sys/cam/ctl/ctl.c

   1 /*-
   2  * Copyright (c) 2003-2009 Silicon Graphics International Corp.
   3  * Copyright (c) 2012 The FreeBSD Foundation
   4  * Copyright (c) 2015 Alexander Motin <mav@FreeBSD.org>
   5  * All rights reserved.
   6  *
   7  * Portions of this software were developed by Edward Tomasz Napierala
   8  * under sponsorship from the FreeBSD Foundation.
   9  *
  10  * Redistribution and use in source and binary forms, with or without
  11  * modification, are permitted provided that the following conditions
  12  * are met:
  13  * 1. Redistributions of source code must retain the above copyright
  14  *    notice, this list of conditions, and the following disclaimer,
  15  *    without modification.
  16  * 2. Redistributions in binary form must reproduce at minimum a disclaimer
  17  *    substantially similar to the "NO WARRANTY" disclaimer below
  18  *    ("Disclaimer") and any redistribution must be conditioned upon
  19  *    including a substantially similar Disclaimer requirement for further
  20  *    binary redistribution.
  21  *
  22  * NO WARRANTY
  23  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  24  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  25  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR
  26  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  27  * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  28  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  29  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  30  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
  31  * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
  32  * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
  33  * POSSIBILITY OF SUCH DAMAGES.
  34  *
  35  * $Id$
  36  */
  37 /*
  38  * CAM Target Layer, a SCSI device emulation subsystem.
  39  *
  40  * Author: Ken Merry <ken@FreeBSD.org>
  41  */
  42
  43 #define _CTL_C
  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 copan_debugconf_subpage debugconf_page_default = {
  97         DBGCNF_PAGE_CODE | SMPH_SPF,    /* page_code */
  98         DBGCNF_SUBPAGE_CODE,            /* subpage */
  99         {(sizeof(struct copan_debugconf_subpage) - 4) >> 8,
 100          (sizeof(struct copan_debugconf_subpage) - 4) >> 0}, /* page_length */
 101         DBGCNF_VERSION,                 /* page_version */
 102         {CTL_TIME_IO_DEFAULT_SECS>>8,
 103          CTL_TIME_IO_DEFAULT_SECS>>0},  /* ctl_time_io_secs */
 104 };
 105
 106 const static struct copan_debugconf_subpage debugconf_page_changeable = {
 107         DBGCNF_PAGE_CODE | SMPH_SPF,    /* page_code */
 108         DBGCNF_SUBPAGE_CODE,            /* subpage */
 109         {(sizeof(struct copan_debugconf_subpage) - 4) >> 8,
 110          (sizeof(struct copan_debugconf_subpage) - 4) >> 0}, /* page_length */
 111         0,                              /* page_version */
 112         {0xff,0xff},                    /* ctl_time_io_secs */
 113 };
 114
 115 const static struct scsi_da_rw_recovery_page rw_er_page_default = {
 116         /*page_code*/SMS_RW_ERROR_RECOVERY_PAGE,
 117         /*page_length*/sizeof(struct scsi_da_rw_recovery_page) - 2,
 118         /*byte3*/SMS_RWER_AWRE|SMS_RWER_ARRE,
 119         /*read_retry_count*/0,
 120         /*correction_span*/0,
 121         /*head_offset_count*/0,
 122         /*data_strobe_offset_cnt*/0,
 123         /*byte8*/SMS_RWER_LBPERE,
 124         /*write_retry_count*/0,
 125         /*reserved2*/0,
 126         /*recovery_time_limit*/{0, 0},
 127 };
 128
 129 const static struct scsi_da_rw_recovery_page rw_er_page_changeable = {
 130         /*page_code*/SMS_RW_ERROR_RECOVERY_PAGE,
 131         /*page_length*/sizeof(struct scsi_da_rw_recovery_page) - 2,
 132         /*byte3*/0,
 133         /*read_retry_count*/0,
 134         /*correction_span*/0,
 135         /*head_offset_count*/0,
 136         /*data_strobe_offset_cnt*/0,
 137         /*byte8*/0,
 138         /*write_retry_count*/0,
 139         /*reserved2*/0,
 140         /*recovery_time_limit*/{0, 0},
 141 };
 142
 143 const static struct scsi_format_page format_page_default = {
 144         /*page_code*/SMS_FORMAT_DEVICE_PAGE,
 145         /*page_length*/sizeof(struct scsi_format_page) - 2,
 146         /*tracks_per_zone*/ {0, 0},
 147         /*alt_sectors_per_zone*/ {0, 0},
 148         /*alt_tracks_per_zone*/ {0, 0},
 149         /*alt_tracks_per_lun*/ {0, 0},
 150         /*sectors_per_track*/ {(CTL_DEFAULT_SECTORS_PER_TRACK >> 8) & 0xff,
 151                                 CTL_DEFAULT_SECTORS_PER_TRACK & 0xff},
 152         /*bytes_per_sector*/ {0, 0},
 153         /*interleave*/ {0, 0},
 154         /*track_skew*/ {0, 0},
 155         /*cylinder_skew*/ {0, 0},
 156         /*flags*/ SFP_HSEC,
 157         /*reserved*/ {0, 0, 0}
 158 };
 159
 160 const static struct scsi_format_page format_page_changeable = {
 161         /*page_code*/SMS_FORMAT_DEVICE_PAGE,
 162         /*page_length*/sizeof(struct scsi_format_page) - 2,
 163         /*tracks_per_zone*/ {0, 0},
 164         /*alt_sectors_per_zone*/ {0, 0},
 165         /*alt_tracks_per_zone*/ {0, 0},
 166         /*alt_tracks_per_lun*/ {0, 0},
 167         /*sectors_per_track*/ {0, 0},
 168         /*bytes_per_sector*/ {0, 0},
 169         /*interleave*/ {0, 0},
 170         /*track_skew*/ {0, 0},
 171         /*cylinder_skew*/ {0, 0},
 172         /*flags*/ 0,
 173         /*reserved*/ {0, 0, 0}
 174 };
 175
 176 const static struct scsi_rigid_disk_page rigid_disk_page_default = {
 177         /*page_code*/SMS_RIGID_DISK_PAGE,
 178         /*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
 179         /*cylinders*/ {0, 0, 0},
 180         /*heads*/ CTL_DEFAULT_HEADS,
 181         /*start_write_precomp*/ {0, 0, 0},
 182         /*start_reduced_current*/ {0, 0, 0},
 183         /*step_rate*/ {0, 0},
 184         /*landing_zone_cylinder*/ {0, 0, 0},
 185         /*rpl*/ SRDP_RPL_DISABLED,
 186         /*rotational_offset*/ 0,
 187         /*reserved1*/ 0,
 188         /*rotation_rate*/ {(CTL_DEFAULT_ROTATION_RATE >> 8) & 0xff,
 189                            CTL_DEFAULT_ROTATION_RATE & 0xff},
 190         /*reserved2*/ {0, 0}
 191 };
 192
 193 const static struct scsi_rigid_disk_page rigid_disk_page_changeable = {
 194         /*page_code*/SMS_RIGID_DISK_PAGE,
 195         /*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
 196         /*cylinders*/ {0, 0, 0},
 197         /*heads*/ 0,
 198         /*start_write_precomp*/ {0, 0, 0},
 199         /*start_reduced_current*/ {0, 0, 0},
 200         /*step_rate*/ {0, 0},
 201         /*landing_zone_cylinder*/ {0, 0, 0},
 202         /*rpl*/ 0,
 203         /*rotational_offset*/ 0,
 204         /*reserved1*/ 0,
 205         /*rotation_rate*/ {0, 0},
 206         /*reserved2*/ {0, 0}
 207 };
 208
 209 const static struct scsi_caching_page caching_page_default = {
 210         /*page_code*/SMS_CACHING_PAGE,
 211         /*page_length*/sizeof(struct scsi_caching_page) - 2,
 212         /*flags1*/ SCP_DISC | SCP_WCE,
 213         /*ret_priority*/ 0,
 214         /*disable_pf_transfer_len*/ {0xff, 0xff},
 215         /*min_prefetch*/ {0, 0},
 216         /*max_prefetch*/ {0xff, 0xff},
 217         /*max_pf_ceiling*/ {0xff, 0xff},
 218         /*flags2*/ 0,
 219         /*cache_segments*/ 0,
 220         /*cache_seg_size*/ {0, 0},
 221         /*reserved*/ 0,
 222         /*non_cache_seg_size*/ {0, 0, 0}
 223 };
 224
 225 const static struct scsi_caching_page caching_page_changeable = {
 226         /*page_code*/SMS_CACHING_PAGE,
 227         /*page_length*/sizeof(struct scsi_caching_page) - 2,
 228         /*flags1*/ SCP_WCE | SCP_RCD,
 229         /*ret_priority*/ 0,
 230         /*disable_pf_transfer_len*/ {0, 0},
 231         /*min_prefetch*/ {0, 0},
 232         /*max_prefetch*/ {0, 0},
 233         /*max_pf_ceiling*/ {0, 0},
 234         /*flags2*/ 0,
 235         /*cache_segments*/ 0,
 236         /*cache_seg_size*/ {0, 0},
 237         /*reserved*/ 0,
 238         /*non_cache_seg_size*/ {0, 0, 0}
 239 };
 240
 241 const static struct scsi_control_page control_page_default = {
 242         /*page_code*/SMS_CONTROL_MODE_PAGE,
 243         /*page_length*/sizeof(struct scsi_control_page) - 2,
 244         /*rlec*/0,
 245         /*queue_flags*/SCP_QUEUE_ALG_RESTRICTED,
 246         /*eca_and_aen*/0,
 247         /*flags4*/SCP_TAS,
 248         /*aen_holdoff_period*/{0, 0},
 249         /*busy_timeout_period*/{0, 0},
 250         /*extended_selftest_completion_time*/{0, 0}
 251 };
 252
 253 const static struct scsi_control_page control_page_changeable = {
 254         /*page_code*/SMS_CONTROL_MODE_PAGE,
 255         /*page_length*/sizeof(struct scsi_control_page) - 2,
 256         /*rlec*/SCP_DSENSE,
 257         /*queue_flags*/SCP_QUEUE_ALG_MASK,
 258         /*eca_and_aen*/SCP_SWP,
 259         /*flags4*/0,
 260         /*aen_holdoff_period*/{0, 0},
 261         /*busy_timeout_period*/{0, 0},
 262         /*extended_selftest_completion_time*/{0, 0}
 263 };
 264
 265 #define CTL_CEM_LEN     (sizeof(struct scsi_control_ext_page) - 4)
 266
 267 const static struct scsi_control_ext_page control_ext_page_default = {
 268         /*page_code*/SMS_CONTROL_MODE_PAGE | SMPH_SPF,
 269         /*subpage_code*/0x01,
 270         /*page_length*/{CTL_CEM_LEN >> 8, CTL_CEM_LEN},
 271         /*flags*/0,
 272         /*prio*/0,
 273         /*max_sense*/0
 274 };
 275
 276 const static struct scsi_control_ext_page control_ext_page_changeable = {
 277         /*page_code*/SMS_CONTROL_MODE_PAGE | SMPH_SPF,
 278         /*subpage_code*/0x01,
 279         /*page_length*/{CTL_CEM_LEN >> 8, CTL_CEM_LEN},
 280         /*flags*/0,
 281         /*prio*/0,
 282         /*max_sense*/0
 283 };
 284
 285 const static struct scsi_info_exceptions_page ie_page_default = {
 286         /*page_code*/SMS_INFO_EXCEPTIONS_PAGE,
 287         /*page_length*/sizeof(struct scsi_info_exceptions_page) - 2,
 288         /*info_flags*/SIEP_FLAGS_DEXCPT,
 289         /*mrie*/0,
 290         /*interval_timer*/{0, 0, 0, 0},
 291         /*report_count*/{0, 0, 0, 0}
 292 };
 293
 294 const static struct scsi_info_exceptions_page ie_page_changeable = {
 295         /*page_code*/SMS_INFO_EXCEPTIONS_PAGE,
 296         /*page_length*/sizeof(struct scsi_info_exceptions_page) - 2,
 297         /*info_flags*/0,
 298         /*mrie*/0,
 299         /*interval_timer*/{0, 0, 0, 0},
 300         /*report_count*/{0, 0, 0, 0}
 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*/0,
 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
 411 /*
 412  * Supported pages (0x00), Serial number (0x80), Device ID (0x83),
 413  * Extended INQUIRY Data (0x86), Mode Page Policy (0x87),
 414  * SCSI Ports (0x88), Third-party Copy (0x8F), Block limits (0xB0),
 415  * Block Device Characteristics (0xB1) and Logical Block Provisioning (0xB2)
 416  */
 417 #define SCSI_EVPD_NUM_SUPPORTED_PAGES   10
 418
 419 static void ctl_isc_event_handler(ctl_ha_channel chanel, ctl_ha_event event,
 420                                   int param);
 421 static void ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest);
 422 static void ctl_copy_sense_data_back(union ctl_io *src, union ctl_ha_msg *dest);
 423 static int ctl_init(void);
 424 void ctl_shutdown(void);
 425 static int ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td);
 426 static int ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td);
 427 static int ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio);
 428 static void ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
 429                               struct ctl_ooa *ooa_hdr,
 430                               struct ctl_ooa_entry *kern_entries);
 431 static int ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
 432                      struct thread *td);
 433 static int ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
 434                          struct ctl_be_lun *be_lun);
 435 static int ctl_free_lun(struct ctl_lun *lun);
 436 static void ctl_create_lun(struct ctl_be_lun *be_lun);
 437 static struct ctl_port * ctl_io_port(struct ctl_io_hdr *io_hdr);
 438
 439 static int ctl_do_mode_select(union ctl_io *io);
 440 static int ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun,
 441                            uint64_t res_key, uint64_t sa_res_key,
 442                            uint8_t type, uint32_t residx,
 443                            struct ctl_scsiio *ctsio,
 444                            struct scsi_per_res_out *cdb,
 445                            struct scsi_per_res_out_parms* param);
 446 static void ctl_pro_preempt_other(struct ctl_lun *lun,
 447                                   union ctl_ha_msg *msg);
 448 static void ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg);
 449 static int ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len);
 450 static int ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len);
 451 static int ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len);
 452 static int ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len);
 453 static int ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len);
 454 static int ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio,
 455                                          int alloc_len);
 456 static int ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio,
 457                                          int alloc_len);
 458 static int ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len);
 459 static int ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len);
 460 static int ctl_inquiry_evpd(struct ctl_scsiio *ctsio);
 461 static int ctl_inquiry_std(struct ctl_scsiio *ctsio);
 462 static int ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len);
 463 static ctl_action ctl_extent_check(union ctl_io *io1, union ctl_io *io2,
 464     bool seq);
 465 static ctl_action ctl_extent_check_seq(union ctl_io *io1, union ctl_io *io2);
 466 static ctl_action ctl_check_for_blockage(struct ctl_lun *lun,
 467     union ctl_io *pending_io, union ctl_io *ooa_io);
 468 static ctl_action ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
 469                                 union ctl_io *starting_io);
 470 static int ctl_check_blocked(struct ctl_lun *lun);
 471 static int ctl_scsiio_lun_check(struct ctl_lun *lun,
 472                                 const struct ctl_cmd_entry *entry,
 473                                 struct ctl_scsiio *ctsio);
 474 static void ctl_failover_lun(union ctl_io *io);
 475 static int ctl_scsiio_precheck(struct ctl_softc *ctl_softc,
 476                                struct ctl_scsiio *ctsio);
 477 static int ctl_scsiio(struct ctl_scsiio *ctsio);
 478
 479 static int ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io);
 480 static int ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
 481                             ctl_ua_type ua_type);
 482 static int ctl_do_lun_reset(struct ctl_lun *lun, union ctl_io *io,
 483                          ctl_ua_type ua_type);
 484 static int ctl_lun_reset(struct ctl_softc *ctl_softc, union ctl_io *io);
 485 static int ctl_abort_task(union ctl_io *io);
 486 static int ctl_abort_task_set(union ctl_io *io);
 487 static int ctl_query_task(union ctl_io *io, int task_set);
 488 static int ctl_i_t_nexus_reset(union ctl_io *io);
 489 static int ctl_query_async_event(union ctl_io *io);
 490 static void ctl_run_task(union ctl_io *io);
 491 #ifdef CTL_IO_DELAY
 492 static void ctl_datamove_timer_wakeup(void *arg);
 493 static void ctl_done_timer_wakeup(void *arg);
 494 #endif /* CTL_IO_DELAY */
 495
 496 static void ctl_send_datamove_done(union ctl_io *io, int have_lock);
 497 static void ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq);
 498 static int ctl_datamove_remote_dm_write_cb(union ctl_io *io);
 499 static void ctl_datamove_remote_write(union ctl_io *io);
 500 static int ctl_datamove_remote_dm_read_cb(union ctl_io *io);
 501 static void ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq);
 502 static int ctl_datamove_remote_sgl_setup(union ctl_io *io);
 503 static int ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
 504                                     ctl_ha_dt_cb callback);
 505 static void ctl_datamove_remote_read(union ctl_io *io);
 506 static void ctl_datamove_remote(union ctl_io *io);
 507 static void ctl_process_done(union ctl_io *io);
 508 static void ctl_lun_thread(void *arg);
 509 static void ctl_thresh_thread(void *arg);
 510 static void ctl_work_thread(void *arg);
 511 static void ctl_enqueue_incoming(union ctl_io *io);
 512 static void ctl_enqueue_rtr(union ctl_io *io);
 513 static void ctl_enqueue_done(union ctl_io *io);
 514 static void ctl_enqueue_isc(union ctl_io *io);
 515 static const struct ctl_cmd_entry *
 516     ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa);
 517 static const struct ctl_cmd_entry *
 518     ctl_validate_command(struct ctl_scsiio *ctsio);
 519 static int ctl_cmd_applicable(uint8_t lun_type,
 520     const struct ctl_cmd_entry *entry);
 521
 522 static uint64_t ctl_get_prkey(struct ctl_lun *lun, uint32_t residx);
 523 static void ctl_clr_prkey(struct ctl_lun *lun, uint32_t residx);
 524 static void ctl_alloc_prkey(struct ctl_lun *lun, uint32_t residx);
 525 static void ctl_set_prkey(struct ctl_lun *lun, uint32_t residx, uint64_t key);
 526
 527 /*
 528  * Load the serialization table.  This isn't very pretty, but is probably
 529  * the easiest way to do it.
 530  */
 531 #include "ctl_ser_table.c"
 532
 533 /*
 534  * We only need to define open, close and ioctl routines for this driver.
 535  */
 536 static struct cdevsw ctl_cdevsw = {
 537         .d_version =    D_VERSION,
 538         .d_flags =      0,
 539         .d_open =       ctl_open,
 540         .d_close =      ctl_close,
 541         .d_ioctl =      ctl_ioctl,
 542         .d_name =       "ctl",
 543 };
 544
 545
 546 MALLOC_DEFINE(M_CTL, "ctlmem", "Memory used for CTL");
 547
 548 static int ctl_module_event_handler(module_t, int /*modeventtype_t*/, void *);
 549
 550 static moduledata_t ctl_moduledata = {
 551         "ctl",
 552         ctl_module_event_handler,
 553         NULL
 554 };
 555
 556 DECLARE_MODULE(ctl, ctl_moduledata, SI_SUB_CONFIGURE, SI_ORDER_THIRD);
 557 MODULE_VERSION(ctl, 1);
 558
 559 static struct ctl_frontend ha_frontend =
 560 {
 561         .name = "ha",
 562 };
 563
 564 static void
 565 ctl_ha_datamove(union ctl_io *io)
 566 {
 567         struct ctl_lun *lun;
 568         struct ctl_sg_entry *sgl;
 569         union ctl_ha_msg msg;
 570         uint32_t sg_entries_sent;
 571         int do_sg_copy, i, j;
 572
 573         lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
 574         memset(&msg.dt, 0, sizeof(msg.dt));
 575         msg.hdr.msg_type = CTL_MSG_DATAMOVE;
 576         msg.hdr.original_sc = io->io_hdr.original_sc;
 577         msg.hdr.serializing_sc = io;
 578         msg.hdr.nexus = io->io_hdr.nexus;
 579         msg.hdr.status = io->io_hdr.status;
 580         msg.dt.flags = io->io_hdr.flags;
 581
 582         /*
 583          * We convert everything into a S/G list here.  We can't
 584          * pass by reference, only by value between controllers.
 585          * So we can't pass a pointer to the S/G list, only as many
 586          * S/G entries as we can fit in here.  If it's possible for
 587          * us to get more than CTL_HA_MAX_SG_ENTRIES S/G entries,
 588          * then we need to break this up into multiple transfers.
 589          */
 590         if (io->scsiio.kern_sg_entries == 0) {
 591                 msg.dt.kern_sg_entries = 1;
 592 #if 0
 593                 if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
 594                         msg.dt.sg_list[0].addr = io->scsiio.kern_data_ptr;
 595                 } else {
 596                         /* XXX KDM use busdma here! */
 597                         msg.dt.sg_list[0].addr =
 598                             (void *)vtophys(io->scsiio.kern_data_ptr);
 599                 }
 600 #else
 601                 KASSERT((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0,
 602                     ("HA does not support BUS_ADDR"));
 603                 msg.dt.sg_list[0].addr = io->scsiio.kern_data_ptr;
 604 #endif
 605                 msg.dt.sg_list[0].len = io->scsiio.kern_data_len;
 606                 do_sg_copy = 0;
 607         } else {
 608                 msg.dt.kern_sg_entries = io->scsiio.kern_sg_entries;
 609                 do_sg_copy = 1;
 610         }
 611
 612         msg.dt.kern_data_len = io->scsiio.kern_data_len;
 613         msg.dt.kern_total_len = io->scsiio.kern_total_len;
 614         msg.dt.kern_data_resid = io->scsiio.kern_data_resid;
 615         msg.dt.kern_rel_offset = io->scsiio.kern_rel_offset;
 616         msg.dt.sg_sequence = 0;
 617
 618         /*
 619          * Loop until we've sent all of the S/G entries.  On the
 620          * other end, we'll recompose these S/G entries into one
 621          * contiguous list before processing.
 622          */
 623         for (sg_entries_sent = 0; sg_entries_sent < msg.dt.kern_sg_entries;
 624             msg.dt.sg_sequence++) {
 625                 msg.dt.cur_sg_entries = MIN((sizeof(msg.dt.sg_list) /
 626                     sizeof(msg.dt.sg_list[0])),
 627                     msg.dt.kern_sg_entries - sg_entries_sent);
 628                 if (do_sg_copy != 0) {
 629                         sgl = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr;
 630                         for (i = sg_entries_sent, j = 0;
 631                              i < msg.dt.cur_sg_entries; i++, j++) {
 632 #if 0
 633                                 if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
 634                                         msg.dt.sg_list[j].addr = sgl[i].addr;
 635                                 } else {
 636                                         /* XXX KDM use busdma here! */
 637                                         msg.dt.sg_list[j].addr =
 638                                             (void *)vtophys(sgl[i].addr);
 639                                 }
 640 #else
 641                                 KASSERT((io->io_hdr.flags &
 642                                     CTL_FLAG_BUS_ADDR) == 0,
 643                                     ("HA does not support BUS_ADDR"));
 644                                 msg.dt.sg_list[j].addr = sgl[i].addr;
 645 #endif
 646                                 msg.dt.sg_list[j].len = sgl[i].len;
 647                         }
 648                 }
 649
 650                 sg_entries_sent += msg.dt.cur_sg_entries;
 651                 msg.dt.sg_last = (sg_entries_sent >= msg.dt.kern_sg_entries);
 652                 if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
 653                     sizeof(msg.dt) - sizeof(msg.dt.sg_list) +
 654                     sizeof(struct ctl_sg_entry) * msg.dt.cur_sg_entries,
 655                     M_WAITOK) > CTL_HA_STATUS_SUCCESS) {
 656                         io->io_hdr.port_status = 31341;
 657                         io->scsiio.be_move_done(io);
 658                         return;
 659                 }
 660                 msg.dt.sent_sg_entries = sg_entries_sent;
 661         }
 662
 663         /*
 664          * Officially handover the request from us to peer.
 665          * If failover has just happened, then we must return error.
 666          * If failover happen just after, then it is not our problem.
 667          */
 668         if (lun)
 669                 mtx_lock(&lun->lun_lock);
 670         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
 671                 if (lun)
 672                         mtx_unlock(&lun->lun_lock);
 673                 io->io_hdr.port_status = 31342;
 674                 io->scsiio.be_move_done(io);
 675                 return;
 676         }
 677         io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
 678         io->io_hdr.flags |= CTL_FLAG_DMA_INPROG;
 679         if (lun)
 680                 mtx_unlock(&lun->lun_lock);
 681 }
 682
 683 static void
 684 ctl_ha_done(union ctl_io *io)
 685 {
 686         union ctl_ha_msg msg;
 687
 688         if (io->io_hdr.io_type == CTL_IO_SCSI) {
 689                 memset(&msg, 0, sizeof(msg));
 690                 msg.hdr.msg_type = CTL_MSG_FINISH_IO;
 691                 msg.hdr.original_sc = io->io_hdr.original_sc;
 692                 msg.hdr.nexus = io->io_hdr.nexus;
 693                 msg.hdr.status = io->io_hdr.status;
 694                 msg.scsi.scsi_status = io->scsiio.scsi_status;
 695                 msg.scsi.tag_num = io->scsiio.tag_num;
 696                 msg.scsi.tag_type = io->scsiio.tag_type;
 697                 msg.scsi.sense_len = io->scsiio.sense_len;
 698                 msg.scsi.sense_residual = io->scsiio.sense_residual;
 699                 msg.scsi.residual = io->scsiio.residual;
 700                 memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
 701                     io->scsiio.sense_len);
 702                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
 703                     sizeof(msg.scsi) - sizeof(msg.scsi.sense_data) +
 704                     msg.scsi.sense_len, M_WAITOK);
 705         }
 706         ctl_free_io(io);
 707 }
 708
 709 static void
 710 ctl_isc_handler_finish_xfer(struct ctl_softc *ctl_softc,
 711                             union ctl_ha_msg *msg_info)
 712 {
 713         struct ctl_scsiio *ctsio;
 714
 715         if (msg_info->hdr.original_sc == NULL) {
 716                 printf("%s: original_sc == NULL!\n", __func__);
 717                 /* XXX KDM now what? */
 718                 return;
 719         }
 720
 721         ctsio = &msg_info->hdr.original_sc->scsiio;
 722         ctsio->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
 723         ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
 724         ctsio->io_hdr.status = msg_info->hdr.status;
 725         ctsio->scsi_status = msg_info->scsi.scsi_status;
 726         ctsio->sense_len = msg_info->scsi.sense_len;
 727         ctsio->sense_residual = msg_info->scsi.sense_residual;
 728         ctsio->residual = msg_info->scsi.residual;
 729         memcpy(&ctsio->sense_data, &msg_info->scsi.sense_data,
 730                msg_info->scsi.sense_len);
 731         ctl_enqueue_isc((union ctl_io *)ctsio);
 732 }
 733
 734 static void
 735 ctl_isc_handler_finish_ser_only(struct ctl_softc *ctl_softc,
 736                                 union ctl_ha_msg *msg_info)
 737 {
 738         struct ctl_scsiio *ctsio;
 739
 740         if (msg_info->hdr.serializing_sc == NULL) {
 741                 printf("%s: serializing_sc == NULL!\n", __func__);
 742                 /* XXX KDM now what? */
 743                 return;
 744         }
 745
 746         ctsio = &msg_info->hdr.serializing_sc->scsiio;
 747         ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
 748         ctl_enqueue_isc((union ctl_io *)ctsio);
 749 }
 750
 751 void
 752 ctl_isc_announce_lun(struct ctl_lun *lun)
 753 {
 754         struct ctl_softc *softc = lun->ctl_softc;
 755         union ctl_ha_msg *msg;
 756         struct ctl_ha_msg_lun_pr_key pr_key;
 757         int i, k;
 758
 759         if (softc->ha_link != CTL_HA_LINK_ONLINE)
 760                 return;
 761         mtx_lock(&lun->lun_lock);
 762         i = sizeof(msg->lun);
 763         if (lun->lun_devid)
 764                 i += lun->lun_devid->len;
 765         i += sizeof(pr_key) * lun->pr_key_count;
 766 alloc:
 767         mtx_unlock(&lun->lun_lock);
 768         msg = malloc(i, M_CTL, M_WAITOK);
 769         mtx_lock(&lun->lun_lock);
 770         k = sizeof(msg->lun);
 771         if (lun->lun_devid)
 772                 k += lun->lun_devid->len;
 773         k += sizeof(pr_key) * lun->pr_key_count;
 774         if (i < k) {
 775                 free(msg, M_CTL);
 776                 i = k;
 777                 goto alloc;
 778         }
 779         bzero(&msg->lun, sizeof(msg->lun));
 780         msg->hdr.msg_type = CTL_MSG_LUN_SYNC;
 781         msg->hdr.nexus.targ_lun = lun->lun;
 782         msg->hdr.nexus.targ_mapped_lun = lun->lun;
 783         msg->lun.flags = lun->flags;
 784         msg->lun.pr_generation = lun->pr_generation;
 785         msg->lun.pr_res_idx = lun->pr_res_idx;
 786         msg->lun.pr_res_type = lun->pr_res_type;
 787         msg->lun.pr_key_count = lun->pr_key_count;
 788         i = 0;
 789         if (lun->lun_devid) {
 790                 msg->lun.lun_devid_len = lun->lun_devid->len;
 791                 memcpy(&msg->lun.data[i], lun->lun_devid->data,
 792                     msg->lun.lun_devid_len);
 793                 i += msg->lun.lun_devid_len;
 794         }
 795         for (k = 0; k < CTL_MAX_INITIATORS; k++) {
 796                 if ((pr_key.pr_key = ctl_get_prkey(lun, k)) == 0)
 797                         continue;
 798                 pr_key.pr_iid = k;
 799                 memcpy(&msg->lun.data[i], &pr_key, sizeof(pr_key));
 800                 i += sizeof(pr_key);
 801         }
 802         mtx_unlock(&lun->lun_lock);
 803         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->port, sizeof(msg->port) + i,
 804             M_WAITOK);
 805         free(msg, M_CTL);
 806
 807         if (lun->flags & CTL_LUN_PRIMARY_SC) {
 808                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
 809                         ctl_isc_announce_mode(lun, -1,
 810                             lun->mode_pages.index[i].page_code & SMPH_PC_MASK,
 811                             lun->mode_pages.index[i].subpage);
 812                 }
 813         }
 814 }
 815
 816 void
 817 ctl_isc_announce_port(struct ctl_port *port)
 818 {
 819         struct ctl_softc *softc = port->ctl_softc;
 820         union ctl_ha_msg *msg;
 821         int i;
 822
 823         if (port->targ_port < softc->port_min ||
 824             port->targ_port >= softc->port_max ||
 825             softc->ha_link != CTL_HA_LINK_ONLINE)
 826                 return;
 827         i = sizeof(msg->port) + strlen(port->port_name) + 1;
 828         if (port->lun_map)
 829                 i += sizeof(uint32_t) * CTL_MAX_LUNS;
 830         if (port->port_devid)
 831                 i += port->port_devid->len;
 832         if (port->target_devid)
 833                 i += port->target_devid->len;
 834         if (port->init_devid)
 835                 i += port->init_devid->len;
 836         msg = malloc(i, M_CTL, M_WAITOK);
 837         bzero(&msg->port, sizeof(msg->port));
 838         msg->hdr.msg_type = CTL_MSG_PORT_SYNC;
 839         msg->hdr.nexus.targ_port = port->targ_port;
 840         msg->port.port_type = port->port_type;
 841         msg->port.physical_port = port->physical_port;
 842         msg->port.virtual_port = port->virtual_port;
 843         msg->port.status = port->status;
 844         i = 0;
 845         msg->port.name_len = sprintf(&msg->port.data[i],
 846             "%d:%s", softc->ha_id, port->port_name) + 1;
 847         i += msg->port.name_len;
 848         if (port->lun_map) {
 849                 msg->port.lun_map_len = sizeof(uint32_t) * CTL_MAX_LUNS;
 850                 memcpy(&msg->port.data[i], port->lun_map,
 851                     msg->port.lun_map_len);
 852                 i += msg->port.lun_map_len;
 853         }
 854         if (port->port_devid) {
 855                 msg->port.port_devid_len = port->port_devid->len;
 856                 memcpy(&msg->port.data[i], port->port_devid->data,
 857                     msg->port.port_devid_len);
 858                 i += msg->port.port_devid_len;
 859         }
 860         if (port->target_devid) {
 861                 msg->port.target_devid_len = port->target_devid->len;
 862                 memcpy(&msg->port.data[i], port->target_devid->data,
 863                     msg->port.target_devid_len);
 864                 i += msg->port.target_devid_len;
 865         }
 866         if (port->init_devid) {
 867                 msg->port.init_devid_len = port->init_devid->len;
 868                 memcpy(&msg->port.data[i], port->init_devid->data,
 869                     msg->port.init_devid_len);
 870                 i += msg->port.init_devid_len;
 871         }
 872         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->port, sizeof(msg->port) + i,
 873             M_WAITOK);
 874         free(msg, M_CTL);
 875 }
 876
 877 void
 878 ctl_isc_announce_iid(struct ctl_port *port, int iid)
 879 {
 880         struct ctl_softc *softc = port->ctl_softc;
 881         union ctl_ha_msg *msg;
 882         int i, l;
 883
 884         if (port->targ_port < softc->port_min ||
 885             port->targ_port >= softc->port_max ||
 886             softc->ha_link != CTL_HA_LINK_ONLINE)
 887                 return;
 888         mtx_lock(&softc->ctl_lock);
 889         i = sizeof(msg->iid);
 890         l = 0;
 891         if (port->wwpn_iid[iid].name)
 892                 l = strlen(port->wwpn_iid[iid].name) + 1;
 893         i += l;
 894         msg = malloc(i, M_CTL, M_NOWAIT);
 895         if (msg == NULL) {
 896                 mtx_unlock(&softc->ctl_lock);
 897                 return;
 898         }
 899         bzero(&msg->iid, sizeof(msg->iid));
 900         msg->hdr.msg_type = CTL_MSG_IID_SYNC;
 901         msg->hdr.nexus.targ_port = port->targ_port;
 902         msg->hdr.nexus.initid = iid;
 903         msg->iid.in_use = port->wwpn_iid[iid].in_use;
 904         msg->iid.name_len = l;
 905         msg->iid.wwpn = port->wwpn_iid[iid].wwpn;
 906         if (port->wwpn_iid[iid].name)
 907                 strlcpy(msg->iid.data, port->wwpn_iid[iid].name, l);
 908         mtx_unlock(&softc->ctl_lock);
 909         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->iid, i, M_NOWAIT);
 910         free(msg, M_CTL);
 911 }
 912
 913 void
 914 ctl_isc_announce_mode(struct ctl_lun *lun, uint32_t initidx,
 915     uint8_t page, uint8_t subpage)
 916 {
 917         struct ctl_softc *softc = lun->ctl_softc;
 918         union ctl_ha_msg msg;
 919         u_int i;
 920
 921         if (softc->ha_link != CTL_HA_LINK_ONLINE)
 922                 return;
 923         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
 924                 if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) ==
 925                     page && lun->mode_pages.index[i].subpage == subpage)
 926                         break;
 927         }
 928         if (i == CTL_NUM_MODE_PAGES)
 929                 return;
 930
 931         /* Don't try to replicate pages not present on this device. */
 932         if (lun->mode_pages.index[i].page_data == NULL)
 933                 return;
 934
 935         bzero(&msg.mode, sizeof(msg.mode));
 936         msg.hdr.msg_type = CTL_MSG_MODE_SYNC;
 937         msg.hdr.nexus.targ_port = initidx / CTL_MAX_INIT_PER_PORT;
 938         msg.hdr.nexus.initid = initidx % CTL_MAX_INIT_PER_PORT;
 939         msg.hdr.nexus.targ_lun = lun->lun;
 940         msg.hdr.nexus.targ_mapped_lun = lun->lun;
 941         msg.mode.page_code = page;
 942         msg.mode.subpage = subpage;
 943         msg.mode.page_len = lun->mode_pages.index[i].page_len;
 944         memcpy(msg.mode.data, lun->mode_pages.index[i].page_data,
 945             msg.mode.page_len);
 946         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg.mode, sizeof(msg.mode),
 947             M_WAITOK);
 948 }
 949
 950 static void
 951 ctl_isc_ha_link_up(struct ctl_softc *softc)
 952 {
 953         struct ctl_port *port;
 954         struct ctl_lun *lun;
 955         union ctl_ha_msg msg;
 956         int i;
 957
 958         /* Announce this node parameters to peer for validation. */
 959         msg.login.msg_type = CTL_MSG_LOGIN;
 960         msg.login.version = CTL_HA_VERSION;
 961         msg.login.ha_mode = softc->ha_mode;
 962         msg.login.ha_id = softc->ha_id;
 963         msg.login.max_luns = CTL_MAX_LUNS;
 964         msg.login.max_ports = CTL_MAX_PORTS;
 965         msg.login.max_init_per_port = CTL_MAX_INIT_PER_PORT;
 966         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg.login, sizeof(msg.login),
 967             M_WAITOK);
 968
 969         STAILQ_FOREACH(port, &softc->port_list, links) {
 970                 ctl_isc_announce_port(port);
 971                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
 972                         if (port->wwpn_iid[i].in_use)
 973                                 ctl_isc_announce_iid(port, i);
 974                 }
 975         }
 976         STAILQ_FOREACH(lun, &softc->lun_list, links)
 977                 ctl_isc_announce_lun(lun);
 978 }
 979
 980 static void
 981 ctl_isc_ha_link_down(struct ctl_softc *softc)
 982 {
 983         struct ctl_port *port;
 984         struct ctl_lun *lun;
 985         union ctl_io *io;
 986         int i;
 987
 988         mtx_lock(&softc->ctl_lock);
 989         STAILQ_FOREACH(lun, &softc->lun_list, links) {
 990                 mtx_lock(&lun->lun_lock);
 991                 if (lun->flags & CTL_LUN_PEER_SC_PRIMARY) {
 992                         lun->flags &= ~CTL_LUN_PEER_SC_PRIMARY;
 993                         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
 994                 }
 995                 mtx_unlock(&lun->lun_lock);
 996
 997                 mtx_unlock(&softc->ctl_lock);
 998                 io = ctl_alloc_io(softc->othersc_pool);
 999                 mtx_lock(&softc->ctl_lock);
1000                 ctl_zero_io(io);
1001                 io->io_hdr.msg_type = CTL_MSG_FAILOVER;
1002                 io->io_hdr.nexus.targ_mapped_lun = lun->lun;
1003                 ctl_enqueue_isc(io);
1004         }
1005
1006         STAILQ_FOREACH(port, &softc->port_list, links) {
1007                 if (port->targ_port >= softc->port_min &&
1008                     port->targ_port < softc->port_max)
1009                         continue;
1010                 port->status &= ~CTL_PORT_STATUS_ONLINE;
1011                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1012                         port->wwpn_iid[i].in_use = 0;
1013                         free(port->wwpn_iid[i].name, M_CTL);
1014                         port->wwpn_iid[i].name = NULL;
1015                 }
1016         }
1017         mtx_unlock(&softc->ctl_lock);
1018 }
1019
1020 static void
1021 ctl_isc_ua(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1022 {
1023         struct ctl_lun *lun;
1024         uint32_t iid = ctl_get_initindex(&msg->hdr.nexus);
1025
1026         mtx_lock(&softc->ctl_lock);
1027         if (msg->hdr.nexus.targ_lun < CTL_MAX_LUNS &&
1028             (lun = softc->ctl_luns[msg->hdr.nexus.targ_mapped_lun]) != NULL) {
1029                 mtx_lock(&lun->lun_lock);
1030                 mtx_unlock(&softc->ctl_lock);
1031                 if (msg->ua.ua_type == CTL_UA_THIN_PROV_THRES &&
1032                     msg->ua.ua_set)
1033                         memcpy(lun->ua_tpt_info, msg->ua.ua_info, 8);
1034                 if (msg->ua.ua_all) {
1035                         if (msg->ua.ua_set)
1036                                 ctl_est_ua_all(lun, iid, msg->ua.ua_type);
1037                         else
1038                                 ctl_clr_ua_all(lun, iid, msg->ua.ua_type);
1039                 } else {
1040                         if (msg->ua.ua_set)
1041                                 ctl_est_ua(lun, iid, msg->ua.ua_type);
1042                         else
1043                                 ctl_clr_ua(lun, iid, msg->ua.ua_type);
1044                 }
1045                 mtx_unlock(&lun->lun_lock);
1046         } else
1047                 mtx_unlock(&softc->ctl_lock);
1048 }
1049
1050 static void
1051 ctl_isc_lun_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1052 {
1053         struct ctl_lun *lun;
1054         struct ctl_ha_msg_lun_pr_key pr_key;
1055         int i, k;
1056         ctl_lun_flags oflags;
1057         uint32_t targ_lun;
1058
1059         targ_lun = msg->hdr.nexus.targ_mapped_lun;
1060         mtx_lock(&softc->ctl_lock);
1061         if ((targ_lun >= CTL_MAX_LUNS) ||
1062             ((lun = softc->ctl_luns[targ_lun]) == NULL)) {
1063                 mtx_unlock(&softc->ctl_lock);
1064                 return;
1065         }
1066         mtx_lock(&lun->lun_lock);
1067         mtx_unlock(&softc->ctl_lock);
1068         if (lun->flags & CTL_LUN_DISABLED) {
1069                 mtx_unlock(&lun->lun_lock);
1070                 return;
1071         }
1072         i = (lun->lun_devid != NULL) ? lun->lun_devid->len : 0;
1073         if (msg->lun.lun_devid_len != i || (i > 0 &&
1074             memcmp(&msg->lun.data[0], lun->lun_devid->data, i) != 0)) {
1075                 mtx_unlock(&lun->lun_lock);
1076                 printf("%s: Received conflicting HA LUN %d\n",
1077                     __func__, msg->hdr.nexus.targ_lun);
1078                 return;
1079         } else {
1080                 /* Record whether peer is primary. */
1081                 oflags = lun->flags;
1082                 if ((msg->lun.flags & CTL_LUN_PRIMARY_SC) &&
1083                     (msg->lun.flags & CTL_LUN_DISABLED) == 0)
1084                         lun->flags |= CTL_LUN_PEER_SC_PRIMARY;
1085                 else
1086                         lun->flags &= ~CTL_LUN_PEER_SC_PRIMARY;
1087                 if (oflags != lun->flags)
1088                         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
1089
1090                 /* If peer is primary and we are not -- use data */
1091                 if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
1092                     (lun->flags & CTL_LUN_PEER_SC_PRIMARY)) {
1093                         lun->pr_generation = msg->lun.pr_generation;
1094                         lun->pr_res_idx = msg->lun.pr_res_idx;
1095                         lun->pr_res_type = msg->lun.pr_res_type;
1096                         lun->pr_key_count = msg->lun.pr_key_count;
1097                         for (k = 0; k < CTL_MAX_INITIATORS; k++)
1098                                 ctl_clr_prkey(lun, k);
1099                         for (k = 0; k < msg->lun.pr_key_count; k++) {
1100                                 memcpy(&pr_key, &msg->lun.data[i],
1101                                     sizeof(pr_key));
1102                                 ctl_alloc_prkey(lun, pr_key.pr_iid);
1103                                 ctl_set_prkey(lun, pr_key.pr_iid,
1104                                     pr_key.pr_key);
1105                                 i += sizeof(pr_key);
1106                         }
1107                 }
1108
1109                 mtx_unlock(&lun->lun_lock);
1110                 CTL_DEBUG_PRINT(("%s: Known LUN %d, peer is %s\n",
1111                     __func__, msg->hdr.nexus.targ_lun,
1112                     (msg->lun.flags & CTL_LUN_PRIMARY_SC) ?
1113                     "primary" : "secondary"));
1114
1115                 /* If we are primary but peer doesn't know -- notify */
1116                 if ((lun->flags & CTL_LUN_PRIMARY_SC) &&
1117                     (msg->lun.flags & CTL_LUN_PEER_SC_PRIMARY) == 0)
1118                         ctl_isc_announce_lun(lun);
1119         }
1120 }
1121
1122 static void
1123 ctl_isc_port_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1124 {
1125         struct ctl_port *port;
1126         struct ctl_lun *lun;
1127         int i, new;
1128
1129         port = softc->ctl_ports[msg->hdr.nexus.targ_port];
1130         if (port == NULL) {
1131                 CTL_DEBUG_PRINT(("%s: New port %d\n", __func__,
1132                     msg->hdr.nexus.targ_port));
1133                 new = 1;
1134                 port = malloc(sizeof(*port), M_CTL, M_WAITOK | M_ZERO);
1135                 port->frontend = &ha_frontend;
1136                 port->targ_port = msg->hdr.nexus.targ_port;
1137                 port->fe_datamove = ctl_ha_datamove;
1138                 port->fe_done = ctl_ha_done;
1139         } else if (port->frontend == &ha_frontend) {
1140                 CTL_DEBUG_PRINT(("%s: Updated port %d\n", __func__,
1141                     msg->hdr.nexus.targ_port));
1142                 new = 0;
1143         } else {
1144                 printf("%s: Received conflicting HA port %d\n",
1145                     __func__, msg->hdr.nexus.targ_port);
1146                 return;
1147         }
1148         port->port_type = msg->port.port_type;
1149         port->physical_port = msg->port.physical_port;
1150         port->virtual_port = msg->port.virtual_port;
1151         port->status = msg->port.status;
1152         i = 0;
1153         free(port->port_name, M_CTL);
1154         port->port_name = strndup(&msg->port.data[i], msg->port.name_len,
1155             M_CTL);
1156         i += msg->port.name_len;
1157         if (msg->port.lun_map_len != 0) {
1158                 if (port->lun_map == NULL)
1159                         port->lun_map = malloc(sizeof(uint32_t) * CTL_MAX_LUNS,
1160                             M_CTL, M_WAITOK);
1161                 memcpy(port->lun_map, &msg->port.data[i],
1162                     sizeof(uint32_t) * CTL_MAX_LUNS);
1163                 i += msg->port.lun_map_len;
1164         } else {
1165                 free(port->lun_map, M_CTL);
1166                 port->lun_map = NULL;
1167         }
1168         if (msg->port.port_devid_len != 0) {
1169                 if (port->port_devid == NULL ||
1170                     port->port_devid->len != msg->port.port_devid_len) {
1171                         free(port->port_devid, M_CTL);
1172                         port->port_devid = malloc(sizeof(struct ctl_devid) +
1173                             msg->port.port_devid_len, M_CTL, M_WAITOK);
1174                 }
1175                 memcpy(port->port_devid->data, &msg->port.data[i],
1176                     msg->port.port_devid_len);
1177                 port->port_devid->len = msg->port.port_devid_len;
1178                 i += msg->port.port_devid_len;
1179         } else {
1180                 free(port->port_devid, M_CTL);
1181                 port->port_devid = NULL;
1182         }
1183         if (msg->port.target_devid_len != 0) {
1184                 if (port->target_devid == NULL ||
1185                     port->target_devid->len != msg->port.target_devid_len) {
1186                         free(port->target_devid, M_CTL);
1187                         port->target_devid = malloc(sizeof(struct ctl_devid) +
1188                             msg->port.target_devid_len, M_CTL, M_WAITOK);
1189                 }
1190                 memcpy(port->target_devid->data, &msg->port.data[i],
1191                     msg->port.target_devid_len);
1192                 port->target_devid->len = msg->port.target_devid_len;
1193                 i += msg->port.target_devid_len;
1194         } else {
1195                 free(port->target_devid, M_CTL);
1196                 port->target_devid = NULL;
1197         }
1198         if (msg->port.init_devid_len != 0) {
1199                 if (port->init_devid == NULL ||
1200                     port->init_devid->len != msg->port.init_devid_len) {
1201                         free(port->init_devid, M_CTL);
1202                         port->init_devid = malloc(sizeof(struct ctl_devid) +
1203                             msg->port.init_devid_len, M_CTL, M_WAITOK);
1204                 }
1205                 memcpy(port->init_devid->data, &msg->port.data[i],
1206                     msg->port.init_devid_len);
1207                 port->init_devid->len = msg->port.init_devid_len;
1208                 i += msg->port.init_devid_len;
1209         } else {
1210                 free(port->init_devid, M_CTL);
1211                 port->init_devid = NULL;
1212         }
1213         if (new) {
1214                 if (ctl_port_register(port) != 0) {
1215                         printf("%s: ctl_port_register() failed with error\n",
1216                             __func__);
1217                 }
1218         }
1219         mtx_lock(&softc->ctl_lock);
1220         STAILQ_FOREACH(lun, &softc->lun_list, links) {
1221                 if (ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
1222                         continue;
1223                 mtx_lock(&lun->lun_lock);
1224                 ctl_est_ua_all(lun, -1, CTL_UA_INQ_CHANGE);
1225                 mtx_unlock(&lun->lun_lock);
1226         }
1227         mtx_unlock(&softc->ctl_lock);
1228 }
1229
1230 static void
1231 ctl_isc_iid_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1232 {
1233         struct ctl_port *port;
1234         int iid;
1235
1236         port = softc->ctl_ports[msg->hdr.nexus.targ_port];
1237         if (port == NULL) {
1238                 printf("%s: Received IID for unknown port %d\n",
1239                     __func__, msg->hdr.nexus.targ_port);
1240                 return;
1241         }
1242         iid = msg->hdr.nexus.initid;
1243         port->wwpn_iid[iid].in_use = msg->iid.in_use;
1244         port->wwpn_iid[iid].wwpn = msg->iid.wwpn;
1245         free(port->wwpn_iid[iid].name, M_CTL);
1246         if (msg->iid.name_len) {
1247                 port->wwpn_iid[iid].name = strndup(&msg->iid.data[0],
1248                     msg->iid.name_len, M_CTL);
1249         } else
1250                 port->wwpn_iid[iid].name = NULL;
1251 }
1252
1253 static void
1254 ctl_isc_login(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1255 {
1256
1257         if (msg->login.version != CTL_HA_VERSION) {
1258                 printf("CTL HA peers have different versions %d != %d\n",
1259                     msg->login.version, CTL_HA_VERSION);
1260                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1261                 return;
1262         }
1263         if (msg->login.ha_mode != softc->ha_mode) {
1264                 printf("CTL HA peers have different ha_mode %d != %d\n",
1265                     msg->login.ha_mode, softc->ha_mode);
1266                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1267                 return;
1268         }
1269         if (msg->login.ha_id == softc->ha_id) {
1270                 printf("CTL HA peers have same ha_id %d\n", msg->login.ha_id);
1271                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1272                 return;
1273         }
1274         if (msg->login.max_luns != CTL_MAX_LUNS ||
1275             msg->login.max_ports != CTL_MAX_PORTS ||
1276             msg->login.max_init_per_port != CTL_MAX_INIT_PER_PORT) {
1277                 printf("CTL HA peers have different limits\n");
1278                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1279                 return;
1280         }
1281 }
1282
1283 static void
1284 ctl_isc_mode_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1285 {
1286         struct ctl_lun *lun;
1287         u_int i;
1288         uint32_t initidx, targ_lun;
1289
1290         targ_lun = msg->hdr.nexus.targ_mapped_lun;
1291         mtx_lock(&softc->ctl_lock);
1292         if ((targ_lun >= CTL_MAX_LUNS) ||
1293             ((lun = softc->ctl_luns[targ_lun]) == NULL)) {
1294                 mtx_unlock(&softc->ctl_lock);
1295                 return;
1296         }
1297         mtx_lock(&lun->lun_lock);
1298         mtx_unlock(&softc->ctl_lock);
1299         if (lun->flags & CTL_LUN_DISABLED) {
1300                 mtx_unlock(&lun->lun_lock);
1301                 return;
1302         }
1303         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
1304                 if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) ==
1305                     msg->mode.page_code &&
1306                     lun->mode_pages.index[i].subpage == msg->mode.subpage)
1307                         break;
1308         }
1309         if (i == CTL_NUM_MODE_PAGES) {
1310                 mtx_unlock(&lun->lun_lock);
1311                 return;
1312         }
1313         memcpy(lun->mode_pages.index[i].page_data, msg->mode.data,
1314             lun->mode_pages.index[i].page_len);
1315         initidx = ctl_get_initindex(&msg->hdr.nexus);
1316         if (initidx != -1)
1317                 ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
1318         mtx_unlock(&lun->lun_lock);
1319 }
1320
1321 /*
1322  * ISC (Inter Shelf Communication) event handler.  Events from the HA
1323  * subsystem come in here.
1324  */
1325 static void
1326 ctl_isc_event_handler(ctl_ha_channel channel, ctl_ha_event event, int param)
1327 {
1328         struct ctl_softc *softc = control_softc;
1329         union ctl_io *io;
1330         struct ctl_prio *presio;
1331         ctl_ha_status isc_status;
1332
1333         CTL_DEBUG_PRINT(("CTL: Isc Msg event %d\n", event));
1334         if (event == CTL_HA_EVT_MSG_RECV) {
1335                 union ctl_ha_msg *msg, msgbuf;
1336
1337                 if (param > sizeof(msgbuf))
1338                         msg = malloc(param, M_CTL, M_WAITOK);
1339                 else
1340                         msg = &msgbuf;
1341                 isc_status = ctl_ha_msg_recv(CTL_HA_CHAN_CTL, msg, param,
1342                     M_WAITOK);
1343                 if (isc_status != CTL_HA_STATUS_SUCCESS) {
1344                         printf("%s: Error receiving message: %d\n",
1345                             __func__, isc_status);
1346                         if (msg != &msgbuf)
1347                                 free(msg, M_CTL);
1348                         return;
1349                 }
1350
1351                 CTL_DEBUG_PRINT(("CTL: msg_type %d\n", msg->msg_type));
1352                 switch (msg->hdr.msg_type) {
1353                 case CTL_MSG_SERIALIZE:
1354                         io = ctl_alloc_io(softc->othersc_pool);
1355                         ctl_zero_io(io);
1356                         // populate ctsio from msg
1357                         io->io_hdr.io_type = CTL_IO_SCSI;
1358                         io->io_hdr.msg_type = CTL_MSG_SERIALIZE;
1359                         io->io_hdr.original_sc = msg->hdr.original_sc;
1360                         io->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC |
1361                                             CTL_FLAG_IO_ACTIVE;
1362                         /*
1363                          * If we're in serialization-only mode, we don't
1364                          * want to go through full done processing.  Thus
1365                          * the COPY flag.
1366                          *
1367                          * XXX KDM add another flag that is more specific.
1368                          */
1369                         if (softc->ha_mode != CTL_HA_MODE_XFER)
1370                                 io->io_hdr.flags |= CTL_FLAG_INT_COPY;
1371                         io->io_hdr.nexus = msg->hdr.nexus;
1372 #if 0
1373                         printf("port %u, iid %u, lun %u\n",
1374                                io->io_hdr.nexus.targ_port,
1375                                io->io_hdr.nexus.initid,
1376                                io->io_hdr.nexus.targ_lun);
1377 #endif
1378                         io->scsiio.tag_num = msg->scsi.tag_num;
1379                         io->scsiio.tag_type = msg->scsi.tag_type;
1380 #ifdef CTL_TIME_IO
1381                         io->io_hdr.start_time = time_uptime;
1382                         getbinuptime(&io->io_hdr.start_bt);
1383 #endif /* CTL_TIME_IO */
1384                         io->scsiio.cdb_len = msg->scsi.cdb_len;
1385                         memcpy(io->scsiio.cdb, msg->scsi.cdb,
1386                                CTL_MAX_CDBLEN);
1387                         if (softc->ha_mode == CTL_HA_MODE_XFER) {
1388                                 const struct ctl_cmd_entry *entry;
1389
1390                                 entry = ctl_get_cmd_entry(&io->scsiio, NULL);
1391                                 io->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
1392                                 io->io_hdr.flags |=
1393                                         entry->flags & CTL_FLAG_DATA_MASK;
1394                         }
1395                         ctl_enqueue_isc(io);
1396                         break;
1397
1398                 /* Performed on the Originating SC, XFER mode only */
1399                 case CTL_MSG_DATAMOVE: {
1400                         struct ctl_sg_entry *sgl;
1401                         int i, j;
1402
1403                         io = msg->hdr.original_sc;
1404                         if (io == NULL) {
1405                                 printf("%s: original_sc == NULL!\n", __func__);
1406                                 /* XXX KDM do something here */
1407                                 break;
1408                         }
1409                         io->io_hdr.msg_type = CTL_MSG_DATAMOVE;
1410                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1411                         /*
1412                          * Keep track of this, we need to send it back over
1413                          * when the datamove is complete.
1414                          */
1415                         io->io_hdr.serializing_sc = msg->hdr.serializing_sc;
1416                         if (msg->hdr.status == CTL_SUCCESS)
1417                                 io->io_hdr.status = msg->hdr.status;
1418
1419                         if (msg->dt.sg_sequence == 0) {
1420 #ifdef CTL_TIME_IO
1421                                 getbinuptime(&io->io_hdr.dma_start_bt);
1422 #endif
1423                                 i = msg->dt.kern_sg_entries +
1424                                     msg->dt.kern_data_len /
1425                                     CTL_HA_DATAMOVE_SEGMENT + 1;
1426                                 sgl = malloc(sizeof(*sgl) * i, M_CTL,
1427                                     M_WAITOK | M_ZERO);
1428                                 io->io_hdr.remote_sglist = sgl;
1429                                 io->io_hdr.local_sglist =
1430                                     &sgl[msg->dt.kern_sg_entries];
1431
1432                                 io->scsiio.kern_data_ptr = (uint8_t *)sgl;
1433
1434                                 io->scsiio.kern_sg_entries =
1435                                         msg->dt.kern_sg_entries;
1436                                 io->scsiio.rem_sg_entries =
1437                                         msg->dt.kern_sg_entries;
1438                                 io->scsiio.kern_data_len =
1439                                         msg->dt.kern_data_len;
1440                                 io->scsiio.kern_total_len =
1441                                         msg->dt.kern_total_len;
1442                                 io->scsiio.kern_data_resid =
1443                                         msg->dt.kern_data_resid;
1444                                 io->scsiio.kern_rel_offset =
1445                                         msg->dt.kern_rel_offset;
1446                                 io->io_hdr.flags &= ~CTL_FLAG_BUS_ADDR;
1447                                 io->io_hdr.flags |= msg->dt.flags &
1448                                     CTL_FLAG_BUS_ADDR;
1449                         } else
1450                                 sgl = (struct ctl_sg_entry *)
1451                                         io->scsiio.kern_data_ptr;
1452
1453                         for (i = msg->dt.sent_sg_entries, j = 0;
1454                              i < (msg->dt.sent_sg_entries +
1455                              msg->dt.cur_sg_entries); i++, j++) {
1456                                 sgl[i].addr = msg->dt.sg_list[j].addr;
1457                                 sgl[i].len = msg->dt.sg_list[j].len;
1458
1459 #if 0
1460                                 printf("%s: DATAMOVE: %p,%lu j=%d, i=%d\n",
1461                                     __func__, sgl[i].addr, sgl[i].len, j, i);
1462 #endif
1463                         }
1464
1465                         /*
1466                          * If this is the last piece of the I/O, we've got
1467                          * the full S/G list.  Queue processing in the thread.
1468                          * Otherwise wait for the next piece.
1469                          */
1470                         if (msg->dt.sg_last != 0)
1471                                 ctl_enqueue_isc(io);
1472                         break;
1473                 }
1474                 /* Performed on the Serializing (primary) SC, XFER mode only */
1475                 case CTL_MSG_DATAMOVE_DONE: {
1476                         if (msg->hdr.serializing_sc == NULL) {
1477                                 printf("%s: serializing_sc == NULL!\n",
1478                                        __func__);
1479                                 /* XXX KDM now what? */
1480                                 break;
1481                         }
1482                         /*
1483                          * We grab the sense information here in case
1484                          * there was a failure, so we can return status
1485                          * back to the initiator.
1486                          */
1487                         io = msg->hdr.serializing_sc;
1488                         io->io_hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
1489                         io->io_hdr.flags &= ~CTL_FLAG_DMA_INPROG;
1490                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1491                         io->io_hdr.port_status = msg->scsi.fetd_status;
1492                         io->scsiio.residual = msg->scsi.residual;
1493                         if (msg->hdr.status != CTL_STATUS_NONE) {
1494                                 io->io_hdr.status = msg->hdr.status;
1495                                 io->scsiio.scsi_status = msg->scsi.scsi_status;
1496                                 io->scsiio.sense_len = msg->scsi.sense_len;
1497                                 io->scsiio.sense_residual =msg->scsi.sense_residual;
1498                                 memcpy(&io->scsiio.sense_data,
1499                                     &msg->scsi.sense_data,
1500                                     msg->scsi.sense_len);
1501                                 if (msg->hdr.status == CTL_SUCCESS)
1502                                         io->io_hdr.flags |= CTL_FLAG_STATUS_SENT;
1503                         }
1504                         ctl_enqueue_isc(io);
1505                         break;
1506                 }
1507
1508                 /* Preformed on Originating SC, SER_ONLY mode */
1509                 case CTL_MSG_R2R:
1510                         io = msg->hdr.original_sc;
1511                         if (io == NULL) {
1512                                 printf("%s: original_sc == NULL!\n",
1513                                     __func__);
1514                                 break;
1515                         }
1516                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1517                         io->io_hdr.msg_type = CTL_MSG_R2R;
1518                         io->io_hdr.serializing_sc = msg->hdr.serializing_sc;
1519                         ctl_enqueue_isc(io);
1520                         break;
1521
1522                 /*
1523                  * Performed on Serializing(i.e. primary SC) SC in SER_ONLY
1524                  * mode.
1525                  * Performed on the Originating (i.e. secondary) SC in XFER
1526                  * mode
1527                  */
1528                 case CTL_MSG_FINISH_IO:
1529                         if (softc->ha_mode == CTL_HA_MODE_XFER)
1530                                 ctl_isc_handler_finish_xfer(softc, msg);
1531                         else
1532                                 ctl_isc_handler_finish_ser_only(softc, msg);
1533                         break;
1534
1535                 /* Preformed on Originating SC */
1536                 case CTL_MSG_BAD_JUJU:
1537                         io = msg->hdr.original_sc;
1538                         if (io == NULL) {
1539                                 printf("%s: Bad JUJU!, original_sc is NULL!\n",
1540                                        __func__);
1541                                 break;
1542                         }
1543                         ctl_copy_sense_data(msg, io);
1544                         /*
1545                          * IO should have already been cleaned up on other
1546                          * SC so clear this flag so we won't send a message
1547                          * back to finish the IO there.
1548                          */
1549                         io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
1550                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1551
1552                         /* io = msg->hdr.serializing_sc; */
1553                         io->io_hdr.msg_type = CTL_MSG_BAD_JUJU;
1554                         ctl_enqueue_isc(io);
1555                         break;
1556
1557                 /* Handle resets sent from the other side */
1558                 case CTL_MSG_MANAGE_TASKS: {
1559                         struct ctl_taskio *taskio;
1560                         taskio = (struct ctl_taskio *)ctl_alloc_io(
1561                             softc->othersc_pool);
1562                         ctl_zero_io((union ctl_io *)taskio);
1563                         taskio->io_hdr.io_type = CTL_IO_TASK;
1564                         taskio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
1565                         taskio->io_hdr.nexus = msg->hdr.nexus;
1566                         taskio->task_action = msg->task.task_action;
1567                         taskio->tag_num = msg->task.tag_num;
1568                         taskio->tag_type = msg->task.tag_type;
1569 #ifdef CTL_TIME_IO
1570                         taskio->io_hdr.start_time = time_uptime;
1571                         getbinuptime(&taskio->io_hdr.start_bt);
1572 #endif /* CTL_TIME_IO */
1573                         ctl_run_task((union ctl_io *)taskio);
1574                         break;
1575                 }
1576                 /* Persistent Reserve action which needs attention */
1577                 case CTL_MSG_PERS_ACTION:
1578                         presio = (struct ctl_prio *)ctl_alloc_io(
1579                             softc->othersc_pool);
1580                         ctl_zero_io((union ctl_io *)presio);
1581                         presio->io_hdr.msg_type = CTL_MSG_PERS_ACTION;
1582                         presio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
1583                         presio->io_hdr.nexus = msg->hdr.nexus;
1584                         presio->pr_msg = msg->pr;
1585                         ctl_enqueue_isc((union ctl_io *)presio);
1586                         break;
1587                 case CTL_MSG_UA:
1588                         ctl_isc_ua(softc, msg, param);
1589                         break;
1590                 case CTL_MSG_PORT_SYNC:
1591                         ctl_isc_port_sync(softc, msg, param);
1592                         break;
1593                 case CTL_MSG_LUN_SYNC:
1594                         ctl_isc_lun_sync(softc, msg, param);
1595                         break;
1596                 case CTL_MSG_IID_SYNC:
1597                         ctl_isc_iid_sync(softc, msg, param);
1598                         break;
1599                 case CTL_MSG_LOGIN:
1600                         ctl_isc_login(softc, msg, param);
1601                         break;
1602                 case CTL_MSG_MODE_SYNC:
1603                         ctl_isc_mode_sync(softc, msg, param);
1604                         break;
1605                 default:
1606                         printf("Received HA message of unknown type %d\n",
1607                             msg->hdr.msg_type);
1608                         ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1609                         break;
1610                 }
1611                 if (msg != &msgbuf)
1612                         free(msg, M_CTL);
1613         } else if (event == CTL_HA_EVT_LINK_CHANGE) {
1614                 printf("CTL: HA link status changed from %d to %d\n",
1615                     softc->ha_link, param);
1616                 if (param == softc->ha_link)
1617                         return;
1618                 if (softc->ha_link == CTL_HA_LINK_ONLINE) {
1619                         softc->ha_link = param;
1620                         ctl_isc_ha_link_down(softc);
1621                 } else {
1622                         softc->ha_link = param;
1623                         if (softc->ha_link == CTL_HA_LINK_ONLINE)
1624                                 ctl_isc_ha_link_up(softc);
1625                 }
1626                 return;
1627         } else {
1628                 printf("ctl_isc_event_handler: Unknown event %d\n", event);
1629                 return;
1630         }
1631 }
1632
1633 static void
1634 ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest)
1635 {
1636
1637         memcpy(&dest->scsiio.sense_data, &src->scsi.sense_data,
1638             src->scsi.sense_len);
1639         dest->scsiio.scsi_status = src->scsi.scsi_status;
1640         dest->scsiio.sense_len = src->scsi.sense_len;
1641         dest->io_hdr.status = src->hdr.status;
1642 }
1643
1644 static void
1645 ctl_copy_sense_data_back(union ctl_io *src, union ctl_ha_msg *dest)
1646 {
1647
1648         memcpy(&dest->scsi.sense_data, &src->scsiio.sense_data,
1649             src->scsiio.sense_len);
1650         dest->scsi.scsi_status = src->scsiio.scsi_status;
1651         dest->scsi.sense_len = src->scsiio.sense_len;
1652         dest->hdr.status = src->io_hdr.status;
1653 }
1654
1655 void
1656 ctl_est_ua(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua)
1657 {
1658         struct ctl_softc *softc = lun->ctl_softc;
1659         ctl_ua_type *pu;
1660
1661         if (initidx < softc->init_min || initidx >= softc->init_max)
1662                 return;
1663         mtx_assert(&lun->lun_lock, MA_OWNED);
1664         pu = lun->pending_ua[initidx / CTL_MAX_INIT_PER_PORT];
1665         if (pu == NULL)
1666                 return;
1667         pu[initidx % CTL_MAX_INIT_PER_PORT] |= ua;
1668 }
1669
1670 void
1671 ctl_est_ua_port(struct ctl_lun *lun, int port, uint32_t except, ctl_ua_type ua)
1672 {
1673         int i;
1674
1675         mtx_assert(&lun->lun_lock, MA_OWNED);
1676         if (lun->pending_ua[port] == NULL)
1677                 return;
1678         for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1679                 if (port * CTL_MAX_INIT_PER_PORT + i == except)
1680                         continue;
1681                 lun->pending_ua[port][i] |= ua;
1682         }
1683 }
1684
1685 void
1686 ctl_est_ua_all(struct ctl_lun *lun, uint32_t except, ctl_ua_type ua)
1687 {
1688         struct ctl_softc *softc = lun->ctl_softc;
1689         int i;
1690
1691         mtx_assert(&lun->lun_lock, MA_OWNED);
1692         for (i = softc->port_min; i < softc->port_max; i++)
1693                 ctl_est_ua_port(lun, i, except, ua);
1694 }
1695
1696 void
1697 ctl_clr_ua(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua)
1698 {
1699         struct ctl_softc *softc = lun->ctl_softc;
1700         ctl_ua_type *pu;
1701
1702         if (initidx < softc->init_min || initidx >= softc->init_max)
1703                 return;
1704         mtx_assert(&lun->lun_lock, MA_OWNED);
1705         pu = lun->pending_ua[initidx / CTL_MAX_INIT_PER_PORT];
1706         if (pu == NULL)
1707                 return;
1708         pu[initidx % CTL_MAX_INIT_PER_PORT] &= ~ua;
1709 }
1710
1711 void
1712 ctl_clr_ua_all(struct ctl_lun *lun, uint32_t except, ctl_ua_type ua)
1713 {
1714         struct ctl_softc *softc = lun->ctl_softc;
1715         int i, j;
1716
1717         mtx_assert(&lun->lun_lock, MA_OWNED);
1718         for (i = softc->port_min; i < softc->port_max; i++) {
1719                 if (lun->pending_ua[i] == NULL)
1720                         continue;
1721                 for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
1722                         if (i * CTL_MAX_INIT_PER_PORT + j == except)
1723                                 continue;
1724                         lun->pending_ua[i][j] &= ~ua;
1725                 }
1726         }
1727 }
1728
1729 void
1730 ctl_clr_ua_allluns(struct ctl_softc *ctl_softc, uint32_t initidx,
1731     ctl_ua_type ua_type)
1732 {
1733         struct ctl_lun *lun;
1734
1735         mtx_assert(&ctl_softc->ctl_lock, MA_OWNED);
1736         STAILQ_FOREACH(lun, &ctl_softc->lun_list, links) {
1737                 mtx_lock(&lun->lun_lock);
1738                 ctl_clr_ua(lun, initidx, ua_type);
1739                 mtx_unlock(&lun->lun_lock);
1740         }
1741 }
1742
1743 static int
1744 ctl_ha_role_sysctl(SYSCTL_HANDLER_ARGS)
1745 {
1746         struct ctl_softc *softc = (struct ctl_softc *)arg1;
1747         struct ctl_lun *lun;
1748         struct ctl_lun_req ireq;
1749         int error, value;
1750
1751         value = (softc->flags & CTL_FLAG_ACTIVE_SHELF) ? 0 : 1;
1752         error = sysctl_handle_int(oidp, &value, 0, req);
1753         if ((error != 0) || (req->newptr == NULL))
1754                 return (error);
1755
1756         mtx_lock(&softc->ctl_lock);
1757         if (value == 0)
1758                 softc->flags |= CTL_FLAG_ACTIVE_SHELF;
1759         else
1760                 softc->flags &= ~CTL_FLAG_ACTIVE_SHELF;
1761         STAILQ_FOREACH(lun, &softc->lun_list, links) {
1762                 mtx_unlock(&softc->ctl_lock);
1763                 bzero(&ireq, sizeof(ireq));
1764                 ireq.reqtype = CTL_LUNREQ_MODIFY;
1765                 ireq.reqdata.modify.lun_id = lun->lun;
1766                 lun->backend->ioctl(NULL, CTL_LUN_REQ, (caddr_t)&ireq, 0,
1767                     curthread);
1768                 if (ireq.status != CTL_LUN_OK) {
1769                         printf("%s: CTL_LUNREQ_MODIFY returned %d '%s'\n",
1770                             __func__, ireq.status, ireq.error_str);
1771                 }
1772                 mtx_lock(&softc->ctl_lock);
1773         }
1774         mtx_unlock(&softc->ctl_lock);
1775         return (0);
1776 }
1777
1778 static int
1779 ctl_init(void)
1780 {
1781         struct make_dev_args args;
1782         struct ctl_softc *softc;
1783         void *other_pool;
1784         int i, error;
1785
1786         softc = control_softc = malloc(sizeof(*control_softc), M_DEVBUF,
1787                                M_WAITOK | M_ZERO);
1788
1789         make_dev_args_init(&args);
1790         args.mda_devsw = &ctl_cdevsw;
1791         args.mda_uid = UID_ROOT;
1792         args.mda_gid = GID_OPERATOR;
1793         args.mda_mode = 0600;
1794         args.mda_si_drv1 = softc;
1795         error = make_dev_s(&args, &softc->dev, "cam/ctl");
1796         if (error != 0) {
1797                 free(control_softc, M_DEVBUF);
1798                 return (error);
1799         }
1800
1801         sysctl_ctx_init(&softc->sysctl_ctx);
1802         softc->sysctl_tree = SYSCTL_ADD_NODE(&softc->sysctl_ctx,
1803                 SYSCTL_STATIC_CHILDREN(_kern_cam), OID_AUTO, "ctl",
1804                 CTLFLAG_RD, 0, "CAM Target Layer");
1805
1806         if (softc->sysctl_tree == NULL) {
1807                 printf("%s: unable to allocate sysctl tree\n", __func__);
1808                 destroy_dev(softc->dev);
1809                 free(control_softc, M_DEVBUF);
1810                 control_softc = NULL;
1811                 return (ENOMEM);
1812         }
1813
1814         mtx_init(&softc->ctl_lock, "CTL mutex", NULL, MTX_DEF);
1815         softc->io_zone = uma_zcreate("CTL IO", sizeof(union ctl_io),
1816             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1817         softc->flags = 0;
1818
1819         SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1820             OID_AUTO, "ha_mode", CTLFLAG_RDTUN, (int *)&softc->ha_mode, 0,
1821             "HA mode (0 - act/stby, 1 - serialize only, 2 - xfer)");
1822
1823         /*
1824          * In Copan's HA scheme, the "master" and "slave" roles are
1825          * figured out through the slot the controller is in.  Although it
1826          * is an active/active system, someone has to be in charge.
1827          */
1828         SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1829             OID_AUTO, "ha_id", CTLFLAG_RDTUN, &softc->ha_id, 0,
1830             "HA head ID (0 - no HA)");
1831         if (softc->ha_id == 0 || softc->ha_id > NUM_HA_SHELVES) {
1832                 softc->flags |= CTL_FLAG_ACTIVE_SHELF;
1833                 softc->is_single = 1;
1834                 softc->port_cnt = CTL_MAX_PORTS;
1835                 softc->port_min = 0;
1836         } else {
1837                 softc->port_cnt = CTL_MAX_PORTS / NUM_HA_SHELVES;
1838                 softc->port_min = (softc->ha_id - 1) * softc->port_cnt;
1839         }
1840         softc->port_max = softc->port_min + softc->port_cnt;
1841         softc->init_min = softc->port_min * CTL_MAX_INIT_PER_PORT;
1842         softc->init_max = softc->port_max * CTL_MAX_INIT_PER_PORT;
1843
1844         SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1845             OID_AUTO, "ha_link", CTLFLAG_RD, (int *)&softc->ha_link, 0,
1846             "HA link state (0 - offline, 1 - unknown, 2 - online)");
1847
1848         STAILQ_INIT(&softc->lun_list);
1849         STAILQ_INIT(&softc->pending_lun_queue);
1850         STAILQ_INIT(&softc->fe_list);
1851         STAILQ_INIT(&softc->port_list);
1852         STAILQ_INIT(&softc->be_list);
1853         ctl_tpc_init(softc);
1854
1855         if (ctl_pool_create(softc, "othersc", CTL_POOL_ENTRIES_OTHER_SC,
1856                             &other_pool) != 0)
1857         {
1858                 printf("ctl: can't allocate %d entry other SC pool, "
1859                        "exiting\n", CTL_POOL_ENTRIES_OTHER_SC);
1860                 return (ENOMEM);
1861         }
1862         softc->othersc_pool = other_pool;
1863
1864         if (worker_threads <= 0)
1865                 worker_threads = max(1, mp_ncpus / 4);
1866         if (worker_threads > CTL_MAX_THREADS)
1867                 worker_threads = CTL_MAX_THREADS;
1868
1869         for (i = 0; i < worker_threads; i++) {
1870                 struct ctl_thread *thr = &softc->threads[i];
1871
1872                 mtx_init(&thr->queue_lock, "CTL queue mutex", NULL, MTX_DEF);
1873                 thr->ctl_softc = softc;
1874                 STAILQ_INIT(&thr->incoming_queue);
1875                 STAILQ_INIT(&thr->rtr_queue);
1876                 STAILQ_INIT(&thr->done_queue);
1877                 STAILQ_INIT(&thr->isc_queue);
1878
1879                 error = kproc_kthread_add(ctl_work_thread, thr,
1880                     &softc->ctl_proc, &thr->thread, 0, 0, "ctl", "work%d", i);
1881                 if (error != 0) {
1882                         printf("error creating CTL work thread!\n");
1883                         ctl_pool_free(other_pool);
1884                         return (error);
1885                 }
1886         }
1887         error = kproc_kthread_add(ctl_lun_thread, softc,
1888             &softc->ctl_proc, NULL, 0, 0, "ctl", "lun");
1889         if (error != 0) {
1890                 printf("error creating CTL lun thread!\n");
1891                 ctl_pool_free(other_pool);
1892                 return (error);
1893         }
1894         error = kproc_kthread_add(ctl_thresh_thread, softc,
1895             &softc->ctl_proc, NULL, 0, 0, "ctl", "thresh");
1896         if (error != 0) {
1897                 printf("error creating CTL threshold thread!\n");
1898                 ctl_pool_free(other_pool);
1899                 return (error);
1900         }
1901
1902         SYSCTL_ADD_PROC(&softc->sysctl_ctx,SYSCTL_CHILDREN(softc->sysctl_tree),
1903             OID_AUTO, "ha_role", CTLTYPE_INT | CTLFLAG_RWTUN,
1904             softc, 0, ctl_ha_role_sysctl, "I", "HA role for this head");
1905
1906         if (softc->is_single == 0) {
1907                 ctl_frontend_register(&ha_frontend);
1908                 if (ctl_ha_msg_init(softc) != CTL_HA_STATUS_SUCCESS) {
1909                         printf("ctl_init: ctl_ha_msg_init failed.\n");
1910                         softc->is_single = 1;
1911                 } else
1912                 if (ctl_ha_msg_register(CTL_HA_CHAN_CTL, ctl_isc_event_handler)
1913                     != CTL_HA_STATUS_SUCCESS) {
1914                         printf("ctl_init: ctl_ha_msg_register failed.\n");
1915                         softc->is_single = 1;
1916                 }
1917         }
1918         return (0);
1919 }
1920
1921 void
1922 ctl_shutdown(void)
1923 {
1924         struct ctl_softc *softc = control_softc;
1925         struct ctl_lun *lun, *next_lun;
1926
1927         if (softc->is_single == 0) {
1928                 ctl_ha_msg_shutdown(softc);
1929                 if (ctl_ha_msg_deregister(CTL_HA_CHAN_CTL)
1930                     != CTL_HA_STATUS_SUCCESS)
1931                         printf("%s: ctl_ha_msg_deregister failed.\n", __func__);
1932                 if (ctl_ha_msg_destroy(softc) != CTL_HA_STATUS_SUCCESS)
1933                         printf("%s: ctl_ha_msg_destroy failed.\n", __func__);
1934                 ctl_frontend_deregister(&ha_frontend);
1935         }
1936
1937         mtx_lock(&softc->ctl_lock);
1938
1939         STAILQ_FOREACH_SAFE(lun, &softc->lun_list, links, next_lun)
1940                 ctl_free_lun(lun);
1941
1942         mtx_unlock(&softc->ctl_lock);
1943
1944 #if 0
1945         ctl_shutdown_thread(softc->work_thread);
1946         mtx_destroy(&softc->queue_lock);
1947 #endif
1948
1949         ctl_tpc_shutdown(softc);
1950         uma_zdestroy(softc->io_zone);
1951         mtx_destroy(&softc->ctl_lock);
1952
1953         destroy_dev(softc->dev);
1954
1955         sysctl_ctx_free(&softc->sysctl_ctx);
1956
1957         free(control_softc, M_DEVBUF);
1958         control_softc = NULL;
1959 }
1960
1961 static int
1962 ctl_module_event_handler(module_t mod, int what, void *arg)
1963 {
1964
1965         switch (what) {
1966         case MOD_LOAD:
1967                 return (ctl_init());
1968         case MOD_UNLOAD:
1969                 return (EBUSY);
1970         default:
1971                 return (EOPNOTSUPP);
1972         }
1973 }
1974
1975 /*
1976  * XXX KDM should we do some access checks here?  Bump a reference count to
1977  * prevent a CTL module from being unloaded while someone has it open?
1978  */
1979 static int
1980 ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td)
1981 {
1982         return (0);
1983 }
1984
1985 static int
1986 ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td)
1987 {
1988         return (0);
1989 }
1990
1991 /*
1992  * Remove an initiator by port number and initiator ID.
1993  * Returns 0 for success, -1 for failure.
1994  */
1995 int
1996 ctl_remove_initiator(struct ctl_port *port, int iid)
1997 {
1998         struct ctl_softc *softc = port->ctl_softc;
1999
2000         mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
2001
2002         if (iid > CTL_MAX_INIT_PER_PORT) {
2003                 printf("%s: initiator ID %u > maximun %u!\n",
2004                        __func__, iid, CTL_MAX_INIT_PER_PORT);
2005                 return (-1);
2006         }
2007
2008         mtx_lock(&softc->ctl_lock);
2009         port->wwpn_iid[iid].in_use--;
2010         port->wwpn_iid[iid].last_use = time_uptime;
2011         mtx_unlock(&softc->ctl_lock);
2012         ctl_isc_announce_iid(port, iid);
2013
2014         return (0);
2015 }
2016
2017 /*
2018  * Add an initiator to the initiator map.
2019  * Returns iid for success, < 0 for failure.
2020  */
2021 int
2022 ctl_add_initiator(struct ctl_port *port, int iid, uint64_t wwpn, char *name)
2023 {
2024         struct ctl_softc *softc = port->ctl_softc;
2025         time_t best_time;
2026         int i, best;
2027
2028         mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
2029
2030         if (iid >= CTL_MAX_INIT_PER_PORT) {
2031                 printf("%s: WWPN %#jx initiator ID %u > maximum %u!\n",
2032                        __func__, wwpn, iid, CTL_MAX_INIT_PER_PORT);
2033                 free(name, M_CTL);
2034                 return (-1);
2035         }
2036
2037         mtx_lock(&softc->ctl_lock);
2038
2039         if (iid < 0 && (wwpn != 0 || name != NULL)) {
2040                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
2041                         if (wwpn != 0 && wwpn == port->wwpn_iid[i].wwpn) {
2042                                 iid = i;
2043                                 break;
2044                         }
2045                         if (name != NULL && port->wwpn_iid[i].name != NULL &&
2046                             strcmp(name, port->wwpn_iid[i].name) == 0) {
2047                                 iid = i;
2048                                 break;
2049                         }
2050                 }
2051         }
2052
2053         if (iid < 0) {
2054                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
2055                         if (port->wwpn_iid[i].in_use == 0 &&
2056                             port->wwpn_iid[i].wwpn == 0 &&
2057                             port->wwpn_iid[i].name == NULL) {
2058                                 iid = i;
2059                                 break;
2060                         }
2061                 }
2062         }
2063
2064         if (iid < 0) {
2065                 best = -1;
2066                 best_time = INT32_MAX;
2067                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
2068                         if (port->wwpn_iid[i].in_use == 0) {
2069                                 if (port->wwpn_iid[i].last_use < best_time) {
2070                                         best = i;
2071                                         best_time = port->wwpn_iid[i].last_use;
2072                                 }
2073                         }
2074                 }
2075                 iid = best;
2076         }
2077
2078         if (iid < 0) {
2079                 mtx_unlock(&softc->ctl_lock);
2080                 free(name, M_CTL);
2081                 return (-2);
2082         }
2083
2084         if (port->wwpn_iid[iid].in_use > 0 && (wwpn != 0 || name != NULL)) {
2085                 /*
2086                  * This is not an error yet.
2087                  */
2088                 if (wwpn != 0 && wwpn == port->wwpn_iid[iid].wwpn) {
2089 #if 0
2090                         printf("%s: port %d iid %u WWPN %#jx arrived"
2091                             " again\n", __func__, port->targ_port,
2092                             iid, (uintmax_t)wwpn);
2093 #endif
2094                         goto take;
2095                 }
2096                 if (name != NULL && port->wwpn_iid[iid].name != NULL &&
2097                     strcmp(name, port->wwpn_iid[iid].name) == 0) {
2098 #if 0
2099                         printf("%s: port %d iid %u name '%s' arrived"
2100                             " again\n", __func__, port->targ_port,
2101                             iid, name);
2102 #endif
2103                         goto take;
2104                 }
2105
2106                 /*
2107                  * This is an error, but what do we do about it?  The
2108                  * driver is telling us we have a new WWPN for this
2109                  * initiator ID, so we pretty much need to use it.
2110                  */
2111                 printf("%s: port %d iid %u WWPN %#jx '%s' arrived,"
2112                     " but WWPN %#jx '%s' is still at that address\n",
2113                     __func__, port->targ_port, iid, wwpn, name,
2114                     (uintmax_t)port->wwpn_iid[iid].wwpn,
2115                     port->wwpn_iid[iid].name);
2116
2117                 /*
2118                  * XXX KDM clear have_ca and ua_pending on each LUN for
2119                  * this initiator.
2120                  */
2121         }
2122 take:
2123         free(port->wwpn_iid[iid].name, M_CTL);
2124         port->wwpn_iid[iid].name = name;
2125         port->wwpn_iid[iid].wwpn = wwpn;
2126         port->wwpn_iid[iid].in_use++;
2127         mtx_unlock(&softc->ctl_lock);
2128         ctl_isc_announce_iid(port, iid);
2129
2130         return (iid);
2131 }
2132
2133 static int
2134 ctl_create_iid(struct ctl_port *port, int iid, uint8_t *buf)
2135 {
2136         int len;
2137
2138         switch (port->port_type) {
2139         case CTL_PORT_FC:
2140         {
2141                 struct scsi_transportid_fcp *id =
2142                     (struct scsi_transportid_fcp *)buf;
2143                 if (port->wwpn_iid[iid].wwpn == 0)
2144                         return (0);
2145                 memset(id, 0, sizeof(*id));
2146                 id->format_protocol = SCSI_PROTO_FC;
2147                 scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->n_port_name);
2148                 return (sizeof(*id));
2149         }
2150         case CTL_PORT_ISCSI:
2151         {
2152                 struct scsi_transportid_iscsi_port *id =
2153                     (struct scsi_transportid_iscsi_port *)buf;
2154                 if (port->wwpn_iid[iid].name == NULL)
2155                         return (0);
2156                 memset(id, 0, 256);
2157                 id->format_protocol = SCSI_TRN_ISCSI_FORMAT_PORT |
2158                     SCSI_PROTO_ISCSI;
2159                 len = strlcpy(id->iscsi_name, port->wwpn_iid[iid].name, 252) + 1;
2160                 len = roundup2(min(len, 252), 4);
2161                 scsi_ulto2b(len, id->additional_length);
2162                 return (sizeof(*id) + len);
2163         }
2164         case CTL_PORT_SAS:
2165         {
2166                 struct scsi_transportid_sas *id =
2167                     (struct scsi_transportid_sas *)buf;
2168                 if (port->wwpn_iid[iid].wwpn == 0)
2169                         return (0);
2170                 memset(id, 0, sizeof(*id));
2171                 id->format_protocol = SCSI_PROTO_SAS;
2172                 scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->sas_address);
2173                 return (sizeof(*id));
2174         }
2175         default:
2176         {
2177                 struct scsi_transportid_spi *id =
2178                     (struct scsi_transportid_spi *)buf;
2179                 memset(id, 0, sizeof(*id));
2180                 id->format_protocol = SCSI_PROTO_SPI;
2181                 scsi_ulto2b(iid, id->scsi_addr);
2182                 scsi_ulto2b(port->targ_port, id->rel_trgt_port_id);
2183                 return (sizeof(*id));
2184         }
2185         }
2186 }
2187
2188 /*
2189  * Serialize a command that went down the "wrong" side, and so was sent to
2190  * this controller for execution.  The logic is a little different than the
2191  * standard case in ctl_scsiio_precheck().  Errors in this case need to get
2192  * sent back to the other side, but in the success case, we execute the
2193  * command on this side (XFER mode) or tell the other side to execute it
2194  * (SER_ONLY mode).
2195  */
2196 static int
2197 ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio)
2198 {
2199         struct ctl_softc *softc = control_softc;
2200         union ctl_ha_msg msg_info;
2201         struct ctl_port *port;
2202         struct ctl_lun *lun;
2203         const struct ctl_cmd_entry *entry;
2204         int retval = 0;
2205         uint32_t targ_lun;
2206
2207         targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
2208         mtx_lock(&softc->ctl_lock);
2209
2210         /* Make sure that we know about this port. */
2211         port = ctl_io_port(&ctsio->io_hdr);
2212         if (port == NULL || (port->status & CTL_PORT_STATUS_ONLINE) == 0) {
2213                 ctl_set_internal_failure(ctsio, /*sks_valid*/ 0,
2214                                          /*retry_count*/ 1);
2215                 goto badjuju;
2216         }
2217
2218         /* Make sure that we know about this LUN. */
2219         if ((targ_lun < CTL_MAX_LUNS) &&
2220             ((lun = softc->ctl_luns[targ_lun]) != NULL)) {
2221                 mtx_lock(&lun->lun_lock);
2222                 mtx_unlock(&softc->ctl_lock);
2223                 /*
2224                  * If the LUN is invalid, pretend that it doesn't exist.
2225                  * It will go away as soon as all pending I/O has been
2226                  * completed.
2227                  */
2228                 if (lun->flags & CTL_LUN_DISABLED) {
2229                         mtx_unlock(&lun->lun_lock);
2230                         lun = NULL;
2231                 }
2232         } else {
2233                 mtx_unlock(&softc->ctl_lock);
2234                 lun = NULL;
2235         }
2236         if (lun == NULL) {
2237                 /*
2238                  * The other node would not send this request to us unless
2239                  * received announce that we are primary node for this LUN.
2240                  * If this LUN does not exist now, it is probably result of
2241                  * a race, so respond to initiator in the most opaque way.
2242                  */
2243                 ctl_set_busy(ctsio);
2244                 goto badjuju;
2245         }
2246
2247         entry = ctl_get_cmd_entry(ctsio, NULL);
2248         if (ctl_scsiio_lun_check(lun, entry, ctsio) != 0) {
2249                 mtx_unlock(&lun->lun_lock);
2250                 goto badjuju;
2251         }
2252
2253         ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = lun;
2254         ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = lun->be_lun;
2255
2256         /*
2257          * Every I/O goes into the OOA queue for a
2258          * particular LUN, and stays there until completion.
2259          */
2260 #ifdef CTL_TIME_IO
2261         if (TAILQ_EMPTY(&lun->ooa_queue))
2262                 lun->idle_time += getsbinuptime() - lun->last_busy;
2263 #endif
2264         TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2265
2266         switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
2267                 (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr, ctl_ooaq,
2268                  ooa_links))) {
2269         case CTL_ACTION_BLOCK:
2270                 ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
2271                 TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
2272                                   blocked_links);
2273                 mtx_unlock(&lun->lun_lock);
2274                 break;
2275         case CTL_ACTION_PASS:
2276         case CTL_ACTION_SKIP:
2277                 if (softc->ha_mode == CTL_HA_MODE_XFER) {
2278                         ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
2279                         ctl_enqueue_rtr((union ctl_io *)ctsio);
2280                         mtx_unlock(&lun->lun_lock);
2281                 } else {
2282                         ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
2283                         mtx_unlock(&lun->lun_lock);
2284
2285                         /* send msg back to other side */
2286                         msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
2287                         msg_info.hdr.serializing_sc = (union ctl_io *)ctsio;
2288                         msg_info.hdr.msg_type = CTL_MSG_R2R;
2289                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
2290                             sizeof(msg_info.hdr), M_WAITOK);
2291                 }
2292                 break;
2293         case CTL_ACTION_OVERLAP:
2294                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2295                 mtx_unlock(&lun->lun_lock);
2296                 ctl_set_overlapped_cmd(ctsio);
2297                 goto badjuju;
2298         case CTL_ACTION_OVERLAP_TAG:
2299                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2300                 mtx_unlock(&lun->lun_lock);
2301                 ctl_set_overlapped_tag(ctsio, ctsio->tag_num);
2302                 goto badjuju;
2303         case CTL_ACTION_ERROR:
2304         default:
2305                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2306                 mtx_unlock(&lun->lun_lock);
2307
2308                 ctl_set_internal_failure(ctsio, /*sks_valid*/ 0,
2309                                          /*retry_count*/ 0);
2310 badjuju:
2311                 ctl_copy_sense_data_back((union ctl_io *)ctsio, &msg_info);
2312                 msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
2313                 msg_info.hdr.serializing_sc = NULL;
2314                 msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
2315                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
2316                     sizeof(msg_info.scsi), M_WAITOK);
2317                 retval = 1;
2318                 break;
2319         }
2320         return (retval);
2321 }
2322
2323 /*
2324  * Returns 0 for success, errno for failure.
2325  */
2326 static void
2327 ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
2328                    struct ctl_ooa *ooa_hdr, struct ctl_ooa_entry *kern_entries)
2329 {
2330         union ctl_io *io;
2331
2332         mtx_lock(&lun->lun_lock);
2333         for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); (io != NULL);
2334              (*cur_fill_num)++, io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2335              ooa_links)) {
2336                 struct ctl_ooa_entry *entry;
2337
2338                 /*
2339                  * If we've got more than we can fit, just count the
2340                  * remaining entries.
2341                  */
2342                 if (*cur_fill_num >= ooa_hdr->alloc_num)
2343                         continue;
2344
2345                 entry = &kern_entries[*cur_fill_num];
2346
2347                 entry->tag_num = io->scsiio.tag_num;
2348                 entry->lun_num = lun->lun;
2349 #ifdef CTL_TIME_IO
2350                 entry->start_bt = io->io_hdr.start_bt;
2351 #endif
2352                 bcopy(io->scsiio.cdb, entry->cdb, io->scsiio.cdb_len);
2353                 entry->cdb_len = io->scsiio.cdb_len;
2354                 if (io->io_hdr.flags & CTL_FLAG_BLOCKED)
2355                         entry->cmd_flags |= CTL_OOACMD_FLAG_BLOCKED;
2356
2357                 if (io->io_hdr.flags & CTL_FLAG_DMA_INPROG)
2358                         entry->cmd_flags |= CTL_OOACMD_FLAG_DMA;
2359
2360                 if (io->io_hdr.flags & CTL_FLAG_ABORT)
2361                         entry->cmd_flags |= CTL_OOACMD_FLAG_ABORT;
2362
2363                 if (io->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR)
2364                         entry->cmd_flags |= CTL_OOACMD_FLAG_RTR;
2365
2366                 if (io->io_hdr.flags & CTL_FLAG_DMA_QUEUED)
2367                         entry->cmd_flags |= CTL_OOACMD_FLAG_DMA_QUEUED;
2368         }
2369         mtx_unlock(&lun->lun_lock);
2370 }
2371
2372 static void *
2373 ctl_copyin_alloc(void *user_addr, unsigned int len, char *error_str,
2374                  size_t error_str_len)
2375 {
2376         void *kptr;
2377
2378         kptr = malloc(len, M_CTL, M_WAITOK | M_ZERO);
2379
2380         if (copyin(user_addr, kptr, len) != 0) {
2381                 snprintf(error_str, error_str_len, "Error copying %d bytes "
2382                          "from user address %p to kernel address %p", len,
2383                          user_addr, kptr);
2384                 free(kptr, M_CTL);
2385                 return (NULL);
2386         }
2387
2388         return (kptr);
2389 }
2390
2391 static void
2392 ctl_free_args(int num_args, struct ctl_be_arg *args)
2393 {
2394         int i;
2395
2396         if (args == NULL)
2397                 return;
2398
2399         for (i = 0; i < num_args; i++) {
2400                 free(args[i].kname, M_CTL);
2401                 free(args[i].kvalue, M_CTL);
2402         }
2403
2404         free(args, M_CTL);
2405 }
2406
2407 static struct ctl_be_arg *
2408 ctl_copyin_args(int num_args, struct ctl_be_arg *uargs,
2409                 char *error_str, size_t error_str_len)
2410 {
2411         struct ctl_be_arg *args;
2412         int i;
2413
2414         args = ctl_copyin_alloc(uargs, num_args * sizeof(*args),
2415                                 error_str, error_str_len);
2416
2417         if (args == NULL)
2418                 goto bailout;
2419
2420         for (i = 0; i < num_args; i++) {
2421                 args[i].kname = NULL;
2422                 args[i].kvalue = NULL;
2423         }
2424
2425         for (i = 0; i < num_args; i++) {
2426                 uint8_t *tmpptr;
2427
2428                 if (args[i].namelen == 0) {
2429                         snprintf(error_str, error_str_len, "Argument %d "
2430                                  "name length is zero", i);
2431                         goto bailout;
2432                 }
2433
2434                 args[i].kname = ctl_copyin_alloc(args[i].name,
2435                         args[i].namelen, error_str, error_str_len);
2436                 if (args[i].kname == NULL)
2437                         goto bailout;
2438
2439                 if (args[i].kname[args[i].namelen - 1] != '\0') {
2440                         snprintf(error_str, error_str_len, "Argument %d "
2441                                  "name is not NUL-terminated", i);
2442                         goto bailout;
2443                 }
2444
2445                 if (args[i].flags & CTL_BEARG_RD) {
2446                         if (args[i].vallen == 0) {
2447                                 snprintf(error_str, error_str_len, "Argument %d "
2448                                          "value length is zero", i);
2449                                 goto bailout;
2450                         }
2451
2452                         tmpptr = ctl_copyin_alloc(args[i].value,
2453                                 args[i].vallen, error_str, error_str_len);
2454                         if (tmpptr == NULL)
2455                                 goto bailout;
2456
2457                         if ((args[i].flags & CTL_BEARG_ASCII)
2458                          && (tmpptr[args[i].vallen - 1] != '\0')) {
2459                                 snprintf(error_str, error_str_len, "Argument "
2460                                     "%d value is not NUL-terminated", i);
2461                                 free(tmpptr, M_CTL);
2462                                 goto bailout;
2463                         }
2464                         args[i].kvalue = tmpptr;
2465                 } else {
2466                         args[i].kvalue = malloc(args[i].vallen,
2467                             M_CTL, M_WAITOK | M_ZERO);
2468                 }
2469         }
2470
2471         return (args);
2472 bailout:
2473
2474         ctl_free_args(num_args, args);
2475
2476         return (NULL);
2477 }
2478
2479 static void
2480 ctl_copyout_args(int num_args, struct ctl_be_arg *args)
2481 {
2482         int i;
2483
2484         for (i = 0; i < num_args; i++) {
2485                 if (args[i].flags & CTL_BEARG_WR)
2486                         copyout(args[i].kvalue, args[i].value, args[i].vallen);
2487         }
2488 }
2489
2490 /*
2491  * Escape characters that are illegal or not recommended in XML.
2492  */
2493 int
2494 ctl_sbuf_printf_esc(struct sbuf *sb, char *str, int size)
2495 {
2496         char *end = str + size;
2497         int retval;
2498
2499         retval = 0;
2500
2501         for (; *str && str < end; str++) {
2502                 switch (*str) {
2503                 case '&':
2504                         retval = sbuf_printf(sb, "&amp;");
2505                         break;
2506                 case '>':
2507                         retval = sbuf_printf(sb, "&gt;");
2508                         break;
2509                 case '<':
2510                         retval = sbuf_printf(sb, "&lt;");
2511                         break;
2512                 default:
2513                         retval = sbuf_putc(sb, *str);
2514                         break;
2515                 }
2516
2517                 if (retval != 0)
2518                         break;
2519
2520         }
2521
2522         return (retval);
2523 }
2524
2525 static void
2526 ctl_id_sbuf(struct ctl_devid *id, struct sbuf *sb)
2527 {
2528         struct scsi_vpd_id_descriptor *desc;
2529         int i;
2530
2531         if (id == NULL || id->len < 4)
2532                 return;
2533         desc = (struct scsi_vpd_id_descriptor *)id->data;
2534         switch (desc->id_type & SVPD_ID_TYPE_MASK) {
2535         case SVPD_ID_TYPE_T10:
2536                 sbuf_printf(sb, "t10.");
2537                 break;
2538         case SVPD_ID_TYPE_EUI64:
2539                 sbuf_printf(sb, "eui.");
2540                 break;
2541         case SVPD_ID_TYPE_NAA:
2542                 sbuf_printf(sb, "naa.");
2543                 break;
2544         case SVPD_ID_TYPE_SCSI_NAME:
2545                 break;
2546         }
2547         switch (desc->proto_codeset & SVPD_ID_CODESET_MASK) {
2548         case SVPD_ID_CODESET_BINARY:
2549                 for (i = 0; i < desc->length; i++)
2550                         sbuf_printf(sb, "%02x", desc->identifier[i]);
2551                 break;
2552         case SVPD_ID_CODESET_ASCII:
2553                 sbuf_printf(sb, "%.*s", (int)desc->length,
2554                     (char *)desc->identifier);
2555                 break;
2556         case SVPD_ID_CODESET_UTF8:
2557                 sbuf_printf(sb, "%s", (char *)desc->identifier);
2558                 break;
2559         }
2560 }
2561
2562 static int
2563 ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
2564           struct thread *td)
2565 {
2566         struct ctl_softc *softc = dev->si_drv1;
2567         struct ctl_lun *lun;
2568         int retval;
2569
2570         retval = 0;
2571
2572         switch (cmd) {
2573         case CTL_IO:
2574                 retval = ctl_ioctl_io(dev, cmd, addr, flag, td);
2575                 break;
2576         case CTL_ENABLE_PORT:
2577         case CTL_DISABLE_PORT:
2578         case CTL_SET_PORT_WWNS: {
2579                 struct ctl_port *port;
2580                 struct ctl_port_entry *entry;
2581
2582                 entry = (struct ctl_port_entry *)addr;
2583
2584                 mtx_lock(&softc->ctl_lock);
2585                 STAILQ_FOREACH(port, &softc->port_list, links) {
2586                         int action, done;
2587
2588                         if (port->targ_port < softc->port_min ||
2589                             port->targ_port >= softc->port_max)
2590                                 continue;
2591
2592                         action = 0;
2593                         done = 0;
2594                         if ((entry->port_type == CTL_PORT_NONE)
2595                          && (entry->targ_port == port->targ_port)) {
2596                                 /*
2597                                  * If the user only wants to enable or
2598                                  * disable or set WWNs on a specific port,
2599                                  * do the operation and we're done.
2600                                  */
2601                                 action = 1;
2602                                 done = 1;
2603                         } else if (entry->port_type & port->port_type) {
2604                                 /*
2605                                  * Compare the user's type mask with the
2606                                  * particular frontend type to see if we
2607                                  * have a match.
2608                                  */
2609                                 action = 1;
2610                                 done = 0;
2611
2612                                 /*
2613                                  * Make sure the user isn't trying to set
2614                                  * WWNs on multiple ports at the same time.
2615                                  */
2616                                 if (cmd == CTL_SET_PORT_WWNS) {
2617                                         printf("%s: Can't set WWNs on "
2618                                                "multiple ports\n", __func__);
2619                                         retval = EINVAL;
2620                                         break;
2621                                 }
2622                         }
2623                         if (action == 0)
2624                                 continue;
2625
2626                         /*
2627                          * XXX KDM we have to drop the lock here, because
2628                          * the online/offline operations can potentially
2629                          * block.  We need to reference count the frontends
2630                          * so they can't go away,
2631                          */
2632                         if (cmd == CTL_ENABLE_PORT) {
2633                                 mtx_unlock(&softc->ctl_lock);
2634                                 ctl_port_online(port);
2635                                 mtx_lock(&softc->ctl_lock);
2636                         } else if (cmd == CTL_DISABLE_PORT) {
2637                                 mtx_unlock(&softc->ctl_lock);
2638                                 ctl_port_offline(port);
2639                                 mtx_lock(&softc->ctl_lock);
2640                         } else if (cmd == CTL_SET_PORT_WWNS) {
2641                                 ctl_port_set_wwns(port,
2642                                     (entry->flags & CTL_PORT_WWNN_VALID) ?
2643                                     1 : 0, entry->wwnn,
2644                                     (entry->flags & CTL_PORT_WWPN_VALID) ?
2645                                     1 : 0, entry->wwpn);
2646                         }
2647                         if (done != 0)
2648                                 break;
2649                 }
2650                 mtx_unlock(&softc->ctl_lock);
2651                 break;
2652         }
2653         case CTL_GET_OOA: {
2654                 struct ctl_ooa *ooa_hdr;
2655                 struct ctl_ooa_entry *entries;
2656                 uint32_t cur_fill_num;
2657
2658                 ooa_hdr = (struct ctl_ooa *)addr;
2659
2660                 if ((ooa_hdr->alloc_len == 0)
2661                  || (ooa_hdr->alloc_num == 0)) {
2662                         printf("%s: CTL_GET_OOA: alloc len %u and alloc num %u "
2663                                "must be non-zero\n", __func__,
2664                                ooa_hdr->alloc_len, ooa_hdr->alloc_num);
2665                         retval = EINVAL;
2666                         break;
2667                 }
2668
2669                 if (ooa_hdr->alloc_len != (ooa_hdr->alloc_num *
2670                     sizeof(struct ctl_ooa_entry))) {
2671                         printf("%s: CTL_GET_OOA: alloc len %u must be alloc "
2672                                "num %d * sizeof(struct ctl_ooa_entry) %zd\n",
2673                                __func__, ooa_hdr->alloc_len,
2674                                ooa_hdr->alloc_num,sizeof(struct ctl_ooa_entry));
2675                         retval = EINVAL;
2676                         break;
2677                 }
2678
2679                 entries = malloc(ooa_hdr->alloc_len, M_CTL, M_WAITOK | M_ZERO);
2680                 if (entries == NULL) {
2681                         printf("%s: could not allocate %d bytes for OOA "
2682                                "dump\n", __func__, ooa_hdr->alloc_len);
2683                         retval = ENOMEM;
2684                         break;
2685                 }
2686
2687                 mtx_lock(&softc->ctl_lock);
2688                 if (((ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) == 0)
2689                  && ((ooa_hdr->lun_num >= CTL_MAX_LUNS)
2690                   || (softc->ctl_luns[ooa_hdr->lun_num] == NULL))) {
2691                         mtx_unlock(&softc->ctl_lock);
2692                         free(entries, M_CTL);
2693                         printf("%s: CTL_GET_OOA: invalid LUN %ju\n",
2694                                __func__, (uintmax_t)ooa_hdr->lun_num);
2695                         retval = EINVAL;
2696                         break;
2697                 }
2698
2699                 cur_fill_num = 0;
2700
2701                 if (ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) {
2702                         STAILQ_FOREACH(lun, &softc->lun_list, links) {
2703                                 ctl_ioctl_fill_ooa(lun, &cur_fill_num,
2704                                     ooa_hdr, entries);
2705                         }
2706                 } else {
2707                         lun = softc->ctl_luns[ooa_hdr->lun_num];
2708                         ctl_ioctl_fill_ooa(lun, &cur_fill_num, ooa_hdr,
2709                             entries);
2710                 }
2711                 mtx_unlock(&softc->ctl_lock);
2712
2713                 ooa_hdr->fill_num = min(cur_fill_num, ooa_hdr->alloc_num);
2714                 ooa_hdr->fill_len = ooa_hdr->fill_num *
2715                         sizeof(struct ctl_ooa_entry);
2716                 retval = copyout(entries, ooa_hdr->entries, ooa_hdr->fill_len);
2717                 if (retval != 0) {
2718                         printf("%s: error copying out %d bytes for OOA dump\n",
2719                                __func__, ooa_hdr->fill_len);
2720                 }
2721
2722                 getbinuptime(&ooa_hdr->cur_bt);
2723
2724                 if (cur_fill_num > ooa_hdr->alloc_num) {
2725                         ooa_hdr->dropped_num = cur_fill_num -ooa_hdr->alloc_num;
2726                         ooa_hdr->status = CTL_OOA_NEED_MORE_SPACE;
2727                 } else {
2728                         ooa_hdr->dropped_num = 0;
2729                         ooa_hdr->status = CTL_OOA_OK;
2730                 }
2731
2732                 free(entries, M_CTL);
2733                 break;
2734         }
2735         case CTL_DELAY_IO: {
2736                 struct ctl_io_delay_info *delay_info;
2737
2738                 delay_info = (struct ctl_io_delay_info *)addr;
2739
2740 #ifdef CTL_IO_DELAY
2741                 mtx_lock(&softc->ctl_lock);
2742
2743                 if ((delay_info->lun_id >= CTL_MAX_LUNS)
2744                  || (softc->ctl_luns[delay_info->lun_id] == NULL)) {
2745                         delay_info->status = CTL_DELAY_STATUS_INVALID_LUN;
2746                 } else {
2747                         lun = softc->ctl_luns[delay_info->lun_id];
2748                         mtx_lock(&lun->lun_lock);
2749
2750                         delay_info->status = CTL_DELAY_STATUS_OK;
2751
2752                         switch (delay_info->delay_type) {
2753                         case CTL_DELAY_TYPE_CONT:
2754                                 break;
2755                         case CTL_DELAY_TYPE_ONESHOT:
2756                                 break;
2757                         default:
2758                                 delay_info->status =
2759                                         CTL_DELAY_STATUS_INVALID_TYPE;
2760                                 break;
2761                         }
2762
2763                         switch (delay_info->delay_loc) {
2764                         case CTL_DELAY_LOC_DATAMOVE:
2765                                 lun->delay_info.datamove_type =
2766                                         delay_info->delay_type;
2767                                 lun->delay_info.datamove_delay =
2768                                         delay_info->delay_secs;
2769                                 break;
2770                         case CTL_DELAY_LOC_DONE:
2771                                 lun->delay_info.done_type =
2772                                         delay_info->delay_type;
2773                                 lun->delay_info.done_delay =
2774                                         delay_info->delay_secs;
2775                                 break;
2776                         default:
2777                                 delay_info->status =
2778                                         CTL_DELAY_STATUS_INVALID_LOC;
2779                                 break;
2780                         }
2781                         mtx_unlock(&lun->lun_lock);
2782                 }
2783
2784                 mtx_unlock(&softc->ctl_lock);
2785 #else
2786                 delay_info->status = CTL_DELAY_STATUS_NOT_IMPLEMENTED;
2787 #endif /* CTL_IO_DELAY */
2788                 break;
2789         }
2790         case CTL_GETSTATS: {
2791                 struct ctl_stats *stats;
2792                 int i;
2793
2794                 stats = (struct ctl_stats *)addr;
2795
2796                 if ((sizeof(struct ctl_lun_io_stats) * softc->num_luns) >
2797                      stats->alloc_len) {
2798                         stats->status = CTL_SS_NEED_MORE_SPACE;
2799                         stats->num_luns = softc->num_luns;
2800                         break;
2801                 }
2802                 /*
2803                  * XXX KDM no locking here.  If the LUN list changes,
2804                  * things can blow up.
2805                  */
2806                 i = 0;
2807                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
2808                         retval = copyout(&lun->stats, &stats->lun_stats[i++],
2809                                          sizeof(lun->stats));
2810                         if (retval != 0)
2811                                 break;
2812                 }
2813                 stats->num_luns = softc->num_luns;
2814                 stats->fill_len = sizeof(struct ctl_lun_io_stats) *
2815                                  softc->num_luns;
2816                 stats->status = CTL_SS_OK;
2817 #ifdef CTL_TIME_IO
2818                 stats->flags = CTL_STATS_FLAG_TIME_VALID;
2819 #else
2820                 stats->flags = CTL_STATS_FLAG_NONE;
2821 #endif
2822                 getnanouptime(&stats->timestamp);
2823                 break;
2824         }
2825         case CTL_ERROR_INJECT: {
2826                 struct ctl_error_desc *err_desc, *new_err_desc;
2827
2828                 err_desc = (struct ctl_error_desc *)addr;
2829
2830                 new_err_desc = malloc(sizeof(*new_err_desc), M_CTL,
2831                                       M_WAITOK | M_ZERO);
2832                 bcopy(err_desc, new_err_desc, sizeof(*new_err_desc));
2833
2834                 mtx_lock(&softc->ctl_lock);
2835                 lun = softc->ctl_luns[err_desc->lun_id];
2836                 if (lun == NULL) {
2837                         mtx_unlock(&softc->ctl_lock);
2838                         free(new_err_desc, M_CTL);
2839                         printf("%s: CTL_ERROR_INJECT: invalid LUN %ju\n",
2840                                __func__, (uintmax_t)err_desc->lun_id);
2841                         retval = EINVAL;
2842                         break;
2843                 }
2844                 mtx_lock(&lun->lun_lock);
2845                 mtx_unlock(&softc->ctl_lock);
2846
2847                 /*
2848                  * We could do some checking here to verify the validity
2849                  * of the request, but given the complexity of error
2850                  * injection requests, the checking logic would be fairly
2851                  * complex.
2852                  *
2853                  * For now, if the request is invalid, it just won't get
2854                  * executed and might get deleted.
2855                  */
2856                 STAILQ_INSERT_TAIL(&lun->error_list, new_err_desc, links);
2857
2858                 /*
2859                  * XXX KDM check to make sure the serial number is unique,
2860                  * in case we somehow manage to wrap.  That shouldn't
2861                  * happen for a very long time, but it's the right thing to
2862                  * do.
2863                  */
2864                 new_err_desc->serial = lun->error_serial;
2865                 err_desc->serial = lun->error_serial;
2866                 lun->error_serial++;
2867
2868                 mtx_unlock(&lun->lun_lock);
2869                 break;
2870         }
2871         case CTL_ERROR_INJECT_DELETE: {
2872                 struct ctl_error_desc *delete_desc, *desc, *desc2;
2873                 int delete_done;
2874
2875                 delete_desc = (struct ctl_error_desc *)addr;
2876                 delete_done = 0;
2877
2878                 mtx_lock(&softc->ctl_lock);
2879                 lun = softc->ctl_luns[delete_desc->lun_id];
2880                 if (lun == NULL) {
2881                         mtx_unlock(&softc->ctl_lock);
2882                         printf("%s: CTL_ERROR_INJECT_DELETE: invalid LUN %ju\n",
2883                                __func__, (uintmax_t)delete_desc->lun_id);
2884                         retval = EINVAL;
2885                         break;
2886                 }
2887                 mtx_lock(&lun->lun_lock);
2888                 mtx_unlock(&softc->ctl_lock);
2889                 STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
2890                         if (desc->serial != delete_desc->serial)
2891                                 continue;
2892
2893                         STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc,
2894                                       links);
2895                         free(desc, M_CTL);
2896                         delete_done = 1;
2897                 }
2898                 mtx_unlock(&lun->lun_lock);
2899                 if (delete_done == 0) {
2900                         printf("%s: CTL_ERROR_INJECT_DELETE: can't find "
2901                                "error serial %ju on LUN %u\n", __func__,
2902                                delete_desc->serial, delete_desc->lun_id);
2903                         retval = EINVAL;
2904                         break;
2905                 }
2906                 break;
2907         }
2908         case CTL_DUMP_STRUCTS: {
2909                 int i, j, k;
2910                 struct ctl_port *port;
2911                 struct ctl_frontend *fe;
2912
2913                 mtx_lock(&softc->ctl_lock);
2914                 printf("CTL Persistent Reservation information start:\n");
2915                 for (i = 0; i < CTL_MAX_LUNS; i++) {
2916                         lun = softc->ctl_luns[i];
2917
2918                         if ((lun == NULL)
2919                          || ((lun->flags & CTL_LUN_DISABLED) != 0))
2920                                 continue;
2921
2922                         for (j = 0; j < CTL_MAX_PORTS; j++) {
2923                                 if (lun->pr_keys[j] == NULL)
2924                                         continue;
2925                                 for (k = 0; k < CTL_MAX_INIT_PER_PORT; k++){
2926                                         if (lun->pr_keys[j][k] == 0)
2927                                                 continue;
2928                                         printf("  LUN %d port %d iid %d key "
2929                                                "%#jx\n", i, j, k,
2930                                                (uintmax_t)lun->pr_keys[j][k]);
2931                                 }
2932                         }
2933                 }
2934                 printf("CTL Persistent Reservation information end\n");
2935                 printf("CTL Ports:\n");
2936                 STAILQ_FOREACH(port, &softc->port_list, links) {
2937                         printf("  Port %d '%s' Frontend '%s' Type %u pp %d vp %d WWNN "
2938                                "%#jx WWPN %#jx\n", port->targ_port, port->port_name,
2939                                port->frontend->name, port->port_type,
2940                                port->physical_port, port->virtual_port,
2941                                (uintmax_t)port->wwnn, (uintmax_t)port->wwpn);
2942                         for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
2943                                 if (port->wwpn_iid[j].in_use == 0 &&
2944                                     port->wwpn_iid[j].wwpn == 0 &&
2945                                     port->wwpn_iid[j].name == NULL)
2946                                         continue;
2947
2948                                 printf("    iid %u use %d WWPN %#jx '%s'\n",
2949                                     j, port->wwpn_iid[j].in_use,
2950                                     (uintmax_t)port->wwpn_iid[j].wwpn,
2951                                     port->wwpn_iid[j].name);
2952                         }
2953                 }
2954                 printf("CTL Port information end\n");
2955                 mtx_unlock(&softc->ctl_lock);
2956                 /*
2957                  * XXX KDM calling this without a lock.  We'd likely want
2958                  * to drop the lock before calling the frontend's dump
2959                  * routine anyway.
2960                  */
2961                 printf("CTL Frontends:\n");
2962                 STAILQ_FOREACH(fe, &softc->fe_list, links) {
2963                         printf("  Frontend '%s'\n", fe->name);
2964                         if (fe->fe_dump != NULL)
2965                                 fe->fe_dump();
2966                 }
2967                 printf("CTL Frontend information end\n");
2968                 break;
2969         }
2970         case CTL_LUN_REQ: {
2971                 struct ctl_lun_req *lun_req;
2972                 struct ctl_backend_driver *backend;
2973
2974                 lun_req = (struct ctl_lun_req *)addr;
2975
2976                 backend = ctl_backend_find(lun_req->backend);
2977                 if (backend == NULL) {
2978                         lun_req->status = CTL_LUN_ERROR;
2979                         snprintf(lun_req->error_str,
2980                                  sizeof(lun_req->error_str),
2981                                  "Backend \"%s\" not found.",
2982                                  lun_req->backend);
2983                         break;
2984                 }
2985                 if (lun_req->num_be_args > 0) {
2986                         lun_req->kern_be_args = ctl_copyin_args(
2987                                 lun_req->num_be_args,
2988                                 lun_req->be_args,
2989                                 lun_req->error_str,
2990                                 sizeof(lun_req->error_str));
2991                         if (lun_req->kern_be_args == NULL) {
2992                                 lun_req->status = CTL_LUN_ERROR;
2993                                 break;
2994                         }
2995                 }
2996
2997                 retval = backend->ioctl(dev, cmd, addr, flag, td);
2998
2999                 if (lun_req->num_be_args > 0) {
3000                         ctl_copyout_args(lun_req->num_be_args,
3001                                       lun_req->kern_be_args);
3002                         ctl_free_args(lun_req->num_be_args,
3003                                       lun_req->kern_be_args);
3004                 }
3005                 break;
3006         }
3007         case CTL_LUN_LIST: {
3008                 struct sbuf *sb;
3009                 struct ctl_lun_list *list;
3010                 struct ctl_option *opt;
3011
3012                 list = (struct ctl_lun_list *)addr;
3013
3014                 /*
3015                  * Allocate a fixed length sbuf here, based on the length
3016                  * of the user's buffer.  We could allocate an auto-extending
3017                  * buffer, and then tell the user how much larger our
3018                  * amount of data is than his buffer, but that presents
3019                  * some problems:
3020                  *
3021                  * 1.  The sbuf(9) routines use a blocking malloc, and so
3022                  *     we can't hold a lock while calling them with an
3023                  *     auto-extending buffer.
3024                  *
3025                  * 2.  There is not currently a LUN reference counting
3026                  *     mechanism, outside of outstanding transactions on
3027                  *     the LUN's OOA queue.  So a LUN could go away on us
3028                  *     while we're getting the LUN number, backend-specific
3029                  *     information, etc.  Thus, given the way things
3030                  *     currently work, we need to hold the CTL lock while
3031                  *     grabbing LUN information.
3032                  *
3033                  * So, from the user's standpoint, the best thing to do is
3034                  * allocate what he thinks is a reasonable buffer length,
3035                  * and then if he gets a CTL_LUN_LIST_NEED_MORE_SPACE error,
3036                  * double the buffer length and try again.  (And repeat
3037                  * that until he succeeds.)
3038                  */
3039                 sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3040                 if (sb == NULL) {
3041                         list->status = CTL_LUN_LIST_ERROR;
3042                         snprintf(list->error_str, sizeof(list->error_str),
3043                                  "Unable to allocate %d bytes for LUN list",
3044                                  list->alloc_len);
3045                         break;
3046                 }
3047
3048                 sbuf_printf(sb, "<ctllunlist>\n");
3049
3050                 mtx_lock(&softc->ctl_lock);
3051                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
3052                         mtx_lock(&lun->lun_lock);
3053                         retval = sbuf_printf(sb, "<lun id=\"%ju\">\n",
3054                                              (uintmax_t)lun->lun);
3055
3056                         /*
3057                          * Bail out as soon as we see that we've overfilled
3058                          * the buffer.
3059                          */
3060                         if (retval != 0)
3061                                 break;
3062
3063                         retval = sbuf_printf(sb, "\t<backend_type>%s"
3064                                              "</backend_type>\n",
3065                                              (lun->backend == NULL) ?  "none" :
3066                                              lun->backend->name);
3067
3068                         if (retval != 0)
3069                                 break;
3070
3071                         retval = sbuf_printf(sb, "\t<lun_type>%d</lun_type>\n",
3072                                              lun->be_lun->lun_type);
3073
3074                         if (retval != 0)
3075                                 break;
3076
3077                         if (lun->backend == NULL) {
3078                                 retval = sbuf_printf(sb, "</lun>\n");
3079                                 if (retval != 0)
3080                                         break;
3081                                 continue;
3082                         }
3083
3084                         retval = sbuf_printf(sb, "\t<size>%ju</size>\n",
3085                                              (lun->be_lun->maxlba > 0) ?
3086                                              lun->be_lun->maxlba + 1 : 0);
3087
3088                         if (retval != 0)
3089                                 break;
3090
3091                         retval = sbuf_printf(sb, "\t<blocksize>%u</blocksize>\n",
3092                                              lun->be_lun->blocksize);
3093
3094                         if (retval != 0)
3095                                 break;
3096
3097                         retval = sbuf_printf(sb, "\t<serial_number>");
3098
3099                         if (retval != 0)
3100                                 break;
3101
3102                         retval = ctl_sbuf_printf_esc(sb,
3103                             lun->be_lun->serial_num,
3104                             sizeof(lun->be_lun->serial_num));
3105
3106                         if (retval != 0)
3107                                 break;
3108
3109                         retval = sbuf_printf(sb, "</serial_number>\n");
3110
3111                         if (retval != 0)
3112                                 break;
3113
3114                         retval = sbuf_printf(sb, "\t<device_id>");
3115
3116                         if (retval != 0)
3117                                 break;
3118
3119                         retval = ctl_sbuf_printf_esc(sb,
3120                             lun->be_lun->device_id,
3121                             sizeof(lun->be_lun->device_id));
3122
3123                         if (retval != 0)
3124                                 break;
3125
3126                         retval = sbuf_printf(sb, "</device_id>\n");
3127
3128                         if (retval != 0)
3129                                 break;
3130
3131                         if (lun->backend->lun_info != NULL) {
3132                                 retval = lun->backend->lun_info(lun->be_lun->be_lun, sb);
3133                                 if (retval != 0)
3134                                         break;
3135                         }
3136                         STAILQ_FOREACH(opt, &lun->be_lun->options, links) {
3137                                 retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3138                                     opt->name, opt->value, opt->name);
3139                                 if (retval != 0)
3140                                         break;
3141                         }
3142
3143                         retval = sbuf_printf(sb, "</lun>\n");
3144
3145                         if (retval != 0)
3146                                 break;
3147                         mtx_unlock(&lun->lun_lock);
3148                 }
3149                 if (lun != NULL)
3150                         mtx_unlock(&lun->lun_lock);
3151                 mtx_unlock(&softc->ctl_lock);
3152
3153                 if ((retval != 0)
3154                  || ((retval = sbuf_printf(sb, "</ctllunlist>\n")) != 0)) {
3155                         retval = 0;
3156                         sbuf_delete(sb);
3157                         list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3158                         snprintf(list->error_str, sizeof(list->error_str),
3159                                  "Out of space, %d bytes is too small",
3160                                  list->alloc_len);
3161                         break;
3162                 }
3163
3164                 sbuf_finish(sb);
3165
3166                 retval = copyout(sbuf_data(sb), list->lun_xml,
3167                                  sbuf_len(sb) + 1);
3168
3169                 list->fill_len = sbuf_len(sb) + 1;
3170                 list->status = CTL_LUN_LIST_OK;
3171                 sbuf_delete(sb);
3172                 break;
3173         }
3174         case CTL_ISCSI: {
3175                 struct ctl_iscsi *ci;
3176                 struct ctl_frontend *fe;
3177
3178                 ci = (struct ctl_iscsi *)addr;
3179
3180                 fe = ctl_frontend_find("iscsi");
3181                 if (fe == NULL) {
3182                         ci->status = CTL_ISCSI_ERROR;
3183                         snprintf(ci->error_str, sizeof(ci->error_str),
3184                             "Frontend \"iscsi\" not found.");
3185                         break;
3186                 }
3187
3188                 retval = fe->ioctl(dev, cmd, addr, flag, td);
3189                 break;
3190         }
3191         case CTL_PORT_REQ: {
3192                 struct ctl_req *req;
3193                 struct ctl_frontend *fe;
3194
3195                 req = (struct ctl_req *)addr;
3196
3197                 fe = ctl_frontend_find(req->driver);
3198                 if (fe == NULL) {
3199                         req->status = CTL_LUN_ERROR;
3200                         snprintf(req->error_str, sizeof(req->error_str),
3201                             "Frontend \"%s\" not found.", req->driver);
3202                         break;
3203                 }
3204                 if (req->num_args > 0) {
3205                         req->kern_args = ctl_copyin_args(req->num_args,
3206                             req->args, req->error_str, sizeof(req->error_str));
3207                         if (req->kern_args == NULL) {
3208                                 req->status = CTL_LUN_ERROR;
3209                                 break;
3210                         }
3211                 }
3212
3213                 if (fe->ioctl)
3214                         retval = fe->ioctl(dev, cmd, addr, flag, td);
3215                 else
3216                         retval = ENODEV;
3217
3218                 if (req->num_args > 0) {
3219                         ctl_copyout_args(req->num_args, req->kern_args);
3220                         ctl_free_args(req->num_args, req->kern_args);
3221                 }
3222                 break;
3223         }
3224         case CTL_PORT_LIST: {
3225                 struct sbuf *sb;
3226                 struct ctl_port *port;
3227                 struct ctl_lun_list *list;
3228                 struct ctl_option *opt;
3229                 int j;
3230                 uint32_t plun;
3231
3232                 list = (struct ctl_lun_list *)addr;
3233
3234                 sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3235                 if (sb == NULL) {
3236                         list->status = CTL_LUN_LIST_ERROR;
3237                         snprintf(list->error_str, sizeof(list->error_str),
3238                                  "Unable to allocate %d bytes for LUN list",
3239                                  list->alloc_len);
3240                         break;
3241                 }
3242
3243                 sbuf_printf(sb, "<ctlportlist>\n");
3244
3245                 mtx_lock(&softc->ctl_lock);
3246                 STAILQ_FOREACH(port, &softc->port_list, links) {
3247                         retval = sbuf_printf(sb, "<targ_port id=\"%ju\">\n",
3248                                              (uintmax_t)port->targ_port);
3249
3250                         /*
3251                          * Bail out as soon as we see that we've overfilled
3252                          * the buffer.
3253                          */
3254                         if (retval != 0)
3255                                 break;
3256
3257                         retval = sbuf_printf(sb, "\t<frontend_type>%s"
3258                             "</frontend_type>\n", port->frontend->name);
3259                         if (retval != 0)
3260                                 break;
3261
3262                         retval = sbuf_printf(sb, "\t<port_type>%d</port_type>\n",
3263                                              port->port_type);
3264                         if (retval != 0)
3265                                 break;
3266
3267                         retval = sbuf_printf(sb, "\t<online>%s</online>\n",
3268                             (port->status & CTL_PORT_STATUS_ONLINE) ? "YES" : "NO");
3269                         if (retval != 0)
3270                                 break;
3271
3272                         retval = sbuf_printf(sb, "\t<port_name>%s</port_name>\n",
3273                             port->port_name);
3274                         if (retval != 0)
3275                                 break;
3276
3277                         retval = sbuf_printf(sb, "\t<physical_port>%d</physical_port>\n",
3278                             port->physical_port);
3279                         if (retval != 0)
3280                                 break;
3281
3282                         retval = sbuf_printf(sb, "\t<virtual_port>%d</virtual_port>\n",
3283                             port->virtual_port);
3284                         if (retval != 0)
3285                                 break;
3286
3287                         if (port->target_devid != NULL) {
3288                                 sbuf_printf(sb, "\t<target>");
3289                                 ctl_id_sbuf(port->target_devid, sb);
3290                                 sbuf_printf(sb, "</target>\n");
3291                         }
3292
3293                         if (port->port_devid != NULL) {
3294                                 sbuf_printf(sb, "\t<port>");
3295                                 ctl_id_sbuf(port->port_devid, sb);
3296                                 sbuf_printf(sb, "</port>\n");
3297                         }
3298
3299                         if (port->port_info != NULL) {
3300                                 retval = port->port_info(port->onoff_arg, sb);
3301                                 if (retval != 0)
3302                                         break;
3303                         }
3304                         STAILQ_FOREACH(opt, &port->options, links) {
3305                                 retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3306                                     opt->name, opt->value, opt->name);
3307                                 if (retval != 0)
3308                                         break;
3309                         }
3310
3311                         if (port->lun_map != NULL) {
3312                                 sbuf_printf(sb, "\t<lun_map>on</lun_map>\n");
3313                                 for (j = 0; j < CTL_MAX_LUNS; j++) {
3314                                         plun = ctl_lun_map_from_port(port, j);
3315                                         if (plun >= CTL_MAX_LUNS)
3316                                                 continue;
3317                                         sbuf_printf(sb,
3318                                             "\t<lun id=\"%u\">%u</lun>\n",
3319                                             j, plun);
3320                                 }
3321                         }
3322
3323                         for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
3324                                 if (port->wwpn_iid[j].in_use == 0 ||
3325                                     (port->wwpn_iid[j].wwpn == 0 &&
3326                                      port->wwpn_iid[j].name == NULL))
3327                                         continue;
3328
3329                                 if (port->wwpn_iid[j].name != NULL)
3330                                         retval = sbuf_printf(sb,
3331                                             "\t<initiator id=\"%u\">%s</initiator>\n",
3332                                             j, port->wwpn_iid[j].name);
3333                                 else
3334                                         retval = sbuf_printf(sb,
3335                                             "\t<initiator id=\"%u\">naa.%08jx</initiator>\n",
3336                                             j, port->wwpn_iid[j].wwpn);
3337                                 if (retval != 0)
3338                                         break;
3339                         }
3340                         if (retval != 0)
3341                                 break;
3342
3343                         retval = sbuf_printf(sb, "</targ_port>\n");
3344                         if (retval != 0)
3345                                 break;
3346                 }
3347                 mtx_unlock(&softc->ctl_lock);
3348
3349                 if ((retval != 0)
3350                  || ((retval = sbuf_printf(sb, "</ctlportlist>\n")) != 0)) {
3351                         retval = 0;
3352                         sbuf_delete(sb);
3353                         list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3354                         snprintf(list->error_str, sizeof(list->error_str),
3355                                  "Out of space, %d bytes is too small",
3356                                  list->alloc_len);
3357                         break;
3358                 }
3359
3360                 sbuf_finish(sb);
3361
3362                 retval = copyout(sbuf_data(sb), list->lun_xml,
3363                                  sbuf_len(sb) + 1);
3364
3365                 list->fill_len = sbuf_len(sb) + 1;
3366                 list->status = CTL_LUN_LIST_OK;
3367                 sbuf_delete(sb);
3368                 break;
3369         }
3370         case CTL_LUN_MAP: {
3371                 struct ctl_lun_map *lm  = (struct ctl_lun_map *)addr;
3372                 struct ctl_port *port;
3373
3374                 mtx_lock(&softc->ctl_lock);
3375                 if (lm->port < softc->port_min ||
3376                     lm->port >= softc->port_max ||
3377                     (port = softc->ctl_ports[lm->port]) == NULL) {
3378                         mtx_unlock(&softc->ctl_lock);
3379                         return (ENXIO);
3380                 }
3381                 if (port->status & CTL_PORT_STATUS_ONLINE) {
3382                         STAILQ_FOREACH(lun, &softc->lun_list, links) {
3383                                 if (ctl_lun_map_to_port(port, lun->lun) >=
3384                                     CTL_MAX_LUNS)
3385                                         continue;
3386                                 mtx_lock(&lun->lun_lock);
3387                                 ctl_est_ua_port(lun, lm->port, -1,
3388                                     CTL_UA_LUN_CHANGE);
3389                                 mtx_unlock(&lun->lun_lock);
3390                         }
3391                 }
3392                 mtx_unlock(&softc->ctl_lock); // XXX: port_enable sleeps
3393                 if (lm->plun < CTL_MAX_LUNS) {
3394                         if (lm->lun == UINT32_MAX)
3395                                 retval = ctl_lun_map_unset(port, lm->plun);
3396                         else if (lm->lun < CTL_MAX_LUNS &&
3397                             softc->ctl_luns[lm->lun] != NULL)
3398                                 retval = ctl_lun_map_set(port, lm->plun, lm->lun);
3399                         else
3400                                 return (ENXIO);
3401                 } else if (lm->plun == UINT32_MAX) {
3402                         if (lm->lun == UINT32_MAX)
3403                                 retval = ctl_lun_map_deinit(port);
3404                         else
3405                                 retval = ctl_lun_map_init(port);
3406                 } else
3407                         return (ENXIO);
3408                 if (port->status & CTL_PORT_STATUS_ONLINE)
3409                         ctl_isc_announce_port(port);
3410                 break;
3411         }
3412         default: {
3413                 /* XXX KDM should we fix this? */
3414 #if 0
3415                 struct ctl_backend_driver *backend;
3416                 unsigned int type;
3417                 int found;
3418
3419                 found = 0;
3420
3421                 /*
3422                  * We encode the backend type as the ioctl type for backend
3423                  * ioctls.  So parse it out here, and then search for a
3424                  * backend of this type.
3425                  */
3426                 type = _IOC_TYPE(cmd);
3427
3428                 STAILQ_FOREACH(backend, &softc->be_list, links) {
3429                         if (backend->type == type) {
3430                                 found = 1;
3431                                 break;
3432                         }
3433                 }
3434                 if (found == 0) {
3435                         printf("ctl: unknown ioctl command %#lx or backend "
3436                                "%d\n", cmd, type);
3437                         retval = EINVAL;
3438                         break;
3439                 }
3440                 retval = backend->ioctl(dev, cmd, addr, flag, td);
3441 #endif
3442                 retval = ENOTTY;
3443                 break;
3444         }
3445         }
3446         return (retval);
3447 }
3448
3449 uint32_t
3450 ctl_get_initindex(struct ctl_nexus *nexus)
3451 {
3452         return (nexus->initid + (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3453 }
3454
3455 int
3456 ctl_lun_map_init(struct ctl_port *port)
3457 {
3458         struct ctl_softc *softc = port->ctl_softc;
3459         struct ctl_lun *lun;
3460         uint32_t i;
3461
3462         if (port->lun_map == NULL)
3463                 port->lun_map = malloc(sizeof(uint32_t) * CTL_MAX_LUNS,
3464                     M_CTL, M_NOWAIT);
3465         if (port->lun_map == NULL)
3466                 return (ENOMEM);
3467         for (i = 0; i < CTL_MAX_LUNS; i++)
3468                 port->lun_map[i] = UINT32_MAX;
3469         if (port->status & CTL_PORT_STATUS_ONLINE) {
3470                 if (port->lun_disable != NULL) {
3471                         STAILQ_FOREACH(lun, &softc->lun_list, links)
3472                                 port->lun_disable(port->targ_lun_arg, lun->lun);
3473                 }
3474                 ctl_isc_announce_port(port);
3475         }
3476         return (0);
3477 }
3478
3479 int
3480 ctl_lun_map_deinit(struct ctl_port *port)
3481 {
3482         struct ctl_softc *softc = port->ctl_softc;
3483         struct ctl_lun *lun;
3484
3485         if (port->lun_map == NULL)
3486                 return (0);
3487         free(port->lun_map, M_CTL);
3488         port->lun_map = NULL;
3489         if (port->status & CTL_PORT_STATUS_ONLINE) {
3490                 if (port->lun_enable != NULL) {
3491                         STAILQ_FOREACH(lun, &softc->lun_list, links)
3492                                 port->lun_enable(port->targ_lun_arg, lun->lun);
3493                 }
3494                 ctl_isc_announce_port(port);
3495         }
3496         return (0);
3497 }
3498
3499 int
3500 ctl_lun_map_set(struct ctl_port *port, uint32_t plun, uint32_t glun)
3501 {
3502         int status;
3503         uint32_t old;
3504
3505         if (port->lun_map == NULL) {
3506                 status = ctl_lun_map_init(port);
3507                 if (status != 0)
3508                         return (status);
3509         }
3510         old = port->lun_map[plun];
3511         port->lun_map[plun] = glun;
3512         if ((port->status & CTL_PORT_STATUS_ONLINE) && old >= CTL_MAX_LUNS) {
3513                 if (port->lun_enable != NULL)
3514                         port->lun_enable(port->targ_lun_arg, plun);
3515                 ctl_isc_announce_port(port);
3516         }
3517         return (0);
3518 }
3519
3520 int
3521 ctl_lun_map_unset(struct ctl_port *port, uint32_t plun)
3522 {
3523         uint32_t old;
3524
3525         if (port->lun_map == NULL)
3526                 return (0);
3527         old = port->lun_map[plun];
3528         port->lun_map[plun] = UINT32_MAX;
3529         if ((port->status & CTL_PORT_STATUS_ONLINE) && old < CTL_MAX_LUNS) {
3530                 if (port->lun_disable != NULL)
3531                         port->lun_disable(port->targ_lun_arg, plun);
3532                 ctl_isc_announce_port(port);
3533         }
3534         return (0);
3535 }
3536
3537 uint32_t
3538 ctl_lun_map_from_port(struct ctl_port *port, uint32_t lun_id)
3539 {
3540
3541         if (port == NULL)
3542                 return (UINT32_MAX);
3543         if (port->lun_map == NULL || lun_id >= CTL_MAX_LUNS)
3544                 return (lun_id);
3545         return (port->lun_map[lun_id]);
3546 }
3547
3548 uint32_t
3549 ctl_lun_map_to_port(struct ctl_port *port, uint32_t lun_id)
3550 {
3551         uint32_t i;
3552
3553         if (port == NULL)
3554                 return (UINT32_MAX);
3555         if (port->lun_map == NULL)
3556                 return (lun_id);
3557         for (i = 0; i < CTL_MAX_LUNS; i++) {
3558                 if (port->lun_map[i] == lun_id)
3559                         return (i);
3560         }
3561         return (UINT32_MAX);
3562 }
3563
3564 uint32_t
3565 ctl_decode_lun(uint64_t encoded)
3566 {
3567         uint8_t lun[8];
3568         uint32_t result = 0xffffffff;
3569
3570         be64enc(lun, encoded);
3571         switch (lun[0] & RPL_LUNDATA_ATYP_MASK) {
3572         case RPL_LUNDATA_ATYP_PERIPH:
3573                 if ((lun[0] & 0x3f) == 0 && lun[2] == 0 && lun[3] == 0 &&
3574                     lun[4] == 0 && lun[5] == 0 && lun[6] == 0 && lun[7] == 0)
3575                         result = lun[1];
3576                 break;
3577         case RPL_LUNDATA_ATYP_FLAT:
3578                 if (lun[2] == 0 && lun[3] == 0 && lun[4] == 0 && lun[5] == 0 &&
3579                     lun[6] == 0 && lun[7] == 0)
3580                         result = ((lun[0] & 0x3f) << 8) + lun[1];
3581                 break;
3582         case RPL_LUNDATA_ATYP_EXTLUN:
3583                 switch (lun[0] & RPL_LUNDATA_EXT_EAM_MASK) {
3584                 case 0x02:
3585                         switch (lun[0] & RPL_LUNDATA_EXT_LEN_MASK) {
3586                         case 0x00:
3587                                 result = lun[1];
3588                                 break;
3589                         case 0x10:
3590                                 result = (lun[1] << 16) + (lun[2] << 8) +
3591                                     lun[3];
3592                                 break;
3593                         case 0x20:
3594                                 if (lun[1] == 0 && lun[6] == 0 && lun[7] == 0)
3595                                         result = (lun[2] << 24) +
3596                                             (lun[3] << 16) + (lun[4] << 8) +
3597                                             lun[5];
3598                                 break;
3599                         }
3600                         break;
3601                 case RPL_LUNDATA_EXT_EAM_NOT_SPEC:
3602                         result = 0xffffffff;
3603                         break;
3604                 }
3605                 break;
3606         }
3607         return (result);
3608 }
3609
3610 uint64_t
3611 ctl_encode_lun(uint32_t decoded)
3612 {
3613         uint64_t l = decoded;
3614
3615         if (l <= 0xff)
3616                 return (((uint64_t)RPL_LUNDATA_ATYP_PERIPH << 56) | (l << 48));
3617         if (l <= 0x3fff)
3618                 return (((uint64_t)RPL_LUNDATA_ATYP_FLAT << 56) | (l << 48));
3619         if (l <= 0xffffff)
3620                 return (((uint64_t)(RPL_LUNDATA_ATYP_EXTLUN | 0x12) << 56) |
3621                     (l << 32));
3622         return ((((uint64_t)RPL_LUNDATA_ATYP_EXTLUN | 0x22) << 56) | (l << 16));
3623 }
3624
3625 static struct ctl_port *
3626 ctl_io_port(struct ctl_io_hdr *io_hdr)
3627 {
3628
3629         return (control_softc->ctl_ports[io_hdr->nexus.targ_port]);
3630 }
3631
3632 int
3633 ctl_ffz(uint32_t *mask, uint32_t first, uint32_t last)
3634 {
3635         int i;
3636
3637         for (i = first; i < last; i++) {
3638                 if ((mask[i / 32] & (1 << (i % 32))) == 0)
3639                         return (i);
3640         }
3641         return (-1);
3642 }
3643
3644 int
3645 ctl_set_mask(uint32_t *mask, uint32_t bit)
3646 {
3647         uint32_t chunk, piece;
3648
3649         chunk = bit >> 5;
3650         piece = bit % (sizeof(uint32_t) * 8);
3651
3652         if ((mask[chunk] & (1 << piece)) != 0)
3653                 return (-1);
3654         else
3655                 mask[chunk] |= (1 << piece);
3656
3657         return (0);
3658 }
3659
3660 int
3661 ctl_clear_mask(uint32_t *mask, uint32_t bit)
3662 {
3663         uint32_t chunk, piece;
3664
3665         chunk = bit >> 5;
3666         piece = bit % (sizeof(uint32_t) * 8);
3667
3668         if ((mask[chunk] & (1 << piece)) == 0)
3669                 return (-1);
3670         else
3671                 mask[chunk] &= ~(1 << piece);
3672
3673         return (0);
3674 }
3675
3676 int
3677 ctl_is_set(uint32_t *mask, uint32_t bit)
3678 {
3679         uint32_t chunk, piece;
3680
3681         chunk = bit >> 5;
3682         piece = bit % (sizeof(uint32_t) * 8);
3683
3684         if ((mask[chunk] & (1 << piece)) == 0)
3685                 return (0);
3686         else
3687                 return (1);
3688 }
3689
3690 static uint64_t
3691 ctl_get_prkey(struct ctl_lun *lun, uint32_t residx)
3692 {
3693         uint64_t *t;
3694
3695         t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3696         if (t == NULL)
3697                 return (0);
3698         return (t[residx % CTL_MAX_INIT_PER_PORT]);
3699 }
3700
3701 static void
3702 ctl_clr_prkey(struct ctl_lun *lun, uint32_t residx)
3703 {
3704         uint64_t *t;
3705
3706         t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3707         if (t == NULL)
3708                 return;
3709         t[residx % CTL_MAX_INIT_PER_PORT] = 0;
3710 }
3711
3712 static void
3713 ctl_alloc_prkey(struct ctl_lun *lun, uint32_t residx)
3714 {
3715         uint64_t *p;
3716         u_int i;
3717
3718         i = residx/CTL_MAX_INIT_PER_PORT;
3719         if (lun->pr_keys[i] != NULL)
3720                 return;
3721         mtx_unlock(&lun->lun_lock);
3722         p = malloc(sizeof(uint64_t) * CTL_MAX_INIT_PER_PORT, M_CTL,
3723             M_WAITOK | M_ZERO);
3724         mtx_lock(&lun->lun_lock);
3725         if (lun->pr_keys[i] == NULL)
3726                 lun->pr_keys[i] = p;
3727         else
3728                 free(p, M_CTL);
3729 }
3730
3731 static void
3732 ctl_set_prkey(struct ctl_lun *lun, uint32_t residx, uint64_t key)
3733 {
3734         uint64_t *t;
3735
3736         t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3737         KASSERT(t != NULL, ("prkey %d is not allocated", residx));
3738         t[residx % CTL_MAX_INIT_PER_PORT] = key;
3739 }
3740
3741 /*
3742  * ctl_softc, pool_name, total_ctl_io are passed in.
3743  * npool is passed out.
3744  */
3745 int
3746 ctl_pool_create(struct ctl_softc *ctl_softc, const char *pool_name,
3747                 uint32_t total_ctl_io, void **npool)
3748 {
3749 #ifdef IO_POOLS
3750         struct ctl_io_pool *pool;
3751
3752         pool = (struct ctl_io_pool *)malloc(sizeof(*pool), M_CTL,
3753                                             M_NOWAIT | M_ZERO);
3754         if (pool == NULL)
3755                 return (ENOMEM);
3756
3757         snprintf(pool->name, sizeof(pool->name), "CTL IO %s", pool_name);
3758         pool->ctl_softc = ctl_softc;
3759         pool->zone = uma_zsecond_create(pool->name, NULL,
3760             NULL, NULL, NULL, ctl_softc->io_zone);
3761         /* uma_prealloc(pool->zone, total_ctl_io); */
3762
3763         *npool = pool;
3764 #else
3765         *npool = ctl_softc->io_zone;
3766 #endif
3767         return (0);
3768 }
3769
3770 void
3771 ctl_pool_free(struct ctl_io_pool *pool)
3772 {
3773
3774         if (pool == NULL)
3775                 return;
3776
3777 #ifdef IO_POOLS
3778         uma_zdestroy(pool->zone);
3779         free(pool, M_CTL);
3780 #endif
3781 }
3782
3783 union ctl_io *
3784 ctl_alloc_io(void *pool_ref)
3785 {
3786         union ctl_io *io;
3787 #ifdef IO_POOLS
3788         struct ctl_io_pool *pool = (struct ctl_io_pool *)pool_ref;
3789
3790         io = uma_zalloc(pool->zone, M_WAITOK);
3791 #else
3792         io = uma_zalloc((uma_zone_t)pool_ref, M_WAITOK);
3793 #endif
3794         if (io != NULL)
3795                 io->io_hdr.pool = pool_ref;
3796         return (io);
3797 }
3798
3799 union ctl_io *
3800 ctl_alloc_io_nowait(void *pool_ref)
3801 {
3802         union ctl_io *io;
3803 #ifdef IO_POOLS
3804         struct ctl_io_pool *pool = (struct ctl_io_pool *)pool_ref;
3805
3806         io = uma_zalloc(pool->zone, M_NOWAIT);
3807 #else
3808         io = uma_zalloc((uma_zone_t)pool_ref, M_NOWAIT);
3809 #endif
3810         if (io != NULL)
3811                 io->io_hdr.pool = pool_ref;
3812         return (io);
3813 }
3814
3815 void
3816 ctl_free_io(union ctl_io *io)
3817 {
3818 #ifdef IO_POOLS
3819         struct ctl_io_pool *pool;
3820 #endif
3821
3822         if (io == NULL)
3823                 return;
3824
3825 #ifdef IO_POOLS
3826         pool = (struct ctl_io_pool *)io->io_hdr.pool;
3827         uma_zfree(pool->zone, io);
3828 #else
3829         uma_zfree((uma_zone_t)io->io_hdr.pool, io);
3830 #endif
3831 }
3832
3833 void
3834 ctl_zero_io(union ctl_io *io)
3835 {
3836         void *pool_ref;
3837
3838         if (io == NULL)
3839                 return;
3840
3841         /*
3842          * May need to preserve linked list pointers at some point too.
3843          */
3844         pool_ref = io->io_hdr.pool;
3845         memset(io, 0, sizeof(*io));
3846         io->io_hdr.pool = pool_ref;
3847 }
3848
3849 int
3850 ctl_expand_number(const char *buf, uint64_t *num)
3851 {
3852         char *endptr;
3853         uint64_t number;
3854         unsigned shift;
3855
3856         number = strtoq(buf, &endptr, 0);
3857
3858         switch (tolower((unsigned char)*endptr)) {
3859         case 'e':
3860                 shift = 60;
3861                 break;
3862         case 'p':
3863                 shift = 50;
3864                 break;
3865         case 't':
3866                 shift = 40;
3867                 break;
3868         case 'g':
3869                 shift = 30;
3870                 break;
3871         case 'm':
3872                 shift = 20;
3873                 break;
3874         case 'k':
3875                 shift = 10;
3876                 break;
3877         case 'b':
3878         case '\0': /* No unit. */
3879                 *num = number;
3880                 return (0);
3881         default:
3882                 /* Unrecognized unit. */
3883                 return (-1);
3884         }
3885
3886         if ((number << shift) >> shift != number) {
3887                 /* Overflow */
3888                 return (-1);
3889         }
3890         *num = number << shift;
3891         return (0);
3892 }
3893
3894
3895 /*
3896  * This routine could be used in the future to load default and/or saved
3897  * mode page parameters for a particuar lun.
3898  */
3899 static int
3900 ctl_init_page_index(struct ctl_lun *lun)
3901 {
3902         int i, page_code;
3903         struct ctl_page_index *page_index;
3904         const char *value;
3905         uint64_t ival;
3906
3907         memcpy(&lun->mode_pages.index, page_index_template,
3908                sizeof(page_index_template));
3909
3910         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
3911
3912                 page_index = &lun->mode_pages.index[i];
3913                 if (lun->be_lun->lun_type == T_DIRECT &&
3914                     (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
3915                         continue;
3916                 if (lun->be_lun->lun_type == T_PROCESSOR &&
3917                     (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
3918                         continue;
3919                 if (lun->be_lun->lun_type == T_CDROM &&
3920                     (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
3921                         continue;
3922
3923                 page_code = page_index->page_code & SMPH_PC_MASK;
3924                 switch (page_code) {
3925                 case SMS_RW_ERROR_RECOVERY_PAGE: {
3926                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
3927                             ("subpage %#x for page %#x is incorrect!",
3928                             page_index->subpage, page_code));
3929                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_CURRENT],
3930                                &rw_er_page_default,
3931                                sizeof(rw_er_page_default));
3932                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_CHANGEABLE],
3933                                &rw_er_page_changeable,
3934                                sizeof(rw_er_page_changeable));
3935                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_DEFAULT],
3936                                &rw_er_page_default,
3937                                sizeof(rw_er_page_default));
3938                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_SAVED],
3939                                &rw_er_page_default,
3940                                sizeof(rw_er_page_default));
3941                         page_index->page_data =
3942                                 (uint8_t *)lun->mode_pages.rw_er_page;
3943                         break;
3944                 }
3945                 case SMS_FORMAT_DEVICE_PAGE: {
3946                         struct scsi_format_page *format_page;
3947
3948                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
3949                             ("subpage %#x for page %#x is incorrect!",
3950                             page_index->subpage, page_code));
3951
3952                         /*
3953                          * Sectors per track are set above.  Bytes per
3954                          * sector need to be set here on a per-LUN basis.
3955                          */
3956                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_CURRENT],
3957                                &format_page_default,
3958                                sizeof(format_page_default));
3959                         memcpy(&lun->mode_pages.format_page[
3960                                CTL_PAGE_CHANGEABLE], &format_page_changeable,
3961                                sizeof(format_page_changeable));
3962                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_DEFAULT],
3963                                &format_page_default,
3964                                sizeof(format_page_default));
3965                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_SAVED],
3966                                &format_page_default,
3967                                sizeof(format_page_default));
3968
3969                         format_page = &lun->mode_pages.format_page[
3970                                 CTL_PAGE_CURRENT];
3971                         scsi_ulto2b(lun->be_lun->blocksize,
3972                                     format_page->bytes_per_sector);
3973
3974                         format_page = &lun->mode_pages.format_page[
3975                                 CTL_PAGE_DEFAULT];
3976                         scsi_ulto2b(lun->be_lun->blocksize,
3977                                     format_page->bytes_per_sector);
3978
3979                         format_page = &lun->mode_pages.format_page[
3980                                 CTL_PAGE_SAVED];
3981                         scsi_ulto2b(lun->be_lun->blocksize,
3982                                     format_page->bytes_per_sector);
3983
3984                         page_index->page_data =
3985                                 (uint8_t *)lun->mode_pages.format_page;
3986                         break;
3987                 }
3988                 case SMS_RIGID_DISK_PAGE: {
3989                         struct scsi_rigid_disk_page *rigid_disk_page;
3990                         uint32_t sectors_per_cylinder;
3991                         uint64_t cylinders;
3992 #ifndef __XSCALE__
3993                         int shift;
3994 #endif /* !__XSCALE__ */
3995
3996                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
3997                             ("subpage %#x for page %#x is incorrect!",
3998                             page_index->subpage, page_code));
3999
4000                         /*
4001                          * Rotation rate and sectors per track are set
4002                          * above.  We calculate the cylinders here based on
4003                          * capacity.  Due to the number of heads and
4004                          * sectors per track we're using, smaller arrays
4005                          * may turn out to have 0 cylinders.  Linux and
4006                          * FreeBSD don't pay attention to these mode pages
4007                          * to figure out capacity, but Solaris does.  It
4008                          * seems to deal with 0 cylinders just fine, and
4009                          * works out a fake geometry based on the capacity.
4010                          */
4011                         memcpy(&lun->mode_pages.rigid_disk_page[
4012                                CTL_PAGE_DEFAULT], &rigid_disk_page_default,
4013                                sizeof(rigid_disk_page_default));
4014                         memcpy(&lun->mode_pages.rigid_disk_page[
4015                                CTL_PAGE_CHANGEABLE],&rigid_disk_page_changeable,
4016                                sizeof(rigid_disk_page_changeable));
4017
4018                         sectors_per_cylinder = CTL_DEFAULT_SECTORS_PER_TRACK *
4019                                 CTL_DEFAULT_HEADS;
4020
4021                         /*
4022                          * The divide method here will be more accurate,
4023                          * probably, but results in floating point being
4024                          * used in the kernel on i386 (__udivdi3()).  On the
4025                          * XScale, though, __udivdi3() is implemented in
4026                          * software.
4027                          *
4028                          * The shift method for cylinder calculation is
4029                          * accurate if sectors_per_cylinder is a power of
4030                          * 2.  Otherwise it might be slightly off -- you
4031                          * might have a bit of a truncation problem.
4032                          */
4033 #ifdef  __XSCALE__
4034                         cylinders = (lun->be_lun->maxlba + 1) /
4035                                 sectors_per_cylinder;
4036 #else
4037                         for (shift = 31; shift > 0; shift--) {
4038                                 if (sectors_per_cylinder & (1 << shift))
4039                                         break;
4040                         }
4041                         cylinders = (lun->be_lun->maxlba + 1) >> shift;
4042 #endif
4043
4044                         /*
4045                          * We've basically got 3 bytes, or 24 bits for the
4046                          * cylinder size in the mode page.  If we're over,
4047                          * just round down to 2^24.
4048                          */
4049                         if (cylinders > 0xffffff)
4050                                 cylinders = 0xffffff;
4051
4052                         rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4053                                 CTL_PAGE_DEFAULT];
4054                         scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4055
4056                         if ((value = ctl_get_opt(&lun->be_lun->options,
4057                             "rpm")) != NULL) {
4058                                 scsi_ulto2b(strtol(value, NULL, 0),
4059                                      rigid_disk_page->rotation_rate);
4060                         }
4061
4062                         memcpy(&lun->mode_pages.rigid_disk_page[CTL_PAGE_CURRENT],
4063                                &lun->mode_pages.rigid_disk_page[CTL_PAGE_DEFAULT],
4064                                sizeof(rigid_disk_page_default));
4065                         memcpy(&lun->mode_pages.rigid_disk_page[CTL_PAGE_SAVED],
4066                                &lun->mode_pages.rigid_disk_page[CTL_PAGE_DEFAULT],
4067                                sizeof(rigid_disk_page_default));
4068
4069                         page_index->page_data =
4070                                 (uint8_t *)lun->mode_pages.rigid_disk_page;
4071                         break;
4072                 }
4073                 case SMS_CACHING_PAGE: {
4074                         struct scsi_caching_page *caching_page;
4075
4076                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4077                             ("subpage %#x for page %#x is incorrect!",
4078                             page_index->subpage, page_code));
4079                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_DEFAULT],
4080                                &caching_page_default,
4081                                sizeof(caching_page_default));
4082                         memcpy(&lun->mode_pages.caching_page[
4083                                CTL_PAGE_CHANGEABLE], &caching_page_changeable,
4084                                sizeof(caching_page_changeable));
4085                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4086                                &caching_page_default,
4087                                sizeof(caching_page_default));
4088                         caching_page = &lun->mode_pages.caching_page[
4089                             CTL_PAGE_SAVED];
4090                         value = ctl_get_opt(&lun->be_lun->options, "writecache");
4091                         if (value != NULL && strcmp(value, "off") == 0)
4092                                 caching_page->flags1 &= ~SCP_WCE;
4093                         value = ctl_get_opt(&lun->be_lun->options, "readcache");
4094                         if (value != NULL && strcmp(value, "off") == 0)
4095                                 caching_page->flags1 |= SCP_RCD;
4096                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_CURRENT],
4097                                &lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4098                                sizeof(caching_page_default));
4099                         page_index->page_data =
4100                                 (uint8_t *)lun->mode_pages.caching_page;
4101                         break;
4102                 }
4103                 case SMS_CONTROL_MODE_PAGE: {
4104                         switch (page_index->subpage) {
4105                         case SMS_SUBPAGE_PAGE_0: {
4106                                 struct scsi_control_page *control_page;
4107
4108                                 memcpy(&lun->mode_pages.control_page[
4109                                     CTL_PAGE_DEFAULT],
4110                                        &control_page_default,
4111                                        sizeof(control_page_default));
4112                                 memcpy(&lun->mode_pages.control_page[
4113                                     CTL_PAGE_CHANGEABLE],
4114                                        &control_page_changeable,
4115                                        sizeof(control_page_changeable));
4116                                 memcpy(&lun->mode_pages.control_page[
4117                                     CTL_PAGE_SAVED],
4118                                        &control_page_default,
4119                                        sizeof(control_page_default));
4120                                 control_page = &lun->mode_pages.control_page[
4121                                     CTL_PAGE_SAVED];
4122                                 value = ctl_get_opt(&lun->be_lun->options,
4123                                     "reordering");
4124                                 if (value != NULL &&
4125                                     strcmp(value, "unrestricted") == 0) {
4126                                         control_page->queue_flags &=
4127                                             ~SCP_QUEUE_ALG_MASK;
4128                                         control_page->queue_flags |=
4129                                             SCP_QUEUE_ALG_UNRESTRICTED;
4130                                 }
4131                                 memcpy(&lun->mode_pages.control_page[
4132                                     CTL_PAGE_CURRENT],
4133                                        &lun->mode_pages.control_page[
4134                                     CTL_PAGE_SAVED],
4135                                        sizeof(control_page_default));
4136                                 page_index->page_data =
4137                                     (uint8_t *)lun->mode_pages.control_page;
4138                                 break;
4139                         }
4140                         case 0x01:
4141                                 memcpy(&lun->mode_pages.control_ext_page[
4142                                     CTL_PAGE_DEFAULT],
4143                                        &control_ext_page_default,
4144                                        sizeof(control_ext_page_default));
4145                                 memcpy(&lun->mode_pages.control_ext_page[
4146                                     CTL_PAGE_CHANGEABLE],
4147                                        &control_ext_page_changeable,
4148                                        sizeof(control_ext_page_changeable));
4149                                 memcpy(&lun->mode_pages.control_ext_page[
4150                                     CTL_PAGE_SAVED],
4151                                        &control_ext_page_default,
4152                                        sizeof(control_ext_page_default));
4153                                 memcpy(&lun->mode_pages.control_ext_page[
4154                                     CTL_PAGE_CURRENT],
4155                                        &lun->mode_pages.control_ext_page[
4156                                     CTL_PAGE_SAVED],
4157                                        sizeof(control_ext_page_default));
4158                                 page_index->page_data =
4159                                     (uint8_t *)lun->mode_pages.control_ext_page;
4160                                 break;
4161                         default:
4162                                 panic("subpage %#x for page %#x is incorrect!",
4163                                       page_index->subpage, page_code);
4164                         }
4165                         break;
4166                 }
4167                 case SMS_INFO_EXCEPTIONS_PAGE: {
4168                         switch (page_index->subpage) {
4169                         case SMS_SUBPAGE_PAGE_0:
4170                                 memcpy(&lun->mode_pages.ie_page[CTL_PAGE_CURRENT],
4171                                        &ie_page_default,
4172                                        sizeof(ie_page_default));
4173                                 memcpy(&lun->mode_pages.ie_page[
4174                                        CTL_PAGE_CHANGEABLE], &ie_page_changeable,
4175                                        sizeof(ie_page_changeable));
4176                                 memcpy(&lun->mode_pages.ie_page[CTL_PAGE_DEFAULT],
4177                                        &ie_page_default,
4178                                        sizeof(ie_page_default));
4179                                 memcpy(&lun->mode_pages.ie_page[CTL_PAGE_SAVED],
4180                                        &ie_page_default,
4181                                        sizeof(ie_page_default));
4182                                 page_index->page_data =
4183                                         (uint8_t *)lun->mode_pages.ie_page;
4184                                 break;
4185                         case 0x02: {
4186                                 struct ctl_logical_block_provisioning_page *page;
4187
4188                                 memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_DEFAULT],
4189                                        &lbp_page_default,
4190                                        sizeof(lbp_page_default));
4191                                 memcpy(&lun->mode_pages.lbp_page[
4192                                        CTL_PAGE_CHANGEABLE], &lbp_page_changeable,
4193                                        sizeof(lbp_page_changeable));
4194                                 memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_SAVED],
4195                                        &lbp_page_default,
4196                                        sizeof(lbp_page_default));
4197                                 page = &lun->mode_pages.lbp_page[CTL_PAGE_SAVED];
4198                                 value = ctl_get_opt(&lun->be_lun->options,
4199                                     "avail-threshold");
4200                                 if (value != NULL &&
4201                                     ctl_expand_number(value, &ival) == 0) {
4202                                         page->descr[0].flags |= SLBPPD_ENABLED |
4203                                             SLBPPD_ARMING_DEC;
4204                                         if (lun->be_lun->blocksize)
4205                                                 ival /= lun->be_lun->blocksize;
4206                                         else
4207                                                 ival /= 512;
4208                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4209                                             page->descr[0].count);
4210                                 }
4211                                 value = ctl_get_opt(&lun->be_lun->options,
4212                                     "used-threshold");
4213                                 if (value != NULL &&
4214                                     ctl_expand_number(value, &ival) == 0) {
4215                                         page->descr[1].flags |= SLBPPD_ENABLED |
4216                                             SLBPPD_ARMING_INC;
4217                                         if (lun->be_lun->blocksize)
4218                                                 ival /= lun->be_lun->blocksize;
4219                                         else
4220                                                 ival /= 512;
4221                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4222                                             page->descr[1].count);
4223                                 }
4224                                 value = ctl_get_opt(&lun->be_lun->options,
4225                                     "pool-avail-threshold");
4226                                 if (value != NULL &&
4227                                     ctl_expand_number(value, &ival) == 0) {
4228                                         page->descr[2].flags |= SLBPPD_ENABLED |
4229                                             SLBPPD_ARMING_DEC;
4230                                         if (lun->be_lun->blocksize)
4231                                                 ival /= lun->be_lun->blocksize;
4232                                         else
4233                                                 ival /= 512;
4234                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4235                                             page->descr[2].count);
4236                                 }
4237                                 value = ctl_get_opt(&lun->be_lun->options,
4238                                     "pool-used-threshold");
4239                                 if (value != NULL &&
4240                                     ctl_expand_number(value, &ival) == 0) {
4241                                         page->descr[3].flags |= SLBPPD_ENABLED |
4242                                             SLBPPD_ARMING_INC;
4243                                         if (lun->be_lun->blocksize)
4244                                                 ival /= lun->be_lun->blocksize;
4245                                         else
4246                                                 ival /= 512;
4247                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4248                                             page->descr[3].count);
4249                                 }
4250                                 memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_CURRENT],
4251                                        &lun->mode_pages.lbp_page[CTL_PAGE_SAVED],
4252                                        sizeof(lbp_page_default));
4253                                 page_index->page_data =
4254                                         (uint8_t *)lun->mode_pages.lbp_page;
4255                                 break;
4256                         }
4257                         default:
4258                                 panic("subpage %#x for page %#x is incorrect!",
4259                                       page_index->subpage, page_code);
4260                         }
4261                         break;
4262                 }
4263                 case SMS_CDDVD_CAPS_PAGE:{
4264                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4265                             ("subpage %#x for page %#x is incorrect!",
4266                             page_index->subpage, page_code));
4267                         memcpy(&lun->mode_pages.cddvd_page[CTL_PAGE_DEFAULT],
4268                                &cddvd_page_default,
4269                                sizeof(cddvd_page_default));
4270                         memcpy(&lun->mode_pages.cddvd_page[
4271                                CTL_PAGE_CHANGEABLE], &cddvd_page_changeable,
4272                                sizeof(cddvd_page_changeable));
4273                         memcpy(&lun->mode_pages.cddvd_page[CTL_PAGE_SAVED],
4274                                &cddvd_page_default,
4275                                sizeof(cddvd_page_default));
4276                         memcpy(&lun->mode_pages.cddvd_page[CTL_PAGE_CURRENT],
4277                                &lun->mode_pages.cddvd_page[CTL_PAGE_SAVED],
4278                                sizeof(cddvd_page_default));
4279                         page_index->page_data =
4280                                 (uint8_t *)lun->mode_pages.cddvd_page;
4281                         break;
4282                 }
4283                 case SMS_VENDOR_SPECIFIC_PAGE:{
4284                         switch (page_index->subpage) {
4285                         case DBGCNF_SUBPAGE_CODE: {
4286                                 memcpy(&lun->mode_pages.debugconf_subpage[
4287                                        CTL_PAGE_CURRENT],
4288                                        &debugconf_page_default,
4289                                        sizeof(debugconf_page_default));
4290                                 memcpy(&lun->mode_pages.debugconf_subpage[
4291                                        CTL_PAGE_CHANGEABLE],
4292                                        &debugconf_page_changeable,
4293                                        sizeof(debugconf_page_changeable));
4294                                 memcpy(&lun->mode_pages.debugconf_subpage[
4295                                        CTL_PAGE_DEFAULT],
4296                                        &debugconf_page_default,
4297                                        sizeof(debugconf_page_default));
4298                                 memcpy(&lun->mode_pages.debugconf_subpage[
4299                                        CTL_PAGE_SAVED],
4300                                        &debugconf_page_default,
4301                                        sizeof(debugconf_page_default));
4302                                 page_index->page_data =
4303                                     (uint8_t *)lun->mode_pages.debugconf_subpage;
4304                                 break;
4305                         }
4306                         default:
4307                                 panic("subpage %#x for page %#x is incorrect!",
4308                                       page_index->subpage, page_code);
4309                         }
4310                         break;
4311                 }
4312                 default:
4313                         panic("invalid page code value %#x", page_code);
4314                 }
4315         }
4316
4317         return (CTL_RETVAL_COMPLETE);
4318 }
4319
4320 static int
4321 ctl_init_log_page_index(struct ctl_lun *lun)
4322 {
4323         struct ctl_page_index *page_index;
4324         int i, j, k, prev;
4325
4326         memcpy(&lun->log_pages.index, log_page_index_template,
4327                sizeof(log_page_index_template));
4328
4329         prev = -1;
4330         for (i = 0, j = 0, k = 0; i < CTL_NUM_LOG_PAGES; i++) {
4331
4332                 page_index = &lun->log_pages.index[i];
4333                 if (lun->be_lun->lun_type == T_DIRECT &&
4334                     (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
4335                         continue;
4336                 if (lun->be_lun->lun_type == T_PROCESSOR &&
4337                     (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
4338                         continue;
4339                 if (lun->be_lun->lun_type == T_CDROM &&
4340                     (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
4341                         continue;
4342
4343                 if (page_index->page_code == SLS_LOGICAL_BLOCK_PROVISIONING &&
4344                     lun->backend->lun_attr == NULL)
4345                         continue;
4346
4347                 if (page_index->page_code != prev) {
4348                         lun->log_pages.pages_page[j] = page_index->page_code;
4349                         prev = page_index->page_code;
4350                         j++;
4351                 }
4352                 lun->log_pages.subpages_page[k*2] = page_index->page_code;
4353                 lun->log_pages.subpages_page[k*2+1] = page_index->subpage;
4354                 k++;
4355         }
4356         lun->log_pages.index[0].page_data = &lun->log_pages.pages_page[0];
4357         lun->log_pages.index[0].page_len = j;
4358         lun->log_pages.index[1].page_data = &lun->log_pages.subpages_page[0];
4359         lun->log_pages.index[1].page_len = k * 2;
4360         lun->log_pages.index[2].page_data = &lun->log_pages.lbp_page[0];
4361         lun->log_pages.index[2].page_len = 12*CTL_NUM_LBP_PARAMS;
4362         lun->log_pages.index[3].page_data = (uint8_t *)&lun->log_pages.stat_page;
4363         lun->log_pages.index[3].page_len = sizeof(lun->log_pages.stat_page);
4364
4365         return (CTL_RETVAL_COMPLETE);
4366 }
4367
4368 static int
4369 hex2bin(const char *str, uint8_t *buf, int buf_size)
4370 {
4371         int i;
4372         u_char c;
4373
4374         memset(buf, 0, buf_size);
4375         while (isspace(str[0]))
4376                 str++;
4377         if (str[0] == '0' && (str[1] == 'x' || str[1] == 'X'))
4378                 str += 2;
4379         buf_size *= 2;
4380         for (i = 0; str[i] != 0 && i < buf_size; i++) {
4381                 c = str[i];
4382                 if (isdigit(c))
4383                         c -= '0';
4384                 else if (isalpha(c))
4385                         c -= isupper(c) ? 'A' - 10 : 'a' - 10;
4386                 else
4387                         break;
4388                 if (c >= 16)
4389                         break;
4390                 if ((i & 1) == 0)
4391                         buf[i / 2] |= (c << 4);
4392                 else
4393                         buf[i / 2] |= c;
4394         }
4395         return ((i + 1) / 2);
4396 }
4397
4398 /*
4399  * LUN allocation.
4400  *
4401  * Requirements:
4402  * - caller allocates and zeros LUN storage, or passes in a NULL LUN if he
4403  *   wants us to allocate the LUN and he can block.
4404  * - ctl_softc is always set
4405  * - be_lun is set if the LUN has a backend (needed for disk LUNs)
4406  *
4407  * Returns 0 for success, non-zero (errno) for failure.
4408  */
4409 static int
4410 ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *ctl_lun,
4411               struct ctl_be_lun *const be_lun)
4412 {
4413         struct ctl_lun *nlun, *lun;
4414         struct scsi_vpd_id_descriptor *desc;
4415         struct scsi_vpd_id_t10 *t10id;
4416         const char *eui, *naa, *scsiname, *vendor, *value;
4417         int lun_number, i, lun_malloced;
4418         int devidlen, idlen1, idlen2 = 0, len;
4419
4420         if (be_lun == NULL)
4421                 return (EINVAL);
4422
4423         /*
4424          * We currently only support Direct Access or Processor LUN types.
4425          */
4426         switch (be_lun->lun_type) {
4427         case T_DIRECT:
4428         case T_PROCESSOR:
4429         case T_CDROM:
4430                 break;
4431         case T_SEQUENTIAL:
4432         case T_CHANGER:
4433         default:
4434                 be_lun->lun_config_status(be_lun->be_lun,
4435                                           CTL_LUN_CONFIG_FAILURE);
4436                 break;
4437         }
4438         if (ctl_lun == NULL) {
4439                 lun = malloc(sizeof(*lun), M_CTL, M_WAITOK);
4440                 lun_malloced = 1;
4441         } else {
4442                 lun_malloced = 0;
4443                 lun = ctl_lun;
4444         }
4445
4446         memset(lun, 0, sizeof(*lun));
4447         if (lun_malloced)
4448                 lun->flags = CTL_LUN_MALLOCED;
4449
4450         /* Generate LUN ID. */
4451         devidlen = max(CTL_DEVID_MIN_LEN,
4452             strnlen(be_lun->device_id, CTL_DEVID_LEN));
4453         idlen1 = sizeof(*t10id) + devidlen;
4454         len = sizeof(struct scsi_vpd_id_descriptor) + idlen1;
4455         scsiname = ctl_get_opt(&be_lun->options, "scsiname");
4456         if (scsiname != NULL) {
4457                 idlen2 = roundup2(strlen(scsiname) + 1, 4);
4458                 len += sizeof(struct scsi_vpd_id_descriptor) + idlen2;
4459         }
4460         eui = ctl_get_opt(&be_lun->options, "eui");
4461         if (eui != NULL) {
4462                 len += sizeof(struct scsi_vpd_id_descriptor) + 16;
4463         }
4464         naa = ctl_get_opt(&be_lun->options, "naa");
4465         if (naa != NULL) {
4466                 len += sizeof(struct scsi_vpd_id_descriptor) + 16;
4467         }
4468         lun->lun_devid = malloc(sizeof(struct ctl_devid) + len,
4469             M_CTL, M_WAITOK | M_ZERO);
4470         desc = (struct scsi_vpd_id_descriptor *)lun->lun_devid->data;
4471         desc->proto_codeset = SVPD_ID_CODESET_ASCII;
4472         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_T10;
4473         desc->length = idlen1;
4474         t10id = (struct scsi_vpd_id_t10 *)&desc->identifier[0];
4475         memset(t10id->vendor, ' ', sizeof(t10id->vendor));
4476         if ((vendor = ctl_get_opt(&be_lun->options, "vendor")) == NULL) {
4477                 strncpy((char *)t10id->vendor, CTL_VENDOR, sizeof(t10id->vendor));
4478         } else {
4479                 strncpy(t10id->vendor, vendor,
4480                     min(sizeof(t10id->vendor), strlen(vendor)));
4481         }
4482         strncpy((char *)t10id->vendor_spec_id,
4483             (char *)be_lun->device_id, devidlen);
4484         if (scsiname != NULL) {
4485                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4486                     desc->length);
4487                 desc->proto_codeset = SVPD_ID_CODESET_UTF8;
4488                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4489                     SVPD_ID_TYPE_SCSI_NAME;
4490                 desc->length = idlen2;
4491                 strlcpy(desc->identifier, scsiname, idlen2);
4492         }
4493         if (eui != NULL) {
4494                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4495                     desc->length);
4496                 desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4497                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4498                     SVPD_ID_TYPE_EUI64;
4499                 desc->length = hex2bin(eui, desc->identifier, 16);
4500                 desc->length = desc->length > 12 ? 16 :
4501                     (desc->length > 8 ? 12 : 8);
4502                 len -= 16 - desc->length;
4503         }
4504         if (naa != NULL) {
4505                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4506                     desc->length);
4507                 desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4508                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4509                     SVPD_ID_TYPE_NAA;
4510                 desc->length = hex2bin(naa, desc->identifier, 16);
4511                 desc->length = desc->length > 8 ? 16 : 8;
4512                 len -= 16 - desc->length;
4513         }
4514         lun->lun_devid->len = len;
4515
4516         mtx_lock(&ctl_softc->ctl_lock);
4517         /*
4518          * See if the caller requested a particular LUN number.  If so, see
4519          * if it is available.  Otherwise, allocate the first available LUN.
4520          */
4521         if (be_lun->flags & CTL_LUN_FLAG_ID_REQ) {
4522                 if ((be_lun->req_lun_id > (CTL_MAX_LUNS - 1))
4523                  || (ctl_is_set(ctl_softc->ctl_lun_mask, be_lun->req_lun_id))) {
4524                         mtx_unlock(&ctl_softc->ctl_lock);
4525                         if (be_lun->req_lun_id > (CTL_MAX_LUNS - 1)) {
4526                                 printf("ctl: requested LUN ID %d is higher "
4527                                        "than CTL_MAX_LUNS - 1 (%d)\n",
4528                                        be_lun->req_lun_id, CTL_MAX_LUNS - 1);
4529                         } else {
4530                                 /*
4531                                  * XXX KDM return an error, or just assign
4532                                  * another LUN ID in this case??
4533                                  */
4534                                 printf("ctl: requested LUN ID %d is already "
4535                                        "in use\n", be_lun->req_lun_id);
4536                         }
4537                         if (lun->flags & CTL_LUN_MALLOCED)
4538                                 free(lun, M_CTL);
4539                         be_lun->lun_config_status(be_lun->be_lun,
4540                                                   CTL_LUN_CONFIG_FAILURE);
4541                         return (ENOSPC);
4542                 }
4543                 lun_number = be_lun->req_lun_id;
4544         } else {
4545                 lun_number = ctl_ffz(ctl_softc->ctl_lun_mask, 0, CTL_MAX_LUNS);
4546                 if (lun_number == -1) {
4547                         mtx_unlock(&ctl_softc->ctl_lock);
4548                         printf("ctl: can't allocate LUN, out of LUNs\n");
4549                         if (lun->flags & CTL_LUN_MALLOCED)
4550                                 free(lun, M_CTL);
4551                         be_lun->lun_config_status(be_lun->be_lun,
4552                                                   CTL_LUN_CONFIG_FAILURE);
4553                         return (ENOSPC);
4554                 }
4555         }
4556         ctl_set_mask(ctl_softc->ctl_lun_mask, lun_number);
4557
4558         mtx_init(&lun->lun_lock, "CTL LUN", NULL, MTX_DEF);
4559         lun->lun = lun_number;
4560         lun->be_lun = be_lun;
4561         /*
4562          * The processor LUN is always enabled.  Disk LUNs come on line
4563          * disabled, and must be enabled by the backend.
4564          */
4565         lun->flags |= CTL_LUN_DISABLED;
4566         lun->backend = be_lun->be;
4567         be_lun->ctl_lun = lun;
4568         be_lun->lun_id = lun_number;
4569         atomic_add_int(&be_lun->be->num_luns, 1);
4570         if (be_lun->flags & CTL_LUN_FLAG_EJECTED)
4571                 lun->flags |= CTL_LUN_EJECTED;
4572         if (be_lun->flags & CTL_LUN_FLAG_NO_MEDIA)
4573                 lun->flags |= CTL_LUN_NO_MEDIA;
4574         if (be_lun->flags & CTL_LUN_FLAG_STOPPED)
4575                 lun->flags |= CTL_LUN_STOPPED;
4576
4577         if (be_lun->flags & CTL_LUN_FLAG_PRIMARY)
4578                 lun->flags |= CTL_LUN_PRIMARY_SC;
4579
4580         value = ctl_get_opt(&be_lun->options, "removable");
4581         if (value != NULL) {
4582                 if (strcmp(value, "on") == 0)
4583                         lun->flags |= CTL_LUN_REMOVABLE;
4584         } else if (be_lun->lun_type == T_CDROM)
4585                 lun->flags |= CTL_LUN_REMOVABLE;
4586
4587         lun->ctl_softc = ctl_softc;
4588 #ifdef CTL_TIME_IO
4589         lun->last_busy = getsbinuptime();
4590 #endif
4591         TAILQ_INIT(&lun->ooa_queue);
4592         TAILQ_INIT(&lun->blocked_queue);
4593         STAILQ_INIT(&lun->error_list);
4594         ctl_tpc_lun_init(lun);
4595
4596         /*
4597          * Initialize the mode and log page index.
4598          */
4599         ctl_init_page_index(lun);
4600         ctl_init_log_page_index(lun);
4601
4602         /*
4603          * Now, before we insert this lun on the lun list, set the lun
4604          * inventory changed UA for all other luns.
4605          */
4606         STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) {
4607                 mtx_lock(&nlun->lun_lock);
4608                 ctl_est_ua_all(nlun, -1, CTL_UA_LUN_CHANGE);
4609                 mtx_unlock(&nlun->lun_lock);
4610         }
4611
4612         STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links);
4613
4614         ctl_softc->ctl_luns[lun_number] = lun;
4615
4616         ctl_softc->num_luns++;
4617
4618         /* Setup statistics gathering */
4619         lun->stats.device_type = be_lun->lun_type;
4620         lun->stats.lun_number = lun_number;
4621         lun->stats.blocksize = be_lun->blocksize;
4622         if (be_lun->blocksize == 0)
4623                 lun->stats.flags = CTL_LUN_STATS_NO_BLOCKSIZE;
4624         for (i = 0;i < CTL_MAX_PORTS;i++)
4625                 lun->stats.ports[i].targ_port = i;
4626
4627         mtx_unlock(&ctl_softc->ctl_lock);
4628
4629         lun->be_lun->lun_config_status(lun->be_lun->be_lun, CTL_LUN_CONFIG_OK);
4630         return (0);
4631 }
4632
4633 /*
4634  * Delete a LUN.
4635  * Assumptions:
4636  * - LUN has already been marked invalid and any pending I/O has been taken
4637  *   care of.
4638  */
4639 static int
4640 ctl_free_lun(struct ctl_lun *lun)
4641 {
4642         struct ctl_softc *softc;
4643         struct ctl_lun *nlun;
4644         int i;
4645
4646         softc = lun->ctl_softc;
4647
4648         mtx_assert(&softc->ctl_lock, MA_OWNED);
4649
4650         STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links);
4651
4652         ctl_clear_mask(softc->ctl_lun_mask, lun->lun);
4653
4654         softc->ctl_luns[lun->lun] = NULL;
4655
4656         if (!TAILQ_EMPTY(&lun->ooa_queue))
4657                 panic("Freeing a LUN %p with outstanding I/O!!\n", lun);
4658
4659         softc->num_luns--;
4660
4661         /*
4662          * Tell the backend to free resources, if this LUN has a backend.
4663          */
4664         atomic_subtract_int(&lun->be_lun->be->num_luns, 1);
4665         lun->be_lun->lun_shutdown(lun->be_lun->be_lun);
4666
4667         ctl_tpc_lun_shutdown(lun);
4668         mtx_destroy(&lun->lun_lock);
4669         free(lun->lun_devid, M_CTL);
4670         for (i = 0; i < CTL_MAX_PORTS; i++)
4671                 free(lun->pending_ua[i], M_CTL);
4672         for (i = 0; i < CTL_MAX_PORTS; i++)
4673                 free(lun->pr_keys[i], M_CTL);
4674         free(lun->write_buffer, M_CTL);
4675         if (lun->flags & CTL_LUN_MALLOCED)
4676                 free(lun, M_CTL);
4677
4678         STAILQ_FOREACH(nlun, &softc->lun_list, links) {
4679                 mtx_lock(&nlun->lun_lock);
4680                 ctl_est_ua_all(nlun, -1, CTL_UA_LUN_CHANGE);
4681                 mtx_unlock(&nlun->lun_lock);
4682         }
4683
4684         return (0);
4685 }
4686
4687 static void
4688 ctl_create_lun(struct ctl_be_lun *be_lun)
4689 {
4690
4691         /*
4692          * ctl_alloc_lun() should handle all potential failure cases.
4693          */
4694         ctl_alloc_lun(control_softc, NULL, be_lun);
4695 }
4696
4697 int
4698 ctl_add_lun(struct ctl_be_lun *be_lun)
4699 {
4700         struct ctl_softc *softc = control_softc;
4701
4702         mtx_lock(&softc->ctl_lock);
4703         STAILQ_INSERT_TAIL(&softc->pending_lun_queue, be_lun, links);
4704         mtx_unlock(&softc->ctl_lock);
4705         wakeup(&softc->pending_lun_queue);
4706
4707         return (0);
4708 }
4709
4710 int
4711 ctl_enable_lun(struct ctl_be_lun *be_lun)
4712 {
4713         struct ctl_softc *softc;
4714         struct ctl_port *port, *nport;
4715         struct ctl_lun *lun;
4716         int retval;
4717
4718         lun = (struct ctl_lun *)be_lun->ctl_lun;
4719         softc = lun->ctl_softc;
4720
4721         mtx_lock(&softc->ctl_lock);
4722         mtx_lock(&lun->lun_lock);
4723         if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4724                 /*
4725                  * eh?  Why did we get called if the LUN is already
4726                  * enabled?
4727                  */
4728                 mtx_unlock(&lun->lun_lock);
4729                 mtx_unlock(&softc->ctl_lock);
4730                 return (0);
4731         }
4732         lun->flags &= ~CTL_LUN_DISABLED;
4733         mtx_unlock(&lun->lun_lock);
4734
4735         STAILQ_FOREACH_SAFE(port, &softc->port_list, links, nport) {
4736                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0 ||
4737                     port->lun_map != NULL || port->lun_enable == NULL)
4738                         continue;
4739
4740                 /*
4741                  * Drop the lock while we call the FETD's enable routine.
4742                  * This can lead to a callback into CTL (at least in the
4743                  * case of the internal initiator frontend.
4744                  */
4745                 mtx_unlock(&softc->ctl_lock);
4746                 retval = port->lun_enable(port->targ_lun_arg, lun->lun);
4747                 mtx_lock(&softc->ctl_lock);
4748                 if (retval != 0) {
4749                         printf("%s: FETD %s port %d returned error "
4750                                "%d for lun_enable on lun %jd\n",
4751                                __func__, port->port_name, port->targ_port,
4752                                retval, (intmax_t)lun->lun);
4753                 }
4754         }
4755
4756         mtx_unlock(&softc->ctl_lock);
4757         ctl_isc_announce_lun(lun);
4758
4759         return (0);
4760 }
4761
4762 int
4763 ctl_disable_lun(struct ctl_be_lun *be_lun)
4764 {
4765         struct ctl_softc *softc;
4766         struct ctl_port *port;
4767         struct ctl_lun *lun;
4768         int retval;
4769
4770         lun = (struct ctl_lun *)be_lun->ctl_lun;
4771         softc = lun->ctl_softc;
4772
4773         mtx_lock(&softc->ctl_lock);
4774         mtx_lock(&lun->lun_lock);
4775         if (lun->flags & CTL_LUN_DISABLED) {
4776                 mtx_unlock(&lun->lun_lock);
4777                 mtx_unlock(&softc->ctl_lock);
4778                 return (0);
4779         }
4780         lun->flags |= CTL_LUN_DISABLED;
4781         mtx_unlock(&lun->lun_lock);
4782
4783         STAILQ_FOREACH(port, &softc->port_list, links) {
4784                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0 ||
4785                     port->lun_map != NULL || port->lun_disable == NULL)
4786                         continue;
4787
4788                 /*
4789                  * Drop the lock before we call the frontend's disable
4790                  * routine, to avoid lock order reversals.
4791                  *
4792                  * XXX KDM what happens if the frontend list changes while
4793                  * we're traversing it?  It's unlikely, but should be handled.
4794                  */
4795                 mtx_unlock(&softc->ctl_lock);
4796                 retval = port->lun_disable(port->targ_lun_arg, lun->lun);
4797                 mtx_lock(&softc->ctl_lock);
4798                 if (retval != 0) {
4799                         printf("%s: FETD %s port %d returned error "
4800                                "%d for lun_disable on lun %jd\n",
4801                                __func__, port->port_name, port->targ_port,
4802                                retval, (intmax_t)lun->lun);
4803                 }
4804         }
4805
4806         mtx_unlock(&softc->ctl_lock);
4807         ctl_isc_announce_lun(lun);
4808
4809         return (0);
4810 }
4811
4812 int
4813 ctl_start_lun(struct ctl_be_lun *be_lun)
4814 {
4815         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4816
4817         mtx_lock(&lun->lun_lock);
4818         lun->flags &= ~CTL_LUN_STOPPED;
4819         mtx_unlock(&lun->lun_lock);
4820         return (0);
4821 }
4822
4823 int
4824 ctl_stop_lun(struct ctl_be_lun *be_lun)
4825 {
4826         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4827
4828         mtx_lock(&lun->lun_lock);
4829         lun->flags |= CTL_LUN_STOPPED;
4830         mtx_unlock(&lun->lun_lock);
4831         return (0);
4832 }
4833
4834 int
4835 ctl_lun_no_media(struct ctl_be_lun *be_lun)
4836 {
4837         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4838
4839         mtx_lock(&lun->lun_lock);
4840         lun->flags |= CTL_LUN_NO_MEDIA;
4841         mtx_unlock(&lun->lun_lock);
4842         return (0);
4843 }
4844
4845 int
4846 ctl_lun_has_media(struct ctl_be_lun *be_lun)
4847 {
4848         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4849         union ctl_ha_msg msg;
4850
4851         mtx_lock(&lun->lun_lock);
4852         lun->flags &= ~(CTL_LUN_NO_MEDIA | CTL_LUN_EJECTED);
4853         if (lun->flags & CTL_LUN_REMOVABLE)
4854                 ctl_est_ua_all(lun, -1, CTL_UA_MEDIUM_CHANGE);
4855         mtx_unlock(&lun->lun_lock);
4856         if ((lun->flags & CTL_LUN_REMOVABLE) &&
4857             lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
4858                 bzero(&msg.ua, sizeof(msg.ua));
4859                 msg.hdr.msg_type = CTL_MSG_UA;
4860                 msg.hdr.nexus.initid = -1;
4861                 msg.hdr.nexus.targ_port = -1;
4862                 msg.hdr.nexus.targ_lun = lun->lun;
4863                 msg.hdr.nexus.targ_mapped_lun = lun->lun;
4864                 msg.ua.ua_all = 1;
4865                 msg.ua.ua_set = 1;
4866                 msg.ua.ua_type = CTL_UA_MEDIUM_CHANGE;
4867                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg.ua),
4868                     M_WAITOK);
4869         }
4870         return (0);
4871 }
4872
4873 int
4874 ctl_lun_ejected(struct ctl_be_lun *be_lun)
4875 {
4876         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4877
4878         mtx_lock(&lun->lun_lock);
4879         lun->flags |= CTL_LUN_EJECTED;
4880         mtx_unlock(&lun->lun_lock);
4881         return (0);
4882 }
4883
4884 int
4885 ctl_lun_primary(struct ctl_be_lun *be_lun)
4886 {
4887         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4888
4889         mtx_lock(&lun->lun_lock);
4890         lun->flags |= CTL_LUN_PRIMARY_SC;
4891         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
4892         mtx_unlock(&lun->lun_lock);
4893         ctl_isc_announce_lun(lun);
4894         return (0);
4895 }
4896
4897 int
4898 ctl_lun_secondary(struct ctl_be_lun *be_lun)
4899 {
4900         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4901
4902         mtx_lock(&lun->lun_lock);
4903         lun->flags &= ~CTL_LUN_PRIMARY_SC;
4904         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
4905         mtx_unlock(&lun->lun_lock);
4906         ctl_isc_announce_lun(lun);
4907         return (0);
4908 }
4909
4910 int
4911 ctl_invalidate_lun(struct ctl_be_lun *be_lun)
4912 {
4913         struct ctl_softc *softc;
4914         struct ctl_lun *lun;
4915
4916         lun = (struct ctl_lun *)be_lun->ctl_lun;
4917         softc = lun->ctl_softc;
4918
4919         mtx_lock(&lun->lun_lock);
4920
4921         /*
4922          * The LUN needs to be disabled before it can be marked invalid.
4923          */
4924         if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4925                 mtx_unlock(&lun->lun_lock);
4926                 return (-1);
4927         }
4928         /*
4929          * Mark the LUN invalid.
4930          */
4931         lun->flags |= CTL_LUN_INVALID;
4932
4933         /*
4934          * If there is nothing in the OOA queue, go ahead and free the LUN.
4935          * If we have something in the OOA queue, we'll free it when the
4936          * last I/O completes.
4937          */
4938         if (TAILQ_EMPTY(&lun->ooa_queue)) {
4939                 mtx_unlock(&lun->lun_lock);
4940                 mtx_lock(&softc->ctl_lock);
4941                 ctl_free_lun(lun);
4942                 mtx_unlock(&softc->ctl_lock);
4943         } else
4944                 mtx_unlock(&lun->lun_lock);
4945
4946         return (0);
4947 }
4948
4949 void
4950 ctl_lun_capacity_changed(struct ctl_be_lun *be_lun)
4951 {
4952         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4953         union ctl_ha_msg msg;
4954
4955         mtx_lock(&lun->lun_lock);
4956         ctl_est_ua_all(lun, -1, CTL_UA_CAPACITY_CHANGE);
4957         mtx_unlock(&lun->lun_lock);
4958         if (lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
4959                 /* Send msg to other side. */
4960                 bzero(&msg.ua, sizeof(msg.ua));
4961                 msg.hdr.msg_type = CTL_MSG_UA;
4962                 msg.hdr.nexus.initid = -1;
4963                 msg.hdr.nexus.targ_port = -1;
4964                 msg.hdr.nexus.targ_lun = lun->lun;
4965                 msg.hdr.nexus.targ_mapped_lun = lun->lun;
4966                 msg.ua.ua_all = 1;
4967                 msg.ua.ua_set = 1;
4968                 msg.ua.ua_type = CTL_UA_CAPACITY_CHANGE;
4969                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg.ua),
4970                     M_WAITOK);
4971         }
4972 }
4973
4974 /*
4975  * Backend "memory move is complete" callback for requests that never
4976  * make it down to say RAIDCore's configuration code.
4977  */
4978 int
4979 ctl_config_move_done(union ctl_io *io)
4980 {
4981         int retval;
4982
4983         CTL_DEBUG_PRINT(("ctl_config_move_done\n"));
4984         KASSERT(io->io_hdr.io_type == CTL_IO_SCSI,
4985             ("Config I/O type isn't CTL_IO_SCSI (%d)!", io->io_hdr.io_type));
4986
4987         if ((io->io_hdr.port_status != 0) &&
4988             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
4989              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
4990                 /*
4991                  * For hardware error sense keys, the sense key
4992                  * specific value is defined to be a retry count,
4993                  * but we use it to pass back an internal FETD
4994                  * error code.  XXX KDM  Hopefully the FETD is only
4995                  * using 16 bits for an error code, since that's
4996                  * all the space we have in the sks field.
4997                  */
4998                 ctl_set_internal_failure(&io->scsiio,
4999                                          /*sks_valid*/ 1,
5000                                          /*retry_count*/
5001                                          io->io_hdr.port_status);
5002         }
5003
5004         if (ctl_debug & CTL_DEBUG_CDB_DATA)
5005                 ctl_data_print(io);
5006         if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN) ||
5007             ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
5008              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS) ||
5009             ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) {
5010                 /*
5011                  * XXX KDM just assuming a single pointer here, and not a
5012                  * S/G list.  If we start using S/G lists for config data,
5013                  * we'll need to know how to clean them up here as well.
5014                  */
5015                 if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5016                         free(io->scsiio.kern_data_ptr, M_CTL);
5017                 ctl_done(io);
5018                 retval = CTL_RETVAL_COMPLETE;
5019         } else {
5020                 /*
5021                  * XXX KDM now we need to continue data movement.  Some
5022                  * options:
5023                  * - call ctl_scsiio() again?  We don't do this for data
5024                  *   writes, because for those at least we know ahead of
5025                  *   time where the write will go and how long it is.  For
5026                  *   config writes, though, that information is largely
5027                  *   contained within the write itself, thus we need to
5028                  *   parse out the data again.
5029                  *
5030                  * - Call some other function once the data is in?
5031                  */
5032
5033                 /*
5034                  * XXX KDM call ctl_scsiio() again for now, and check flag
5035                  * bits to see whether we're allocated or not.
5036                  */
5037                 retval = ctl_scsiio(&io->scsiio);
5038         }
5039         return (retval);
5040 }
5041
5042 /*
5043  * This gets called by a backend driver when it is done with a
5044  * data_submit method.
5045  */
5046 void
5047 ctl_data_submit_done(union ctl_io *io)
5048 {
5049         /*
5050          * If the IO_CONT flag is set, we need to call the supplied
5051          * function to continue processing the I/O, instead of completing
5052          * the I/O just yet.
5053          *
5054          * If there is an error, though, we don't want to keep processing.
5055          * Instead, just send status back to the initiator.
5056          */
5057         if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5058             (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5059             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5060              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5061                 io->scsiio.io_cont(io);
5062                 return;
5063         }
5064         ctl_done(io);
5065 }
5066
5067 /*
5068  * This gets called by a backend driver when it is done with a
5069  * configuration write.
5070  */
5071 void
5072 ctl_config_write_done(union ctl_io *io)
5073 {
5074         uint8_t *buf;
5075
5076         /*
5077          * If the IO_CONT flag is set, we need to call the supplied
5078          * function to continue processing the I/O, instead of completing
5079          * the I/O just yet.
5080          *
5081          * If there is an error, though, we don't want to keep processing.
5082          * Instead, just send status back to the initiator.
5083          */
5084         if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5085             (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5086             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5087              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5088                 io->scsiio.io_cont(io);
5089                 return;
5090         }
5091         /*
5092          * Since a configuration write can be done for commands that actually
5093          * have data allocated, like write buffer, and commands that have
5094          * no data, like start/stop unit, we need to check here.
5095          */
5096         if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5097                 buf = io->scsiio.kern_data_ptr;
5098         else
5099                 buf = NULL;
5100         ctl_done(io);
5101         if (buf)
5102                 free(buf, M_CTL);
5103 }
5104
5105 void
5106 ctl_config_read_done(union ctl_io *io)
5107 {
5108         uint8_t *buf;
5109
5110         /*
5111          * If there is some error -- we are done, skip data transfer.
5112          */
5113         if ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0 ||
5114             ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
5115              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) {
5116                 if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5117                         buf = io->scsiio.kern_data_ptr;
5118                 else
5119                         buf = NULL;
5120                 ctl_done(io);
5121                 if (buf)
5122                         free(buf, M_CTL);
5123                 return;
5124         }
5125
5126         /*
5127          * If the IO_CONT flag is set, we need to call the supplied
5128          * function to continue processing the I/O, instead of completing
5129          * the I/O just yet.
5130          */
5131         if (io->io_hdr.flags & CTL_FLAG_IO_CONT) {
5132                 io->scsiio.io_cont(io);
5133                 return;
5134         }
5135
5136         ctl_datamove(io);
5137 }
5138
5139 /*
5140  * SCSI release command.
5141  */
5142 int
5143 ctl_scsi_release(struct ctl_scsiio *ctsio)
5144 {
5145         struct ctl_lun *lun;
5146         uint32_t residx;
5147
5148         CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
5149
5150         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5151         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5152
5153         /*
5154          * XXX KDM right now, we only support LUN reservation.  We don't
5155          * support 3rd party reservations, or extent reservations, which
5156          * might actually need the parameter list.  If we've gotten this
5157          * far, we've got a LUN reservation.  Anything else got kicked out
5158          * above.  So, according to SPC, ignore the length.
5159          */
5160
5161         mtx_lock(&lun->lun_lock);
5162
5163         /*
5164          * According to SPC, it is not an error for an intiator to attempt
5165          * to release a reservation on a LUN that isn't reserved, or that
5166          * is reserved by another initiator.  The reservation can only be
5167          * released, though, by the initiator who made it or by one of
5168          * several reset type events.
5169          */
5170         if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == residx))
5171                         lun->flags &= ~CTL_LUN_RESERVED;
5172
5173         mtx_unlock(&lun->lun_lock);
5174
5175         ctl_set_success(ctsio);
5176         ctl_done((union ctl_io *)ctsio);
5177         return (CTL_RETVAL_COMPLETE);
5178 }
5179
5180 int
5181 ctl_scsi_reserve(struct ctl_scsiio *ctsio)
5182 {
5183         struct ctl_lun *lun;
5184         uint32_t residx;
5185
5186         CTL_DEBUG_PRINT(("ctl_reserve\n"));
5187
5188         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5189         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5190
5191         /*
5192          * XXX KDM right now, we only support LUN reservation.  We don't
5193          * support 3rd party reservations, or extent reservations, which
5194          * might actually need the parameter list.  If we've gotten this
5195          * far, we've got a LUN reservation.  Anything else got kicked out
5196          * above.  So, according to SPC, ignore the length.
5197          */
5198
5199         mtx_lock(&lun->lun_lock);
5200         if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx != residx)) {
5201                 ctl_set_reservation_conflict(ctsio);
5202                 goto bailout;
5203         }
5204
5205         /* SPC-3 exceptions to SPC-2 RESERVE and RELEASE behavior. */
5206         if (lun->flags & CTL_LUN_PR_RESERVED) {
5207                 ctl_set_success(ctsio);
5208                 goto bailout;
5209         }
5210
5211         lun->flags |= CTL_LUN_RESERVED;
5212         lun->res_idx = residx;
5213         ctl_set_success(ctsio);
5214
5215 bailout:
5216         mtx_unlock(&lun->lun_lock);
5217         ctl_done((union ctl_io *)ctsio);
5218         return (CTL_RETVAL_COMPLETE);
5219 }
5220
5221 int
5222 ctl_start_stop(struct ctl_scsiio *ctsio)
5223 {
5224         struct scsi_start_stop_unit *cdb;
5225         struct ctl_lun *lun;
5226         int retval;
5227
5228         CTL_DEBUG_PRINT(("ctl_start_stop\n"));
5229
5230         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5231         cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
5232
5233         if ((cdb->how & SSS_PC_MASK) == 0) {
5234                 if ((lun->flags & CTL_LUN_PR_RESERVED) &&
5235                     (cdb->how & SSS_START) == 0) {
5236                         uint32_t residx;
5237
5238                         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5239                         if (ctl_get_prkey(lun, residx) == 0 ||
5240                             (lun->pr_res_idx != residx && lun->pr_res_type < 4)) {
5241
5242                                 ctl_set_reservation_conflict(ctsio);
5243                                 ctl_done((union ctl_io *)ctsio);
5244                                 return (CTL_RETVAL_COMPLETE);
5245                         }
5246                 }
5247
5248                 if ((cdb->how & SSS_LOEJ) &&
5249                     (lun->flags & CTL_LUN_REMOVABLE) == 0) {
5250                         ctl_set_invalid_field(ctsio,
5251                                               /*sks_valid*/ 1,
5252                                               /*command*/ 1,
5253                                               /*field*/ 4,
5254                                               /*bit_valid*/ 1,
5255                                               /*bit*/ 1);
5256                         ctl_done((union ctl_io *)ctsio);
5257                         return (CTL_RETVAL_COMPLETE);
5258                 }
5259
5260                 if ((cdb->how & SSS_START) == 0 && (cdb->how & SSS_LOEJ) &&
5261                     lun->prevent_count > 0) {
5262                         /* "Medium removal prevented" */
5263                         ctl_set_sense(ctsio, /*current_error*/ 1,
5264                             /*sense_key*/(lun->flags & CTL_LUN_NO_MEDIA) ?
5265                              SSD_KEY_NOT_READY : SSD_KEY_ILLEGAL_REQUEST,
5266                             /*asc*/ 0x53, /*ascq*/ 0x02, SSD_ELEM_NONE);
5267                         ctl_done((union ctl_io *)ctsio);
5268                         return (CTL_RETVAL_COMPLETE);
5269                 }
5270         }
5271
5272         retval = lun->backend->config_write((union ctl_io *)ctsio);
5273         return (retval);
5274 }
5275
5276 int
5277 ctl_prevent_allow(struct ctl_scsiio *ctsio)
5278 {
5279         struct ctl_lun *lun;
5280         struct scsi_prevent *cdb;
5281         int retval;
5282         uint32_t initidx;
5283
5284         CTL_DEBUG_PRINT(("ctl_prevent_allow\n"));
5285
5286         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5287         cdb = (struct scsi_prevent *)ctsio->cdb;
5288
5289         if ((lun->flags & CTL_LUN_REMOVABLE) == 0) {
5290                 ctl_set_invalid_opcode(ctsio);
5291                 ctl_done((union ctl_io *)ctsio);
5292                 return (CTL_RETVAL_COMPLETE);
5293         }
5294
5295         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5296         mtx_lock(&lun->lun_lock);
5297         if ((cdb->how & PR_PREVENT) &&
5298             ctl_is_set(lun->prevent, initidx) == 0) {
5299                 ctl_set_mask(lun->prevent, initidx);
5300                 lun->prevent_count++;
5301         } else if ((cdb->how & PR_PREVENT) == 0 &&
5302             ctl_is_set(lun->prevent, initidx)) {
5303                 ctl_clear_mask(lun->prevent, initidx);
5304                 lun->prevent_count--;
5305         }
5306         mtx_unlock(&lun->lun_lock);
5307         retval = lun->backend->config_write((union ctl_io *)ctsio);
5308         return (retval);
5309 }
5310
5311 /*
5312  * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
5313  * we don't really do anything with the LBA and length fields if the user
5314  * passes them in.  Instead we'll just flush out the cache for the entire
5315  * LUN.
5316  */
5317 int
5318 ctl_sync_cache(struct ctl_scsiio *ctsio)
5319 {
5320         struct ctl_lun *lun;
5321         struct ctl_softc *softc;
5322         struct ctl_lba_len_flags *lbalen;
5323         uint64_t starting_lba;
5324         uint32_t block_count;
5325         int retval;
5326         uint8_t byte2;
5327
5328         CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
5329
5330         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5331         softc = lun->ctl_softc;
5332         retval = 0;
5333
5334         switch (ctsio->cdb[0]) {
5335         case SYNCHRONIZE_CACHE: {
5336                 struct scsi_sync_cache *cdb;
5337                 cdb = (struct scsi_sync_cache *)ctsio->cdb;
5338
5339                 starting_lba = scsi_4btoul(cdb->begin_lba);
5340                 block_count = scsi_2btoul(cdb->lb_count);
5341                 byte2 = cdb->byte2;
5342                 break;
5343         }
5344         case SYNCHRONIZE_CACHE_16: {
5345                 struct scsi_sync_cache_16 *cdb;
5346                 cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5347
5348                 starting_lba = scsi_8btou64(cdb->begin_lba);
5349                 block_count = scsi_4btoul(cdb->lb_count);
5350                 byte2 = cdb->byte2;
5351                 break;
5352         }
5353         default:
5354                 ctl_set_invalid_opcode(ctsio);
5355                 ctl_done((union ctl_io *)ctsio);
5356                 goto bailout;
5357                 break; /* NOTREACHED */
5358         }
5359
5360         /*
5361          * We check the LBA and length, but don't do anything with them.
5362          * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5363          * get flushed.  This check will just help satisfy anyone who wants
5364          * to see an error for an out of range LBA.
5365          */
5366         if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5367                 ctl_set_lba_out_of_range(ctsio);
5368                 ctl_done((union ctl_io *)ctsio);
5369                 goto bailout;
5370         }
5371
5372         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5373         lbalen->lba = starting_lba;
5374         lbalen->len = block_count;
5375         lbalen->flags = byte2;
5376         retval = lun->backend->config_write((union ctl_io *)ctsio);
5377
5378 bailout:
5379         return (retval);
5380 }
5381
5382 int
5383 ctl_format(struct ctl_scsiio *ctsio)
5384 {
5385         struct scsi_format *cdb;
5386         struct ctl_lun *lun;
5387         int length, defect_list_len;
5388
5389         CTL_DEBUG_PRINT(("ctl_format\n"));
5390
5391         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5392
5393         cdb = (struct scsi_format *)ctsio->cdb;
5394
5395         length = 0;
5396         if (cdb->byte2 & SF_FMTDATA) {
5397                 if (cdb->byte2 & SF_LONGLIST)
5398                         length = sizeof(struct scsi_format_header_long);
5399                 else
5400                         length = sizeof(struct scsi_format_header_short);
5401         }
5402
5403         if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5404          && (length > 0)) {
5405                 ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5406                 ctsio->kern_data_len = length;
5407                 ctsio->kern_total_len = length;
5408                 ctsio->kern_data_resid = 0;
5409                 ctsio->kern_rel_offset = 0;
5410                 ctsio->kern_sg_entries = 0;
5411                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5412                 ctsio->be_move_done = ctl_config_move_done;
5413                 ctl_datamove((union ctl_io *)ctsio);
5414
5415                 return (CTL_RETVAL_COMPLETE);
5416         }
5417
5418         defect_list_len = 0;
5419
5420         if (cdb->byte2 & SF_FMTDATA) {
5421                 if (cdb->byte2 & SF_LONGLIST) {
5422                         struct scsi_format_header_long *header;
5423
5424                         header = (struct scsi_format_header_long *)
5425                                 ctsio->kern_data_ptr;
5426
5427                         defect_list_len = scsi_4btoul(header->defect_list_len);
5428                         if (defect_list_len != 0) {
5429                                 ctl_set_invalid_field(ctsio,
5430                                                       /*sks_valid*/ 1,
5431                                                       /*command*/ 0,
5432                                                       /*field*/ 2,
5433                                                       /*bit_valid*/ 0,
5434                                                       /*bit*/ 0);
5435                                 goto bailout;
5436                         }
5437                 } else {
5438                         struct scsi_format_header_short *header;
5439
5440                         header = (struct scsi_format_header_short *)
5441                                 ctsio->kern_data_ptr;
5442
5443                         defect_list_len = scsi_2btoul(header->defect_list_len);
5444                         if (defect_list_len != 0) {
5445                                 ctl_set_invalid_field(ctsio,
5446                                                       /*sks_valid*/ 1,
5447                                                       /*command*/ 0,
5448                                                       /*field*/ 2,
5449                                                       /*bit_valid*/ 0,
5450                                                       /*bit*/ 0);
5451                                 goto bailout;
5452                         }
5453                 }
5454         }
5455
5456         ctl_set_success(ctsio);
5457 bailout:
5458
5459         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5460                 free(ctsio->kern_data_ptr, M_CTL);
5461                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5462         }
5463
5464         ctl_done((union ctl_io *)ctsio);
5465         return (CTL_RETVAL_COMPLETE);
5466 }
5467
5468 int
5469 ctl_read_buffer(struct ctl_scsiio *ctsio)
5470 {
5471         struct ctl_lun *lun;
5472         uint64_t buffer_offset;
5473         uint32_t len;
5474         uint8_t byte2;
5475         static uint8_t descr[4];
5476         static uint8_t echo_descr[4] = { 0 };
5477
5478         CTL_DEBUG_PRINT(("ctl_read_buffer\n"));
5479         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5480         switch (ctsio->cdb[0]) {
5481         case READ_BUFFER: {
5482                 struct scsi_read_buffer *cdb;
5483
5484                 cdb = (struct scsi_read_buffer *)ctsio->cdb;
5485                 buffer_offset = scsi_3btoul(cdb->offset);
5486                 len = scsi_3btoul(cdb->length);
5487                 byte2 = cdb->byte2;
5488                 break;
5489         }
5490         case READ_BUFFER_16: {
5491                 struct scsi_read_buffer_16 *cdb;
5492
5493                 cdb = (struct scsi_read_buffer_16 *)ctsio->cdb;
5494                 buffer_offset = scsi_8btou64(cdb->offset);
5495                 len = scsi_4btoul(cdb->length);
5496                 byte2 = cdb->byte2;
5497                 break;
5498         }
5499         default: /* This shouldn't happen. */
5500                 ctl_set_invalid_opcode(ctsio);
5501                 ctl_done((union ctl_io *)ctsio);
5502                 return (CTL_RETVAL_COMPLETE);
5503         }
5504
5505         if ((byte2 & RWB_MODE) != RWB_MODE_DATA &&
5506             (byte2 & RWB_MODE) != RWB_MODE_ECHO_DESCR &&
5507             (byte2 & RWB_MODE) != RWB_MODE_DESCR) {
5508                 ctl_set_invalid_field(ctsio,
5509                                       /*sks_valid*/ 1,
5510                                       /*command*/ 1,
5511                                       /*field*/ 1,
5512                                       /*bit_valid*/ 1,
5513                                       /*bit*/ 4);
5514                 ctl_done((union ctl_io *)ctsio);
5515                 return (CTL_RETVAL_COMPLETE);
5516         }
5517
5518         if (buffer_offset > CTL_WRITE_BUFFER_SIZE ||
5519             buffer_offset + len > CTL_WRITE_BUFFER_SIZE) {
5520                 ctl_set_invalid_field(ctsio,
5521                                       /*sks_valid*/ 1,
5522                                       /*command*/ 1,
5523                                       /*field*/ 6,
5524                                       /*bit_valid*/ 0,
5525                                       /*bit*/ 0);
5526                 ctl_done((union ctl_io *)ctsio);
5527                 return (CTL_RETVAL_COMPLETE);
5528         }
5529
5530         if ((byte2 & RWB_MODE) == RWB_MODE_DESCR) {
5531                 descr[0] = 0;
5532                 scsi_ulto3b(CTL_WRITE_BUFFER_SIZE, &descr[1]);
5533                 ctsio->kern_data_ptr = descr;
5534                 len = min(len, sizeof(descr));
5535         } else if ((byte2 & RWB_MODE) == RWB_MODE_ECHO_DESCR) {
5536                 ctsio->kern_data_ptr = echo_descr;
5537                 len = min(len, sizeof(echo_descr));
5538         } else {
5539                 if (lun->write_buffer == NULL) {
5540                         lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE,
5541                             M_CTL, M_WAITOK);
5542                 }
5543                 ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5544         }
5545         ctsio->kern_data_len = len;
5546         ctsio->kern_total_len = len;
5547         ctsio->kern_data_resid = 0;
5548         ctsio->kern_rel_offset = 0;
5549         ctsio->kern_sg_entries = 0;
5550         ctl_set_success(ctsio);
5551         ctsio->be_move_done = ctl_config_move_done;
5552         ctl_datamove((union ctl_io *)ctsio);
5553         return (CTL_RETVAL_COMPLETE);
5554 }
5555
5556 int
5557 ctl_write_buffer(struct ctl_scsiio *ctsio)
5558 {
5559         struct scsi_write_buffer *cdb;
5560         struct ctl_lun *lun;
5561         int buffer_offset, len;
5562
5563         CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5564
5565         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5566         cdb = (struct scsi_write_buffer *)ctsio->cdb;
5567
5568         if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA) {
5569                 ctl_set_invalid_field(ctsio,
5570                                       /*sks_valid*/ 1,
5571                                       /*command*/ 1,
5572                                       /*field*/ 1,
5573                                       /*bit_valid*/ 1,
5574                                       /*bit*/ 4);
5575                 ctl_done((union ctl_io *)ctsio);
5576                 return (CTL_RETVAL_COMPLETE);
5577         }
5578
5579         len = scsi_3btoul(cdb->length);
5580         buffer_offset = scsi_3btoul(cdb->offset);
5581
5582         if (buffer_offset + len > CTL_WRITE_BUFFER_SIZE) {
5583                 ctl_set_invalid_field(ctsio,
5584                                       /*sks_valid*/ 1,
5585                                       /*command*/ 1,
5586                                       /*field*/ 6,
5587                                       /*bit_valid*/ 0,
5588                                       /*bit*/ 0);
5589                 ctl_done((union ctl_io *)ctsio);
5590                 return (CTL_RETVAL_COMPLETE);
5591         }
5592
5593         /*
5594          * If we've got a kernel request that hasn't been malloced yet,
5595          * malloc it and tell the caller the data buffer is here.
5596          */
5597         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5598                 if (lun->write_buffer == NULL) {
5599                         lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE,
5600                             M_CTL, M_WAITOK);
5601                 }
5602                 ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5603                 ctsio->kern_data_len = len;
5604                 ctsio->kern_total_len = len;
5605                 ctsio->kern_data_resid = 0;
5606                 ctsio->kern_rel_offset = 0;
5607                 ctsio->kern_sg_entries = 0;
5608                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5609                 ctsio->be_move_done = ctl_config_move_done;
5610                 ctl_datamove((union ctl_io *)ctsio);
5611
5612                 return (CTL_RETVAL_COMPLETE);
5613         }
5614
5615         ctl_set_success(ctsio);
5616         ctl_done((union ctl_io *)ctsio);
5617         return (CTL_RETVAL_COMPLETE);
5618 }
5619
5620 int
5621 ctl_write_same(struct ctl_scsiio *ctsio)
5622 {
5623         struct ctl_lun *lun;
5624         struct ctl_lba_len_flags *lbalen;
5625         uint64_t lba;
5626         uint32_t num_blocks;
5627         int len, retval;
5628         uint8_t byte2;
5629
5630         CTL_DEBUG_PRINT(("ctl_write_same\n"));
5631
5632         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5633
5634         switch (ctsio->cdb[0]) {
5635         case WRITE_SAME_10: {
5636                 struct scsi_write_same_10 *cdb;
5637
5638                 cdb = (struct scsi_write_same_10 *)ctsio->cdb;
5639
5640                 lba = scsi_4btoul(cdb->addr);
5641                 num_blocks = scsi_2btoul(cdb->length);
5642                 byte2 = cdb->byte2;
5643                 break;
5644         }
5645         case WRITE_SAME_16: {
5646                 struct scsi_write_same_16 *cdb;
5647
5648                 cdb = (struct scsi_write_same_16 *)ctsio->cdb;
5649
5650                 lba = scsi_8btou64(cdb->addr);
5651                 num_blocks = scsi_4btoul(cdb->length);
5652                 byte2 = cdb->byte2;
5653                 break;
5654         }
5655         default:
5656                 /*
5657                  * We got a command we don't support.  This shouldn't
5658                  * happen, commands should be filtered out above us.
5659                  */
5660                 ctl_set_invalid_opcode(ctsio);
5661                 ctl_done((union ctl_io *)ctsio);
5662
5663                 return (CTL_RETVAL_COMPLETE);
5664                 break; /* NOTREACHED */
5665         }
5666
5667         /* ANCHOR flag can be used only together with UNMAP */
5668         if ((byte2 & SWS_UNMAP) == 0 && (byte2 & SWS_ANCHOR) != 0) {
5669                 ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
5670                     /*command*/ 1, /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 0);
5671                 ctl_done((union ctl_io *)ctsio);
5672                 return (CTL_RETVAL_COMPLETE);
5673         }
5674
5675         /*
5676          * The first check is to make sure we're in bounds, the second
5677          * check is to catch wrap-around problems.  If the lba + num blocks
5678          * is less than the lba, then we've wrapped around and the block
5679          * range is invalid anyway.
5680          */
5681         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5682          || ((lba + num_blocks) < lba)) {
5683                 ctl_set_lba_out_of_range(ctsio);
5684                 ctl_done((union ctl_io *)ctsio);
5685                 return (CTL_RETVAL_COMPLETE);
5686         }
5687
5688         /* Zero number of blocks means "to the last logical block" */
5689         if (num_blocks == 0) {
5690                 if ((lun->be_lun->maxlba + 1) - lba > UINT32_MAX) {
5691                         ctl_set_invalid_field(ctsio,
5692                                               /*sks_valid*/ 0,
5693                                               /*command*/ 1,
5694                                               /*field*/ 0,
5695                                               /*bit_valid*/ 0,
5696                                               /*bit*/ 0);
5697                         ctl_done((union ctl_io *)ctsio);
5698                         return (CTL_RETVAL_COMPLETE);
5699                 }
5700                 num_blocks = (lun->be_lun->maxlba + 1) - lba;
5701         }
5702
5703         len = lun->be_lun->blocksize;
5704
5705         /*
5706          * If we've got a kernel request that hasn't been malloced yet,
5707          * malloc it and tell the caller the data buffer is here.
5708          */
5709         if ((byte2 & SWS_NDOB) == 0 &&
5710             (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5711                 ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);
5712                 ctsio->kern_data_len = len;
5713                 ctsio->kern_total_len = len;
5714                 ctsio->kern_data_resid = 0;
5715                 ctsio->kern_rel_offset = 0;
5716                 ctsio->kern_sg_entries = 0;
5717                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5718                 ctsio->be_move_done = ctl_config_move_done;
5719                 ctl_datamove((union ctl_io *)ctsio);
5720
5721                 return (CTL_RETVAL_COMPLETE);
5722         }
5723
5724         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5725         lbalen->lba = lba;
5726         lbalen->len = num_blocks;
5727         lbalen->flags = byte2;
5728         retval = lun->backend->config_write((union ctl_io *)ctsio);
5729
5730         return (retval);
5731 }
5732
5733 int
5734 ctl_unmap(struct ctl_scsiio *ctsio)
5735 {
5736         struct ctl_lun *lun;
5737         struct scsi_unmap *cdb;
5738         struct ctl_ptr_len_flags *ptrlen;
5739         struct scsi_unmap_header *hdr;
5740         struct scsi_unmap_desc *buf, *end, *endnz, *range;
5741         uint64_t lba;
5742         uint32_t num_blocks;
5743         int len, retval;
5744         uint8_t byte2;
5745
5746         CTL_DEBUG_PRINT(("ctl_unmap\n"));
5747
5748         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5749         cdb = (struct scsi_unmap *)ctsio->cdb;
5750
5751         len = scsi_2btoul(cdb->length);
5752         byte2 = cdb->byte2;
5753
5754         /*
5755          * If we've got a kernel request that hasn't been malloced yet,
5756          * malloc it and tell the caller the data buffer is here.
5757          */
5758         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5759                 ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);
5760                 ctsio->kern_data_len = len;
5761                 ctsio->kern_total_len = len;
5762                 ctsio->kern_data_resid = 0;
5763                 ctsio->kern_rel_offset = 0;
5764                 ctsio->kern_sg_entries = 0;
5765                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5766                 ctsio->be_move_done = ctl_config_move_done;
5767                 ctl_datamove((union ctl_io *)ctsio);
5768
5769                 return (CTL_RETVAL_COMPLETE);
5770         }
5771
5772         len = ctsio->kern_total_len - ctsio->kern_data_resid;
5773         hdr = (struct scsi_unmap_header *)ctsio->kern_data_ptr;
5774         if (len < sizeof (*hdr) ||
5775             len < (scsi_2btoul(hdr->length) + sizeof(hdr->length)) ||
5776             len < (scsi_2btoul(hdr->desc_length) + sizeof (*hdr)) ||
5777             scsi_2btoul(hdr->desc_length) % sizeof(*buf) != 0) {
5778                 ctl_set_invalid_field(ctsio,
5779                                       /*sks_valid*/ 0,
5780                                       /*command*/ 0,
5781                                       /*field*/ 0,
5782                                       /*bit_valid*/ 0,
5783                                       /*bit*/ 0);
5784                 goto done;
5785         }
5786         len = scsi_2btoul(hdr->desc_length);
5787         buf = (struct scsi_unmap_desc *)(hdr + 1);
5788         end = buf + len / sizeof(*buf);
5789
5790         endnz = buf;
5791         for (range = buf; range < end; range++) {
5792                 lba = scsi_8btou64(range->lba);
5793                 num_blocks = scsi_4btoul(range->length);
5794                 if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5795                  || ((lba + num_blocks) < lba)) {
5796                         ctl_set_lba_out_of_range(ctsio);
5797                         ctl_done((union ctl_io *)ctsio);
5798                         return (CTL_RETVAL_COMPLETE);
5799                 }
5800                 if (num_blocks != 0)
5801                         endnz = range + 1;
5802         }
5803
5804         /*
5805          * Block backend can not handle zero last range.
5806          * Filter it out and return if there is nothing left.
5807          */
5808         len = (uint8_t *)endnz - (uint8_t *)buf;
5809         if (len == 0) {
5810                 ctl_set_success(ctsio);
5811                 goto done;
5812         }
5813
5814         mtx_lock(&lun->lun_lock);
5815         ptrlen = (struct ctl_ptr_len_flags *)
5816             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5817         ptrlen->ptr = (void *)buf;
5818         ptrlen->len = len;
5819         ptrlen->flags = byte2;
5820         ctl_check_blocked(lun);
5821         mtx_unlock(&lun->lun_lock);
5822
5823         retval = lun->backend->config_write((union ctl_io *)ctsio);
5824         return (retval);
5825
5826 done:
5827         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5828                 free(ctsio->kern_data_ptr, M_CTL);
5829                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5830         }
5831         ctl_done((union ctl_io *)ctsio);
5832         return (CTL_RETVAL_COMPLETE);
5833 }
5834
5835 /*
5836  * Note that this function currently doesn't actually do anything inside
5837  * CTL to enforce things if the DQue bit is turned on.
5838  *
5839  * Also note that this function can't be used in the default case, because
5840  * the DQue bit isn't set in the changeable mask for the control mode page
5841  * anyway.  This is just here as an example for how to implement a page
5842  * handler, and a placeholder in case we want to allow the user to turn
5843  * tagged queueing on and off.
5844  *
5845  * The D_SENSE bit handling is functional, however, and will turn
5846  * descriptor sense on and off for a given LUN.
5847  */
5848 int
5849 ctl_control_page_handler(struct ctl_scsiio *ctsio,
5850                          struct ctl_page_index *page_index, uint8_t *page_ptr)
5851 {
5852         struct scsi_control_page *current_cp, *saved_cp, *user_cp;
5853         struct ctl_lun *lun;
5854         int set_ua;
5855         uint32_t initidx;
5856
5857         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5858         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5859         set_ua = 0;
5860
5861         user_cp = (struct scsi_control_page *)page_ptr;
5862         current_cp = (struct scsi_control_page *)
5863                 (page_index->page_data + (page_index->page_len *
5864                 CTL_PAGE_CURRENT));
5865         saved_cp = (struct scsi_control_page *)
5866                 (page_index->page_data + (page_index->page_len *
5867                 CTL_PAGE_SAVED));
5868
5869         mtx_lock(&lun->lun_lock);
5870         if (((current_cp->rlec & SCP_DSENSE) == 0)
5871          && ((user_cp->rlec & SCP_DSENSE) != 0)) {
5872                 /*
5873                  * Descriptor sense is currently turned off and the user
5874                  * wants to turn it on.
5875                  */
5876                 current_cp->rlec |= SCP_DSENSE;
5877                 saved_cp->rlec |= SCP_DSENSE;
5878                 lun->flags |= CTL_LUN_SENSE_DESC;
5879                 set_ua = 1;
5880         } else if (((current_cp->rlec & SCP_DSENSE) != 0)
5881                 && ((user_cp->rlec & SCP_DSENSE) == 0)) {
5882                 /*
5883                  * Descriptor sense is currently turned on, and the user
5884                  * wants to turn it off.
5885                  */
5886                 current_cp->rlec &= ~SCP_DSENSE;
5887                 saved_cp->rlec &= ~SCP_DSENSE;
5888                 lun->flags &= ~CTL_LUN_SENSE_DESC;
5889                 set_ua = 1;
5890         }
5891         if ((current_cp->queue_flags & SCP_QUEUE_ALG_MASK) !=
5892             (user_cp->queue_flags & SCP_QUEUE_ALG_MASK)) {
5893                 current_cp->queue_flags &= ~SCP_QUEUE_ALG_MASK;
5894                 current_cp->queue_flags |= user_cp->queue_flags & SCP_QUEUE_ALG_MASK;
5895                 saved_cp->queue_flags &= ~SCP_QUEUE_ALG_MASK;
5896                 saved_cp->queue_flags |= user_cp->queue_flags & SCP_QUEUE_ALG_MASK;
5897                 set_ua = 1;
5898         }
5899         if ((current_cp->eca_and_aen & SCP_SWP) !=
5900             (user_cp->eca_and_aen & SCP_SWP)) {
5901                 current_cp->eca_and_aen &= ~SCP_SWP;
5902                 current_cp->eca_and_aen |= user_cp->eca_and_aen & SCP_SWP;
5903                 saved_cp->eca_and_aen &= ~SCP_SWP;
5904                 saved_cp->eca_and_aen |= user_cp->eca_and_aen & SCP_SWP;
5905                 set_ua = 1;
5906         }
5907         if (set_ua != 0)
5908                 ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
5909         mtx_unlock(&lun->lun_lock);
5910         if (set_ua) {
5911                 ctl_isc_announce_mode(lun,
5912                     ctl_get_initindex(&ctsio->io_hdr.nexus),
5913                     page_index->page_code, page_index->subpage);
5914         }
5915         return (0);
5916 }
5917
5918 int
5919 ctl_caching_sp_handler(struct ctl_scsiio *ctsio,
5920                      struct ctl_page_index *page_index, uint8_t *page_ptr)
5921 {
5922         struct scsi_caching_page *current_cp, *saved_cp, *user_cp;
5923         struct ctl_lun *lun;
5924         int set_ua;
5925         uint32_t initidx;
5926
5927         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5928         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5929         set_ua = 0;
5930
5931         user_cp = (struct scsi_caching_page *)page_ptr;
5932         current_cp = (struct scsi_caching_page *)
5933                 (page_index->page_data + (page_index->page_len *
5934                 CTL_PAGE_CURRENT));
5935         saved_cp = (struct scsi_caching_page *)
5936                 (page_index->page_data + (page_index->page_len *
5937                 CTL_PAGE_SAVED));
5938
5939         mtx_lock(&lun->lun_lock);
5940         if ((current_cp->flags1 & (SCP_WCE | SCP_RCD)) !=
5941             (user_cp->flags1 & (SCP_WCE | SCP_RCD))) {
5942                 current_cp->flags1 &= ~(SCP_WCE | SCP_RCD);
5943                 current_cp->flags1 |= user_cp->flags1 & (SCP_WCE | SCP_RCD);
5944                 saved_cp->flags1 &= ~(SCP_WCE | SCP_RCD);
5945                 saved_cp->flags1 |= user_cp->flags1 & (SCP_WCE | SCP_RCD);
5946                 set_ua = 1;
5947         }
5948         if (set_ua != 0)
5949                 ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
5950         mtx_unlock(&lun->lun_lock);
5951         if (set_ua) {
5952                 ctl_isc_announce_mode(lun,
5953                     ctl_get_initindex(&ctsio->io_hdr.nexus),
5954                     page_index->page_code, page_index->subpage);
5955         }
5956         return (0);
5957 }
5958
5959 int
5960 ctl_debugconf_sp_select_handler(struct ctl_scsiio *ctsio,
5961                                 struct ctl_page_index *page_index,
5962                                 uint8_t *page_ptr)
5963 {
5964         uint8_t *c;
5965         int i;
5966
5967         c = ((struct copan_debugconf_subpage *)page_ptr)->ctl_time_io_secs;
5968         ctl_time_io_secs =
5969                 (c[0] << 8) |
5970                 (c[1] << 0) |
5971                 0;
5972         CTL_DEBUG_PRINT(("set ctl_time_io_secs to %d\n", ctl_time_io_secs));
5973         printf("set ctl_time_io_secs to %d\n", ctl_time_io_secs);
5974         printf("page data:");
5975         for (i=0; i<8; i++)
5976                 printf(" %.2x",page_ptr[i]);
5977         printf("\n");
5978         return (0);
5979 }
5980
5981 int
5982 ctl_debugconf_sp_sense_handler(struct ctl_scsiio *ctsio,
5983                                struct ctl_page_index *page_index,
5984                                int pc)
5985 {
5986         struct copan_debugconf_subpage *page;
5987
5988         page = (struct copan_debugconf_subpage *)page_index->page_data +
5989                 (page_index->page_len * pc);
5990
5991         switch (pc) {
5992         case SMS_PAGE_CTRL_CHANGEABLE >> 6:
5993         case SMS_PAGE_CTRL_DEFAULT >> 6:
5994         case SMS_PAGE_CTRL_SAVED >> 6:
5995                 /*
5996                  * We don't update the changeable or default bits for this page.
5997                  */
5998                 break;
5999         case SMS_PAGE_CTRL_CURRENT >> 6:
6000                 page->ctl_time_io_secs[0] = ctl_time_io_secs >> 8;
6001                 page->ctl_time_io_secs[1] = ctl_time_io_secs >> 0;
6002                 break;
6003         default:
6004                 break;
6005         }
6006         return (0);
6007 }
6008
6009
6010 static int
6011 ctl_do_mode_select(union ctl_io *io)
6012 {
6013         struct scsi_mode_page_header *page_header;
6014         struct ctl_page_index *page_index;
6015         struct ctl_scsiio *ctsio;
6016         int page_len, page_len_offset, page_len_size;
6017         union ctl_modepage_info *modepage_info;
6018         struct ctl_lun *lun;
6019         int *len_left, *len_used;
6020         int retval, i;
6021
6022         ctsio = &io->scsiio;
6023         page_index = NULL;
6024         page_len = 0;
6025         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6026
6027         modepage_info = (union ctl_modepage_info *)
6028                 ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6029         len_left = &modepage_info->header.len_left;
6030         len_used = &modepage_info->header.len_used;
6031
6032 do_next_page:
6033
6034         page_header = (struct scsi_mode_page_header *)
6035                 (ctsio->kern_data_ptr + *len_used);
6036
6037         if (*len_left == 0) {
6038                 free(ctsio->kern_data_ptr, M_CTL);
6039                 ctl_set_success(ctsio);
6040                 ctl_done((union ctl_io *)ctsio);
6041                 return (CTL_RETVAL_COMPLETE);
6042         } else if (*len_left < sizeof(struct scsi_mode_page_header)) {
6043
6044                 free(ctsio->kern_data_ptr, M_CTL);
6045                 ctl_set_param_len_error(ctsio);
6046                 ctl_done((union ctl_io *)ctsio);
6047                 return (CTL_RETVAL_COMPLETE);
6048
6049         } else if ((page_header->page_code & SMPH_SPF)
6050                 && (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
6051
6052                 free(ctsio->kern_data_ptr, M_CTL);
6053                 ctl_set_param_len_error(ctsio);
6054                 ctl_done((union ctl_io *)ctsio);
6055                 return (CTL_RETVAL_COMPLETE);
6056         }
6057
6058
6059         /*
6060          * XXX KDM should we do something with the block descriptor?
6061          */
6062         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6063                 page_index = &lun->mode_pages.index[i];
6064                 if (lun->be_lun->lun_type == T_DIRECT &&
6065                     (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6066                         continue;
6067                 if (lun->be_lun->lun_type == T_PROCESSOR &&
6068                     (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6069                         continue;
6070                 if (lun->be_lun->lun_type == T_CDROM &&
6071                     (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6072                         continue;
6073
6074                 if ((page_index->page_code & SMPH_PC_MASK) !=
6075                     (page_header->page_code & SMPH_PC_MASK))
6076                         continue;
6077
6078                 /*
6079                  * If neither page has a subpage code, then we've got a
6080                  * match.
6081                  */
6082                 if (((page_index->page_code & SMPH_SPF) == 0)
6083                  && ((page_header->page_code & SMPH_SPF) == 0)) {
6084                         page_len = page_header->page_length;
6085                         break;
6086                 }
6087
6088                 /*
6089                  * If both pages have subpages, then the subpage numbers
6090                  * have to match.
6091                  */
6092                 if ((page_index->page_code & SMPH_SPF)
6093                   && (page_header->page_code & SMPH_SPF)) {
6094                         struct scsi_mode_page_header_sp *sph;
6095
6096                         sph = (struct scsi_mode_page_header_sp *)page_header;
6097                         if (page_index->subpage == sph->subpage) {
6098                                 page_len = scsi_2btoul(sph->page_length);
6099                                 break;
6100                         }
6101                 }
6102         }
6103
6104         /*
6105          * If we couldn't find the page, or if we don't have a mode select
6106          * handler for it, send back an error to the user.
6107          */
6108         if ((i >= CTL_NUM_MODE_PAGES)
6109          || (page_index->select_handler == NULL)) {
6110                 ctl_set_invalid_field(ctsio,
6111                                       /*sks_valid*/ 1,
6112                                       /*command*/ 0,
6113                                       /*field*/ *len_used,
6114                                       /*bit_valid*/ 0,
6115                                       /*bit*/ 0);
6116                 free(ctsio->kern_data_ptr, M_CTL);
6117                 ctl_done((union ctl_io *)ctsio);
6118                 return (CTL_RETVAL_COMPLETE);
6119         }
6120
6121         if (page_index->page_code & SMPH_SPF) {
6122                 page_len_offset = 2;
6123                 page_len_size = 2;
6124         } else {
6125                 page_len_size = 1;
6126                 page_len_offset = 1;
6127         }
6128
6129         /*
6130          * If the length the initiator gives us isn't the one we specify in
6131          * the mode page header, or if they didn't specify enough data in
6132          * the CDB to avoid truncating this page, kick out the request.
6133          */
6134         if ((page_len != (page_index->page_len - page_len_offset -
6135                           page_len_size))
6136          || (*len_left < page_index->page_len)) {
6137
6138
6139                 ctl_set_invalid_field(ctsio,
6140                                       /*sks_valid*/ 1,
6141                                       /*command*/ 0,
6142                                       /*field*/ *len_used + page_len_offset,
6143                                       /*bit_valid*/ 0,
6144                                       /*bit*/ 0);
6145                 free(ctsio->kern_data_ptr, M_CTL);
6146                 ctl_done((union ctl_io *)ctsio);
6147                 return (CTL_RETVAL_COMPLETE);
6148         }
6149
6150         /*
6151          * Run through the mode page, checking to make sure that the bits
6152          * the user changed are actually legal for him to change.
6153          */
6154         for (i = 0; i < page_index->page_len; i++) {
6155                 uint8_t *user_byte, *change_mask, *current_byte;
6156                 int bad_bit;
6157                 int j;
6158
6159                 user_byte = (uint8_t *)page_header + i;
6160                 change_mask = page_index->page_data +
6161                               (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
6162                 current_byte = page_index->page_data +
6163                                (page_index->page_len * CTL_PAGE_CURRENT) + i;
6164
6165                 /*
6166                  * Check to see whether the user set any bits in this byte
6167                  * that he is not allowed to set.
6168                  */
6169                 if ((*user_byte & ~(*change_mask)) ==
6170                     (*current_byte & ~(*change_mask)))
6171                         continue;
6172
6173                 /*
6174                  * Go through bit by bit to determine which one is illegal.
6175                  */
6176                 bad_bit = 0;
6177                 for (j = 7; j >= 0; j--) {
6178                         if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
6179                             (((1 << i) & ~(*change_mask)) & *current_byte)) {
6180                                 bad_bit = i;
6181                                 break;
6182                         }
6183                 }
6184                 ctl_set_invalid_field(ctsio,
6185                                       /*sks_valid*/ 1,
6186                                       /*command*/ 0,
6187                                       /*field*/ *len_used + i,
6188                                       /*bit_valid*/ 1,
6189                                       /*bit*/ bad_bit);
6190                 free(ctsio->kern_data_ptr, M_CTL);
6191                 ctl_done((union ctl_io *)ctsio);
6192                 return (CTL_RETVAL_COMPLETE);
6193         }
6194
6195         /*
6196          * Decrement these before we call the page handler, since we may
6197          * end up getting called back one way or another before the handler
6198          * returns to this context.
6199          */
6200         *len_left -= page_index->page_len;
6201         *len_used += page_index->page_len;
6202
6203         retval = page_index->select_handler(ctsio, page_index,
6204                                             (uint8_t *)page_header);
6205
6206         /*
6207          * If the page handler returns CTL_RETVAL_QUEUED, then we need to
6208          * wait until this queued command completes to finish processing
6209          * the mode page.  If it returns anything other than
6210          * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
6211          * already set the sense information, freed the data pointer, and
6212          * completed the io for us.
6213          */
6214         if (retval != CTL_RETVAL_COMPLETE)
6215                 goto bailout_no_done;
6216
6217         /*
6218          * If the initiator sent us more than one page, parse the next one.
6219          */
6220         if (*len_left > 0)
6221                 goto do_next_page;
6222
6223         ctl_set_success(ctsio);
6224         free(ctsio->kern_data_ptr, M_CTL);
6225         ctl_done((union ctl_io *)ctsio);
6226
6227 bailout_no_done:
6228
6229         return (CTL_RETVAL_COMPLETE);
6230
6231 }
6232
6233 int
6234 ctl_mode_select(struct ctl_scsiio *ctsio)
6235 {
6236         int param_len, pf, sp;
6237         int header_size, bd_len;
6238         union ctl_modepage_info *modepage_info;
6239
6240         switch (ctsio->cdb[0]) {
6241         case MODE_SELECT_6: {
6242                 struct scsi_mode_select_6 *cdb;
6243
6244                 cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
6245
6246                 pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6247                 sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6248                 param_len = cdb->length;
6249                 header_size = sizeof(struct scsi_mode_header_6);
6250                 break;
6251         }
6252         case MODE_SELECT_10: {
6253                 struct scsi_mode_select_10 *cdb;
6254
6255                 cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
6256
6257                 pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6258                 sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6259                 param_len = scsi_2btoul(cdb->length);
6260                 header_size = sizeof(struct scsi_mode_header_10);
6261                 break;
6262         }
6263         default:
6264                 ctl_set_invalid_opcode(ctsio);
6265                 ctl_done((union ctl_io *)ctsio);
6266                 return (CTL_RETVAL_COMPLETE);
6267         }
6268
6269         /*
6270          * From SPC-3:
6271          * "A parameter list length of zero indicates that the Data-Out Buffer
6272          * shall be empty. This condition shall not be considered as an error."
6273          */
6274         if (param_len == 0) {
6275                 ctl_set_success(ctsio);
6276                 ctl_done((union ctl_io *)ctsio);
6277                 return (CTL_RETVAL_COMPLETE);
6278         }
6279
6280         /*
6281          * Since we'll hit this the first time through, prior to
6282          * allocation, we don't need to free a data buffer here.
6283          */
6284         if (param_len < header_size) {
6285                 ctl_set_param_len_error(ctsio);
6286                 ctl_done((union ctl_io *)ctsio);
6287                 return (CTL_RETVAL_COMPLETE);
6288         }
6289
6290         /*
6291          * Allocate the data buffer and grab the user's data.  In theory,
6292          * we shouldn't have to sanity check the parameter list length here
6293          * because the maximum size is 64K.  We should be able to malloc
6294          * that much without too many problems.
6295          */
6296         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6297                 ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6298                 ctsio->kern_data_len = param_len;
6299                 ctsio->kern_total_len = param_len;
6300                 ctsio->kern_data_resid = 0;
6301                 ctsio->kern_rel_offset = 0;
6302                 ctsio->kern_sg_entries = 0;
6303                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6304                 ctsio->be_move_done = ctl_config_move_done;
6305                 ctl_datamove((union ctl_io *)ctsio);
6306
6307                 return (CTL_RETVAL_COMPLETE);
6308         }
6309
6310         switch (ctsio->cdb[0]) {
6311         case MODE_SELECT_6: {
6312                 struct scsi_mode_header_6 *mh6;
6313
6314                 mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6315                 bd_len = mh6->blk_desc_len;
6316                 break;
6317         }
6318         case MODE_SELECT_10: {
6319                 struct scsi_mode_header_10 *mh10;
6320
6321                 mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6322                 bd_len = scsi_2btoul(mh10->blk_desc_len);
6323                 break;
6324         }
6325         default:
6326                 panic("%s: Invalid CDB type %#x", __func__, ctsio->cdb[0]);
6327         }
6328
6329         if (param_len < (header_size + bd_len)) {
6330                 free(ctsio->kern_data_ptr, M_CTL);
6331                 ctl_set_param_len_error(ctsio);
6332                 ctl_done((union ctl_io *)ctsio);
6333                 return (CTL_RETVAL_COMPLETE);
6334         }
6335
6336         /*
6337          * Set the IO_CONT flag, so that if this I/O gets passed to
6338          * ctl_config_write_done(), it'll get passed back to
6339          * ctl_do_mode_select() for further processing, or completion if
6340          * we're all done.
6341          */
6342         ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6343         ctsio->io_cont = ctl_do_mode_select;
6344
6345         modepage_info = (union ctl_modepage_info *)
6346                 ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6347         memset(modepage_info, 0, sizeof(*modepage_info));
6348         modepage_info->header.len_left = param_len - header_size - bd_len;
6349         modepage_info->header.len_used = header_size + bd_len;
6350
6351         return (ctl_do_mode_select((union ctl_io *)ctsio));
6352 }
6353
6354 int
6355 ctl_mode_sense(struct ctl_scsiio *ctsio)
6356 {
6357         struct ctl_lun *lun;
6358         int pc, page_code, dbd, llba, subpage;
6359         int alloc_len, page_len, header_len, total_len;
6360         struct scsi_mode_block_descr *block_desc;
6361         struct ctl_page_index *page_index;
6362
6363         dbd = 0;
6364         llba = 0;
6365         block_desc = NULL;
6366
6367         CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6368
6369         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6370         switch (ctsio->cdb[0]) {
6371         case MODE_SENSE_6: {
6372                 struct scsi_mode_sense_6 *cdb;
6373
6374                 cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6375
6376                 header_len = sizeof(struct scsi_mode_hdr_6);
6377                 if (cdb->byte2 & SMS_DBD)
6378                         dbd = 1;
6379                 else
6380                         header_len += sizeof(struct scsi_mode_block_descr);
6381
6382                 pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6383                 page_code = cdb->page & SMS_PAGE_CODE;
6384                 subpage = cdb->subpage;
6385                 alloc_len = cdb->length;
6386                 break;
6387         }
6388         case MODE_SENSE_10: {
6389                 struct scsi_mode_sense_10 *cdb;
6390
6391                 cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6392
6393                 header_len = sizeof(struct scsi_mode_hdr_10);
6394
6395                 if (cdb->byte2 & SMS_DBD)
6396                         dbd = 1;
6397                 else
6398                         header_len += sizeof(struct scsi_mode_block_descr);
6399                 if (cdb->byte2 & SMS10_LLBAA)
6400                         llba = 1;
6401                 pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6402                 page_code = cdb->page & SMS_PAGE_CODE;
6403                 subpage = cdb->subpage;
6404                 alloc_len = scsi_2btoul(cdb->length);
6405                 break;
6406         }
6407         default:
6408                 ctl_set_invalid_opcode(ctsio);
6409                 ctl_done((union ctl_io *)ctsio);
6410                 return (CTL_RETVAL_COMPLETE);
6411                 break; /* NOTREACHED */
6412         }
6413
6414         /*
6415          * We have to make a first pass through to calculate the size of
6416          * the pages that match the user's query.  Then we allocate enough
6417          * memory to hold it, and actually copy the data into the buffer.
6418          */
6419         switch (page_code) {
6420         case SMS_ALL_PAGES_PAGE: {
6421                 u_int i;
6422
6423                 page_len = 0;
6424
6425                 /*
6426                  * At the moment, values other than 0 and 0xff here are
6427                  * reserved according to SPC-3.
6428                  */
6429                 if ((subpage != SMS_SUBPAGE_PAGE_0)
6430                  && (subpage != SMS_SUBPAGE_ALL)) {
6431                         ctl_set_invalid_field(ctsio,
6432                                               /*sks_valid*/ 1,
6433                                               /*command*/ 1,
6434                                               /*field*/ 3,
6435                                               /*bit_valid*/ 0,
6436                                               /*bit*/ 0);
6437                         ctl_done((union ctl_io *)ctsio);
6438                         return (CTL_RETVAL_COMPLETE);
6439                 }
6440
6441                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6442                         page_index = &lun->mode_pages.index[i];
6443
6444                         /* Make sure the page is supported for this dev type */
6445                         if (lun->be_lun->lun_type == T_DIRECT &&
6446                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6447                                 continue;
6448                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6449                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6450                                 continue;
6451                         if (lun->be_lun->lun_type == T_CDROM &&
6452                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6453                                 continue;
6454
6455                         /*
6456                          * We don't use this subpage if the user didn't
6457                          * request all subpages.
6458                          */
6459                         if ((page_index->subpage != 0)
6460                          && (subpage == SMS_SUBPAGE_PAGE_0))
6461                                 continue;
6462
6463 #if 0
6464                         printf("found page %#x len %d\n",
6465                                page_index->page_code & SMPH_PC_MASK,
6466                                page_index->page_len);
6467 #endif
6468                         page_len += page_index->page_len;
6469                 }
6470                 break;
6471         }
6472         default: {
6473                 u_int i;
6474
6475                 page_len = 0;
6476
6477                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6478                         page_index = &lun->mode_pages.index[i];
6479
6480                         /* Make sure the page is supported for this dev type */
6481                         if (lun->be_lun->lun_type == T_DIRECT &&
6482                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6483                                 continue;
6484                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6485                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6486                                 continue;
6487                         if (lun->be_lun->lun_type == T_CDROM &&
6488                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6489                                 continue;
6490
6491                         /* Look for the right page code */
6492                         if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6493                                 continue;
6494
6495                         /* Look for the right subpage or the subpage wildcard*/
6496                         if ((page_index->subpage != subpage)
6497                          && (subpage != SMS_SUBPAGE_ALL))
6498                                 continue;
6499
6500 #if 0
6501                         printf("found page %#x len %d\n",
6502                                page_index->page_code & SMPH_PC_MASK,
6503                                page_index->page_len);
6504 #endif
6505
6506                         page_len += page_index->page_len;
6507                 }
6508
6509                 if (page_len == 0) {
6510                         ctl_set_invalid_field(ctsio,
6511                                               /*sks_valid*/ 1,
6512                                               /*command*/ 1,
6513                                               /*field*/ 2,
6514                                               /*bit_valid*/ 1,
6515                                               /*bit*/ 5);
6516                         ctl_done((union ctl_io *)ctsio);
6517                         return (CTL_RETVAL_COMPLETE);
6518                 }
6519                 break;
6520         }
6521         }
6522
6523         total_len = header_len + page_len;
6524 #if 0
6525         printf("header_len = %d, page_len = %d, total_len = %d\n",
6526                header_len, page_len, total_len);
6527 #endif
6528
6529         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6530         ctsio->kern_sg_entries = 0;
6531         ctsio->kern_data_resid = 0;
6532         ctsio->kern_rel_offset = 0;
6533         if (total_len < alloc_len) {
6534                 ctsio->residual = alloc_len - total_len;
6535                 ctsio->kern_data_len = total_len;
6536                 ctsio->kern_total_len = total_len;
6537         } else {
6538                 ctsio->residual = 0;
6539                 ctsio->kern_data_len = alloc_len;
6540                 ctsio->kern_total_len = alloc_len;
6541         }
6542
6543         switch (ctsio->cdb[0]) {
6544         case MODE_SENSE_6: {
6545                 struct scsi_mode_hdr_6 *header;
6546
6547                 header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
6548
6549                 header->datalen = MIN(total_len - 1, 254);
6550                 if (lun->be_lun->lun_type == T_DIRECT) {
6551                         header->dev_specific = 0x10; /* DPOFUA */
6552                         if ((lun->be_lun->flags & CTL_LUN_FLAG_READONLY) ||
6553                             (lun->mode_pages.control_page[CTL_PAGE_CURRENT]
6554                             .eca_and_aen & SCP_SWP) != 0)
6555                                     header->dev_specific |= 0x80; /* WP */
6556                 }
6557                 if (dbd)
6558                         header->block_descr_len = 0;
6559                 else
6560                         header->block_descr_len =
6561                                 sizeof(struct scsi_mode_block_descr);
6562                 block_desc = (struct scsi_mode_block_descr *)&header[1];
6563                 break;
6564         }
6565         case MODE_SENSE_10: {
6566                 struct scsi_mode_hdr_10 *header;
6567                 int datalen;
6568
6569                 header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
6570
6571                 datalen = MIN(total_len - 2, 65533);
6572                 scsi_ulto2b(datalen, header->datalen);
6573                 if (lun->be_lun->lun_type == T_DIRECT) {
6574                         header->dev_specific = 0x10; /* DPOFUA */
6575                         if ((lun->be_lun->flags & CTL_LUN_FLAG_READONLY) ||
6576                             (lun->mode_pages.control_page[CTL_PAGE_CURRENT]
6577                             .eca_and_aen & SCP_SWP) != 0)
6578                                     header->dev_specific |= 0x80; /* WP */
6579                 }
6580                 if (dbd)
6581                         scsi_ulto2b(0, header->block_descr_len);
6582                 else
6583                         scsi_ulto2b(sizeof(struct scsi_mode_block_descr),
6584                                     header->block_descr_len);
6585                 block_desc = (struct scsi_mode_block_descr *)&header[1];
6586                 break;
6587         }
6588         default:
6589                 panic("%s: Invalid CDB type %#x", __func__, ctsio->cdb[0]);
6590         }
6591
6592         /*
6593          * If we've got a disk, use its blocksize in the block
6594          * descriptor.  Otherwise, just set it to 0.
6595          */
6596         if (dbd == 0) {
6597                 if (lun->be_lun->lun_type == T_DIRECT)
6598                         scsi_ulto3b(lun->be_lun->blocksize,
6599                                     block_desc->block_len);
6600                 else
6601                         scsi_ulto3b(0, block_desc->block_len);
6602         }
6603
6604         switch (page_code) {
6605         case SMS_ALL_PAGES_PAGE: {
6606                 int i, data_used;
6607
6608                 data_used = header_len;
6609                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6610                         struct ctl_page_index *page_index;
6611
6612                         page_index = &lun->mode_pages.index[i];
6613                         if (lun->be_lun->lun_type == T_DIRECT &&
6614                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6615                                 continue;
6616                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6617                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6618                                 continue;
6619                         if (lun->be_lun->lun_type == T_CDROM &&
6620                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6621                                 continue;
6622
6623                         /*
6624                          * We don't use this subpage if the user didn't
6625                          * request all subpages.  We already checked (above)
6626                          * to make sure the user only specified a subpage
6627                          * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case.
6628                          */
6629                         if ((page_index->subpage != 0)
6630                          && (subpage == SMS_SUBPAGE_PAGE_0))
6631                                 continue;
6632
6633                         /*
6634                          * Call the handler, if it exists, to update the
6635                          * page to the latest values.
6636                          */
6637                         if (page_index->sense_handler != NULL)
6638                                 page_index->sense_handler(ctsio, page_index,pc);
6639
6640                         memcpy(ctsio->kern_data_ptr + data_used,
6641                                page_index->page_data +
6642                                (page_index->page_len * pc),
6643                                page_index->page_len);
6644                         data_used += page_index->page_len;
6645                 }
6646                 break;
6647         }
6648         default: {
6649                 int i, data_used;
6650
6651                 data_used = header_len;
6652
6653                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6654                         struct ctl_page_index *page_index;
6655
6656                         page_index = &lun->mode_pages.index[i];
6657
6658                         /* Look for the right page code */
6659                         if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6660                                 continue;
6661
6662                         /* Look for the right subpage or the subpage wildcard*/
6663                         if ((page_index->subpage != subpage)
6664                          && (subpage != SMS_SUBPAGE_ALL))
6665                                 continue;
6666
6667                         /* Make sure the page is supported for this dev type */
6668                         if (lun->be_lun->lun_type == T_DIRECT &&
6669                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6670                                 continue;
6671                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6672                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6673                                 continue;
6674                         if (lun->be_lun->lun_type == T_CDROM &&
6675                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6676                                 continue;
6677
6678                         /*
6679                          * Call the handler, if it exists, to update the
6680                          * page to the latest values.
6681                          */
6682                         if (page_index->sense_handler != NULL)
6683                                 page_index->sense_handler(ctsio, page_index,pc);
6684
6685                         memcpy(ctsio->kern_data_ptr + data_used,
6686                                page_index->page_data +
6687                                (page_index->page_len * pc),
6688                                page_index->page_len);
6689                         data_used += page_index->page_len;
6690                 }
6691                 break;
6692         }
6693         }
6694
6695         ctl_set_success(ctsio);
6696         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6697         ctsio->be_move_done = ctl_config_move_done;
6698         ctl_datamove((union ctl_io *)ctsio);
6699         return (CTL_RETVAL_COMPLETE);
6700 }
6701
6702 int
6703 ctl_lbp_log_sense_handler(struct ctl_scsiio *ctsio,
6704                                struct ctl_page_index *page_index,
6705                                int pc)
6706 {
6707         struct ctl_lun *lun;
6708         struct scsi_log_param_header *phdr;
6709         uint8_t *data;
6710         uint64_t val;
6711
6712         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6713         data = page_index->page_data;
6714
6715         if (lun->backend->lun_attr != NULL &&
6716             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "blocksavail"))
6717              != UINT64_MAX) {
6718                 phdr = (struct scsi_log_param_header *)data;
6719                 scsi_ulto2b(0x0001, phdr->param_code);
6720                 phdr->param_control = SLP_LBIN | SLP_LP;
6721                 phdr->param_len = 8;
6722                 data = (uint8_t *)(phdr + 1);
6723                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6724                 data[4] = 0x02; /* per-pool */
6725                 data += phdr->param_len;
6726         }
6727
6728         if (lun->backend->lun_attr != NULL &&
6729             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "blocksused"))
6730              != UINT64_MAX) {
6731                 phdr = (struct scsi_log_param_header *)data;
6732                 scsi_ulto2b(0x0002, phdr->param_code);
6733                 phdr->param_control = SLP_LBIN | SLP_LP;
6734                 phdr->param_len = 8;
6735                 data = (uint8_t *)(phdr + 1);
6736                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6737                 data[4] = 0x01; /* per-LUN */
6738                 data += phdr->param_len;
6739         }
6740
6741         if (lun->backend->lun_attr != NULL &&
6742             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "poolblocksavail"))
6743              != UINT64_MAX) {
6744                 phdr = (struct scsi_log_param_header *)data;
6745                 scsi_ulto2b(0x00f1, phdr->param_code);
6746                 phdr->param_control = SLP_LBIN | SLP_LP;
6747                 phdr->param_len = 8;
6748                 data = (uint8_t *)(phdr + 1);
6749                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6750                 data[4] = 0x02; /* per-pool */
6751                 data += phdr->param_len;
6752         }
6753
6754         if (lun->backend->lun_attr != NULL &&
6755             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "poolblocksused"))
6756              != UINT64_MAX) {
6757                 phdr = (struct scsi_log_param_header *)data;
6758                 scsi_ulto2b(0x00f2, 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         page_index->page_len = data - page_index->page_data;
6768         return (0);
6769 }
6770
6771 int
6772 ctl_sap_log_sense_handler(struct ctl_scsiio *ctsio,
6773                                struct ctl_page_index *page_index,
6774                                int pc)
6775 {
6776         struct ctl_lun *lun;
6777         struct stat_page *data;
6778         uint64_t rn, wn, rb, wb;
6779         struct bintime rt, wt;
6780         int i;
6781
6782         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6783         data = (struct stat_page *)page_index->page_data;
6784
6785         scsi_ulto2b(SLP_SAP, data->sap.hdr.param_code);
6786         data->sap.hdr.param_control = SLP_LBIN;
6787         data->sap.hdr.param_len = sizeof(struct scsi_log_stat_and_perf) -
6788             sizeof(struct scsi_log_param_header);
6789         rn = wn = rb = wb = 0;
6790         bintime_clear(&rt);
6791         bintime_clear(&wt);
6792         for (i = 0; i < CTL_MAX_PORTS; i++) {
6793                 rn += lun->stats.ports[i].operations[CTL_STATS_READ];
6794                 wn += lun->stats.ports[i].operations[CTL_STATS_WRITE];
6795                 rb += lun->stats.ports[i].bytes[CTL_STATS_READ];
6796                 wb += lun->stats.ports[i].bytes[CTL_STATS_WRITE];
6797                 bintime_add(&rt, &lun->stats.ports[i].time[CTL_STATS_READ]);
6798                 bintime_add(&wt, &lun->stats.ports[i].time[CTL_STATS_WRITE]);
6799         }
6800         scsi_u64to8b(rn, data->sap.read_num);
6801         scsi_u64to8b(wn, data->sap.write_num);
6802         if (lun->stats.blocksize > 0) {
6803                 scsi_u64to8b(wb / lun->stats.blocksize,
6804                     data->sap.recvieved_lba);
6805                 scsi_u64to8b(rb / lun->stats.blocksize,
6806                     data->sap.transmitted_lba);
6807         }
6808         scsi_u64to8b((uint64_t)rt.sec * 1000 + rt.frac / (UINT64_MAX / 1000),
6809             data->sap.read_int);
6810         scsi_u64to8b((uint64_t)wt.sec * 1000 + wt.frac / (UINT64_MAX / 1000),
6811             data->sap.write_int);
6812         scsi_u64to8b(0, data->sap.weighted_num);
6813         scsi_u64to8b(0, data->sap.weighted_int);
6814         scsi_ulto2b(SLP_IT, data->it.hdr.param_code);
6815         data->it.hdr.param_control = SLP_LBIN;
6816         data->it.hdr.param_len = sizeof(struct scsi_log_idle_time) -
6817             sizeof(struct scsi_log_param_header);
6818 #ifdef CTL_TIME_IO
6819         scsi_u64to8b(lun->idle_time / SBT_1MS, data->it.idle_int);
6820 #endif
6821         scsi_ulto2b(SLP_TI, data->ti.hdr.param_code);
6822         data->it.hdr.param_control = SLP_LBIN;
6823         data->ti.hdr.param_len = sizeof(struct scsi_log_time_interval) -
6824             sizeof(struct scsi_log_param_header);
6825         scsi_ulto4b(3, data->ti.exponent);
6826         scsi_ulto4b(1, data->ti.integer);
6827
6828         page_index->page_len = sizeof(*data);
6829         return (0);
6830 }
6831
6832 int
6833 ctl_log_sense(struct ctl_scsiio *ctsio)
6834 {
6835         struct ctl_lun *lun;
6836         int i, pc, page_code, subpage;
6837         int alloc_len, total_len;
6838         struct ctl_page_index *page_index;
6839         struct scsi_log_sense *cdb;
6840         struct scsi_log_header *header;
6841
6842         CTL_DEBUG_PRINT(("ctl_log_sense\n"));
6843
6844         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6845         cdb = (struct scsi_log_sense *)ctsio->cdb;
6846         pc = (cdb->page & SLS_PAGE_CTRL_MASK) >> 6;
6847         page_code = cdb->page & SLS_PAGE_CODE;
6848         subpage = cdb->subpage;
6849         alloc_len = scsi_2btoul(cdb->length);
6850
6851         page_index = NULL;
6852         for (i = 0; i < CTL_NUM_LOG_PAGES; i++) {
6853                 page_index = &lun->log_pages.index[i];
6854
6855                 /* Look for the right page code */
6856                 if ((page_index->page_code & SL_PAGE_CODE) != page_code)
6857                         continue;
6858
6859                 /* Look for the right subpage or the subpage wildcard*/
6860                 if (page_index->subpage != subpage)
6861                         continue;
6862
6863                 break;
6864         }
6865         if (i >= CTL_NUM_LOG_PAGES) {
6866                 ctl_set_invalid_field(ctsio,
6867                                       /*sks_valid*/ 1,
6868                                       /*command*/ 1,
6869                                       /*field*/ 2,
6870                                       /*bit_valid*/ 0,
6871                                       /*bit*/ 0);
6872                 ctl_done((union ctl_io *)ctsio);
6873                 return (CTL_RETVAL_COMPLETE);
6874         }
6875
6876         total_len = sizeof(struct scsi_log_header) + page_index->page_len;
6877
6878         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6879         ctsio->kern_sg_entries = 0;
6880         ctsio->kern_data_resid = 0;
6881         ctsio->kern_rel_offset = 0;
6882         if (total_len < alloc_len) {
6883                 ctsio->residual = alloc_len - total_len;
6884                 ctsio->kern_data_len = total_len;
6885                 ctsio->kern_total_len = total_len;
6886         } else {
6887                 ctsio->residual = 0;
6888                 ctsio->kern_data_len = alloc_len;
6889                 ctsio->kern_total_len = alloc_len;
6890         }
6891
6892         header = (struct scsi_log_header *)ctsio->kern_data_ptr;
6893         header->page = page_index->page_code;
6894         if (page_index->page_code == SLS_LOGICAL_BLOCK_PROVISIONING)
6895                 header->page |= SL_DS;
6896         if (page_index->subpage) {
6897                 header->page |= SL_SPF;
6898                 header->subpage = page_index->subpage;
6899         }
6900         scsi_ulto2b(page_index->page_len, header->datalen);
6901
6902         /*
6903          * Call the handler, if it exists, to update the
6904          * page to the latest values.
6905          */
6906         if (page_index->sense_handler != NULL)
6907                 page_index->sense_handler(ctsio, page_index, pc);
6908
6909         memcpy(header + 1, page_index->page_data, page_index->page_len);
6910
6911         ctl_set_success(ctsio);
6912         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6913         ctsio->be_move_done = ctl_config_move_done;
6914         ctl_datamove((union ctl_io *)ctsio);
6915         return (CTL_RETVAL_COMPLETE);
6916 }
6917
6918 int
6919 ctl_read_capacity(struct ctl_scsiio *ctsio)
6920 {
6921         struct scsi_read_capacity *cdb;
6922         struct scsi_read_capacity_data *data;
6923         struct ctl_lun *lun;
6924         uint32_t lba;
6925
6926         CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
6927
6928         cdb = (struct scsi_read_capacity *)ctsio->cdb;
6929
6930         lba = scsi_4btoul(cdb->addr);
6931         if (((cdb->pmi & SRC_PMI) == 0)
6932          && (lba != 0)) {
6933                 ctl_set_invalid_field(/*ctsio*/ ctsio,
6934                                       /*sks_valid*/ 1,
6935                                       /*command*/ 1,
6936                                       /*field*/ 2,
6937                                       /*bit_valid*/ 0,
6938                                       /*bit*/ 0);
6939                 ctl_done((union ctl_io *)ctsio);
6940                 return (CTL_RETVAL_COMPLETE);
6941         }
6942
6943         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6944
6945         ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
6946         data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
6947         ctsio->residual = 0;
6948         ctsio->kern_data_len = sizeof(*data);
6949         ctsio->kern_total_len = sizeof(*data);
6950         ctsio->kern_data_resid = 0;
6951         ctsio->kern_rel_offset = 0;
6952         ctsio->kern_sg_entries = 0;
6953
6954         /*
6955          * If the maximum LBA is greater than 0xfffffffe, the user must
6956          * issue a SERVICE ACTION IN (16) command, with the read capacity
6957          * serivce action set.
6958          */
6959         if (lun->be_lun->maxlba > 0xfffffffe)
6960                 scsi_ulto4b(0xffffffff, data->addr);
6961         else
6962                 scsi_ulto4b(lun->be_lun->maxlba, data->addr);
6963
6964         /*
6965          * XXX KDM this may not be 512 bytes...
6966          */
6967         scsi_ulto4b(lun->be_lun->blocksize, data->length);
6968
6969         ctl_set_success(ctsio);
6970         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6971         ctsio->be_move_done = ctl_config_move_done;
6972         ctl_datamove((union ctl_io *)ctsio);
6973         return (CTL_RETVAL_COMPLETE);
6974 }
6975
6976 int
6977 ctl_read_capacity_16(struct ctl_scsiio *ctsio)
6978 {
6979         struct scsi_read_capacity_16 *cdb;
6980         struct scsi_read_capacity_data_long *data;
6981         struct ctl_lun *lun;
6982         uint64_t lba;
6983         uint32_t alloc_len;
6984
6985         CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
6986
6987         cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
6988
6989         alloc_len = scsi_4btoul(cdb->alloc_len);
6990         lba = scsi_8btou64(cdb->addr);
6991
6992         if ((cdb->reladr & SRC16_PMI)
6993          && (lba != 0)) {
6994                 ctl_set_invalid_field(/*ctsio*/ ctsio,
6995                                       /*sks_valid*/ 1,
6996                                       /*command*/ 1,
6997                                       /*field*/ 2,
6998                                       /*bit_valid*/ 0,
6999                                       /*bit*/ 0);
7000                 ctl_done((union ctl_io *)ctsio);
7001                 return (CTL_RETVAL_COMPLETE);
7002         }
7003
7004         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7005
7006         ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
7007         data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
7008
7009         if (sizeof(*data) < alloc_len) {
7010                 ctsio->residual = alloc_len - sizeof(*data);
7011                 ctsio->kern_data_len = sizeof(*data);
7012                 ctsio->kern_total_len = sizeof(*data);
7013         } else {
7014                 ctsio->residual = 0;
7015                 ctsio->kern_data_len = alloc_len;
7016                 ctsio->kern_total_len = alloc_len;
7017         }
7018         ctsio->kern_data_resid = 0;
7019         ctsio->kern_rel_offset = 0;
7020         ctsio->kern_sg_entries = 0;
7021
7022         scsi_u64to8b(lun->be_lun->maxlba, data->addr);
7023         /* XXX KDM this may not be 512 bytes... */
7024         scsi_ulto4b(lun->be_lun->blocksize, data->length);
7025         data->prot_lbppbe = lun->be_lun->pblockexp & SRC16_LBPPBE;
7026         scsi_ulto2b(lun->be_lun->pblockoff & SRC16_LALBA_A, data->lalba_lbp);
7027         if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
7028                 data->lalba_lbp[0] |= SRC16_LBPME | SRC16_LBPRZ;
7029
7030         ctl_set_success(ctsio);
7031         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7032         ctsio->be_move_done = ctl_config_move_done;
7033         ctl_datamove((union ctl_io *)ctsio);
7034         return (CTL_RETVAL_COMPLETE);
7035 }
7036
7037 int
7038 ctl_get_lba_status(struct ctl_scsiio *ctsio)
7039 {
7040         struct scsi_get_lba_status *cdb;
7041         struct scsi_get_lba_status_data *data;
7042         struct ctl_lun *lun;
7043         struct ctl_lba_len_flags *lbalen;
7044         uint64_t lba;
7045         uint32_t alloc_len, total_len;
7046         int retval;
7047
7048         CTL_DEBUG_PRINT(("ctl_get_lba_status\n"));
7049
7050         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7051         cdb = (struct scsi_get_lba_status *)ctsio->cdb;
7052         lba = scsi_8btou64(cdb->addr);
7053         alloc_len = scsi_4btoul(cdb->alloc_len);
7054
7055         if (lba > lun->be_lun->maxlba) {
7056                 ctl_set_lba_out_of_range(ctsio);
7057                 ctl_done((union ctl_io *)ctsio);
7058                 return (CTL_RETVAL_COMPLETE);
7059         }
7060
7061         total_len = sizeof(*data) + sizeof(data->descr[0]);
7062         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7063         data = (struct scsi_get_lba_status_data *)ctsio->kern_data_ptr;
7064
7065         if (total_len < alloc_len) {
7066                 ctsio->residual = alloc_len - total_len;
7067                 ctsio->kern_data_len = total_len;
7068                 ctsio->kern_total_len = total_len;
7069         } else {
7070                 ctsio->residual = 0;
7071                 ctsio->kern_data_len = alloc_len;
7072                 ctsio->kern_total_len = alloc_len;
7073         }
7074         ctsio->kern_data_resid = 0;
7075         ctsio->kern_rel_offset = 0;
7076         ctsio->kern_sg_entries = 0;
7077
7078         /* Fill dummy data in case backend can't tell anything. */
7079         scsi_ulto4b(4 + sizeof(data->descr[0]), data->length);
7080         scsi_u64to8b(lba, data->descr[0].addr);
7081         scsi_ulto4b(MIN(UINT32_MAX, lun->be_lun->maxlba + 1 - lba),
7082             data->descr[0].length);
7083         data->descr[0].status = 0; /* Mapped or unknown. */
7084
7085         ctl_set_success(ctsio);
7086         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7087         ctsio->be_move_done = ctl_config_move_done;
7088
7089         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
7090         lbalen->lba = lba;
7091         lbalen->len = total_len;
7092         lbalen->flags = 0;
7093         retval = lun->backend->config_read((union ctl_io *)ctsio);
7094         return (CTL_RETVAL_COMPLETE);
7095 }
7096
7097 int
7098 ctl_read_defect(struct ctl_scsiio *ctsio)
7099 {
7100         struct scsi_read_defect_data_10 *ccb10;
7101         struct scsi_read_defect_data_12 *ccb12;
7102         struct scsi_read_defect_data_hdr_10 *data10;
7103         struct scsi_read_defect_data_hdr_12 *data12;
7104         uint32_t alloc_len, data_len;
7105         uint8_t format;
7106
7107         CTL_DEBUG_PRINT(("ctl_read_defect\n"));
7108
7109         if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
7110                 ccb10 = (struct scsi_read_defect_data_10 *)&ctsio->cdb;
7111                 format = ccb10->format;
7112                 alloc_len = scsi_2btoul(ccb10->alloc_length);
7113                 data_len = sizeof(*data10);
7114         } else {
7115                 ccb12 = (struct scsi_read_defect_data_12 *)&ctsio->cdb;
7116                 format = ccb12->format;
7117                 alloc_len = scsi_4btoul(ccb12->alloc_length);
7118                 data_len = sizeof(*data12);
7119         }
7120         if (alloc_len == 0) {
7121                 ctl_set_success(ctsio);
7122                 ctl_done((union ctl_io *)ctsio);
7123                 return (CTL_RETVAL_COMPLETE);
7124         }
7125
7126         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
7127         if (data_len < alloc_len) {
7128                 ctsio->residual = alloc_len - data_len;
7129                 ctsio->kern_data_len = data_len;
7130                 ctsio->kern_total_len = data_len;
7131         } else {
7132                 ctsio->residual = 0;
7133                 ctsio->kern_data_len = alloc_len;
7134                 ctsio->kern_total_len = alloc_len;
7135         }
7136         ctsio->kern_data_resid = 0;
7137         ctsio->kern_rel_offset = 0;
7138         ctsio->kern_sg_entries = 0;
7139
7140         if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
7141                 data10 = (struct scsi_read_defect_data_hdr_10 *)
7142                     ctsio->kern_data_ptr;
7143                 data10->format = format;
7144                 scsi_ulto2b(0, data10->length);
7145         } else {
7146                 data12 = (struct scsi_read_defect_data_hdr_12 *)
7147                     ctsio->kern_data_ptr;
7148                 data12->format = format;
7149                 scsi_ulto2b(0, data12->generation);
7150                 scsi_ulto4b(0, data12->length);
7151         }
7152
7153         ctl_set_success(ctsio);
7154         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7155         ctsio->be_move_done = ctl_config_move_done;
7156         ctl_datamove((union ctl_io *)ctsio);
7157         return (CTL_RETVAL_COMPLETE);
7158 }
7159
7160 int
7161 ctl_report_tagret_port_groups(struct ctl_scsiio *ctsio)
7162 {
7163         struct scsi_maintenance_in *cdb;
7164         int retval;
7165         int alloc_len, ext, total_len = 0, g, pc, pg, ts, os;
7166         int num_ha_groups, num_target_ports, shared_group;
7167         struct ctl_lun *lun;
7168         struct ctl_softc *softc;
7169         struct ctl_port *port;
7170         struct scsi_target_group_data *rtg_ptr;
7171         struct scsi_target_group_data_extended *rtg_ext_ptr;
7172         struct scsi_target_port_group_descriptor *tpg_desc;
7173
7174         CTL_DEBUG_PRINT(("ctl_report_tagret_port_groups\n"));
7175
7176         cdb = (struct scsi_maintenance_in *)ctsio->cdb;
7177         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7178         softc = lun->ctl_softc;
7179
7180         retval = CTL_RETVAL_COMPLETE;
7181
7182         switch (cdb->byte2 & STG_PDF_MASK) {
7183         case STG_PDF_LENGTH:
7184                 ext = 0;
7185                 break;
7186         case STG_PDF_EXTENDED:
7187                 ext = 1;
7188                 break;
7189         default:
7190                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7191                                       /*sks_valid*/ 1,
7192                                       /*command*/ 1,
7193                                       /*field*/ 2,
7194                                       /*bit_valid*/ 1,
7195                                       /*bit*/ 5);
7196                 ctl_done((union ctl_io *)ctsio);
7197                 return(retval);
7198         }
7199
7200         num_target_ports = 0;
7201         shared_group = (softc->is_single != 0);
7202         mtx_lock(&softc->ctl_lock);
7203         STAILQ_FOREACH(port, &softc->port_list, links) {
7204                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7205                         continue;
7206                 if (ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
7207                         continue;
7208                 num_target_ports++;
7209                 if (port->status & CTL_PORT_STATUS_HA_SHARED)
7210                         shared_group = 1;
7211         }
7212         mtx_unlock(&softc->ctl_lock);
7213         num_ha_groups = (softc->is_single) ? 0 : NUM_HA_SHELVES;
7214
7215         if (ext)
7216                 total_len = sizeof(struct scsi_target_group_data_extended);
7217         else
7218                 total_len = sizeof(struct scsi_target_group_data);
7219         total_len += sizeof(struct scsi_target_port_group_descriptor) *
7220                 (shared_group + num_ha_groups) +
7221             sizeof(struct scsi_target_port_descriptor) * num_target_ports;
7222
7223         alloc_len = scsi_4btoul(cdb->length);
7224
7225         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7226
7227         ctsio->kern_sg_entries = 0;
7228
7229         if (total_len < alloc_len) {
7230                 ctsio->residual = alloc_len - total_len;
7231                 ctsio->kern_data_len = total_len;
7232                 ctsio->kern_total_len = total_len;
7233         } else {
7234                 ctsio->residual = 0;
7235                 ctsio->kern_data_len = alloc_len;
7236                 ctsio->kern_total_len = alloc_len;
7237         }
7238         ctsio->kern_data_resid = 0;
7239         ctsio->kern_rel_offset = 0;
7240
7241         if (ext) {
7242                 rtg_ext_ptr = (struct scsi_target_group_data_extended *)
7243                     ctsio->kern_data_ptr;
7244                 scsi_ulto4b(total_len - 4, rtg_ext_ptr->length);
7245                 rtg_ext_ptr->format_type = 0x10;
7246                 rtg_ext_ptr->implicit_transition_time = 0;
7247                 tpg_desc = &rtg_ext_ptr->groups[0];
7248         } else {
7249                 rtg_ptr = (struct scsi_target_group_data *)
7250                     ctsio->kern_data_ptr;
7251                 scsi_ulto4b(total_len - 4, rtg_ptr->length);
7252                 tpg_desc = &rtg_ptr->groups[0];
7253         }
7254
7255         mtx_lock(&softc->ctl_lock);
7256         pg = softc->port_min / softc->port_cnt;
7257         if (lun->flags & (CTL_LUN_PRIMARY_SC | CTL_LUN_PEER_SC_PRIMARY)) {
7258                 /* Some shelf is known to be primary. */
7259                 if (softc->ha_link == CTL_HA_LINK_OFFLINE)
7260                         os = TPG_ASYMMETRIC_ACCESS_UNAVAILABLE;
7261                 else if (softc->ha_link == CTL_HA_LINK_UNKNOWN)
7262                         os = TPG_ASYMMETRIC_ACCESS_TRANSITIONING;
7263                 else if (softc->ha_mode == CTL_HA_MODE_ACT_STBY)
7264                         os = TPG_ASYMMETRIC_ACCESS_STANDBY;
7265                 else
7266                         os = TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7267                 if (lun->flags & CTL_LUN_PRIMARY_SC) {
7268                         ts = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7269                 } else {
7270                         ts = os;
7271                         os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7272                 }
7273         } else {
7274                 /* No known primary shelf. */
7275                 if (softc->ha_link == CTL_HA_LINK_OFFLINE) {
7276                         ts = TPG_ASYMMETRIC_ACCESS_UNAVAILABLE;
7277                         os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7278                 } else if (softc->ha_link == CTL_HA_LINK_UNKNOWN) {
7279                         ts = TPG_ASYMMETRIC_ACCESS_TRANSITIONING;
7280                         os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7281                 } else {
7282                         ts = os = TPG_ASYMMETRIC_ACCESS_TRANSITIONING;
7283                 }
7284         }
7285         if (shared_group) {
7286                 tpg_desc->pref_state = ts;
7287                 tpg_desc->support = TPG_AO_SUP | TPG_AN_SUP | TPG_S_SUP |
7288                     TPG_U_SUP | TPG_T_SUP;
7289                 scsi_ulto2b(1, tpg_desc->target_port_group);
7290                 tpg_desc->status = TPG_IMPLICIT;
7291                 pc = 0;
7292                 STAILQ_FOREACH(port, &softc->port_list, links) {
7293                         if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7294                                 continue;
7295                         if (!softc->is_single &&
7296                             (port->status & CTL_PORT_STATUS_HA_SHARED) == 0)
7297                                 continue;
7298                         if (ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
7299                                 continue;
7300                         scsi_ulto2b(port->targ_port, tpg_desc->descriptors[pc].
7301                             relative_target_port_identifier);
7302                         pc++;
7303                 }
7304                 tpg_desc->target_port_count = pc;
7305                 tpg_desc = (struct scsi_target_port_group_descriptor *)
7306                     &tpg_desc->descriptors[pc];
7307         }
7308         for (g = 0; g < num_ha_groups; g++) {
7309                 tpg_desc->pref_state = (g == pg) ? ts : os;
7310                 tpg_desc->support = TPG_AO_SUP | TPG_AN_SUP | TPG_S_SUP |
7311                     TPG_U_SUP | TPG_T_SUP;
7312                 scsi_ulto2b(2 + g, tpg_desc->target_port_group);
7313                 tpg_desc->status = TPG_IMPLICIT;
7314                 pc = 0;
7315                 STAILQ_FOREACH(port, &softc->port_list, links) {
7316                         if (port->targ_port < g * softc->port_cnt ||
7317                             port->targ_port >= (g + 1) * softc->port_cnt)
7318                                 continue;
7319                         if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7320                                 continue;
7321                         if (port->status & CTL_PORT_STATUS_HA_SHARED)
7322                                 continue;
7323                         if (ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
7324                                 continue;
7325                         scsi_ulto2b(port->targ_port, tpg_desc->descriptors[pc].
7326                             relative_target_port_identifier);
7327                         pc++;
7328                 }
7329                 tpg_desc->target_port_count = pc;
7330                 tpg_desc = (struct scsi_target_port_group_descriptor *)
7331                     &tpg_desc->descriptors[pc];
7332         }
7333         mtx_unlock(&softc->ctl_lock);
7334
7335         ctl_set_success(ctsio);
7336         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7337         ctsio->be_move_done = ctl_config_move_done;
7338         ctl_datamove((union ctl_io *)ctsio);
7339         return(retval);
7340 }
7341
7342 int
7343 ctl_report_supported_opcodes(struct ctl_scsiio *ctsio)
7344 {
7345         struct ctl_lun *lun;
7346         struct scsi_report_supported_opcodes *cdb;
7347         const struct ctl_cmd_entry *entry, *sentry;
7348         struct scsi_report_supported_opcodes_all *all;
7349         struct scsi_report_supported_opcodes_descr *descr;
7350         struct scsi_report_supported_opcodes_one *one;
7351         int retval;
7352         int alloc_len, total_len;
7353         int opcode, service_action, i, j, num;
7354
7355         CTL_DEBUG_PRINT(("ctl_report_supported_opcodes\n"));
7356
7357         cdb = (struct scsi_report_supported_opcodes *)ctsio->cdb;
7358         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7359
7360         retval = CTL_RETVAL_COMPLETE;
7361
7362         opcode = cdb->requested_opcode;
7363         service_action = scsi_2btoul(cdb->requested_service_action);
7364         switch (cdb->options & RSO_OPTIONS_MASK) {
7365         case RSO_OPTIONS_ALL:
7366                 num = 0;
7367                 for (i = 0; i < 256; i++) {
7368                         entry = &ctl_cmd_table[i];
7369                         if (entry->flags & CTL_CMD_FLAG_SA5) {
7370                                 for (j = 0; j < 32; j++) {
7371                                         sentry = &((const struct ctl_cmd_entry *)
7372                                             entry->execute)[j];
7373                                         if (ctl_cmd_applicable(
7374                                             lun->be_lun->lun_type, sentry))
7375                                                 num++;
7376                                 }
7377                         } else {
7378                                 if (ctl_cmd_applicable(lun->be_lun->lun_type,
7379                                     entry))
7380                                         num++;
7381                         }
7382                 }
7383                 total_len = sizeof(struct scsi_report_supported_opcodes_all) +
7384                     num * sizeof(struct scsi_report_supported_opcodes_descr);
7385                 break;
7386         case RSO_OPTIONS_OC:
7387                 if (ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) {
7388                         ctl_set_invalid_field(/*ctsio*/ ctsio,
7389                                               /*sks_valid*/ 1,
7390                                               /*command*/ 1,
7391                                               /*field*/ 2,
7392                                               /*bit_valid*/ 1,
7393                                               /*bit*/ 2);
7394                         ctl_done((union ctl_io *)ctsio);
7395                         return (CTL_RETVAL_COMPLETE);
7396                 }
7397                 total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7398                 break;
7399         case RSO_OPTIONS_OC_SA:
7400                 if ((ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) == 0 ||
7401                     service_action >= 32) {
7402                         ctl_set_invalid_field(/*ctsio*/ ctsio,
7403                                               /*sks_valid*/ 1,
7404                                               /*command*/ 1,
7405                                               /*field*/ 2,
7406                                               /*bit_valid*/ 1,
7407                                               /*bit*/ 2);
7408                         ctl_done((union ctl_io *)ctsio);
7409                         return (CTL_RETVAL_COMPLETE);
7410                 }
7411                 total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7412                 break;
7413         default:
7414                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7415                                       /*sks_valid*/ 1,
7416                                       /*command*/ 1,
7417                                       /*field*/ 2,
7418                                       /*bit_valid*/ 1,
7419                                       /*bit*/ 2);
7420                 ctl_done((union ctl_io *)ctsio);
7421                 return (CTL_RETVAL_COMPLETE);
7422         }
7423
7424         alloc_len = scsi_4btoul(cdb->length);
7425
7426         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7427
7428         ctsio->kern_sg_entries = 0;
7429
7430         if (total_len < alloc_len) {
7431                 ctsio->residual = alloc_len - total_len;
7432                 ctsio->kern_data_len = total_len;
7433                 ctsio->kern_total_len = total_len;
7434         } else {
7435                 ctsio->residual = 0;
7436                 ctsio->kern_data_len = alloc_len;
7437                 ctsio->kern_total_len = alloc_len;
7438         }
7439         ctsio->kern_data_resid = 0;
7440         ctsio->kern_rel_offset = 0;
7441
7442         switch (cdb->options & RSO_OPTIONS_MASK) {
7443         case RSO_OPTIONS_ALL:
7444                 all = (struct scsi_report_supported_opcodes_all *)
7445                     ctsio->kern_data_ptr;
7446                 num = 0;
7447                 for (i = 0; i < 256; i++) {
7448                         entry = &ctl_cmd_table[i];
7449                         if (entry->flags & CTL_CMD_FLAG_SA5) {
7450                                 for (j = 0; j < 32; j++) {
7451                                         sentry = &((const struct ctl_cmd_entry *)
7452                                             entry->execute)[j];
7453                                         if (!ctl_cmd_applicable(
7454                                             lun->be_lun->lun_type, sentry))
7455                                                 continue;
7456                                         descr = &all->descr[num++];
7457                                         descr->opcode = i;
7458                                         scsi_ulto2b(j, descr->service_action);
7459                                         descr->flags = RSO_SERVACTV;
7460                                         scsi_ulto2b(sentry->length,
7461                                             descr->cdb_length);
7462                                 }
7463                         } else {
7464                                 if (!ctl_cmd_applicable(lun->be_lun->lun_type,
7465                                     entry))
7466                                         continue;
7467                                 descr = &all->descr[num++];
7468                                 descr->opcode = i;
7469                                 scsi_ulto2b(0, descr->service_action);
7470                                 descr->flags = 0;
7471                                 scsi_ulto2b(entry->length, descr->cdb_length);
7472                         }
7473                 }
7474                 scsi_ulto4b(
7475                     num * sizeof(struct scsi_report_supported_opcodes_descr),
7476                     all->length);
7477                 break;
7478         case RSO_OPTIONS_OC:
7479                 one = (struct scsi_report_supported_opcodes_one *)
7480                     ctsio->kern_data_ptr;
7481                 entry = &ctl_cmd_table[opcode];
7482                 goto fill_one;
7483         case RSO_OPTIONS_OC_SA:
7484                 one = (struct scsi_report_supported_opcodes_one *)
7485                     ctsio->kern_data_ptr;
7486                 entry = &ctl_cmd_table[opcode];
7487                 entry = &((const struct ctl_cmd_entry *)
7488                     entry->execute)[service_action];
7489 fill_one:
7490                 if (ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
7491                         one->support = 3;
7492                         scsi_ulto2b(entry->length, one->cdb_length);
7493                         one->cdb_usage[0] = opcode;
7494                         memcpy(&one->cdb_usage[1], entry->usage,
7495                             entry->length - 1);
7496                 } else
7497                         one->support = 1;
7498                 break;
7499         }
7500
7501         ctl_set_success(ctsio);
7502         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7503         ctsio->be_move_done = ctl_config_move_done;
7504         ctl_datamove((union ctl_io *)ctsio);
7505         return(retval);
7506 }
7507
7508 int
7509 ctl_report_supported_tmf(struct ctl_scsiio *ctsio)
7510 {
7511         struct scsi_report_supported_tmf *cdb;
7512         struct scsi_report_supported_tmf_data *data;
7513         int retval;
7514         int alloc_len, total_len;
7515
7516         CTL_DEBUG_PRINT(("ctl_report_supported_tmf\n"));
7517
7518         cdb = (struct scsi_report_supported_tmf *)ctsio->cdb;
7519
7520         retval = CTL_RETVAL_COMPLETE;
7521
7522         total_len = sizeof(struct scsi_report_supported_tmf_data);
7523         alloc_len = scsi_4btoul(cdb->length);
7524
7525         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7526
7527         ctsio->kern_sg_entries = 0;
7528
7529         if (total_len < alloc_len) {
7530                 ctsio->residual = alloc_len - total_len;
7531                 ctsio->kern_data_len = total_len;
7532                 ctsio->kern_total_len = total_len;
7533         } else {
7534                 ctsio->residual = 0;
7535                 ctsio->kern_data_len = alloc_len;
7536                 ctsio->kern_total_len = alloc_len;
7537         }
7538         ctsio->kern_data_resid = 0;
7539         ctsio->kern_rel_offset = 0;
7540
7541         data = (struct scsi_report_supported_tmf_data *)ctsio->kern_data_ptr;
7542         data->byte1 |= RST_ATS | RST_ATSS | RST_CTSS | RST_LURS | RST_QTS |
7543             RST_TRS;
7544         data->byte2 |= RST_QAES | RST_QTSS | RST_ITNRS;
7545
7546         ctl_set_success(ctsio);
7547         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7548         ctsio->be_move_done = ctl_config_move_done;
7549         ctl_datamove((union ctl_io *)ctsio);
7550         return (retval);
7551 }
7552
7553 int
7554 ctl_report_timestamp(struct ctl_scsiio *ctsio)
7555 {
7556         struct scsi_report_timestamp *cdb;
7557         struct scsi_report_timestamp_data *data;
7558         struct timeval tv;
7559         int64_t timestamp;
7560         int retval;
7561         int alloc_len, total_len;
7562
7563         CTL_DEBUG_PRINT(("ctl_report_timestamp\n"));
7564
7565         cdb = (struct scsi_report_timestamp *)ctsio->cdb;
7566
7567         retval = CTL_RETVAL_COMPLETE;
7568
7569         total_len = sizeof(struct scsi_report_timestamp_data);
7570         alloc_len = scsi_4btoul(cdb->length);
7571
7572         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7573
7574         ctsio->kern_sg_entries = 0;
7575
7576         if (total_len < alloc_len) {
7577                 ctsio->residual = alloc_len - total_len;
7578                 ctsio->kern_data_len = total_len;
7579                 ctsio->kern_total_len = total_len;
7580         } else {
7581                 ctsio->residual = 0;
7582                 ctsio->kern_data_len = alloc_len;
7583                 ctsio->kern_total_len = alloc_len;
7584         }
7585         ctsio->kern_data_resid = 0;
7586         ctsio->kern_rel_offset = 0;
7587
7588         data = (struct scsi_report_timestamp_data *)ctsio->kern_data_ptr;
7589         scsi_ulto2b(sizeof(*data) - 2, data->length);
7590         data->origin = RTS_ORIG_OUTSIDE;
7591         getmicrotime(&tv);
7592         timestamp = (int64_t)tv.tv_sec * 1000 + tv.tv_usec / 1000;
7593         scsi_ulto4b(timestamp >> 16, data->timestamp);
7594         scsi_ulto2b(timestamp & 0xffff, &data->timestamp[4]);
7595
7596         ctl_set_success(ctsio);
7597         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7598         ctsio->be_move_done = ctl_config_move_done;
7599         ctl_datamove((union ctl_io *)ctsio);
7600         return (retval);
7601 }
7602
7603 int
7604 ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7605 {
7606         struct scsi_per_res_in *cdb;
7607         int alloc_len, total_len = 0;
7608         /* struct scsi_per_res_in_rsrv in_data; */
7609         struct ctl_lun *lun;
7610         struct ctl_softc *softc;
7611         uint64_t key;
7612
7613         CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7614
7615         cdb = (struct scsi_per_res_in *)ctsio->cdb;
7616
7617         alloc_len = scsi_2btoul(cdb->length);
7618
7619         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7620         softc = lun->ctl_softc;
7621
7622 retry:
7623         mtx_lock(&lun->lun_lock);
7624         switch (cdb->action) {
7625         case SPRI_RK: /* read keys */
7626                 total_len = sizeof(struct scsi_per_res_in_keys) +
7627                         lun->pr_key_count *
7628                         sizeof(struct scsi_per_res_key);
7629                 break;
7630         case SPRI_RR: /* read reservation */
7631                 if (lun->flags & CTL_LUN_PR_RESERVED)
7632                         total_len = sizeof(struct scsi_per_res_in_rsrv);
7633                 else
7634                         total_len = sizeof(struct scsi_per_res_in_header);
7635                 break;
7636         case SPRI_RC: /* report capabilities */
7637                 total_len = sizeof(struct scsi_per_res_cap);
7638                 break;
7639         case SPRI_RS: /* read full status */
7640                 total_len = sizeof(struct scsi_per_res_in_header) +
7641                     (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7642                     lun->pr_key_count;
7643                 break;
7644         default:
7645                 panic("%s: Invalid PR type %#x", __func__, cdb->action);
7646         }
7647         mtx_unlock(&lun->lun_lock);
7648
7649         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7650
7651         if (total_len < alloc_len) {
7652                 ctsio->residual = alloc_len - total_len;
7653                 ctsio->kern_data_len = total_len;
7654                 ctsio->kern_total_len = total_len;
7655         } else {
7656                 ctsio->residual = 0;
7657                 ctsio->kern_data_len = alloc_len;
7658                 ctsio->kern_total_len = alloc_len;
7659         }
7660
7661         ctsio->kern_data_resid = 0;
7662         ctsio->kern_rel_offset = 0;
7663         ctsio->kern_sg_entries = 0;
7664
7665         mtx_lock(&lun->lun_lock);
7666         switch (cdb->action) {
7667         case SPRI_RK: { // read keys
7668         struct scsi_per_res_in_keys *res_keys;
7669                 int i, key_count;
7670
7671                 res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7672
7673                 /*
7674                  * We had to drop the lock to allocate our buffer, which
7675                  * leaves time for someone to come in with another
7676                  * persistent reservation.  (That is unlikely, though,
7677                  * since this should be the only persistent reservation
7678                  * command active right now.)
7679                  */
7680                 if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7681                     (lun->pr_key_count *
7682                      sizeof(struct scsi_per_res_key)))){
7683                         mtx_unlock(&lun->lun_lock);
7684                         free(ctsio->kern_data_ptr, M_CTL);
7685                         printf("%s: reservation length changed, retrying\n",
7686                                __func__);
7687                         goto retry;
7688                 }
7689
7690                 scsi_ulto4b(lun->pr_generation, res_keys->header.generation);
7691
7692                 scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7693                              lun->pr_key_count, res_keys->header.length);
7694
7695                 for (i = 0, key_count = 0; i < CTL_MAX_INITIATORS; i++) {
7696                         if ((key = ctl_get_prkey(lun, i)) == 0)
7697                                 continue;
7698
7699                         /*
7700                          * We used lun->pr_key_count to calculate the
7701                          * size to allocate.  If it turns out the number of
7702                          * initiators with the registered flag set is
7703                          * larger than that (i.e. they haven't been kept in
7704                          * sync), we've got a problem.
7705                          */
7706                         if (key_count >= lun->pr_key_count) {
7707                                 key_count++;
7708                                 continue;
7709                         }
7710                         scsi_u64to8b(key, res_keys->keys[key_count].key);
7711                         key_count++;
7712                 }
7713                 break;
7714         }
7715         case SPRI_RR: { // read reservation
7716                 struct scsi_per_res_in_rsrv *res;
7717                 int tmp_len, header_only;
7718
7719                 res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7720
7721                 scsi_ulto4b(lun->pr_generation, res->header.generation);
7722
7723                 if (lun->flags & CTL_LUN_PR_RESERVED)
7724                 {
7725                         tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7726                         scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7727                                     res->header.length);
7728                         header_only = 0;
7729                 } else {
7730                         tmp_len = sizeof(struct scsi_per_res_in_header);
7731                         scsi_ulto4b(0, res->header.length);
7732                         header_only = 1;
7733                 }
7734
7735                 /*
7736                  * We had to drop the lock to allocate our buffer, which
7737                  * leaves time for someone to come in with another
7738                  * persistent reservation.  (That is unlikely, though,
7739                  * since this should be the only persistent reservation
7740                  * command active right now.)
7741                  */
7742                 if (tmp_len != total_len) {
7743                         mtx_unlock(&lun->lun_lock);
7744                         free(ctsio->kern_data_ptr, M_CTL);
7745                         printf("%s: reservation status changed, retrying\n",
7746                                __func__);
7747                         goto retry;
7748                 }
7749
7750                 /*
7751                  * No reservation held, so we're done.
7752                  */
7753                 if (header_only != 0)
7754                         break;
7755
7756                 /*
7757                  * If the registration is an All Registrants type, the key
7758                  * is 0, since it doesn't really matter.
7759                  */
7760                 if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7761                         scsi_u64to8b(ctl_get_prkey(lun, lun->pr_res_idx),
7762                             res->data.reservation);
7763                 }
7764                 res->data.scopetype = lun->pr_res_type;
7765                 break;
7766         }
7767         case SPRI_RC:     //report capabilities
7768         {
7769                 struct scsi_per_res_cap *res_cap;
7770                 uint16_t type_mask;
7771
7772                 res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7773                 scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7774                 res_cap->flags1 = SPRI_CRH;
7775                 res_cap->flags2 = SPRI_TMV | SPRI_ALLOW_5;
7776                 type_mask = SPRI_TM_WR_EX_AR |
7777                             SPRI_TM_EX_AC_RO |
7778                             SPRI_TM_WR_EX_RO |
7779                             SPRI_TM_EX_AC |
7780                             SPRI_TM_WR_EX |
7781                             SPRI_TM_EX_AC_AR;
7782                 scsi_ulto2b(type_mask, res_cap->type_mask);
7783                 break;
7784         }
7785         case SPRI_RS: { // read full status
7786                 struct scsi_per_res_in_full *res_status;
7787                 struct scsi_per_res_in_full_desc *res_desc;
7788                 struct ctl_port *port;
7789                 int i, len;
7790
7791                 res_status = (struct scsi_per_res_in_full*)ctsio->kern_data_ptr;
7792
7793                 /*
7794                  * We had to drop the lock to allocate our buffer, which
7795                  * leaves time for someone to come in with another
7796                  * persistent reservation.  (That is unlikely, though,
7797                  * since this should be the only persistent reservation
7798                  * command active right now.)
7799                  */
7800                 if (total_len < (sizeof(struct scsi_per_res_in_header) +
7801                     (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7802                      lun->pr_key_count)){
7803                         mtx_unlock(&lun->lun_lock);
7804                         free(ctsio->kern_data_ptr, M_CTL);
7805                         printf("%s: reservation length changed, retrying\n",
7806                                __func__);
7807                         goto retry;
7808                 }
7809
7810                 scsi_ulto4b(lun->pr_generation, res_status->header.generation);
7811
7812                 res_desc = &res_status->desc[0];
7813                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
7814                         if ((key = ctl_get_prkey(lun, i)) == 0)
7815                                 continue;
7816
7817                         scsi_u64to8b(key, res_desc->res_key.key);
7818                         if ((lun->flags & CTL_LUN_PR_RESERVED) &&
7819                             (lun->pr_res_idx == i ||
7820                              lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS)) {
7821                                 res_desc->flags = SPRI_FULL_R_HOLDER;
7822                                 res_desc->scopetype = lun->pr_res_type;
7823                         }
7824                         scsi_ulto2b(i / CTL_MAX_INIT_PER_PORT,
7825                             res_desc->rel_trgt_port_id);
7826                         len = 0;
7827                         port = softc->ctl_ports[i / CTL_MAX_INIT_PER_PORT];
7828                         if (port != NULL)
7829                                 len = ctl_create_iid(port,
7830                                     i % CTL_MAX_INIT_PER_PORT,
7831                                     res_desc->transport_id);
7832                         scsi_ulto4b(len, res_desc->additional_length);
7833                         res_desc = (struct scsi_per_res_in_full_desc *)
7834                             &res_desc->transport_id[len];
7835                 }
7836                 scsi_ulto4b((uint8_t *)res_desc - (uint8_t *)&res_status->desc[0],
7837                     res_status->header.length);
7838                 break;
7839         }
7840         default:
7841                 panic("%s: Invalid PR type %#x", __func__, cdb->action);
7842         }
7843         mtx_unlock(&lun->lun_lock);
7844
7845         ctl_set_success(ctsio);
7846         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7847         ctsio->be_move_done = ctl_config_move_done;
7848         ctl_datamove((union ctl_io *)ctsio);
7849         return (CTL_RETVAL_COMPLETE);
7850 }
7851
7852 /*
7853  * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
7854  * it should return.
7855  */
7856 static int
7857 ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
7858                 uint64_t sa_res_key, uint8_t type, uint32_t residx,
7859                 struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
7860                 struct scsi_per_res_out_parms* param)
7861 {
7862         union ctl_ha_msg persis_io;
7863         int i;
7864
7865         mtx_lock(&lun->lun_lock);
7866         if (sa_res_key == 0) {
7867                 if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
7868                         /* validate scope and type */
7869                         if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7870                              SPR_LU_SCOPE) {
7871                                 mtx_unlock(&lun->lun_lock);
7872                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7873                                                       /*sks_valid*/ 1,
7874                                                       /*command*/ 1,
7875                                                       /*field*/ 2,
7876                                                       /*bit_valid*/ 1,
7877                                                       /*bit*/ 4);
7878                                 ctl_done((union ctl_io *)ctsio);
7879                                 return (1);
7880                         }
7881
7882                         if (type>8 || type==2 || type==4 || type==0) {
7883                                 mtx_unlock(&lun->lun_lock);
7884                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7885                                                       /*sks_valid*/ 1,
7886                                                       /*command*/ 1,
7887                                                       /*field*/ 2,
7888                                                       /*bit_valid*/ 1,
7889                                                       /*bit*/ 0);
7890                                 ctl_done((union ctl_io *)ctsio);
7891                                 return (1);
7892                         }
7893
7894                         /*
7895                          * Unregister everybody else and build UA for
7896                          * them
7897                          */
7898                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
7899                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
7900                                         continue;
7901
7902                                 ctl_clr_prkey(lun, i);
7903                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7904                         }
7905                         lun->pr_key_count = 1;
7906                         lun->pr_res_type = type;
7907                         if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
7908                             lun->pr_res_type != SPR_TYPE_EX_AC_AR)
7909                                 lun->pr_res_idx = residx;
7910                         lun->pr_generation++;
7911                         mtx_unlock(&lun->lun_lock);
7912
7913                         /* send msg to other side */
7914                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7915                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7916                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7917                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
7918                         persis_io.pr.pr_info.res_type = type;
7919                         memcpy(persis_io.pr.pr_info.sa_res_key,
7920                                param->serv_act_res_key,
7921                                sizeof(param->serv_act_res_key));
7922                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
7923                             sizeof(persis_io.pr), M_WAITOK);
7924                 } else {
7925                         /* not all registrants */
7926                         mtx_unlock(&lun->lun_lock);
7927                         free(ctsio->kern_data_ptr, M_CTL);
7928                         ctl_set_invalid_field(ctsio,
7929                                               /*sks_valid*/ 1,
7930                                               /*command*/ 0,
7931                                               /*field*/ 8,
7932                                               /*bit_valid*/ 0,
7933                                               /*bit*/ 0);
7934                         ctl_done((union ctl_io *)ctsio);
7935                         return (1);
7936                 }
7937         } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
7938                 || !(lun->flags & CTL_LUN_PR_RESERVED)) {
7939                 int found = 0;
7940
7941                 if (res_key == sa_res_key) {
7942                         /* special case */
7943                         /*
7944                          * The spec implies this is not good but doesn't
7945                          * say what to do. There are two choices either
7946                          * generate a res conflict or check condition
7947                          * with illegal field in parameter data. Since
7948                          * that is what is done when the sa_res_key is
7949                          * zero I'll take that approach since this has
7950                          * to do with the sa_res_key.
7951                          */
7952                         mtx_unlock(&lun->lun_lock);
7953                         free(ctsio->kern_data_ptr, M_CTL);
7954                         ctl_set_invalid_field(ctsio,
7955                                               /*sks_valid*/ 1,
7956                                               /*command*/ 0,
7957                                               /*field*/ 8,
7958                                               /*bit_valid*/ 0,
7959                                               /*bit*/ 0);
7960                         ctl_done((union ctl_io *)ctsio);
7961                         return (1);
7962                 }
7963
7964                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
7965                         if (ctl_get_prkey(lun, i) != sa_res_key)
7966                                 continue;
7967
7968                         found = 1;
7969                         ctl_clr_prkey(lun, i);
7970                         lun->pr_key_count--;
7971                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7972                 }
7973                 if (!found) {
7974                         mtx_unlock(&lun->lun_lock);
7975                         free(ctsio->kern_data_ptr, M_CTL);
7976                         ctl_set_reservation_conflict(ctsio);
7977                         ctl_done((union ctl_io *)ctsio);
7978                         return (CTL_RETVAL_COMPLETE);
7979                 }
7980                 lun->pr_generation++;
7981                 mtx_unlock(&lun->lun_lock);
7982
7983                 /* send msg to other side */
7984                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7985                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7986                 persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7987                 persis_io.pr.pr_info.residx = lun->pr_res_idx;
7988                 persis_io.pr.pr_info.res_type = type;
7989                 memcpy(persis_io.pr.pr_info.sa_res_key,
7990                        param->serv_act_res_key,
7991                        sizeof(param->serv_act_res_key));
7992                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
7993                     sizeof(persis_io.pr), M_WAITOK);
7994         } else {
7995                 /* Reserved but not all registrants */
7996                 /* sa_res_key is res holder */
7997                 if (sa_res_key == ctl_get_prkey(lun, lun->pr_res_idx)) {
7998                         /* validate scope and type */
7999                         if ((cdb->scope_type & SPR_SCOPE_MASK) !=
8000                              SPR_LU_SCOPE) {
8001                                 mtx_unlock(&lun->lun_lock);
8002                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
8003                                                       /*sks_valid*/ 1,
8004                                                       /*command*/ 1,
8005                                                       /*field*/ 2,
8006                                                       /*bit_valid*/ 1,
8007                                                       /*bit*/ 4);
8008                                 ctl_done((union ctl_io *)ctsio);
8009                                 return (1);
8010                         }
8011
8012                         if (type>8 || type==2 || type==4 || type==0) {
8013                                 mtx_unlock(&lun->lun_lock);
8014                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
8015                                                       /*sks_valid*/ 1,
8016                                                       /*command*/ 1,
8017                                                       /*field*/ 2,
8018                                                       /*bit_valid*/ 1,
8019                                                       /*bit*/ 0);
8020                                 ctl_done((union ctl_io *)ctsio);
8021                                 return (1);
8022                         }
8023
8024                         /*
8025                          * Do the following:
8026                          * if sa_res_key != res_key remove all
8027                          * registrants w/sa_res_key and generate UA
8028                          * for these registrants(Registrations
8029                          * Preempted) if it wasn't an exclusive
8030                          * reservation generate UA(Reservations
8031                          * Preempted) for all other registered nexuses
8032                          * if the type has changed. Establish the new
8033                          * reservation and holder. If res_key and
8034                          * sa_res_key are the same do the above
8035                          * except don't unregister the res holder.
8036                          */
8037
8038                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
8039                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
8040                                         continue;
8041
8042                                 if (sa_res_key == ctl_get_prkey(lun, i)) {
8043                                         ctl_clr_prkey(lun, i);
8044                                         lun->pr_key_count--;
8045                                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8046                                 } else if (type != lun->pr_res_type &&
8047                                     (lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8048                                      lun->pr_res_type == SPR_TYPE_EX_AC_RO)) {
8049                                         ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8050                                 }
8051                         }
8052                         lun->pr_res_type = type;
8053                         if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
8054                             lun->pr_res_type != SPR_TYPE_EX_AC_AR)
8055                                 lun->pr_res_idx = residx;
8056                         else
8057                                 lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8058                         lun->pr_generation++;
8059                         mtx_unlock(&lun->lun_lock);
8060
8061                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8062                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8063                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8064                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8065                         persis_io.pr.pr_info.res_type = type;
8066                         memcpy(persis_io.pr.pr_info.sa_res_key,
8067                                param->serv_act_res_key,
8068                                sizeof(param->serv_act_res_key));
8069                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8070                             sizeof(persis_io.pr), M_WAITOK);
8071                 } else {
8072                         /*
8073                          * sa_res_key is not the res holder just
8074                          * remove registrants
8075                          */
8076                         int found=0;
8077
8078                         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8079                                 if (sa_res_key != ctl_get_prkey(lun, i))
8080                                         continue;
8081
8082                                 found = 1;
8083                                 ctl_clr_prkey(lun, i);
8084                                 lun->pr_key_count--;
8085                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8086                         }
8087
8088                         if (!found) {
8089                                 mtx_unlock(&lun->lun_lock);
8090                                 free(ctsio->kern_data_ptr, M_CTL);
8091                                 ctl_set_reservation_conflict(ctsio);
8092                                 ctl_done((union ctl_io *)ctsio);
8093                                 return (1);
8094                         }
8095                         lun->pr_generation++;
8096                         mtx_unlock(&lun->lun_lock);
8097
8098                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8099                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8100                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8101                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8102                         persis_io.pr.pr_info.res_type = type;
8103                         memcpy(persis_io.pr.pr_info.sa_res_key,
8104                                param->serv_act_res_key,
8105                                sizeof(param->serv_act_res_key));
8106                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8107                             sizeof(persis_io.pr), M_WAITOK);
8108                 }
8109         }
8110         return (0);
8111 }
8112
8113 static void
8114 ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
8115 {
8116         uint64_t sa_res_key;
8117         int i;
8118
8119         sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
8120
8121         if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8122          || lun->pr_res_idx == CTL_PR_NO_RESERVATION
8123          || sa_res_key != ctl_get_prkey(lun, lun->pr_res_idx)) {
8124                 if (sa_res_key == 0) {
8125                         /*
8126                          * Unregister everybody else and build UA for
8127                          * them
8128                          */
8129                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
8130                                 if (i == msg->pr.pr_info.residx ||
8131                                     ctl_get_prkey(lun, i) == 0)
8132                                         continue;
8133
8134                                 ctl_clr_prkey(lun, i);
8135                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8136                         }
8137
8138                         lun->pr_key_count = 1;
8139                         lun->pr_res_type = msg->pr.pr_info.res_type;
8140                         if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
8141                             lun->pr_res_type != SPR_TYPE_EX_AC_AR)
8142                                 lun->pr_res_idx = msg->pr.pr_info.residx;
8143                 } else {
8144                         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8145                                 if (sa_res_key == ctl_get_prkey(lun, i))
8146                                         continue;
8147
8148                                 ctl_clr_prkey(lun, i);
8149                                 lun->pr_key_count--;
8150                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8151                         }
8152                 }
8153         } else {
8154                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8155                         if (i == msg->pr.pr_info.residx ||
8156                             ctl_get_prkey(lun, i) == 0)
8157                                 continue;
8158
8159                         if (sa_res_key == ctl_get_prkey(lun, i)) {
8160                                 ctl_clr_prkey(lun, i);
8161                                 lun->pr_key_count--;
8162                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8163                         } else if (msg->pr.pr_info.res_type != lun->pr_res_type
8164                             && (lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8165                              lun->pr_res_type == SPR_TYPE_EX_AC_RO)) {
8166                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8167                         }
8168                 }
8169                 lun->pr_res_type = msg->pr.pr_info.res_type;
8170                 if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
8171                     lun->pr_res_type != SPR_TYPE_EX_AC_AR)
8172                         lun->pr_res_idx = msg->pr.pr_info.residx;
8173                 else
8174                         lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8175         }
8176         lun->pr_generation++;
8177
8178 }
8179
8180
8181 int
8182 ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
8183 {
8184         int retval;
8185         u_int32_t param_len;
8186         struct scsi_per_res_out *cdb;
8187         struct ctl_lun *lun;
8188         struct scsi_per_res_out_parms* param;
8189         struct ctl_softc *softc;
8190         uint32_t residx;
8191         uint64_t res_key, sa_res_key, key;
8192         uint8_t type;
8193         union ctl_ha_msg persis_io;
8194         int    i;
8195
8196         CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n"));
8197
8198         retval = CTL_RETVAL_COMPLETE;
8199
8200         cdb = (struct scsi_per_res_out *)ctsio->cdb;
8201         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8202         softc = lun->ctl_softc;
8203
8204         /*
8205          * We only support whole-LUN scope.  The scope & type are ignored for
8206          * register, register and ignore existing key and clear.
8207          * We sometimes ignore scope and type on preempts too!!
8208          * Verify reservation type here as well.
8209          */
8210         type = cdb->scope_type & SPR_TYPE_MASK;
8211         if ((cdb->action == SPRO_RESERVE)
8212          || (cdb->action == SPRO_RELEASE)) {
8213                 if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
8214                         ctl_set_invalid_field(/*ctsio*/ ctsio,
8215                                               /*sks_valid*/ 1,
8216                                               /*command*/ 1,
8217                                               /*field*/ 2,
8218                                               /*bit_valid*/ 1,
8219                                               /*bit*/ 4);
8220                         ctl_done((union ctl_io *)ctsio);
8221                         return (CTL_RETVAL_COMPLETE);
8222                 }
8223
8224                 if (type>8 || type==2 || type==4 || type==0) {
8225                         ctl_set_invalid_field(/*ctsio*/ ctsio,
8226                                               /*sks_valid*/ 1,
8227                                               /*command*/ 1,
8228                                               /*field*/ 2,
8229                                               /*bit_valid*/ 1,
8230                                               /*bit*/ 0);
8231                         ctl_done((union ctl_io *)ctsio);
8232                         return (CTL_RETVAL_COMPLETE);
8233                 }
8234         }
8235
8236         param_len = scsi_4btoul(cdb->length);
8237
8238         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
8239                 ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
8240                 ctsio->kern_data_len = param_len;
8241                 ctsio->kern_total_len = param_len;
8242                 ctsio->kern_data_resid = 0;
8243                 ctsio->kern_rel_offset = 0;
8244                 ctsio->kern_sg_entries = 0;
8245                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
8246                 ctsio->be_move_done = ctl_config_move_done;
8247                 ctl_datamove((union ctl_io *)ctsio);
8248
8249                 return (CTL_RETVAL_COMPLETE);
8250         }
8251
8252         param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
8253
8254         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
8255         res_key = scsi_8btou64(param->res_key.key);
8256         sa_res_key = scsi_8btou64(param->serv_act_res_key);
8257
8258         /*
8259          * Validate the reservation key here except for SPRO_REG_IGNO
8260          * This must be done for all other service actions
8261          */
8262         if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
8263                 mtx_lock(&lun->lun_lock);
8264                 if ((key = ctl_get_prkey(lun, residx)) != 0) {
8265                         if (res_key != key) {
8266                                 /*
8267                                  * The current key passed in doesn't match
8268                                  * the one the initiator previously
8269                                  * registered.
8270                                  */
8271                                 mtx_unlock(&lun->lun_lock);
8272                                 free(ctsio->kern_data_ptr, M_CTL);
8273                                 ctl_set_reservation_conflict(ctsio);
8274                                 ctl_done((union ctl_io *)ctsio);
8275                                 return (CTL_RETVAL_COMPLETE);
8276                         }
8277                 } else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
8278                         /*
8279                          * We are not registered
8280                          */
8281                         mtx_unlock(&lun->lun_lock);
8282                         free(ctsio->kern_data_ptr, M_CTL);
8283                         ctl_set_reservation_conflict(ctsio);
8284                         ctl_done((union ctl_io *)ctsio);
8285                         return (CTL_RETVAL_COMPLETE);
8286                 } else if (res_key != 0) {
8287                         /*
8288                          * We are not registered and trying to register but
8289                          * the register key isn't zero.
8290                          */
8291                         mtx_unlock(&lun->lun_lock);
8292                         free(ctsio->kern_data_ptr, M_CTL);
8293                         ctl_set_reservation_conflict(ctsio);
8294                         ctl_done((union ctl_io *)ctsio);
8295                         return (CTL_RETVAL_COMPLETE);
8296                 }
8297                 mtx_unlock(&lun->lun_lock);
8298         }
8299
8300         switch (cdb->action & SPRO_ACTION_MASK) {
8301         case SPRO_REGISTER:
8302         case SPRO_REG_IGNO: {
8303
8304 #if 0
8305                 printf("Registration received\n");
8306 #endif
8307
8308                 /*
8309                  * We don't support any of these options, as we report in
8310                  * the read capabilities request (see
8311                  * ctl_persistent_reserve_in(), above).
8312                  */
8313                 if ((param->flags & SPR_SPEC_I_PT)
8314                  || (param->flags & SPR_ALL_TG_PT)
8315                  || (param->flags & SPR_APTPL)) {
8316                         int bit_ptr;
8317
8318                         if (param->flags & SPR_APTPL)
8319                                 bit_ptr = 0;
8320                         else if (param->flags & SPR_ALL_TG_PT)
8321                                 bit_ptr = 2;
8322                         else /* SPR_SPEC_I_PT */
8323                                 bit_ptr = 3;
8324
8325                         free(ctsio->kern_data_ptr, M_CTL);
8326                         ctl_set_invalid_field(ctsio,
8327                                               /*sks_valid*/ 1,
8328                                               /*command*/ 0,
8329                                               /*field*/ 20,
8330                                               /*bit_valid*/ 1,
8331                                               /*bit*/ bit_ptr);
8332                         ctl_done((union ctl_io *)ctsio);
8333                         return (CTL_RETVAL_COMPLETE);
8334                 }
8335
8336                 mtx_lock(&lun->lun_lock);
8337
8338                 /*
8339                  * The initiator wants to clear the
8340                  * key/unregister.
8341                  */
8342                 if (sa_res_key == 0) {
8343                         if ((res_key == 0
8344                           && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
8345                          || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
8346                           && ctl_get_prkey(lun, residx) == 0)) {
8347                                 mtx_unlock(&lun->lun_lock);
8348                                 goto done;
8349                         }
8350
8351                         ctl_clr_prkey(lun, residx);
8352                         lun->pr_key_count--;
8353
8354                         if (residx == lun->pr_res_idx) {
8355                                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8356                                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8357
8358                                 if ((lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8359                                      lun->pr_res_type == SPR_TYPE_EX_AC_RO) &&
8360                                     lun->pr_key_count) {
8361                                         /*
8362                                          * If the reservation is a registrants
8363                                          * only type we need to generate a UA
8364                                          * for other registered inits.  The
8365                                          * sense code should be RESERVATIONS
8366                                          * RELEASED
8367                                          */
8368
8369                                         for (i = softc->init_min; i < softc->init_max; i++){
8370                                                 if (ctl_get_prkey(lun, i) == 0)
8371                                                         continue;
8372                                                 ctl_est_ua(lun, i,
8373                                                     CTL_UA_RES_RELEASE);
8374                                         }
8375                                 }
8376                                 lun->pr_res_type = 0;
8377                         } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8378                                 if (lun->pr_key_count==0) {
8379                                         lun->flags &= ~CTL_LUN_PR_RESERVED;
8380                                         lun->pr_res_type = 0;
8381                                         lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8382                                 }
8383                         }
8384                         lun->pr_generation++;
8385                         mtx_unlock(&lun->lun_lock);
8386
8387                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8388                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8389                         persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
8390                         persis_io.pr.pr_info.residx = residx;
8391                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8392                             sizeof(persis_io.pr), M_WAITOK);
8393                 } else /* sa_res_key != 0 */ {
8394
8395                         /*
8396                          * If we aren't registered currently then increment
8397                          * the key count and set the registered flag.
8398                          */
8399                         ctl_alloc_prkey(lun, residx);
8400                         if (ctl_get_prkey(lun, residx) == 0)
8401                                 lun->pr_key_count++;
8402                         ctl_set_prkey(lun, residx, sa_res_key);
8403                         lun->pr_generation++;
8404                         mtx_unlock(&lun->lun_lock);
8405
8406                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8407                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8408                         persis_io.pr.pr_info.action = CTL_PR_REG_KEY;
8409                         persis_io.pr.pr_info.residx = residx;
8410                         memcpy(persis_io.pr.pr_info.sa_res_key,
8411                                param->serv_act_res_key,
8412                                sizeof(param->serv_act_res_key));
8413                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8414                             sizeof(persis_io.pr), M_WAITOK);
8415                 }
8416
8417                 break;
8418         }
8419         case SPRO_RESERVE:
8420 #if 0
8421                 printf("Reserve executed type %d\n", type);
8422 #endif
8423                 mtx_lock(&lun->lun_lock);
8424                 if (lun->flags & CTL_LUN_PR_RESERVED) {
8425                         /*
8426                          * if this isn't the reservation holder and it's
8427                          * not a "all registrants" type or if the type is
8428                          * different then we have a conflict
8429                          */
8430                         if ((lun->pr_res_idx != residx
8431                           && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8432                          || lun->pr_res_type != type) {
8433                                 mtx_unlock(&lun->lun_lock);
8434                                 free(ctsio->kern_data_ptr, M_CTL);
8435                                 ctl_set_reservation_conflict(ctsio);
8436                                 ctl_done((union ctl_io *)ctsio);
8437                                 return (CTL_RETVAL_COMPLETE);
8438                         }
8439                         mtx_unlock(&lun->lun_lock);
8440                 } else /* create a reservation */ {
8441                         /*
8442                          * If it's not an "all registrants" type record
8443                          * reservation holder
8444                          */
8445                         if (type != SPR_TYPE_WR_EX_AR
8446                          && type != SPR_TYPE_EX_AC_AR)
8447                                 lun->pr_res_idx = residx; /* Res holder */
8448                         else
8449                                 lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8450
8451                         lun->flags |= CTL_LUN_PR_RESERVED;
8452                         lun->pr_res_type = type;
8453
8454                         mtx_unlock(&lun->lun_lock);
8455
8456                         /* send msg to other side */
8457                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8458                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8459                         persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8460                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8461                         persis_io.pr.pr_info.res_type = type;
8462                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8463                             sizeof(persis_io.pr), M_WAITOK);
8464                 }
8465                 break;
8466
8467         case SPRO_RELEASE:
8468                 mtx_lock(&lun->lun_lock);
8469                 if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8470                         /* No reservation exists return good status */
8471                         mtx_unlock(&lun->lun_lock);
8472                         goto done;
8473                 }
8474                 /*
8475                  * Is this nexus a reservation holder?
8476                  */
8477                 if (lun->pr_res_idx != residx
8478                  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8479                         /*
8480                          * not a res holder return good status but
8481                          * do nothing
8482                          */
8483                         mtx_unlock(&lun->lun_lock);
8484                         goto done;
8485                 }
8486
8487                 if (lun->pr_res_type != type) {
8488                         mtx_unlock(&lun->lun_lock);
8489                         free(ctsio->kern_data_ptr, M_CTL);
8490                         ctl_set_illegal_pr_release(ctsio);
8491                         ctl_done((union ctl_io *)ctsio);
8492                         return (CTL_RETVAL_COMPLETE);
8493                 }
8494
8495                 /* okay to release */
8496                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8497                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8498                 lun->pr_res_type = 0;
8499
8500                 /*
8501                  * if this isn't an exclusive access
8502                  * res generate UA for all other
8503                  * registrants.
8504                  */
8505                 if (type != SPR_TYPE_EX_AC
8506                  && type != SPR_TYPE_WR_EX) {
8507                         for (i = softc->init_min; i < softc->init_max; i++) {
8508                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
8509                                         continue;
8510                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8511                         }
8512                 }
8513                 mtx_unlock(&lun->lun_lock);
8514
8515                 /* Send msg to other side */
8516                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8517                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8518                 persis_io.pr.pr_info.action = CTL_PR_RELEASE;
8519                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8520                      sizeof(persis_io.pr), M_WAITOK);
8521                 break;
8522
8523         case SPRO_CLEAR:
8524                 /* send msg to other side */
8525
8526                 mtx_lock(&lun->lun_lock);
8527                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8528                 lun->pr_res_type = 0;
8529                 lun->pr_key_count = 0;
8530                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8531
8532                 ctl_clr_prkey(lun, residx);
8533                 for (i = 0; i < CTL_MAX_INITIATORS; i++)
8534                         if (ctl_get_prkey(lun, i) != 0) {
8535                                 ctl_clr_prkey(lun, i);
8536                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8537                         }
8538                 lun->pr_generation++;
8539                 mtx_unlock(&lun->lun_lock);
8540
8541                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8542                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8543                 persis_io.pr.pr_info.action = CTL_PR_CLEAR;
8544                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8545                      sizeof(persis_io.pr), M_WAITOK);
8546                 break;
8547
8548         case SPRO_PREEMPT:
8549         case SPRO_PRE_ABO: {
8550                 int nretval;
8551
8552                 nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8553                                           residx, ctsio, cdb, param);
8554                 if (nretval != 0)
8555                         return (CTL_RETVAL_COMPLETE);
8556                 break;
8557         }
8558         default:
8559                 panic("%s: Invalid PR type %#x", __func__, cdb->action);
8560         }
8561
8562 done:
8563         free(ctsio->kern_data_ptr, M_CTL);
8564         ctl_set_success(ctsio);
8565         ctl_done((union ctl_io *)ctsio);
8566
8567         return (retval);
8568 }
8569
8570 /*
8571  * This routine is for handling a message from the other SC pertaining to
8572  * persistent reserve out. All the error checking will have been done
8573  * so only perorming the action need be done here to keep the two
8574  * in sync.
8575  */
8576 static void
8577 ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg)
8578 {
8579         struct ctl_softc *softc = control_softc;
8580         struct ctl_lun *lun;
8581         int i;
8582         uint32_t residx, targ_lun;
8583
8584         targ_lun = msg->hdr.nexus.targ_mapped_lun;
8585         mtx_lock(&softc->ctl_lock);
8586         if ((targ_lun >= CTL_MAX_LUNS) ||
8587             ((lun = softc->ctl_luns[targ_lun]) == NULL)) {
8588                 mtx_unlock(&softc->ctl_lock);
8589                 return;
8590         }
8591         mtx_lock(&lun->lun_lock);
8592         mtx_unlock(&softc->ctl_lock);
8593         if (lun->flags & CTL_LUN_DISABLED) {
8594                 mtx_unlock(&lun->lun_lock);
8595                 return;
8596         }
8597         residx = ctl_get_initindex(&msg->hdr.nexus);
8598         switch(msg->pr.pr_info.action) {
8599         case CTL_PR_REG_KEY:
8600                 ctl_alloc_prkey(lun, msg->pr.pr_info.residx);
8601                 if (ctl_get_prkey(lun, msg->pr.pr_info.residx) == 0)
8602                         lun->pr_key_count++;
8603                 ctl_set_prkey(lun, msg->pr.pr_info.residx,
8604                     scsi_8btou64(msg->pr.pr_info.sa_res_key));
8605                 lun->pr_generation++;
8606                 break;
8607
8608         case CTL_PR_UNREG_KEY:
8609                 ctl_clr_prkey(lun, msg->pr.pr_info.residx);
8610                 lun->pr_key_count--;
8611
8612                 /* XXX Need to see if the reservation has been released */
8613                 /* if so do we need to generate UA? */
8614                 if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8615                         lun->flags &= ~CTL_LUN_PR_RESERVED;
8616                         lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8617
8618                         if ((lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8619                              lun->pr_res_type == SPR_TYPE_EX_AC_RO) &&
8620                             lun->pr_key_count) {
8621                                 /*
8622                                  * If the reservation is a registrants
8623                                  * only type we need to generate a UA
8624                                  * for other registered inits.  The
8625                                  * sense code should be RESERVATIONS
8626                                  * RELEASED
8627                                  */
8628
8629                                 for (i = softc->init_min; i < softc->init_max; i++) {
8630                                         if (ctl_get_prkey(lun, i) == 0)
8631                                                 continue;
8632
8633                                         ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8634                                 }
8635                         }
8636                         lun->pr_res_type = 0;
8637                 } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8638                         if (lun->pr_key_count==0) {
8639                                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8640                                 lun->pr_res_type = 0;
8641                                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8642                         }
8643                 }
8644                 lun->pr_generation++;
8645                 break;
8646
8647         case CTL_PR_RESERVE:
8648                 lun->flags |= CTL_LUN_PR_RESERVED;
8649                 lun->pr_res_type = msg->pr.pr_info.res_type;
8650                 lun->pr_res_idx = msg->pr.pr_info.residx;
8651
8652                 break;
8653
8654         case CTL_PR_RELEASE:
8655                 /*
8656                  * if this isn't an exclusive access res generate UA for all
8657                  * other registrants.
8658                  */
8659                 if (lun->pr_res_type != SPR_TYPE_EX_AC &&
8660                     lun->pr_res_type != SPR_TYPE_WR_EX) {
8661                         for (i = softc->init_min; i < softc->init_max; i++)
8662                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
8663                                         continue;
8664                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8665                 }
8666
8667                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8668                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8669                 lun->pr_res_type = 0;
8670                 break;
8671
8672         case CTL_PR_PREEMPT:
8673                 ctl_pro_preempt_other(lun, msg);
8674                 break;
8675         case CTL_PR_CLEAR:
8676                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8677                 lun->pr_res_type = 0;
8678                 lun->pr_key_count = 0;
8679                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8680
8681                 for (i=0; i < CTL_MAX_INITIATORS; i++) {
8682                         if (ctl_get_prkey(lun, i) == 0)
8683                                 continue;
8684                         ctl_clr_prkey(lun, i);
8685                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8686                 }
8687                 lun->pr_generation++;
8688                 break;
8689         }
8690
8691         mtx_unlock(&lun->lun_lock);
8692 }
8693
8694 int
8695 ctl_read_write(struct ctl_scsiio *ctsio)
8696 {
8697         struct ctl_lun *lun;
8698         struct ctl_lba_len_flags *lbalen;
8699         uint64_t lba;
8700         uint32_t num_blocks;
8701         int flags, retval;
8702         int isread;
8703
8704         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8705
8706         CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
8707
8708         flags = 0;
8709         isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
8710               || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
8711         switch (ctsio->cdb[0]) {
8712         case READ_6:
8713         case WRITE_6: {
8714                 struct scsi_rw_6 *cdb;
8715
8716                 cdb = (struct scsi_rw_6 *)ctsio->cdb;
8717
8718                 lba = scsi_3btoul(cdb->addr);
8719                 /* only 5 bits are valid in the most significant address byte */
8720                 lba &= 0x1fffff;
8721                 num_blocks = cdb->length;
8722                 /*
8723                  * This is correct according to SBC-2.
8724                  */
8725                 if (num_blocks == 0)
8726                         num_blocks = 256;
8727                 break;
8728         }
8729         case READ_10:
8730         case WRITE_10: {
8731                 struct scsi_rw_10 *cdb;
8732
8733                 cdb = (struct scsi_rw_10 *)ctsio->cdb;
8734                 if (cdb->byte2 & SRW10_FUA)
8735                         flags |= CTL_LLF_FUA;
8736                 if (cdb->byte2 & SRW10_DPO)
8737                         flags |= CTL_LLF_DPO;
8738                 lba = scsi_4btoul(cdb->addr);
8739                 num_blocks = scsi_2btoul(cdb->length);
8740                 break;
8741         }
8742         case WRITE_VERIFY_10: {
8743                 struct scsi_write_verify_10 *cdb;
8744
8745                 cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
8746                 flags |= CTL_LLF_FUA;
8747                 if (cdb->byte2 & SWV_DPO)
8748                         flags |= CTL_LLF_DPO;
8749                 lba = scsi_4btoul(cdb->addr);
8750                 num_blocks = scsi_2btoul(cdb->length);
8751                 break;
8752         }
8753         case READ_12:
8754         case WRITE_12: {
8755                 struct scsi_rw_12 *cdb;
8756
8757                 cdb = (struct scsi_rw_12 *)ctsio->cdb;
8758                 if (cdb->byte2 & SRW12_FUA)
8759                         flags |= CTL_LLF_FUA;
8760                 if (cdb->byte2 & SRW12_DPO)
8761                         flags |= CTL_LLF_DPO;
8762                 lba = scsi_4btoul(cdb->addr);
8763                 num_blocks = scsi_4btoul(cdb->length);
8764                 break;
8765         }
8766         case WRITE_VERIFY_12: {
8767                 struct scsi_write_verify_12 *cdb;
8768
8769                 cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
8770                 flags |= CTL_LLF_FUA;
8771                 if (cdb->byte2 & SWV_DPO)
8772                         flags |= CTL_LLF_DPO;
8773                 lba = scsi_4btoul(cdb->addr);
8774                 num_blocks = scsi_4btoul(cdb->length);
8775                 break;
8776         }
8777         case READ_16:
8778         case WRITE_16: {
8779                 struct scsi_rw_16 *cdb;
8780
8781                 cdb = (struct scsi_rw_16 *)ctsio->cdb;
8782                 if (cdb->byte2 & SRW12_FUA)
8783                         flags |= CTL_LLF_FUA;
8784                 if (cdb->byte2 & SRW12_DPO)
8785                         flags |= CTL_LLF_DPO;
8786                 lba = scsi_8btou64(cdb->addr);
8787                 num_blocks = scsi_4btoul(cdb->length);
8788                 break;
8789         }
8790         case WRITE_ATOMIC_16: {
8791                 struct scsi_write_atomic_16 *cdb;
8792
8793                 if (lun->be_lun->atomicblock == 0) {
8794                         ctl_set_invalid_opcode(ctsio);
8795                         ctl_done((union ctl_io *)ctsio);
8796                         return (CTL_RETVAL_COMPLETE);
8797                 }
8798
8799                 cdb = (struct scsi_write_atomic_16 *)ctsio->cdb;
8800                 if (cdb->byte2 & SRW12_FUA)
8801                         flags |= CTL_LLF_FUA;
8802                 if (cdb->byte2 & SRW12_DPO)
8803                         flags |= CTL_LLF_DPO;
8804                 lba = scsi_8btou64(cdb->addr);
8805                 num_blocks = scsi_2btoul(cdb->length);
8806                 if (num_blocks > lun->be_lun->atomicblock) {
8807                         ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
8808                             /*command*/ 1, /*field*/ 12, /*bit_valid*/ 0,
8809                             /*bit*/ 0);
8810                         ctl_done((union ctl_io *)ctsio);
8811                         return (CTL_RETVAL_COMPLETE);
8812                 }
8813                 break;
8814         }
8815         case WRITE_VERIFY_16: {
8816                 struct scsi_write_verify_16 *cdb;
8817
8818                 cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
8819                 flags |= CTL_LLF_FUA;
8820                 if (cdb->byte2 & SWV_DPO)
8821                         flags |= CTL_LLF_DPO;
8822                 lba = scsi_8btou64(cdb->addr);
8823                 num_blocks = scsi_4btoul(cdb->length);
8824                 break;
8825         }
8826         default:
8827                 /*
8828                  * We got a command we don't support.  This shouldn't
8829                  * happen, commands should be filtered out above us.
8830                  */
8831                 ctl_set_invalid_opcode(ctsio);
8832                 ctl_done((union ctl_io *)ctsio);
8833
8834                 return (CTL_RETVAL_COMPLETE);
8835                 break; /* NOTREACHED */
8836         }
8837
8838         /*
8839          * The first check is to make sure we're in bounds, the second
8840          * check is to catch wrap-around problems.  If the lba + num blocks
8841          * is less than the lba, then we've wrapped around and the block
8842          * range is invalid anyway.
8843          */
8844         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8845          || ((lba + num_blocks) < lba)) {
8846                 ctl_set_lba_out_of_range(ctsio);
8847                 ctl_done((union ctl_io *)ctsio);
8848                 return (CTL_RETVAL_COMPLETE);
8849         }
8850
8851         /*
8852          * According to SBC-3, a transfer length of 0 is not an error.
8853          * Note that this cannot happen with WRITE(6) or READ(6), since 0
8854          * translates to 256 blocks for those commands.
8855          */
8856         if (num_blocks == 0) {
8857                 ctl_set_success(ctsio);
8858                 ctl_done((union ctl_io *)ctsio);
8859                 return (CTL_RETVAL_COMPLETE);
8860         }
8861
8862         /* Set FUA and/or DPO if caches are disabled. */
8863         if (isread) {
8864                 if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
8865                     SCP_RCD) != 0)
8866                         flags |= CTL_LLF_FUA | CTL_LLF_DPO;
8867         } else {
8868                 if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
8869                     SCP_WCE) == 0)
8870                         flags |= CTL_LLF_FUA;
8871         }
8872
8873         lbalen = (struct ctl_lba_len_flags *)
8874             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8875         lbalen->lba = lba;
8876         lbalen->len = num_blocks;
8877         lbalen->flags = (isread ? CTL_LLF_READ : CTL_LLF_WRITE) | flags;
8878
8879         ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
8880         ctsio->kern_rel_offset = 0;
8881
8882         CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
8883
8884         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8885         return (retval);
8886 }
8887
8888 static int
8889 ctl_cnw_cont(union ctl_io *io)
8890 {
8891         struct ctl_scsiio *ctsio;
8892         struct ctl_lun *lun;
8893         struct ctl_lba_len_flags *lbalen;
8894         int retval;
8895
8896         ctsio = &io->scsiio;
8897         ctsio->io_hdr.status = CTL_STATUS_NONE;
8898         ctsio->io_hdr.flags &= ~CTL_FLAG_IO_CONT;
8899         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8900         lbalen = (struct ctl_lba_len_flags *)
8901             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8902         lbalen->flags &= ~CTL_LLF_COMPARE;
8903         lbalen->flags |= CTL_LLF_WRITE;
8904
8905         CTL_DEBUG_PRINT(("ctl_cnw_cont: calling data_submit()\n"));
8906         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8907         return (retval);
8908 }
8909
8910 int
8911 ctl_cnw(struct ctl_scsiio *ctsio)
8912 {
8913         struct ctl_lun *lun;
8914         struct ctl_lba_len_flags *lbalen;
8915         uint64_t lba;
8916         uint32_t num_blocks;
8917         int flags, retval;
8918
8919         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8920
8921         CTL_DEBUG_PRINT(("ctl_cnw: command: %#x\n", ctsio->cdb[0]));
8922
8923         flags = 0;
8924         switch (ctsio->cdb[0]) {
8925         case COMPARE_AND_WRITE: {
8926                 struct scsi_compare_and_write *cdb;
8927
8928                 cdb = (struct scsi_compare_and_write *)ctsio->cdb;
8929                 if (cdb->byte2 & SRW10_FUA)
8930                         flags |= CTL_LLF_FUA;
8931                 if (cdb->byte2 & SRW10_DPO)
8932                         flags |= CTL_LLF_DPO;
8933                 lba = scsi_8btou64(cdb->addr);
8934                 num_blocks = cdb->length;
8935                 break;
8936         }
8937         default:
8938                 /*
8939                  * We got a command we don't support.  This shouldn't
8940                  * happen, commands should be filtered out above us.
8941                  */
8942                 ctl_set_invalid_opcode(ctsio);
8943                 ctl_done((union ctl_io *)ctsio);
8944
8945                 return (CTL_RETVAL_COMPLETE);
8946                 break; /* NOTREACHED */
8947         }
8948
8949         /*
8950          * The first check is to make sure we're in bounds, the second
8951          * check is to catch wrap-around problems.  If the lba + num blocks
8952          * is less than the lba, then we've wrapped around and the block
8953          * range is invalid anyway.
8954          */
8955         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8956          || ((lba + num_blocks) < lba)) {
8957                 ctl_set_lba_out_of_range(ctsio);
8958                 ctl_done((union ctl_io *)ctsio);
8959                 return (CTL_RETVAL_COMPLETE);
8960         }
8961
8962         /*
8963          * According to SBC-3, a transfer length of 0 is not an error.
8964          */
8965         if (num_blocks == 0) {
8966                 ctl_set_success(ctsio);
8967                 ctl_done((union ctl_io *)ctsio);
8968                 return (CTL_RETVAL_COMPLETE);
8969         }
8970
8971         /* Set FUA if write cache is disabled. */
8972         if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
8973             SCP_WCE) == 0)
8974                 flags |= CTL_LLF_FUA;
8975
8976         ctsio->kern_total_len = 2 * num_blocks * lun->be_lun->blocksize;
8977         ctsio->kern_rel_offset = 0;
8978
8979         /*
8980          * Set the IO_CONT flag, so that if this I/O gets passed to
8981          * ctl_data_submit_done(), it'll get passed back to
8982          * ctl_ctl_cnw_cont() for further processing.
8983          */
8984         ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
8985         ctsio->io_cont = ctl_cnw_cont;
8986
8987         lbalen = (struct ctl_lba_len_flags *)
8988             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8989         lbalen->lba = lba;
8990         lbalen->len = num_blocks;
8991         lbalen->flags = CTL_LLF_COMPARE | flags;
8992
8993         CTL_DEBUG_PRINT(("ctl_cnw: calling data_submit()\n"));
8994         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8995         return (retval);
8996 }
8997
8998 int
8999 ctl_verify(struct ctl_scsiio *ctsio)
9000 {
9001         struct ctl_lun *lun;
9002         struct ctl_lba_len_flags *lbalen;
9003         uint64_t lba;
9004         uint32_t num_blocks;
9005         int bytchk, flags;
9006         int retval;
9007
9008         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9009
9010         CTL_DEBUG_PRINT(("ctl_verify: command: %#x\n", ctsio->cdb[0]));
9011
9012         bytchk = 0;
9013         flags = CTL_LLF_FUA;
9014         switch (ctsio->cdb[0]) {
9015         case VERIFY_10: {
9016                 struct scsi_verify_10 *cdb;
9017
9018                 cdb = (struct scsi_verify_10 *)ctsio->cdb;
9019                 if (cdb->byte2 & SVFY_BYTCHK)
9020                         bytchk = 1;
9021                 if (cdb->byte2 & SVFY_DPO)
9022                         flags |= CTL_LLF_DPO;
9023                 lba = scsi_4btoul(cdb->addr);
9024                 num_blocks = scsi_2btoul(cdb->length);
9025                 break;
9026         }
9027         case VERIFY_12: {
9028                 struct scsi_verify_12 *cdb;
9029
9030                 cdb = (struct scsi_verify_12 *)ctsio->cdb;
9031                 if (cdb->byte2 & SVFY_BYTCHK)
9032                         bytchk = 1;
9033                 if (cdb->byte2 & SVFY_DPO)
9034                         flags |= CTL_LLF_DPO;
9035                 lba = scsi_4btoul(cdb->addr);
9036                 num_blocks = scsi_4btoul(cdb->length);
9037                 break;
9038         }
9039         case VERIFY_16: {
9040                 struct scsi_rw_16 *cdb;
9041
9042                 cdb = (struct scsi_rw_16 *)ctsio->cdb;
9043                 if (cdb->byte2 & SVFY_BYTCHK)
9044                         bytchk = 1;
9045                 if (cdb->byte2 & SVFY_DPO)
9046                         flags |= CTL_LLF_DPO;
9047                 lba = scsi_8btou64(cdb->addr);
9048                 num_blocks = scsi_4btoul(cdb->length);
9049                 break;
9050         }
9051         default:
9052                 /*
9053                  * We got a command we don't support.  This shouldn't
9054                  * happen, commands should be filtered out above us.
9055                  */
9056                 ctl_set_invalid_opcode(ctsio);
9057                 ctl_done((union ctl_io *)ctsio);
9058                 return (CTL_RETVAL_COMPLETE);
9059         }
9060
9061         /*
9062          * The first check is to make sure we're in bounds, the second
9063          * check is to catch wrap-around problems.  If the lba + num blocks
9064          * is less than the lba, then we've wrapped around and the block
9065          * range is invalid anyway.
9066          */
9067         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9068          || ((lba + num_blocks) < lba)) {
9069                 ctl_set_lba_out_of_range(ctsio);
9070                 ctl_done((union ctl_io *)ctsio);
9071                 return (CTL_RETVAL_COMPLETE);
9072         }
9073
9074         /*
9075          * According to SBC-3, a transfer length of 0 is not an error.
9076          */
9077         if (num_blocks == 0) {
9078                 ctl_set_success(ctsio);
9079                 ctl_done((union ctl_io *)ctsio);
9080                 return (CTL_RETVAL_COMPLETE);
9081         }
9082
9083         lbalen = (struct ctl_lba_len_flags *)
9084             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9085         lbalen->lba = lba;
9086         lbalen->len = num_blocks;
9087         if (bytchk) {
9088                 lbalen->flags = CTL_LLF_COMPARE | flags;
9089                 ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9090         } else {
9091                 lbalen->flags = CTL_LLF_VERIFY | flags;
9092                 ctsio->kern_total_len = 0;
9093         }
9094         ctsio->kern_rel_offset = 0;
9095
9096         CTL_DEBUG_PRINT(("ctl_verify: calling data_submit()\n"));
9097         retval = lun->backend->data_submit((union ctl_io *)ctsio);
9098         return (retval);
9099 }
9100
9101 int
9102 ctl_report_luns(struct ctl_scsiio *ctsio)
9103 {
9104         struct ctl_softc *softc;
9105         struct scsi_report_luns *cdb;
9106         struct scsi_report_luns_data *lun_data;
9107         struct ctl_lun *lun, *request_lun;
9108         struct ctl_port *port;
9109         int num_luns, retval;
9110         uint32_t alloc_len, lun_datalen;
9111         int num_filled;
9112         uint32_t initidx, targ_lun_id, lun_id;
9113
9114         retval = CTL_RETVAL_COMPLETE;
9115         cdb = (struct scsi_report_luns *)ctsio->cdb;
9116         port = ctl_io_port(&ctsio->io_hdr);
9117         softc = port->ctl_softc;
9118
9119         CTL_DEBUG_PRINT(("ctl_report_luns\n"));
9120
9121         mtx_lock(&softc->ctl_lock);
9122         num_luns = 0;
9123         for (targ_lun_id = 0; targ_lun_id < CTL_MAX_LUNS; targ_lun_id++) {
9124                 if (ctl_lun_map_from_port(port, targ_lun_id) < CTL_MAX_LUNS)
9125                         num_luns++;
9126         }
9127         mtx_unlock(&softc->ctl_lock);
9128
9129         switch (cdb->select_report) {
9130         case RPL_REPORT_DEFAULT:
9131         case RPL_REPORT_ALL:
9132         case RPL_REPORT_NONSUBSID:
9133                 break;
9134         case RPL_REPORT_WELLKNOWN:
9135         case RPL_REPORT_ADMIN:
9136         case RPL_REPORT_CONGLOM:
9137                 num_luns = 0;
9138                 break;
9139         default:
9140                 ctl_set_invalid_field(ctsio,
9141                                       /*sks_valid*/ 1,
9142                                       /*command*/ 1,
9143                                       /*field*/ 2,
9144                                       /*bit_valid*/ 0,
9145                                       /*bit*/ 0);
9146                 ctl_done((union ctl_io *)ctsio);
9147                 return (retval);
9148                 break; /* NOTREACHED */
9149         }
9150
9151         alloc_len = scsi_4btoul(cdb->length);
9152         /*
9153          * The initiator has to allocate at least 16 bytes for this request,
9154          * so he can at least get the header and the first LUN.  Otherwise
9155          * we reject the request (per SPC-3 rev 14, section 6.21).
9156          */
9157         if (alloc_len < (sizeof(struct scsi_report_luns_data) +
9158             sizeof(struct scsi_report_luns_lundata))) {
9159                 ctl_set_invalid_field(ctsio,
9160                                       /*sks_valid*/ 1,
9161                                       /*command*/ 1,
9162                                       /*field*/ 6,
9163                                       /*bit_valid*/ 0,
9164                                       /*bit*/ 0);
9165                 ctl_done((union ctl_io *)ctsio);
9166                 return (retval);
9167         }
9168
9169         request_lun = (struct ctl_lun *)
9170                 ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9171
9172         lun_datalen = sizeof(*lun_data) +
9173                 (num_luns * sizeof(struct scsi_report_luns_lundata));
9174
9175         ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO);
9176         lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
9177         ctsio->kern_sg_entries = 0;
9178
9179         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9180
9181         mtx_lock(&softc->ctl_lock);
9182         for (targ_lun_id = 0, num_filled = 0; targ_lun_id < CTL_MAX_LUNS && num_filled < num_luns; targ_lun_id++) {
9183                 lun_id = ctl_lun_map_from_port(port, targ_lun_id);
9184                 if (lun_id >= CTL_MAX_LUNS)
9185                         continue;
9186                 lun = softc->ctl_luns[lun_id];
9187                 if (lun == NULL)
9188                         continue;
9189
9190                 be64enc(lun_data->luns[num_filled++].lundata,
9191                     ctl_encode_lun(targ_lun_id));
9192
9193                 /*
9194                  * According to SPC-3, rev 14 section 6.21:
9195                  *
9196                  * "The execution of a REPORT LUNS command to any valid and
9197                  * installed logical unit shall clear the REPORTED LUNS DATA
9198                  * HAS CHANGED unit attention condition for all logical
9199                  * units of that target with respect to the requesting
9200                  * initiator. A valid and installed logical unit is one
9201                  * having a PERIPHERAL QUALIFIER of 000b in the standard
9202                  * INQUIRY data (see 6.4.2)."
9203                  *
9204                  * If request_lun is NULL, the LUN this report luns command
9205                  * was issued to is either disabled or doesn't exist. In that
9206                  * case, we shouldn't clear any pending lun change unit
9207                  * attention.
9208                  */
9209                 if (request_lun != NULL) {
9210                         mtx_lock(&lun->lun_lock);
9211                         ctl_clr_ua(lun, initidx, CTL_UA_LUN_CHANGE);
9212                         mtx_unlock(&lun->lun_lock);
9213                 }
9214         }
9215         mtx_unlock(&softc->ctl_lock);
9216
9217         /*
9218          * It's quite possible that we've returned fewer LUNs than we allocated
9219          * space for.  Trim it.
9220          */
9221         lun_datalen = sizeof(*lun_data) +
9222                 (num_filled * sizeof(struct scsi_report_luns_lundata));
9223
9224         if (lun_datalen < alloc_len) {
9225                 ctsio->residual = alloc_len - lun_datalen;
9226                 ctsio->kern_data_len = lun_datalen;
9227                 ctsio->kern_total_len = lun_datalen;
9228         } else {
9229                 ctsio->residual = 0;
9230                 ctsio->kern_data_len = alloc_len;
9231                 ctsio->kern_total_len = alloc_len;
9232         }
9233         ctsio->kern_data_resid = 0;
9234         ctsio->kern_rel_offset = 0;
9235         ctsio->kern_sg_entries = 0;
9236
9237         /*
9238          * We set this to the actual data length, regardless of how much
9239          * space we actually have to return results.  If the user looks at
9240          * this value, he'll know whether or not he allocated enough space
9241          * and reissue the command if necessary.  We don't support well
9242          * known logical units, so if the user asks for that, return none.
9243          */
9244         scsi_ulto4b(lun_datalen - 8, lun_data->length);
9245
9246         /*
9247          * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
9248          * this request.
9249          */
9250         ctl_set_success(ctsio);
9251         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9252         ctsio->be_move_done = ctl_config_move_done;
9253         ctl_datamove((union ctl_io *)ctsio);
9254         return (retval);
9255 }
9256
9257 int
9258 ctl_request_sense(struct ctl_scsiio *ctsio)
9259 {
9260         struct scsi_request_sense *cdb;
9261         struct scsi_sense_data *sense_ptr;
9262         struct ctl_softc *ctl_softc;
9263         struct ctl_lun *lun;
9264         uint32_t initidx;
9265         int have_error;
9266         scsi_sense_data_type sense_format;
9267         ctl_ua_type ua_type;
9268
9269         cdb = (struct scsi_request_sense *)ctsio->cdb;
9270
9271         ctl_softc = control_softc;
9272         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9273
9274         CTL_DEBUG_PRINT(("ctl_request_sense\n"));
9275
9276         /*
9277          * Determine which sense format the user wants.
9278          */
9279         if (cdb->byte2 & SRS_DESC)
9280                 sense_format = SSD_TYPE_DESC;
9281         else
9282                 sense_format = SSD_TYPE_FIXED;
9283
9284         ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
9285         sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
9286         ctsio->kern_sg_entries = 0;
9287
9288         /*
9289          * struct scsi_sense_data, which is currently set to 256 bytes, is
9290          * larger than the largest allowed value for the length field in the
9291          * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
9292          */
9293         ctsio->residual = 0;
9294         ctsio->kern_data_len = cdb->length;
9295         ctsio->kern_total_len = cdb->length;
9296
9297         ctsio->kern_data_resid = 0;
9298         ctsio->kern_rel_offset = 0;
9299         ctsio->kern_sg_entries = 0;
9300
9301         /*
9302          * If we don't have a LUN, we don't have any pending sense.
9303          */
9304         if (lun == NULL)
9305                 goto no_sense;
9306
9307         have_error = 0;
9308         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9309         /*
9310          * Check for pending sense, and then for pending unit attentions.
9311          * Pending sense gets returned first, then pending unit attentions.
9312          */
9313         mtx_lock(&lun->lun_lock);
9314 #ifdef CTL_WITH_CA
9315         if (ctl_is_set(lun->have_ca, initidx)) {
9316                 scsi_sense_data_type stored_format;
9317
9318                 /*
9319                  * Check to see which sense format was used for the stored
9320                  * sense data.
9321                  */
9322                 stored_format = scsi_sense_type(&lun->pending_sense[initidx]);
9323
9324                 /*
9325                  * If the user requested a different sense format than the
9326                  * one we stored, then we need to convert it to the other
9327                  * format.  If we're going from descriptor to fixed format
9328                  * sense data, we may lose things in translation, depending
9329                  * on what options were used.
9330                  *
9331                  * If the stored format is SSD_TYPE_NONE (i.e. invalid),
9332                  * for some reason we'll just copy it out as-is.
9333                  */
9334                 if ((stored_format == SSD_TYPE_FIXED)
9335                  && (sense_format == SSD_TYPE_DESC))
9336                         ctl_sense_to_desc((struct scsi_sense_data_fixed *)
9337                             &lun->pending_sense[initidx],
9338                             (struct scsi_sense_data_desc *)sense_ptr);
9339                 else if ((stored_format == SSD_TYPE_DESC)
9340                       && (sense_format == SSD_TYPE_FIXED))
9341                         ctl_sense_to_fixed((struct scsi_sense_data_desc *)
9342                             &lun->pending_sense[initidx],
9343                             (struct scsi_sense_data_fixed *)sense_ptr);
9344                 else
9345                         memcpy(sense_ptr, &lun->pending_sense[initidx],
9346                                MIN(sizeof(*sense_ptr),
9347                                sizeof(lun->pending_sense[initidx])));
9348
9349                 ctl_clear_mask(lun->have_ca, initidx);
9350                 have_error = 1;
9351         } else
9352 #endif
9353         {
9354                 ua_type = ctl_build_ua(lun, initidx, sense_ptr, sense_format);
9355                 if (ua_type != CTL_UA_NONE)
9356                         have_error = 1;
9357                 if (ua_type == CTL_UA_LUN_CHANGE) {
9358                         mtx_unlock(&lun->lun_lock);
9359                         mtx_lock(&ctl_softc->ctl_lock);
9360                         ctl_clr_ua_allluns(ctl_softc, initidx, ua_type);
9361                         mtx_unlock(&ctl_softc->ctl_lock);
9362                         mtx_lock(&lun->lun_lock);
9363                 }
9364
9365         }
9366         mtx_unlock(&lun->lun_lock);
9367
9368         /*
9369          * We already have a pending error, return it.
9370          */
9371         if (have_error != 0) {
9372                 /*
9373                  * We report the SCSI status as OK, since the status of the
9374                  * request sense command itself is OK.
9375                  * We report 0 for the sense length, because we aren't doing
9376                  * autosense in this case.  We're reporting sense as
9377                  * parameter data.
9378                  */
9379                 ctl_set_success(ctsio);
9380                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9381                 ctsio->be_move_done = ctl_config_move_done;
9382                 ctl_datamove((union ctl_io *)ctsio);
9383                 return (CTL_RETVAL_COMPLETE);
9384         }
9385
9386 no_sense:
9387
9388         /*
9389          * No sense information to report, so we report that everything is
9390          * okay.
9391          */
9392         ctl_set_sense_data(sense_ptr,
9393                            lun,
9394                            sense_format,
9395                            /*current_error*/ 1,
9396                            /*sense_key*/ SSD_KEY_NO_SENSE,
9397                            /*asc*/ 0x00,
9398                            /*ascq*/ 0x00,
9399                            SSD_ELEM_NONE);
9400
9401         /*
9402          * We report 0 for the sense length, because we aren't doing
9403          * autosense in this case.  We're reporting sense as parameter data.
9404          */
9405         ctl_set_success(ctsio);
9406         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9407         ctsio->be_move_done = ctl_config_move_done;
9408         ctl_datamove((union ctl_io *)ctsio);
9409         return (CTL_RETVAL_COMPLETE);
9410 }
9411
9412 int
9413 ctl_tur(struct ctl_scsiio *ctsio)
9414 {
9415
9416         CTL_DEBUG_PRINT(("ctl_tur\n"));
9417
9418         ctl_set_success(ctsio);
9419         ctl_done((union ctl_io *)ctsio);
9420
9421         return (CTL_RETVAL_COMPLETE);
9422 }
9423
9424 /*
9425  * SCSI VPD page 0x00, the Supported VPD Pages page.
9426  */
9427 static int
9428 ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
9429 {
9430         struct scsi_vpd_supported_pages *pages;
9431         int sup_page_size;
9432         struct ctl_lun *lun;
9433         int p;
9434
9435         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9436
9437         sup_page_size = sizeof(struct scsi_vpd_supported_pages) *
9438             SCSI_EVPD_NUM_SUPPORTED_PAGES;
9439         ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO);
9440         pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
9441         ctsio->kern_sg_entries = 0;
9442
9443         if (sup_page_size < alloc_len) {
9444                 ctsio->residual = alloc_len - sup_page_size;
9445                 ctsio->kern_data_len = sup_page_size;
9446                 ctsio->kern_total_len = sup_page_size;
9447         } else {
9448                 ctsio->residual = 0;
9449                 ctsio->kern_data_len = alloc_len;
9450                 ctsio->kern_total_len = alloc_len;
9451         }
9452         ctsio->kern_data_resid = 0;
9453         ctsio->kern_rel_offset = 0;
9454         ctsio->kern_sg_entries = 0;
9455
9456         /*
9457          * The control device is always connected.  The disk device, on the
9458          * other hand, may not be online all the time.  Need to change this
9459          * to figure out whether the disk device is actually online or not.
9460          */
9461         if (lun != NULL)
9462                 pages->device = (SID_QUAL_LU_CONNECTED << 5) |
9463                                 lun->be_lun->lun_type;
9464         else
9465                 pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9466
9467         p = 0;
9468         /* Supported VPD pages */
9469         pages->page_list[p++] = SVPD_SUPPORTED_PAGES;
9470         /* Serial Number */
9471         pages->page_list[p++] = SVPD_UNIT_SERIAL_NUMBER;
9472         /* Device Identification */
9473         pages->page_list[p++] = SVPD_DEVICE_ID;
9474         /* Extended INQUIRY Data */
9475         pages->page_list[p++] = SVPD_EXTENDED_INQUIRY_DATA;
9476         /* Mode Page Policy */
9477         pages->page_list[p++] = SVPD_MODE_PAGE_POLICY;
9478         /* SCSI Ports */
9479         pages->page_list[p++] = SVPD_SCSI_PORTS;
9480         /* Third-party Copy */
9481         pages->page_list[p++] = SVPD_SCSI_TPC;
9482         if (lun != NULL && lun->be_lun->lun_type == T_DIRECT) {
9483                 /* Block limits */
9484                 pages->page_list[p++] = SVPD_BLOCK_LIMITS;
9485                 /* Block Device Characteristics */
9486                 pages->page_list[p++] = SVPD_BDC;
9487                 /* Logical Block Provisioning */
9488                 pages->page_list[p++] = SVPD_LBP;
9489         }
9490         pages->length = p;
9491
9492         ctl_set_success(ctsio);
9493         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9494         ctsio->be_move_done = ctl_config_move_done;
9495         ctl_datamove((union ctl_io *)ctsio);
9496         return (CTL_RETVAL_COMPLETE);
9497 }
9498
9499 /*
9500  * SCSI VPD page 0x80, the Unit Serial Number page.
9501  */
9502 static int
9503 ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
9504 {
9505         struct scsi_vpd_unit_serial_number *sn_ptr;
9506         struct ctl_lun *lun;
9507         int data_len;
9508
9509         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9510
9511         data_len = 4 + CTL_SN_LEN;
9512         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9513         sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
9514         if (data_len < alloc_len) {
9515                 ctsio->residual = alloc_len - data_len;
9516                 ctsio->kern_data_len = data_len;
9517                 ctsio->kern_total_len = data_len;
9518         } else {
9519                 ctsio->residual = 0;
9520                 ctsio->kern_data_len = alloc_len;
9521                 ctsio->kern_total_len = alloc_len;
9522         }
9523         ctsio->kern_data_resid = 0;
9524         ctsio->kern_rel_offset = 0;
9525         ctsio->kern_sg_entries = 0;
9526
9527         /*
9528          * The control device is always connected.  The disk device, on the
9529          * other hand, may not be online all the time.  Need to change this
9530          * to figure out whether the disk device is actually online or not.
9531          */
9532         if (lun != NULL)
9533                 sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9534                                   lun->be_lun->lun_type;
9535         else
9536                 sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9537
9538         sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9539         sn_ptr->length = CTL_SN_LEN;
9540         /*
9541          * If we don't have a LUN, we just leave the serial number as
9542          * all spaces.
9543          */
9544         if (lun != NULL) {
9545                 strncpy((char *)sn_ptr->serial_num,
9546                         (char *)lun->be_lun->serial_num, CTL_SN_LEN);
9547         } else
9548                 memset(sn_ptr->serial_num, 0x20, CTL_SN_LEN);
9549
9550         ctl_set_success(ctsio);
9551         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9552         ctsio->be_move_done = ctl_config_move_done;
9553         ctl_datamove((union ctl_io *)ctsio);
9554         return (CTL_RETVAL_COMPLETE);
9555 }
9556
9557
9558 /*
9559  * SCSI VPD page 0x86, the Extended INQUIRY Data page.
9560  */
9561 static int
9562 ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len)
9563 {
9564         struct scsi_vpd_extended_inquiry_data *eid_ptr;
9565         struct ctl_lun *lun;
9566         int data_len;
9567
9568         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9569
9570         data_len = sizeof(struct scsi_vpd_extended_inquiry_data);
9571         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9572         eid_ptr = (struct scsi_vpd_extended_inquiry_data *)ctsio->kern_data_ptr;
9573         ctsio->kern_sg_entries = 0;
9574
9575         if (data_len < alloc_len) {
9576                 ctsio->residual = alloc_len - data_len;
9577                 ctsio->kern_data_len = data_len;
9578                 ctsio->kern_total_len = data_len;
9579         } else {
9580                 ctsio->residual = 0;
9581                 ctsio->kern_data_len = alloc_len;
9582                 ctsio->kern_total_len = alloc_len;
9583         }
9584         ctsio->kern_data_resid = 0;
9585         ctsio->kern_rel_offset = 0;
9586         ctsio->kern_sg_entries = 0;
9587
9588         /*
9589          * The control device is always connected.  The disk device, on the
9590          * other hand, may not be online all the time.
9591          */
9592         if (lun != NULL)
9593                 eid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9594                                      lun->be_lun->lun_type;
9595         else
9596                 eid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9597         eid_ptr->page_code = SVPD_EXTENDED_INQUIRY_DATA;
9598         scsi_ulto2b(data_len - 4, eid_ptr->page_length);
9599         /*
9600          * We support head of queue, ordered and simple tags.
9601          */
9602         eid_ptr->flags2 = SVPD_EID_HEADSUP | SVPD_EID_ORDSUP | SVPD_EID_SIMPSUP;
9603         /*
9604          * Volatile cache supported.
9605          */
9606         eid_ptr->flags3 = SVPD_EID_V_SUP;
9607
9608         /*
9609          * This means that we clear the REPORTED LUNS DATA HAS CHANGED unit
9610          * attention for a particular IT nexus on all LUNs once we report
9611          * it to that nexus once.  This bit is required as of SPC-4.
9612          */
9613         eid_ptr->flags4 = SVPD_EID_LUICLT;
9614
9615         /*
9616          * XXX KDM in order to correctly answer this, we would need
9617          * information from the SIM to determine how much sense data it
9618          * can send.  So this would really be a path inquiry field, most
9619          * likely.  This can be set to a maximum of 252 according to SPC-4,
9620          * but the hardware may or may not be able to support that much.
9621          * 0 just means that the maximum sense data length is not reported.
9622          */
9623         eid_ptr->max_sense_length = 0;
9624
9625         ctl_set_success(ctsio);
9626         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9627         ctsio->be_move_done = ctl_config_move_done;
9628         ctl_datamove((union ctl_io *)ctsio);
9629         return (CTL_RETVAL_COMPLETE);
9630 }
9631
9632 static int
9633 ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len)
9634 {
9635         struct scsi_vpd_mode_page_policy *mpp_ptr;
9636         struct ctl_lun *lun;
9637         int data_len;
9638
9639         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9640
9641         data_len = sizeof(struct scsi_vpd_mode_page_policy) +
9642             sizeof(struct scsi_vpd_mode_page_policy_descr);
9643
9644         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9645         mpp_ptr = (struct scsi_vpd_mode_page_policy *)ctsio->kern_data_ptr;
9646         ctsio->kern_sg_entries = 0;
9647
9648         if (data_len < alloc_len) {
9649                 ctsio->residual = alloc_len - data_len;
9650                 ctsio->kern_data_len = data_len;
9651                 ctsio->kern_total_len = data_len;
9652         } else {
9653                 ctsio->residual = 0;
9654                 ctsio->kern_data_len = alloc_len;
9655                 ctsio->kern_total_len = alloc_len;
9656         }
9657         ctsio->kern_data_resid = 0;
9658         ctsio->kern_rel_offset = 0;
9659         ctsio->kern_sg_entries = 0;
9660
9661         /*
9662          * The control device is always connected.  The disk device, on the
9663          * other hand, may not be online all the time.
9664          */
9665         if (lun != NULL)
9666                 mpp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9667                                      lun->be_lun->lun_type;
9668         else
9669                 mpp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9670         mpp_ptr->page_code = SVPD_MODE_PAGE_POLICY;
9671         scsi_ulto2b(data_len - 4, mpp_ptr->page_length);
9672         mpp_ptr->descr[0].page_code = 0x3f;
9673         mpp_ptr->descr[0].subpage_code = 0xff;
9674         mpp_ptr->descr[0].policy = SVPD_MPP_SHARED;
9675
9676         ctl_set_success(ctsio);
9677         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9678         ctsio->be_move_done = ctl_config_move_done;
9679         ctl_datamove((union ctl_io *)ctsio);
9680         return (CTL_RETVAL_COMPLETE);
9681 }
9682
9683 /*
9684  * SCSI VPD page 0x83, the Device Identification page.
9685  */
9686 static int
9687 ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
9688 {
9689         struct scsi_vpd_device_id *devid_ptr;
9690         struct scsi_vpd_id_descriptor *desc;
9691         struct ctl_softc *softc;
9692         struct ctl_lun *lun;
9693         struct ctl_port *port;
9694         int data_len, g;
9695         uint8_t proto;
9696
9697         softc = control_softc;
9698
9699         port = ctl_io_port(&ctsio->io_hdr);
9700         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9701
9702         data_len = sizeof(struct scsi_vpd_device_id) +
9703             sizeof(struct scsi_vpd_id_descriptor) +
9704                 sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
9705             sizeof(struct scsi_vpd_id_descriptor) +
9706                 sizeof(struct scsi_vpd_id_trgt_port_grp_id);
9707         if (lun && lun->lun_devid)
9708                 data_len += lun->lun_devid->len;
9709         if (port && port->port_devid)
9710                 data_len += port->port_devid->len;
9711         if (port && port->target_devid)
9712                 data_len += port->target_devid->len;
9713
9714         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9715         devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
9716         ctsio->kern_sg_entries = 0;
9717
9718         if (data_len < alloc_len) {
9719                 ctsio->residual = alloc_len - data_len;
9720                 ctsio->kern_data_len = data_len;
9721                 ctsio->kern_total_len = data_len;
9722         } else {
9723                 ctsio->residual = 0;
9724                 ctsio->kern_data_len = alloc_len;
9725                 ctsio->kern_total_len = alloc_len;
9726         }
9727         ctsio->kern_data_resid = 0;
9728         ctsio->kern_rel_offset = 0;
9729         ctsio->kern_sg_entries = 0;
9730
9731         /*
9732          * The control device is always connected.  The disk device, on the
9733          * other hand, may not be online all the time.
9734          */
9735         if (lun != NULL)
9736                 devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9737                                      lun->be_lun->lun_type;
9738         else
9739                 devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9740         devid_ptr->page_code = SVPD_DEVICE_ID;
9741         scsi_ulto2b(data_len - 4, devid_ptr->length);
9742
9743         if (port && port->port_type == CTL_PORT_FC)
9744                 proto = SCSI_PROTO_FC << 4;
9745         else if (port && port->port_type == CTL_PORT_ISCSI)
9746                 proto = SCSI_PROTO_ISCSI << 4;
9747         else
9748                 proto = SCSI_PROTO_SPI << 4;
9749         desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
9750
9751         /*
9752          * We're using a LUN association here.  i.e., this device ID is a
9753          * per-LUN identifier.
9754          */
9755         if (lun && lun->lun_devid) {
9756                 memcpy(desc, lun->lun_devid->data, lun->lun_devid->len);
9757                 desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9758                     lun->lun_devid->len);
9759         }
9760
9761         /*
9762          * This is for the WWPN which is a port association.
9763          */
9764         if (port && port->port_devid) {
9765                 memcpy(desc, port->port_devid->data, port->port_devid->len);
9766                 desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9767                     port->port_devid->len);
9768         }
9769
9770         /*
9771          * This is for the Relative Target Port(type 4h) identifier
9772          */
9773         desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9774         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9775             SVPD_ID_TYPE_RELTARG;
9776         desc->length = 4;
9777         scsi_ulto2b(ctsio->io_hdr.nexus.targ_port, &desc->identifier[2]);
9778         desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9779             sizeof(struct scsi_vpd_id_rel_trgt_port_id));
9780
9781         /*
9782          * This is for the Target Port Group(type 5h) identifier
9783          */
9784         desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9785         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9786             SVPD_ID_TYPE_TPORTGRP;
9787         desc->length = 4;
9788         if (softc->is_single ||
9789             (port && port->status & CTL_PORT_STATUS_HA_SHARED))
9790                 g = 1;
9791         else
9792                 g = 2 + ctsio->io_hdr.nexus.targ_port / softc->port_cnt;
9793         scsi_ulto2b(g, &desc->identifier[2]);
9794         desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9795             sizeof(struct scsi_vpd_id_trgt_port_grp_id));
9796
9797         /*
9798          * This is for the Target identifier
9799          */
9800         if (port && port->target_devid) {
9801                 memcpy(desc, port->target_devid->data, port->target_devid->len);
9802         }
9803
9804         ctl_set_success(ctsio);
9805         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9806         ctsio->be_move_done = ctl_config_move_done;
9807         ctl_datamove((union ctl_io *)ctsio);
9808         return (CTL_RETVAL_COMPLETE);
9809 }
9810
9811 static int
9812 ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio, int alloc_len)
9813 {
9814         struct ctl_softc *softc = control_softc;
9815         struct scsi_vpd_scsi_ports *sp;
9816         struct scsi_vpd_port_designation *pd;
9817         struct scsi_vpd_port_designation_cont *pdc;
9818         struct ctl_lun *lun;
9819         struct ctl_port *port;
9820         int data_len, num_target_ports, iid_len, id_len;
9821
9822         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9823
9824         num_target_ports = 0;
9825         iid_len = 0;
9826         id_len = 0;
9827         mtx_lock(&softc->ctl_lock);
9828         STAILQ_FOREACH(port, &softc->port_list, links) {
9829                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
9830                         continue;
9831                 if (lun != NULL &&
9832                     ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
9833                         continue;
9834                 num_target_ports++;
9835                 if (port->init_devid)
9836                         iid_len += port->init_devid->len;
9837                 if (port->port_devid)
9838                         id_len += port->port_devid->len;
9839         }
9840         mtx_unlock(&softc->ctl_lock);
9841
9842         data_len = sizeof(struct scsi_vpd_scsi_ports) +
9843             num_target_ports * (sizeof(struct scsi_vpd_port_designation) +
9844              sizeof(struct scsi_vpd_port_designation_cont)) + iid_len + id_len;
9845         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9846         sp = (struct scsi_vpd_scsi_ports *)ctsio->kern_data_ptr;
9847         ctsio->kern_sg_entries = 0;
9848
9849         if (data_len < alloc_len) {
9850                 ctsio->residual = alloc_len - data_len;
9851                 ctsio->kern_data_len = data_len;
9852                 ctsio->kern_total_len = data_len;
9853         } else {
9854                 ctsio->residual = 0;
9855                 ctsio->kern_data_len = alloc_len;
9856                 ctsio->kern_total_len = alloc_len;
9857         }
9858         ctsio->kern_data_resid = 0;
9859         ctsio->kern_rel_offset = 0;
9860         ctsio->kern_sg_entries = 0;
9861
9862         /*
9863          * The control device is always connected.  The disk device, on the
9864          * other hand, may not be online all the time.  Need to change this
9865          * to figure out whether the disk device is actually online or not.
9866          */
9867         if (lun != NULL)
9868                 sp->device = (SID_QUAL_LU_CONNECTED << 5) |
9869                                   lun->be_lun->lun_type;
9870         else
9871                 sp->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9872
9873         sp->page_code = SVPD_SCSI_PORTS;
9874         scsi_ulto2b(data_len - sizeof(struct scsi_vpd_scsi_ports),
9875             sp->page_length);
9876         pd = &sp->design[0];
9877
9878         mtx_lock(&softc->ctl_lock);
9879         STAILQ_FOREACH(port, &softc->port_list, links) {
9880                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
9881                         continue;
9882                 if (lun != NULL &&
9883                     ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
9884                         continue;
9885                 scsi_ulto2b(port->targ_port, pd->relative_port_id);
9886                 if (port->init_devid) {
9887                         iid_len = port->init_devid->len;
9888                         memcpy(pd->initiator_transportid,
9889                             port->init_devid->data, port->init_devid->len);
9890                 } else
9891                         iid_len = 0;
9892                 scsi_ulto2b(iid_len, pd->initiator_transportid_length);
9893                 pdc = (struct scsi_vpd_port_designation_cont *)
9894                     (&pd->initiator_transportid[iid_len]);
9895                 if (port->port_devid) {
9896                         id_len = port->port_devid->len;
9897                         memcpy(pdc->target_port_descriptors,
9898                             port->port_devid->data, port->port_devid->len);
9899                 } else
9900                         id_len = 0;
9901                 scsi_ulto2b(id_len, pdc->target_port_descriptors_length);
9902                 pd = (struct scsi_vpd_port_designation *)
9903                     ((uint8_t *)pdc->target_port_descriptors + id_len);
9904         }
9905         mtx_unlock(&softc->ctl_lock);
9906
9907         ctl_set_success(ctsio);
9908         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9909         ctsio->be_move_done = ctl_config_move_done;
9910         ctl_datamove((union ctl_io *)ctsio);
9911         return (CTL_RETVAL_COMPLETE);
9912 }
9913
9914 static int
9915 ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio, int alloc_len)
9916 {
9917         struct scsi_vpd_block_limits *bl_ptr;
9918         struct ctl_lun *lun;
9919         uint64_t ival;
9920
9921         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9922
9923         ctsio->kern_data_ptr = malloc(sizeof(*bl_ptr), M_CTL, M_WAITOK | M_ZERO);
9924         bl_ptr = (struct scsi_vpd_block_limits *)ctsio->kern_data_ptr;
9925         ctsio->kern_sg_entries = 0;
9926
9927         if (sizeof(*bl_ptr) < alloc_len) {
9928                 ctsio->residual = alloc_len - sizeof(*bl_ptr);
9929                 ctsio->kern_data_len = sizeof(*bl_ptr);
9930                 ctsio->kern_total_len = sizeof(*bl_ptr);
9931         } else {
9932                 ctsio->residual = 0;
9933                 ctsio->kern_data_len = alloc_len;
9934                 ctsio->kern_total_len = alloc_len;
9935         }
9936         ctsio->kern_data_resid = 0;
9937         ctsio->kern_rel_offset = 0;
9938         ctsio->kern_sg_entries = 0;
9939
9940         /*
9941          * The control device is always connected.  The disk device, on the
9942          * other hand, may not be online all the time.  Need to change this
9943          * to figure out whether the disk device is actually online or not.
9944          */
9945         if (lun != NULL)
9946                 bl_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9947                                   lun->be_lun->lun_type;
9948         else
9949                 bl_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9950
9951         bl_ptr->page_code = SVPD_BLOCK_LIMITS;
9952         scsi_ulto2b(sizeof(*bl_ptr) - 4, bl_ptr->page_length);
9953         bl_ptr->max_cmp_write_len = 0xff;
9954         scsi_ulto4b(0xffffffff, bl_ptr->max_txfer_len);
9955         if (lun != NULL) {
9956                 scsi_ulto4b(lun->be_lun->opttxferlen, bl_ptr->opt_txfer_len);
9957                 if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
9958                         ival = 0xffffffff;
9959                         ctl_get_opt_number(&lun->be_lun->options,
9960                             "unmap_max_lba", &ival);
9961                         scsi_ulto4b(ival, bl_ptr->max_unmap_lba_cnt);
9962                         ival = 0xffffffff;
9963                         ctl_get_opt_number(&lun->be_lun->options,
9964                             "unmap_max_descr", &ival);
9965                         scsi_ulto4b(ival, bl_ptr->max_unmap_blk_cnt);
9966                         if (lun->be_lun->ublockexp != 0) {
9967                                 scsi_ulto4b((1 << lun->be_lun->ublockexp),
9968                                     bl_ptr->opt_unmap_grain);
9969                                 scsi_ulto4b(0x80000000 | lun->be_lun->ublockoff,
9970                                     bl_ptr->unmap_grain_align);
9971                         }
9972                 }
9973                 scsi_ulto4b(lun->be_lun->atomicblock,
9974                     bl_ptr->max_atomic_transfer_length);
9975                 scsi_ulto4b(0, bl_ptr->atomic_alignment);
9976                 scsi_ulto4b(0, bl_ptr->atomic_transfer_length_granularity);
9977                 scsi_ulto4b(0, bl_ptr->max_atomic_transfer_length_with_atomic_boundary);
9978                 scsi_ulto4b(0, bl_ptr->max_atomic_boundary_size);
9979                 ival = UINT64_MAX;
9980                 ctl_get_opt_number(&lun->be_lun->options, "write_same_max_lba", &ival);
9981                 scsi_u64to8b(ival, bl_ptr->max_write_same_length);
9982         }
9983
9984         ctl_set_success(ctsio);
9985         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9986         ctsio->be_move_done = ctl_config_move_done;
9987         ctl_datamove((union ctl_io *)ctsio);
9988         return (CTL_RETVAL_COMPLETE);
9989 }
9990
9991 static int
9992 ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len)
9993 {
9994         struct scsi_vpd_block_device_characteristics *bdc_ptr;
9995         struct ctl_lun *lun;
9996         const char *value;
9997         u_int i;
9998
9999         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10000
10001         ctsio->kern_data_ptr = malloc(sizeof(*bdc_ptr), M_CTL, M_WAITOK | M_ZERO);
10002         bdc_ptr = (struct scsi_vpd_block_device_characteristics *)ctsio->kern_data_ptr;
10003         ctsio->kern_sg_entries = 0;
10004
10005         if (sizeof(*bdc_ptr) < alloc_len) {
10006                 ctsio->residual = alloc_len - sizeof(*bdc_ptr);
10007                 ctsio->kern_data_len = sizeof(*bdc_ptr);
10008                 ctsio->kern_total_len = sizeof(*bdc_ptr);
10009         } else {
10010                 ctsio->residual = 0;
10011                 ctsio->kern_data_len = alloc_len;
10012                 ctsio->kern_total_len = alloc_len;
10013         }
10014         ctsio->kern_data_resid = 0;
10015         ctsio->kern_rel_offset = 0;
10016         ctsio->kern_sg_entries = 0;
10017
10018         /*
10019          * The control device is always connected.  The disk device, on the
10020          * other hand, may not be online all the time.  Need to change this
10021          * to figure out whether the disk device is actually online or not.
10022          */
10023         if (lun != NULL)
10024                 bdc_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10025                                   lun->be_lun->lun_type;
10026         else
10027                 bdc_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10028         bdc_ptr->page_code = SVPD_BDC;
10029         scsi_ulto2b(sizeof(*bdc_ptr) - 4, bdc_ptr->page_length);
10030         if (lun != NULL &&
10031             (value = ctl_get_opt(&lun->be_lun->options, "rpm")) != NULL)
10032                 i = strtol(value, NULL, 0);
10033         else
10034                 i = CTL_DEFAULT_ROTATION_RATE;
10035         scsi_ulto2b(i, bdc_ptr->medium_rotation_rate);
10036         if (lun != NULL &&
10037             (value = ctl_get_opt(&lun->be_lun->options, "formfactor")) != NULL)
10038                 i = strtol(value, NULL, 0);
10039         else
10040                 i = 0;
10041         bdc_ptr->wab_wac_ff = (i & 0x0f);
10042         bdc_ptr->flags = SVPD_FUAB | SVPD_VBULS;
10043
10044         ctl_set_success(ctsio);
10045         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10046         ctsio->be_move_done = ctl_config_move_done;
10047         ctl_datamove((union ctl_io *)ctsio);
10048         return (CTL_RETVAL_COMPLETE);
10049 }
10050
10051 static int
10052 ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len)
10053 {
10054         struct scsi_vpd_logical_block_prov *lbp_ptr;
10055         struct ctl_lun *lun;
10056         const char *value;
10057
10058         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10059
10060         ctsio->kern_data_ptr = malloc(sizeof(*lbp_ptr), M_CTL, M_WAITOK | M_ZERO);
10061         lbp_ptr = (struct scsi_vpd_logical_block_prov *)ctsio->kern_data_ptr;
10062         ctsio->kern_sg_entries = 0;
10063
10064         if (sizeof(*lbp_ptr) < alloc_len) {
10065                 ctsio->residual = alloc_len - sizeof(*lbp_ptr);
10066                 ctsio->kern_data_len = sizeof(*lbp_ptr);
10067                 ctsio->kern_total_len = sizeof(*lbp_ptr);
10068         } else {
10069                 ctsio->residual = 0;
10070                 ctsio->kern_data_len = alloc_len;
10071                 ctsio->kern_total_len = alloc_len;
10072         }
10073         ctsio->kern_data_resid = 0;
10074         ctsio->kern_rel_offset = 0;
10075         ctsio->kern_sg_entries = 0;
10076
10077         /*
10078          * The control device is always connected.  The disk device, on the
10079          * other hand, may not be online all the time.  Need to change this
10080          * to figure out whether the disk device is actually online or not.
10081          */
10082         if (lun != NULL)
10083                 lbp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10084                                   lun->be_lun->lun_type;
10085         else
10086                 lbp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10087
10088         lbp_ptr->page_code = SVPD_LBP;
10089         scsi_ulto2b(sizeof(*lbp_ptr) - 4, lbp_ptr->page_length);
10090         lbp_ptr->threshold_exponent = CTL_LBP_EXPONENT;
10091         if (lun != NULL && lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
10092                 lbp_ptr->flags = SVPD_LBP_UNMAP | SVPD_LBP_WS16 |
10093                     SVPD_LBP_WS10 | SVPD_LBP_RZ | SVPD_LBP_ANC_SUP;
10094                 value = ctl_get_opt(&lun->be_lun->options, "provisioning_type");
10095                 if (value != NULL) {
10096                         if (strcmp(value, "resource") == 0)
10097                                 lbp_ptr->prov_type = SVPD_LBP_RESOURCE;
10098                         else if (strcmp(value, "thin") == 0)
10099                                 lbp_ptr->prov_type = SVPD_LBP_THIN;
10100                 } else
10101                         lbp_ptr->prov_type = SVPD_LBP_THIN;
10102         }
10103
10104         ctl_set_success(ctsio);
10105         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10106         ctsio->be_move_done = ctl_config_move_done;
10107         ctl_datamove((union ctl_io *)ctsio);
10108         return (CTL_RETVAL_COMPLETE);
10109 }
10110
10111 /*
10112  * INQUIRY with the EVPD bit set.
10113  */
10114 static int
10115 ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
10116 {
10117         struct ctl_lun *lun;
10118         struct scsi_inquiry *cdb;
10119         int alloc_len, retval;
10120
10121         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10122         cdb = (struct scsi_inquiry *)ctsio->cdb;
10123         alloc_len = scsi_2btoul(cdb->length);
10124
10125         switch (cdb->page_code) {
10126         case SVPD_SUPPORTED_PAGES:
10127                 retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
10128                 break;
10129         case SVPD_UNIT_SERIAL_NUMBER:
10130                 retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
10131                 break;
10132         case SVPD_DEVICE_ID:
10133                 retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
10134                 break;
10135         case SVPD_EXTENDED_INQUIRY_DATA:
10136                 retval = ctl_inquiry_evpd_eid(ctsio, alloc_len);
10137                 break;
10138         case SVPD_MODE_PAGE_POLICY:
10139                 retval = ctl_inquiry_evpd_mpp(ctsio, alloc_len);
10140                 break;
10141         case SVPD_SCSI_PORTS:
10142                 retval = ctl_inquiry_evpd_scsi_ports(ctsio, alloc_len);
10143                 break;
10144         case SVPD_SCSI_TPC:
10145                 retval = ctl_inquiry_evpd_tpc(ctsio, alloc_len);
10146                 break;
10147         case SVPD_BLOCK_LIMITS:
10148                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
10149                         goto err;
10150                 retval = ctl_inquiry_evpd_block_limits(ctsio, alloc_len);
10151                 break;
10152         case SVPD_BDC:
10153                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
10154                         goto err;
10155                 retval = ctl_inquiry_evpd_bdc(ctsio, alloc_len);
10156                 break;
10157         case SVPD_LBP:
10158                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
10159                         goto err;
10160                 retval = ctl_inquiry_evpd_lbp(ctsio, alloc_len);
10161                 break;
10162         default:
10163 err:
10164                 ctl_set_invalid_field(ctsio,
10165                                       /*sks_valid*/ 1,
10166                                       /*command*/ 1,
10167                                       /*field*/ 2,
10168                                       /*bit_valid*/ 0,
10169                                       /*bit*/ 0);
10170                 ctl_done((union ctl_io *)ctsio);
10171                 retval = CTL_RETVAL_COMPLETE;
10172                 break;
10173         }
10174
10175         return (retval);
10176 }
10177
10178 /*
10179  * Standard INQUIRY data.
10180  */
10181 static int
10182 ctl_inquiry_std(struct ctl_scsiio *ctsio)
10183 {
10184         struct scsi_inquiry_data *inq_ptr;
10185         struct scsi_inquiry *cdb;
10186         struct ctl_softc *softc = control_softc;
10187         struct ctl_port *port;
10188         struct ctl_lun *lun;
10189         char *val;
10190         uint32_t alloc_len, data_len;
10191         ctl_port_type port_type;
10192
10193         port = ctl_io_port(&ctsio->io_hdr);
10194         port_type = port->port_type;
10195         if (port_type == CTL_PORT_IOCTL || port_type == CTL_PORT_INTERNAL)
10196                 port_type = CTL_PORT_SCSI;
10197
10198         lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10199         cdb = (struct scsi_inquiry *)ctsio->cdb;
10200         alloc_len = scsi_2btoul(cdb->length);
10201
10202         /*
10203          * We malloc the full inquiry data size here and fill it
10204          * in.  If the user only asks for less, we'll give him
10205          * that much.
10206          */
10207         data_len = offsetof(struct scsi_inquiry_data, vendor_specific1);
10208         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10209         inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
10210         ctsio->kern_sg_entries = 0;
10211         ctsio->kern_data_resid = 0;
10212         ctsio->kern_rel_offset = 0;
10213
10214         if (data_len < alloc_len) {
10215                 ctsio->residual = alloc_len - data_len;
10216                 ctsio->kern_data_len = data_len;
10217                 ctsio->kern_total_len = data_len;
10218         } else {
10219                 ctsio->residual = 0;
10220                 ctsio->kern_data_len = alloc_len;
10221                 ctsio->kern_total_len = alloc_len;
10222         }
10223
10224         if (lun != NULL) {
10225                 if ((lun->flags & CTL_LUN_PRIMARY_SC) ||
10226                     softc->ha_link >= CTL_HA_LINK_UNKNOWN) {
10227                         inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10228                             lun->be_lun->lun_type;
10229                 } else {
10230                         inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) |
10231                             lun->be_lun->lun_type;
10232                 }
10233                 if (lun->flags & CTL_LUN_REMOVABLE)
10234                         inq_ptr->dev_qual2 |= SID_RMB;
10235         } else
10236                 inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE;
10237
10238         /* RMB in byte 2 is 0 */
10239         inq_ptr->version = SCSI_REV_SPC4;
10240
10241         /*
10242          * According to SAM-3, even if a device only supports a single
10243          * level of LUN addressing, it should still set the HISUP bit:
10244          *
10245          * 4.9.1 Logical unit numbers overview
10246          *
10247          * All logical unit number formats described in this standard are
10248          * hierarchical in structure even when only a single level in that
10249          * hierarchy is used. The HISUP bit shall be set to one in the
10250          * standard INQUIRY data (see SPC-2) when any logical unit number
10251          * format described in this standard is used.  Non-hierarchical
10252          * formats are outside the scope of this standard.
10253          *
10254          * Therefore we set the HiSup bit here.
10255          *
10256          * The response format is 2, per SPC-3.
10257          */
10258         inq_ptr->response_format = SID_HiSup | 2;
10259
10260         inq_ptr->additional_length = data_len -
10261             (offsetof(struct scsi_inquiry_data, additional_length) + 1);
10262         CTL_DEBUG_PRINT(("additional_length = %d\n",
10263                          inq_ptr->additional_length));
10264
10265         inq_ptr->spc3_flags = SPC3_SID_3PC | SPC3_SID_TPGS_IMPLICIT;
10266         if (port_type == CTL_PORT_SCSI)
10267                 inq_ptr->spc2_flags = SPC2_SID_ADDR16;
10268         inq_ptr->spc2_flags |= SPC2_SID_MultiP;
10269         inq_ptr->flags = SID_CmdQue;
10270         if (port_type == CTL_PORT_SCSI)
10271                 inq_ptr->flags |= SID_WBus16 | SID_Sync;
10272
10273         /*
10274          * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
10275          * We have 8 bytes for the vendor name, and 16 bytes for the device
10276          * name and 4 bytes for the revision.
10277          */
10278         if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10279             "vendor")) == NULL) {
10280                 strncpy(inq_ptr->vendor, CTL_VENDOR, sizeof(inq_ptr->vendor));
10281         } else {
10282                 memset(inq_ptr->vendor, ' ', sizeof(inq_ptr->vendor));
10283                 strncpy(inq_ptr->vendor, val,
10284                     min(sizeof(inq_ptr->vendor), strlen(val)));
10285         }
10286         if (lun == NULL) {
10287                 strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10288                     sizeof(inq_ptr->product));
10289         } else if ((val = ctl_get_opt(&lun->be_lun->options, "product")) == NULL) {
10290                 switch (lun->be_lun->lun_type) {
10291                 case T_DIRECT:
10292                         strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10293                             sizeof(inq_ptr->product));
10294                         break;
10295                 case T_PROCESSOR:
10296                         strncpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT,
10297                             sizeof(inq_ptr->product));
10298                         break;
10299                 case T_CDROM:
10300                         strncpy(inq_ptr->product, CTL_CDROM_PRODUCT,
10301                             sizeof(inq_ptr->product));
10302                         break;
10303                 default:
10304                         strncpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT,
10305                             sizeof(inq_ptr->product));
10306                         break;
10307                 }
10308         } else {
10309                 memset(inq_ptr->product, ' ', sizeof(inq_ptr->product));
10310                 strncpy(inq_ptr->product, val,
10311                     min(sizeof(inq_ptr->product), strlen(val)));
10312         }
10313
10314         /*
10315          * XXX make this a macro somewhere so it automatically gets
10316          * incremented when we make changes.
10317          */
10318         if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10319             "revision")) == NULL) {
10320                 strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
10321         } else {
10322                 memset(inq_ptr->revision, ' ', sizeof(inq_ptr->revision));
10323                 strncpy(inq_ptr->revision, val,
10324                     min(sizeof(inq_ptr->revision), strlen(val)));
10325         }
10326
10327         /*
10328          * For parallel SCSI, we support double transition and single
10329          * transition clocking.  We also support QAS (Quick Arbitration
10330          * and Selection) and Information Unit transfers on both the
10331          * control and array devices.
10332          */
10333         if (port_type == CTL_PORT_SCSI)
10334                 inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
10335                                     SID_SPI_IUS;
10336
10337         /* SAM-5 (no version claimed) */
10338         scsi_ulto2b(0x00A0, inq_ptr->version1);
10339         /* SPC-4 (no version claimed) */
10340         scsi_ulto2b(0x0460, inq_ptr->version2);
10341         if (port_type == CTL_PORT_FC) {
10342                 /* FCP-2 ANSI INCITS.350:2003 */
10343                 scsi_ulto2b(0x0917, inq_ptr->version3);
10344         } else if (port_type == CTL_PORT_SCSI) {
10345                 /* SPI-4 ANSI INCITS.362:200x */
10346                 scsi_ulto2b(0x0B56, inq_ptr->version3);
10347         } else if (port_type == CTL_PORT_ISCSI) {
10348                 /* iSCSI (no version claimed) */
10349                 scsi_ulto2b(0x0960, inq_ptr->version3);
10350         } else if (port_type == CTL_PORT_SAS) {
10351                 /* SAS (no version claimed) */
10352                 scsi_ulto2b(0x0BE0, inq_ptr->version3);
10353         }
10354
10355         if (lun == NULL) {
10356                 /* SBC-4 (no version claimed) */
10357                 scsi_ulto2b(0x0600, inq_ptr->version4);
10358         } else {
10359                 switch (lun->be_lun->lun_type) {
10360                 case T_DIRECT:
10361                         /* SBC-4 (no version claimed) */
10362                         scsi_ulto2b(0x0600, inq_ptr->version4);
10363                         break;
10364                 case T_PROCESSOR:
10365                         break;
10366                 case T_CDROM:
10367                         /* MMC-6 (no version claimed) */
10368                         scsi_ulto2b(0x04E0, inq_ptr->version4);
10369                         break;
10370                 default:
10371                         break;
10372                 }
10373         }
10374
10375         ctl_set_success(ctsio);
10376         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10377         ctsio->be_move_done = ctl_config_move_done;
10378         ctl_datamove((union ctl_io *)ctsio);
10379         return (CTL_RETVAL_COMPLETE);
10380 }
10381
10382 int
10383 ctl_inquiry(struct ctl_scsiio *ctsio)
10384 {
10385         struct scsi_inquiry *cdb;
10386         int retval;
10387
10388         CTL_DEBUG_PRINT(("ctl_inquiry\n"));
10389
10390         cdb = (struct scsi_inquiry *)ctsio->cdb;
10391         if (cdb->byte2 & SI_EVPD)
10392                 retval = ctl_inquiry_evpd(ctsio);
10393         else if (cdb->page_code == 0)
10394                 retval = ctl_inquiry_std(ctsio);
10395         else {
10396                 ctl_set_invalid_field(ctsio,
10397                                       /*sks_valid*/ 1,
10398                                       /*command*/ 1,
10399                                       /*field*/ 2,
10400                                       /*bit_valid*/ 0,
10401                                       /*bit*/ 0);
10402                 ctl_done((union ctl_io *)ctsio);
10403                 return (CTL_RETVAL_COMPLETE);
10404         }
10405
10406         return (retval);
10407 }
10408
10409 int
10410 ctl_get_config(struct ctl_scsiio *ctsio)
10411 {
10412         struct scsi_get_config_header *hdr;
10413         struct scsi_get_config_feature *feature;
10414         struct scsi_get_config *cdb;
10415         struct ctl_lun *lun;
10416         uint32_t alloc_len, data_len;
10417         int rt, starting;
10418
10419         lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10420         cdb = (struct scsi_get_config *)ctsio->cdb;
10421         rt = (cdb->rt & SGC_RT_MASK);
10422         starting = scsi_2btoul(cdb->starting_feature);
10423         alloc_len = scsi_2btoul(cdb->length);
10424
10425         data_len = sizeof(struct scsi_get_config_header) +
10426             sizeof(struct scsi_get_config_feature) + 8 +
10427             sizeof(struct scsi_get_config_feature) + 8 +
10428             sizeof(struct scsi_get_config_feature) + 4 +
10429             sizeof(struct scsi_get_config_feature) + 4 +
10430             sizeof(struct scsi_get_config_feature) + 8 +
10431             sizeof(struct scsi_get_config_feature) +
10432             sizeof(struct scsi_get_config_feature) + 4 +
10433             sizeof(struct scsi_get_config_feature) + 4 +
10434             sizeof(struct scsi_get_config_feature) + 4 +
10435             sizeof(struct scsi_get_config_feature) + 4 +
10436             sizeof(struct scsi_get_config_feature) + 4 +
10437             sizeof(struct scsi_get_config_feature) + 4;
10438         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10439         ctsio->kern_sg_entries = 0;
10440         ctsio->kern_data_resid = 0;
10441         ctsio->kern_rel_offset = 0;
10442
10443         hdr = (struct scsi_get_config_header *)ctsio->kern_data_ptr;
10444         if (lun->flags & CTL_LUN_NO_MEDIA)
10445                 scsi_ulto2b(0x0000, hdr->current_profile);
10446         else
10447                 scsi_ulto2b(0x0010, hdr->current_profile);
10448         feature = (struct scsi_get_config_feature *)(hdr + 1);
10449
10450         if (starting > 0x003b)
10451                 goto done;
10452         if (starting > 0x003a)
10453                 goto f3b;
10454         if (starting > 0x002b)
10455                 goto f3a;
10456         if (starting > 0x002a)
10457                 goto f2b;
10458         if (starting > 0x001f)
10459                 goto f2a;
10460         if (starting > 0x001e)
10461                 goto f1f;
10462         if (starting > 0x001d)
10463                 goto f1e;
10464         if (starting > 0x0010)
10465                 goto f1d;
10466         if (starting > 0x0003)
10467                 goto f10;
10468         if (starting > 0x0002)
10469                 goto f3;
10470         if (starting > 0x0001)
10471                 goto f2;
10472         if (starting > 0x0000)
10473                 goto f1;
10474
10475         /* Profile List */
10476         scsi_ulto2b(0x0000, feature->feature_code);
10477         feature->flags = SGC_F_PERSISTENT | SGC_F_CURRENT;
10478         feature->add_length = 8;
10479         scsi_ulto2b(0x0008, &feature->feature_data[0]); /* CD-ROM */
10480         feature->feature_data[2] = 0x00;
10481         scsi_ulto2b(0x0010, &feature->feature_data[4]); /* DVD-ROM */
10482         feature->feature_data[6] = 0x01;
10483         feature = (struct scsi_get_config_feature *)
10484             &feature->feature_data[feature->add_length];
10485
10486 f1:     /* Core */
10487         scsi_ulto2b(0x0001, feature->feature_code);
10488         feature->flags = 0x08 | SGC_F_PERSISTENT | SGC_F_CURRENT;
10489         feature->add_length = 8;
10490         scsi_ulto4b(0x00000000, &feature->feature_data[0]);
10491         feature->feature_data[4] = 0x03;
10492         feature = (struct scsi_get_config_feature *)
10493             &feature->feature_data[feature->add_length];
10494
10495 f2:     /* Morphing */
10496         scsi_ulto2b(0x0002, feature->feature_code);
10497         feature->flags = 0x04 | SGC_F_PERSISTENT | SGC_F_CURRENT;
10498         feature->add_length = 4;
10499         feature->feature_data[0] = 0x02;
10500         feature = (struct scsi_get_config_feature *)
10501             &feature->feature_data[feature->add_length];
10502
10503 f3:     /* Removable Medium */
10504         scsi_ulto2b(0x0003, feature->feature_code);
10505         feature->flags = 0x04 | SGC_F_PERSISTENT | SGC_F_CURRENT;
10506         feature->add_length = 4;
10507         feature->feature_data[0] = 0x39;
10508         feature = (struct scsi_get_config_feature *)
10509             &feature->feature_data[feature->add_length];
10510
10511         if (rt == SGC_RT_CURRENT && (lun->flags & CTL_LUN_NO_MEDIA))
10512                 goto done;
10513
10514 f10:    /* Random Read */
10515         scsi_ulto2b(0x0010, feature->feature_code);
10516         feature->flags = 0x00;
10517         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10518                 feature->flags |= SGC_F_CURRENT;
10519         feature->add_length = 8;
10520         scsi_ulto4b(lun->be_lun->blocksize, &feature->feature_data[0]);
10521         scsi_ulto2b(1, &feature->feature_data[4]);
10522         feature->feature_data[6] = 0x00;
10523         feature = (struct scsi_get_config_feature *)
10524             &feature->feature_data[feature->add_length];
10525
10526 f1d:    /* Multi-Read */
10527         scsi_ulto2b(0x001D, feature->feature_code);
10528         feature->flags = 0x00;
10529         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10530                 feature->flags |= SGC_F_CURRENT;
10531         feature->add_length = 0;
10532         feature = (struct scsi_get_config_feature *)
10533             &feature->feature_data[feature->add_length];
10534
10535 f1e:    /* CD Read */
10536         scsi_ulto2b(0x001E, feature->feature_code);
10537         feature->flags = 0x00;
10538         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10539                 feature->flags |= SGC_F_CURRENT;
10540         feature->add_length = 4;
10541         feature->feature_data[0] = 0x00;
10542         feature = (struct scsi_get_config_feature *)
10543             &feature->feature_data[feature->add_length];
10544
10545 f1f:    /* DVD Read */
10546         scsi_ulto2b(0x001F, feature->feature_code);
10547         feature->flags = 0x08;
10548         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10549                 feature->flags |= SGC_F_CURRENT;
10550         feature->add_length = 4;
10551         feature->feature_data[0] = 0x01;
10552         feature->feature_data[2] = 0x03;
10553         feature = (struct scsi_get_config_feature *)
10554             &feature->feature_data[feature->add_length];
10555
10556 f2a:    /* DVD+RW */
10557         scsi_ulto2b(0x002A, feature->feature_code);
10558         feature->flags = 0x04;
10559         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10560                 feature->flags |= SGC_F_CURRENT;
10561         feature->add_length = 4;
10562         feature->feature_data[0] = 0x00;
10563         feature->feature_data[1] = 0x00;
10564         feature = (struct scsi_get_config_feature *)
10565             &feature->feature_data[feature->add_length];
10566
10567 f2b:    /* DVD+R */
10568         scsi_ulto2b(0x002B, feature->feature_code);
10569         feature->flags = 0x00;
10570         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10571                 feature->flags |= SGC_F_CURRENT;
10572         feature->add_length = 4;
10573         feature->feature_data[0] = 0x00;
10574         feature = (struct scsi_get_config_feature *)
10575             &feature->feature_data[feature->add_length];
10576
10577 f3a:    /* DVD+RW Dual Layer */
10578         scsi_ulto2b(0x003A, feature->feature_code);
10579         feature->flags = 0x00;
10580         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10581                 feature->flags |= SGC_F_CURRENT;
10582         feature->add_length = 4;
10583         feature->feature_data[0] = 0x00;
10584         feature->feature_data[1] = 0x00;
10585         feature = (struct scsi_get_config_feature *)
10586             &feature->feature_data[feature->add_length];
10587
10588 f3b:    /* DVD+R Dual Layer */
10589         scsi_ulto2b(0x003B, feature->feature_code);
10590         feature->flags = 0x00;
10591         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10592                 feature->flags |= SGC_F_CURRENT;
10593         feature->add_length = 4;
10594         feature->feature_data[0] = 0x00;
10595         feature = (struct scsi_get_config_feature *)
10596             &feature->feature_data[feature->add_length];
10597
10598 done:
10599         data_len = (uint8_t *)feature - (uint8_t *)hdr;
10600         if (rt == SGC_RT_SPECIFIC && data_len > 4) {
10601                 feature = (struct scsi_get_config_feature *)(hdr + 1);
10602                 if (scsi_2btoul(feature->feature_code) == starting)
10603                         feature = (struct scsi_get_config_feature *)
10604                             &feature->feature_data[feature->add_length];
10605                 data_len = (uint8_t *)feature - (uint8_t *)hdr;
10606         }
10607         scsi_ulto4b(data_len - 4, hdr->data_length);
10608         if (data_len < alloc_len) {
10609                 ctsio->residual = alloc_len - data_len;
10610                 ctsio->kern_data_len = data_len;
10611                 ctsio->kern_total_len = data_len;
10612         } else {
10613                 ctsio->residual = 0;
10614                 ctsio->kern_data_len = alloc_len;
10615                 ctsio->kern_total_len = alloc_len;
10616         }
10617
10618         ctl_set_success(ctsio);
10619         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10620         ctsio->be_move_done = ctl_config_move_done;
10621         ctl_datamove((union ctl_io *)ctsio);
10622         return (CTL_RETVAL_COMPLETE);
10623 }
10624
10625 int
10626 ctl_get_event_status(struct ctl_scsiio *ctsio)
10627 {
10628         struct scsi_get_event_status_header *hdr;
10629         struct scsi_get_event_status *cdb;
10630         struct ctl_lun *lun;
10631         uint32_t alloc_len, data_len;
10632         int notif_class;
10633
10634         lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10635         cdb = (struct scsi_get_event_status *)ctsio->cdb;
10636         if ((cdb->byte2 & SGESN_POLLED) == 0) {
10637                 ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 1,
10638                     /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 0);
10639                 ctl_done((union ctl_io *)ctsio);
10640                 return (CTL_RETVAL_COMPLETE);
10641         }
10642         notif_class = cdb->notif_class;
10643         alloc_len = scsi_2btoul(cdb->length);
10644
10645         data_len = sizeof(struct scsi_get_event_status_header);
10646         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10647         ctsio->kern_sg_entries = 0;
10648         ctsio->kern_data_resid = 0;
10649         ctsio->kern_rel_offset = 0;
10650
10651         if (data_len < alloc_len) {
10652                 ctsio->residual = alloc_len - data_len;
10653                 ctsio->kern_data_len = data_len;
10654                 ctsio->kern_total_len = data_len;
10655         } else {
10656                 ctsio->residual = 0;
10657                 ctsio->kern_data_len = alloc_len;
10658                 ctsio->kern_total_len = alloc_len;
10659         }
10660
10661         hdr = (struct scsi_get_event_status_header *)ctsio->kern_data_ptr;
10662         scsi_ulto2b(0, hdr->descr_length);
10663         hdr->nea_class = SGESN_NEA;
10664         hdr->supported_class = 0;
10665
10666         ctl_set_success(ctsio);
10667         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10668         ctsio->be_move_done = ctl_config_move_done;
10669         ctl_datamove((union ctl_io *)ctsio);
10670         return (CTL_RETVAL_COMPLETE);
10671 }
10672
10673 int
10674 ctl_mechanism_status(struct ctl_scsiio *ctsio)
10675 {
10676         struct scsi_mechanism_status_header *hdr;
10677         struct scsi_mechanism_status *cdb;
10678         struct ctl_lun *lun;
10679         uint32_t alloc_len, data_len;
10680
10681         lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10682         cdb = (struct scsi_mechanism_status *)ctsio->cdb;
10683         alloc_len = scsi_2btoul(cdb->length);
10684
10685         data_len = sizeof(struct scsi_mechanism_status_header);
10686         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10687         ctsio->kern_sg_entries = 0;
10688         ctsio->kern_data_resid = 0;
10689         ctsio->kern_rel_offset = 0;
10690
10691         if (data_len < alloc_len) {
10692                 ctsio->residual = alloc_len - data_len;
10693                 ctsio->kern_data_len = data_len;
10694                 ctsio->kern_total_len = data_len;
10695         } else {
10696                 ctsio->residual = 0;
10697                 ctsio->kern_data_len = alloc_len;
10698                 ctsio->kern_total_len = alloc_len;
10699         }
10700
10701         hdr = (struct scsi_mechanism_status_header *)ctsio->kern_data_ptr;
10702         hdr->state1 = 0x00;
10703         hdr->state2 = 0xe0;
10704         scsi_ulto3b(0, hdr->lba);
10705         hdr->slots_num = 0;
10706         scsi_ulto2b(0, hdr->slots_length);
10707
10708         ctl_set_success(ctsio);
10709         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10710         ctsio->be_move_done = ctl_config_move_done;
10711         ctl_datamove((union ctl_io *)ctsio);
10712         return (CTL_RETVAL_COMPLETE);
10713 }
10714
10715 static void
10716 ctl_ultomsf(uint32_t lba, uint8_t *buf)
10717 {
10718
10719         lba += 150;
10720         buf[0] = 0;
10721         buf[1] = bin2bcd((lba / 75) / 60);
10722         buf[2] = bin2bcd((lba / 75) % 60);
10723         buf[3] = bin2bcd(lba % 75);
10724 }
10725
10726 int
10727 ctl_read_toc(struct ctl_scsiio *ctsio)
10728 {
10729         struct scsi_read_toc_hdr *hdr;
10730         struct scsi_read_toc_type01_descr *descr;
10731         struct scsi_read_toc *cdb;
10732         struct ctl_lun *lun;
10733         uint32_t alloc_len, data_len;
10734         int format, msf;
10735
10736         lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10737         cdb = (struct scsi_read_toc *)ctsio->cdb;
10738         msf = (cdb->byte2 & CD_MSF) != 0;
10739         format = cdb->format;
10740         alloc_len = scsi_2btoul(cdb->data_len);
10741
10742         data_len = sizeof(struct scsi_read_toc_hdr);
10743         if (format == 0)
10744                 data_len += 2 * sizeof(struct scsi_read_toc_type01_descr);
10745         else
10746                 data_len += sizeof(struct scsi_read_toc_type01_descr);
10747         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10748         ctsio->kern_sg_entries = 0;
10749         ctsio->kern_data_resid = 0;
10750         ctsio->kern_rel_offset = 0;
10751
10752         if (data_len < alloc_len) {
10753                 ctsio->residual = alloc_len - data_len;
10754                 ctsio->kern_data_len = data_len;
10755                 ctsio->kern_total_len = data_len;
10756         } else {
10757                 ctsio->residual = 0;
10758                 ctsio->kern_data_len = alloc_len;
10759                 ctsio->kern_total_len = alloc_len;
10760         }
10761
10762         hdr = (struct scsi_read_toc_hdr *)ctsio->kern_data_ptr;
10763         if (format == 0) {
10764                 scsi_ulto2b(0x12, hdr->data_length);
10765                 hdr->first = 1;
10766                 hdr->last = 1;
10767                 descr = (struct scsi_read_toc_type01_descr *)(hdr + 1);
10768                 descr->addr_ctl = 0x14;
10769                 descr->track_number = 1;
10770                 if (msf)
10771                         ctl_ultomsf(0, descr->track_start);
10772                 else
10773                         scsi_ulto4b(0, descr->track_start);
10774                 descr++;
10775                 descr->addr_ctl = 0x14;
10776                 descr->track_number = 0xaa;
10777                 if (msf)
10778                         ctl_ultomsf(lun->be_lun->maxlba+1, descr->track_start);
10779                 else
10780                         scsi_ulto4b(lun->be_lun->maxlba+1, descr->track_start);
10781         } else {
10782                 scsi_ulto2b(0x0a, hdr->data_length);
10783                 hdr->first = 1;
10784                 hdr->last = 1;
10785                 descr = (struct scsi_read_toc_type01_descr *)(hdr + 1);
10786                 descr->addr_ctl = 0x14;
10787                 descr->track_number = 1;
10788                 if (msf)
10789                         ctl_ultomsf(0, descr->track_start);
10790                 else
10791                         scsi_ulto4b(0, descr->track_start);
10792         }
10793
10794         ctl_set_success(ctsio);
10795         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10796         ctsio->be_move_done = ctl_config_move_done;
10797         ctl_datamove((union ctl_io *)ctsio);
10798         return (CTL_RETVAL_COMPLETE);
10799 }
10800
10801 /*
10802  * For known CDB types, parse the LBA and length.
10803  */
10804 static int
10805 ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len)
10806 {
10807         if (io->io_hdr.io_type != CTL_IO_SCSI)
10808                 return (1);
10809
10810         switch (io->scsiio.cdb[0]) {
10811         case COMPARE_AND_WRITE: {
10812                 struct scsi_compare_and_write *cdb;
10813
10814                 cdb = (struct scsi_compare_and_write *)io->scsiio.cdb;
10815
10816                 *lba = scsi_8btou64(cdb->addr);
10817                 *len = cdb->length;
10818                 break;
10819         }
10820         case READ_6:
10821         case WRITE_6: {
10822                 struct scsi_rw_6 *cdb;
10823
10824                 cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
10825
10826                 *lba = scsi_3btoul(cdb->addr);
10827                 /* only 5 bits are valid in the most significant address byte */
10828                 *lba &= 0x1fffff;
10829                 *len = cdb->length;
10830                 break;
10831         }
10832         case READ_10:
10833         case WRITE_10: {
10834                 struct scsi_rw_10 *cdb;
10835
10836                 cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
10837
10838                 *lba = scsi_4btoul(cdb->addr);
10839                 *len = scsi_2btoul(cdb->length);
10840                 break;
10841         }
10842         case WRITE_VERIFY_10: {
10843                 struct scsi_write_verify_10 *cdb;
10844
10845                 cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
10846
10847                 *lba = scsi_4btoul(cdb->addr);
10848                 *len = scsi_2btoul(cdb->length);
10849                 break;
10850         }
10851         case READ_12:
10852         case WRITE_12: {
10853                 struct scsi_rw_12 *cdb;
10854
10855                 cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
10856
10857                 *lba = scsi_4btoul(cdb->addr);
10858                 *len = scsi_4btoul(cdb->length);
10859                 break;
10860         }
10861         case WRITE_VERIFY_12: {
10862                 struct scsi_write_verify_12 *cdb;
10863
10864                 cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
10865
10866                 *lba = scsi_4btoul(cdb->addr);
10867                 *len = scsi_4btoul(cdb->length);
10868                 break;
10869         }
10870         case READ_16:
10871         case WRITE_16: {
10872                 struct scsi_rw_16 *cdb;
10873
10874                 cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
10875
10876                 *lba = scsi_8btou64(cdb->addr);
10877                 *len = scsi_4btoul(cdb->length);
10878                 break;
10879         }
10880         case WRITE_ATOMIC_16: {
10881                 struct scsi_write_atomic_16 *cdb;
10882
10883                 cdb = (struct scsi_write_atomic_16 *)io->scsiio.cdb;
10884
10885                 *lba = scsi_8btou64(cdb->addr);
10886                 *len = scsi_2btoul(cdb->length);
10887                 break;
10888         }
10889         case WRITE_VERIFY_16: {
10890                 struct scsi_write_verify_16 *cdb;
10891
10892                 cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
10893
10894                 *lba = scsi_8btou64(cdb->addr);
10895                 *len = scsi_4btoul(cdb->length);
10896                 break;
10897         }
10898         case WRITE_SAME_10: {
10899                 struct scsi_write_same_10 *cdb;
10900
10901                 cdb = (struct scsi_write_same_10 *)io->scsiio.cdb;
10902
10903                 *lba = scsi_4btoul(cdb->addr);
10904                 *len = scsi_2btoul(cdb->length);
10905                 break;
10906         }
10907         case WRITE_SAME_16: {
10908                 struct scsi_write_same_16 *cdb;
10909
10910                 cdb = (struct scsi_write_same_16 *)io->scsiio.cdb;
10911
10912                 *lba = scsi_8btou64(cdb->addr);
10913                 *len = scsi_4btoul(cdb->length);
10914                 break;
10915         }
10916         case VERIFY_10: {
10917                 struct scsi_verify_10 *cdb;
10918
10919                 cdb = (struct scsi_verify_10 *)io->scsiio.cdb;
10920
10921                 *lba = scsi_4btoul(cdb->addr);
10922                 *len = scsi_2btoul(cdb->length);
10923                 break;
10924         }
10925         case VERIFY_12: {
10926                 struct scsi_verify_12 *cdb;
10927
10928                 cdb = (struct scsi_verify_12 *)io->scsiio.cdb;
10929
10930                 *lba = scsi_4btoul(cdb->addr);
10931                 *len = scsi_4btoul(cdb->length);
10932                 break;
10933         }
10934         case VERIFY_16: {
10935                 struct scsi_verify_16 *cdb;
10936
10937                 cdb = (struct scsi_verify_16 *)io->scsiio.cdb;
10938
10939                 *lba = scsi_8btou64(cdb->addr);
10940                 *len = scsi_4btoul(cdb->length);
10941                 break;
10942         }
10943         case UNMAP: {
10944                 *lba = 0;
10945                 *len = UINT64_MAX;
10946                 break;
10947         }
10948         case SERVICE_ACTION_IN: {       /* GET LBA STATUS */
10949                 struct scsi_get_lba_status *cdb;
10950
10951                 cdb = (struct scsi_get_lba_status *)io->scsiio.cdb;
10952                 *lba = scsi_8btou64(cdb->addr);
10953                 *len = UINT32_MAX;
10954                 break;
10955         }
10956         default:
10957                 return (1);
10958                 break; /* NOTREACHED */
10959         }
10960
10961         return (0);
10962 }
10963
10964 static ctl_action
10965 ctl_extent_check_lba(uint64_t lba1, uint64_t len1, uint64_t lba2, uint64_t len2,
10966     bool seq)
10967 {
10968         uint64_t endlba1, endlba2;
10969
10970         endlba1 = lba1 + len1 - (seq ? 0 : 1);
10971         endlba2 = lba2 + len2 - 1;
10972
10973         if ((endlba1 < lba2) || (endlba2 < lba1))
10974                 return (CTL_ACTION_PASS);
10975         else
10976                 return (CTL_ACTION_BLOCK);
10977 }
10978
10979 static int
10980 ctl_extent_check_unmap(union ctl_io *io, uint64_t lba2, uint64_t len2)
10981 {
10982         struct ctl_ptr_len_flags *ptrlen;
10983         struct scsi_unmap_desc *buf, *end, *range;
10984         uint64_t lba;
10985         uint32_t len;
10986
10987         /* If not UNMAP -- go other way. */
10988         if (io->io_hdr.io_type != CTL_IO_SCSI ||
10989             io->scsiio.cdb[0] != UNMAP)
10990                 return (CTL_ACTION_ERROR);
10991
10992         /* If UNMAP without data -- block and wait for data. */
10993         ptrlen = (struct ctl_ptr_len_flags *)
10994             &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
10995         if ((io->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0 ||
10996             ptrlen->ptr == NULL)
10997                 return (CTL_ACTION_BLOCK);
10998
10999         /* UNMAP with data -- check for collision. */
11000         buf = (struct scsi_unmap_desc *)ptrlen->ptr;
11001         end = buf + ptrlen->len / sizeof(*buf);
11002         for (range = buf; range < end; range++) {
11003                 lba = scsi_8btou64(range->lba);
11004                 len = scsi_4btoul(range->length);
11005                 if ((lba < lba2 + len2) && (lba + len > lba2))
11006                         return (CTL_ACTION_BLOCK);
11007         }
11008         return (CTL_ACTION_PASS);
11009 }
11010
11011 static ctl_action
11012 ctl_extent_check(union ctl_io *io1, union ctl_io *io2, bool seq)
11013 {
11014         uint64_t lba1, lba2;
11015         uint64_t len1, len2;
11016         int retval;
11017
11018         if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
11019                 return (CTL_ACTION_ERROR);
11020
11021         retval = ctl_extent_check_unmap(io1, lba2, len2);
11022         if (retval != CTL_ACTION_ERROR)
11023                 return (retval);
11024
11025         if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
11026                 return (CTL_ACTION_ERROR);
11027
11028         if (io1->io_hdr.flags & CTL_FLAG_SERSEQ_DONE)
11029                 seq = FALSE;
11030         return (ctl_extent_check_lba(lba1, len1, lba2, len2, seq));
11031 }
11032
11033 static ctl_action
11034 ctl_extent_check_seq(union ctl_io *io1, union ctl_io *io2)
11035 {
11036         uint64_t lba1, lba2;
11037         uint64_t len1, len2;
11038
11039         if (io1->io_hdr.flags & CTL_FLAG_SERSEQ_DONE)
11040                 return (CTL_ACTION_PASS);
11041         if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
11042                 return (CTL_ACTION_ERROR);
11043         if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
11044                 return (CTL_ACTION_ERROR);
11045
11046         if (lba1 + len1 == lba2)
11047                 return (CTL_ACTION_BLOCK);
11048         return (CTL_ACTION_PASS);
11049 }
11050
11051 static ctl_action
11052 ctl_check_for_blockage(struct ctl_lun *lun, union ctl_io *pending_io,
11053     union ctl_io *ooa_io)
11054 {
11055         const struct ctl_cmd_entry *pending_entry, *ooa_entry;
11056         const ctl_serialize_action *serialize_row;
11057
11058         /*
11059          * The initiator attempted multiple untagged commands at the same
11060          * time.  Can't do that.
11061          */
11062         if ((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
11063          && (ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
11064          && ((pending_io->io_hdr.nexus.targ_port ==
11065               ooa_io->io_hdr.nexus.targ_port)
11066           && (pending_io->io_hdr.nexus.initid ==
11067               ooa_io->io_hdr.nexus.initid))
11068          && ((ooa_io->io_hdr.flags & (CTL_FLAG_ABORT |
11069               CTL_FLAG_STATUS_SENT)) == 0))
11070                 return (CTL_ACTION_OVERLAP);
11071
11072         /*
11073          * The initiator attempted to send multiple tagged commands with
11074          * the same ID.  (It's fine if different initiators have the same
11075          * tag ID.)
11076          *
11077          * Even if all of those conditions are true, we don't kill the I/O
11078          * if the command ahead of us has been aborted.  We won't end up
11079          * sending it to the FETD, and it's perfectly legal to resend a
11080          * command with the same tag number as long as the previous
11081          * instance of this tag number has been aborted somehow.
11082          */
11083         if ((pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
11084          && (ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
11085          && (pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
11086          && ((pending_io->io_hdr.nexus.targ_port ==
11087               ooa_io->io_hdr.nexus.targ_port)
11088           && (pending_io->io_hdr.nexus.initid ==
11089               ooa_io->io_hdr.nexus.initid))
11090          && ((ooa_io->io_hdr.flags & (CTL_FLAG_ABORT |
11091               CTL_FLAG_STATUS_SENT)) == 0))
11092                 return (CTL_ACTION_OVERLAP_TAG);
11093
11094         /*
11095          * If we get a head of queue tag, SAM-3 says that we should
11096          * immediately execute it.
11097          *
11098          * What happens if this command would normally block for some other
11099          * reason?  e.g. a request sense with a head of queue tag
11100          * immediately after a write.  Normally that would block, but this
11101          * will result in its getting executed immediately...
11102          *
11103          * We currently return "pass" instead of "skip", so we'll end up
11104          * going through the rest of the queue to check for overlapped tags.
11105          *
11106          * XXX KDM check for other types of blockage first??
11107          */
11108         if (pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
11109                 return (CTL_ACTION_PASS);
11110
11111         /*
11112          * Ordered tags have to block until all items ahead of them
11113          * have completed.  If we get called with an ordered tag, we always
11114          * block, if something else is ahead of us in the queue.
11115          */
11116         if (pending_io->scsiio.tag_type == CTL_TAG_ORDERED)
11117                 return (CTL_ACTION_BLOCK);
11118
11119         /*
11120          * Simple tags get blocked until all head of queue and ordered tags
11121          * ahead of them have completed.  I'm lumping untagged commands in
11122          * with simple tags here.  XXX KDM is that the right thing to do?
11123          */
11124         if (((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
11125           || (pending_io->scsiio.tag_type == CTL_TAG_SIMPLE))
11126          && ((ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
11127           || (ooa_io->scsiio.tag_type == CTL_TAG_ORDERED)))
11128                 return (CTL_ACTION_BLOCK);
11129
11130         pending_entry = ctl_get_cmd_entry(&pending_io->scsiio, NULL);
11131         KASSERT(pending_entry->seridx < CTL_SERIDX_COUNT,
11132             ("%s: Invalid seridx %d for pending CDB %02x %02x @ %p",
11133              __func__, pending_entry->seridx, pending_io->scsiio.cdb[0],
11134              pending_io->scsiio.cdb[1], pending_io));
11135         ooa_entry = ctl_get_cmd_entry(&ooa_io->scsiio, NULL);
11136         if (ooa_entry->seridx == CTL_SERIDX_INVLD)
11137                 return (CTL_ACTION_PASS); /* Unsupported command in OOA queue */
11138         KASSERT(ooa_entry->seridx < CTL_SERIDX_COUNT,
11139             ("%s: Invalid seridx %d for ooa CDB %02x %02x @ %p",
11140              __func__, ooa_entry->seridx, ooa_io->scsiio.cdb[0],
11141              ooa_io->scsiio.cdb[1], ooa_io));
11142
11143         serialize_row = ctl_serialize_table[ooa_entry->seridx];
11144
11145         switch (serialize_row[pending_entry->seridx]) {
11146         case CTL_SER_BLOCK:
11147                 return (CTL_ACTION_BLOCK);
11148         case CTL_SER_EXTENT:
11149                 return (ctl_extent_check(ooa_io, pending_io,
11150                     (lun->be_lun && lun->be_lun->serseq == CTL_LUN_SERSEQ_ON)));
11151         case CTL_SER_EXTENTOPT:
11152                 if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT].queue_flags
11153                     & SCP_QUEUE_ALG_MASK) != SCP_QUEUE_ALG_UNRESTRICTED)
11154                         return (ctl_extent_check(ooa_io, pending_io,
11155                             (lun->be_lun &&
11156                              lun->be_lun->serseq == CTL_LUN_SERSEQ_ON)));
11157                 return (CTL_ACTION_PASS);
11158         case CTL_SER_EXTENTSEQ:
11159                 if (lun->be_lun && lun->be_lun->serseq != CTL_LUN_SERSEQ_OFF)
11160                         return (ctl_extent_check_seq(ooa_io, pending_io));
11161                 return (CTL_ACTION_PASS);
11162         case CTL_SER_PASS:
11163                 return (CTL_ACTION_PASS);
11164         case CTL_SER_BLOCKOPT:
11165                 if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT].queue_flags
11166                     & SCP_QUEUE_ALG_MASK) != SCP_QUEUE_ALG_UNRESTRICTED)
11167                         return (CTL_ACTION_BLOCK);
11168                 return (CTL_ACTION_PASS);
11169         case CTL_SER_SKIP:
11170                 return (CTL_ACTION_SKIP);
11171         default:
11172                 panic("%s: Invalid serialization value %d for %d => %d",
11173                     __func__, serialize_row[pending_entry->seridx],
11174                     pending_entry->seridx, ooa_entry->seridx);
11175         }
11176
11177         return (CTL_ACTION_ERROR);
11178 }
11179
11180 /*
11181  * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
11182  * Assumptions:
11183  * - pending_io is generally either incoming, or on the blocked queue
11184  * - starting I/O is the I/O we want to start the check with.
11185  */
11186 static ctl_action
11187 ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
11188               union ctl_io *starting_io)
11189 {
11190         union ctl_io *ooa_io;
11191         ctl_action action;
11192
11193         mtx_assert(&lun->lun_lock, MA_OWNED);
11194
11195         /*
11196          * Run back along the OOA queue, starting with the current
11197          * blocked I/O and going through every I/O before it on the
11198          * queue.  If starting_io is NULL, we'll just end up returning
11199          * CTL_ACTION_PASS.
11200          */
11201         for (ooa_io = starting_io; ooa_io != NULL;
11202              ooa_io = (union ctl_io *)TAILQ_PREV(&ooa_io->io_hdr, ctl_ooaq,
11203              ooa_links)){
11204
11205                 /*
11206                  * This routine just checks to see whether
11207                  * cur_blocked is blocked by ooa_io, which is ahead
11208                  * of it in the queue.  It doesn't queue/dequeue
11209                  * cur_blocked.
11210                  */
11211                 action = ctl_check_for_blockage(lun, pending_io, ooa_io);
11212                 switch (action) {
11213                 case CTL_ACTION_BLOCK:
11214                 case CTL_ACTION_OVERLAP:
11215                 case CTL_ACTION_OVERLAP_TAG:
11216                 case CTL_ACTION_SKIP:
11217                 case CTL_ACTION_ERROR:
11218                         return (action);
11219                         break; /* NOTREACHED */
11220                 case CTL_ACTION_PASS:
11221                         break;
11222                 default:
11223                         panic("%s: Invalid action %d\n", __func__, action);
11224                 }
11225         }
11226
11227         return (CTL_ACTION_PASS);
11228 }
11229
11230 /*
11231  * Assumptions:
11232  * - An I/O has just completed, and has been removed from the per-LUN OOA
11233  *   queue, so some items on the blocked queue may now be unblocked.
11234  */
11235 static int
11236 ctl_check_blocked(struct ctl_lun *lun)
11237 {
11238         struct ctl_softc *softc = lun->ctl_softc;
11239         union ctl_io *cur_blocked, *next_blocked;
11240
11241         mtx_assert(&lun->lun_lock, MA_OWNED);
11242
11243         /*
11244          * Run forward from the head of the blocked queue, checking each
11245          * entry against the I/Os prior to it on the OOA queue to see if
11246          * there is still any blockage.
11247          *
11248          * We cannot use the TAILQ_FOREACH() macro, because it can't deal
11249          * with our removing a variable on it while it is traversing the
11250          * list.
11251          */
11252         for (cur_blocked = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue);
11253              cur_blocked != NULL; cur_blocked = next_blocked) {
11254                 union ctl_io *prev_ooa;
11255                 ctl_action action;
11256
11257                 next_blocked = (union ctl_io *)TAILQ_NEXT(&cur_blocked->io_hdr,
11258                                                           blocked_links);
11259
11260                 prev_ooa = (union ctl_io *)TAILQ_PREV(&cur_blocked->io_hdr,
11261                                                       ctl_ooaq, ooa_links);
11262
11263                 /*
11264                  * If cur_blocked happens to be the first item in the OOA
11265                  * queue now, prev_ooa will be NULL, and the action
11266                  * returned will just be CTL_ACTION_PASS.
11267                  */
11268                 action = ctl_check_ooa(lun, cur_blocked, prev_ooa);
11269
11270                 switch (action) {
11271                 case CTL_ACTION_BLOCK:
11272                         /* Nothing to do here, still blocked */
11273                         break;
11274                 case CTL_ACTION_OVERLAP:
11275                 case CTL_ACTION_OVERLAP_TAG:
11276                         /*
11277                          * This shouldn't happen!  In theory we've already
11278                          * checked this command for overlap...
11279                          */
11280                         break;
11281                 case CTL_ACTION_PASS:
11282                 case CTL_ACTION_SKIP: {
11283                         const struct ctl_cmd_entry *entry;
11284
11285                         /*
11286                          * The skip case shouldn't happen, this transaction
11287                          * should have never made it onto the blocked queue.
11288                          */
11289                         /*
11290                          * This I/O is no longer blocked, we can remove it
11291                          * from the blocked queue.  Since this is a TAILQ
11292                          * (doubly linked list), we can do O(1) removals
11293                          * from any place on the list.
11294                          */
11295                         TAILQ_REMOVE(&lun->blocked_queue, &cur_blocked->io_hdr,
11296                                      blocked_links);
11297                         cur_blocked->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
11298
11299                         if ((softc->ha_mode != CTL_HA_MODE_XFER) &&
11300                             (cur_blocked->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)){
11301                                 /*
11302                                  * Need to send IO back to original side to
11303                                  * run
11304                                  */
11305                                 union ctl_ha_msg msg_info;
11306
11307                                 cur_blocked->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11308                                 msg_info.hdr.original_sc =
11309                                         cur_blocked->io_hdr.original_sc;
11310                                 msg_info.hdr.serializing_sc = cur_blocked;
11311                                 msg_info.hdr.msg_type = CTL_MSG_R2R;
11312                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11313                                     sizeof(msg_info.hdr), M_NOWAIT);
11314                                 break;
11315                         }
11316                         entry = ctl_get_cmd_entry(&cur_blocked->scsiio, NULL);
11317
11318                         /*
11319                          * Check this I/O for LUN state changes that may
11320                          * have happened while this command was blocked.
11321                          * The LUN state may have been changed by a command
11322                          * ahead of us in the queue, so we need to re-check
11323                          * for any states that can be caused by SCSI
11324                          * commands.
11325                          */
11326                         if (ctl_scsiio_lun_check(lun, entry,
11327                                                  &cur_blocked->scsiio) == 0) {
11328                                 cur_blocked->io_hdr.flags |=
11329                                                       CTL_FLAG_IS_WAS_ON_RTR;
11330                                 ctl_enqueue_rtr(cur_blocked);
11331                         } else
11332                                 ctl_done(cur_blocked);
11333                         break;
11334                 }
11335                 default:
11336                         /*
11337                          * This probably shouldn't happen -- we shouldn't
11338                          * get CTL_ACTION_ERROR, or anything else.
11339                          */
11340                         break;
11341                 }
11342         }
11343
11344         return (CTL_RETVAL_COMPLETE);
11345 }
11346
11347 /*
11348  * This routine (with one exception) checks LUN flags that can be set by
11349  * commands ahead of us in the OOA queue.  These flags have to be checked
11350  * when a command initially comes in, and when we pull a command off the
11351  * blocked queue and are preparing to execute it.  The reason we have to
11352  * check these flags for commands on the blocked queue is that the LUN
11353  * state may have been changed by a command ahead of us while we're on the
11354  * blocked queue.
11355  *
11356  * Ordering is somewhat important with these checks, so please pay
11357  * careful attention to the placement of any new checks.
11358  */
11359 static int
11360 ctl_scsiio_lun_check(struct ctl_lun *lun,
11361     const struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
11362 {
11363         struct ctl_softc *softc = lun->ctl_softc;
11364         int retval;
11365         uint32_t residx;
11366
11367         retval = 0;
11368
11369         mtx_assert(&lun->lun_lock, MA_OWNED);
11370
11371         /*
11372          * If this shelf is a secondary shelf controller, we may have to
11373          * reject some commands disallowed by HA mode and link state.
11374          */
11375         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
11376                 if (softc->ha_link == CTL_HA_LINK_OFFLINE &&
11377                     (entry->flags & CTL_CMD_FLAG_OK_ON_UNAVAIL) == 0) {
11378                         ctl_set_lun_unavail(ctsio);
11379                         retval = 1;
11380                         goto bailout;
11381                 }
11382                 if ((lun->flags & CTL_LUN_PEER_SC_PRIMARY) == 0 &&
11383                     (entry->flags & CTL_CMD_FLAG_OK_ON_UNAVAIL) == 0) {
11384                         ctl_set_lun_transit(ctsio);
11385                         retval = 1;
11386                         goto bailout;
11387                 }
11388                 if (softc->ha_mode == CTL_HA_MODE_ACT_STBY &&
11389                     (entry->flags & CTL_CMD_FLAG_OK_ON_STANDBY) == 0) {
11390                         ctl_set_lun_standby(ctsio);
11391                         retval = 1;
11392                         goto bailout;
11393                 }
11394
11395                 /* The rest of checks are only done on executing side */
11396                 if (softc->ha_mode == CTL_HA_MODE_XFER)
11397                         goto bailout;
11398         }
11399
11400         if (entry->pattern & CTL_LUN_PAT_WRITE) {
11401                 if (lun->be_lun &&
11402                     lun->be_lun->flags & CTL_LUN_FLAG_READONLY) {
11403                         ctl_set_hw_write_protected(ctsio);
11404                         retval = 1;
11405                         goto bailout;
11406                 }
11407                 if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT]
11408                     .eca_and_aen & SCP_SWP) != 0) {
11409                         ctl_set_sense(ctsio, /*current_error*/ 1,
11410                             /*sense_key*/ SSD_KEY_DATA_PROTECT,
11411                             /*asc*/ 0x27, /*ascq*/ 0x02, SSD_ELEM_NONE);
11412                         retval = 1;
11413                         goto bailout;
11414                 }
11415         }
11416
11417         /*
11418          * Check for a reservation conflict.  If this command isn't allowed
11419          * even on reserved LUNs, and if this initiator isn't the one who
11420          * reserved us, reject the command with a reservation conflict.
11421          */
11422         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11423         if ((lun->flags & CTL_LUN_RESERVED)
11424          && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
11425                 if (lun->res_idx != residx) {
11426                         ctl_set_reservation_conflict(ctsio);
11427                         retval = 1;
11428                         goto bailout;
11429                 }
11430         }
11431
11432         if ((lun->flags & CTL_LUN_PR_RESERVED) == 0 ||
11433             (entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV)) {
11434                 /* No reservation or command is allowed. */;
11435         } else if ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_WRESV) &&
11436             (lun->pr_res_type == SPR_TYPE_WR_EX ||
11437              lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
11438              lun->pr_res_type == SPR_TYPE_WR_EX_AR)) {
11439                 /* The command is allowed for Write Exclusive resv. */;
11440         } else {
11441                 /*
11442                  * if we aren't registered or it's a res holder type
11443                  * reservation and this isn't the res holder then set a
11444                  * conflict.
11445                  */
11446                 if (ctl_get_prkey(lun, residx) == 0 ||
11447                     (residx != lun->pr_res_idx && lun->pr_res_type < 4)) {
11448                         ctl_set_reservation_conflict(ctsio);
11449                         retval = 1;
11450                         goto bailout;
11451                 }
11452         }
11453
11454         if ((entry->flags & CTL_CMD_FLAG_OK_ON_NO_MEDIA) == 0) {
11455                 if (lun->flags & CTL_LUN_EJECTED)
11456                         ctl_set_lun_ejected(ctsio);
11457                 else if (lun->flags & CTL_LUN_NO_MEDIA) {
11458                         if (lun->flags & CTL_LUN_REMOVABLE)
11459                                 ctl_set_lun_no_media(ctsio);
11460                         else
11461                                 ctl_set_lun_int_reqd(ctsio);
11462                 } else if (lun->flags & CTL_LUN_STOPPED)
11463                         ctl_set_lun_stopped(ctsio);
11464                 else
11465                         goto bailout;
11466                 retval = 1;
11467                 goto bailout;
11468         }
11469
11470 bailout:
11471         return (retval);
11472 }
11473
11474 static void
11475 ctl_failover_io(union ctl_io *io, int have_lock)
11476 {
11477         ctl_set_busy(&io->scsiio);
11478         ctl_done(io);
11479 }
11480
11481 static void
11482 ctl_failover_lun(union ctl_io *rio)
11483 {
11484         struct ctl_softc *softc = control_softc;
11485         struct ctl_lun *lun;
11486         struct ctl_io_hdr *io, *next_io;
11487         uint32_t targ_lun;
11488
11489         targ_lun = rio->io_hdr.nexus.targ_mapped_lun;
11490         CTL_DEBUG_PRINT(("FAILOVER for lun %ju\n", targ_lun));
11491
11492         /* Find and lock the LUN. */
11493         mtx_lock(&softc->ctl_lock);
11494         if ((targ_lun < CTL_MAX_LUNS) &&
11495             ((lun = softc->ctl_luns[targ_lun]) != NULL)) {
11496                 mtx_lock(&lun->lun_lock);
11497                 mtx_unlock(&softc->ctl_lock);
11498                 if (lun->flags & CTL_LUN_DISABLED) {
11499                         mtx_unlock(&lun->lun_lock);
11500                         return;
11501                 }
11502         } else {
11503                 mtx_unlock(&softc->ctl_lock);
11504                 return;
11505         }
11506
11507         if (softc->ha_mode == CTL_HA_MODE_XFER) {
11508                 TAILQ_FOREACH_SAFE(io, &lun->ooa_queue, ooa_links, next_io) {
11509                         /* We are master */
11510                         if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
11511                                 if (io->flags & CTL_FLAG_IO_ACTIVE) {
11512                                         io->flags |= CTL_FLAG_ABORT;
11513                                         io->flags |= CTL_FLAG_FAILOVER;
11514                                 } else { /* This can be only due to DATAMOVE */
11515                                         io->msg_type = CTL_MSG_DATAMOVE_DONE;
11516                                         io->flags &= ~CTL_FLAG_DMA_INPROG;
11517                                         io->flags |= CTL_FLAG_IO_ACTIVE;
11518                                         io->port_status = 31340;
11519                                         ctl_enqueue_isc((union ctl_io *)io);
11520                                 }
11521                         }
11522                         /* We are slave */
11523                         if (io->flags & CTL_FLAG_SENT_2OTHER_SC) {
11524                                 io->flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11525                                 if (io->flags & CTL_FLAG_IO_ACTIVE) {
11526                                         io->flags |= CTL_FLAG_FAILOVER;
11527                                 } else {
11528                                         ctl_set_busy(&((union ctl_io *)io)->
11529                                             scsiio);
11530                                         ctl_done((union ctl_io *)io);
11531                                 }
11532                         }
11533                 }
11534         } else { /* SERIALIZE modes */
11535                 TAILQ_FOREACH_SAFE(io, &lun->blocked_queue, blocked_links,
11536                     next_io) {
11537                         /* We are master */
11538                         if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
11539                                 TAILQ_REMOVE(&lun->blocked_queue, io,
11540                                     blocked_links);
11541                                 io->flags &= ~CTL_FLAG_BLOCKED;
11542                                 TAILQ_REMOVE(&lun->ooa_queue, io, ooa_links);
11543                                 ctl_free_io((union ctl_io *)io);
11544                         }
11545                 }
11546                 TAILQ_FOREACH_SAFE(io, &lun->ooa_queue, ooa_links, next_io) {
11547                         /* We are master */
11548                         if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
11549                                 TAILQ_REMOVE(&lun->ooa_queue, io, ooa_links);
11550                                 ctl_free_io((union ctl_io *)io);
11551                         }
11552                         /* We are slave */
11553                         if (io->flags & CTL_FLAG_SENT_2OTHER_SC) {
11554                                 io->flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11555                                 if (!(io->flags & CTL_FLAG_IO_ACTIVE)) {
11556                                         ctl_set_busy(&((union ctl_io *)io)->
11557                                             scsiio);
11558                                         ctl_done((union ctl_io *)io);
11559                                 }
11560                         }
11561                 }
11562                 ctl_check_blocked(lun);
11563         }
11564         mtx_unlock(&lun->lun_lock);
11565 }
11566
11567 static int
11568 ctl_scsiio_precheck(struct ctl_softc *softc, struct ctl_scsiio *ctsio)
11569 {
11570         struct ctl_lun *lun;
11571         const struct ctl_cmd_entry *entry;
11572         uint32_t initidx, targ_lun;
11573         int retval;
11574
11575         retval = 0;
11576
11577         lun = NULL;
11578
11579         targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
11580         if ((targ_lun < CTL_MAX_LUNS)
11581          && ((lun = softc->ctl_luns[targ_lun]) != NULL)) {
11582                 /*
11583                  * If the LUN is invalid, pretend that it doesn't exist.
11584                  * It will go away as soon as all pending I/O has been
11585                  * completed.
11586                  */
11587                 mtx_lock(&lun->lun_lock);
11588                 if (lun->flags & CTL_LUN_DISABLED) {
11589                         mtx_unlock(&lun->lun_lock);
11590                         lun = NULL;
11591                         ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
11592                         ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
11593                 } else {
11594                         ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = lun;
11595                         ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr =
11596                                 lun->be_lun;
11597
11598                         /*
11599                          * Every I/O goes into the OOA queue for a
11600                          * particular LUN, and stays there until completion.
11601                          */
11602 #ifdef CTL_TIME_IO
11603                         if (TAILQ_EMPTY(&lun->ooa_queue)) {
11604                                 lun->idle_time += getsbinuptime() -
11605                                     lun->last_busy;
11606                         }
11607 #endif
11608                         TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr,
11609                             ooa_links);
11610                 }
11611         } else {
11612                 ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
11613                 ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
11614         }
11615
11616         /* Get command entry and return error if it is unsuppotyed. */
11617         entry = ctl_validate_command(ctsio);
11618         if (entry == NULL) {
11619                 if (lun)
11620                         mtx_unlock(&lun->lun_lock);
11621                 return (retval);
11622         }
11623
11624         ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
11625         ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
11626
11627         /*
11628          * Check to see whether we can send this command to LUNs that don't
11629          * exist.  This should pretty much only be the case for inquiry
11630          * and request sense.  Further checks, below, really require having
11631          * a LUN, so we can't really check the command anymore.  Just put
11632          * it on the rtr queue.
11633          */
11634         if (lun == NULL) {
11635                 if (entry->flags & CTL_CMD_FLAG_OK_ON_NO_LUN) {
11636                         ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11637                         ctl_enqueue_rtr((union ctl_io *)ctsio);
11638                         return (retval);
11639                 }
11640
11641                 ctl_set_unsupported_lun(ctsio);
11642                 ctl_done((union ctl_io *)ctsio);
11643                 CTL_DEBUG_PRINT(("ctl_scsiio_precheck: bailing out due to invalid LUN\n"));
11644                 return (retval);
11645         } else {
11646                 /*
11647                  * Make sure we support this particular command on this LUN.
11648                  * e.g., we don't support writes to the control LUN.
11649                  */
11650                 if (!ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
11651                         mtx_unlock(&lun->lun_lock);
11652                         ctl_set_invalid_opcode(ctsio);
11653                         ctl_done((union ctl_io *)ctsio);
11654                         return (retval);
11655                 }
11656         }
11657
11658         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11659
11660 #ifdef CTL_WITH_CA
11661         /*
11662          * If we've got a request sense, it'll clear the contingent
11663          * allegiance condition.  Otherwise, if we have a CA condition for
11664          * this initiator, clear it, because it sent down a command other
11665          * than request sense.
11666          */
11667         if ((ctsio->cdb[0] != REQUEST_SENSE)
11668          && (ctl_is_set(lun->have_ca, initidx)))
11669                 ctl_clear_mask(lun->have_ca, initidx);
11670 #endif
11671
11672         /*
11673          * If the command has this flag set, it handles its own unit
11674          * attention reporting, we shouldn't do anything.  Otherwise we
11675          * check for any pending unit attentions, and send them back to the
11676          * initiator.  We only do this when a command initially comes in,
11677          * not when we pull it off the blocked queue.
11678          *
11679          * According to SAM-3, section 5.3.2, the order that things get
11680          * presented back to the host is basically unit attentions caused
11681          * by some sort of reset event, busy status, reservation conflicts
11682          * or task set full, and finally any other status.
11683          *
11684          * One issue here is that some of the unit attentions we report
11685          * don't fall into the "reset" category (e.g. "reported luns data
11686          * has changed").  So reporting it here, before the reservation
11687          * check, may be technically wrong.  I guess the only thing to do
11688          * would be to check for and report the reset events here, and then
11689          * check for the other unit attention types after we check for a
11690          * reservation conflict.
11691          *
11692          * XXX KDM need to fix this
11693          */
11694         if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
11695                 ctl_ua_type ua_type;
11696
11697                 ua_type = ctl_build_ua(lun, initidx, &ctsio->sense_data,
11698                     SSD_TYPE_NONE);
11699                 if (ua_type != CTL_UA_NONE) {
11700                         mtx_unlock(&lun->lun_lock);
11701                         ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
11702                         ctsio->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
11703                         ctsio->sense_len = SSD_FULL_SIZE;
11704                         ctl_done((union ctl_io *)ctsio);
11705                         return (retval);
11706                 }
11707         }
11708
11709
11710         if (ctl_scsiio_lun_check(lun, entry, ctsio) != 0) {
11711                 mtx_unlock(&lun->lun_lock);
11712                 ctl_done((union ctl_io *)ctsio);
11713                 return (retval);
11714         }
11715
11716         /*
11717          * XXX CHD this is where we want to send IO to other side if
11718          * this LUN is secondary on this SC. We will need to make a copy
11719          * of the IO and flag the IO on this side as SENT_2OTHER and the flag
11720          * the copy we send as FROM_OTHER.
11721          * We also need to stuff the address of the original IO so we can
11722          * find it easily. Something similar will need be done on the other
11723          * side so when we are done we can find the copy.
11724          */
11725         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
11726             (lun->flags & CTL_LUN_PEER_SC_PRIMARY) != 0 &&
11727             (entry->flags & CTL_CMD_FLAG_RUN_HERE) == 0) {
11728                 union ctl_ha_msg msg_info;
11729                 int isc_retval;
11730
11731                 ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11732                 ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11733                 mtx_unlock(&lun->lun_lock);
11734
11735                 msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
11736                 msg_info.hdr.original_sc = (union ctl_io *)ctsio;
11737                 msg_info.hdr.serializing_sc = NULL;
11738                 msg_info.hdr.nexus = ctsio->io_hdr.nexus;
11739                 msg_info.scsi.tag_num = ctsio->tag_num;
11740                 msg_info.scsi.tag_type = ctsio->tag_type;
11741                 msg_info.scsi.cdb_len = ctsio->cdb_len;
11742                 memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
11743
11744                 if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11745                     sizeof(msg_info.scsi) - sizeof(msg_info.scsi.sense_data),
11746                     M_WAITOK)) > CTL_HA_STATUS_SUCCESS) {
11747                         ctl_set_busy(ctsio);
11748                         ctl_done((union ctl_io *)ctsio);
11749                         return (retval);
11750                 }
11751                 return (retval);
11752         }
11753
11754         switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
11755                               (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr,
11756                               ctl_ooaq, ooa_links))) {
11757         case CTL_ACTION_BLOCK:
11758                 ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
11759                 TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
11760                                   blocked_links);
11761                 mtx_unlock(&lun->lun_lock);
11762                 return (retval);
11763         case CTL_ACTION_PASS:
11764         case CTL_ACTION_SKIP:
11765                 ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11766                 mtx_unlock(&lun->lun_lock);
11767                 ctl_enqueue_rtr((union ctl_io *)ctsio);
11768                 break;
11769         case CTL_ACTION_OVERLAP:
11770                 mtx_unlock(&lun->lun_lock);
11771                 ctl_set_overlapped_cmd(ctsio);
11772                 ctl_done((union ctl_io *)ctsio);
11773                 break;
11774         case CTL_ACTION_OVERLAP_TAG:
11775                 mtx_unlock(&lun->lun_lock);
11776                 ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
11777                 ctl_done((union ctl_io *)ctsio);
11778                 break;
11779         case CTL_ACTION_ERROR:
11780         default:
11781                 mtx_unlock(&lun->lun_lock);
11782                 ctl_set_internal_failure(ctsio,
11783                                          /*sks_valid*/ 0,
11784                                          /*retry_count*/ 0);
11785                 ctl_done((union ctl_io *)ctsio);
11786                 break;
11787         }
11788         return (retval);
11789 }
11790
11791 const struct ctl_cmd_entry *
11792 ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa)
11793 {
11794         const struct ctl_cmd_entry *entry;
11795         int service_action;
11796
11797         entry = &ctl_cmd_table[ctsio->cdb[0]];
11798         if (sa)
11799                 *sa = ((entry->flags & CTL_CMD_FLAG_SA5) != 0);
11800         if (entry->flags & CTL_CMD_FLAG_SA5) {
11801                 service_action = ctsio->cdb[1] & SERVICE_ACTION_MASK;
11802                 entry = &((const struct ctl_cmd_entry *)
11803                     entry->execute)[service_action];
11804         }
11805         return (entry);
11806 }
11807
11808 const struct ctl_cmd_entry *
11809 ctl_validate_command(struct ctl_scsiio *ctsio)
11810 {
11811         const struct ctl_cmd_entry *entry;
11812         int i, sa;
11813         uint8_t diff;
11814
11815         entry = ctl_get_cmd_entry(ctsio, &sa);
11816         if (entry->execute == NULL) {
11817                 if (sa)
11818                         ctl_set_invalid_field(ctsio,
11819                                               /*sks_valid*/ 1,
11820                                               /*command*/ 1,
11821                                               /*field*/ 1,
11822                                               /*bit_valid*/ 1,
11823                                               /*bit*/ 4);
11824                 else
11825                         ctl_set_invalid_opcode(ctsio);
11826                 ctl_done((union ctl_io *)ctsio);
11827                 return (NULL);
11828         }
11829         KASSERT(entry->length > 0,
11830             ("Not defined length for command 0x%02x/0x%02x",
11831              ctsio->cdb[0], ctsio->cdb[1]));
11832         for (i = 1; i < entry->length; i++) {
11833                 diff = ctsio->cdb[i] & ~entry->usage[i - 1];
11834                 if (diff == 0)
11835                         continue;
11836                 ctl_set_invalid_field(ctsio,
11837                                       /*sks_valid*/ 1,
11838                                       /*command*/ 1,
11839                                       /*field*/ i,
11840                                       /*bit_valid*/ 1,
11841                                       /*bit*/ fls(diff) - 1);
11842                 ctl_done((union ctl_io *)ctsio);
11843                 return (NULL);
11844         }
11845         return (entry);
11846 }
11847
11848 static int
11849 ctl_cmd_applicable(uint8_t lun_type, const struct ctl_cmd_entry *entry)
11850 {
11851
11852         switch (lun_type) {
11853         case T_DIRECT:
11854                 if ((entry->flags & CTL_CMD_FLAG_OK_ON_DIRECT) == 0)
11855                         return (0);
11856                 break;
11857         case T_PROCESSOR:
11858                 if ((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0)
11859                         return (0);
11860                 break;
11861         case T_CDROM:
11862                 if ((entry->flags & CTL_CMD_FLAG_OK_ON_CDROM) == 0)
11863                         return (0);
11864                 break;
11865         default:
11866                 return (0);
11867         }
11868         return (1);
11869 }
11870
11871 static int
11872 ctl_scsiio(struct ctl_scsiio *ctsio)
11873 {
11874         int retval;
11875         const struct ctl_cmd_entry *entry;
11876
11877         retval = CTL_RETVAL_COMPLETE;
11878
11879         CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
11880
11881         entry = ctl_get_cmd_entry(ctsio, NULL);
11882
11883         /*
11884          * If this I/O has been aborted, just send it straight to
11885          * ctl_done() without executing it.
11886          */
11887         if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
11888                 ctl_done((union ctl_io *)ctsio);
11889                 goto bailout;
11890         }
11891
11892         /*
11893          * All the checks should have been handled by ctl_scsiio_precheck().
11894          * We should be clear now to just execute the I/O.
11895          */
11896         retval = entry->execute(ctsio);
11897
11898 bailout:
11899         return (retval);
11900 }
11901
11902 /*
11903  * Since we only implement one target right now, a bus reset simply resets
11904  * our single target.
11905  */
11906 static int
11907 ctl_bus_reset(struct ctl_softc *softc, union ctl_io *io)
11908 {
11909         return(ctl_target_reset(softc, io, CTL_UA_BUS_RESET));
11910 }
11911
11912 static int
11913 ctl_target_reset(struct ctl_softc *softc, union ctl_io *io,
11914                  ctl_ua_type ua_type)
11915 {
11916         struct ctl_port *port;
11917         struct ctl_lun *lun;
11918         int retval;
11919
11920         if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11921                 union ctl_ha_msg msg_info;
11922
11923                 msg_info.hdr.nexus = io->io_hdr.nexus;
11924                 if (ua_type==CTL_UA_TARG_RESET)
11925                         msg_info.task.task_action = CTL_TASK_TARGET_RESET;
11926                 else
11927                         msg_info.task.task_action = CTL_TASK_BUS_RESET;
11928                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11929                 msg_info.hdr.original_sc = NULL;
11930                 msg_info.hdr.serializing_sc = NULL;
11931                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11932                     sizeof(msg_info.task), M_WAITOK);
11933         }
11934         retval = 0;
11935
11936         mtx_lock(&softc->ctl_lock);
11937         port = ctl_io_port(&io->io_hdr);
11938         STAILQ_FOREACH(lun, &softc->lun_list, links) {
11939                 if (port != NULL &&
11940                     ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
11941                         continue;
11942                 retval += ctl_do_lun_reset(lun, io, ua_type);
11943         }
11944         mtx_unlock(&softc->ctl_lock);
11945         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11946         return (retval);
11947 }
11948
11949 /*
11950  * The LUN should always be set.  The I/O is optional, and is used to
11951  * distinguish between I/Os sent by this initiator, and by other
11952  * initiators.  We set unit attention for initiators other than this one.
11953  * SAM-3 is vague on this point.  It does say that a unit attention should
11954  * be established for other initiators when a LUN is reset (see section
11955  * 5.7.3), but it doesn't specifically say that the unit attention should
11956  * be established for this particular initiator when a LUN is reset.  Here
11957  * is the relevant text, from SAM-3 rev 8:
11958  *
11959  * 5.7.2 When a SCSI initiator port aborts its own tasks
11960  *
11961  * When a SCSI initiator port causes its own task(s) to be aborted, no
11962  * notification that the task(s) have been aborted shall be returned to
11963  * the SCSI initiator port other than the completion response for the
11964  * command or task management function action that caused the task(s) to
11965  * be aborted and notification(s) associated with related effects of the
11966  * action (e.g., a reset unit attention condition).
11967  *
11968  * XXX KDM for now, we're setting unit attention for all initiators.
11969  */
11970 static int
11971 ctl_do_lun_reset(struct ctl_lun *lun, union ctl_io *io, ctl_ua_type ua_type)
11972 {
11973         union ctl_io *xio;
11974 #if 0
11975         uint32_t initidx;
11976 #endif
11977         int i;
11978
11979         mtx_lock(&lun->lun_lock);
11980         /*
11981          * Run through the OOA queue and abort each I/O.
11982          */
11983         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11984              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11985                 xio->io_hdr.flags |= CTL_FLAG_ABORT | CTL_FLAG_ABORT_STATUS;
11986         }
11987
11988         /*
11989          * This version sets unit attention for every
11990          */
11991 #if 0
11992         initidx = ctl_get_initindex(&io->io_hdr.nexus);
11993         ctl_est_ua_all(lun, initidx, ua_type);
11994 #else
11995         ctl_est_ua_all(lun, -1, ua_type);
11996 #endif
11997
11998         /*
11999          * A reset (any kind, really) clears reservations established with
12000          * RESERVE/RELEASE.  It does not clear reservations established
12001          * with PERSISTENT RESERVE OUT, but we don't support that at the
12002          * moment anyway.  See SPC-2, section 5.6.  SPC-3 doesn't address
12003          * reservations made with the RESERVE/RELEASE commands, because
12004          * those commands are obsolete in SPC-3.
12005          */
12006         lun->flags &= ~CTL_LUN_RESERVED;
12007
12008 #ifdef CTL_WITH_CA
12009         for (i = 0; i < CTL_MAX_INITIATORS; i++)
12010                 ctl_clear_mask(lun->have_ca, i);
12011 #endif
12012         lun->prevent_count = 0;
12013         for (i = 0; i < CTL_MAX_INITIATORS; i++)
12014                 ctl_clear_mask(lun->prevent, i);
12015         mtx_unlock(&lun->lun_lock);
12016
12017         return (0);
12018 }
12019
12020 static int
12021 ctl_lun_reset(struct ctl_softc *softc, union ctl_io *io)
12022 {
12023         struct ctl_lun *lun;
12024         uint32_t targ_lun;
12025         int retval;
12026
12027         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12028         mtx_lock(&softc->ctl_lock);
12029         if ((targ_lun >= CTL_MAX_LUNS) ||
12030             (lun = softc->ctl_luns[targ_lun]) == NULL) {
12031                 mtx_unlock(&softc->ctl_lock);
12032                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
12033                 return (1);
12034         }
12035         retval = ctl_do_lun_reset(lun, io, CTL_UA_LUN_RESET);
12036         mtx_unlock(&softc->ctl_lock);
12037         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12038
12039         if ((io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) == 0) {
12040                 union ctl_ha_msg msg_info;
12041
12042                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
12043                 msg_info.hdr.nexus = io->io_hdr.nexus;
12044                 msg_info.task.task_action = CTL_TASK_LUN_RESET;
12045                 msg_info.hdr.original_sc = NULL;
12046                 msg_info.hdr.serializing_sc = NULL;
12047                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
12048                     sizeof(msg_info.task), M_WAITOK);
12049         }
12050         return (retval);
12051 }
12052
12053 static void
12054 ctl_abort_tasks_lun(struct ctl_lun *lun, uint32_t targ_port, uint32_t init_id,
12055     int other_sc)
12056 {
12057         union ctl_io *xio;
12058
12059         mtx_assert(&lun->lun_lock, MA_OWNED);
12060
12061         /*
12062          * Run through the OOA queue and attempt to find the given I/O.
12063          * The target port, initiator ID, tag type and tag number have to
12064          * match the values that we got from the initiator.  If we have an
12065          * untagged command to abort, simply abort the first untagged command
12066          * we come to.  We only allow one untagged command at a time of course.
12067          */
12068         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12069              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12070
12071                 if ((targ_port == UINT32_MAX ||
12072                      targ_port == xio->io_hdr.nexus.targ_port) &&
12073                     (init_id == UINT32_MAX ||
12074                      init_id == xio->io_hdr.nexus.initid)) {
12075                         if (targ_port != xio->io_hdr.nexus.targ_port ||
12076                             init_id != xio->io_hdr.nexus.initid)
12077                                 xio->io_hdr.flags |= CTL_FLAG_ABORT_STATUS;
12078                         xio->io_hdr.flags |= CTL_FLAG_ABORT;
12079                         if (!other_sc && !(lun->flags & CTL_LUN_PRIMARY_SC)) {
12080                                 union ctl_ha_msg msg_info;
12081
12082                                 msg_info.hdr.nexus = xio->io_hdr.nexus;
12083                                 msg_info.task.task_action = CTL_TASK_ABORT_TASK;
12084                                 msg_info.task.tag_num = xio->scsiio.tag_num;
12085                                 msg_info.task.tag_type = xio->scsiio.tag_type;
12086                                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
12087                                 msg_info.hdr.original_sc = NULL;
12088                                 msg_info.hdr.serializing_sc = NULL;
12089                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
12090                                     sizeof(msg_info.task), M_NOWAIT);
12091                         }
12092                 }
12093         }
12094 }
12095
12096 static int
12097 ctl_abort_task_set(union ctl_io *io)
12098 {
12099         struct ctl_softc *softc = control_softc;
12100         struct ctl_lun *lun;
12101         uint32_t targ_lun;
12102
12103         /*
12104          * Look up the LUN.
12105          */
12106         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12107         mtx_lock(&softc->ctl_lock);
12108         if ((targ_lun >= CTL_MAX_LUNS) ||
12109             (lun = softc->ctl_luns[targ_lun]) == NULL) {
12110                 mtx_unlock(&softc->ctl_lock);
12111                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
12112                 return (1);
12113         }
12114
12115         mtx_lock(&lun->lun_lock);
12116         mtx_unlock(&softc->ctl_lock);
12117         if (io->taskio.task_action == CTL_TASK_ABORT_TASK_SET) {
12118                 ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
12119                     io->io_hdr.nexus.initid,
12120                     (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12121         } else { /* CTL_TASK_CLEAR_TASK_SET */
12122                 ctl_abort_tasks_lun(lun, UINT32_MAX, UINT32_MAX,
12123                     (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12124         }
12125         mtx_unlock(&lun->lun_lock);
12126         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12127         return (0);
12128 }
12129
12130 static int
12131 ctl_i_t_nexus_reset(union ctl_io *io)
12132 {
12133         struct ctl_softc *softc = control_softc;
12134         struct ctl_lun *lun;
12135         uint32_t initidx;
12136
12137         if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
12138                 union ctl_ha_msg msg_info;
12139
12140                 msg_info.hdr.nexus = io->io_hdr.nexus;
12141                 msg_info.task.task_action = CTL_TASK_I_T_NEXUS_RESET;
12142                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
12143                 msg_info.hdr.original_sc = NULL;
12144                 msg_info.hdr.serializing_sc = NULL;
12145                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
12146                     sizeof(msg_info.task), M_WAITOK);
12147         }
12148
12149         initidx = ctl_get_initindex(&io->io_hdr.nexus);
12150         mtx_lock(&softc->ctl_lock);
12151         STAILQ_FOREACH(lun, &softc->lun_list, links) {
12152                 mtx_lock(&lun->lun_lock);
12153                 ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
12154                     io->io_hdr.nexus.initid, 1);
12155 #ifdef CTL_WITH_CA
12156                 ctl_clear_mask(lun->have_ca, initidx);
12157 #endif
12158                 if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == initidx))
12159                         lun->flags &= ~CTL_LUN_RESERVED;
12160                 if (ctl_is_set(lun->prevent, initidx)) {
12161                         ctl_clear_mask(lun->prevent, initidx);
12162                         lun->prevent_count--;
12163                 }
12164                 ctl_est_ua(lun, initidx, CTL_UA_I_T_NEXUS_LOSS);
12165                 mtx_unlock(&lun->lun_lock);
12166         }
12167         mtx_unlock(&softc->ctl_lock);
12168         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12169         return (0);
12170 }
12171
12172 static int
12173 ctl_abort_task(union ctl_io *io)
12174 {
12175         union ctl_io *xio;
12176         struct ctl_lun *lun;
12177         struct ctl_softc *softc;
12178 #if 0
12179         struct sbuf sb;
12180         char printbuf[128];
12181 #endif
12182         int found;
12183         uint32_t targ_lun;
12184
12185         softc = control_softc;
12186         found = 0;
12187
12188         /*
12189          * Look up the LUN.
12190          */
12191         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12192         mtx_lock(&softc->ctl_lock);
12193         if ((targ_lun >= CTL_MAX_LUNS) ||
12194             (lun = softc->ctl_luns[targ_lun]) == NULL) {
12195                 mtx_unlock(&softc->ctl_lock);
12196                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
12197                 return (1);
12198         }
12199
12200 #if 0
12201         printf("ctl_abort_task: called for lun %lld, tag %d type %d\n",
12202                lun->lun, io->taskio.tag_num, io->taskio.tag_type);
12203 #endif
12204
12205         mtx_lock(&lun->lun_lock);
12206         mtx_unlock(&softc->ctl_lock);
12207         /*
12208          * Run through the OOA queue and attempt to find the given I/O.
12209          * The target port, initiator ID, tag type and tag number have to
12210          * match the values that we got from the initiator.  If we have an
12211          * untagged command to abort, simply abort the first untagged command
12212          * we come to.  We only allow one untagged command at a time of course.
12213          */
12214         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12215              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12216 #if 0
12217                 sbuf_new(&sb, printbuf, sizeof(printbuf), SBUF_FIXEDLEN);
12218
12219                 sbuf_printf(&sb, "LUN %lld tag %d type %d%s%s%s%s: ",
12220                             lun->lun, xio->scsiio.tag_num,
12221                             xio->scsiio.tag_type,
12222                             (xio->io_hdr.blocked_links.tqe_prev
12223                             == NULL) ? "" : " BLOCKED",
12224                             (xio->io_hdr.flags &
12225                             CTL_FLAG_DMA_INPROG) ? " DMA" : "",
12226                             (xio->io_hdr.flags &
12227                             CTL_FLAG_ABORT) ? " ABORT" : "",
12228                             (xio->io_hdr.flags &
12229                             CTL_FLAG_IS_WAS_ON_RTR ? " RTR" : ""));
12230                 ctl_scsi_command_string(&xio->scsiio, NULL, &sb);
12231                 sbuf_finish(&sb);
12232                 printf("%s\n", sbuf_data(&sb));
12233 #endif
12234
12235                 if ((xio->io_hdr.nexus.targ_port != io->io_hdr.nexus.targ_port)
12236                  || (xio->io_hdr.nexus.initid != io->io_hdr.nexus.initid)
12237                  || (xio->io_hdr.flags & CTL_FLAG_ABORT))
12238                         continue;
12239
12240                 /*
12241                  * If the abort says that the task is untagged, the
12242                  * task in the queue must be untagged.  Otherwise,
12243                  * we just check to see whether the tag numbers
12244                  * match.  This is because the QLogic firmware
12245                  * doesn't pass back the tag type in an abort
12246                  * request.
12247                  */
12248 #if 0
12249                 if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
12250                   && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
12251                  || (xio->scsiio.tag_num == io->taskio.tag_num))
12252 #endif
12253                 /*
12254                  * XXX KDM we've got problems with FC, because it
12255                  * doesn't send down a tag type with aborts.  So we
12256                  * can only really go by the tag number...
12257                  * This may cause problems with parallel SCSI.
12258                  * Need to figure that out!!
12259                  */
12260                 if (xio->scsiio.tag_num == io->taskio.tag_num) {
12261                         xio->io_hdr.flags |= CTL_FLAG_ABORT;
12262                         found = 1;
12263                         if ((io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) == 0 &&
12264                             !(lun->flags & CTL_LUN_PRIMARY_SC)) {
12265                                 union ctl_ha_msg msg_info;
12266
12267                                 msg_info.hdr.nexus = io->io_hdr.nexus;
12268                                 msg_info.task.task_action = CTL_TASK_ABORT_TASK;
12269                                 msg_info.task.tag_num = io->taskio.tag_num;
12270                                 msg_info.task.tag_type = io->taskio.tag_type;
12271                                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
12272                                 msg_info.hdr.original_sc = NULL;
12273                                 msg_info.hdr.serializing_sc = NULL;
12274 #if 0
12275                                 printf("Sent Abort to other side\n");
12276 #endif
12277                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
12278                                     sizeof(msg_info.task), M_NOWAIT);
12279                         }
12280 #if 0
12281                         printf("ctl_abort_task: found I/O to abort\n");
12282 #endif
12283                 }
12284         }
12285         mtx_unlock(&lun->lun_lock);
12286
12287         if (found == 0) {
12288                 /*
12289                  * This isn't really an error.  It's entirely possible for
12290                  * the abort and command completion to cross on the wire.
12291                  * This is more of an informative/diagnostic error.
12292                  */
12293 #if 0
12294                 printf("ctl_abort_task: ABORT sent for nonexistent I/O: "
12295                        "%u:%u:%u tag %d type %d\n",
12296                        io->io_hdr.nexus.initid,
12297                        io->io_hdr.nexus.targ_port,
12298                        io->io_hdr.nexus.targ_lun, io->taskio.tag_num,
12299                        io->taskio.tag_type);
12300 #endif
12301         }
12302         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12303         return (0);
12304 }
12305
12306 static int
12307 ctl_query_task(union ctl_io *io, int task_set)
12308 {
12309         union ctl_io *xio;
12310         struct ctl_lun *lun;
12311         struct ctl_softc *softc;
12312         int found = 0;
12313         uint32_t targ_lun;
12314
12315         softc = control_softc;
12316         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12317         mtx_lock(&softc->ctl_lock);
12318         if ((targ_lun >= CTL_MAX_LUNS) ||
12319             (lun = softc->ctl_luns[targ_lun]) == NULL) {
12320                 mtx_unlock(&softc->ctl_lock);
12321                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
12322                 return (1);
12323         }
12324         mtx_lock(&lun->lun_lock);
12325         mtx_unlock(&softc->ctl_lock);
12326         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12327              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12328
12329                 if ((xio->io_hdr.nexus.targ_port != io->io_hdr.nexus.targ_port)
12330                  || (xio->io_hdr.nexus.initid != io->io_hdr.nexus.initid)
12331                  || (xio->io_hdr.flags & CTL_FLAG_ABORT))
12332                         continue;
12333
12334                 if (task_set || xio->scsiio.tag_num == io->taskio.tag_num) {
12335                         found = 1;
12336                         break;
12337                 }
12338         }
12339         mtx_unlock(&lun->lun_lock);
12340         if (found)
12341                 io->taskio.task_status = CTL_TASK_FUNCTION_SUCCEEDED;
12342         else
12343                 io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12344         return (0);
12345 }
12346
12347 static int
12348 ctl_query_async_event(union ctl_io *io)
12349 {
12350         struct ctl_lun *lun;
12351         struct ctl_softc *softc;
12352         ctl_ua_type ua;
12353         uint32_t targ_lun, initidx;
12354
12355         softc = control_softc;
12356         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12357         mtx_lock(&softc->ctl_lock);
12358         if ((targ_lun >= CTL_MAX_LUNS) ||
12359             (lun = softc->ctl_luns[targ_lun]) == NULL) {
12360                 mtx_unlock(&softc->ctl_lock);
12361                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
12362                 return (1);
12363         }
12364         mtx_lock(&lun->lun_lock);
12365         mtx_unlock(&softc->ctl_lock);
12366         initidx = ctl_get_initindex(&io->io_hdr.nexus);
12367         ua = ctl_build_qae(lun, initidx, io->taskio.task_resp);
12368         mtx_unlock(&lun->lun_lock);
12369         if (ua != CTL_UA_NONE)
12370                 io->taskio.task_status = CTL_TASK_FUNCTION_SUCCEEDED;
12371         else
12372                 io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12373         return (0);
12374 }
12375
12376 static void
12377 ctl_run_task(union ctl_io *io)
12378 {
12379         struct ctl_softc *softc = control_softc;
12380         int retval = 1;
12381
12382         CTL_DEBUG_PRINT(("ctl_run_task\n"));
12383         KASSERT(io->io_hdr.io_type == CTL_IO_TASK,
12384             ("ctl_run_task: Unextected io_type %d\n", io->io_hdr.io_type));
12385         io->taskio.task_status = CTL_TASK_FUNCTION_NOT_SUPPORTED;
12386         bzero(io->taskio.task_resp, sizeof(io->taskio.task_resp));
12387         switch (io->taskio.task_action) {
12388         case CTL_TASK_ABORT_TASK:
12389                 retval = ctl_abort_task(io);
12390                 break;
12391         case CTL_TASK_ABORT_TASK_SET:
12392         case CTL_TASK_CLEAR_TASK_SET:
12393                 retval = ctl_abort_task_set(io);
12394                 break;
12395         case CTL_TASK_CLEAR_ACA:
12396                 break;
12397         case CTL_TASK_I_T_NEXUS_RESET:
12398                 retval = ctl_i_t_nexus_reset(io);
12399                 break;
12400         case CTL_TASK_LUN_RESET:
12401                 retval = ctl_lun_reset(softc, io);
12402                 break;
12403         case CTL_TASK_TARGET_RESET:
12404                 retval = ctl_target_reset(softc, io, CTL_UA_TARG_RESET);
12405                 break;
12406         case CTL_TASK_BUS_RESET:
12407                 retval = ctl_bus_reset(softc, io);
12408                 break;
12409         case CTL_TASK_PORT_LOGIN:
12410                 break;
12411         case CTL_TASK_PORT_LOGOUT:
12412                 break;
12413         case CTL_TASK_QUERY_TASK:
12414                 retval = ctl_query_task(io, 0);
12415                 break;
12416         case CTL_TASK_QUERY_TASK_SET:
12417                 retval = ctl_query_task(io, 1);
12418                 break;
12419         case CTL_TASK_QUERY_ASYNC_EVENT:
12420                 retval = ctl_query_async_event(io);
12421                 break;
12422         default:
12423                 printf("%s: got unknown task management event %d\n",
12424                        __func__, io->taskio.task_action);
12425                 break;
12426         }
12427         if (retval == 0)
12428                 io->io_hdr.status = CTL_SUCCESS;
12429         else
12430                 io->io_hdr.status = CTL_ERROR;
12431         ctl_done(io);
12432 }
12433
12434 /*
12435  * For HA operation.  Handle commands that come in from the other
12436  * controller.
12437  */
12438 static void
12439 ctl_handle_isc(union ctl_io *io)
12440 {
12441         int free_io;
12442         struct ctl_lun *lun;
12443         struct ctl_softc *softc = control_softc;
12444         uint32_t targ_lun;
12445
12446         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12447         lun = softc->ctl_luns[targ_lun];
12448
12449         switch (io->io_hdr.msg_type) {
12450         case CTL_MSG_SERIALIZE:
12451                 free_io = ctl_serialize_other_sc_cmd(&io->scsiio);
12452                 break;
12453         case CTL_MSG_R2R: {
12454                 const struct ctl_cmd_entry *entry;
12455
12456                 /*
12457                  * This is only used in SER_ONLY mode.
12458                  */
12459                 free_io = 0;
12460                 entry = ctl_get_cmd_entry(&io->scsiio, NULL);
12461                 mtx_lock(&lun->lun_lock);
12462                 if (ctl_scsiio_lun_check(lun,
12463                     entry, (struct ctl_scsiio *)io) != 0) {
12464                         mtx_unlock(&lun->lun_lock);
12465                         ctl_done(io);
12466                         break;
12467                 }
12468                 io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
12469                 mtx_unlock(&lun->lun_lock);
12470                 ctl_enqueue_rtr(io);
12471                 break;
12472         }
12473         case CTL_MSG_FINISH_IO:
12474                 if (softc->ha_mode == CTL_HA_MODE_XFER) {
12475                         free_io = 0;
12476                         ctl_done(io);
12477                 } else {
12478                         free_io = 1;
12479                         mtx_lock(&lun->lun_lock);
12480                         TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr,
12481                                      ooa_links);
12482                         ctl_check_blocked(lun);
12483                         mtx_unlock(&lun->lun_lock);
12484                 }
12485                 break;
12486         case CTL_MSG_PERS_ACTION:
12487                 ctl_hndl_per_res_out_on_other_sc(
12488                         (union ctl_ha_msg *)&io->presio.pr_msg);
12489                 free_io = 1;
12490                 break;
12491         case CTL_MSG_BAD_JUJU:
12492                 free_io = 0;
12493                 ctl_done(io);
12494                 break;
12495         case CTL_MSG_DATAMOVE:
12496                 /* Only used in XFER mode */
12497                 free_io = 0;
12498                 ctl_datamove_remote(io);
12499                 break;
12500         case CTL_MSG_DATAMOVE_DONE:
12501                 /* Only used in XFER mode */
12502                 free_io = 0;
12503                 io->scsiio.be_move_done(io);
12504                 break;
12505         case CTL_MSG_FAILOVER:
12506                 ctl_failover_lun(io);
12507                 free_io = 1;
12508                 break;
12509         default:
12510                 free_io = 1;
12511                 printf("%s: Invalid message type %d\n",
12512                        __func__, io->io_hdr.msg_type);
12513                 break;
12514         }
12515         if (free_io)
12516                 ctl_free_io(io);
12517
12518 }
12519
12520
12521 /*
12522  * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
12523  * there is no match.
12524  */
12525 static ctl_lun_error_pattern
12526 ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
12527 {
12528         const struct ctl_cmd_entry *entry;
12529         ctl_lun_error_pattern filtered_pattern, pattern;
12530
12531         pattern = desc->error_pattern;
12532
12533         /*
12534          * XXX KDM we need more data passed into this function to match a
12535          * custom pattern, and we actually need to implement custom pattern
12536          * matching.
12537          */
12538         if (pattern & CTL_LUN_PAT_CMD)
12539                 return (CTL_LUN_PAT_CMD);
12540
12541         if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
12542                 return (CTL_LUN_PAT_ANY);
12543
12544         entry = ctl_get_cmd_entry(ctsio, NULL);
12545
12546         filtered_pattern = entry->pattern & pattern;
12547
12548         /*
12549          * If the user requested specific flags in the pattern (e.g.
12550          * CTL_LUN_PAT_RANGE), make sure the command supports all of those
12551          * flags.
12552          *
12553          * If the user did not specify any flags, it doesn't matter whether
12554          * or not the command supports the flags.
12555          */
12556         if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
12557              (pattern & ~CTL_LUN_PAT_MASK))
12558                 return (CTL_LUN_PAT_NONE);
12559
12560         /*
12561          * If the user asked for a range check, see if the requested LBA
12562          * range overlaps with this command's LBA range.
12563          */
12564         if (filtered_pattern & CTL_LUN_PAT_RANGE) {
12565                 uint64_t lba1;
12566                 uint64_t len1;
12567                 ctl_action action;
12568                 int retval;
12569
12570                 retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
12571                 if (retval != 0)
12572                         return (CTL_LUN_PAT_NONE);
12573
12574                 action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
12575                                               desc->lba_range.len, FALSE);
12576                 /*
12577                  * A "pass" means that the LBA ranges don't overlap, so
12578                  * this doesn't match the user's range criteria.
12579                  */
12580                 if (action == CTL_ACTION_PASS)
12581                         return (CTL_LUN_PAT_NONE);
12582         }
12583
12584         return (filtered_pattern);
12585 }
12586
12587 static void
12588 ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
12589 {
12590         struct ctl_error_desc *desc, *desc2;
12591
12592         mtx_assert(&lun->lun_lock, MA_OWNED);
12593
12594         STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
12595                 ctl_lun_error_pattern pattern;
12596                 /*
12597                  * Check to see whether this particular command matches
12598                  * the pattern in the descriptor.
12599                  */
12600                 pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
12601                 if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
12602                         continue;
12603
12604                 switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
12605                 case CTL_LUN_INJ_ABORTED:
12606                         ctl_set_aborted(&io->scsiio);
12607                         break;
12608                 case CTL_LUN_INJ_MEDIUM_ERR:
12609                         ctl_set_medium_error(&io->scsiio,
12610                             (io->io_hdr.flags & CTL_FLAG_DATA_MASK) !=
12611                              CTL_FLAG_DATA_OUT);
12612                         break;
12613                 case CTL_LUN_INJ_UA:
12614                         /* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
12615                          * OCCURRED */
12616                         ctl_set_ua(&io->scsiio, 0x29, 0x00);
12617                         break;
12618                 case CTL_LUN_INJ_CUSTOM:
12619                         /*
12620                          * We're assuming the user knows what he is doing.
12621                          * Just copy the sense information without doing
12622                          * checks.
12623                          */
12624                         bcopy(&desc->custom_sense, &io->scsiio.sense_data,
12625                               MIN(sizeof(desc->custom_sense),
12626                                   sizeof(io->scsiio.sense_data)));
12627                         io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
12628                         io->scsiio.sense_len = SSD_FULL_SIZE;
12629                         io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
12630                         break;
12631                 case CTL_LUN_INJ_NONE:
12632                 default:
12633                         /*
12634                          * If this is an error injection type we don't know
12635                          * about, clear the continuous flag (if it is set)
12636                          * so it will get deleted below.
12637                          */
12638                         desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
12639                         break;
12640                 }
12641                 /*
12642                  * By default, each error injection action is a one-shot
12643                  */
12644                 if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
12645                         continue;
12646
12647                 STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
12648
12649                 free(desc, M_CTL);
12650         }
12651 }
12652
12653 #ifdef CTL_IO_DELAY
12654 static void
12655 ctl_datamove_timer_wakeup(void *arg)
12656 {
12657         union ctl_io *io;
12658
12659         io = (union ctl_io *)arg;
12660
12661         ctl_datamove(io);
12662 }
12663 #endif /* CTL_IO_DELAY */
12664
12665 void
12666 ctl_datamove(union ctl_io *io)
12667 {
12668         struct ctl_lun *lun;
12669         void (*fe_datamove)(union ctl_io *io);
12670
12671         mtx_assert(&control_softc->ctl_lock, MA_NOTOWNED);
12672
12673         CTL_DEBUG_PRINT(("ctl_datamove\n"));
12674
12675         lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12676 #ifdef CTL_TIME_IO
12677         if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12678                 char str[256];
12679                 char path_str[64];
12680                 struct sbuf sb;
12681
12682                 ctl_scsi_path_string(io, path_str, sizeof(path_str));
12683                 sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12684
12685                 sbuf_cat(&sb, path_str);
12686                 switch (io->io_hdr.io_type) {
12687                 case CTL_IO_SCSI:
12688                         ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12689                         sbuf_printf(&sb, "\n");
12690                         sbuf_cat(&sb, path_str);
12691                         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12692                                     io->scsiio.tag_num, io->scsiio.tag_type);
12693                         break;
12694                 case CTL_IO_TASK:
12695                         sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12696                                     "Tag Type: %d\n", io->taskio.task_action,
12697                                     io->taskio.tag_num, io->taskio.tag_type);
12698                         break;
12699                 default:
12700                         panic("%s: Invalid CTL I/O type %d\n",
12701                             __func__, io->io_hdr.io_type);
12702                 }
12703                 sbuf_cat(&sb, path_str);
12704                 sbuf_printf(&sb, "ctl_datamove: %jd seconds\n",
12705                             (intmax_t)time_uptime - io->io_hdr.start_time);
12706                 sbuf_finish(&sb);
12707                 printf("%s", sbuf_data(&sb));
12708         }
12709 #endif /* CTL_TIME_IO */
12710
12711 #ifdef CTL_IO_DELAY
12712         if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
12713                 io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
12714         } else {
12715                 if ((lun != NULL)
12716                  && (lun->delay_info.datamove_delay > 0)) {
12717
12718                         callout_init(&io->io_hdr.delay_callout, /*mpsafe*/ 1);
12719                         io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
12720                         callout_reset(&io->io_hdr.delay_callout,
12721                                       lun->delay_info.datamove_delay * hz,
12722                                       ctl_datamove_timer_wakeup, io);
12723                         if (lun->delay_info.datamove_type ==
12724                             CTL_DELAY_TYPE_ONESHOT)
12725                                 lun->delay_info.datamove_delay = 0;
12726                         return;
12727                 }
12728         }
12729 #endif
12730
12731         /*
12732          * This command has been aborted.  Set the port status, so we fail
12733          * the data move.
12734          */
12735         if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12736                 printf("ctl_datamove: tag 0x%04x on (%u:%u:%u) aborted\n",
12737                        io->scsiio.tag_num, io->io_hdr.nexus.initid,
12738                        io->io_hdr.nexus.targ_port,
12739                        io->io_hdr.nexus.targ_lun);
12740                 io->io_hdr.port_status = 31337;
12741                 /*
12742                  * Note that the backend, in this case, will get the
12743                  * callback in its context.  In other cases it may get
12744                  * called in the frontend's interrupt thread context.
12745                  */
12746                 io->scsiio.be_move_done(io);
12747                 return;
12748         }
12749
12750         /* Don't confuse frontend with zero length data move. */
12751         if (io->scsiio.kern_data_len == 0) {
12752                 io->scsiio.be_move_done(io);
12753                 return;
12754         }
12755
12756         fe_datamove = ctl_io_port(&io->io_hdr)->fe_datamove;
12757         fe_datamove(io);
12758 }
12759
12760 static void
12761 ctl_send_datamove_done(union ctl_io *io, int have_lock)
12762 {
12763         union ctl_ha_msg msg;
12764 #ifdef CTL_TIME_IO
12765         struct bintime cur_bt;
12766 #endif
12767
12768         memset(&msg, 0, sizeof(msg));
12769         msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
12770         msg.hdr.original_sc = io;
12771         msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
12772         msg.hdr.nexus = io->io_hdr.nexus;
12773         msg.hdr.status = io->io_hdr.status;
12774         msg.scsi.tag_num = io->scsiio.tag_num;
12775         msg.scsi.tag_type = io->scsiio.tag_type;
12776         msg.scsi.scsi_status = io->scsiio.scsi_status;
12777         memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
12778                io->scsiio.sense_len);
12779         msg.scsi.sense_len = io->scsiio.sense_len;
12780         msg.scsi.sense_residual = io->scsiio.sense_residual;
12781         msg.scsi.fetd_status = io->io_hdr.port_status;
12782         msg.scsi.residual = io->scsiio.residual;
12783         io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12784         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12785                 ctl_failover_io(io, /*have_lock*/ have_lock);
12786                 return;
12787         }
12788         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12789             sizeof(msg.scsi) - sizeof(msg.scsi.sense_data) +
12790             msg.scsi.sense_len, M_WAITOK);
12791
12792 #ifdef CTL_TIME_IO
12793         getbinuptime(&cur_bt);
12794         bintime_sub(&cur_bt, &io->io_hdr.dma_start_bt);
12795         bintime_add(&io->io_hdr.dma_bt, &cur_bt);
12796 #endif
12797         io->io_hdr.num_dmas++;
12798 }
12799
12800 /*
12801  * The DMA to the remote side is done, now we need to tell the other side
12802  * we're done so it can continue with its data movement.
12803  */
12804 static void
12805 ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
12806 {
12807         union ctl_io *io;
12808         uint32_t i;
12809
12810         io = rq->context;
12811
12812         if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12813                 printf("%s: ISC DMA write failed with error %d", __func__,
12814                        rq->ret);
12815                 ctl_set_internal_failure(&io->scsiio,
12816                                          /*sks_valid*/ 1,
12817                                          /*retry_count*/ rq->ret);
12818         }
12819
12820         ctl_dt_req_free(rq);
12821
12822         for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12823                 free(io->io_hdr.local_sglist[i].addr, M_CTL);
12824         free(io->io_hdr.remote_sglist, M_CTL);
12825         io->io_hdr.remote_sglist = NULL;
12826         io->io_hdr.local_sglist = NULL;
12827
12828         /*
12829          * The data is in local and remote memory, so now we need to send
12830          * status (good or back) back to the other side.
12831          */
12832         ctl_send_datamove_done(io, /*have_lock*/ 0);
12833 }
12834
12835 /*
12836  * We've moved the data from the host/controller into local memory.  Now we
12837  * need to push it over to the remote controller's memory.
12838  */
12839 static int
12840 ctl_datamove_remote_dm_write_cb(union ctl_io *io)
12841 {
12842         int retval;
12843
12844         retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
12845                                           ctl_datamove_remote_write_cb);
12846         return (retval);
12847 }
12848
12849 static void
12850 ctl_datamove_remote_write(union ctl_io *io)
12851 {
12852         int retval;
12853         void (*fe_datamove)(union ctl_io *io);
12854
12855         /*
12856          * - Get the data from the host/HBA into local memory.
12857          * - DMA memory from the local controller to the remote controller.
12858          * - Send status back to the remote controller.
12859          */
12860
12861         retval = ctl_datamove_remote_sgl_setup(io);
12862         if (retval != 0)
12863                 return;
12864
12865         /* Switch the pointer over so the FETD knows what to do */
12866         io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12867
12868         /*
12869          * Use a custom move done callback, since we need to send completion
12870          * back to the other controller, not to the backend on this side.
12871          */
12872         io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
12873
12874         fe_datamove = ctl_io_port(&io->io_hdr)->fe_datamove;
12875         fe_datamove(io);
12876 }
12877
12878 static int
12879 ctl_datamove_remote_dm_read_cb(union ctl_io *io)
12880 {
12881 #if 0
12882         char str[256];
12883         char path_str[64];
12884         struct sbuf sb;
12885 #endif
12886         uint32_t i;
12887
12888         for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12889                 free(io->io_hdr.local_sglist[i].addr, M_CTL);
12890         free(io->io_hdr.remote_sglist, M_CTL);
12891         io->io_hdr.remote_sglist = NULL;
12892         io->io_hdr.local_sglist = NULL;
12893
12894 #if 0
12895         scsi_path_string(io, path_str, sizeof(path_str));
12896         sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12897         sbuf_cat(&sb, path_str);
12898         scsi_command_string(&io->scsiio, NULL, &sb);
12899         sbuf_printf(&sb, "\n");
12900         sbuf_cat(&sb, path_str);
12901         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12902                     io->scsiio.tag_num, io->scsiio.tag_type);
12903         sbuf_cat(&sb, path_str);
12904         sbuf_printf(&sb, "%s: flags %#x, status %#x\n", __func__,
12905                     io->io_hdr.flags, io->io_hdr.status);
12906         sbuf_finish(&sb);
12907         printk("%s", sbuf_data(&sb));
12908 #endif
12909
12910
12911         /*
12912          * The read is done, now we need to send status (good or bad) back
12913          * to the other side.
12914          */
12915         ctl_send_datamove_done(io, /*have_lock*/ 0);
12916
12917         return (0);
12918 }
12919
12920 static void
12921 ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
12922 {
12923         union ctl_io *io;
12924         void (*fe_datamove)(union ctl_io *io);
12925
12926         io = rq->context;
12927
12928         if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12929                 printf("%s: ISC DMA read failed with error %d\n", __func__,
12930                        rq->ret);
12931                 ctl_set_internal_failure(&io->scsiio,
12932                                          /*sks_valid*/ 1,
12933                                          /*retry_count*/ rq->ret);
12934         }
12935
12936         ctl_dt_req_free(rq);
12937
12938         /* Switch the pointer over so the FETD knows what to do */
12939         io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12940
12941         /*
12942          * Use a custom move done callback, since we need to send completion
12943          * back to the other controller, not to the backend on this side.
12944          */
12945         io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
12946
12947         /* XXX KDM add checks like the ones in ctl_datamove? */
12948
12949         fe_datamove = ctl_io_port(&io->io_hdr)->fe_datamove;
12950         fe_datamove(io);
12951 }
12952
12953 static int
12954 ctl_datamove_remote_sgl_setup(union ctl_io *io)
12955 {
12956         struct ctl_sg_entry *local_sglist;
12957         uint32_t len_to_go;
12958         int retval;
12959         int i;
12960
12961         retval = 0;
12962         local_sglist = io->io_hdr.local_sglist;
12963         len_to_go = io->scsiio.kern_data_len;
12964
12965         /*
12966          * The difficult thing here is that the size of the various
12967          * S/G segments may be different than the size from the
12968          * remote controller.  That'll make it harder when DMAing
12969          * the data back to the other side.
12970          */
12971         for (i = 0; len_to_go > 0; i++) {
12972                 local_sglist[i].len = MIN(len_to_go, CTL_HA_DATAMOVE_SEGMENT);
12973                 local_sglist[i].addr =
12974                     malloc(local_sglist[i].len, M_CTL, M_WAITOK);
12975
12976                 len_to_go -= local_sglist[i].len;
12977         }
12978         /*
12979          * Reset the number of S/G entries accordingly.  The original
12980          * number of S/G entries is available in rem_sg_entries.
12981          */
12982         io->scsiio.kern_sg_entries = i;
12983
12984 #if 0
12985         printf("%s: kern_sg_entries = %d\n", __func__,
12986                io->scsiio.kern_sg_entries);
12987         for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12988                 printf("%s: sg[%d] = %p, %lu\n", __func__, i,
12989                        local_sglist[i].addr, local_sglist[i].len);
12990 #endif
12991
12992         return (retval);
12993 }
12994
12995 static int
12996 ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
12997                          ctl_ha_dt_cb callback)
12998 {
12999         struct ctl_ha_dt_req *rq;
13000         struct ctl_sg_entry *remote_sglist, *local_sglist;
13001         uint32_t local_used, remote_used, total_used;
13002         int i, j, isc_ret;
13003
13004         rq = ctl_dt_req_alloc();
13005
13006         /*
13007          * If we failed to allocate the request, and if the DMA didn't fail
13008          * anyway, set busy status.  This is just a resource allocation
13009          * failure.
13010          */
13011         if ((rq == NULL)
13012          && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
13013              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS))
13014                 ctl_set_busy(&io->scsiio);
13015
13016         if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
13017             (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS) {
13018
13019                 if (rq != NULL)
13020                         ctl_dt_req_free(rq);
13021
13022                 /*
13023                  * The data move failed.  We need to return status back
13024                  * to the other controller.  No point in trying to DMA
13025                  * data to the remote controller.
13026                  */
13027
13028                 ctl_send_datamove_done(io, /*have_lock*/ 0);
13029
13030                 return (1);
13031         }
13032
13033         local_sglist = io->io_hdr.local_sglist;
13034         remote_sglist = io->io_hdr.remote_sglist;
13035         local_used = 0;
13036         remote_used = 0;
13037         total_used = 0;
13038
13039         /*
13040          * Pull/push the data over the wire from/to the other controller.
13041          * This takes into account the possibility that the local and
13042          * remote sglists may not be identical in terms of the size of
13043          * the elements and the number of elements.
13044          *
13045          * One fundamental assumption here is that the length allocated for
13046          * both the local and remote sglists is identical.  Otherwise, we've
13047          * essentially got a coding error of some sort.
13048          */
13049         isc_ret = CTL_HA_STATUS_SUCCESS;
13050         for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
13051                 uint32_t cur_len;
13052                 uint8_t *tmp_ptr;
13053
13054                 rq->command = command;
13055                 rq->context = io;
13056
13057                 /*
13058                  * Both pointers should be aligned.  But it is possible
13059                  * that the allocation length is not.  They should both
13060                  * also have enough slack left over at the end, though,
13061                  * to round up to the next 8 byte boundary.
13062                  */
13063                 cur_len = MIN(local_sglist[i].len - local_used,
13064                               remote_sglist[j].len - remote_used);
13065                 rq->size = cur_len;
13066
13067                 tmp_ptr = (uint8_t *)local_sglist[i].addr;
13068                 tmp_ptr += local_used;
13069
13070 #if 0
13071                 /* Use physical addresses when talking to ISC hardware */
13072                 if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
13073                         /* XXX KDM use busdma */
13074                         rq->local = vtophys(tmp_ptr);
13075                 } else
13076                         rq->local = tmp_ptr;
13077 #else
13078                 KASSERT((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0,
13079                     ("HA does not support BUS_ADDR"));
13080                 rq->local = tmp_ptr;
13081 #endif
13082
13083                 tmp_ptr = (uint8_t *)remote_sglist[j].addr;
13084                 tmp_ptr += remote_used;
13085                 rq->remote = tmp_ptr;
13086
13087                 rq->callback = NULL;
13088
13089                 local_used += cur_len;
13090                 if (local_used >= local_sglist[i].len) {
13091                         i++;
13092                         local_used = 0;
13093                 }
13094
13095                 remote_used += cur_len;
13096                 if (remote_used >= remote_sglist[j].len) {
13097                         j++;
13098                         remote_used = 0;
13099                 }
13100                 total_used += cur_len;
13101
13102                 if (total_used >= io->scsiio.kern_data_len)
13103                         rq->callback = callback;
13104
13105 #if 0
13106                 printf("%s: %s: local %p remote %p size %d\n", __func__,
13107                        (command == CTL_HA_DT_CMD_WRITE) ? "WRITE" : "READ",
13108                        rq->local, rq->remote, rq->size);
13109 #endif
13110
13111                 isc_ret = ctl_dt_single(rq);
13112                 if (isc_ret > CTL_HA_STATUS_SUCCESS)
13113                         break;
13114         }
13115         if (isc_ret != CTL_HA_STATUS_WAIT) {
13116                 rq->ret = isc_ret;
13117                 callback(rq);
13118         }
13119
13120         return (0);
13121 }
13122
13123 static void
13124 ctl_datamove_remote_read(union ctl_io *io)
13125 {
13126         int retval;
13127         uint32_t i;
13128
13129         /*
13130          * This will send an error to the other controller in the case of a
13131          * failure.
13132          */
13133         retval = ctl_datamove_remote_sgl_setup(io);
13134         if (retval != 0)
13135                 return;
13136
13137         retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
13138                                           ctl_datamove_remote_read_cb);
13139         if (retval != 0) {
13140                 /*
13141                  * Make sure we free memory if there was an error..  The
13142                  * ctl_datamove_remote_xfer() function will send the
13143                  * datamove done message, or call the callback with an
13144                  * error if there is a problem.
13145                  */
13146                 for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13147                         free(io->io_hdr.local_sglist[i].addr, M_CTL);
13148                 free(io->io_hdr.remote_sglist, M_CTL);
13149                 io->io_hdr.remote_sglist = NULL;
13150                 io->io_hdr.local_sglist = NULL;
13151         }
13152 }
13153
13154 /*
13155  * Process a datamove request from the other controller.  This is used for
13156  * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
13157  * first.  Once that is complete, the data gets DMAed into the remote
13158  * controller's memory.  For reads, we DMA from the remote controller's
13159  * memory into our memory first, and then move it out to the FETD.
13160  */
13161 static void
13162 ctl_datamove_remote(union ctl_io *io)
13163 {
13164
13165         mtx_assert(&control_softc->ctl_lock, MA_NOTOWNED);
13166
13167         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
13168                 ctl_failover_io(io, /*have_lock*/ 0);
13169                 return;
13170         }
13171
13172         /*
13173          * Note that we look for an aborted I/O here, but don't do some of
13174          * the other checks that ctl_datamove() normally does.
13175          * We don't need to run the datamove delay code, since that should
13176          * have been done if need be on the other controller.
13177          */
13178         if (io->io_hdr.flags & CTL_FLAG_ABORT) {
13179                 printf("%s: tag 0x%04x on (%u:%u:%u) aborted\n", __func__,
13180                        io->scsiio.tag_num, io->io_hdr.nexus.initid,
13181                        io->io_hdr.nexus.targ_port,
13182                        io->io_hdr.nexus.targ_lun);
13183                 io->io_hdr.port_status = 31338;
13184                 ctl_send_datamove_done(io, /*have_lock*/ 0);
13185                 return;
13186         }
13187
13188         if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT)
13189                 ctl_datamove_remote_write(io);
13190         else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN)
13191                 ctl_datamove_remote_read(io);
13192         else {
13193                 io->io_hdr.port_status = 31339;
13194                 ctl_send_datamove_done(io, /*have_lock*/ 0);
13195         }
13196 }
13197
13198 static void
13199 ctl_process_done(union ctl_io *io)
13200 {
13201         struct ctl_lun *lun;
13202         struct ctl_softc *softc = control_softc;
13203         void (*fe_done)(union ctl_io *io);
13204         union ctl_ha_msg msg;
13205         uint32_t targ_port = io->io_hdr.nexus.targ_port;
13206
13207         CTL_DEBUG_PRINT(("ctl_process_done\n"));
13208         fe_done = softc->ctl_ports[targ_port]->fe_done;
13209
13210 #ifdef CTL_TIME_IO
13211         if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
13212                 char str[256];
13213                 char path_str[64];
13214                 struct sbuf sb;
13215
13216                 ctl_scsi_path_string(io, path_str, sizeof(path_str));
13217                 sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
13218
13219                 sbuf_cat(&sb, path_str);
13220                 switch (io->io_hdr.io_type) {
13221                 case CTL_IO_SCSI:
13222                         ctl_scsi_command_string(&io->scsiio, NULL, &sb);
13223                         sbuf_printf(&sb, "\n");
13224                         sbuf_cat(&sb, path_str);
13225                         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
13226                                     io->scsiio.tag_num, io->scsiio.tag_type);
13227                         break;
13228                 case CTL_IO_TASK:
13229                         sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
13230                                     "Tag Type: %d\n", io->taskio.task_action,
13231                                     io->taskio.tag_num, io->taskio.tag_type);
13232                         break;
13233                 default:
13234                         panic("%s: Invalid CTL I/O type %d\n",
13235                             __func__, io->io_hdr.io_type);
13236                 }
13237                 sbuf_cat(&sb, path_str);
13238                 sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
13239                             (intmax_t)time_uptime - io->io_hdr.start_time);
13240                 sbuf_finish(&sb);
13241                 printf("%s", sbuf_data(&sb));
13242         }
13243 #endif /* CTL_TIME_IO */
13244
13245         switch (io->io_hdr.io_type) {
13246         case CTL_IO_SCSI:
13247                 break;
13248         case CTL_IO_TASK:
13249                 if (ctl_debug & CTL_DEBUG_INFO)
13250                         ctl_io_error_print(io, NULL);
13251                 fe_done(io);
13252                 return;
13253         default:
13254                 panic("%s: Invalid CTL I/O type %d\n",
13255                     __func__, io->io_hdr.io_type);
13256         }
13257
13258         lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13259         if (lun == NULL) {
13260                 CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
13261                                  io->io_hdr.nexus.targ_mapped_lun));
13262                 goto bailout;
13263         }
13264
13265         mtx_lock(&lun->lun_lock);
13266
13267         /*
13268          * Check to see if we have any errors to inject here.  We only
13269          * inject errors for commands that don't already have errors set.
13270          */
13271         if (!STAILQ_EMPTY(&lun->error_list) &&
13272             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS) &&
13273             ((io->io_hdr.flags & CTL_FLAG_STATUS_SENT) == 0))
13274                 ctl_inject_error(lun, io);
13275
13276         /*
13277          * XXX KDM how do we treat commands that aren't completed
13278          * successfully?
13279          *
13280          * XXX KDM should we also track I/O latency?
13281          */
13282         if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS &&
13283             io->io_hdr.io_type == CTL_IO_SCSI) {
13284 #ifdef CTL_TIME_IO
13285                 struct bintime cur_bt;
13286 #endif
13287                 int type;
13288
13289                 if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13290                     CTL_FLAG_DATA_IN)
13291                         type = CTL_STATS_READ;
13292                 else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13293                     CTL_FLAG_DATA_OUT)
13294                         type = CTL_STATS_WRITE;
13295                 else
13296                         type = CTL_STATS_NO_IO;
13297
13298                 lun->stats.ports[targ_port].bytes[type] +=
13299                     io->scsiio.kern_total_len;
13300                 lun->stats.ports[targ_port].operations[type]++;
13301 #ifdef CTL_TIME_IO
13302                 bintime_add(&lun->stats.ports[targ_port].dma_time[type],
13303                    &io->io_hdr.dma_bt);
13304                 getbinuptime(&cur_bt);
13305                 bintime_sub(&cur_bt, &io->io_hdr.start_bt);
13306                 bintime_add(&lun->stats.ports[targ_port].time[type], &cur_bt);
13307 #endif
13308                 lun->stats.ports[targ_port].num_dmas[type] +=
13309                     io->io_hdr.num_dmas;
13310         }
13311
13312         /*
13313          * Remove this from the OOA queue.
13314          */
13315         TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
13316 #ifdef CTL_TIME_IO
13317         if (TAILQ_EMPTY(&lun->ooa_queue))
13318                 lun->last_busy = getsbinuptime();
13319 #endif
13320
13321         /*
13322          * Run through the blocked queue on this LUN and see if anything
13323          * has become unblocked, now that this transaction is done.
13324          */
13325         ctl_check_blocked(lun);
13326
13327         /*
13328          * If the LUN has been invalidated, free it if there is nothing
13329          * left on its OOA queue.
13330          */
13331         if ((lun->flags & CTL_LUN_INVALID)
13332          && TAILQ_EMPTY(&lun->ooa_queue)) {
13333                 mtx_unlock(&lun->lun_lock);
13334                 mtx_lock(&softc->ctl_lock);
13335                 ctl_free_lun(lun);
13336                 mtx_unlock(&softc->ctl_lock);
13337         } else
13338                 mtx_unlock(&lun->lun_lock);
13339
13340 bailout:
13341
13342         /*
13343          * If this command has been aborted, make sure we set the status
13344          * properly.  The FETD is responsible for freeing the I/O and doing
13345          * whatever it needs to do to clean up its state.
13346          */
13347         if (io->io_hdr.flags & CTL_FLAG_ABORT)
13348                 ctl_set_task_aborted(&io->scsiio);
13349
13350         /*
13351          * If enabled, print command error status.
13352          */
13353         if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS &&
13354             (ctl_debug & CTL_DEBUG_INFO) != 0)
13355                 ctl_io_error_print(io, NULL);
13356
13357         /*
13358          * Tell the FETD or the other shelf controller we're done with this
13359          * command.  Note that only SCSI commands get to this point.  Task
13360          * management commands are completed above.
13361          */
13362         if ((softc->ha_mode != CTL_HA_MODE_XFER) &&
13363             (io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)) {
13364                 memset(&msg, 0, sizeof(msg));
13365                 msg.hdr.msg_type = CTL_MSG_FINISH_IO;
13366                 msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
13367                 msg.hdr.nexus = io->io_hdr.nexus;
13368                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13369                     sizeof(msg.scsi) - sizeof(msg.scsi.sense_data),
13370                     M_WAITOK);
13371         }
13372
13373         fe_done(io);
13374 }
13375
13376 #ifdef CTL_WITH_CA
13377 /*
13378  * Front end should call this if it doesn't do autosense.  When the request
13379  * sense comes back in from the initiator, we'll dequeue this and send it.
13380  */
13381 int
13382 ctl_queue_sense(union ctl_io *io)
13383 {
13384         struct ctl_lun *lun;
13385         struct ctl_port *port;
13386         struct ctl_softc *softc;
13387         uint32_t initidx, targ_lun;
13388
13389         softc = control_softc;
13390
13391         CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
13392
13393         /*
13394          * LUN lookup will likely move to the ctl_work_thread() once we
13395          * have our new queueing infrastructure (that doesn't put things on
13396          * a per-LUN queue initially).  That is so that we can handle
13397          * things like an INQUIRY to a LUN that we don't have enabled.  We
13398          * can't deal with that right now.
13399          */
13400         mtx_lock(&softc->ctl_lock);
13401
13402         /*
13403          * If we don't have a LUN for this, just toss the sense
13404          * information.
13405          */
13406         port = ctl_io_port(&ctsio->io_hdr);
13407         targ_lun = ctl_lun_map_from_port(port, io->io_hdr.nexus.targ_lun);
13408         if ((targ_lun < CTL_MAX_LUNS)
13409          && (softc->ctl_luns[targ_lun] != NULL))
13410                 lun = softc->ctl_luns[targ_lun];
13411         else
13412                 goto bailout;
13413
13414         initidx = ctl_get_initindex(&io->io_hdr.nexus);
13415
13416         mtx_lock(&lun->lun_lock);
13417         /*
13418          * Already have CA set for this LUN...toss the sense information.
13419          */
13420         if (ctl_is_set(lun->have_ca, initidx)) {
13421                 mtx_unlock(&lun->lun_lock);
13422                 goto bailout;
13423         }
13424
13425         memcpy(&lun->pending_sense[initidx], &io->scsiio.sense_data,
13426                MIN(sizeof(lun->pending_sense[initidx]),
13427                sizeof(io->scsiio.sense_data)));
13428         ctl_set_mask(lun->have_ca, initidx);
13429         mtx_unlock(&lun->lun_lock);
13430
13431 bailout:
13432         mtx_unlock(&softc->ctl_lock);
13433
13434         ctl_free_io(io);
13435
13436         return (CTL_RETVAL_COMPLETE);
13437 }
13438 #endif
13439
13440 /*
13441  * Primary command inlet from frontend ports.  All SCSI and task I/O
13442  * requests must go through this function.
13443  */
13444 int
13445 ctl_queue(union ctl_io *io)
13446 {
13447         struct ctl_port *port;
13448
13449         CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
13450
13451 #ifdef CTL_TIME_IO
13452         io->io_hdr.start_time = time_uptime;
13453         getbinuptime(&io->io_hdr.start_bt);
13454 #endif /* CTL_TIME_IO */
13455
13456         /* Map FE-specific LUN ID into global one. */
13457         port = ctl_io_port(&io->io_hdr);
13458         io->io_hdr.nexus.targ_mapped_lun =
13459             ctl_lun_map_from_port(port, io->io_hdr.nexus.targ_lun);
13460
13461         switch (io->io_hdr.io_type) {
13462         case CTL_IO_SCSI:
13463         case CTL_IO_TASK:
13464                 if (ctl_debug & CTL_DEBUG_CDB)
13465                         ctl_io_print(io);
13466                 ctl_enqueue_incoming(io);
13467                 break;
13468         default:
13469                 printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
13470                 return (EINVAL);
13471         }
13472
13473         return (CTL_RETVAL_COMPLETE);
13474 }
13475
13476 #ifdef CTL_IO_DELAY
13477 static void
13478 ctl_done_timer_wakeup(void *arg)
13479 {
13480         union ctl_io *io;
13481
13482         io = (union ctl_io *)arg;
13483         ctl_done(io);
13484 }
13485 #endif /* CTL_IO_DELAY */
13486
13487 void
13488 ctl_serseq_done(union ctl_io *io)
13489 {
13490         struct ctl_lun *lun;
13491
13492         lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13493         if (lun->be_lun == NULL ||
13494             lun->be_lun->serseq == CTL_LUN_SERSEQ_OFF)
13495                 return;
13496         mtx_lock(&lun->lun_lock);
13497         io->io_hdr.flags |= CTL_FLAG_SERSEQ_DONE;
13498         ctl_check_blocked(lun);
13499         mtx_unlock(&lun->lun_lock);
13500 }
13501
13502 void
13503 ctl_done(union ctl_io *io)
13504 {
13505
13506         /*
13507          * Enable this to catch duplicate completion issues.
13508          */
13509 #if 0
13510         if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
13511                 printf("%s: type %d msg %d cdb %x iptl: "
13512                        "%u:%u:%u tag 0x%04x "
13513                        "flag %#x status %x\n",
13514                         __func__,
13515                         io->io_hdr.io_type,
13516                         io->io_hdr.msg_type,
13517                         io->scsiio.cdb[0],
13518                         io->io_hdr.nexus.initid,
13519                         io->io_hdr.nexus.targ_port,
13520                         io->io_hdr.nexus.targ_lun,
13521                         (io->io_hdr.io_type ==
13522                         CTL_IO_TASK) ?
13523                         io->taskio.tag_num :
13524                         io->scsiio.tag_num,
13525                         io->io_hdr.flags,
13526                         io->io_hdr.status);
13527         } else
13528                 io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
13529 #endif
13530
13531         /*
13532          * This is an internal copy of an I/O, and should not go through
13533          * the normal done processing logic.
13534          */
13535         if (io->io_hdr.flags & CTL_FLAG_INT_COPY)
13536                 return;
13537
13538 #ifdef CTL_IO_DELAY
13539         if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
13540                 struct ctl_lun *lun;
13541
13542                 lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13543
13544                 io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
13545         } else {
13546                 struct ctl_lun *lun;
13547
13548                 lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13549
13550                 if ((lun != NULL)
13551                  && (lun->delay_info.done_delay > 0)) {
13552
13553                         callout_init(&io->io_hdr.delay_callout, /*mpsafe*/ 1);
13554                         io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
13555                         callout_reset(&io->io_hdr.delay_callout,
13556                                       lun->delay_info.done_delay * hz,
13557                                       ctl_done_timer_wakeup, io);
13558                         if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
13559                                 lun->delay_info.done_delay = 0;
13560                         return;
13561                 }
13562         }
13563 #endif /* CTL_IO_DELAY */
13564
13565         ctl_enqueue_done(io);
13566 }
13567
13568 static void
13569 ctl_work_thread(void *arg)
13570 {
13571         struct ctl_thread *thr = (struct ctl_thread *)arg;
13572         struct ctl_softc *softc = thr->ctl_softc;
13573         union ctl_io *io;
13574         int retval;
13575
13576         CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
13577
13578         for (;;) {
13579                 /*
13580                  * We handle the queues in this order:
13581                  * - ISC
13582                  * - done queue (to free up resources, unblock other commands)
13583                  * - RtR queue
13584                  * - incoming queue
13585                  *
13586                  * If those queues are empty, we break out of the loop and
13587                  * go to sleep.
13588                  */
13589                 mtx_lock(&thr->queue_lock);
13590                 io = (union ctl_io *)STAILQ_FIRST(&thr->isc_queue);
13591                 if (io != NULL) {
13592                         STAILQ_REMOVE_HEAD(&thr->isc_queue, links);
13593                         mtx_unlock(&thr->queue_lock);
13594                         ctl_handle_isc(io);
13595                         continue;
13596                 }
13597                 io = (union ctl_io *)STAILQ_FIRST(&thr->done_queue);
13598                 if (io != NULL) {
13599                         STAILQ_REMOVE_HEAD(&thr->done_queue, links);
13600                         /* clear any blocked commands, call fe_done */
13601                         mtx_unlock(&thr->queue_lock);
13602                         ctl_process_done(io);
13603                         continue;
13604                 }
13605                 io = (union ctl_io *)STAILQ_FIRST(&thr->incoming_queue);
13606                 if (io != NULL) {
13607                         STAILQ_REMOVE_HEAD(&thr->incoming_queue, links);
13608                         mtx_unlock(&thr->queue_lock);
13609                         if (io->io_hdr.io_type == CTL_IO_TASK)
13610                                 ctl_run_task(io);
13611                         else
13612                                 ctl_scsiio_precheck(softc, &io->scsiio);
13613                         continue;
13614                 }
13615                 io = (union ctl_io *)STAILQ_FIRST(&thr->rtr_queue);
13616                 if (io != NULL) {
13617                         STAILQ_REMOVE_HEAD(&thr->rtr_queue, links);
13618                         mtx_unlock(&thr->queue_lock);
13619                         retval = ctl_scsiio(&io->scsiio);
13620                         if (retval != CTL_RETVAL_COMPLETE)
13621                                 CTL_DEBUG_PRINT(("ctl_scsiio failed\n"));
13622                         continue;
13623                 }
13624
13625                 /* Sleep until we have something to do. */
13626                 mtx_sleep(thr, &thr->queue_lock, PDROP | PRIBIO, "-", 0);
13627         }
13628 }
13629
13630 static void
13631 ctl_lun_thread(void *arg)
13632 {
13633         struct ctl_softc *softc = (struct ctl_softc *)arg;
13634         struct ctl_be_lun *be_lun;
13635
13636         CTL_DEBUG_PRINT(("ctl_lun_thread starting\n"));
13637
13638         for (;;) {
13639                 mtx_lock(&softc->ctl_lock);
13640                 be_lun = STAILQ_FIRST(&softc->pending_lun_queue);
13641                 if (be_lun != NULL) {
13642                         STAILQ_REMOVE_HEAD(&softc->pending_lun_queue, links);
13643                         mtx_unlock(&softc->ctl_lock);
13644                         ctl_create_lun(be_lun);
13645                         continue;
13646                 }
13647
13648                 /* Sleep until we have something to do. */
13649                 mtx_sleep(&softc->pending_lun_queue, &softc->ctl_lock,
13650                     PDROP | PRIBIO, "-", 0);
13651         }
13652 }
13653
13654 static void
13655 ctl_thresh_thread(void *arg)
13656 {
13657         struct ctl_softc *softc = (struct ctl_softc *)arg;
13658         struct ctl_lun *lun;
13659         struct scsi_da_rw_recovery_page *rwpage;
13660         struct ctl_logical_block_provisioning_page *page;
13661         const char *attr;
13662         union ctl_ha_msg msg;
13663         uint64_t thres, val;
13664         int i, e, set;
13665
13666         CTL_DEBUG_PRINT(("ctl_thresh_thread starting\n"));
13667
13668         for (;;) {
13669                 mtx_lock(&softc->ctl_lock);
13670                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
13671                         if ((lun->flags & CTL_LUN_DISABLED) ||
13672                             (lun->flags & CTL_LUN_NO_MEDIA) ||
13673                             lun->backend->lun_attr == NULL)
13674                                 continue;
13675                         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
13676                             softc->ha_mode == CTL_HA_MODE_XFER)
13677                                 continue;
13678                         rwpage = &lun->mode_pages.rw_er_page[CTL_PAGE_CURRENT];
13679                         if ((rwpage->byte8 & SMS_RWER_LBPERE) == 0)
13680                                 continue;
13681                         e = 0;
13682                         page = &lun->mode_pages.lbp_page[CTL_PAGE_CURRENT];
13683                         for (i = 0; i < CTL_NUM_LBP_THRESH; i++) {
13684                                 if ((page->descr[i].flags & SLBPPD_ENABLED) == 0)
13685                                         continue;
13686                                 thres = scsi_4btoul(page->descr[i].count);
13687                                 thres <<= CTL_LBP_EXPONENT;
13688                                 switch (page->descr[i].resource) {
13689                                 case 0x01:
13690                                         attr = "blocksavail";
13691                                         break;
13692                                 case 0x02:
13693                                         attr = "blocksused";
13694                                         break;
13695                                 case 0xf1:
13696                                         attr = "poolblocksavail";
13697                                         break;
13698                                 case 0xf2:
13699                                         attr = "poolblocksused";
13700                                         break;
13701                                 default:
13702                                         continue;
13703                                 }
13704                                 mtx_unlock(&softc->ctl_lock); // XXX
13705                                 val = lun->backend->lun_attr(
13706                                     lun->be_lun->be_lun, attr);
13707                                 mtx_lock(&softc->ctl_lock);
13708                                 if (val == UINT64_MAX)
13709                                         continue;
13710                                 if ((page->descr[i].flags & SLBPPD_ARMING_MASK)
13711                                     == SLBPPD_ARMING_INC)
13712                                         e = (val >= thres);
13713                                 else
13714                                         e = (val <= thres);
13715                                 if (e)
13716                                         break;
13717                         }
13718                         mtx_lock(&lun->lun_lock);
13719                         if (e) {
13720                                 scsi_u64to8b((uint8_t *)&page->descr[i] -
13721                                     (uint8_t *)page, lun->ua_tpt_info);
13722                                 if (lun->lasttpt == 0 ||
13723                                     time_uptime - lun->lasttpt >= CTL_LBP_UA_PERIOD) {
13724                                         lun->lasttpt = time_uptime;
13725                                         ctl_est_ua_all(lun, -1, CTL_UA_THIN_PROV_THRES);
13726                                         set = 1;
13727                                 } else
13728                                         set = 0;
13729                         } else {
13730                                 lun->lasttpt = 0;
13731                                 ctl_clr_ua_all(lun, -1, CTL_UA_THIN_PROV_THRES);
13732                                 set = -1;
13733                         }
13734                         mtx_unlock(&lun->lun_lock);
13735                         if (set != 0 &&
13736                             lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
13737                                 /* Send msg to other side. */
13738                                 bzero(&msg.ua, sizeof(msg.ua));
13739                                 msg.hdr.msg_type = CTL_MSG_UA;
13740                                 msg.hdr.nexus.initid = -1;
13741                                 msg.hdr.nexus.targ_port = -1;
13742                                 msg.hdr.nexus.targ_lun = lun->lun;
13743                                 msg.hdr.nexus.targ_mapped_lun = lun->lun;
13744                                 msg.ua.ua_all = 1;
13745                                 msg.ua.ua_set = (set > 0);
13746                                 msg.ua.ua_type = CTL_UA_THIN_PROV_THRES;
13747                                 memcpy(msg.ua.ua_info, lun->ua_tpt_info, 8);
13748                                 mtx_unlock(&softc->ctl_lock); // XXX
13749                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13750                                     sizeof(msg.ua), M_WAITOK);
13751                                 mtx_lock(&softc->ctl_lock);
13752                         }
13753                 }
13754                 mtx_unlock(&softc->ctl_lock);
13755                 pause("-", CTL_LBP_PERIOD * hz);
13756         }
13757 }
13758
13759 static void
13760 ctl_enqueue_incoming(union ctl_io *io)
13761 {
13762         struct ctl_softc *softc = control_softc;
13763         struct ctl_thread *thr;
13764         u_int idx;
13765
13766         idx = (io->io_hdr.nexus.targ_port * 127 +
13767                io->io_hdr.nexus.initid) % worker_threads;
13768         thr = &softc->threads[idx];
13769         mtx_lock(&thr->queue_lock);
13770         STAILQ_INSERT_TAIL(&thr->incoming_queue, &io->io_hdr, links);
13771         mtx_unlock(&thr->queue_lock);
13772         wakeup(thr);
13773 }
13774
13775 static void
13776 ctl_enqueue_rtr(union ctl_io *io)
13777 {
13778         struct ctl_softc *softc = control_softc;
13779         struct ctl_thread *thr;
13780
13781         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13782         mtx_lock(&thr->queue_lock);
13783         STAILQ_INSERT_TAIL(&thr->rtr_queue, &io->io_hdr, links);
13784         mtx_unlock(&thr->queue_lock);
13785         wakeup(thr);
13786 }
13787
13788 static void
13789 ctl_enqueue_done(union ctl_io *io)
13790 {
13791         struct ctl_softc *softc = control_softc;
13792         struct ctl_thread *thr;
13793
13794         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13795         mtx_lock(&thr->queue_lock);
13796         STAILQ_INSERT_TAIL(&thr->done_queue, &io->io_hdr, links);
13797         mtx_unlock(&thr->queue_lock);
13798         wakeup(thr);
13799 }
13800
13801 static void
13802 ctl_enqueue_isc(union ctl_io *io)
13803 {
13804         struct ctl_softc *softc = control_softc;
13805         struct ctl_thread *thr;
13806
13807         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13808         mtx_lock(&thr->queue_lock);
13809         STAILQ_INSERT_TAIL(&thr->isc_queue, &io->io_hdr, links);
13810         mtx_unlock(&thr->queue_lock);
13811         wakeup(thr);
13812 }
13813
13814 /*
13815  *  vim: ts=8
13816  */