PRINT(("ctl_report_timestamp\n"));
7552
7553         cdb = (struct scsi_report_timestamp *)ctsio->cdb;
7554
7555         retval = CTL_RETVAL_COMPLETE;
7556
7557         total_len = sizeof(struct scsi_report_timestamp_data);
7558         alloc_len = scsi_4btoul(cdb->length);
7559
7560         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7561         ctsio->kern_sg_entries = 0;
7562         ctsio->kern_rel_offset = 0;
7563         ctsio->kern_data_len = min(total_len, alloc_len);
7564         ctsio->kern_total_len = ctsio->kern_data_len;
7565
7566         data = (struct scsi_report_timestamp_data *)ctsio->kern_data_ptr;
7567         scsi_ulto2b(sizeof(*data) - 2, data->length);
7568         data->origin = RTS_ORIG_OUTSIDE;
7569         getmicrotime(&tv);
7570         timestamp = (int64_t)tv.tv_sec * 1000 + tv.tv_usec / 1000;
7571         scsi_ulto4b(timestamp >> 16, data->timestamp);
7572         scsi_ulto2b(timestamp & 0xffff, &data->timestamp[4]);
7573
7574         ctl_set_success(ctsio);
7575         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7576         ctsio->be_move_done = ctl_config_move_done;
7577         ctl_datamove((union ctl_io *)ctsio);
7578         return (retval);
7579 }
7580
7581 int
7582 ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7583 {
7584         struct ctl_softc *softc = CTL_SOFTC(ctsio);
7585         struct ctl_lun *lun = CTL_LUN(ctsio);
7586         struct scsi_per_res_in *cdb;
7587         int alloc_len, total_len = 0;
7588         /* struct scsi_per_res_in_rsrv in_data; */
7589         uint64_t key;
7590
7591         CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7592
7593         cdb = (struct scsi_per_res_in *)ctsio->cdb;
7594
7595         alloc_len = scsi_2btoul(cdb->length);
7596
7597 retry:
7598         mtx_lock(&lun->lun_lock);
7599         switch (cdb->action) {
7600         case SPRI_RK: /* read keys */
7601                 total_len = sizeof(struct scsi_per_res_in_keys) +
7602                         lun->pr_key_count *
7603                         sizeof(struct scsi_per_res_key);
7604                 break;
7605         case SPRI_RR: /* read reservation */
7606                 if (lun->flags & CTL_LUN_PR_RESERVED)
7607                         total_len = sizeof(struct scsi_per_res_in_rsrv);
7608                 else
7609                         total_len = sizeof(struct scsi_per_res_in_header);
7610                 break;
7611         case SPRI_RC: /* report capabilities */
7612                 total_len = sizeof(struct scsi_per_res_cap);
7613                 break;
7614         case SPRI_RS: /* read full status */
7615                 total_len = sizeof(struct scsi_per_res_in_header) +
7616                     (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7617                     lun->pr_key_count;
7618                 break;
7619         default:
7620                 panic("%s: Invalid PR type %#x", __func__, cdb->action);
7621         }
7622         mtx_unlock(&lun->lun_lock);
7623
7624         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7625         ctsio->kern_rel_offset = 0;
7626         ctsio->kern_sg_entries = 0;
7627         ctsio->kern_data_len = min(total_len, alloc_len);
7628         ctsio->kern_total_len = ctsio->kern_data_len;
7629
7630         mtx_lock(&lun->lun_lock);
7631         switch (cdb->action) {
7632         case SPRI_RK: { // read keys
7633         struct scsi_per_res_in_keys *res_keys;
7634                 int i, key_count;
7635
7636                 res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7637
7638                 /*
7639                  * We had to drop the lock to allocate our buffer, which
7640                  * leaves time for someone to come in with another
7641                  * persistent reservation.  (That is unlikely, though,
7642                  * since this should be the only persistent reservation
7643                  * command active right now.)
7644                  */
7645                 if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7646                     (lun->pr_key_count *
7647                      sizeof(struct scsi_per_res_key)))){
7648                         mtx_unlock(&lun->lun_lock);
7649                         free(ctsio->kern_data_ptr, M_CTL);
7650                         printf("%s: reservation length changed, retrying\n",
7651                                __func__);
7652                         goto retry;
7653                 }
7654
7655                 scsi_ulto4b(lun->pr_generation, res_keys->header.generation);
7656
7657                 scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7658                              lun->pr_key_count, res_keys->header.length);
7659
7660                 for (i = 0, key_count = 0; i < CTL_MAX_INITIATORS; i++) {
7661                         if ((key = ctl_get_prkey(lun, i)) == 0)
7662                                 continue;
7663
7664                         /*
7665                          * We used lun->pr_key_count to calculate the
7666                          * size to allocate.  If it turns out the number of
7667                          * initiators with the registered flag set is
7668                          * larger than that (i.e. they haven't been kept in
7669                          * sync), we've got a problem.
7670                          */
7671                         if (key_count >= lun->pr_key_count) {
7672                                 key_count++;
7673                                 continue;
7674                         }
7675                         scsi_u64to8b(key, res_keys->keys[key_count].key);
7676                         key_count++;
7677                 }
7678                 break;
7679         }
7680         case SPRI_RR: { // read reservation
7681                 struct scsi_per_res_in_rsrv *res;
7682                 int tmp_len, header_only;
7683
7684                 res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7685
7686                 scsi_ulto4b(lun->pr_generation, res->header.generation);
7687
7688                 if (lun->flags & CTL_LUN_PR_RESERVED)
7689                 {
7690                         tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7691                         scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7692                                     res->header.length);
7693                         header_only = 0;
7694                 } else {
7695                         tmp_len = sizeof(struct scsi_per_res_in_header);
7696                         scsi_ulto4b(0, res->header.length);
7697                         header_only = 1;
7698                 }
7699
7700                 /*
7701                  * We had to drop the lock to allocate our buffer, which
7702                  * leaves time for someone to come in with another
7703                  * persistent reservation.  (That is unlikely, though,
7704                  * since this should be the only persistent reservation
7705                  * command active right now.)
7706                  */
7707                 if (tmp_len != total_len) {
7708                         mtx_unlock(&lun->lun_lock);
7709                         free(ctsio->kern_data_ptr, M_CTL);
7710                         printf("%s: reservation status changed, retrying\n",
7711                                __func__);
7712                         goto retry;
7713                 }
7714
7715                 /*
7716                  * No reservation held, so we're done.
7717                  */
7718                 if (header_only != 0)
7719                         break;
7720
7721                 /*
7722                  * If the registration is an All Registrants type, the key
7723                  * is 0, since it doesn't really matter.
7724                  */
7725                 if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7726                         scsi_u64to8b(ctl_get_prkey(lun, lun->pr_res_idx),
7727                             res->data.reservation);
7728                 }
7729                 res->data.scopetype = lun->pr_res_type;
7730                 break;
7731         }
7732         case SPRI_RC:     //report capabilities
7733         {
7734                 struct scsi_per_res_cap *res_cap;
7735                 uint16_t type_mask;
7736
7737                 res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7738                 scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7739                 res_cap->flags1 = SPRI_CRH;
7740                 res_cap->flags2 = SPRI_TMV | SPRI_ALLOW_5;
7741                 type_mask = SPRI_TM_WR_EX_AR |
7742                             SPRI_TM_EX_AC_RO |
7743                             SPRI_TM_WR_EX_RO |
7744                             SPRI_TM_EX_AC |
7745                             SPRI_TM_WR_EX |
7746                             SPRI_TM_EX_AC_AR;
7747                 scsi_ulto2b(type_mask, res_cap->type_mask);
7748                 break;
7749         }
7750         case SPRI_RS: { // read full status
7751                 struct scsi_per_res_in_full *res_status;
7752                 struct scsi_per_res_in_full_desc *res_desc;
7753                 struct ctl_port *port;
7754                 int i, len;
7755
7756                 res_status = (struct scsi_per_res_in_full*)ctsio->kern_data_ptr;
7757
7758                 /*
7759                  * We had to drop the lock to allocate our buffer, which
7760                  * leaves time for someone to come in with another
7761                  * persistent reservation.  (That is unlikely, though,
7762                  * since this should be the only persistent reservation
7763                  * command active right now.)
7764                  */
7765                 if (total_len < (sizeof(struct scsi_per_res_in_header) +
7766                     (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7767                      lun->pr_key_count)){
7768                         mtx_unlock(&lun->lun_lock);
7769                         free(ctsio->kern_data_ptr, M_CTL);
7770                         printf("%s: reservation length changed, retrying\n",
7771                                __func__);
7772                         goto retry;
7773                 }
7774
7775                 scsi_ulto4b(lun->pr_generation, res_status->header.generation);
7776
7777                 res_desc = &res_status->desc[0];
7778                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
7779                         if ((key = ctl_get_prkey(lun, i)) == 0)
7780                                 continue;
7781
7782                         scsi_u64to8b(key, res_desc->res_key.key);
7783                         if ((lun->flags & CTL_LUN_PR_RESERVED) &&
7784                             (lun->pr_res_idx == i ||
7785                              lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS)) {
7786                                 res_desc->flags = SPRI_FULL_R_HOLDER;
7787                                 res_desc->scopetype = lun->pr_res_type;
7788                         }
7789                         scsi_ulto2b(i / CTL_MAX_INIT_PER_PORT,
7790                             res_desc->rel_trgt_port_id);
7791                         len = 0;
7792                         port = softc->ctl_ports[i / CTL_MAX_INIT_PER_PORT];
7793                         if (port != NULL)
7794                                 len = ctl_create_iid(port,
7795                                     i % CTL_MAX_INIT_PER_PORT,
7796                                     res_desc->transport_id);
7797                         scsi_ulto4b(len, res_desc->additional_length);
7798                         res_desc = (struct scsi_per_res_in_full_desc *)
7799                             &res_desc->transport_id[len];
7800                 }
7801                 scsi_ulto4b((uint8_t *)res_desc - (uint8_t *)&res_status->desc[0],
7802                     res_status->header.length);
7803                 break;
7804         }
7805         default:
7806                 panic("%s: Invalid PR type %#x", __func__, cdb->action);
7807         }
7808         mtx_unlock(&lun->lun_lock);
7809
7810         ctl_set_success(ctsio);
7811         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7812         ctsio->be_move_done = ctl_config_move_done;
7813         ctl_datamove((union ctl_io *)ctsio);
7814         return (CTL_RETVAL_COMPLETE);
7815 }
7816
7817 /*
7818  * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
7819  * it should return.
7820  */
7821 static int
7822 ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
7823                 uint64_t sa_res_key, uint8_t type, uint32_t residx,
7824                 struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
7825                 struct scsi_per_res_out_parms* param)
7826 {
7827         union ctl_ha_msg persis_io;
7828         int i;
7829
7830         mtx_lock(&lun->lun_lock);
7831         if (sa_res_key == 0) {
7832                 if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
7833                         /* validate scope and type */
7834                         if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7835                              SPR_LU_SCOPE) {
7836                                 mtx_unlock(&lun->lun_lock);
7837                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7838                                                       /*sks_valid*/ 1,
7839                                                       /*command*/ 1,
7840                                                       /*field*/ 2,
7841                                                       /*bit_valid*/ 1,
7842                                                       /*bit*/ 4);
7843                                 ctl_done((union ctl_io *)ctsio);
7844                                 return (1);
7845                         }
7846
7847                         if (type>8 || type==2 || type==4 || type==0) {
7848                                 mtx_unlock(&lun->lun_lock);
7849                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7850                                                       /*sks_valid*/ 1,
7851                                                       /*command*/ 1,
7852                                                       /*field*/ 2,
7853                                                       /*bit_valid*/ 1,
7854                                                       /*bit*/ 0);
7855                                 ctl_done((union ctl_io *)ctsio);
7856                                 return (1);
7857                         }
7858
7859                         /*
7860                          * Unregister everybody else and build UA for
7861                          * them
7862                          */
7863                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
7864                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
7865                                         continue;
7866
7867                                 ctl_clr_prkey(lun, i);
7868                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7869                         }
7870                         lun->pr_key_count = 1;
7871                         lun->pr_res_type = type;
7872                         if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
7873                             lun->pr_res_type != SPR_TYPE_EX_AC_AR)
7874                                 lun->pr_res_idx = residx;
7875                         lun->pr_generation++;
7876                         mtx_unlock(&lun->lun_lock);
7877
7878                         /* send msg to other side */
7879                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7880                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7881                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7882                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
7883                         persis_io.pr.pr_info.res_type = type;
7884                         memcpy(persis_io.pr.pr_info.sa_res_key,
7885                                param->serv_act_res_key,
7886                                sizeof(param->serv_act_res_key));
7887                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
7888                             sizeof(persis_io.pr), M_WAITOK);
7889                 } else {
7890                         /* not all registrants */
7891                         mtx_unlock(&lun->lun_lock);
7892                         free(ctsio->kern_data_ptr, M_CTL);
7893                         ctl_set_invalid_field(ctsio,
7894                                               /*sks_valid*/ 1,
7895                                               /*command*/ 0,
7896                                               /*field*/ 8,
7897                                               /*bit_valid*/ 0,
7898                                               /*bit*/ 0);
7899                         ctl_done((union ctl_io *)ctsio);
7900                         return (1);
7901                 }
7902         } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
7903                 || !(lun->flags & CTL_LUN_PR_RESERVED)) {
7904                 int found = 0;
7905
7906                 if (res_key == sa_res_key) {
7907                         /* special case */
7908                         /*
7909                          * The spec implies this is not good but doesn't
7910                          * say what to do. There are two choices either
7911                          * generate a res conflict or check condition
7912                          * with illegal field in parameter data. Since
7913                          * that is what is done when the sa_res_key is
7914                          * zero I'll take that approach since this has
7915                          * to do with the sa_res_key.
7916                          */
7917                         mtx_unlock(&lun->lun_lock);
7918                         free(ctsio->kern_data_ptr, M_CTL);
7919                         ctl_set_invalid_field(ctsio,
7920                                               /*sks_valid*/ 1,
7921                                               /*command*/ 0,
7922                                               /*field*/ 8,
7923                                               /*bit_valid*/ 0,
7924                                               /*bit*/ 0);
7925                         ctl_done((union ctl_io *)ctsio);
7926                         return (1);
7927                 }
7928
7929                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
7930                         if (ctl_get_prkey(lun, i) != sa_res_key)
7931                                 continue;
7932
7933                         found = 1;
7934                         ctl_clr_prkey(lun, i);
7935                         lun->pr_key_count--;
7936                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7937                 }
7938                 if (!found) {
7939                         mtx_unlock(&lun->lun_lock);
7940                         free(ctsio->kern_data_ptr, M_CTL);
7941                         ctl_set_reservation_conflict(ctsio);
7942                         ctl_done((union ctl_io *)ctsio);
7943                         return (CTL_RETVAL_COMPLETE);
7944                 }
7945                 lun->pr_generation++;
7946                 mtx_unlock(&lun->lun_lock);
7947
7948                 /* send msg to other side */
7949                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7950                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7951                 persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7952                 persis_io.pr.pr_info.residx = lun->pr_res_idx;
7953                 persis_io.pr.pr_info.res_type = type;
7954                 memcpy(persis_io.pr.pr_info.sa_res_key,
7955                        param->serv_act_res_key,
7956                        sizeof(param->serv_act_res_key));
7957                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
7958                     sizeof(persis_io.pr), M_WAITOK);
7959         } else {
7960                 /* Reserved but not all registrants */
7961                 /* sa_res_key is res holder */
7962                 if (sa_res_key == ctl_get_prkey(lun, lun->pr_res_idx)) {
7963                         /* validate scope and type */
7964                         if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7965                              SPR_LU_SCOPE) {
7966                                 mtx_unlock(&lun->lun_lock);
7967                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7968                                                       /*sks_valid*/ 1,
7969                                                       /*command*/ 1,
7970                                                       /*field*/ 2,
7971                                                       /*bit_valid*/ 1,
7972                                                       /*bit*/ 4);
7973                                 ctl_done((union ctl_io *)ctsio);
7974                                 return (1);
7975                         }
7976
7977                         if (type>8 || type==2 || type==4 || type==0) {
7978                                 mtx_unlock(&lun->lun_lock);
7979                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7980                                                       /*sks_valid*/ 1,
7981                                                       /*command*/ 1,
7982                                                       /*field*/ 2,
7983                                                       /*bit_valid*/ 1,
7984                                                       /*bit*/ 0);
7985                                 ctl_done((union ctl_io *)ctsio);
7986                                 return (1);
7987                         }
7988
7989                         /*
7990                          * Do the following:
7991                          * if sa_res_key != res_key remove all
7992                          * registrants w/sa_res_key and generate UA
7993                          * for these registrants(Registrations
7994                          * Preempted) if it wasn't an exclusive
7995                          * reservation generate UA(Reservations
7996                          * Preempted) for all other registered nexuses
7997                          * if the type has changed. Establish the new
7998                          * reservation and holder. If res_key and
7999                          * sa_res_key are the same do the above
8000                          * except don't unregister the res holder.
8001                          */
8002
8003                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
8004                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
8005                                         continue;
8006
8007                                 if (sa_res_key == ctl_get_prkey(lun, i)) {
8008                                         ctl_clr_prkey(lun, i);
8009                                         lun->pr_key_count--;
8010                                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8011                                 } else if (type != lun->pr_res_type &&
8012                                     (lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8013                                      lun->pr_res_type == SPR_TYPE_EX_AC_RO)) {
8014                                         ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8015                                 }
8016                         }
8017                         lun->pr_res_type = type;
8018                         if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
8019                             lun->pr_res_type != SPR_TYPE_EX_AC_AR)
8020                                 lun->pr_res_idx = residx;
8021                         else
8022                                 lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8023                         lun->pr_generation++;
8024                         mtx_unlock(&lun->lun_lock);
8025
8026                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8027                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8028                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8029                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8030                         persis_io.pr.pr_info.res_type = type;
8031                         memcpy(persis_io.pr.pr_info.sa_res_key,
8032                                param->serv_act_res_key,
8033                                sizeof(param->serv_act_res_key));
8034                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8035                             sizeof(persis_io.pr), M_WAITOK);
8036                 } else {
8037                         /*
8038                          * sa_res_key is not the res holder just
8039                          * remove registrants
8040                          */
8041                         int found=0;
8042
8043                         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8044                                 if (sa_res_key != ctl_get_prkey(lun, i))
8045                                         continue;
8046
8047                                 found = 1;
8048                                 ctl_clr_prkey(lun, i);
8049                                 lun->pr_key_count--;
8050                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8051                         }
8052
8053                         if (!found) {
8054                                 mtx_unlock(&lun->lun_lock);
8055                                 free(ctsio->kern_data_ptr, M_CTL);
8056                                 ctl_set_reservation_conflict(ctsio);
8057                                 ctl_done((union ctl_io *)ctsio);
8058                                 return (1);
8059                         }
8060                         lun->pr_generation++;
8061                         mtx_unlock(&lun->lun_lock);
8062
8063                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8064                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8065                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8066                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8067                         persis_io.pr.pr_info.res_type = type;
8068                         memcpy(persis_io.pr.pr_info.sa_res_key,
8069                                param->serv_act_res_key,
8070                                sizeof(param->serv_act_res_key));
8071                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8072                             sizeof(persis_io.pr), M_WAITOK);
8073                 }
8074         }
8075         return (0);
8076 }
8077
8078 static void
8079 ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
8080 {
8081         uint64_t sa_res_key;
8082         int i;
8083
8084         sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
8085
8086         if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8087          || lun->pr_res_idx == CTL_PR_NO_RESERVATION
8088          || sa_res_key != ctl_get_prkey(lun, lun->pr_res_idx)) {
8089                 if (sa_res_key == 0) {
8090                         /*
8091                          * Unregister everybody else and build UA for
8092                          * them
8093                          */
8094                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
8095                                 if (i == msg->pr.pr_info.residx ||
8096                                     ctl_get_prkey(lun, i) == 0)
8097                                         continue;
8098
8099                                 ctl_clr_prkey(lun, i);
8100                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8101                         }
8102
8103                         lun->pr_key_count = 1;
8104                         lun->pr_res_type = msg->pr.pr_info.res_type;
8105                         if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
8106                             lun->pr_res_type != SPR_TYPE_EX_AC_AR)
8107                                 lun->pr_res_idx = msg->pr.pr_info.residx;
8108                 } else {
8109                         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8110                                 if (sa_res_key == ctl_get_prkey(lun, i))
8111                                         continue;
8112
8113                                 ctl_clr_prkey(lun, i);
8114                                 lun->pr_key_count--;
8115                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8116                         }
8117                 }
8118         } else {
8119                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8120                         if (i == msg->pr.pr_info.residx ||
8121                             ctl_get_prkey(lun, i) == 0)
8122                                 continue;
8123
8124                         if (sa_res_key == ctl_get_prkey(lun, i)) {
8125                                 ctl_clr_prkey(lun, i);
8126                                 lun->pr_key_count--;
8127                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8128                         } else if (msg->pr.pr_info.res_type != lun->pr_res_type
8129                             && (lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8130                              lun->pr_res_type == SPR_TYPE_EX_AC_RO)) {
8131                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8132                         }
8133                 }
8134                 lun->pr_res_type = msg->pr.pr_info.res_type;
8135                 if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
8136                     lun->pr_res_type != SPR_TYPE_EX_AC_AR)
8137                         lun->pr_res_idx = msg->pr.pr_info.residx;
8138                 else
8139                         lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8140         }
8141         lun->pr_generation++;
8142
8143 }
8144
8145
8146 int
8147 ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
8148 {
8149         struct ctl_softc *softc = CTL_SOFTC(ctsio);
8150         struct ctl_lun *lun = CTL_LUN(ctsio);
8151         int retval;
8152         u_int32_t param_len;
8153         struct scsi_per_res_out *cdb;
8154         struct scsi_per_res_out_parms* param;
8155         uint32_t residx;
8156         uint64_t res_key, sa_res_key, key;
8157         uint8_t type;
8158         union ctl_ha_msg persis_io;
8159         int    i;
8160
8161         CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n"));
8162
8163         cdb = (struct scsi_per_res_out *)ctsio->cdb;
8164         retval = CTL_RETVAL_COMPLETE;
8165
8166         /*
8167          * We only support whole-LUN scope.  The scope & type are ignored for
8168          * register, register and ignore existing key and clear.
8169          * We sometimes ignore scope and type on preempts too!!
8170          * Verify reservation type here as well.
8171          */
8172         type = cdb->scope_type & SPR_TYPE_MASK;
8173         if ((cdb->action == SPRO_RESERVE)
8174          || (cdb->action == SPRO_RELEASE)) {
8175                 if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
8176                         ctl_set_invalid_field(/*ctsio*/ ctsio,
8177                                               /*sks_valid*/ 1,
8178                                               /*command*/ 1,
8179                                               /*field*/ 2,
8180                                               /*bit_valid*/ 1,
8181                                               /*bit*/ 4);
8182                         ctl_done((union ctl_io *)ctsio);
8183                         return (CTL_RETVAL_COMPLETE);
8184                 }
8185
8186                 if (type>8 || type==2 || type==4 || type==0) {
8187                         ctl_set_invalid_field(/*ctsio*/ ctsio,
8188                                               /*sks_valid*/ 1,
8189                                               /*command*/ 1,
8190                                               /*field*/ 2,
8191                                               /*bit_valid*/ 1,
8192                                               /*bit*/ 0);
8193                         ctl_done((union ctl_io *)ctsio);
8194                         return (CTL_RETVAL_COMPLETE);
8195                 }
8196         }
8197
8198         param_len = scsi_4btoul(cdb->length);
8199
8200         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
8201                 ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
8202                 ctsio->kern_data_len = param_len;
8203                 ctsio->kern_total_len = param_len;
8204                 ctsio->kern_rel_offset = 0;
8205                 ctsio->kern_sg_entries = 0;
8206                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
8207                 ctsio->be_move_done = ctl_config_move_done;
8208                 ctl_datamove((union ctl_io *)ctsio);
8209
8210                 return (CTL_RETVAL_COMPLETE);
8211         }
8212
8213         param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
8214
8215         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
8216         res_key = scsi_8btou64(param->res_key.key);
8217         sa_res_key = scsi_8btou64(param->serv_act_res_key);
8218
8219         /*
8220          * Validate the reservation key here except for SPRO_REG_IGNO
8221          * This must be done for all other service actions
8222          */
8223         if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
8224                 mtx_lock(&lun->lun_lock);
8225                 if ((key = ctl_get_prkey(lun, residx)) != 0) {
8226                         if (res_key != key) {
8227                                 /*
8228                                  * The current key passed in doesn't match
8229                                  * the one the initiator previously
8230                                  * registered.
8231                                  */
8232                                 mtx_unlock(&lun->lun_lock);
8233                                 free(ctsio->kern_data_ptr, M_CTL);
8234                                 ctl_set_reservation_conflict(ctsio);
8235                                 ctl_done((union ctl_io *)ctsio);
8236                                 return (CTL_RETVAL_COMPLETE);
8237                         }
8238                 } else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
8239                         /*
8240                          * We are not registered
8241                          */
8242                         mtx_unlock(&lun->lun_lock);
8243                         free(ctsio->kern_data_ptr, M_CTL);
8244                         ctl_set_reservation_conflict(ctsio);
8245                         ctl_done((union ctl_io *)ctsio);
8246                         return (CTL_RETVAL_COMPLETE);
8247                 } else if (res_key != 0) {
8248                         /*
8249                          * We are not registered and trying to register but
8250                          * the register key isn't zero.
8251                          */
8252                         mtx_unlock(&lun->lun_lock);
8253                         free(ctsio->kern_data_ptr, M_CTL);
8254                         ctl_set_reservation_conflict(ctsio);
8255                         ctl_done((union ctl_io *)ctsio);
8256                         return (CTL_RETVAL_COMPLETE);
8257                 }
8258                 mtx_unlock(&lun->lun_lock);
8259         }
8260
8261         switch (cdb->action & SPRO_ACTION_MASK) {
8262         case SPRO_REGISTER:
8263         case SPRO_REG_IGNO: {
8264
8265 #if 0
8266                 printf("Registration received\n");
8267 #endif
8268
8269                 /*
8270                  * We don't support any of these options, as we report in
8271                  * the read capabilities request (see
8272                  * ctl_persistent_reserve_in(), above).
8273                  */
8274                 if ((param->flags & SPR_SPEC_I_PT)
8275                  || (param->flags & SPR_ALL_TG_PT)
8276                  || (param->flags & SPR_APTPL)) {
8277                         int bit_ptr;
8278
8279                         if (param->flags & SPR_APTPL)
8280                                 bit_ptr = 0;
8281                         else if (param->flags & SPR_ALL_TG_PT)
8282                                 bit_ptr = 2;
8283                         else /* SPR_SPEC_I_PT */
8284                                 bit_ptr = 3;
8285
8286                         free(ctsio->kern_data_ptr, M_CTL);
8287                         ctl_set_invalid_field(ctsio,
8288                                               /*sks_valid*/ 1,
8289                                               /*command*/ 0,
8290                                               /*field*/ 20,
8291                                               /*bit_valid*/ 1,
8292                                               /*bit*/ bit_ptr);
8293                         ctl_done((union ctl_io *)ctsio);
8294                         return (CTL_RETVAL_COMPLETE);
8295                 }
8296
8297                 mtx_lock(&lun->lun_lock);
8298
8299                 /*
8300                  * The initiator wants to clear the
8301                  * key/unregister.
8302                  */
8303                 if (sa_res_key == 0) {
8304                         if ((res_key == 0
8305                           && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
8306                          || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
8307                           && ctl_get_prkey(lun, residx) == 0)) {
8308                                 mtx_unlock(&lun->lun_lock);
8309                                 goto done;
8310                         }
8311
8312                         ctl_clr_prkey(lun, residx);
8313                         lun->pr_key_count--;
8314
8315                         if (residx == lun->pr_res_idx) {
8316                                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8317                                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8318
8319                                 if ((lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8320                                      lun->pr_res_type == SPR_TYPE_EX_AC_RO) &&
8321                                     lun->pr_key_count) {
8322                                         /*
8323                                          * If the reservation is a registrants
8324                                          * only type we need to generate a UA
8325                                          * for other registered inits.  The
8326                                          * sense code should be RESERVATIONS
8327                                          * RELEASED
8328                                          */
8329
8330                                         for (i = softc->init_min; i < softc->init_max; i++){
8331                                                 if (ctl_get_prkey(lun, i) == 0)
8332                                                         continue;
8333                                                 ctl_est_ua(lun, i,
8334                                                     CTL_UA_RES_RELEASE);
8335                                         }
8336                                 }
8337                                 lun->pr_res_type = 0;
8338                         } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8339                                 if (lun->pr_key_count==0) {
8340                                         lun->flags &= ~CTL_LUN_PR_RESERVED;
8341                                         lun->pr_res_type = 0;
8342                                         lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8343                                 }
8344                         }
8345                         lun->pr_generation++;
8346                         mtx_unlock(&lun->lun_lock);
8347
8348                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8349                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8350                         persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
8351                         persis_io.pr.pr_info.residx = residx;
8352                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8353                             sizeof(persis_io.pr), M_WAITOK);
8354                 } else /* sa_res_key != 0 */ {
8355
8356                         /*
8357                          * If we aren't registered currently then increment
8358                          * the key count and set the registered flag.
8359                          */
8360                         ctl_alloc_prkey(lun, residx);
8361                         if (ctl_get_prkey(lun, residx) == 0)
8362                                 lun->pr_key_count++;
8363                         ctl_set_prkey(lun, residx, sa_res_key);
8364                         lun->pr_generation++;
8365                         mtx_unlock(&lun->lun_lock);
8366
8367                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8368                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8369                         persis_io.pr.pr_info.action = CTL_PR_REG_KEY;
8370                         persis_io.pr.pr_info.residx = residx;
8371                         memcpy(persis_io.pr.pr_info.sa_res_key,
8372                                param->serv_act_res_key,
8373                                sizeof(param->serv_act_res_key));
8374                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8375                             sizeof(persis_io.pr), M_WAITOK);
8376                 }
8377
8378                 break;
8379         }
8380         case SPRO_RESERVE:
8381 #if 0
8382                 printf("Reserve executed type %d\n", type);
8383 #endif
8384                 mtx_lock(&lun->lun_lock);
8385                 if (lun->flags & CTL_LUN_PR_RESERVED) {
8386                         /*
8387                          * if this isn't the reservation holder and it's
8388                          * not a "all registrants" type or if the type is
8389                          * different then we have a conflict
8390                          */
8391                         if ((lun->pr_res_idx != residx
8392                           && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8393                          || lun->pr_res_type != type) {
8394                                 mtx_unlock(&lun->lun_lock);
8395                                 free(ctsio->kern_data_ptr, M_CTL);
8396                                 ctl_set_reservation_conflict(ctsio);
8397                                 ctl_done((union ctl_io *)ctsio);
8398                                 return (CTL_RETVAL_COMPLETE);
8399                         }
8400                         mtx_unlock(&lun->lun_lock);
8401                 } else /* create a reservation */ {
8402                         /*
8403                          * If it's not an "all registrants" type record
8404                          * reservation holder
8405                          */
8406                         if (type != SPR_TYPE_WR_EX_AR
8407                          && type != SPR_TYPE_EX_AC_AR)
8408                                 lun->pr_res_idx = residx; /* Res holder */
8409                         else
8410                                 lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8411
8412                         lun->flags |= CTL_LUN_PR_RESERVED;
8413                         lun->pr_res_type = type;
8414
8415                         mtx_unlock(&lun->lun_lock);
8416
8417                         /* send msg to other side */
8418                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8419                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8420                         persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8421                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8422                         persis_io.pr.pr_info.res_type = type;
8423                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8424                             sizeof(persis_io.pr), M_WAITOK);
8425                 }
8426                 break;
8427
8428         case SPRO_RELEASE:
8429                 mtx_lock(&lun->lun_lock);
8430                 if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8431                         /* No reservation exists return good status */
8432                         mtx_unlock(&lun->lun_lock);
8433                         goto done;
8434                 }
8435                 /*
8436                  * Is this nexus a reservation holder?
8437                  */
8438                 if (lun->pr_res_idx != residx
8439                  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8440                         /*
8441                          * not a res holder return good status but
8442                          * do nothing
8443                          */
8444                         mtx_unlock(&lun->lun_lock);
8445                         goto done;
8446                 }
8447
8448                 if (lun->pr_res_type != type) {
8449                         mtx_unlock(&lun->lun_lock);
8450                         free(ctsio->kern_data_ptr, M_CTL);
8451                         ctl_set_illegal_pr_release(ctsio);
8452                         ctl_done((union ctl_io *)ctsio);
8453                         return (CTL_RETVAL_COMPLETE);
8454                 }
8455
8456                 /* okay to release */
8457                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8458                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8459                 lun->pr_res_type = 0;
8460
8461                 /*
8462                  * If this isn't an exclusive access reservation and NUAR
8463                  * is not set, generate UA for all other registrants.
8464                  */
8465                 if (type != SPR_TYPE_EX_AC && type != SPR_TYPE_WR_EX &&
8466                     (lun->MODE_CTRL.queue_flags & SCP_NUAR) == 0) {
8467                         for (i = softc->init_min; i < softc->init_max; i++) {
8468                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
8469                                         continue;
8470                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8471                         }
8472                 }
8473                 mtx_unlock(&lun->lun_lock);
8474
8475                 /* Send msg to other side */
8476                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8477                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8478                 persis_io.pr.pr_info.action = CTL_PR_RELEASE;
8479                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8480                      sizeof(persis_io.pr), M_WAITOK);
8481                 break;
8482
8483         case SPRO_CLEAR:
8484                 /* send msg to other side */
8485
8486                 mtx_lock(&lun->lun_lock);
8487                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8488                 lun->pr_res_type = 0;
8489                 lun->pr_key_count = 0;
8490                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8491
8492                 ctl_clr_prkey(lun, residx);
8493                 for (i = 0; i < CTL_MAX_INITIATORS; i++)
8494                         if (ctl_get_prkey(lun, i) != 0) {
8495                                 ctl_clr_prkey(lun, i);
8496                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8497                         }
8498                 lun->pr_generation++;
8499                 mtx_unlock(&lun->lun_lock);
8500
8501                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8502                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8503                 persis_io.pr.pr_info.action = CTL_PR_CLEAR;
8504                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8505                      sizeof(persis_io.pr), M_WAITOK);
8506                 break;
8507
8508         case SPRO_PREEMPT:
8509         case SPRO_PRE_ABO: {
8510                 int nretval;
8511
8512                 nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8513                                           residx, ctsio, cdb, param);
8514                 if (nretval != 0)
8515                         return (CTL_RETVAL_COMPLETE);
8516                 break;
8517         }
8518         default:
8519                 panic("%s: Invalid PR type %#x", __func__, cdb->action);
8520         }
8521
8522 done:
8523         free(ctsio->kern_data_ptr, M_CTL);
8524         ctl_set_success(ctsio);
8525         ctl_done((union ctl_io *)ctsio);
8526
8527         return (retval);
8528 }
8529
8530 /*
8531  * This routine is for handling a message from the other SC pertaining to
8532  * persistent reserve out. All the error checking will have been done
8533  * so only perorming the action need be done here to keep the two
8534  * in sync.
8535  */
8536 static void
8537 ctl_hndl_per_res_out_on_other_sc(union ctl_io *io)
8538 {
8539         struct ctl_softc *softc = CTL_SOFTC(io);
8540         union ctl_ha_msg *msg = (union ctl_ha_msg *)&io->presio.pr_msg;
8541         struct ctl_lun *lun;
8542         int i;
8543         uint32_t residx, targ_lun;
8544
8545         targ_lun = msg->hdr.nexus.targ_mapped_lun;
8546         mtx_lock(&softc->ctl_lock);
8547         if (targ_lun >= ctl_max_luns ||
8548             (lun = softc->ctl_luns[targ_lun]) == NULL) {
8549                 mtx_unlock(&softc->ctl_lock);
8550                 return;
8551         }
8552         mtx_lock(&lun->lun_lock);
8553         mtx_unlock(&softc->ctl_lock);
8554         if (lun->flags & CTL_LUN_DISABLED) {
8555                 mtx_unlock(&lun->lun_lock);
8556                 return;
8557         }
8558         residx = ctl_get_initindex(&msg->hdr.nexus);
8559         switch(msg->pr.pr_info.action) {
8560         case CTL_PR_REG_KEY:
8561                 ctl_alloc_prkey(lun, msg->pr.pr_info.residx);
8562                 if (ctl_get_prkey(lun, msg->pr.pr_info.residx) == 0)
8563                         lun->pr_key_count++;
8564                 ctl_set_prkey(lun, msg->pr.pr_info.residx,
8565                     scsi_8btou64(msg->pr.pr_info.sa_res_key));
8566                 lun->pr_generation++;
8567                 break;
8568
8569         case CTL_PR_UNREG_KEY:
8570                 ctl_clr_prkey(lun, msg->pr.pr_info.residx);
8571                 lun->pr_key_count--;
8572
8573                 /* XXX Need to see if the reservation has been released */
8574                 /* if so do we need to generate UA? */
8575                 if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8576                         lun->flags &= ~CTL_LUN_PR_RESERVED;
8577                         lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8578
8579                         if ((lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8580                              lun->pr_res_type == SPR_TYPE_EX_AC_RO) &&
8581                             lun->pr_key_count) {
8582                                 /*
8583                                  * If the reservation is a registrants
8584                                  * only type we need to generate a UA
8585                                  * for other registered inits.  The
8586                                  * sense code should be RESERVATIONS
8587                                  * RELEASED
8588                                  */
8589
8590                                 for (i = softc->init_min; i < softc->init_max; i++) {
8591                                         if (ctl_get_prkey(lun, i) == 0)
8592                                                 continue;
8593
8594                                         ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8595                                 }
8596                         }
8597                         lun->pr_res_type = 0;
8598                 } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8599                         if (lun->pr_key_count==0) {
8600                                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8601                                 lun->pr_res_type = 0;
8602                                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8603                         }
8604                 }
8605                 lun->pr_generation++;
8606                 break;
8607
8608         case CTL_PR_RESERVE:
8609                 lun->flags |= CTL_LUN_PR_RESERVED;
8610                 lun->pr_res_type = msg->pr.pr_info.res_type;
8611                 lun->pr_res_idx = msg->pr.pr_info.residx;
8612
8613                 break;
8614
8615         case CTL_PR_RELEASE:
8616                 /*
8617                  * If this isn't an exclusive access reservation and NUAR
8618                  * is not set, generate UA for all other registrants.
8619                  */
8620                 if (lun->pr_res_type != SPR_TYPE_EX_AC &&
8621                     lun->pr_res_type != SPR_TYPE_WR_EX &&
8622                     (lun->MODE_CTRL.queue_flags & SCP_NUAR) == 0) {
8623                         for (i = softc->init_min; i < softc->init_max; i++)
8624                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
8625                                         continue;
8626                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8627                 }
8628
8629                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8630                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8631                 lun->pr_res_type = 0;
8632                 break;
8633
8634         case CTL_PR_PREEMPT:
8635                 ctl_pro_preempt_other(lun, msg);
8636                 break;
8637         case CTL_PR_CLEAR:
8638                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8639                 lun->pr_res_type = 0;
8640                 lun->pr_key_count = 0;
8641                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8642
8643                 for (i=0; i < CTL_MAX_INITIATORS; i++) {
8644                         if (ctl_get_prkey(lun, i) == 0)
8645                                 continue;
8646                         ctl_clr_prkey(lun, i);
8647                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8648                 }
8649                 lun->pr_generation++;
8650                 break;
8651         }
8652
8653         mtx_unlock(&lun->lun_lock);
8654 }
8655
8656 int
8657 ctl_read_write(struct ctl_scsiio *ctsio)
8658 {
8659         struct ctl_lun *lun = CTL_LUN(ctsio);
8660         struct ctl_lba_len_flags *lbalen;
8661         uint64_t lba;
8662         uint32_t num_blocks;
8663         int flags, retval;
8664         int isread;
8665
8666         CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
8667
8668         flags = 0;
8669         isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
8670               || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
8671         switch (ctsio->cdb[0]) {
8672         case READ_6:
8673         case WRITE_6: {
8674                 struct scsi_rw_6 *cdb;
8675
8676                 cdb = (struct scsi_rw_6 *)ctsio->cdb;
8677
8678                 lba = scsi_3btoul(cdb->addr);
8679                 /* only 5 bits are valid in the most significant address byte */
8680                 lba &= 0x1fffff;
8681                 num_blocks = cdb->length;
8682                 /*
8683                  * This is correct according to SBC-2.
8684                  */
8685                 if (num_blocks == 0)
8686                         num_blocks = 256;
8687                 break;
8688         }
8689         case READ_10:
8690         case WRITE_10: {
8691                 struct scsi_rw_10 *cdb;
8692
8693                 cdb = (struct scsi_rw_10 *)ctsio->cdb;
8694                 if (cdb->byte2 & SRW10_FUA)
8695                         flags |= CTL_LLF_FUA;
8696                 if (cdb->byte2 & SRW10_DPO)
8697                         flags |= CTL_LLF_DPO;
8698                 lba = scsi_4btoul(cdb->addr);
8699                 num_blocks = scsi_2btoul(cdb->length);
8700                 break;
8701         }
8702         case WRITE_VERIFY_10: {
8703                 struct scsi_write_verify_10 *cdb;
8704
8705                 cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
8706                 flags |= CTL_LLF_FUA;
8707                 if (cdb->byte2 & SWV_DPO)
8708                         flags |= CTL_LLF_DPO;
8709                 lba = scsi_4btoul(cdb->addr);
8710                 num_blocks = scsi_2btoul(cdb->length);
8711                 break;
8712         }
8713         case READ_12:
8714         case WRITE_12: {
8715                 struct scsi_rw_12 *cdb;
8716
8717                 cdb = (struct scsi_rw_12 *)ctsio->cdb;
8718                 if (cdb->byte2 & SRW12_FUA)
8719                         flags |= CTL_LLF_FUA;
8720                 if (cdb->byte2 & SRW12_DPO)
8721                         flags |= CTL_LLF_DPO;
8722                 lba = scsi_4btoul(cdb->addr);
8723                 num_blocks = scsi_4btoul(cdb->length);
8724                 break;
8725         }
8726         case WRITE_VERIFY_12: {
8727                 struct scsi_write_verify_12 *cdb;
8728
8729                 cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
8730                 flags |= CTL_LLF_FUA;
8731                 if (cdb->byte2 & SWV_DPO)
8732                         flags |= CTL_LLF_DPO;
8733                 lba = scsi_4btoul(cdb->addr);
8734                 num_blocks = scsi_4btoul(cdb->length);
8735                 break;
8736         }
8737         case READ_16:
8738         case WRITE_16: {
8739                 struct scsi_rw_16 *cdb;
8740
8741                 cdb = (struct scsi_rw_16 *)ctsio->cdb;
8742                 if (cdb->byte2 & SRW12_FUA)
8743                         flags |= CTL_LLF_FUA;
8744                 if (cdb->byte2 & SRW12_DPO)
8745                         flags |= CTL_LLF_DPO;
8746                 lba = scsi_8btou64(cdb->addr);
8747                 num_blocks = scsi_4btoul(cdb->length);
8748                 break;
8749         }
8750         case WRITE_ATOMIC_16: {
8751                 struct scsi_write_atomic_16 *cdb;
8752
8753                 if (lun->be_lun->atomicblock == 0) {
8754                         ctl_set_invalid_opcode(ctsio);
8755                         ctl_done((union ctl_io *)ctsio);
8756                         return (CTL_RETVAL_COMPLETE);
8757                 }
8758
8759                 cdb = (struct scsi_write_atomic_16 *)ctsio->cdb;
8760                 if (cdb->byte2 & SRW12_FUA)
8761                         flags |= CTL_LLF_FUA;
8762                 if (cdb->byte2 & SRW12_DPO)
8763                         flags |= CTL_LLF_DPO;
8764                 lba = scsi_8btou64(cdb->addr);
8765                 num_blocks = scsi_2btoul(cdb->length);
8766                 if (num_blocks > lun->be_lun->atomicblock) {
8767                         ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
8768                             /*command*/ 1, /*field*/ 12, /*bit_valid*/ 0,
8769                             /*bit*/ 0);
8770                         ctl_done((union ctl_io *)ctsio);
8771                         return (CTL_RETVAL_COMPLETE);
8772                 }
8773                 break;
8774         }
8775         case WRITE_VERIFY_16: {
8776                 struct scsi_write_verify_16 *cdb;
8777
8778                 cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
8779                 flags |= CTL_LLF_FUA;
8780                 if (cdb->byte2 & SWV_DPO)
8781                         flags |= CTL_LLF_DPO;
8782                 lba = scsi_8btou64(cdb->addr);
8783                 num_blocks = scsi_4btoul(cdb->length);
8784                 break;
8785         }
8786         default:
8787                 /*
8788                  * We got a command we don't support.  This shouldn't
8789                  * happen, commands should be filtered out above us.
8790                  */
8791                 ctl_set_invalid_opcode(ctsio);
8792                 ctl_done((union ctl_io *)ctsio);
8793
8794                 return (CTL_RETVAL_COMPLETE);
8795                 break; /* NOTREACHED */
8796         }
8797
8798         /*
8799          * The first check is to make sure we're in bounds, the second
8800          * check is to catch wrap-around problems.  If the lba + num blocks
8801          * is less than the lba, then we've wrapped around and the block
8802          * range is invalid anyway.
8803          */
8804         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8805          || ((lba + num_blocks) < lba)) {
8806                 ctl_set_lba_out_of_range(ctsio,
8807                     MAX(lba, lun->be_lun->maxlba + 1));
8808                 ctl_done((union ctl_io *)ctsio);
8809                 return (CTL_RETVAL_COMPLETE);
8810         }
8811
8812         /*
8813          * According to SBC-3, a transfer length of 0 is not an error.
8814          * Note that this cannot happen with WRITE(6) or READ(6), since 0
8815          * translates to 256 blocks for those commands.
8816          */
8817         if (num_blocks == 0) {
8818                 ctl_set_success(ctsio);
8819                 ctl_done((union ctl_io *)ctsio);
8820                 return (CTL_RETVAL_COMPLETE);
8821         }
8822
8823         /* Set FUA and/or DPO if caches are disabled. */
8824         if (isread) {
8825                 if ((lun->MODE_CACHING.flags1 & SCP_RCD) != 0)
8826                         flags |= CTL_LLF_FUA | CTL_LLF_DPO;
8827         } else {
8828                 if ((lun->MODE_CACHING.flags1 & SCP_WCE) == 0)
8829                         flags |= CTL_LLF_FUA;
8830         }
8831
8832         lbalen = (struct ctl_lba_len_flags *)
8833             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8834         lbalen->lba = lba;
8835         lbalen->len = num_blocks;
8836         lbalen->flags = (isread ? CTL_LLF_READ : CTL_LLF_WRITE) | flags;
8837
8838         ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
8839         ctsio->kern_rel_offset = 0;
8840
8841         CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
8842
8843         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8844         return (retval);
8845 }
8846
8847 static int
8848 ctl_cnw_cont(union ctl_io *io)
8849 {
8850         struct ctl_lun *lun = CTL_LUN(io);
8851         struct ctl_scsiio *ctsio;
8852         struct ctl_lba_len_flags *lbalen;
8853         int retval;
8854
8855         ctsio = &io->scsiio;
8856         ctsio->io_hdr.status = CTL_STATUS_NONE;
8857         ctsio->io_hdr.flags &= ~CTL_FLAG_IO_CONT;
8858         lbalen = (struct ctl_lba_len_flags *)
8859             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8860         lbalen->flags &= ~CTL_LLF_COMPARE;
8861         lbalen->flags |= CTL_LLF_WRITE;
8862
8863         CTL_DEBUG_PRINT(("ctl_cnw_cont: calling data_submit()\n"));
8864         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8865         return (retval);
8866 }
8867
8868 int
8869 ctl_cnw(struct ctl_scsiio *ctsio)
8870 {
8871         struct ctl_lun *lun = CTL_LUN(ctsio);
8872         struct ctl_lba_len_flags *lbalen;
8873         uint64_t lba;
8874         uint32_t num_blocks;
8875         int flags, retval;
8876
8877         CTL_DEBUG_PRINT(("ctl_cnw: command: %#x\n", ctsio->cdb[0]));
8878
8879         flags = 0;
8880         switch (ctsio->cdb[0]) {
8881         case COMPARE_AND_WRITE: {
8882                 struct scsi_compare_and_write *cdb;
8883
8884                 cdb = (struct scsi_compare_and_write *)ctsio->cdb;
8885                 if (cdb->byte2 & SRW10_FUA)
8886                         flags |= CTL_LLF_FUA;
8887                 if (cdb->byte2 & SRW10_DPO)
8888                         flags |= CTL_LLF_DPO;
8889                 lba = scsi_8btou64(cdb->addr);
8890                 num_blocks = cdb->length;
8891                 break;
8892         }
8893         default:
8894                 /*
8895                  * We got a command we don't support.  This shouldn't
8896                  * happen, commands should be filtered out above us.
8897                  */
8898                 ctl_set_invalid_opcode(ctsio);
8899                 ctl_done((union ctl_io *)ctsio);
8900
8901                 return (CTL_RETVAL_COMPLETE);
8902                 break; /* NOTREACHED */
8903         }
8904
8905         /*
8906          * The first check is to make sure we're in bounds, the second
8907          * check is to catch wrap-around problems.  If the lba + num blocks
8908          * is less than the lba, then we've wrapped around and the block
8909          * range is invalid anyway.
8910          */
8911         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8912          || ((lba + num_blocks) < lba)) {
8913                 ctl_set_lba_out_of_range(ctsio,
8914                     MAX(lba, lun->be_lun->maxlba + 1));
8915                 ctl_done((union ctl_io *)ctsio);
8916                 return (CTL_RETVAL_COMPLETE);
8917         }
8918
8919         /*
8920          * According to SBC-3, a transfer length of 0 is not an error.
8921          */
8922         if (num_blocks == 0) {
8923                 ctl_set_success(ctsio);
8924                 ctl_done((union ctl_io *)ctsio);
8925                 return (CTL_RETVAL_COMPLETE);
8926         }
8927
8928         /* Set FUA if write cache is disabled. */
8929         if ((lun->MODE_CACHING.flags1 & SCP_WCE) == 0)
8930                 flags |= CTL_LLF_FUA;
8931
8932         ctsio->kern_total_len = 2 * num_blocks * lun->be_lun->blocksize;
8933         ctsio->kern_rel_offset = 0;
8934
8935         /*
8936          * Set the IO_CONT flag, so that if this I/O gets passed to
8937          * ctl_data_submit_done(), it'll get passed back to
8938          * ctl_ctl_cnw_cont() for further processing.
8939          */
8940         ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
8941         ctsio->io_cont = ctl_cnw_cont;
8942
8943         lbalen = (struct ctl_lba_len_flags *)
8944             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8945         lbalen->lba = lba;
8946         lbalen->len = num_blocks;
8947         lbalen->flags = CTL_LLF_COMPARE | flags;
8948
8949         CTL_DEBUG_PRINT(("ctl_cnw: calling data_submit()\n"));
8950         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8951         return (retval);
8952 }
8953
8954 int
8955 ctl_verify(struct ctl_scsiio *ctsio)
8956 {
8957         struct ctl_lun *lun = CTL_LUN(ctsio);
8958         struct ctl_lba_len_flags *lbalen;
8959         uint64_t lba;
8960         uint32_t num_blocks;
8961         int bytchk, flags;
8962         int retval;
8963
8964         CTL_DEBUG_PRINT(("ctl_verify: command: %#x\n", ctsio->cdb[0]));
8965
8966         bytchk = 0;
8967         flags = CTL_LLF_FUA;
8968         switch (ctsio->cdb[0]) {
8969         case VERIFY_10: {
8970                 struct scsi_verify_10 *cdb;
8971
8972                 cdb = (struct scsi_verify_10 *)ctsio->cdb;
8973                 if (cdb->byte2 & SVFY_BYTCHK)
8974                         bytchk = 1;
8975                 if (cdb->byte2 & SVFY_DPO)
8976                         flags |= CTL_LLF_DPO;
8977                 lba = scsi_4btoul(cdb->addr);
8978                 num_blocks = scsi_2btoul(cdb->length);
8979                 break;
8980         }
8981         case VERIFY_12: {
8982                 struct scsi_verify_12 *cdb;
8983
8984                 cdb = (struct scsi_verify_12 *)ctsio->cdb;
8985                 if (cdb->byte2 & SVFY_BYTCHK)
8986                         bytchk = 1;
8987                 if (cdb->byte2 & SVFY_DPO)
8988                         flags |= CTL_LLF_DPO;
8989                 lba = scsi_4btoul(cdb->addr);
8990                 num_blocks = scsi_4btoul(cdb->length);
8991                 break;
8992         }
8993         case VERIFY_16: {
8994                 struct scsi_rw_16 *cdb;
8995
8996                 cdb = (struct scsi_rw_16 *)ctsio->cdb;
8997                 if (cdb->byte2 & SVFY_BYTCHK)
8998                         bytchk = 1;
8999                 if (cdb->byte2 & SVFY_DPO)
9000                         flags |= CTL_LLF_DPO;
9001                 lba = scsi_8btou64(cdb->addr);
9002                 num_blocks = scsi_4btoul(cdb->length);
9003                 break;
9004         }
9005         default:
9006                 /*
9007                  * We got a command we don't support.  This shouldn't
9008                  * happen, commands should be filtered out above us.
9009                  */
9010                 ctl_set_invalid_opcode(ctsio);
9011                 ctl_done((union ctl_io *)ctsio);
9012                 return (CTL_RETVAL_COMPLETE);
9013         }
9014
9015         /*
9016          * The first check is to make sure we're in bounds, the second
9017          * check is to catch wrap-around problems.  If the lba + num blocks
9018          * is less than the lba, then we've wrapped around and the block
9019          * range is invalid anyway.
9020          */
9021         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9022          || ((lba + num_blocks) < lba)) {
9023                 ctl_set_lba_out_of_range(ctsio,
9024                     MAX(lba, lun->be_lun->maxlba + 1));
9025                 ctl_done((union ctl_io *)ctsio);
9026                 return (CTL_RETVAL_COMPLETE);
9027         }
9028
9029         /*
9030          * According to SBC-3, a transfer length of 0 is not an error.
9031          */
9032         if (num_blocks == 0) {
9033                 ctl_set_success(ctsio);
9034                 ctl_done((union ctl_io *)ctsio);
9035                 return (CTL_RETVAL_COMPLETE);
9036         }
9037
9038         lbalen = (struct ctl_lba_len_flags *)
9039             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9040         lbalen->lba = lba;
9041         lbalen->len = num_blocks;
9042         if (bytchk) {
9043                 lbalen->flags = CTL_LLF_COMPARE | flags;
9044                 ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9045         } else {
9046                 lbalen->flags = CTL_LLF_VERIFY | flags;
9047                 ctsio->kern_total_len = 0;
9048         }
9049         ctsio->kern_rel_offset = 0;
9050
9051         CTL_DEBUG_PRINT(("ctl_verify: calling data_submit()\n"));
9052         retval = lun->backend->data_submit((union ctl_io *)ctsio);
9053         return (retval);
9054 }
9055
9056 int
9057 ctl_report_luns(struct ctl_scsiio *ctsio)
9058 {
9059         struct ctl_softc *softc = CTL_SOFTC(ctsio);
9060         struct ctl_port *port = CTL_PORT(ctsio);
9061         struct ctl_lun *lun, *request_lun = CTL_LUN(ctsio);
9062         struct scsi_report_luns *cdb;
9063         struct scsi_report_luns_data *lun_data;
9064         int num_filled, num_luns, num_port_luns, retval;
9065         uint32_t alloc_len, lun_datalen;
9066         uint32_t initidx, targ_lun_id, lun_id;
9067
9068         retval = CTL_RETVAL_COMPLETE;
9069         cdb = (struct scsi_report_luns *)ctsio->cdb;
9070
9071         CTL_DEBUG_PRINT(("ctl_report_luns\n"));
9072
9073         num_luns = 0;
9074         num_port_luns = port->lun_map ? port->lun_map_size : ctl_max_luns;
9075         mtx_lock(&softc->ctl_lock);
9076         for (targ_lun_id = 0; targ_lun_id < num_port_luns; targ_lun_id++) {
9077                 if (ctl_lun_map_from_port(port, targ_lun_id) != UINT32_MAX)
9078                         num_luns++;
9079         }
9080         mtx_unlock(&softc->ctl_lock);
9081
9082         switch (cdb->select_report) {
9083         case RPL_REPORT_DEFAULT:
9084         case RPL_REPORT_ALL:
9085         case RPL_REPORT_NONSUBSID:
9086                 break;
9087         case RPL_REPORT_WELLKNOWN:
9088         case RPL_REPORT_ADMIN:
9089         case RPL_REPORT_CONGLOM:
9090                 num_luns = 0;
9091                 break;
9092         default:
9093                 ctl_set_invalid_field(ctsio,
9094                                       /*sks_valid*/ 1,
9095                                       /*command*/ 1,
9096                                       /*field*/ 2,
9097                                       /*bit_valid*/ 0,
9098                                       /*bit*/ 0);
9099                 ctl_done((union ctl_io *)ctsio);
9100                 return (retval);
9101                 break; /* NOTREACHED */
9102         }
9103
9104         alloc_len = scsi_4btoul(cdb->length);
9105         /*
9106          * The initiator has to allocate at least 16 bytes for this request,
9107          * so he can at least get the header and the first LUN.  Otherwise
9108          * we reject the request (per SPC-3 rev 14, section 6.21).
9109          */
9110         if (alloc_len < (sizeof(struct scsi_report_luns_data) +
9111             sizeof(struct scsi_report_luns_lundata))) {
9112                 ctl_set_invalid_field(ctsio,
9113                                       /*sks_valid*/ 1,
9114                                       /*command*/ 1,
9115                                       /*field*/ 6,
9116                                       /*bit_valid*/ 0,
9117                                       /*bit*/ 0);
9118                 ctl_done((union ctl_io *)ctsio);
9119                 return (retval);
9120         }
9121
9122         lun_datalen = sizeof(*lun_data) +
9123                 (num_luns * sizeof(struct scsi_report_luns_lundata));
9124
9125         ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO);
9126         lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
9127         ctsio->kern_sg_entries = 0;
9128
9129         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9130
9131         mtx_lock(&softc->ctl_lock);
9132         for (targ_lun_id = 0, num_filled = 0;
9133             targ_lun_id < num_port_luns && num_filled < num_luns;
9134             targ_lun_id++) {
9135                 lun_id = ctl_lun_map_from_port(port, targ_lun_id);
9136                 if (lun_id == UINT32_MAX)
9137                         continue;
9138                 lun = softc->ctl_luns[lun_id];
9139                 if (lun == NULL)
9140                         continue;
9141
9142                 be64enc(lun_data->luns[num_filled++].lundata,
9143                     ctl_encode_lun(targ_lun_id));
9144
9145                 /*
9146                  * According to SPC-3, rev 14 section 6.21:
9147                  *
9148                  * "The execution of a REPORT LUNS command to any valid and
9149                  * installed logical unit shall clear the REPORTED LUNS DATA
9150                  * HAS CHANGED unit attention condition for all logical
9151                  * units of that target with respect to the requesting
9152                  * initiator. A valid and installed logical unit is one
9153                  * having a PERIPHERAL QUALIFIER of 000b in the standard
9154                  * INQUIRY data (see 6.4.2)."
9155                  *
9156                  * If request_lun is NULL, the LUN this report luns command
9157                  * was issued to is either disabled or doesn't exist. In that
9158                  * case, we shouldn't clear any pending lun change unit
9159                  * attention.
9160                  */
9161                 if (request_lun != NULL) {
9162                         mtx_lock(&lun->lun_lock);
9163                         ctl_clr_ua(lun, initidx, CTL_UA_LUN_CHANGE);
9164                         mtx_unlock(&lun->lun_lock);
9165                 }
9166         }
9167         mtx_unlock(&softc->ctl_lock);
9168
9169         /*
9170          * It's quite possible that we've returned fewer LUNs than we allocated
9171          * space for.  Trim it.
9172          */
9173         lun_datalen = sizeof(*lun_data) +
9174                 (num_filled * sizeof(struct scsi_report_luns_lundata));
9175         ctsio->kern_rel_offset = 0;
9176         ctsio->kern_sg_entries = 0;
9177         ctsio->kern_data_len = min(lun_datalen, alloc_len);
9178         ctsio->kern_total_len = ctsio->kern_data_len;
9179
9180         /*
9181          * We set this to the actual data length, regardless of how much
9182          * space we actually have to return results.  If the user looks at
9183          * this value, he'll know whether or not he allocated enough space
9184          * and reissue the command if necessary.  We don't support well
9185          * known logical units, so if the user asks for that, return none.
9186          */
9187         scsi_ulto4b(lun_datalen - 8, lun_data->length);
9188
9189         /*
9190          * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
9191          * this request.
9192          */
9193         ctl_set_success(ctsio);
9194         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9195         ctsio->be_move_done = ctl_config_move_done;
9196         ctl_datamove((union ctl_io *)ctsio);
9197         return (retval);
9198 }
9199
9200 int
9201 ctl_request_sense(struct ctl_scsiio *ctsio)
9202 {
9203         struct ctl_softc *softc = CTL_SOFTC(ctsio);
9204         struct ctl_lun *lun = CTL_LUN(ctsio);
9205         struct scsi_request_sense *cdb;
9206         struct scsi_sense_data *sense_ptr, *ps;
9207         uint32_t initidx;
9208         int have_error;
9209         u_int sense_len = SSD_FULL_SIZE;
9210         scsi_sense_data_type sense_format;
9211         ctl_ua_type ua_type;
9212         uint8_t asc = 0, ascq = 0;
9213
9214         cdb = (struct scsi_request_sense *)ctsio->cdb;
9215
9216         CTL_DEBUG_PRINT(("ctl_request_sense\n"));
9217
9218         /*
9219          * Determine which sense format the user wants.
9220          */
9221         if (cdb->byte2 & SRS_DESC)
9222                 sense_format = SSD_TYPE_DESC;
9223         else
9224                 sense_format = SSD_TYPE_FIXED;
9225
9226         ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
9227         sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
9228         ctsio->kern_sg_entries = 0;
9229         ctsio->kern_rel_offset = 0;
9230
9231         /*
9232          * struct scsi_sense_data, which is currently set to 256 bytes, is
9233          * larger than the largest allowed value for the length field in the
9234          * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
9235          */
9236         ctsio->kern_data_len = cdb->length;
9237         ctsio->kern_total_len = cdb->length;
9238
9239         /*
9240          * If we don't have a LUN, we don't have any pending sense.
9241          */
9242         if (lun == NULL ||
9243             ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
9244              softc->ha_link < CTL_HA_LINK_UNKNOWN)) {
9245                 /* "Logical unit not supported" */
9246                 ctl_set_sense_data(sense_ptr, &sense_len, NULL, sense_format,
9247                     /*current_error*/ 1,
9248                     /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
9249                     /*asc*/ 0x25,
9250                     /*ascq*/ 0x00,
9251                     SSD_ELEM_NONE);
9252                 goto send;
9253         }
9254
9255         have_error = 0;
9256         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9257         /*
9258          * Check for pending sense, and then for pending unit attentions.
9259          * Pending sense gets returned first, then pending unit attentions.
9260          */
9261         mtx_lock(&lun->lun_lock);
9262         ps = lun->pending_sense[initidx / CTL_MAX_INIT_PER_PORT];
9263         if (ps != NULL)
9264                 ps += initidx % CTL_MAX_INIT_PER_PORT;
9265         if (ps != NULL && ps->error_code != 0) {
9266                 scsi_sense_data_type stored_format;
9267
9268                 /*
9269                  * Check to see which sense format was used for the stored
9270                  * sense data.
9271                  */
9272                 stored_format = scsi_sense_type(ps);
9273
9274                 /*
9275                  * If the user requested a different sense format than the
9276                  * one we stored, then we need to convert it to the other
9277                  * format.  If we're going from descriptor to fixed format
9278                  * sense data, we may lose things in translation, depending
9279                  * on what options were used.
9280                  *
9281                  * If the stored format is SSD_TYPE_NONE (i.e. invalid),
9282                  * for some reason we'll just copy it out as-is.
9283                  */
9284                 if ((stored_format == SSD_TYPE_FIXED)
9285                  && (sense_format == SSD_TYPE_DESC))
9286                         ctl_sense_to_desc((struct scsi_sense_data_fixed *)
9287                             ps, (struct scsi_sense_data_desc *)sense_ptr);
9288                 else if ((stored_format == SSD_TYPE_DESC)
9289                       && (sense_format == SSD_TYPE_FIXED))
9290                         ctl_sense_to_fixed((struct scsi_sense_data_desc *)
9291                             ps, (struct scsi_sense_data_fixed *)sense_ptr);
9292                 else
9293                         memcpy(sense_ptr, ps, sizeof(*sense_ptr));
9294
9295                 ps->error_code = 0;
9296                 have_error = 1;
9297         } else {
9298                 ua_type = ctl_build_ua(lun, initidx, sense_ptr, &sense_len,
9299                     sense_format);
9300                 if (ua_type != CTL_UA_NONE)
9301                         have_error = 1;
9302         }
9303         if (have_error == 0) {
9304                 /*
9305                  * Report informational exception if have one and allowed.
9306                  */
9307                 if (lun->MODE_IE.mrie != SIEP_MRIE_NO) {
9308                         asc = lun->ie_asc;
9309                         ascq = lun->ie_ascq;
9310                 }
9311                 ctl_set_sense_data(sense_ptr, &sense_len, lun, sense_format,
9312                     /*current_error*/ 1,
9313                     /*sense_key*/ SSD_KEY_NO_SENSE,
9314                     /*asc*/ asc,
9315                     /*ascq*/ ascq,
9316                     SSD_ELEM_NONE);
9317         }
9318         mtx_unlock(&lun->lun_lock);
9319
9320 send:
9321         /*
9322          * We report the SCSI status as OK, since the status of the command
9323          * itself is OK.  We're reporting sense as parameter data.
9324          */
9325         ctl_set_success(ctsio);
9326         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9327         ctsio->be_move_done = ctl_config_move_done;
9328         ctl_datamove((union ctl_io *)ctsio);
9329         return (CTL_RETVAL_COMPLETE);
9330 }
9331
9332 int
9333 ctl_tur(struct ctl_scsiio *ctsio)
9334 {
9335
9336         CTL_DEBUG_PRINT(("ctl_tur\n"));
9337
9338         ctl_set_success(ctsio);
9339         ctl_done((union ctl_io *)ctsio);
9340
9341         return (CTL_RETVAL_COMPLETE);
9342 }
9343
9344 /*
9345  * SCSI VPD page 0x00, the Supported VPD Pages page.
9346  */
9347 static int
9348 ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
9349 {
9350         struct ctl_lun *lun = CTL_LUN(ctsio);
9351         struct scsi_vpd_supported_pages *pages;
9352         int sup_page_size;
9353         int p;
9354
9355         sup_page_size = sizeof(struct scsi_vpd_supported_pages) *
9356             SCSI_EVPD_NUM_SUPPORTED_PAGES;
9357         ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO);
9358         pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
9359         ctsio->kern_rel_offset = 0;
9360         ctsio->kern_sg_entries = 0;
9361         ctsio->kern_data_len = min(sup_page_size, alloc_len);
9362         ctsio->kern_total_len = ctsio->kern_data_len;
9363
9364         /*
9365          * The control device is always connected.  The disk device, on the
9366          * other hand, may not be online all the time.  Need to change this
9367          * to figure out whether the disk device is actually online or not.
9368          */
9369         if (lun != NULL)
9370                 pages->device = (SID_QUAL_LU_CONNECTED << 5) |
9371                                 lun->be_lun->lun_type;
9372         else
9373                 pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9374
9375         p = 0;
9376         /* Supported VPD pages */
9377         pages->page_list[p++] = SVPD_SUPPORTED_PAGES;
9378         /* Serial Number */
9379         pages->page_list[p++] = SVPD_UNIT_SERIAL_NUMBER;
9380         /* Device Identification */
9381         pages->page_list[p++] = SVPD_DEVICE_ID;
9382         /* Extended INQUIRY Data */
9383         pages->page_list[p++] = SVPD_EXTENDED_INQUIRY_DATA;
9384         /* Mode Page Policy */
9385         pages->page_list[p++] = SVPD_MODE_PAGE_POLICY;
9386         /* SCSI Ports */
9387         pages->page_list[p++] = SVPD_SCSI_PORTS;
9388         /* Third-party Copy */
9389         pages->page_list[p++] = SVPD_SCSI_TPC;
9390         if (lun != NULL && lun->be_lun->lun_type == T_DIRECT) {
9391                 /* Block limits */
9392                 pages->page_list[p++] = SVPD_BLOCK_LIMITS;
9393                 /* Block Device Characteristics */
9394                 pages->page_list[p++] = SVPD_BDC;
9395                 /* Logical Block Provisioning */
9396                 pages->page_list[p++] = SVPD_LBP;
9397         }
9398         pages->length = p;
9399
9400         ctl_set_success(ctsio);
9401         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9402         ctsio->be_move_done = ctl_config_move_done;
9403         ctl_datamove((union ctl_io *)ctsio);
9404         return (CTL_RETVAL_COMPLETE);
9405 }
9406
9407 /*
9408  * SCSI VPD page 0x80, the Unit Serial Number page.
9409  */
9410 static int
9411 ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
9412 {
9413         struct ctl_lun *lun = CTL_LUN(ctsio);
9414         struct scsi_vpd_unit_serial_number *sn_ptr;
9415         int data_len;
9416
9417         data_len = 4 + CTL_SN_LEN;
9418         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9419         sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
9420         ctsio->kern_rel_offset = 0;
9421         ctsio->kern_sg_entries = 0;
9422         ctsio->kern_data_len = min(data_len, alloc_len);
9423         ctsio->kern_total_len = ctsio->kern_data_len;
9424
9425         /*
9426          * The control device is always connected.  The disk device, on the
9427          * other hand, may not be online all the time.  Need to change this
9428          * to figure out whether the disk device is actually online or not.
9429          */
9430         if (lun != NULL)
9431                 sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9432                                   lun->be_lun->lun_type;
9433         else
9434                 sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9435
9436         sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9437         sn_ptr->length = CTL_SN_LEN;
9438         /*
9439          * If we don't have a LUN, we just leave the serial number as
9440          * all spaces.
9441          */
9442         if (lun != NULL) {
9443                 strncpy((char *)sn_ptr->serial_num,
9444                         (char *)lun->be_lun->serial_num, CTL_SN_LEN);
9445         } else
9446                 memset(sn_ptr->serial_num, 0x20, CTL_SN_LEN);
9447
9448         ctl_set_success(ctsio);
9449         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9450         ctsio->be_move_done = ctl_config_move_done;
9451         ctl_datamove((union ctl_io *)ctsio);
9452         return (CTL_RETVAL_COMPLETE);
9453 }
9454
9455
9456 /*
9457  * SCSI VPD page 0x86, the Extended INQUIRY Data page.
9458  */
9459 static int
9460 ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len)
9461 {
9462         struct ctl_lun *lun = CTL_LUN(ctsio);
9463         struct scsi_vpd_extended_inquiry_data *eid_ptr;
9464         int data_len;
9465
9466         data_len = sizeof(struct scsi_vpd_extended_inquiry_data);
9467         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9468         eid_ptr = (struct scsi_vpd_extended_inquiry_data *)ctsio->kern_data_ptr;
9469         ctsio->kern_sg_entries = 0;
9470         ctsio->kern_rel_offset = 0;
9471         ctsio->kern_data_len = min(data_len, alloc_len);
9472         ctsio->kern_total_len = ctsio->kern_data_len;
9473
9474         /*
9475          * The control device is always connected.  The disk device, on the
9476          * other hand, may not be online all the time.
9477          */
9478         if (lun != NULL)
9479                 eid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9480                                      lun->be_lun->lun_type;
9481         else
9482                 eid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9483         eid_ptr->page_code = SVPD_EXTENDED_INQUIRY_DATA;
9484         scsi_ulto2b(data_len - 4, eid_ptr->page_length);
9485         /*
9486          * We support head of queue, ordered and simple tags.
9487          */
9488         eid_ptr->flags2 = SVPD_EID_HEADSUP | SVPD_EID_ORDSUP | SVPD_EID_SIMPSUP;
9489         /*
9490          * Volatile cache supported.
9491          */
9492         eid_ptr->flags3 = SVPD_EID_V_SUP;
9493
9494         /*
9495          * This means that we clear the REPORTED LUNS DATA HAS CHANGED unit
9496          * attention for a particular IT nexus on all LUNs once we report
9497          * it to that nexus once.  This bit is required as of SPC-4.
9498          */
9499         eid_ptr->flags4 = SVPD_EID_LUICLR;
9500
9501         /*
9502          * We support revert to defaults (RTD) bit in MODE SELECT.
9503          */
9504         eid_ptr->flags5 = SVPD_EID_RTD_SUP;
9505
9506         /*
9507          * XXX KDM in order to correctly answer this, we would need
9508          * information from the SIM to determine how much sense data it
9509          * can send.  So this would really be a path inquiry field, most
9510          * likely.  This can be set to a maximum of 252 according to SPC-4,
9511          * but the hardware may or may not be able to support that much.
9512          * 0 just means that the maximum sense data length is not reported.
9513          */
9514         eid_ptr->max_sense_length = 0;
9515
9516         ctl_set_success(ctsio);
9517         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9518         ctsio->be_move_done = ctl_config_move_done;
9519         ctl_datamove((union ctl_io *)ctsio);
9520         return (CTL_RETVAL_COMPLETE);
9521 }
9522
9523 static int
9524 ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len)
9525 {
9526         struct ctl_lun *lun = CTL_LUN(ctsio);
9527         struct scsi_vpd_mode_page_policy *mpp_ptr;
9528         int data_len;
9529
9530         data_len = sizeof(struct scsi_vpd_mode_page_policy) +
9531             sizeof(struct scsi_vpd_mode_page_policy_descr);
9532
9533         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9534         mpp_ptr = (struct scsi_vpd_mode_page_policy *)ctsio->kern_data_ptr;
9535         ctsio->kern_rel_offset = 0;
9536         ctsio->kern_sg_entries = 0;
9537         ctsio->kern_data_len = min(data_len, alloc_len);
9538         ctsio->kern_total_len = ctsio->kern_data_len;
9539
9540         /*
9541          * The control device is always connected.  The disk device, on the
9542          * other hand, may not be online all the time.
9543          */
9544         if (lun != NULL)
9545                 mpp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9546                                      lun->be_lun->lun_type;
9547         else
9548                 mpp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9549         mpp_ptr->page_code = SVPD_MODE_PAGE_POLICY;
9550         scsi_ulto2b(data_len - 4, mpp_ptr->page_length);
9551         mpp_ptr->descr[0].page_code = 0x3f;
9552         mpp_ptr->descr[0].subpage_code = 0xff;
9553         mpp_ptr->descr[0].policy = SVPD_MPP_SHARED;
9554
9555         ctl_set_success(ctsio);
9556         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9557         ctsio->be_move_done = ctl_config_move_done;
9558         ctl_datamove((union ctl_io *)ctsio);
9559         return (CTL_RETVAL_COMPLETE);
9560 }
9561
9562 /*
9563  * SCSI VPD page 0x83, the Device Identification page.
9564  */
9565 static int
9566 ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
9567 {
9568         struct ctl_softc *softc = CTL_SOFTC(ctsio);
9569         struct ctl_port *port = CTL_PORT(ctsio);
9570         struct ctl_lun *lun = CTL_LUN(ctsio);
9571         struct scsi_vpd_device_id *devid_ptr;
9572         struct scsi_vpd_id_descriptor *desc;
9573         int data_len, g;
9574         uint8_t proto;
9575
9576         data_len = sizeof(struct scsi_vpd_device_id) +
9577             sizeof(struct scsi_vpd_id_descriptor) +
9578                 sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
9579             sizeof(struct scsi_vpd_id_descriptor) +
9580                 sizeof(struct scsi_vpd_id_trgt_port_grp_id);
9581         if (lun && lun->lun_devid)
9582                 data_len += lun->lun_devid->len;
9583         if (port && port->port_devid)
9584                 data_len += port->port_devid->len;
9585         if (port && port->target_devid)
9586                 data_len += port->target_devid->len;
9587
9588         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9589         devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
9590         ctsio->kern_sg_entries = 0;
9591         ctsio->kern_rel_offset = 0;
9592         ctsio->kern_sg_entries = 0;
9593         ctsio->kern_data_len = min(data_len, alloc_len);
9594         ctsio->kern_total_len = ctsio->kern_data_len;
9595
9596         /*
9597          * The control device is always connected.  The disk device, on the
9598          * other hand, may not be online all the time.
9599          */
9600         if (lun != NULL)
9601                 devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9602                                      lun->be_lun->lun_type;
9603         else
9604                 devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9605         devid_ptr->page_code = SVPD_DEVICE_ID;
9606         scsi_ulto2b(data_len - 4, devid_ptr->length);
9607
9608         if (port && port->port_type == CTL_PORT_FC)
9609                 proto = SCSI_PROTO_FC << 4;
9610         else if (port && port->port_type == CTL_PORT_SAS)
9611                 proto = SCSI_PROTO_SAS << 4;
9612         else if (port && port->port_type == CTL_PORT_ISCSI)
9613                 proto = SCSI_PROTO_ISCSI << 4;
9614         else
9615                 proto = SCSI_PROTO_SPI << 4;
9616         desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
9617
9618         /*
9619          * We're using a LUN association here.  i.e., this device ID is a
9620          * per-LUN identifier.
9621          */
9622         if (lun && lun->lun_devid) {
9623                 memcpy(desc, lun->lun_devid->data, lun->lun_devid->len);
9624                 desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9625                     lun->lun_devid->len);
9626         }
9627
9628         /*
9629          * This is for the WWPN which is a port association.
9630          */
9631         if (port && port->port_devid) {
9632                 memcpy(desc, port->port_devid->data, port->port_devid->len);
9633                 desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9634                     port->port_devid->len);
9635         }
9636
9637         /*
9638          * This is for the Relative Target Port(type 4h) identifier
9639          */
9640         desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9641         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9642             SVPD_ID_TYPE_RELTARG;
9643         desc->length = 4;
9644         scsi_ulto2b(ctsio->io_hdr.nexus.targ_port, &desc->identifier[2]);
9645         desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9646             sizeof(struct scsi_vpd_id_rel_trgt_port_id));
9647
9648         /*
9649          * This is for the Target Port Group(type 5h) identifier
9650          */
9651         desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9652         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9653             SVPD_ID_TYPE_TPORTGRP;
9654         desc->length = 4;
9655         if (softc->is_single ||
9656             (port && port->status & CTL_PORT_STATUS_HA_SHARED))
9657                 g = 1;
9658         else
9659                 g = 2 + ctsio->io_hdr.nexus.targ_port / softc->port_cnt;
9660         scsi_ulto2b(g, &desc->identifier[2]);
9661         desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9662             sizeof(struct scsi_vpd_id_trgt_port_grp_id));
9663
9664         /*
9665          * This is for the Target identifier
9666          */
9667         if (port && port->target_devid) {
9668                 memcpy(desc, port->target_devid->data, port->target_devid->len);
9669         }
9670
9671         ctl_set_success(ctsio);
9672         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9673         ctsio->be_move_done = ctl_config_move_done;
9674         ctl_datamove((union ctl_io *)ctsio);
9675         return (CTL_RETVAL_COMPLETE);
9676 }
9677
9678 static int
9679 ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio, int alloc_len)
9680 {
9681         struct ctl_softc *softc = CTL_SOFTC(ctsio);
9682         struct ctl_lun *lun = CTL_LUN(ctsio);
9683         struct scsi_vpd_scsi_ports *sp;
9684         struct scsi_vpd_port_designation *pd;
9685         struct scsi_vpd_port_designation_cont *pdc;
9686         struct ctl_port *port;
9687         int data_len, num_target_ports, iid_len, id_len;
9688
9689         num_target_ports = 0;
9690         iid_len = 0;
9691         id_len = 0;
9692         mtx_lock(&softc->ctl_lock);
9693         STAILQ_FOREACH(port, &softc->port_list, links) {
9694                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
9695                         continue;
9696                 if (lun != NULL &&
9697                     ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
9698                         continue;
9699                 num_target_ports++;
9700                 if (port->init_devid)
9701                         iid_len += port->init_devid->len;
9702                 if (port->port_devid)
9703                         id_len += port->port_devid->len;
9704         }
9705         mtx_unlock(&softc->ctl_lock);
9706
9707         data_len = sizeof(struct scsi_vpd_scsi_ports) +
9708             num_target_ports * (sizeof(struct scsi_vpd_port_designation) +
9709              sizeof(struct scsi_vpd_port_designation_cont)) + iid_len + id_len;
9710         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9711         sp = (struct scsi_vpd_scsi_ports *)ctsio->kern_data_ptr;
9712         ctsio->kern_sg_entries = 0;
9713         ctsio->kern_rel_offset = 0;
9714         ctsio->kern_sg_entries = 0;
9715         ctsio->kern_data_len = min(data_len, alloc_len);
9716         ctsio->kern_total_len = ctsio->kern_data_len;
9717
9718         /*
9719          * The control device is always connected.  The disk device, on the
9720          * other hand, may not be online all the time.  Need to change this
9721          * to figure out whether the disk device is actually online or not.
9722          */
9723         if (lun != NULL)
9724                 sp->device = (SID_QUAL_LU_CONNECTED << 5) |
9725                                   lun->be_lun->lun_type;
9726         else
9727                 sp->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9728
9729         sp->page_code = SVPD_SCSI_PORTS;
9730         scsi_ulto2b(data_len - sizeof(struct scsi_vpd_scsi_ports),
9731             sp->page_length);
9732         pd = &sp->design[0];
9733
9734         mtx_lock(&softc->ctl_lock);
9735         STAILQ_FOREACH(port, &softc->port_list, links) {
9736                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
9737                         continue;
9738                 if (lun != NULL &&
9739                     ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
9740                         continue;
9741                 scsi_ulto2b(port->targ_port, pd->relative_port_id);
9742                 if (port->init_devid) {
9743                         iid_len = port->init_devid->len;
9744                         memcpy(pd->initiator_transportid,
9745                             port->init_devid->data, port->init_devid->len);
9746                 } else
9747                         iid_len = 0;
9748                 scsi_ulto2b(iid_len, pd->initiator_transportid_length);
9749                 pdc = (struct scsi_vpd_port_designation_cont *)
9750                     (&pd->initiator_transportid[iid_len]);
9751                 if (port->port_devid) {
9752                         id_len = port->port_devid->len;
9753                         memcpy(pdc->target_port_descriptors,
9754                             port->port_devid->data, port->port_devid->len);
9755                 } else
9756                         id_len = 0;
9757                 scsi_ulto2b(id_len, pdc->target_port_descriptors_length);
9758                 pd = (struct scsi_vpd_port_designation *)
9759                     ((uint8_t *)pdc->target_port_descriptors + id_len);
9760         }
9761         mtx_unlock(&softc->ctl_lock);
9762
9763         ctl_set_success(ctsio);
9764         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9765         ctsio->be_move_done = ctl_config_move_done;
9766         ctl_datamove((union ctl_io *)ctsio);
9767         return (CTL_RETVAL_COMPLETE);
9768 }
9769
9770 static int
9771 ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio, int alloc_len)
9772 {
9773         struct ctl_lun *lun = CTL_LUN(ctsio);
9774         struct scsi_vpd_block_limits *bl_ptr;
9775         uint64_t ival;
9776
9777         ctsio->kern_data_ptr = malloc(sizeof(*bl_ptr), M_CTL, M_WAITOK | M_ZERO);
9778         bl_ptr = (struct scsi_vpd_block_limits *)ctsio->kern_data_ptr;
9779         ctsio->kern_sg_entries = 0;
9780         ctsio->kern_rel_offset = 0;
9781         ctsio->kern_sg_entries = 0;
9782         ctsio->kern_data_len = min(sizeof(*bl_ptr), alloc_len);
9783         ctsio->kern_total_len = ctsio->kern_data_len;
9784
9785         /*
9786          * The control device is always connected.  The disk device, on the
9787          * other hand, may not be online all the time.  Need to change this
9788          * to figure out whether the disk device is actually online or not.
9789          */
9790         if (lun != NULL)
9791                 bl_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9792                                   lun->be_lun->lun_type;
9793         else
9794                 bl_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9795
9796         bl_ptr->page_code = SVPD_BLOCK_LIMITS;
9797         scsi_ulto2b(sizeof(*bl_ptr) - 4, bl_ptr->page_length);
9798         bl_ptr->max_cmp_write_len = 0xff;
9799         scsi_ulto4b(0xffffffff, bl_ptr->max_txfer_len);
9800         if (lun != NULL) {
9801                 scsi_ulto4b(lun->be_lun->opttxferlen, bl_ptr->opt_txfer_len);
9802                 if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
9803                         ival = 0xffffffff;
9804                         ctl_get_opt_number(&lun->be_lun->options,
9805                             "unmap_max_lba", &ival);
9806                         scsi_ulto4b(ival, bl_ptr->max_unmap_lba_cnt);
9807                         ival = 0xffffffff;
9808                         ctl_get_opt_number(&lun->be_lun->options,
9809                             "unmap_max_descr", &ival);
9810                         scsi_ulto4b(ival, bl_ptr->max_unmap_blk_cnt);
9811                         if (lun->be_lun->ublockexp != 0) {
9812                                 scsi_ulto4b((1 << lun->be_lun->ublockexp),
9813                                     bl_ptr->opt_unmap_grain);
9814                                 scsi_ulto4b(0x80000000 | lun->be_lun->ublockoff,
9815                                     bl_ptr->unmap_grain_align);
9816                         }
9817                 }
9818                 scsi_ulto4b(lun->be_lun->atomicblock,
9819                     bl_ptr->max_atomic_transfer_length);
9820                 scsi_ulto4b(0, bl_ptr->atomic_alignment);
9821                 scsi_ulto4b(0, bl_ptr->atomic_transfer_length_granularity);
9822                 scsi_ulto4b(0, bl_ptr->max_atomic_transfer_length_with_atomic_boundary);
9823                 scsi_ulto4b(0, bl_ptr->max_atomic_boundary_size);
9824                 ival = UINT64_MAX;
9825                 ctl_get_opt_number(&lun->be_lun->options, "write_same_max_lba", &ival);
9826                 scsi_u64to8b(ival, bl_ptr->max_write_same_length);
9827         }
9828
9829         ctl_set_success(ctsio);
9830         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9831         ctsio->be_move_done = ctl_config_move_done;
9832         ctl_datamove((union ctl_io *)ctsio);
9833         return (CTL_RETVAL_COMPLETE);
9834 }
9835
9836 static int
9837 ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len)
9838 {
9839         struct ctl_lun *lun = CTL_LUN(ctsio);
9840         struct scsi_vpd_block_device_characteristics *bdc_ptr;
9841         const char *value;
9842         u_int i;
9843
9844         ctsio->kern_data_ptr = malloc(sizeof(*bdc_ptr), M_CTL, M_WAITOK | M_ZERO);
9845         bdc_ptr = (struct scsi_vpd_block_device_characteristics *)ctsio->kern_data_ptr;
9846         ctsio->kern_sg_entries = 0;
9847         ctsio->kern_rel_offset = 0;
9848         ctsio->kern_data_len = min(sizeof(*bdc_ptr), alloc_len);
9849         ctsio->kern_total_len = ctsio->kern_data_len;
9850
9851         /*
9852          * The control device is always connected.  The disk device, on the
9853          * other hand, may not be online all the time.  Need to change this
9854          * to figure out whether the disk device is actually online or not.
9855          */
9856         if (lun != NULL)
9857                 bdc_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9858                                   lun->be_lun->lun_type;
9859         else
9860                 bdc_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9861         bdc_ptr->page_code = SVPD_BDC;
9862         scsi_ulto2b(sizeof(*bdc_ptr) - 4, bdc_ptr->page_length);
9863         if (lun != NULL &&
9864             (value = ctl_get_opt(&lun->be_lun->options, "rpm")) != NULL)
9865                 i = strtol(value, NULL, 0);
9866         else
9867                 i = CTL_DEFAULT_ROTATION_RATE;
9868         scsi_ulto2b(i, bdc_ptr->medium_rotation_rate);
9869         if (lun != NULL &&
9870             (value = ctl_get_opt(&lun->be_lun->options, "formfactor")) != NULL)
9871                 i = strtol(value, NULL, 0);
9872         else
9873                 i = 0;
9874         bdc_ptr->wab_wac_ff = (i & 0x0f);
9875         bdc_ptr->flags = SVPD_FUAB | SVPD_VBULS;
9876
9877         ctl_set_success(ctsio);
9878         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9879         ctsio->be_move_done = ctl_config_move_done;
9880         ctl_datamove((union ctl_io *)ctsio);
9881         return (CTL_RETVAL_COMPLETE);
9882 }
9883
9884 static int
9885 ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len)
9886 {
9887         struct ctl_lun *lun = CTL_LUN(ctsio);
9888         struct scsi_vpd_logical_block_prov *lbp_ptr;
9889         const char *value;
9890
9891         ctsio->kern_data_ptr = malloc(sizeof(*lbp_ptr), M_CTL, M_WAITOK | M_ZERO);
9892         lbp_ptr = (struct scsi_vpd_logical_block_prov *)ctsio->kern_data_ptr;
9893         ctsio->kern_sg_entries = 0;
9894         ctsio->kern_rel_offset = 0;
9895         ctsio->kern_data_len = min(sizeof(*lbp_ptr), alloc_len);
9896         ctsio->kern_total_len = ctsio->kern_data_len;
9897
9898         /*
9899          * The control device is always connected.  The disk device, on the
9900          * other hand, may not be online all the time.  Need to change this
9901          * to figure out whether the disk device is actually online or not.
9902          */
9903         if (lun != NULL)
9904                 lbp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9905                                   lun->be_lun->lun_type;
9906         else
9907                 lbp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9908
9909         lbp_ptr->page_code = SVPD_LBP;
9910         scsi_ulto2b(sizeof(*lbp_ptr) - 4, lbp_ptr->page_length);
9911         lbp_ptr->threshold_exponent = CTL_LBP_EXPONENT;
9912         if (lun != NULL && lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
9913                 lbp_ptr->flags = SVPD_LBP_UNMAP | SVPD_LBP_WS16 |
9914                     SVPD_LBP_WS10 | SVPD_LBP_RZ | SVPD_LBP_ANC_SUP;
9915                 value = ctl_get_opt(&lun->be_lun->options, "provisioning_type");
9916                 if (value != NULL) {
9917                         if (strcmp(value, "resource") == 0)
9918                                 lbp_ptr->prov_type = SVPD_LBP_RESOURCE;
9919                         else if (strcmp(value, "thin") == 0)
9920                                 lbp_ptr->prov_type = SVPD_LBP_THIN;
9921                 } else
9922                         lbp_ptr->prov_type = SVPD_LBP_THIN;
9923         }
9924
9925         ctl_set_success(ctsio);
9926         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9927         ctsio->be_move_done = ctl_config_move_done;
9928         ctl_datamove((union ctl_io *)ctsio);
9929         return (CTL_RETVAL_COMPLETE);
9930 }
9931
9932 /*
9933  * INQUIRY with the EVPD bit set.
9934  */
9935 static int
9936 ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
9937 {
9938         struct ctl_lun *lun = CTL_LUN(ctsio);
9939         struct scsi_inquiry *cdb;
9940         int alloc_len, retval;
9941
9942         cdb = (struct scsi_inquiry *)ctsio->cdb;
9943         alloc_len = scsi_2btoul(cdb->length);
9944
9945         switch (cdb->page_code) {
9946         case SVPD_SUPPORTED_PAGES:
9947                 retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
9948                 break;
9949         case SVPD_UNIT_SERIAL_NUMBER:
9950                 retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
9951                 break;
9952         case SVPD_DEVICE_ID:
9953                 retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
9954                 break;
9955         case SVPD_EXTENDED_INQUIRY_DATA:
9956                 retval = ctl_inquiry_evpd_eid(ctsio, alloc_len);
9957                 break;
9958         case SVPD_MODE_PAGE_POLICY:
9959                 retval = ctl_inquiry_evpd_mpp(ctsio, alloc_len);
9960                 break;
9961         case SVPD_SCSI_PORTS:
9962                 retval = ctl_inquiry_evpd_scsi_ports(ctsio, alloc_len);
9963                 break;
9964         case SVPD_SCSI_TPC:
9965                 retval = ctl_inquiry_evpd_tpc(ctsio, alloc_len);
9966                 break;
9967         case SVPD_BLOCK_LIMITS:
9968                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
9969                         goto err;
9970                 retval = ctl_inquiry_evpd_block_limits(ctsio, alloc_len);
9971                 break;
9972         case SVPD_BDC:
9973                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
9974                         goto err;
9975                 retval = ctl_inquiry_evpd_bdc(ctsio, alloc_len);
9976                 break;
9977         case SVPD_LBP:
9978                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
9979                         goto err;
9980                 retval = ctl_inquiry_evpd_lbp(ctsio, alloc_len);
9981                 break;
9982         default:
9983 err:
9984                 ctl_set_invalid_field(ctsio,
9985                                       /*sks_valid*/ 1,
9986                                       /*command*/ 1,
9987                                       /*field*/ 2,
9988                                       /*bit_valid*/ 0,
9989                                       /*bit*/ 0);
9990                 ctl_done((union ctl_io *)ctsio);
9991                 retval = CTL_RETVAL_COMPLETE;
9992                 break;
9993         }
9994
9995         return (retval);
9996 }
9997
9998 /*
9999  * Standard INQUIRY data.
10000  */
10001 static int
10002 ctl_inquiry_std(struct ctl_scsiio *ctsio)
10003 {
10004         struct ctl_softc *softc = CTL_SOFTC(ctsio);
10005         struct ctl_port *port = CTL_PORT(ctsio);
10006         struct ctl_lun *lun = CTL_LUN(ctsio);
10007         struct scsi_inquiry_data *inq_ptr;
10008         struct scsi_inquiry *cdb;
10009         char *val;
10010         uint32_t alloc_len, data_len;
10011         ctl_port_type port_type;
10012
10013         port_type = port->port_type;
10014         if (port_type == CTL_PORT_IOCTL || port_type == CTL_PORT_INTERNAL)
10015                 port_type = CTL_PORT_SCSI;
10016
10017         cdb = (struct scsi_inquiry *)ctsio->cdb;
10018         alloc_len = scsi_2btoul(cdb->length);
10019
10020         /*
10021          * We malloc the full inquiry data size here and fill it
10022          * in.  If the user only asks for less, we'll give him
10023          * that much.
10024          */
10025         data_len = offsetof(struct scsi_inquiry_data, vendor_specific1);
10026         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10027         inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
10028         ctsio->kern_sg_entries = 0;
10029         ctsio->kern_rel_offset = 0;
10030         ctsio->kern_data_len = min(data_len, alloc_len);
10031         ctsio->kern_total_len = ctsio->kern_data_len;
10032
10033         if (lun != NULL) {
10034                 if ((lun->flags & CTL_LUN_PRIMARY_SC) ||
10035                     softc->ha_link >= CTL_HA_LINK_UNKNOWN) {
10036                         inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10037                             lun->be_lun->lun_type;
10038                 } else {
10039                         inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) |
10040                             lun->be_lun->lun_type;
10041                 }
10042                 if (lun->flags & CTL_LUN_REMOVABLE)
10043                         inq_ptr->dev_qual2 |= SID_RMB;
10044         } else
10045                 inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE;
10046
10047         /* RMB in byte 2 is 0 */
10048         inq_ptr->version = SCSI_REV_SPC5;
10049
10050         /*
10051          * According to SAM-3, even if a device only supports a single
10052          * level of LUN addressing, it should still set the HISUP bit:
10053          *
10054          * 4.9.1 Logical unit numbers overview
10055          *
10056          * All logical unit number formats described in this standard are
10057          * hierarchical in structure even when only a single level in that
10058          * hierarchy is used. The HISUP bit shall be set to one in the
10059          * standard INQUIRY data (see SPC-2) when any logical unit number
10060          * format described in this standard is used.  Non-hierarchical
10061          * formats are outside the scope of this standard.
10062          *
10063          * Therefore we set the HiSup bit here.
10064          *
10065          * The response format is 2, per SPC-3.
10066          */
10067         inq_ptr->response_format = SID_HiSup | 2;
10068
10069         inq_ptr->additional_length = data_len -
10070             (offsetof(struct scsi_inquiry_data, additional_length) + 1);
10071         CTL_DEBUG_PRINT(("additional_length = %d\n",
10072                          inq_ptr->additional_length));
10073
10074         inq_ptr->spc3_flags = SPC3_SID_3PC | SPC3_SID_TPGS_IMPLICIT;
10075         if (port_type == CTL_PORT_SCSI)
10076                 inq_ptr->spc2_flags = SPC2_SID_ADDR16;
10077         inq_ptr->spc2_flags |= SPC2_SID_MultiP;
10078         inq_ptr->flags = SID_CmdQue;
10079         if (port_type == CTL_PORT_SCSI)
10080                 inq_ptr->flags |= SID_WBus16 | SID_Sync;
10081
10082         /*
10083          * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
10084          * We have 8 bytes for the vendor name, and 16 bytes for the device
10085          * name and 4 bytes for the revision.
10086          */
10087         if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10088             "vendor")) == NULL) {
10089                 strncpy(inq_ptr->vendor, CTL_VENDOR, sizeof(inq_ptr->vendor));
10090         } else {
10091                 memset(inq_ptr->vendor, ' ', sizeof(inq_ptr->vendor));
10092                 strncpy(inq_ptr->vendor, val,
10093                     min(sizeof(inq_ptr->vendor), strlen(val)));
10094         }
10095         if (lun == NULL) {
10096                 strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10097                     sizeof(inq_ptr->product));
10098         } else if ((val = ctl_get_opt(&lun->be_lun->options, "product")) == NULL) {
10099                 switch (lun->be_lun->lun_type) {
10100                 case T_DIRECT:
10101                         strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10102                             sizeof(inq_ptr->product));
10103                         break;
10104                 case T_PROCESSOR:
10105                         strncpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT,
10106                             sizeof(inq_ptr->product));
10107                         break;
10108                 case T_CDROM:
10109                         strncpy(inq_ptr->product, CTL_CDROM_PRODUCT,
10110                             sizeof(inq_ptr->product));
10111                         break;
10112                 default:
10113                         strncpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT,
10114                             sizeof(inq_ptr->product));
10115                         break;
10116                 }
10117         } else {
10118                 memset(inq_ptr->product, ' ', sizeof(inq_ptr->product));
10119                 strncpy(inq_ptr->product, val,
10120                     min(sizeof(inq_ptr->product), strlen(val)));
10121         }
10122
10123         /*
10124          * XXX make this a macro somewhere so it automatically gets
10125          * incremented when we make changes.
10126          */
10127         if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10128             "revision")) == NULL) {
10129                 strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
10130         } else {
10131                 memset(inq_ptr->revision, ' ', sizeof(inq_ptr->revision));
10132                 strncpy(inq_ptr->revision, val,
10133                     min(sizeof(inq_ptr->revision), strlen(val)));
10134         }
10135
10136         /*
10137          * For parallel SCSI, we support double transition and single
10138          * transition clocking.  We also support QAS (Quick Arbitration
10139          * and Selection) and Information Unit transfers on both the
10140          * control and array devices.
10141          */
10142         if (port_type == CTL_PORT_SCSI)
10143                 inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
10144                                     SID_SPI_IUS;
10145
10146         /* SAM-6 (no version claimed) */
10147         scsi_ulto2b(0x00C0, inq_ptr->version1);
10148         /* SPC-5 (no version claimed) */
10149         scsi_ulto2b(0x05C0, inq_ptr->version2);
10150         if (port_type == CTL_PORT_FC) {
10151                 /* FCP-2 ANSI INCITS.350:2003 */
10152                 scsi_ulto2b(0x0917, inq_ptr->version3);
10153         } else if (port_type == CTL_PORT_SCSI) {
10154                 /* SPI-4 ANSI INCITS.362:200x */
10155                 scsi_ulto2b(0x0B56, inq_ptr->version3);
10156         } else if (port_type == CTL_PORT_ISCSI) {
10157                 /* iSCSI (no version claimed) */
10158                 scsi_ulto2b(0x0960, inq_ptr->version3);
10159         } else if (port_type == CTL_PORT_SAS) {
10160                 /* SAS (no version claimed) */
10161                 scsi_ulto2b(0x0BE0, inq_ptr->version3);
10162         } else if (port_type == CTL_PORT_UMASS) {
10163                 /* USB Mass Storage Class Bulk-Only Transport, Revision 1.0 */
10164                 scsi_ulto2b(0x1730, inq_ptr->version3);
10165         }
10166
10167         if (lun == NULL) {
10168                 /* SBC-4 (no version claimed) */
10169                 scsi_ulto2b(0x0600, inq_ptr->version4);
10170         } else {
10171                 switch (lun->be_lun->lun_type) {
10172                 case T_DIRECT:
10173                         /* SBC-4 (no version claimed) */
10174                         scsi_ulto2b(0x0600, inq_ptr->version4);
10175                         break;
10176                 case T_PROCESSOR:
10177                         break;
10178                 case T_CDROM:
10179                         /* MMC-6 (no version claimed) */
10180                         scsi_ulto2b(0x04E0, inq_ptr->version4);
10181                         break;
10182                 default:
10183                         break;
10184                 }
10185         }
10186
10187         ctl_set_success(ctsio);
10188         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10189         ctsio->be_move_done = ctl_config_move_done;
10190         ctl_datamove((union ctl_io *)ctsio);
10191         return (CTL_RETVAL_COMPLETE);
10192 }
10193
10194 int
10195 ctl_inquiry(struct ctl_scsiio *ctsio)
10196 {
10197         struct scsi_inquiry *cdb;
10198         int retval;
10199
10200         CTL_DEBUG_PRINT(("ctl_inquiry\n"));
10201
10202         cdb = (struct scsi_inquiry *)ctsio->cdb;
10203         if (cdb->byte2 & SI_EVPD)
10204                 retval = ctl_inquiry_evpd(ctsio);
10205         else if (cdb->page_code == 0)
10206                 retval = ctl_inquiry_std(ctsio);
10207         else {
10208                 ctl_set_invalid_field(ctsio,
10209                                       /*sks_valid*/ 1,
10210                                       /*command*/ 1,
10211                                       /*field*/ 2,
10212                                       /*bit_valid*/ 0,
10213                                       /*bit*/ 0);
10214                 ctl_done((union ctl_io *)ctsio);
10215                 return (CTL_RETVAL_COMPLETE);
10216         }
10217
10218         return (retval);
10219 }
10220
10221 int
10222 ctl_get_config(struct ctl_scsiio *ctsio)
10223 {
10224         struct ctl_lun *lun = CTL_LUN(ctsio);
10225         struct scsi_get_config_header *hdr;
10226         struct scsi_get_config_feature *feature;
10227         struct scsi_get_config *cdb;
10228         uint32_t alloc_len, data_len;
10229         int rt, starting;
10230
10231         cdb = (struct scsi_get_config *)ctsio->cdb;
10232         rt = (cdb->rt & SGC_RT_MASK);
10233         starting = scsi_2btoul(cdb->starting_feature);
10234         alloc_len = scsi_2btoul(cdb->length);
10235
10236         data_len = sizeof(struct scsi_get_config_header) +
10237             sizeof(struct scsi_get_config_feature) + 8 +
10238             sizeof(struct scsi_get_config_feature) + 8 +
10239             sizeof(struct scsi_get_config_feature) + 4 +
10240             sizeof(struct scsi_get_config_feature) + 4 +
10241             sizeof(struct scsi_get_config_feature) + 8 +
10242             sizeof(struct scsi_get_config_feature) +
10243             sizeof(struct scsi_get_config_feature) + 4 +
10244             sizeof(struct scsi_get_config_feature) + 4 +
10245             sizeof(struct scsi_get_config_feature) + 4 +
10246             sizeof(struct scsi_get_config_feature) + 4 +
10247             sizeof(struct scsi_get_config_feature) + 4 +
10248             sizeof(struct scsi_get_config_feature) + 4;
10249         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10250         ctsio->kern_sg_entries = 0;
10251         ctsio->kern_rel_offset = 0;
10252
10253         hdr = (struct scsi_get_config_header *)ctsio->kern_data_ptr;
10254         if (lun->flags & CTL_LUN_NO_MEDIA)
10255                 scsi_ulto2b(0x0000, hdr->current_profile);
10256         else
10257                 scsi_ulto2b(0x0010, hdr->current_profile);
10258         feature = (struct scsi_get_config_feature *)(hdr + 1);
10259
10260         if (starting > 0x003b)
10261                 goto done;
10262         if (starting > 0x003a)
10263                 goto f3b;
10264         if (starting > 0x002b)
10265                 goto f3a;
10266         if (starting > 0x002a)
10267                 goto f2b;
10268         if (starting > 0x001f)
10269                 goto f2a;
10270         if (starting > 0x001e)
10271                 goto f1f;
10272         if (starting > 0x001d)
10273                 goto f1e;
10274         if (starting > 0x0010)
10275                 goto f1d;
10276         if (starting > 0x0003)
10277                 goto f10;
10278         if (starting > 0x0002)
10279                 goto f3;
10280         if (starting > 0x0001)
10281                 goto f2;
10282         if (starting > 0x0000)
10283                 goto f1;
10284
10285         /* Profile List */
10286         scsi_ulto2b(0x0000, feature->feature_code);
10287         feature->flags = SGC_F_PERSISTENT | SGC_F_CURRENT;
10288         feature->add_length = 8;
10289         scsi_ulto2b(0x0008, &feature->feature_data[0]); /* CD-ROM */
10290         feature->feature_data[2] = 0x00;
10291         scsi_ulto2b(0x0010, &feature->feature_data[4]); /* DVD-ROM */
10292         feature->feature_data[6] = 0x01;
10293         feature = (struct scsi_get_config_feature *)
10294             &feature->feature_data[feature->add_length];
10295
10296 f1:     /* Core */
10297         scsi_ulto2b(0x0001, feature->feature_code);
10298         feature->flags = 0x08 | SGC_F_PERSISTENT | SGC_F_CURRENT;
10299         feature->add_length = 8;
10300         scsi_ulto4b(0x00000000, &feature->feature_data[0]);
10301         feature->feature_data[4] = 0x03;
10302         feature = (struct scsi_get_config_feature *)
10303             &feature->feature_data[feature->add_length];
10304
10305 f2:     /* Morphing */
10306         scsi_ulto2b(0x0002, feature->feature_code);
10307         feature->flags = 0x04 | SGC_F_PERSISTENT | SGC_F_CURRENT;
10308         feature->add_length = 4;
10309         feature->feature_data[0] = 0x02;
10310         feature = (struct scsi_get_config_feature *)
10311             &feature->feature_data[feature->add_length];
10312
10313 f3:     /* Removable Medium */
10314         scsi_ulto2b(0x0003, feature->feature_code);
10315         feature->flags = 0x04 | SGC_F_PERSISTENT | SGC_F_CURRENT;
10316         feature->add_length = 4;
10317         feature->feature_data[0] = 0x39;
10318         feature = (struct scsi_get_config_feature *)
10319             &feature->feature_data[feature->add_length];
10320
10321         if (rt == SGC_RT_CURRENT && (lun->flags & CTL_LUN_NO_MEDIA))
10322                 goto done;
10323
10324 f10:    /* Random Read */
10325         scsi_ulto2b(0x0010, feature->feature_code);
10326         feature->flags = 0x00;
10327         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10328                 feature->flags |= SGC_F_CURRENT;
10329         feature->add_length = 8;
10330         scsi_ulto4b(lun->be_lun->blocksize, &feature->feature_data[0]);
10331         scsi_ulto2b(1, &feature->feature_data[4]);
10332         feature->feature_data[6] = 0x00;
10333         feature = (struct scsi_get_config_feature *)
10334             &feature->feature_data[feature->add_length];
10335
10336 f1d:    /* Multi-Read */
10337         scsi_ulto2b(0x001D, 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 = 0;
10342         feature = (struct scsi_get_config_feature *)
10343             &feature->feature_data[feature->add_length];
10344
10345 f1e:    /* CD Read */
10346         scsi_ulto2b(0x001E, feature->feature_code);
10347         feature->flags = 0x00;
10348         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10349                 feature->flags |= SGC_F_CURRENT;
10350         feature->add_length = 4;
10351         feature->feature_data[0] = 0x00;
10352         feature = (struct scsi_get_config_feature *)
10353             &feature->feature_data[feature->add_length];
10354
10355 f1f:    /* DVD Read */
10356         scsi_ulto2b(0x001F, feature->feature_code);
10357         feature->flags = 0x08;
10358         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10359                 feature->flags |= SGC_F_CURRENT;
10360         feature->add_length = 4;
10361         feature->feature_data[0] = 0x01;
10362         feature->feature_data[2] = 0x03;
10363         feature = (struct scsi_get_config_feature *)
10364             &feature->feature_data[feature->add_length];
10365
10366 f2a:    /* DVD+RW */
10367         scsi_ulto2b(0x002A, feature->feature_code);
10368         feature->flags = 0x04;
10369         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10370                 feature->flags |= SGC_F_CURRENT;
10371         feature->add_length = 4;
10372         feature->feature_data[0] = 0x00;
10373         feature->feature_data[1] = 0x00;
10374         feature = (struct scsi_get_config_feature *)
10375             &feature->feature_data[feature->add_length];
10376
10377 f2b:    /* DVD+R */
10378         scsi_ulto2b(0x002B, feature->feature_code);
10379         feature->flags = 0x00;
10380         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10381                 feature->flags |= SGC_F_CURRENT;
10382         feature->add_length = 4;
10383         feature->feature_data[0] = 0x00;
10384         feature = (struct scsi_get_config_feature *)
10385             &feature->feature_data[feature->add_length];
10386
10387 f3a:    /* DVD+RW Dual Layer */
10388         scsi_ulto2b(0x003A, feature->feature_code);
10389         feature->flags = 0x00;
10390         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10391                 feature->flags |= SGC_F_CURRENT;
10392         feature->add_length = 4;
10393         feature->feature_data[0] = 0x00;
10394         feature->feature_data[1] = 0x00;
10395         feature = (struct scsi_get_config_feature *)
10396             &feature->feature_data[feature->add_length];
10397
10398 f3b:    /* DVD+R Dual Layer */
10399         scsi_ulto2b(0x003B, feature->feature_code);
10400         feature->flags = 0x00;
10401         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10402                 feature->flags |= SGC_F_CURRENT;
10403         feature->add_length = 4;
10404         feature->feature_data[0] = 0x00;
10405         feature = (struct scsi_get_config_feature *)
10406             &feature->feature_data[feature->add_length];
10407
10408 done:
10409         data_len = (uint8_t *)feature - (uint8_t *)hdr;
10410         if (rt == SGC_RT_SPECIFIC && data_len > 4) {
10411                 feature = (struct scsi_get_config_feature *)(hdr + 1);
10412                 if (scsi_2btoul(feature->feature_code) == starting)
10413                         feature = (struct scsi_get_config_feature *)
10414                             &feature->feature_data[feature->add_length];
10415                 data_len = (uint8_t *)feature - (uint8_t *)hdr;
10416         }
10417         scsi_ulto4b(data_len - 4, hdr->data_length);
10418         ctsio->kern_data_len = min(data_len, alloc_len);
10419         ctsio->kern_total_len = ctsio->kern_data_len;
10420
10421         ctl_set_success(ctsio);
10422         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10423         ctsio->be_move_done = ctl_config_move_done;
10424         ctl_datamove((union ctl_io *)ctsio);
10425         return (CTL_RETVAL_COMPLETE);
10426 }
10427
10428 int
10429 ctl_get_event_status(struct ctl_scsiio *ctsio)
10430 {
10431         struct scsi_get_event_status_header *hdr;
10432         struct scsi_get_event_status *cdb;
10433         uint32_t alloc_len, data_len;
10434         int notif_class;
10435
10436         cdb = (struct scsi_get_event_status *)ctsio->cdb;
10437         if ((cdb->byte2 & SGESN_POLLED) == 0) {
10438                 ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 1,
10439                     /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 0);
10440                 ctl_done((union ctl_io *)ctsio);
10441                 return (CTL_RETVAL_COMPLETE);
10442         }
10443         notif_class = cdb->notif_class;
10444         alloc_len = scsi_2btoul(cdb->length);
10445
10446         data_len = sizeof(struct scsi_get_event_status_header);
10447         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10448         ctsio->kern_sg_entries = 0;
10449         ctsio->kern_rel_offset = 0;
10450         ctsio->kern_data_len = min(data_len, alloc_len);
10451         ctsio->kern_total_len = ctsio->kern_data_len;
10452
10453         hdr = (struct scsi_get_event_status_header *)ctsio->kern_data_ptr;
10454         scsi_ulto2b(0, hdr->descr_length);
10455         hdr->nea_class = SGESN_NEA;
10456         hdr->supported_class = 0;
10457
10458         ctl_set_success(ctsio);
10459         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10460         ctsio->be_move_done = ctl_config_move_done;
10461         ctl_datamove((union ctl_io *)ctsio);
10462         return (CTL_RETVAL_COMPLETE);
10463 }
10464
10465 int
10466 ctl_mechanism_status(struct ctl_scsiio *ctsio)
10467 {
10468         struct scsi_mechanism_status_header *hdr;
10469         struct scsi_mechanism_status *cdb;
10470         uint32_t alloc_len, data_len;
10471
10472         cdb = (struct scsi_mechanism_status *)ctsio->cdb;
10473         alloc_len = scsi_2btoul(cdb->length);
10474
10475         data_len = sizeof(struct scsi_mechanism_status_header);
10476         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10477         ctsio->kern_sg_entries = 0;
10478         ctsio->kern_rel_offset = 0;
10479         ctsio->kern_data_len = min(data_len, alloc_len);
10480         ctsio->kern_total_len = ctsio->kern_data_len;
10481
10482         hdr = (struct scsi_mechanism_status_header *)ctsio->kern_data_ptr;
10483         hdr->state1 = 0x00;
10484         hdr->state2 = 0xe0;
10485         scsi_ulto3b(0, hdr->lba);
10486         hdr->slots_num = 0;
10487         scsi_ulto2b(0, hdr->slots_length);
10488
10489         ctl_set_success(ctsio);
10490         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10491         ctsio->be_move_done = ctl_config_move_done;
10492         ctl_datamove((union ctl_io *)ctsio);
10493         return (CTL_RETVAL_COMPLETE);
10494 }
10495
10496 static void
10497 ctl_ultomsf(uint32_t lba, uint8_t *buf)
10498 {
10499
10500         lba += 150;
10501         buf[0] = 0;
10502         buf[1] = bin2bcd((lba / 75) / 60);
10503         buf[2] = bin2bcd((lba / 75) % 60);
10504         buf[3] = bin2bcd(lba % 75);
10505 }
10506
10507 int
10508 ctl_read_toc(struct ctl_scsiio *ctsio)
10509 {
10510         struct ctl_lun *lun = CTL_LUN(ctsio);
10511         struct scsi_read_toc_hdr *hdr;
10512         struct scsi_read_toc_type01_descr *descr;
10513         struct scsi_read_toc *cdb;
10514         uint32_t alloc_len, data_len;
10515         int format, msf;
10516
10517         cdb = (struct scsi_read_toc *)ctsio->cdb;
10518         msf = (cdb->byte2 & CD_MSF) != 0;
10519         format = cdb->format;
10520         alloc_len = scsi_2btoul(cdb->data_len);
10521
10522         data_len = sizeof(struct scsi_read_toc_hdr);
10523         if (format == 0)
10524                 data_len += 2 * sizeof(struct scsi_read_toc_type01_descr);
10525         else
10526                 data_len += sizeof(struct scsi_read_toc_type01_descr);
10527         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10528         ctsio->kern_sg_entries = 0;
10529         ctsio->kern_rel_offset = 0;
10530         ctsio->kern_data_len = min(data_len, alloc_len);
10531         ctsio->kern_total_len = ctsio->kern_data_len;
10532
10533         hdr = (struct scsi_read_toc_hdr *)ctsio->kern_data_ptr;
10534         if (format == 0) {
10535                 scsi_ulto2b(0x12, hdr->data_length);
10536                 hdr->first = 1;
10537                 hdr->last = 1;
10538                 descr = (struct scsi_read_toc_type01_descr *)(hdr + 1);
10539                 descr->addr_ctl = 0x14;
10540                 descr->track_number = 1;
10541                 if (msf)
10542                         ctl_ultomsf(0, descr->track_start);
10543                 else
10544                         scsi_ulto4b(0, descr->track_start);
10545                 descr++;
10546                 descr->addr_ctl = 0x14;
10547                 descr->track_number = 0xaa;
10548                 if (msf)
10549                         ctl_ultomsf(lun->be_lun->maxlba+1, descr->track_start);
10550                 else
10551                         scsi_ulto4b(lun->be_lun->maxlba+1, descr->track_start);
10552         } else {
10553                 scsi_ulto2b(0x0a, hdr->data_length);
10554                 hdr->first = 1;
10555                 hdr->last = 1;
10556                 descr = (struct scsi_read_toc_type01_descr *)(hdr + 1);
10557                 descr->addr_ctl = 0x14;
10558                 descr->track_number = 1;
10559                 if (msf)
10560                         ctl_ultomsf(0, descr->track_start);
10561                 else
10562                         scsi_ulto4b(0, descr->track_start);
10563         }
10564
10565         ctl_set_success(ctsio);
10566         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10567         ctsio->be_move_done = ctl_config_move_done;
10568         ctl_datamove((union ctl_io *)ctsio);
10569         return (CTL_RETVAL_COMPLETE);
10570 }
10571
10572 /*
10573  * For known CDB types, parse the LBA and length.
10574  */
10575 static int
10576 ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len)
10577 {
10578         if (io->io_hdr.io_type != CTL_IO_SCSI)
10579                 return (1);
10580
10581         switch (io->scsiio.cdb[0]) {
10582         case COMPARE_AND_WRITE: {
10583                 struct scsi_compare_and_write *cdb;
10584
10585                 cdb = (struct scsi_compare_and_write *)io->scsiio.cdb;
10586
10587                 *lba = scsi_8btou64(cdb->addr);
10588                 *len = cdb->length;
10589                 break;
10590         }
10591         case READ_6:
10592         case WRITE_6: {
10593                 struct scsi_rw_6 *cdb;
10594
10595                 cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
10596
10597                 *lba = scsi_3btoul(cdb->addr);
10598                 /* only 5 bits are valid in the most significant address byte */
10599                 *lba &= 0x1fffff;
10600                 *len = cdb->length;
10601                 break;
10602         }
10603         case READ_10:
10604         case WRITE_10: {
10605                 struct scsi_rw_10 *cdb;
10606
10607                 cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
10608
10609                 *lba = scsi_4btoul(cdb->addr);
10610                 *len = scsi_2btoul(cdb->length);
10611                 break;
10612         }
10613         case WRITE_VERIFY_10: {
10614                 struct scsi_write_verify_10 *cdb;
10615
10616                 cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
10617
10618                 *lba = scsi_4btoul(cdb->addr);
10619                 *len = scsi_2btoul(cdb->length);
10620                 break;
10621         }
10622         case READ_12:
10623         case WRITE_12: {
10624                 struct scsi_rw_12 *cdb;
10625
10626                 cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
10627
10628                 *lba = scsi_4btoul(cdb->addr);
10629                 *len = scsi_4btoul(cdb->length);
10630                 break;
10631         }
10632         case WRITE_VERIFY_12: {
10633                 struct scsi_write_verify_12 *cdb;
10634
10635                 cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
10636
10637                 *lba = scsi_4btoul(cdb->addr);
10638                 *len = scsi_4btoul(cdb->length);
10639                 break;
10640         }
10641         case READ_16:
10642         case WRITE_16: {
10643                 struct scsi_rw_16 *cdb;
10644
10645                 cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
10646
10647                 *lba = scsi_8btou64(cdb->addr);
10648                 *len = scsi_4btoul(cdb->length);
10649                 break;
10650         }
10651         case WRITE_ATOMIC_16: {
10652                 struct scsi_write_atomic_16 *cdb;
10653
10654                 cdb = (struct scsi_write_atomic_16 *)io->scsiio.cdb;
10655
10656                 *lba = scsi_8btou64(cdb->addr);
10657                 *len = scsi_2btoul(cdb->length);
10658                 break;
10659         }
10660         case WRITE_VERIFY_16: {
10661                 struct scsi_write_verify_16 *cdb;
10662
10663                 cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
10664
10665                 *lba = scsi_8btou64(cdb->addr);
10666                 *len = scsi_4btoul(cdb->length);
10667                 break;
10668         }
10669         case WRITE_SAME_10: {
10670                 struct scsi_write_same_10 *cdb;
10671
10672                 cdb = (struct scsi_write_same_10 *)io->scsiio.cdb;
10673
10674                 *lba = scsi_4btoul(cdb->addr);
10675                 *len = scsi_2btoul(cdb->length);
10676                 break;
10677         }
10678         case WRITE_SAME_16: {
10679                 struct scsi_write_same_16 *cdb;
10680
10681                 cdb = (struct scsi_write_same_16 *)io->scsiio.cdb;
10682
10683                 *lba = scsi_8btou64(cdb->addr);
10684                 *len = scsi_4btoul(cdb->length);
10685                 break;
10686         }
10687         case VERIFY_10: {
10688                 struct scsi_verify_10 *cdb;
10689
10690                 cdb = (struct scsi_verify_10 *)io->scsiio.cdb;
10691
10692                 *lba = scsi_4btoul(cdb->addr);
10693                 *len = scsi_2btoul(cdb->length);
10694                 break;
10695         }
10696         case VERIFY_12: {
10697                 struct scsi_verify_12 *cdb;
10698
10699                 cdb = (struct scsi_verify_12 *)io->scsiio.cdb;
10700
10701                 *lba = scsi_4btoul(cdb->addr);
10702                 *len = scsi_4btoul(cdb->length);
10703                 break;
10704         }
10705         case VERIFY_16: {
10706                 struct scsi_verify_16 *cdb;
10707
10708                 cdb = (struct scsi_verify_16 *)io->scsiio.cdb;
10709
10710                 *lba = scsi_8btou64(cdb->addr);
10711                 *len = scsi_4btoul(cdb->length);
10712                 break;
10713         }
10714         case UNMAP: {
10715                 *lba = 0;
10716                 *len = UINT64_MAX;
10717                 break;
10718         }
10719         case SERVICE_ACTION_IN: {       /* GET LBA STATUS */
10720                 struct scsi_get_lba_status *cdb;
10721
10722                 cdb = (struct scsi_get_lba_status *)io->scsiio.cdb;
10723                 *lba = scsi_8btou64(cdb->addr);
10724                 *len = UINT32_MAX;
10725                 break;
10726         }
10727         default:
10728                 return (1);
10729                 break; /* NOTREACHED */
10730         }
10731
10732         return (0);
10733 }
10734
10735 static ctl_action
10736 ctl_extent_check_lba(uint64_t lba1, uint64_t len1, uint64_t lba2, uint64_t len2,
10737     bool seq)
10738 {
10739         uint64_t endlba1, endlba2;
10740
10741         endlba1 = lba1 + len1 - (seq ? 0 : 1);
10742         endlba2 = lba2 + len2 - 1;
10743
10744         if ((endlba1 < lba2) || (endlba2 < lba1))
10745                 return (CTL_ACTION_PASS);
10746         else
10747                 return (CTL_ACTION_BLOCK);
10748 }
10749
10750 static int
10751 ctl_extent_check_unmap(union ctl_io *io, uint64_t lba2, uint64_t len2)
10752 {
10753         struct ctl_ptr_len_flags *ptrlen;
10754         struct scsi_unmap_desc *buf, *end, *range;
10755         uint64_t lba;
10756         uint32_t len;
10757
10758         /* If not UNMAP -- go other way. */
10759         if (io->io_hdr.io_type != CTL_IO_SCSI ||
10760             io->scsiio.cdb[0] != UNMAP)
10761                 return (CTL_ACTION_ERROR);
10762
10763         /* If UNMAP without data -- block and wait for data. */
10764         ptrlen = (struct ctl_ptr_len_flags *)
10765             &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
10766         if ((io->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0 ||
10767             ptrlen->ptr == NULL)
10768                 return (CTL_ACTION_BLOCK);
10769
10770         /* UNMAP with data -- check for collision. */
10771         buf = (struct scsi_unmap_desc *)ptrlen->ptr;
10772         end = buf + ptrlen->len / sizeof(*buf);
10773         for (range = buf; range < end; range++) {
10774                 lba = scsi_8btou64(range->lba);
10775                 len = scsi_4btoul(range->length);
10776                 if ((lba < lba2 + len2) && (lba + len > lba2))
10777                         return (CTL_ACTION_BLOCK);
10778         }
10779         return (CTL_ACTION_PASS);
10780 }
10781
10782 static ctl_action
10783 ctl_extent_check(union ctl_io *io1, union ctl_io *io2, bool seq)
10784 {
10785         uint64_t lba1, lba2;
10786         uint64_t len1, len2;
10787         int retval;
10788
10789         if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
10790                 return (CTL_ACTION_ERROR);
10791
10792         retval = ctl_extent_check_unmap(io1, lba2, len2);
10793         if (retval != CTL_ACTION_ERROR)
10794                 return (retval);
10795
10796         if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
10797                 return (CTL_ACTION_ERROR);
10798
10799         if (io1->io_hdr.flags & CTL_FLAG_SERSEQ_DONE)
10800                 seq = FALSE;
10801         return (ctl_extent_check_lba(lba1, len1, lba2, len2, seq));
10802 }
10803
10804 static ctl_action
10805 ctl_extent_check_seq(union ctl_io *io1, union ctl_io *io2)
10806 {
10807         uint64_t lba1, lba2;
10808         uint64_t len1, len2;
10809
10810         if (io1->io_hdr.flags & CTL_FLAG_SERSEQ_DONE)
10811                 return (CTL_ACTION_PASS);
10812         if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
10813                 return (CTL_ACTION_ERROR);
10814         if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
10815                 return (CTL_ACTION_ERROR);
10816
10817         if (lba1 + len1 == lba2)
10818                 return (CTL_ACTION_BLOCK);
10819         return (CTL_ACTION_PASS);
10820 }
10821
10822 static ctl_action
10823 ctl_check_for_blockage(struct ctl_lun *lun, union ctl_io *pending_io,
10824     union ctl_io *ooa_io)
10825 {
10826         const struct ctl_cmd_entry *pending_entry, *ooa_entry;
10827         const ctl_serialize_action *serialize_row;
10828
10829         /*
10830          * The initiator attempted multiple untagged commands at the same
10831          * time.  Can't do that.
10832          */
10833         if ((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10834          && (ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10835          && ((pending_io->io_hdr.nexus.targ_port ==
10836               ooa_io->io_hdr.nexus.targ_port)
10837           && (pending_io->io_hdr.nexus.initid ==
10838               ooa_io->io_hdr.nexus.initid))
10839          && ((ooa_io->io_hdr.flags & (CTL_FLAG_ABORT |
10840               CTL_FLAG_STATUS_SENT)) == 0))
10841                 return (CTL_ACTION_OVERLAP);
10842
10843         /*
10844          * The initiator attempted to send multiple tagged commands with
10845          * the same ID.  (It's fine if different initiators have the same
10846          * tag ID.)
10847          *
10848          * Even if all of those conditions are true, we don't kill the I/O
10849          * if the command ahead of us has been aborted.  We won't end up
10850          * sending it to the FETD, and it's perfectly legal to resend a
10851          * command with the same tag number as long as the previous
10852          * instance of this tag number has been aborted somehow.
10853          */
10854         if ((pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10855          && (ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10856          && (pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
10857          && ((pending_io->io_hdr.nexus.targ_port ==
10858               ooa_io->io_hdr.nexus.targ_port)
10859           && (pending_io->io_hdr.nexus.initid ==
10860               ooa_io->io_hdr.nexus.initid))
10861          && ((ooa_io->io_hdr.flags & (CTL_FLAG_ABORT |
10862               CTL_FLAG_STATUS_SENT)) == 0))
10863                 return (CTL_ACTION_OVERLAP_TAG);
10864
10865         /*
10866          * If we get a head of queue tag, SAM-3 says that we should
10867          * immediately execute it.
10868          *
10869          * What happens if this command would normally block for some other
10870          * reason?  e.g. a request sense with a head of queue tag
10871          * immediately after a write.  Normally that would block, but this
10872          * will result in its getting executed immediately...
10873          *
10874          * We currently return "pass" instead of "skip", so we'll end up
10875          * going through the rest of the queue to check for overlapped tags.
10876          *
10877          * XXX KDM check for other types of blockage first??
10878          */
10879         if (pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10880                 return (CTL_ACTION_PASS);
10881
10882         /*
10883          * Ordered tags have to block until all items ahead of them
10884          * have completed.  If we get called with an ordered tag, we always
10885          * block, if something else is ahead of us in the queue.
10886          */
10887         if (pending_io->scsiio.tag_type == CTL_TAG_ORDERED)
10888                 return (CTL_ACTION_BLOCK);
10889
10890         /*
10891          * Simple tags get blocked until all head of queue and ordered tags
10892          * ahead of them have completed.  I'm lumping untagged commands in
10893          * with simple tags here.  XXX KDM is that the right thing to do?
10894          */
10895         if (((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10896           || (pending_io->scsiio.tag_type == CTL_TAG_SIMPLE))
10897          && ((ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10898           || (ooa_io->scsiio.tag_type == CTL_TAG_ORDERED)))
10899                 return (CTL_ACTION_BLOCK);
10900
10901         pending_entry = ctl_get_cmd_entry(&pending_io->scsiio, NULL);
10902         KASSERT(pending_entry->seridx < CTL_SERIDX_COUNT,
10903             ("%s: Invalid seridx %d for pending CDB %02x %02x @ %p",
10904              __func__, pending_entry->seridx, pending_io->scsiio.cdb[0],
10905              pending_io->scsiio.cdb[1], pending_io));
10906         ooa_entry = ctl_get_cmd_entry(&ooa_io->scsiio, NULL);
10907         if (ooa_entry->seridx == CTL_SERIDX_INVLD)
10908                 return (CTL_ACTION_PASS); /* Unsupported command in OOA queue */
10909         KASSERT(ooa_entry->seridx < CTL_SERIDX_COUNT,
10910             ("%s: Invalid seridx %d for ooa CDB %02x %02x @ %p",
10911              __func__, ooa_entry->seridx, ooa_io->scsiio.cdb[0],
10912              ooa_io->scsiio.cdb[1], ooa_io));
10913
10914         serialize_row = ctl_serialize_table[ooa_entry->seridx];
10915
10916         switch (serialize_row[pending_entry->seridx]) {
10917         case CTL_SER_BLOCK:
10918                 return (CTL_ACTION_BLOCK);
10919         case CTL_SER_EXTENT:
10920                 return (ctl_extent_check(ooa_io, pending_io,
10921                     (lun->be_lun && lun->be_lun->serseq == CTL_LUN_SERSEQ_ON)));
10922         case CTL_SER_EXTENTOPT:
10923                 if ((lun->MODE_CTRL.queue_flags & SCP_QUEUE_ALG_MASK) !=
10924                     SCP_QUEUE_ALG_UNRESTRICTED)
10925                         return (ctl_extent_check(ooa_io, pending_io,
10926                             (lun->be_lun &&
10927                              lun->be_lun->serseq == CTL_LUN_SERSEQ_ON)));
10928                 return (CTL_ACTION_PASS);
10929         case CTL_SER_EXTENTSEQ:
10930                 if (lun->be_lun && lun->be_lun->serseq != CTL_LUN_SERSEQ_OFF)
10931                         return (ctl_extent_check_seq(ooa_io, pending_io));
10932                 return (CTL_ACTION_PASS);
10933         case CTL_SER_PASS:
10934                 return (CTL_ACTION_PASS);
10935         case CTL_SER_BLOCKOPT:
10936                 if ((lun->MODE_CTRL.queue_flags & SCP_QUEUE_ALG_MASK) !=
10937                     SCP_QUEUE_ALG_UNRESTRICTED)
10938                         return (CTL_ACTION_BLOCK);
10939                 return (CTL_ACTION_PASS);
10940         case CTL_SER_SKIP:
10941                 return (CTL_ACTION_SKIP);
10942         default:
10943                 panic("%s: Invalid serialization value %d for %d => %d",
10944                     __func__, serialize_row[pending_entry->seridx],
10945                     pending_entry->seridx, ooa_entry->seridx);
10946         }
10947
10948         return (CTL_ACTION_ERROR);
10949 }
10950
10951 /*
10952  * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
10953  * Assumptions:
10954  * - pending_io is generally either incoming, or on the blocked queue
10955  * - starting I/O is the I/O we want to start the check with.
10956  */
10957 static ctl_action
10958 ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
10959               union ctl_io *starting_io)
10960 {
10961         union ctl_io *ooa_io;
10962         ctl_action action;
10963
10964         mtx_assert(&lun->lun_lock, MA_OWNED);
10965
10966         /*
10967          * Run back along the OOA queue, starting with the current
10968          * blocked I/O and going through every I/O before it on the
10969          * queue.  If starting_io is NULL, we'll just end up returning
10970          * CTL_ACTION_PASS.
10971          */
10972         for (ooa_io = starting_io; ooa_io != NULL;
10973              ooa_io = (union ctl_io *)TAILQ_PREV(&ooa_io->io_hdr, ctl_ooaq,
10974              ooa_links)){
10975
10976                 /*
10977                  * This routine just checks to see whether
10978                  * cur_blocked is blocked by ooa_io, which is ahead
10979                  * of it in the queue.  It doesn't queue/dequeue
10980                  * cur_blocked.
10981                  */
10982                 action = ctl_check_for_blockage(lun, pending_io, ooa_io);
10983                 switch (action) {
10984                 case CTL_ACTION_BLOCK:
10985                 case CTL_ACTION_OVERLAP:
10986                 case CTL_ACTION_OVERLAP_TAG:
10987                 case CTL_ACTION_SKIP:
10988                 case CTL_ACTION_ERROR:
10989                         return (action);
10990                         break; /* NOTREACHED */
10991                 case CTL_ACTION_PASS:
10992                         break;
10993                 default:
10994                         panic("%s: Invalid action %d\n", __func__, action);
10995                 }
10996         }
10997
10998         return (CTL_ACTION_PASS);
10999 }
11000
11001 /*
11002  * Assumptions:
11003  * - An I/O has just completed, and has been removed from the per-LUN OOA
11004  *   queue, so some items on the blocked queue may now be unblocked.
11005  */
11006 static int
11007 ctl_check_blocked(struct ctl_lun *lun)
11008 {
11009         struct ctl_softc *softc = lun->ctl_softc;
11010         union ctl_io *cur_blocked, *next_blocked;
11011
11012         mtx_assert(&lun->lun_lock, MA_OWNED);
11013
11014         /*
11015          * Run forward from the head of the blocked queue, checking each
11016          * entry against the I/Os prior to it on the OOA queue to see if
11017          * there is still any blockage.
11018          *
11019          * We cannot use the TAILQ_FOREACH() macro, because it can't deal
11020          * with our removing a variable on it while it is traversing the
11021          * list.
11022          */
11023         for (cur_blocked = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue);
11024              cur_blocked != NULL; cur_blocked = next_blocked) {
11025                 union ctl_io *prev_ooa;
11026                 ctl_action action;
11027
11028                 next_blocked = (union ctl_io *)TAILQ_NEXT(&cur_blocked->io_hdr,
11029                                                           blocked_links);
11030
11031                 prev_ooa = (union ctl_io *)TAILQ_PREV(&cur_blocked->io_hdr,
11032                                                       ctl_ooaq, ooa_links);
11033
11034                 /*
11035                  * If cur_blocked happens to be the first item in the OOA
11036                  * queue now, prev_ooa will be NULL, and the action
11037                  * returned will just be CTL_ACTION_PASS.
11038                  */
11039                 action = ctl_check_ooa(lun, cur_blocked, prev_ooa);
11040
11041                 switch (action) {
11042                 case CTL_ACTION_BLOCK:
11043                         /* Nothing to do here, still blocked */
11044                         break;
11045                 case CTL_ACTION_OVERLAP:
11046                 case CTL_ACTION_OVERLAP_TAG:
11047                         /*
11048                          * This shouldn't happen!  In theory we've already
11049                          * checked this command for overlap...
11050                          */
11051                         break;
11052                 case CTL_ACTION_PASS:
11053                 case CTL_ACTION_SKIP: {
11054                         const struct ctl_cmd_entry *entry;
11055
11056                         /*
11057                          * The skip case shouldn't happen, this transaction
11058                          * should have never made it onto the blocked queue.
11059                          */
11060                         /*
11061                          * This I/O is no longer blocked, we can remove it
11062                          * from the blocked queue.  Since this is a TAILQ
11063                          * (doubly linked list), we can do O(1) removals
11064                          * from any place on the list.
11065                          */
11066                         TAILQ_REMOVE(&lun->blocked_queue, &cur_blocked->io_hdr,
11067                                      blocked_links);
11068                         cur_blocked->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
11069
11070                         if ((softc->ha_mode != CTL_HA_MODE_XFER) &&
11071                             (cur_blocked->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)){
11072                                 /*
11073                                  * Need to send IO back to original side to
11074                                  * run
11075                                  */
11076                                 union ctl_ha_msg msg_info;
11077
11078                                 cur_blocked->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11079                                 msg_info.hdr.original_sc =
11080                                         cur_blocked->io_hdr.original_sc;
11081                                 msg_info.hdr.serializing_sc = cur_blocked;
11082                                 msg_info.hdr.msg_type = CTL_MSG_R2R;
11083                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11084                                     sizeof(msg_info.hdr), M_NOWAIT);
11085                                 break;
11086                         }
11087                         entry = ctl_get_cmd_entry(&cur_blocked->scsiio, NULL);
11088
11089                         /*
11090                          * Check this I/O for LUN state changes that may
11091                          * have happened while this command was blocked.
11092                          * The LUN state may have been changed by a command
11093                          * ahead of us in the queue, so we need to re-check
11094                          * for any states that can be caused by SCSI
11095                          * commands.
11096                          */
11097                         if (ctl_scsiio_lun_check(lun, entry,
11098                                                  &cur_blocked->scsiio) == 0) {
11099                                 cur_blocked->io_hdr.flags |=
11100                                                       CTL_FLAG_IS_WAS_ON_RTR;
11101                                 ctl_enqueue_rtr(cur_blocked);
11102                         } else
11103                                 ctl_done(cur_blocked);
11104                         break;
11105                 }
11106                 default:
11107                         /*
11108                          * This probably shouldn't happen -- we shouldn't
11109                          * get CTL_ACTION_ERROR, or anything else.
11110                          */
11111                         break;
11112                 }
11113         }
11114
11115         return (CTL_RETVAL_COMPLETE);
11116 }
11117
11118 /*
11119  * This routine (with one exception) checks LUN flags that can be set by
11120  * commands ahead of us in the OOA queue.  These flags have to be checked
11121  * when a command initially comes in, and when we pull a command off the
11122  * blocked queue and are preparing to execute it.  The reason we have to
11123  * check these flags for commands on the blocked queue is that the LUN
11124  * state may have been changed by a command ahead of us while we're on the
11125  * blocked queue.
11126  *
11127  * Ordering is somewhat important with these checks, so please pay
11128  * careful attention to the placement of any new checks.
11129  */
11130 static int
11131 ctl_scsiio_lun_check(struct ctl_lun *lun,
11132     const struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
11133 {
11134         struct ctl_softc *softc = lun->ctl_softc;
11135         int retval;
11136         uint32_t residx;
11137
11138         retval = 0;
11139
11140         mtx_assert(&lun->lun_lock, MA_OWNED);
11141
11142         /*
11143          * If this shelf is a secondary shelf controller, we may have to
11144          * reject some commands disallowed by HA mode and link state.
11145          */
11146         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
11147                 if (softc->ha_link == CTL_HA_LINK_OFFLINE &&
11148                     (entry->flags & CTL_CMD_FLAG_OK_ON_UNAVAIL) == 0) {
11149                         ctl_set_lun_unavail(ctsio);
11150                         retval = 1;
11151                         goto bailout;
11152                 }
11153                 if ((lun->flags & CTL_LUN_PEER_SC_PRIMARY) == 0 &&
11154                     (entry->flags & CTL_CMD_FLAG_OK_ON_UNAVAIL) == 0) {
11155                         ctl_set_lun_transit(ctsio);
11156                         retval = 1;
11157                         goto bailout;
11158                 }
11159                 if (softc->ha_mode == CTL_HA_MODE_ACT_STBY &&
11160                     (entry->flags & CTL_CMD_FLAG_OK_ON_STANDBY) == 0) {
11161                         ctl_set_lun_standby(ctsio);
11162                         retval = 1;
11163                         goto bailout;
11164                 }
11165
11166                 /* The rest of checks are only done on executing side */
11167                 if (softc->ha_mode == CTL_HA_MODE_XFER)
11168                         goto bailout;
11169         }
11170
11171         if (entry->pattern & CTL_LUN_PAT_WRITE) {
11172                 if (lun->be_lun &&
11173                     lun->be_lun->flags & CTL_LUN_FLAG_READONLY) {
11174                         ctl_set_hw_write_protected(ctsio);
11175                         retval = 1;
11176                         goto bailout;
11177                 }
11178                 if ((lun->MODE_CTRL.eca_and_aen & SCP_SWP) != 0) {
11179                         ctl_set_sense(ctsio, /*current_error*/ 1,
11180                             /*sense_key*/ SSD_KEY_DATA_PROTECT,
11181                             /*asc*/ 0x27, /*ascq*/ 0x02, SSD_ELEM_NONE);
11182                         retval = 1;
11183                         goto bailout;
11184                 }
11185         }
11186
11187         /*
11188          * Check for a reservation conflict.  If this command isn't allowed
11189          * even on reserved LUNs, and if this initiator isn't the one who
11190          * reserved us, reject the command with a reservation conflict.
11191          */
11192         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11193         if ((lun->flags & CTL_LUN_RESERVED)
11194          && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
11195                 if (lun->res_idx != residx) {
11196                         ctl_set_reservation_conflict(ctsio);
11197                         retval = 1;
11198                         goto bailout;
11199                 }
11200         }
11201
11202         if ((lun->flags & CTL_LUN_PR_RESERVED) == 0 ||
11203             (entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV)) {
11204                 /* No reservation or command is allowed. */;
11205         } else if ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_WRESV) &&
11206             (lun->pr_res_type == SPR_TYPE_WR_EX ||
11207              lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
11208              lun->pr_res_type == SPR_TYPE_WR_EX_AR)) {
11209                 /* The command is allowed for Write Exclusive resv. */;
11210         } else {
11211                 /*
11212                  * if we aren't registered or it's a res holder type
11213                  * reservation and this isn't the res holder then set a
11214                  * conflict.
11215                  */
11216                 if (ctl_get_prkey(lun, residx) == 0 ||
11217                     (residx != lun->pr_res_idx && lun->pr_res_type < 4)) {
11218                         ctl_set_reservation_conflict(ctsio);
11219                         retval = 1;
11220                         goto bailout;
11221                 }
11222         }
11223
11224         if ((entry->flags & CTL_CMD_FLAG_OK_ON_NO_MEDIA) == 0) {
11225                 if (lun->flags & CTL_LUN_EJECTED)
11226                         ctl_set_lun_ejected(ctsio);
11227                 else if (lun->flags & CTL_LUN_NO_MEDIA) {
11228                         if (lun->flags & CTL_LUN_REMOVABLE)
11229                                 ctl_set_lun_no_media(ctsio);
11230                         else
11231                                 ctl_set_lun_int_reqd(ctsio);
11232                 } else if (lun->flags & CTL_LUN_STOPPED)
11233                         ctl_set_lun_stopped(ctsio);
11234                 else
11235                         goto bailout;
11236                 retval = 1;
11237                 goto bailout;
11238         }
11239
11240 bailout:
11241         return (retval);
11242 }
11243
11244 static void
11245 ctl_failover_io(union ctl_io *io, int have_lock)
11246 {
11247         ctl_set_busy(&io->scsiio);
11248         ctl_done(io);
11249 }
11250
11251 static void
11252 ctl_failover_lun(union ctl_io *rio)
11253 {
11254         struct ctl_softc *softc = CTL_SOFTC(rio);
11255         struct ctl_lun *lun;
11256         struct ctl_io_hdr *io, *next_io;
11257         uint32_t targ_lun;
11258
11259         targ_lun = rio->io_hdr.nexus.targ_mapped_lun;
11260         CTL_DEBUG_PRINT(("FAILOVER for lun %ju\n", targ_lun));
11261
11262         /* Find and lock the LUN. */
11263         mtx_lock(&softc->ctl_lock);
11264         if (targ_lun > ctl_max_luns ||
11265             (lun = softc->ctl_luns[targ_lun]) == NULL) {
11266                 mtx_unlock(&softc->ctl_lock);
11267                 return;
11268         }
11269         mtx_lock(&lun->lun_lock);
11270         mtx_unlock(&softc->ctl_lock);
11271         if (lun->flags & CTL_LUN_DISABLED) {
11272                 mtx_unlock(&lun->lun_lock);
11273                 return;
11274         }
11275
11276         if (softc->ha_mode == CTL_HA_MODE_XFER) {
11277                 TAILQ_FOREACH_SAFE(io, &lun->ooa_queue, ooa_links, next_io) {
11278                         /* We are master */
11279                         if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
11280                                 if (io->flags & CTL_FLAG_IO_ACTIVE) {
11281                                         io->flags |= CTL_FLAG_ABORT;
11282                                         io->flags |= CTL_FLAG_FAILOVER;
11283                                 } else { /* This can be only due to DATAMOVE */
11284                                         io->msg_type = CTL_MSG_DATAMOVE_DONE;
11285                                         io->flags &= ~CTL_FLAG_DMA_INPROG;
11286                                         io->flags |= CTL_FLAG_IO_ACTIVE;
11287                                         io->port_status = 31340;
11288                                         ctl_enqueue_isc((union ctl_io *)io);
11289                                 }
11290                         }
11291                         /* We are slave */
11292                         if (io->flags & CTL_FLAG_SENT_2OTHER_SC) {
11293                                 io->flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11294                                 if (io->flags & CTL_FLAG_IO_ACTIVE) {
11295                                         io->flags |= CTL_FLAG_FAILOVER;
11296                                 } else {
11297                                         ctl_set_busy(&((union ctl_io *)io)->
11298                                             scsiio);
11299                                         ctl_done((union ctl_io *)io);
11300                                 }
11301                         }
11302                 }
11303         } else { /* SERIALIZE modes */
11304                 TAILQ_FOREACH_SAFE(io, &lun->blocked_queue, blocked_links,
11305                     next_io) {
11306                         /* We are master */
11307                         if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
11308                                 TAILQ_REMOVE(&lun->blocked_queue, io,
11309                                     blocked_links);
11310                                 io->flags &= ~CTL_FLAG_BLOCKED;
11311                                 TAILQ_REMOVE(&lun->ooa_queue, io, ooa_links);
11312                                 ctl_free_io((union ctl_io *)io);
11313                         }
11314                 }
11315                 TAILQ_FOREACH_SAFE(io, &lun->ooa_queue, ooa_links, next_io) {
11316                         /* We are master */
11317                         if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
11318                                 TAILQ_REMOVE(&lun->ooa_queue, io, ooa_links);
11319                                 ctl_free_io((union ctl_io *)io);
11320                         }
11321                         /* We are slave */
11322                         if (io->flags & CTL_FLAG_SENT_2OTHER_SC) {
11323                                 io->flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11324                                 if (!(io->flags & CTL_FLAG_IO_ACTIVE)) {
11325                                         ctl_set_busy(&((union ctl_io *)io)->
11326                                             scsiio);
11327                                         ctl_done((union ctl_io *)io);
11328                                 }
11329                         }
11330                 }
11331                 ctl_check_blocked(lun);
11332         }
11333         mtx_unlock(&lun->lun_lock);
11334 }
11335
11336 static int
11337 ctl_scsiio_precheck(struct ctl_softc *softc, struct ctl_scsiio *ctsio)
11338 {
11339         struct ctl_lun *lun;
11340         const struct ctl_cmd_entry *entry;
11341         uint32_t initidx, targ_lun;
11342         int retval = 0;
11343
11344         lun = NULL;
11345         targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
11346         if (targ_lun < ctl_max_luns)
11347                 lun = softc->ctl_luns[targ_lun];
11348         if (lun) {
11349                 /*
11350                  * If the LUN is invalid, pretend that it doesn't exist.
11351                  * It will go away as soon as all pending I/O has been
11352                  * completed.
11353                  */
11354                 mtx_lock(&lun->lun_lock);
11355                 if (lun->flags & CTL_LUN_DISABLED) {
11356                         mtx_unlock(&lun->lun_lock);
11357                         lun = NULL;
11358                 }
11359         }
11360         CTL_LUN(ctsio) = lun;
11361         if (lun) {
11362                 CTL_BACKEND_LUN(ctsio) = lun->be_lun;
11363
11364                 /*
11365                  * Every I/O goes into the OOA queue for a particular LUN,
11366                  * and stays there until completion.
11367                  */
11368 #ifdef CTL_TIME_IO
11369                 if (TAILQ_EMPTY(&lun->ooa_queue))
11370                         lun->idle_time += getsbinuptime() - lun->last_busy;
11371 #endif
11372                 TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
11373         }
11374
11375         /* Get command entry and return error if it is unsuppotyed. */
11376         entry = ctl_validate_command(ctsio);
11377         if (entry == NULL) {
11378                 if (lun)
11379                         mtx_unlock(&lun->lun_lock);
11380                 return (retval);
11381         }
11382
11383         ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
11384         ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
11385
11386         /*
11387          * Check to see whether we can send this command to LUNs that don't
11388          * exist.  This should pretty much only be the case for inquiry
11389          * and request sense.  Further checks, below, really require having
11390          * a LUN, so we can't really check the command anymore.  Just put
11391          * it on the rtr queue.
11392          */
11393         if (lun == NULL) {
11394                 if (entry->flags & CTL_CMD_FLAG_OK_ON_NO_LUN) {
11395                         ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11396                         ctl_enqueue_rtr((union ctl_io *)ctsio);
11397                         return (retval);
11398                 }
11399
11400                 ctl_set_unsupported_lun(ctsio);
11401                 ctl_done((union ctl_io *)ctsio);
11402                 CTL_DEBUG_PRINT(("ctl_scsiio_precheck: bailing out due to invalid LUN\n"));
11403                 return (retval);
11404         } else {
11405                 /*
11406                  * Make sure we support this particular command on this LUN.
11407                  * e.g., we don't support writes to the control LUN.
11408                  */
11409                 if (!ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
11410                         mtx_unlock(&lun->lun_lock);
11411                         ctl_set_invalid_opcode(ctsio);
11412                         ctl_done((union ctl_io *)ctsio);
11413                         return (retval);
11414                 }
11415         }
11416
11417         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11418
11419         /*
11420          * If we've got a request sense, it'll clear the contingent
11421          * allegiance condition.  Otherwise, if we have a CA condition for
11422          * this initiator, clear it, because it sent down a command other
11423          * than request sense.
11424          */
11425         if (ctsio->cdb[0] != REQUEST_SENSE) {
11426                 struct scsi_sense_data *ps;
11427
11428                 ps = lun->pending_sense[initidx / CTL_MAX_INIT_PER_PORT];
11429                 if (ps != NULL)
11430                         ps[initidx % CTL_MAX_INIT_PER_PORT].error_code = 0;
11431         }
11432
11433         /*
11434          * If the command has this flag set, it handles its own unit
11435          * attention reporting, we shouldn't do anything.  Otherwise we
11436          * check for any pending unit attentions, and send them back to the
11437          * initiator.  We only do this when a command initially comes in,
11438          * not when we pull it off the blocked queue.
11439          *
11440          * According to SAM-3, section 5.3.2, the order that things get
11441          * presented back to the host is basically unit attentions caused
11442          * by some sort of reset event, busy status, reservation conflicts
11443          * or task set full, and finally any other status.
11444          *
11445          * One issue here is that some of the unit attentions we report
11446          * don't fall into the "reset" category (e.g. "reported luns data
11447          * has changed").  So reporting it here, before the reservation
11448          * check, may be technically wrong.  I guess the only thing to do
11449          * would be to check for and report the reset events here, and then
11450          * check for the other unit attention types after we check for a
11451          * reservation conflict.
11452          *
11453          * XXX KDM need to fix this
11454          */
11455         if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
11456                 ctl_ua_type ua_type;
11457                 u_int sense_len = 0;
11458
11459                 ua_type = ctl_build_ua(lun, initidx, &ctsio->sense_data,
11460                     &sense_len, SSD_TYPE_NONE);
11461                 if (ua_type != CTL_UA_NONE) {
11462                         mtx_unlock(&lun->lun_lock);
11463                         ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
11464                         ctsio->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
11465                         ctsio->sense_len = sense_len;
11466                         ctl_done((union ctl_io *)ctsio);
11467                         return (retval);
11468                 }
11469         }
11470
11471
11472         if (ctl_scsiio_lun_check(lun, entry, ctsio) != 0) {
11473                 mtx_unlock(&lun->lun_lock);
11474                 ctl_done((union ctl_io *)ctsio);
11475                 return (retval);
11476         }
11477
11478         /*
11479          * XXX CHD this is where we want to send IO to other side if
11480          * this LUN is secondary on this SC. We will need to make a copy
11481          * of the IO and flag the IO on this side as SENT_2OTHER and the flag
11482          * the copy we send as FROM_OTHER.
11483          * We also need to stuff the address of the original IO so we can
11484          * find it easily. Something similar will need be done on the other
11485          * side so when we are done we can find the copy.
11486          */
11487         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
11488             (lun->flags & CTL_LUN_PEER_SC_PRIMARY) != 0 &&
11489             (entry->flags & CTL_CMD_FLAG_RUN_HERE) == 0) {
11490                 union ctl_ha_msg msg_info;
11491                 int isc_retval;
11492
11493                 ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11494                 ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11495                 mtx_unlock(&lun->lun_lock);
11496
11497                 msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
11498                 msg_info.hdr.original_sc = (union ctl_io *)ctsio;
11499                 msg_info.hdr.serializing_sc = NULL;
11500                 msg_info.hdr.nexus = ctsio->io_hdr.nexus;
11501                 msg_info.scsi.tag_num = ctsio->tag_num;
11502                 msg_info.scsi.tag_type = ctsio->tag_type;
11503                 msg_info.scsi.cdb_len = ctsio->cdb_len;
11504                 memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
11505
11506                 if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11507                     sizeof(msg_info.scsi) - sizeof(msg_info.scsi.sense_data),
11508                     M_WAITOK)) > CTL_HA_STATUS_SUCCESS) {
11509                         ctl_set_busy(ctsio);
11510                         ctl_done((union ctl_io *)ctsio);
11511                         return (retval);
11512                 }
11513                 return (retval);
11514         }
11515
11516         switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
11517                               (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr,
11518                               ctl_ooaq, ooa_links))) {
11519         case CTL_ACTION_BLOCK:
11520                 ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
11521                 TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
11522                                   blocked_links);
11523                 mtx_unlock(&lun->lun_lock);
11524                 return (retval);
11525         case CTL_ACTION_PASS:
11526         case CTL_ACTION_SKIP:
11527                 ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11528                 mtx_unlock(&lun->lun_lock);
11529                 ctl_enqueue_rtr((union ctl_io *)ctsio);
11530                 break;
11531         case CTL_ACTION_OVERLAP:
11532                 mtx_unlock(&lun->lun_lock);
11533                 ctl_set_overlapped_cmd(ctsio);
11534                 ctl_done((union ctl_io *)ctsio);
11535                 break;
11536         case CTL_ACTION_OVERLAP_TAG:
11537                 mtx_unlock(&lun->lun_lock);
11538                 ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
11539                 ctl_done((union ctl_io *)ctsio);
11540                 break;
11541         case CTL_ACTION_ERROR:
11542         default:
11543                 mtx_unlock(&lun->lun_lock);
11544                 ctl_set_internal_failure(ctsio,
11545                                          /*sks_valid*/ 0,
11546                                          /*retry_count*/ 0);
11547                 ctl_done((union ctl_io *)ctsio);
11548                 break;
11549         }
11550         return (retval);
11551 }
11552
11553 const struct ctl_cmd_entry *
11554 ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa)
11555 {
11556         const struct ctl_cmd_entry *entry;
11557         int service_action;
11558
11559         entry = &ctl_cmd_table[ctsio->cdb[0]];
11560         if (sa)
11561                 *sa = ((entry->flags & CTL_CMD_FLAG_SA5) != 0);
11562         if (entry->flags & CTL_CMD_FLAG_SA5) {
11563                 service_action = ctsio->cdb[1] & SERVICE_ACTION_MASK;
11564                 entry = &((const struct ctl_cmd_entry *)
11565                     entry->execute)[service_action];
11566         }
11567         return (entry);
11568 }
11569
11570 const struct ctl_cmd_entry *
11571 ctl_validate_command(struct ctl_scsiio *ctsio)
11572 {
11573         const struct ctl_cmd_entry *entry;
11574         int i, sa;
11575         uint8_t diff;
11576
11577         entry = ctl_get_cmd_entry(ctsio, &sa);
11578         if (entry->execute == NULL) {
11579                 if (sa)
11580                         ctl_set_invalid_field(ctsio,
11581                                               /*sks_valid*/ 1,
11582                                               /*command*/ 1,
11583                                               /*field*/ 1,
11584                                               /*bit_valid*/ 1,
11585                                               /*bit*/ 4);
11586                 else
11587                         ctl_set_invalid_opcode(ctsio);
11588                 ctl_done((union ctl_io *)ctsio);
11589                 return (NULL);
11590         }
11591         KASSERT(entry->length > 0,
11592             ("Not defined length for command 0x%02x/0x%02x",
11593              ctsio->cdb[0], ctsio->cdb[1]));
11594         for (i = 1; i < entry->length; i++) {
11595                 diff = ctsio->cdb[i] & ~entry->usage[i - 1];
11596                 if (diff == 0)
11597                         continue;
11598                 ctl_set_invalid_field(ctsio,
11599                                       /*sks_valid*/ 1,
11600                                       /*command*/ 1,
11601                                       /*field*/ i,
11602                                       /*bit_valid*/ 1,
11603                                       /*bit*/ fls(diff) - 1);
11604                 ctl_done((union ctl_io *)ctsio);
11605                 return (NULL);
11606         }
11607         return (entry);
11608 }
11609
11610 static int
11611 ctl_cmd_applicable(uint8_t lun_type, const struct ctl_cmd_entry *entry)
11612 {
11613
11614         switch (lun_type) {
11615         case T_DIRECT:
11616                 if ((entry->flags & CTL_CMD_FLAG_OK_ON_DIRECT) == 0)
11617                         return (0);
11618                 break;
11619         case T_PROCESSOR:
11620                 if ((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0)
11621                         return (0);
11622                 break;
11623         case T_CDROM:
11624                 if ((entry->flags & CTL_CMD_FLAG_OK_ON_CDROM) == 0)
11625                         return (0);
11626                 break;
11627         default:
11628                 return (0);
11629         }
11630         return (1);
11631 }
11632
11633 static int
11634 ctl_scsiio(struct ctl_scsiio *ctsio)
11635 {
11636         int retval;
11637         const struct ctl_cmd_entry *entry;
11638
11639         retval = CTL_RETVAL_COMPLETE;
11640
11641         CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
11642
11643         entry = ctl_get_cmd_entry(ctsio, NULL);
11644
11645         /*
11646          * If this I/O has been aborted, just send it straight to
11647          * ctl_done() without executing it.
11648          */
11649         if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
11650                 ctl_done((union ctl_io *)ctsio);
11651                 goto bailout;
11652         }
11653
11654         /*
11655          * All the checks should have been handled by ctl_scsiio_precheck().
11656          * We should be clear now to just execute the I/O.
11657          */
11658         retval = entry->execute(ctsio);
11659
11660 bailout:
11661         return (retval);
11662 }
11663
11664 static int
11665 ctl_target_reset(union ctl_io *io)
11666 {
11667         struct ctl_softc *softc = CTL_SOFTC(io);
11668         struct ctl_port *port = CTL_PORT(io);
11669         struct ctl_lun *lun;
11670         uint32_t initidx;
11671         ctl_ua_type ua_type;
11672
11673         if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11674                 union ctl_ha_msg msg_info;
11675
11676                 msg_info.hdr.nexus = io->io_hdr.nexus;
11677                 msg_info.task.task_action = io->taskio.task_action;
11678                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11679                 msg_info.hdr.original_sc = NULL;
11680                 msg_info.hdr.serializing_sc = NULL;
11681                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11682                     sizeof(msg_info.task), M_WAITOK);
11683         }
11684
11685         initidx = ctl_get_initindex(&io->io_hdr.nexus);
11686         if (io->taskio.task_action == CTL_TASK_TARGET_RESET)
11687                 ua_type = CTL_UA_TARG_RESET;
11688         else
11689                 ua_type = CTL_UA_BUS_RESET;
11690         mtx_lock(&softc->ctl_lock);
11691         STAILQ_FOREACH(lun, &softc->lun_list, links) {
11692                 if (port != NULL &&
11693                     ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX)
11694                         continue;
11695                 ctl_do_lun_reset(lun, initidx, ua_type);
11696         }
11697         mtx_unlock(&softc->ctl_lock);
11698         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11699         return (0);
11700 }
11701
11702 /*
11703  * The LUN should always be set.  The I/O is optional, and is used to
11704  * distinguish between I/Os sent by this initiator, and by other
11705  * initiators.  We set unit attention for initiators other than this one.
11706  * SAM-3 is vague on this point.  It does say that a unit attention should
11707  * be established for other initiators when a LUN is reset (see section
11708  * 5.7.3), but it doesn't specifically say that the unit attention should
11709  * be established for this particular initiator when a LUN is reset.  Here
11710  * is the relevant text, from SAM-3 rev 8:
11711  *
11712  * 5.7.2 When a SCSI initiator port aborts its own tasks
11713  *
11714  * When a SCSI initiator port causes its own task(s) to be aborted, no
11715  * notification that the task(s) have been aborted shall be returned to
11716  * the SCSI initiator port other than the completion response for the
11717  * command or task management function action that caused the task(s) to
11718  * be aborted and notification(s) associated with related effects of the
11719  * action (e.g., a reset unit attention condition).
11720  *
11721  * XXX KDM for now, we're setting unit attention for all initiators.
11722  */
11723 static void
11724 ctl_do_lun_reset(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua_type)
11725 {
11726         union ctl_io *xio;
11727         int i;
11728
11729         mtx_lock(&lun->lun_lock);
11730         /* Abort tasks. */
11731         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11732              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11733                 xio->io_hdr.flags |= CTL_FLAG_ABORT | CTL_FLAG_ABORT_STATUS;
11734         }
11735         /* Clear CA. */
11736         for (i = 0; i < ctl_max_ports; i++) {
11737                 free(lun->pending_sense[i], M_CTL);
11738                 lun->pending_sense[i] = NULL;
11739         }
11740         /* Clear reservation. */
11741         lun->flags &= ~CTL_LUN_RESERVED;
11742         /* Clear prevent media removal. */
11743         if (lun->prevent) {
11744                 for (i = 0; i < CTL_MAX_INITIATORS; i++)
11745                         ctl_clear_mask(lun->prevent, i);
11746                 lun->prevent_count = 0;
11747         }
11748         /* Clear TPC status */
11749         ctl_tpc_lun_clear(lun, -1);
11750         /* Establish UA. */
11751 #if 0
11752         ctl_est_ua_all(lun, initidx, ua_type);
11753 #else
11754         ctl_est_ua_all(lun, -1, ua_type);
11755 #endif
11756         mtx_unlock(&lun->lun_lock);
11757 }
11758
11759 static int
11760 ctl_lun_reset(union ctl_io *io)
11761 {
11762         struct ctl_softc *softc = CTL_SOFTC(io);
11763         struct ctl_lun *lun;
11764         uint32_t targ_lun, initidx;
11765
11766         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11767         initidx = ctl_get_initindex(&io->io_hdr.nexus);
11768         mtx_lock(&softc->ctl_lock);
11769         if (targ_lun >= ctl_max_luns ||
11770             (lun = softc->ctl_luns[targ_lun]) == NULL) {
11771                 mtx_unlock(&softc->ctl_lock);
11772                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
11773                 return (1);
11774         }
11775         ctl_do_lun_reset(lun, initidx, CTL_UA_LUN_RESET);
11776         mtx_unlock(&softc->ctl_lock);
11777         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11778
11779         if ((io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) == 0) {
11780                 union ctl_ha_msg msg_info;
11781
11782                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11783                 msg_info.hdr.nexus = io->io_hdr.nexus;
11784                 msg_info.task.task_action = CTL_TASK_LUN_RESET;
11785                 msg_info.hdr.original_sc = NULL;
11786                 msg_info.hdr.serializing_sc = NULL;
11787                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11788                     sizeof(msg_info.task), M_WAITOK);
11789         }
11790         return (0);
11791 }
11792
11793 static void
11794 ctl_abort_tasks_lun(struct ctl_lun *lun, uint32_t targ_port, uint32_t init_id,
11795     int other_sc)
11796 {
11797         union ctl_io *xio;
11798
11799         mtx_assert(&lun->lun_lock, MA_OWNED);
11800
11801         /*
11802          * Run through the OOA queue and attempt to find the given I/O.
11803          * The target port, initiator ID, tag type and tag number have to
11804          * match the values that we got from the initiator.  If we have an
11805          * untagged command to abort, simply abort the first untagged command
11806          * we come to.  We only allow one untagged command at a time of course.
11807          */
11808         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11809              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11810
11811                 if ((targ_port == UINT32_MAX ||
11812                      targ_port == xio->io_hdr.nexus.targ_port) &&
11813                     (init_id == UINT32_MAX ||
11814                      init_id == xio->io_hdr.nexus.initid)) {
11815                         if (targ_port != xio->io_hdr.nexus.targ_port ||
11816                             init_id != xio->io_hdr.nexus.initid)
11817                                 xio->io_hdr.flags |= CTL_FLAG_ABORT_STATUS;
11818                         xio->io_hdr.flags |= CTL_FLAG_ABORT;
11819                         if (!other_sc && !(lun->flags & CTL_LUN_PRIMARY_SC)) {
11820                                 union ctl_ha_msg msg_info;
11821
11822                                 msg_info.hdr.nexus = xio->io_hdr.nexus;
11823                                 msg_info.task.task_action = CTL_TASK_ABORT_TASK;
11824                                 msg_info.task.tag_num = xio->scsiio.tag_num;
11825                                 msg_info.task.tag_type = xio->scsiio.tag_type;
11826                                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11827                                 msg_info.hdr.original_sc = NULL;
11828                                 msg_info.hdr.serializing_sc = NULL;
11829                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11830                                     sizeof(msg_info.task), M_NOWAIT);
11831                         }
11832                 }
11833         }
11834 }
11835
11836 static int
11837 ctl_abort_task_set(union ctl_io *io)
11838 {
11839         struct ctl_softc *softc = CTL_SOFTC(io);
11840         struct ctl_lun *lun;
11841         uint32_t targ_lun;
11842
11843         /*
11844          * Look up the LUN.
11845          */
11846         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11847         mtx_lock(&softc->ctl_lock);
11848         if (targ_lun >= ctl_max_luns ||
11849             (lun = softc->ctl_luns[targ_lun]) == NULL) {
11850                 mtx_unlock(&softc->ctl_lock);
11851                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
11852                 return (1);
11853         }
11854
11855         mtx_lock(&lun->lun_lock);
11856         mtx_unlock(&softc->ctl_lock);
11857         if (io->taskio.task_action == CTL_TASK_ABORT_TASK_SET) {
11858                 ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
11859                     io->io_hdr.nexus.initid,
11860                     (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
11861         } else { /* CTL_TASK_CLEAR_TASK_SET */
11862                 ctl_abort_tasks_lun(lun, UINT32_MAX, UINT32_MAX,
11863                     (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
11864         }
11865         mtx_unlock(&lun->lun_lock);
11866         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11867         return (0);
11868 }
11869
11870 static void
11871 ctl_i_t_nexus_loss(struct ctl_softc *softc, uint32_t initidx,
11872     ctl_ua_type ua_type)
11873 {
11874         struct ctl_lun *lun;
11875         struct scsi_sense_data *ps;
11876         uint32_t p, i;
11877
11878         p = initidx / CTL_MAX_INIT_PER_PORT;
11879         i = initidx % CTL_MAX_INIT_PER_PORT;
11880         mtx_lock(&softc->ctl_lock);
11881         STAILQ_FOREACH(lun, &softc->lun_list, links) {
11882                 mtx_lock(&lun->lun_lock);
11883                 /* Abort tasks. */
11884                 ctl_abort_tasks_lun(lun, p, i, 1);
11885                 /* Clear CA. */
11886                 ps = lun->pending_sense[p];
11887                 if (ps != NULL)
11888                         ps[i].error_code = 0;
11889                 /* Clear reservation. */
11890                 if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == initidx))
11891                         lun->flags &= ~CTL_LUN_RESERVED;
11892                 /* Clear prevent media removal. */
11893                 if (lun->prevent && ctl_is_set(lun->prevent, initidx)) {
11894                         ctl_clear_mask(lun->prevent, initidx);
11895                         lun->prevent_count--;
11896                 }
11897                 /* Clear TPC status */
11898                 ctl_tpc_lun_clear(lun, initidx);
11899                 /* Establish UA. */
11900                 ctl_est_ua(lun, initidx, ua_type);
11901                 mtx_unlock(&lun->lun_lock);
11902         }
11903         mtx_unlock(&softc->ctl_lock);
11904 }
11905
11906 static int
11907 ctl_i_t_nexus_reset(union ctl_io *io)
11908 {
11909         struct ctl_softc *softc = CTL_SOFTC(io);
11910         uint32_t initidx;
11911
11912         if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11913                 union ctl_ha_msg msg_info;
11914
11915                 msg_info.hdr.nexus = io->io_hdr.nexus;
11916                 msg_info.task.task_action = CTL_TASK_I_T_NEXUS_RESET;
11917                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11918                 msg_info.hdr.original_sc = NULL;
11919                 msg_info.hdr.serializing_sc = NULL;
11920                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11921                     sizeof(msg_info.task), M_WAITOK);
11922         }
11923
11924         initidx = ctl_get_initindex(&io->io_hdr.nexus);
11925         ctl_i_t_nexus_loss(softc, initidx, CTL_UA_I_T_NEXUS_LOSS);
11926         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11927         return (0);
11928 }
11929
11930 static int
11931 ctl_abort_task(union ctl_io *io)
11932 {
11933         struct ctl_softc *softc = CTL_SOFTC(io);
11934         union ctl_io *xio;
11935         struct ctl_lun *lun;
11936 #if 0
11937         struct sbuf sb;
11938         char printbuf[128];
11939 #endif
11940         int found;
11941         uint32_t targ_lun;
11942
11943         found = 0;
11944
11945         /*
11946          * Look up the LUN.
11947          */
11948         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11949         mtx_lock(&softc->ctl_lock);
11950         if (targ_lun >= ctl_max_luns ||
11951             (lun = softc->ctl_luns[targ_lun]) == NULL) {
11952                 mtx_unlock(&softc->ctl_lock);
11953                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
11954                 return (1);
11955         }
11956
11957 #if 0
11958         printf("ctl_abort_task: called for lun %lld, tag %d type %d\n",
11959                lun->lun, io->taskio.tag_num, io->taskio.tag_type);
11960 #endif
11961
11962         mtx_lock(&lun->lun_lock);
11963         mtx_unlock(&softc->ctl_lock);
11964         /*
11965          * Run through the OOA queue and attempt to find the given I/O.
11966          * The target port, initiator ID, tag type and tag number have to
11967          * match the values that we got from the initiator.  If we have an
11968          * untagged command to abort, simply abort the first untagged command
11969          * we come to.  We only allow one untagged command at a time of course.
11970          */
11971         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11972              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11973 #if 0
11974                 sbuf_new(&sb, printbuf, sizeof(printbuf), SBUF_FIXEDLEN);
11975
11976                 sbuf_printf(&sb, "LUN %lld tag %d type %d%s%s%s%s: ",
11977                             lun->lun, xio->scsiio.tag_num,
11978                             xio->scsiio.tag_type,
11979                             (xio->io_hdr.blocked_links.tqe_prev
11980                             == NULL) ? "" : " BLOCKED",
11981                             (xio->io_hdr.flags &
11982                             CTL_FLAG_DMA_INPROG) ? " DMA" : "",
11983                             (xio->io_hdr.flags &
11984                             CTL_FLAG_ABORT) ? " ABORT" : "",
11985                             (xio->io_hdr.flags &
11986                             CTL_FLAG_IS_WAS_ON_RTR ? " RTR" : ""));
11987                 ctl_scsi_command_string(&xio->scsiio, NULL, &sb);
11988                 sbuf_finish(&sb);
11989                 printf("%s\n", sbuf_data(&sb));
11990 #endif
11991
11992                 if ((xio->io_hdr.nexus.targ_port != io->io_hdr.nexus.targ_port)
11993                  || (xio->io_hdr.nexus.initid != io->io_hdr.nexus.initid)
11994                  || (xio->io_hdr.flags & CTL_FLAG_ABORT))
11995                         continue;
11996
11997                 /*
11998                  * If the abort says that the task is untagged, the
11999                  * task in the queue must be untagged.  Otherwise,
12000                  * we just check to see whether the tag numbers
12001                  * match.  This is because the QLogic firmware
12002                  * doesn't pass back the tag type in an abort
12003                  * request.
12004                  */
12005 #if 0
12006                 if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
12007                   && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
12008                  || (xio->scsiio.tag_num == io->taskio.tag_num))
12009 #endif
12010                 /*
12011                  * XXX KDM we've got problems with FC, because it
12012                  * doesn't send down a tag type with aborts.  So we
12013                  * can only really go by the tag number...
12014                  * This may cause problems with parallel SCSI.
12015                  * Need to figure that out!!
12016                  */
12017                 if (xio->scsiio.tag_num == io->taskio.tag_num) {
12018                         xio->io_hdr.flags |= CTL_FLAG_ABORT;
12019                         found = 1;
12020                         if ((io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) == 0 &&
12021                             !(lun->flags & CTL_LUN_PRIMARY_SC)) {
12022                                 union ctl_ha_msg msg_info;
12023
12024                                 msg_info.hdr.nexus = io->io_hdr.nexus;
12025                                 msg_info.task.task_action = CTL_TASK_ABORT_TASK;
12026                                 msg_info.task.tag_num = io->taskio.tag_num;
12027                                 msg_info.task.tag_type = io->taskio.tag_type;
12028                                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
12029                                 msg_info.hdr.original_sc = NULL;
12030                                 msg_info.hdr.serializing_sc = NULL;
12031 #if 0
12032                                 printf("Sent Abort to other side\n");
12033 #endif
12034                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
12035                                     sizeof(msg_info.task), M_NOWAIT);
12036                         }
12037 #if 0
12038                         printf("ctl_abort_task: found I/O to abort\n");
12039 #endif
12040                 }
12041         }
12042         mtx_unlock(&lun->lun_lock);
12043
12044         if (found == 0) {
12045                 /*
12046                  * This isn't really an error.  It's entirely possible for
12047                  * the abort and command completion to cross on the wire.
12048                  * This is more of an informative/diagnostic error.
12049                  */
12050 #if 0
12051                 printf("ctl_abort_task: ABORT sent for nonexistent I/O: "
12052                        "%u:%u:%u tag %d type %d\n",
12053                        io->io_hdr.nexus.initid,
12054                        io->io_hdr.nexus.targ_port,
12055                        io->io_hdr.nexus.targ_lun, io->taskio.tag_num,
12056                        io->taskio.tag_type);
12057 #endif
12058         }
12059         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12060         return (0);
12061 }
12062
12063 static int
12064 ctl_query_task(union ctl_io *io, int task_set)
12065 {
12066         struct ctl_softc *softc = CTL_SOFTC(io);
12067         union ctl_io *xio;
12068         struct ctl_lun *lun;
12069         int found = 0;
12070         uint32_t targ_lun;
12071
12072         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12073         mtx_lock(&softc->ctl_lock);
12074         if (targ_lun >= ctl_max_luns ||
12075             (lun = softc->ctl_luns[targ_lun]) == NULL) {
12076                 mtx_unlock(&softc->ctl_lock);
12077                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
12078                 return (1);
12079         }
12080         mtx_lock(&lun->lun_lock);
12081         mtx_unlock(&softc->ctl_lock);
12082         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12083              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12084
12085                 if ((xio->io_hdr.nexus.targ_port != io->io_hdr.nexus.targ_port)
12086                  || (xio->io_hdr.nexus.initid != io->io_hdr.nexus.initid)
12087                  || (xio->io_hdr.flags & CTL_FLAG_ABORT))
12088                         continue;
12089
12090                 if (task_set || xio->scsiio.tag_num == io->taskio.tag_num) {
12091                         found = 1;
12092                         break;
12093                 }
12094         }
12095         mtx_unlock(&lun->lun_lock);
12096         if (found)
12097                 io->taskio.task_status = CTL_TASK_FUNCTION_SUCCEEDED;
12098         else
12099                 io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12100         return (0);
12101 }
12102
12103 static int
12104 ctl_query_async_event(union ctl_io *io)
12105 {
12106         struct ctl_softc *softc = CTL_SOFTC(io);
12107         struct ctl_lun *lun;
12108         ctl_ua_type ua;
12109         uint32_t targ_lun, initidx;
12110
12111         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12112         mtx_lock(&softc->ctl_lock);
12113         if (targ_lun >= ctl_max_luns ||
12114             (lun = softc->ctl_luns[targ_lun]) == NULL) {
12115                 mtx_unlock(&softc->ctl_lock);
12116                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
12117                 return (1);
12118         }
12119         mtx_lock(&lun->lun_lock);
12120         mtx_unlock(&softc->ctl_lock);
12121         initidx = ctl_get_initindex(&io->io_hdr.nexus);
12122         ua = ctl_build_qae(lun, initidx, io->taskio.task_resp);
12123         mtx_unlock(&lun->lun_lock);
12124         if (ua != CTL_UA_NONE)
12125                 io->taskio.task_status = CTL_TASK_FUNCTION_SUCCEEDED;
12126         else
12127                 io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12128         return (0);
12129 }
12130
12131 static void
12132 ctl_run_task(union ctl_io *io)
12133 {
12134         int retval = 1;
12135
12136         CTL_DEBUG_PRINT(("ctl_run_task\n"));
12137         KASSERT(io->io_hdr.io_type == CTL_IO_TASK,
12138             ("ctl_run_task: Unextected io_type %d\n", io->io_hdr.io_type));
12139         io->taskio.task_status = CTL_TASK_FUNCTION_NOT_SUPPORTED;
12140         bzero(io->taskio.task_resp, sizeof(io->taskio.task_resp));
12141         switch (io->taskio.task_action) {
12142         case CTL_TASK_ABORT_TASK:
12143                 retval = ctl_abort_task(io);
12144                 break;
12145         case CTL_TASK_ABORT_TASK_SET:
12146         case CTL_TASK_CLEAR_TASK_SET:
12147                 retval = ctl_abort_task_set(io);
12148                 break;
12149         case CTL_TASK_CLEAR_ACA:
12150                 break;
12151         case CTL_TASK_I_T_NEXUS_RESET:
12152                 retval = ctl_i_t_nexus_reset(io);
12153                 break;
12154         case CTL_TASK_LUN_RESET:
12155                 retval = ctl_lun_reset(io);
12156                 break;
12157         case CTL_TASK_TARGET_RESET:
12158         case CTL_TASK_BUS_RESET:
12159                 retval = ctl_target_reset(io);
12160                 break;
12161         case CTL_TASK_PORT_LOGIN:
12162                 break;
12163         case CTL_TASK_PORT_LOGOUT:
12164                 break;
12165         case CTL_TASK_QUERY_TASK:
12166                 retval = ctl_query_task(io, 0);
12167                 break;
12168         case CTL_TASK_QUERY_TASK_SET:
12169                 retval = ctl_query_task(io, 1);
12170                 break;
12171         case CTL_TASK_QUERY_ASYNC_EVENT:
12172                 retval = ctl_query_async_event(io);
12173                 break;
12174         default:
12175                 printf("%s: got unknown task management event %d\n",
12176                        __func__, io->taskio.task_action);
12177                 break;
12178         }
12179         if (retval == 0)
12180                 io->io_hdr.status = CTL_SUCCESS;
12181         else
12182                 io->io_hdr.status = CTL_ERROR;
12183         ctl_done(io);
12184 }
12185
12186 /*
12187  * For HA operation.  Handle commands that come in from the other
12188  * controller.
12189  */
12190 static void
12191 ctl_handle_isc(union ctl_io *io)
12192 {
12193         struct ctl_softc *softc = CTL_SOFTC(io);
12194         struct ctl_lun *lun;
12195         const struct ctl_cmd_entry *entry;
12196         uint32_t targ_lun;
12197
12198         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12199         switch (io->io_hdr.msg_type) {
12200         case CTL_MSG_SERIALIZE:
12201                 ctl_serialize_other_sc_cmd(&io->scsiio);
12202                 break;
12203         case CTL_MSG_R2R:               /* Only used in SER_ONLY mode. */
12204                 entry = ctl_get_cmd_entry(&io->scsiio, NULL);
12205                 if (targ_lun >= ctl_max_luns ||
12206                     (lun = softc->ctl_luns[targ_lun]) == NULL) {
12207                         ctl_done(io);
12208                         break;
12209                 }
12210                 mtx_lock(&lun->lun_lock);
12211                 if (ctl_scsiio_lun_check(lun, entry, &io->scsiio) != 0) {
12212                         mtx_unlock(&lun->lun_lock);
12213                         ctl_done(io);
12214                         break;
12215                 }
12216                 io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
12217                 mtx_unlock(&lun->lun_lock);
12218                 ctl_enqueue_rtr(io);
12219                 break;
12220         case CTL_MSG_FINISH_IO:
12221                 if (softc->ha_mode == CTL_HA_MODE_XFER) {
12222                         ctl_done(io);
12223                         break;
12224                 }
12225                 if (targ_lun >= ctl_max_luns ||
12226                     (lun = softc->ctl_luns[targ_lun]) == NULL) {
12227                         ctl_free_io(io);
12228                         break;
12229                 }
12230                 mtx_lock(&lun->lun_lock);
12231                 TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
12232                 ctl_check_blocked(lun);
12233                 mtx_unlock(&lun->lun_lock);
12234                 ctl_free_io(io);
12235                 break;
12236         case CTL_MSG_PERS_ACTION:
12237                 ctl_hndl_per_res_out_on_other_sc(io);
12238                 ctl_free_io(io);
12239                 break;
12240         case CTL_MSG_BAD_JUJU:
12241                 ctl_done(io);
12242                 break;
12243         case CTL_MSG_DATAMOVE:          /* Only used in XFER mode */
12244                 ctl_datamove_remote(io);
12245                 break;
12246         case CTL_MSG_DATAMOVE_DONE:     /* Only used in XFER mode */
12247                 io->scsiio.be_move_done(io);
12248                 break;
12249         case CTL_MSG_FAILOVER:
12250                 ctl_failover_lun(io);
12251                 ctl_free_io(io);
12252                 break;
12253         default:
12254                 printf("%s: Invalid message type %d\n",
12255                        __func__, io->io_hdr.msg_type);
12256                 ctl_free_io(io);
12257                 break;
12258         }
12259
12260 }
12261
12262
12263 /*
12264  * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
12265  * there is no match.
12266  */
12267 static ctl_lun_error_pattern
12268 ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
12269 {
12270         const struct ctl_cmd_entry *entry;
12271         ctl_lun_error_pattern filtered_pattern, pattern;
12272
12273         pattern = desc->error_pattern;
12274
12275         /*
12276          * XXX KDM we need more data passed into this function to match a
12277          * custom pattern, and we actually need to implement custom pattern
12278          * matching.
12279          */
12280         if (pattern & CTL_LUN_PAT_CMD)
12281                 return (CTL_LUN_PAT_CMD);
12282
12283         if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
12284                 return (CTL_LUN_PAT_ANY);
12285
12286         entry = ctl_get_cmd_entry(ctsio, NULL);
12287
12288         filtered_pattern = entry->pattern & pattern;
12289
12290         /*
12291          * If the user requested specific flags in the pattern (e.g.
12292          * CTL_LUN_PAT_RANGE), make sure the command supports all of those
12293          * flags.
12294          *
12295          * If the user did not specify any flags, it doesn't matter whether
12296          * or not the command supports the flags.
12297          */
12298         if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
12299              (pattern & ~CTL_LUN_PAT_MASK))
12300                 return (CTL_LUN_PAT_NONE);
12301
12302         /*
12303          * If the user asked for a range check, see if the requested LBA
12304          * range overlaps with this command's LBA range.
12305          */
12306         if (filtered_pattern & CTL_LUN_PAT_RANGE) {
12307                 uint64_t lba1;
12308                 uint64_t len1;
12309                 ctl_action action;
12310                 int retval;
12311
12312                 retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
12313                 if (retval != 0)
12314                         return (CTL_LUN_PAT_NONE);
12315
12316                 action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
12317                                               desc->lba_range.len, FALSE);
12318                 /*
12319                  * A "pass" means that the LBA ranges don't overlap, so
12320                  * this doesn't match the user's range criteria.
12321                  */
12322                 if (action == CTL_ACTION_PASS)
12323                         return (CTL_LUN_PAT_NONE);
12324         }
12325
12326         return (filtered_pattern);
12327 }
12328
12329 static void
12330 ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
12331 {
12332         struct ctl_error_desc *desc, *desc2;
12333
12334         mtx_assert(&lun->lun_lock, MA_OWNED);
12335
12336         STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
12337                 ctl_lun_error_pattern pattern;
12338                 /*
12339                  * Check to see whether this particular command matches
12340                  * the pattern in the descriptor.
12341                  */
12342                 pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
12343                 if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
12344                         continue;
12345
12346                 switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
12347                 case CTL_LUN_INJ_ABORTED:
12348                         ctl_set_aborted(&io->scsiio);
12349                         break;
12350                 case CTL_LUN_INJ_MEDIUM_ERR:
12351                         ctl_set_medium_error(&io->scsiio,
12352                             (io->io_hdr.flags & CTL_FLAG_DATA_MASK) !=
12353                              CTL_FLAG_DATA_OUT);
12354                         break;
12355                 case CTL_LUN_INJ_UA:
12356                         /* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
12357                          * OCCURRED */
12358                         ctl_set_ua(&io->scsiio, 0x29, 0x00);
12359                         break;
12360                 case CTL_LUN_INJ_CUSTOM:
12361                         /*
12362                          * We're assuming the user knows what he is doing.
12363                          * Just copy the sense information without doing
12364                          * checks.
12365                          */
12366                         bcopy(&desc->custom_sense, &io->scsiio.sense_data,
12367                               MIN(sizeof(desc->custom_sense),
12368                                   sizeof(io->scsiio.sense_data)));
12369                         io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
12370                         io->scsiio.sense_len = SSD_FULL_SIZE;
12371                         io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
12372                         break;
12373                 case CTL_LUN_INJ_NONE:
12374                 default:
12375                         /*
12376                          * If this is an error injection type we don't know
12377                          * about, clear the continuous flag (if it is set)
12378                          * so it will get deleted below.
12379                          */
12380                         desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
12381                         break;
12382                 }
12383                 /*
12384                  * By default, each error injection action is a one-shot
12385                  */
12386                 if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
12387                         continue;
12388
12389                 STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
12390
12391                 free(desc, M_CTL);
12392         }
12393 }
12394
12395 #ifdef CTL_IO_DELAY
12396 static void
12397 ctl_datamove_timer_wakeup(void *arg)
12398 {
12399         union ctl_io *io;
12400
12401         io = (union ctl_io *)arg;
12402
12403         ctl_datamove(io);
12404 }
12405 #endif /* CTL_IO_DELAY */
12406
12407 void
12408 ctl_datamove(union ctl_io *io)
12409 {
12410         void (*fe_datamove)(union ctl_io *io);
12411
12412         mtx_assert(&((struct ctl_softc *)CTL_SOFTC(io))->ctl_lock, MA_NOTOWNED);
12413
12414         CTL_DEBUG_PRINT(("ctl_datamove\n"));
12415
12416         /* No data transferred yet.  Frontend must update this when done. */
12417         io->scsiio.kern_data_resid = io->scsiio.kern_data_len;
12418
12419 #ifdef CTL_TIME_IO
12420         if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12421                 char str[256];
12422                 char path_str[64];
12423                 struct sbuf sb;
12424
12425                 ctl_scsi_path_string(io, path_str, sizeof(path_str));
12426                 sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12427
12428                 sbuf_cat(&sb, path_str);
12429                 switch (io->io_hdr.io_type) {
12430                 case CTL_IO_SCSI:
12431                         ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12432                         sbuf_printf(&sb, "\n");
12433                         sbuf_cat(&sb, path_str);
12434                         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12435                                     io->scsiio.tag_num, io->scsiio.tag_type);
12436                         break;
12437                 case CTL_IO_TASK:
12438                         sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12439                                     "Tag Type: %d\n", io->taskio.task_action,
12440                                     io->taskio.tag_num, io->taskio.tag_type);
12441                         break;
12442                 default:
12443                         panic("%s: Invalid CTL I/O type %d\n",
12444                             __func__, io->io_hdr.io_type);
12445                 }
12446                 sbuf_cat(&sb, path_str);
12447                 sbuf_printf(&sb, "ctl_datamove: %jd seconds\n",
12448                             (intmax_t)time_uptime - io->io_hdr.start_time);
12449                 sbuf_finish(&sb);
12450                 printf("%s", sbuf_data(&sb));
12451         }
12452 #endif /* CTL_TIME_IO */
12453
12454 #ifdef CTL_IO_DELAY
12455         if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
12456                 io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
12457         } else {
12458                 struct ctl_lun *lun;
12459
12460                 lun = CTL_LUN(io);
12461                 if ((lun != NULL)
12462                  && (lun->delay_info.datamove_delay > 0)) {
12463
12464                         callout_init(&io->io_hdr.delay_callout, /*mpsafe*/ 1);
12465                         io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
12466                         callout_reset(&io->io_hdr.delay_callout,
12467                                       lun->delay_info.datamove_delay * hz,
12468                                       ctl_datamove_timer_wakeup, io);
12469                         if (lun->delay_info.datamove_type ==
12470                             CTL_DELAY_TYPE_ONESHOT)
12471                                 lun->delay_info.datamove_delay = 0;
12472                         return;
12473                 }
12474         }
12475 #endif
12476
12477         /*
12478          * This command has been aborted.  Set the port status, so we fail
12479          * the data move.
12480          */
12481         if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12482                 printf("ctl_datamove: tag 0x%04x on (%u:%u:%u) aborted\n",
12483                        io->scsiio.tag_num, io->io_hdr.nexus.initid,
12484                        io->io_hdr.nexus.targ_port,
12485                        io->io_hdr.nexus.targ_lun);
12486                 io->io_hdr.port_status = 31337;
12487                 /*
12488                  * Note that the backend, in this case, will get the
12489                  * callback in its context.  In other cases it may get
12490                  * called in the frontend's interrupt thread context.
12491                  */
12492                 io->scsiio.be_move_done(io);
12493                 return;
12494         }
12495
12496         /* Don't confuse frontend with zero length data move. */
12497         if (io->scsiio.kern_data_len == 0) {
12498                 io->scsiio.be_move_done(io);
12499                 return;
12500         }
12501
12502         fe_datamove = CTL_PORT(io)->fe_datamove;
12503         fe_datamove(io);
12504 }
12505
12506 static void
12507 ctl_send_datamove_done(union ctl_io *io, int have_lock)
12508 {
12509         union ctl_ha_msg msg;
12510 #ifdef CTL_TIME_IO
12511         struct bintime cur_bt;
12512 #endif
12513
12514         memset(&msg, 0, sizeof(msg));
12515         msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
12516         msg.hdr.original_sc = io;
12517         msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
12518         msg.hdr.nexus = io->io_hdr.nexus;
12519         msg.hdr.status = io->io_hdr.status;
12520         msg.scsi.kern_data_resid = io->scsiio.kern_data_resid;
12521         msg.scsi.tag_num = io->scsiio.tag_num;
12522         msg.scsi.tag_type = io->scsiio.tag_type;
12523         msg.scsi.scsi_status = io->scsiio.scsi_status;
12524         memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
12525                io->scsiio.sense_len);
12526         msg.scsi.sense_len = io->scsiio.sense_len;
12527         msg.scsi.port_status = io->io_hdr.port_status;
12528         io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12529         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12530                 ctl_failover_io(io, /*have_lock*/ have_lock);
12531                 return;
12532         }
12533         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12534             sizeof(msg.scsi) - sizeof(msg.scsi.sense_data) +
12535             msg.scsi.sense_len, M_WAITOK);
12536
12537 #ifdef CTL_TIME_IO
12538         getbinuptime(&cur_bt);
12539         bintime_sub(&cur_bt, &io->io_hdr.dma_start_bt);
12540         bintime_add(&io->io_hdr.dma_bt, &cur_bt);
12541 #endif
12542         io->io_hdr.num_dmas++;
12543 }
12544
12545 /*
12546  * The DMA to the remote side is done, now we need to tell the other side
12547  * we're done so it can continue with its data movement.
12548  */
12549 static void
12550 ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
12551 {
12552         union ctl_io *io;
12553         uint32_t i;
12554
12555         io = rq->context;
12556
12557         if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12558                 printf("%s: ISC DMA write failed with error %d", __func__,
12559                        rq->ret);
12560                 ctl_set_internal_failure(&io->scsiio,
12561                                          /*sks_valid*/ 1,
12562                                          /*retry_count*/ rq->ret);
12563         }
12564
12565         ctl_dt_req_free(rq);
12566
12567         for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12568                 free(io->io_hdr.local_sglist[i].addr, M_CTL);
12569         free(io->io_hdr.remote_sglist, M_CTL);
12570         io->io_hdr.remote_sglist = NULL;
12571         io->io_hdr.local_sglist = NULL;
12572
12573         /*
12574          * The data is in local and remote memory, so now we need to send
12575          * status (good or back) back to the other side.
12576          */
12577         ctl_send_datamove_done(io, /*have_lock*/ 0);
12578 }
12579
12580 /*
12581  * We've moved the data from the host/controller into local memory.  Now we
12582  * need to push it over to the remote controller's memory.
12583  */
12584 static int
12585 ctl_datamove_remote_dm_write_cb(union ctl_io *io)
12586 {
12587         int retval;
12588
12589         retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
12590                                           ctl_datamove_remote_write_cb);
12591         return (retval);
12592 }
12593
12594 static void
12595 ctl_datamove_remote_write(union ctl_io *io)
12596 {
12597         int retval;
12598         void (*fe_datamove)(union ctl_io *io);
12599
12600         /*
12601          * - Get the data from the host/HBA into local memory.
12602          * - DMA memory from the local controller to the remote controller.
12603          * - Send status back to the remote controller.
12604          */
12605
12606         retval = ctl_datamove_remote_sgl_setup(io);
12607         if (retval != 0)
12608                 return;
12609
12610         /* Switch the pointer over so the FETD knows what to do */
12611         io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12612
12613         /*
12614          * Use a custom move done callback, since we need to send completion
12615          * back to the other controller, not to the backend on this side.
12616          */
12617         io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
12618
12619         fe_datamove = CTL_PORT(io)->fe_datamove;
12620         fe_datamove(io);
12621 }
12622
12623 static int
12624 ctl_datamove_remote_dm_read_cb(union ctl_io *io)
12625 {
12626 #if 0
12627         char str[256];
12628         char path_str[64];
12629         struct sbuf sb;
12630 #endif
12631         uint32_t i;
12632
12633         for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12634                 free(io->io_hdr.local_sglist[i].addr, M_CTL);
12635         free(io->io_hdr.remote_sglist, M_CTL);
12636         io->io_hdr.remote_sglist = NULL;
12637         io->io_hdr.local_sglist = NULL;
12638
12639 #if 0
12640         scsi_path_string(io, path_str, sizeof(path_str));
12641         sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12642         sbuf_cat(&sb, path_str);
12643         scsi_command_string(&io->scsiio, NULL, &sb);
12644         sbuf_printf(&sb, "\n");
12645         sbuf_cat(&sb, path_str);
12646         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12647                     io->scsiio.tag_num, io->scsiio.tag_type);
12648         sbuf_cat(&sb, path_str);
12649         sbuf_printf(&sb, "%s: flags %#x, status %#x\n", __func__,
12650                     io->io_hdr.flags, io->io_hdr.status);
12651         sbuf_finish(&sb);
12652         printk("%s", sbuf_data(&sb));
12653 #endif
12654
12655
12656         /*
12657          * The read is done, now we need to send status (good or bad) back
12658          * to the other side.
12659          */
12660         ctl_send_datamove_done(io, /*have_lock*/ 0);
12661
12662         return (0);
12663 }
12664
12665 static void
12666 ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
12667 {
12668         union ctl_io *io;
12669         void (*fe_datamove)(union ctl_io *io);
12670
12671         io = rq->context;
12672
12673         if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12674                 printf("%s: ISC DMA read failed with error %d\n", __func__,
12675                        rq->ret);
12676                 ctl_set_internal_failure(&io->scsiio,
12677                                          /*sks_valid*/ 1,
12678                                          /*retry_count*/ rq->ret);
12679         }
12680
12681         ctl_dt_req_free(rq);
12682
12683         /* Switch the pointer over so the FETD knows what to do */
12684         io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12685
12686         /*
12687          * Use a custom move done callback, since we need to send completion
12688          * back to the other controller, not to the backend on this side.
12689          */
12690         io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
12691
12692         /* XXX KDM add checks like the ones in ctl_datamove? */
12693
12694         fe_datamove = CTL_PORT(io)->fe_datamove;
12695         fe_datamove(io);
12696 }
12697
12698 static int
12699 ctl_datamove_remote_sgl_setup(union ctl_io *io)
12700 {
12701         struct ctl_sg_entry *local_sglist;
12702         uint32_t len_to_go;
12703         int retval;
12704         int i;
12705
12706         retval = 0;
12707         local_sglist = io->io_hdr.local_sglist;
12708         len_to_go = io->scsiio.kern_data_len;
12709
12710         /*
12711          * The difficult thing here is that the size of the various
12712          * S/G segments may be different than the size from the
12713          * remote controller.  That'll make it harder when DMAing
12714          * the data back to the other side.
12715          */
12716         for (i = 0; len_to_go > 0; i++) {
12717                 local_sglist[i].len = MIN(len_to_go, CTL_HA_DATAMOVE_SEGMENT);
12718                 local_sglist[i].addr =
12719                     malloc(local_sglist[i].len, M_CTL, M_WAITOK);
12720
12721                 len_to_go -= local_sglist[i].len;
12722         }
12723         /*
12724          * Reset the number of S/G entries accordingly.  The original
12725          * number of S/G entries is available in rem_sg_entries.
12726          */
12727         io->scsiio.kern_sg_entries = i;
12728
12729 #if 0
12730         printf("%s: kern_sg_entries = %d\n", __func__,
12731                io->scsiio.kern_sg_entries);
12732         for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12733                 printf("%s: sg[%d] = %p, %lu\n", __func__, i,
12734                        local_sglist[i].addr, local_sglist[i].len);
12735 #endif
12736
12737         return (retval);
12738 }
12739
12740 static int
12741 ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
12742                          ctl_ha_dt_cb callback)
12743 {
12744         struct ctl_ha_dt_req *rq;
12745         struct ctl_sg_entry *remote_sglist, *local_sglist;
12746         uint32_t local_used, remote_used, total_used;
12747         int i, j, isc_ret;
12748
12749         rq = ctl_dt_req_alloc();
12750
12751         /*
12752          * If we failed to allocate the request, and if the DMA didn't fail
12753          * anyway, set busy status.  This is just a resource allocation
12754          * failure.
12755          */
12756         if ((rq == NULL)
12757          && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
12758              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS))
12759                 ctl_set_busy(&io->scsiio);
12760
12761         if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
12762             (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS) {
12763
12764                 if (rq != NULL)
12765                         ctl_dt_req_free(rq);
12766
12767                 /*
12768                  * The data move failed.  We need to return status back
12769                  * to the other controller.  No point in trying to DMA
12770                  * data to the remote controller.
12771                  */
12772
12773                 ctl_send_datamove_done(io, /*have_lock*/ 0);
12774
12775                 return (1);
12776         }
12777
12778         local_sglist = io->io_hdr.local_sglist;
12779         remote_sglist = io->io_hdr.remote_sglist;
12780         local_used = 0;
12781         remote_used = 0;
12782         total_used = 0;
12783
12784         /*
12785          * Pull/push the data over the wire from/to the other controller.
12786          * This takes into account the possibility that the local and
12787          * remote sglists may not be identical in terms of the size of
12788          * the elements and the number of elements.
12789          *
12790          * One fundamental assumption here is that the length allocated for
12791          * both the local and remote sglists is identical.  Otherwise, we've
12792          * essentially got a coding error of some sort.
12793          */
12794         isc_ret = CTL_HA_STATUS_SUCCESS;
12795         for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
12796                 uint32_t cur_len;
12797                 uint8_t *tmp_ptr;
12798
12799                 rq->command = command;
12800                 rq->context = io;
12801
12802                 /*
12803                  * Both pointers should be aligned.  But it is possible
12804                  * that the allocation length is not.  They should both
12805                  * also have enough slack left over at the end, though,
12806                  * to round up to the next 8 byte boundary.
12807                  */
12808                 cur_len = MIN(local_sglist[i].len - local_used,
12809                               remote_sglist[j].len - remote_used);
12810                 rq->size = cur_len;
12811
12812                 tmp_ptr = (uint8_t *)local_sglist[i].addr;
12813                 tmp_ptr += local_used;
12814
12815 #if 0
12816                 /* Use physical addresses when talking to ISC hardware */
12817                 if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
12818                         /* XXX KDM use busdma */
12819                         rq->local = vtophys(tmp_ptr);
12820                 } else
12821                         rq->local = tmp_ptr;
12822 #else
12823                 KASSERT((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0,
12824                     ("HA does not support BUS_ADDR"));
12825                 rq->local = tmp_ptr;
12826 #endif
12827
12828                 tmp_ptr = (uint8_t *)remote_sglist[j].addr;
12829                 tmp_ptr += remote_used;
12830                 rq->remote = tmp_ptr;
12831
12832                 rq->callback = NULL;
12833
12834                 local_used += cur_len;
12835                 if (local_used >= local_sglist[i].len) {
12836                         i++;
12837                         local_used = 0;
12838                 }
12839
12840                 remote_used += cur_len;
12841                 if (remote_used >= remote_sglist[j].len) {
12842                         j++;
12843                         remote_used = 0;
12844                 }
12845                 total_used += cur_len;
12846
12847                 if (total_used >= io->scsiio.kern_data_len)
12848                         rq->callback = callback;
12849
12850 #if 0
12851                 printf("%s: %s: local %p remote %p size %d\n", __func__,
12852                        (command == CTL_HA_DT_CMD_WRITE) ? "WRITE" : "READ",
12853                        rq->local, rq->remote, rq->size);
12854 #endif
12855
12856                 isc_ret = ctl_dt_single(rq);
12857                 if (isc_ret > CTL_HA_STATUS_SUCCESS)
12858                         break;
12859         }
12860         if (isc_ret != CTL_HA_STATUS_WAIT) {
12861                 rq->ret = isc_ret;
12862                 callback(rq);
12863         }
12864
12865         return (0);
12866 }
12867
12868 static void
12869 ctl_datamove_remote_read(union ctl_io *io)
12870 {
12871         int retval;
12872         uint32_t i;
12873
12874         /*
12875          * This will send an error to the other controller in the case of a
12876          * failure.
12877          */
12878         retval = ctl_datamove_remote_sgl_setup(io);
12879         if (retval != 0)
12880                 return;
12881
12882         retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
12883                                           ctl_datamove_remote_read_cb);
12884         if (retval != 0) {
12885                 /*
12886                  * Make sure we free memory if there was an error..  The
12887                  * ctl_datamove_remote_xfer() function will send the
12888                  * datamove done message, or call the callback with an
12889                  * error if there is a problem.
12890                  */
12891                 for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12892                         free(io->io_hdr.local_sglist[i].addr, M_CTL);
12893                 free(io->io_hdr.remote_sglist, M_CTL);
12894                 io->io_hdr.remote_sglist = NULL;
12895                 io->io_hdr.local_sglist = NULL;
12896         }
12897 }
12898
12899 /*
12900  * Process a datamove request from the other controller.  This is used for
12901  * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
12902  * first.  Once that is complete, the data gets DMAed into the remote
12903  * controller's memory.  For reads, we DMA from the remote controller's
12904  * memory into our memory first, and then move it out to the FETD.
12905  */
12906 static void
12907 ctl_datamove_remote(union ctl_io *io)
12908 {
12909
12910         mtx_assert(&((struct ctl_softc *)CTL_SOFTC(io))->ctl_lock, MA_NOTOWNED);
12911
12912         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12913                 ctl_failover_io(io, /*have_lock*/ 0);
12914                 return;
12915         }
12916
12917         /*
12918          * Note that we look for an aborted I/O here, but don't do some of
12919          * the other checks that ctl_datamove() normally does.
12920          * We don't need to run the datamove delay code, since that should
12921          * have been done if need be on the other controller.
12922          */
12923         if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12924                 printf("%s: tag 0x%04x on (%u:%u:%u) aborted\n", __func__,
12925                        io->scsiio.tag_num, io->io_hdr.nexus.initid,
12926                        io->io_hdr.nexus.targ_port,
12927                        io->io_hdr.nexus.targ_lun);
12928                 io->io_hdr.port_status = 31338;
12929                 ctl_send_datamove_done(io, /*have_lock*/ 0);
12930                 return;
12931         }
12932
12933         if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT)
12934                 ctl_datamove_remote_write(io);
12935         else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN)
12936                 ctl_datamove_remote_read(io);
12937         else {
12938                 io->io_hdr.port_status = 31339;
12939                 ctl_send_datamove_done(io, /*have_lock*/ 0);
12940         }
12941 }
12942
12943 static void
12944 ctl_process_done(union ctl_io *io)
12945 {
12946         struct ctl_softc *softc = CTL_SOFTC(io);
12947         struct ctl_port *port = CTL_PORT(io);
12948         struct ctl_lun *lun = CTL_LUN(io);
12949         void (*fe_done)(union ctl_io *io);
12950         union ctl_ha_msg msg;
12951
12952         CTL_DEBUG_PRINT(("ctl_process_done\n"));
12953         fe_done = port->fe_done;
12954
12955 #ifdef CTL_TIME_IO
12956         if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12957                 char str[256];
12958                 char path_str[64];
12959                 struct sbuf sb;
12960
12961                 ctl_scsi_path_string(io, path_str, sizeof(path_str));
12962                 sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12963
12964                 sbuf_cat(&sb, path_str);
12965                 switch (io->io_hdr.io_type) {
12966                 case CTL_IO_SCSI:
12967                         ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12968                         sbuf_printf(&sb, "\n");
12969                         sbuf_cat(&sb, path_str);
12970                         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12971                                     io->scsiio.tag_num, io->scsiio.tag_type);
12972                         break;
12973                 case CTL_IO_TASK:
12974                         sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12975                                     "Tag Type: %d\n", io->taskio.task_action,
12976                                     io->taskio.tag_num, io->taskio.tag_type);
12977                         break;
12978                 default:
12979                         panic("%s: Invalid CTL I/O type %d\n",
12980                             __func__, io->io_hdr.io_type);
12981                 }
12982                 sbuf_cat(&sb, path_str);
12983                 sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
12984                             (intmax_t)time_uptime - io->io_hdr.start_time);
12985                 sbuf_finish(&sb);
12986                 printf("%s", sbuf_data(&sb));
12987         }
12988 #endif /* CTL_TIME_IO */
12989
12990         switch (io->io_hdr.io_type) {
12991         case CTL_IO_SCSI:
12992                 break;
12993         case CTL_IO_TASK:
12994                 if (ctl_debug & CTL_DEBUG_INFO)
12995                         ctl_io_error_print(io, NULL);
12996                 fe_done(io);
12997                 return;
12998         default:
12999                 panic("%s: Invalid CTL I/O type %d\n",
13000                     __func__, io->io_hdr.io_type);
13001         }
13002
13003         if (lun == NULL) {
13004                 CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
13005                                  io->io_hdr.nexus.targ_mapped_lun));
13006                 goto bailout;
13007         }
13008
13009         mtx_lock(&lun->lun_lock);
13010
13011         /*
13012          * Check to see if we have any informational exception and status
13013          * of this command can be modified to report it in form of either
13014          * RECOVERED ERROR or NO SENSE, depending on MRIE mode page field.
13015          */
13016         if (lun->ie_reported == 0 && lun->ie_asc != 0 &&
13017             io->io_hdr.status == CTL_SUCCESS &&
13018             (io->io_hdr.flags & CTL_FLAG_STATUS_SENT) == 0) {
13019                 uint8_t mrie = lun->MODE_IE.mrie;
13020                 uint8_t per = ((lun->MODE_RWER.byte3 & SMS_RWER_PER) ||
13021                     (lun->MODE_VER.byte3 & SMS_VER_PER));
13022                 if (((mrie == SIEP_MRIE_REC_COND && per) ||
13023                      mrie == SIEP_MRIE_REC_UNCOND ||
13024                      mrie == SIEP_MRIE_NO_SENSE) &&
13025                     (ctl_get_cmd_entry(&io->scsiio, NULL)->flags &
13026                      CTL_CMD_FLAG_NO_SENSE) == 0) {
13027                         ctl_set_sense(&io->scsiio,
13028                               /*current_error*/ 1,
13029                               /*sense_key*/ (mrie == SIEP_MRIE_NO_SENSE) ?
13030                                 SSD_KEY_NO_SENSE : SSD_KEY_RECOVERED_ERROR,
13031                               /*asc*/ lun->ie_asc,
13032                               /*ascq*/ lun->ie_ascq,
13033                               SSD_ELEM_NONE);
13034                         lun->ie_reported = 1;
13035                 }
13036         } else if (lun->ie_reported < 0)
13037                 lun->ie_reported = 0;
13038
13039         /*
13040          * Check to see if we have any errors to inject here.  We only
13041          * inject errors for commands that don't already have errors set.
13042          */
13043         if (!STAILQ_EMPTY(&lun->error_list) &&
13044             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS) &&
13045             ((io->io_hdr.flags & CTL_FLAG_STATUS_SENT) == 0))
13046                 ctl_inject_error(lun, io);
13047
13048         /*
13049          * XXX KDM how do we treat commands that aren't completed
13050          * successfully?
13051          *
13052          * XXX KDM should we also track I/O latency?
13053          */
13054         if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS &&
13055             io->io_hdr.io_type == CTL_IO_SCSI) {
13056                 int type;
13057 #ifdef CTL_TIME_IO
13058                 struct bintime bt;
13059
13060                 getbinuptime(&bt);
13061                 bintime_sub(&bt, &io->io_hdr.start_bt);
13062 #endif
13063                 if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13064                     CTL_FLAG_DATA_IN)
13065                         type = CTL_STATS_READ;
13066                 else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13067                     CTL_FLAG_DATA_OUT)
13068                         type = CTL_STATS_WRITE;
13069                 else
13070                         type = CTL_STATS_NO_IO;
13071
13072 #ifdef CTL_LEGACY_STATS
13073                 uint32_t targ_port = port->targ_port;
13074                 lun->legacy_stats.ports[targ_port].bytes[type] +=
13075                     io->scsiio.kern_total_len;
13076                 lun->legacy_stats.ports[targ_port].operations[type] ++;
13077                 lun->legacy_stats.ports[targ_port].num_dmas[type] +=
13078                     io->io_hdr.num_dmas;
13079 #ifdef CTL_TIME_IO
13080                 bintime_add(&lun->legacy_stats.ports[targ_port].dma_time[type],
13081                    &io->io_hdr.dma_bt);
13082                 bintime_add(&lun->legacy_stats.ports[targ_port].time[type],
13083                     &bt);
13084 #endif
13085 #endif /* CTL_LEGACY_STATS */
13086
13087                 lun->stats.bytes[type] += io->scsiio.kern_total_len;
13088                 lun->stats.operations[type] ++;
13089                 lun->stats.dmas[type] += io->io_hdr.num_dmas;
13090 #ifdef CTL_TIME_IO
13091                 bintime_add(&lun->stats.dma_time[type], &io->io_hdr.dma_bt);
13092                 bintime_add(&lun->stats.time[type], &bt);
13093 #endif
13094
13095                 mtx_lock(&port->port_lock);
13096                 port->stats.bytes[type] += io->scsiio.kern_total_len;
13097                 port->stats.operations[type] ++;
13098                 port->stats.dmas[type] += io->io_hdr.num_dmas;
13099 #ifdef CTL_TIME_IO
13100                 bintime_add(&port->stats.dma_time[type], &io->io_hdr.dma_bt);
13101                 bintime_add(&port->stats.time[type], &bt);
13102 #endif
13103                 mtx_unlock(&port->port_lock);
13104         }
13105
13106         /*
13107          * Remove this from the OOA queue.
13108          */
13109         TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
13110 #ifdef CTL_TIME_IO
13111         if (TAILQ_EMPTY(&lun->ooa_queue))
13112                 lun->last_busy = getsbinuptime();
13113 #endif
13114
13115         /*
13116          * Run through the blocked queue on this LUN and see if anything
13117          * has become unblocked, now that this transaction is done.
13118          */
13119         ctl_check_blocked(lun);
13120
13121         /*
13122          * If the LUN has been invalidated, free it if there is nothing
13123          * left on its OOA queue.
13124          */
13125         if ((lun->flags & CTL_LUN_INVALID)
13126          && TAILQ_EMPTY(&lun->ooa_queue)) {
13127                 mtx_unlock(&lun->lun_lock);
13128                 ctl_free_lun(lun);
13129         } else
13130                 mtx_unlock(&lun->lun_lock);
13131
13132 bailout:
13133
13134         /*
13135          * If this command has been aborted, make sure we set the status
13136          * properly.  The FETD is responsible for freeing the I/O and doing
13137          * whatever it needs to do to clean up its state.
13138          */
13139         if (io->io_hdr.flags & CTL_FLAG_ABORT)
13140                 ctl_set_task_aborted(&io->scsiio);
13141
13142         /*
13143          * If enabled, print command error status.
13144          */
13145         if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS &&
13146             (ctl_debug & CTL_DEBUG_INFO) != 0)
13147                 ctl_io_error_print(io, NULL);
13148
13149         /*
13150          * Tell the FETD or the other shelf controller we're done with this
13151          * command.  Note that only SCSI commands get to this point.  Task
13152          * management commands are completed above.
13153          */
13154         if ((softc->ha_mode != CTL_HA_MODE_XFER) &&
13155             (io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)) {
13156                 memset(&msg, 0, sizeof(msg));
13157                 msg.hdr.msg_type = CTL_MSG_FINISH_IO;
13158                 msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
13159                 msg.hdr.nexus = io->io_hdr.nexus;
13160                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13161                     sizeof(msg.scsi) - sizeof(msg.scsi.sense_data),
13162                     M_WAITOK);
13163         }
13164
13165         fe_done(io);
13166 }
13167
13168 /*
13169  * Front end should call this if it doesn't do autosense.  When the request
13170  * sense comes back in from the initiator, we'll dequeue this and send it.
13171  */
13172 int
13173 ctl_queue_sense(union ctl_io *io)
13174 {
13175         struct ctl_softc *softc = CTL_SOFTC(io);
13176         struct ctl_port *port = CTL_PORT(io);
13177         struct ctl_lun *lun;
13178         struct scsi_sense_data *ps;
13179         uint32_t initidx, p, targ_lun;
13180
13181         CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
13182
13183         targ_lun = ctl_lun_map_from_port(port, io->io_hdr.nexus.targ_lun);
13184
13185         /*
13186          * LUN lookup will likely move to the ctl_work_thread() once we
13187          * have our new queueing infrastructure (that doesn't put things on
13188          * a per-LUN queue initially).  That is so that we can handle
13189          * things like an INQUIRY to a LUN that we don't have enabled.  We
13190          * can't deal with that right now.
13191          * If we don't have a LUN for this, just toss the sense information.
13192          */
13193         mtx_lock(&softc->ctl_lock);
13194         if (targ_lun >= ctl_max_luns ||
13195             (lun = softc->ctl_luns[targ_lun]) == NULL) {
13196                 mtx_unlock(&softc->ctl_lock);
13197                 goto bailout;
13198         }
13199         mtx_lock(&lun->lun_lock);
13200         mtx_unlock(&softc->ctl_lock);
13201
13202         initidx = ctl_get_initindex(&io->io_hdr.nexus);
13203         p = initidx / CTL_MAX_INIT_PER_PORT;
13204         if (lun->pending_sense[p] == NULL) {
13205                 lun->pending_sense[p] = malloc(sizeof(*ps) * CTL_MAX_INIT_PER_PORT,
13206                     M_CTL, M_NOWAIT | M_ZERO);
13207         }
13208         if ((ps = lun->pending_sense[p]) != NULL) {
13209                 ps += initidx % CTL_MAX_INIT_PER_PORT;
13210                 memset(ps, 0, sizeof(*ps));
13211                 memcpy(ps, &io->scsiio.sense_data, io->scsiio.sense_len);
13212         }
13213         mtx_unlock(&lun->lun_lock);
13214
13215 bailout:
13216         ctl_free_io(io);
13217         return (CTL_RETVAL_COMPLETE);
13218 }
13219
13220 /*
13221  * Primary command inlet from frontend ports.  All SCSI and task I/O
13222  * requests must go through this function.
13223  */
13224 int
13225 ctl_queue(union ctl_io *io)
13226 {
13227         struct ctl_port *port = CTL_PORT(io);
13228
13229         CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
13230
13231 #ifdef CTL_TIME_IO
13232         io->io_hdr.start_time = time_uptime;
13233         getbinuptime(&io->io_hdr.start_bt);
13234 #endif /* CTL_TIME_IO */
13235
13236         /* Map FE-specific LUN ID into global one. */
13237         io->io_hdr.nexus.targ_mapped_lun =
13238             ctl_lun_map_from_port(port, io->io_hdr.nexus.targ_lun);
13239
13240         switch (io->io_hdr.io_type) {
13241         case CTL_IO_SCSI:
13242         case CTL_IO_TASK:
13243                 if (ctl_debug & CTL_DEBUG_CDB)
13244                         ctl_io_print(io);
13245                 ctl_enqueue_incoming(io);
13246                 break;
13247         default:
13248                 printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
13249                 return (EINVAL);
13250         }
13251
13252         return (CTL_RETVAL_COMPLETE);
13253 }
13254
13255 #ifdef CTL_IO_DELAY
13256 static void
13257 ctl_done_timer_wakeup(void *arg)
13258 {
13259         union ctl_io *io;
13260
13261         io = (union ctl_io *)arg;
13262         ctl_done(io);
13263 }
13264 #endif /* CTL_IO_DELAY */
13265
13266 void
13267 ctl_serseq_done(union ctl_io *io)
13268 {
13269         struct ctl_lun *lun = CTL_LUN(io);;
13270
13271         if (lun->be_lun == NULL ||
13272             lun->be_lun->serseq == CTL_LUN_SERSEQ_OFF)
13273                 return;
13274         mtx_lock(&lun->lun_lock);
13275         io->io_hdr.flags |= CTL_FLAG_SERSEQ_DONE;
13276         ctl_check_blocked(lun);
13277         mtx_unlock(&lun->lun_lock);
13278 }
13279
13280 void
13281 ctl_done(union ctl_io *io)
13282 {
13283
13284         /*
13285          * Enable this to catch duplicate completion issues.
13286          */
13287 #if 0
13288         if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
13289                 printf("%s: type %d msg %d cdb %x iptl: "
13290                        "%u:%u:%u tag 0x%04x "
13291                        "flag %#x status %x\n",
13292                         __func__,
13293                         io->io_hdr.io_type,
13294                         io->io_hdr.msg_type,
13295                         io->scsiio.cdb[0],
13296                         io->io_hdr.nexus.initid,
13297                         io->io_hdr.nexus.targ_port,
13298                         io->io_hdr.nexus.targ_lun,
13299                         (io->io_hdr.io_type ==
13300                         CTL_IO_TASK) ?
13301                         io->taskio.tag_num :
13302                         io->scsiio.tag_num,
13303                         io->io_hdr.flags,
13304                         io->io_hdr.status);
13305         } else
13306                 io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
13307 #endif
13308
13309         /*
13310          * This is an internal copy of an I/O, and should not go through
13311          * the normal done processing logic.
13312          */
13313         if (io->io_hdr.flags & CTL_FLAG_INT_COPY)
13314                 return;
13315
13316 #ifdef CTL_IO_DELAY
13317         if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
13318                 io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
13319         } else {
13320                 struct ctl_lun *lun = CTL_LUN(io);
13321
13322                 if ((lun != NULL)
13323                  && (lun->delay_info.done_delay > 0)) {
13324
13325                         callout_init(&io->io_hdr.delay_callout, /*mpsafe*/ 1);
13326                         io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
13327                         callout_reset(&io->io_hdr.delay_callout,
13328                                       lun->delay_info.done_delay * hz,
13329                                       ctl_done_timer_wakeup, io);
13330                         if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
13331                                 lun->delay_info.done_delay = 0;
13332                         return;
13333                 }
13334         }
13335 #endif /* CTL_IO_DELAY */
13336
13337         ctl_enqueue_done(io);
13338 }
13339
13340 static void
13341 ctl_work_thread(void *arg)
13342 {
13343         struct ctl_thread *thr = (struct ctl_thread *)arg;
13344         struct ctl_softc *softc = thr->ctl_softc;
13345         union ctl_io *io;
13346         int retval;
13347
13348         CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
13349
13350         while (!softc->shutdown) {
13351                 /*
13352                  * We handle the queues in this order:
13353                  * - ISC
13354                  * - done queue (to free up resources, unblock other commands)
13355                  * - incoming queue
13356                  * - RtR queue
13357                  *
13358                  * If those queues are empty, we break out of the loop and
13359                  * go to sleep.
13360                  */
13361                 mtx_lock(&thr->queue_lock);
13362                 io = (union ctl_io *)STAILQ_FIRST(&thr->isc_queue);
13363                 if (io != NULL) {
13364                         STAILQ_REMOVE_HEAD(&thr->isc_queue, links);
13365                         mtx_unlock(&thr->queue_lock);
13366                         ctl_handle_isc(io);
13367                         continue;
13368                 }
13369                 io = (union ctl_io *)STAILQ_FIRST(&thr->done_queue);
13370                 if (io != NULL) {
13371                         STAILQ_REMOVE_HEAD(&thr->done_queue, links);
13372                         /* clear any blocked commands, call fe_done */
13373                         mtx_unlock(&thr->queue_lock);
13374                         ctl_process_done(io);
13375                         continue;
13376                 }
13377                 io = (union ctl_io *)STAILQ_FIRST(&thr->incoming_queue);
13378                 if (io != NULL) {
13379                         STAILQ_REMOVE_HEAD(&thr->incoming_queue, links);
13380                         mtx_unlock(&thr->queue_lock);
13381                         if (io->io_hdr.io_type == CTL_IO_TASK)
13382                                 ctl_run_task(io);
13383                         else
13384                                 ctl_scsiio_precheck(softc, &io->scsiio);
13385                         continue;
13386                 }
13387                 io = (union ctl_io *)STAILQ_FIRST(&thr->rtr_queue);
13388                 if (io != NULL) {
13389                         STAILQ_REMOVE_HEAD(&thr->rtr_queue, links);
13390                         mtx_unlock(&thr->queue_lock);
13391                         retval = ctl_scsiio(&io->scsiio);
13392                         if (retval != CTL_RETVAL_COMPLETE)
13393                                 CTL_DEBUG_PRINT(("ctl_scsiio failed\n"));
13394                         continue;
13395                 }
13396
13397                 /* Sleep until we have something to do. */
13398                 mtx_sleep(thr, &thr->queue_lock, PDROP | PRIBIO, "-", 0);
13399         }
13400         thr->thread = NULL;
13401         kthread_exit();
13402 }
13403
13404 static void
13405 ctl_lun_thread(void *arg)
13406 {
13407         struct ctl_softc *softc = (struct ctl_softc *)arg;
13408         struct ctl_be_lun *be_lun;
13409
13410         CTL_DEBUG_PRINT(("ctl_lun_thread starting\n"));
13411
13412         while (!softc->shutdown) {
13413                 mtx_lock(&softc->ctl_lock);
13414                 be_lun = STAILQ_FIRST(&softc->pending_lun_queue);
13415                 if (be_lun != NULL) {
13416                         STAILQ_REMOVE_HEAD(&softc->pending_lun_queue, links);
13417                         mtx_unlock(&softc->ctl_lock);
13418                         ctl_create_lun(be_lun);
13419                         continue;
13420                 }
13421
13422                 /* Sleep until we have something to do. */
13423                 mtx_sleep(&softc->pending_lun_queue, &softc->ctl_lock,
13424                     PDROP | PRIBIO, "-", 0);
13425         }
13426         softc->lun_thread = NULL;
13427         kthread_exit();
13428 }
13429
13430 static void
13431 ctl_thresh_thread(void *arg)
13432 {
13433         struct ctl_softc *softc = (struct ctl_softc *)arg;
13434         struct ctl_lun *lun;
13435         struct ctl_logical_block_provisioning_page *page;
13436         const char *attr;
13437         union ctl_ha_msg msg;
13438         uint64_t thres, val;
13439         int i, e, set;
13440
13441         CTL_DEBUG_PRINT(("ctl_thresh_thread starting\n"));
13442
13443         while (!softc->shutdown) {
13444                 mtx_lock(&softc->ctl_lock);
13445                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
13446                         if ((lun->flags & CTL_LUN_DISABLED) ||
13447                             (lun->flags & CTL_LUN_NO_MEDIA) ||
13448                             lun->backend->lun_attr == NULL)
13449                                 continue;
13450                         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
13451                             softc->ha_mode == CTL_HA_MODE_XFER)
13452                                 continue;
13453                         if ((lun->MODE_RWER.byte8 & SMS_RWER_LBPERE) == 0)
13454                                 continue;
13455                         e = 0;
13456                         page = &lun->MODE_LBP;
13457                         for (i = 0; i < CTL_NUM_LBP_THRESH; i++) {
13458                                 if ((page->descr[i].flags & SLBPPD_ENABLED) == 0)
13459                                         continue;
13460                                 thres = scsi_4btoul(page->descr[i].count);
13461                                 thres <<= CTL_LBP_EXPONENT;
13462                                 switch (page->descr[i].resource) {
13463                                 case 0x01:
13464                                         attr = "blocksavail";
13465                                         break;
13466                                 case 0x02:
13467                                         attr = "blocksused";
13468                                         break;
13469                                 case 0xf1:
13470                                         attr = "poolblocksavail";
13471                                         break;
13472                                 case 0xf2:
13473                                         attr = "poolblocksused";
13474                                         break;
13475                                 default:
13476                                         continue;
13477                                 }
13478                                 mtx_unlock(&softc->ctl_lock); // XXX
13479                                 val = lun->backend->lun_attr(
13480                                     lun->be_lun->be_lun, attr);
13481                                 mtx_lock(&softc->ctl_lock);
13482                                 if (val == UINT64_MAX)
13483                                         continue;
13484                                 if ((page->descr[i].flags & SLBPPD_ARMING_MASK)
13485                                     == SLBPPD_ARMING_INC)
13486                                         e = (val >= thres);
13487                                 else
13488                                         e = (val <= thres);
13489                                 if (e)
13490                                         break;
13491                         }
13492                         mtx_lock(&lun->lun_lock);
13493                         if (e) {
13494                                 scsi_u64to8b((uint8_t *)&page->descr[i] -
13495                                     (uint8_t *)page, lun->ua_tpt_info);
13496                                 if (lun->lasttpt == 0 ||
13497                                     time_uptime - lun->lasttpt >= CTL_LBP_UA_PERIOD) {
13498                                         lun->lasttpt = time_uptime;
13499                                         ctl_est_ua_all(lun, -1, CTL_UA_THIN_PROV_THRES);
13500                                         set = 1;
13501                                 } else
13502                                         set = 0;
13503                         } else {
13504                                 lun->lasttpt = 0;
13505                                 ctl_clr_ua_all(lun, -1, CTL_UA_THIN_PROV_THRES);
13506                                 set = -1;
13507                         }
13508                         mtx_unlock(&lun->lun_lock);
13509                         if (set != 0 &&
13510                             lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
13511                                 /* Send msg to other side. */
13512                                 bzero(&msg.ua, sizeof(msg.ua));
13513                                 msg.hdr.msg_type = CTL_MSG_UA;
13514                                 msg.hdr.nexus.initid = -1;
13515                                 msg.hdr.nexus.targ_port = -1;
13516                                 msg.hdr.nexus.targ_lun = lun->lun;
13517                                 msg.hdr.nexus.targ_mapped_lun = lun->lun;
13518                                 msg.ua.ua_all = 1;
13519                                 msg.ua.ua_set = (set > 0);
13520                                 msg.ua.ua_type = CTL_UA_THIN_PROV_THRES;
13521                                 memcpy(msg.ua.ua_info, lun->ua_tpt_info, 8);
13522                                 mtx_unlock(&softc->ctl_lock); // XXX
13523                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13524                                     sizeof(msg.ua), M_WAITOK);
13525                                 mtx_lock(&softc->ctl_lock);
13526                         }
13527                 }
13528                 mtx_sleep(&softc->thresh_thread, &softc->ctl_lock,
13529                     PDROP | PRIBIO, "-", CTL_LBP_PERIOD * hz);
13530         }
13531         softc->thresh_thread = NULL;
13532         kthread_exit();
13533 }
13534
13535 static void
13536 ctl_enqueue_incoming(union ctl_io *io)
13537 {
13538         struct ctl_softc *softc = CTL_SOFTC(io);
13539         struct ctl_thread *thr;
13540         u_int idx;
13541
13542         idx = (io->io_hdr.nexus.targ_port * 127 +
13543                io->io_hdr.nexus.initid) % worker_threads;
13544         thr = &softc->threads[idx];
13545         mtx_lock(&thr->queue_lock);
13546         STAILQ_INSERT_TAIL(&thr->incoming_queue, &io->io_hdr, links);
13547         mtx_unlock(&thr->queue_lock);
13548         wakeup(thr);
13549 }
13550
13551 static void
13552 ctl_enqueue_rtr(union ctl_io *io)
13553 {
13554         struct ctl_softc *softc = CTL_SOFTC(io);
13555         struct ctl_thread *thr;
13556
13557         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13558         mtx_lock(&thr->queue_lock);
13559         STAILQ_INSERT_TAIL(&thr->rtr_queue, &io->io_hdr, links);
13560         mtx_unlock(&thr->queue_lock);
13561         wakeup(thr);
13562 }
13563
13564 static void
13565 ctl_enqueue_done(union ctl_io *io)
13566 {
13567         struct ctl_softc *softc = CTL_SOFTC(io);
13568         struct ctl_thread *thr;
13569
13570         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13571         mtx_lock(&thr->queue_lock);
13572         STAILQ_INSERT_TAIL(&thr->done_queue, &io->io_hdr, links);
13573         mtx_unlock(&thr->queue_lock);
13574         wakeup(thr);
13575 }
13576
13577 static void
13578 ctl_enqueue_isc(union ctl_io *io)
13579 {
13580         struct ctl_softc *softc = CTL_SOFTC(io);
13581         struct ctl_thread *thr;
13582
13583         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13584         mtx_lock(&thr->queue_lock);
13585         STAILQ_INSERT_TAIL(&thr->isc_queue, &io->io_hdr, links);
13586         mtx_unlock(&thr->queue_lock);
13587         wakeup(thr);
13588 }
13589
13590 /*
13591  *  vim: ts=8
13592  */