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_da.h>
  73 #include <cam/ctl/ctl_io.h>
  74 #include <cam/ctl/ctl.h>
  75 #include <cam/ctl/ctl_frontend.h>
  76 #include <cam/ctl/ctl_util.h>
  77 #include <cam/ctl/ctl_backend.h>
  78 #include <cam/ctl/ctl_ioctl.h>
  79 #include <cam/ctl/ctl_ha.h>
  80 #include <cam/ctl/ctl_private.h>
  81 #include <cam/ctl/ctl_debug.h>
  82 #include <cam/ctl/ctl_scsi_all.h>
  83 #include <cam/ctl/ctl_error.h>
  84
  85 struct ctl_softc *control_softc = NULL;
  86
  87 /*
  88  * Template mode pages.
  89  */
  90
  91 /*
  92  * Note that these are default values only.  The actual values will be
  93  * filled in when the user does a mode sense.
  94  */
  95 const static struct copan_debugconf_subpage debugconf_page_default = {
  96         DBGCNF_PAGE_CODE | SMPH_SPF,    /* page_code */
  97         DBGCNF_SUBPAGE_CODE,            /* subpage */
  98         {(sizeof(struct copan_debugconf_subpage) - 4) >> 8,
  99          (sizeof(struct copan_debugconf_subpage) - 4) >> 0}, /* page_length */
 100         DBGCNF_VERSION,                 /* page_version */
 101         {CTL_TIME_IO_DEFAULT_SECS>>8,
 102          CTL_TIME_IO_DEFAULT_SECS>>0},  /* ctl_time_io_secs */
 103 };
 104
 105 const static struct copan_debugconf_subpage debugconf_page_changeable = {
 106         DBGCNF_PAGE_CODE | SMPH_SPF,    /* page_code */
 107         DBGCNF_SUBPAGE_CODE,            /* subpage */
 108         {(sizeof(struct copan_debugconf_subpage) - 4) >> 8,
 109          (sizeof(struct copan_debugconf_subpage) - 4) >> 0}, /* page_length */
 110         0,                              /* page_version */
 111         {0xff,0xff},                    /* ctl_time_io_secs */
 112 };
 113
 114 const static struct scsi_da_rw_recovery_page rw_er_page_default = {
 115         /*page_code*/SMS_RW_ERROR_RECOVERY_PAGE,
 116         /*page_length*/sizeof(struct scsi_da_rw_recovery_page) - 2,
 117         /*byte3*/SMS_RWER_AWRE|SMS_RWER_ARRE,
 118         /*read_retry_count*/0,
 119         /*correction_span*/0,
 120         /*head_offset_count*/0,
 121         /*data_strobe_offset_cnt*/0,
 122         /*byte8*/SMS_RWER_LBPERE,
 123         /*write_retry_count*/0,
 124         /*reserved2*/0,
 125         /*recovery_time_limit*/{0, 0},
 126 };
 127
 128 const static struct scsi_da_rw_recovery_page rw_er_page_changeable = {
 129         /*page_code*/SMS_RW_ERROR_RECOVERY_PAGE,
 130         /*page_length*/sizeof(struct scsi_da_rw_recovery_page) - 2,
 131         /*byte3*/0,
 132         /*read_retry_count*/0,
 133         /*correction_span*/0,
 134         /*head_offset_count*/0,
 135         /*data_strobe_offset_cnt*/0,
 136         /*byte8*/0,
 137         /*write_retry_count*/0,
 138         /*reserved2*/0,
 139         /*recovery_time_limit*/{0, 0},
 140 };
 141
 142 const static struct scsi_format_page format_page_default = {
 143         /*page_code*/SMS_FORMAT_DEVICE_PAGE,
 144         /*page_length*/sizeof(struct scsi_format_page) - 2,
 145         /*tracks_per_zone*/ {0, 0},
 146         /*alt_sectors_per_zone*/ {0, 0},
 147         /*alt_tracks_per_zone*/ {0, 0},
 148         /*alt_tracks_per_lun*/ {0, 0},
 149         /*sectors_per_track*/ {(CTL_DEFAULT_SECTORS_PER_TRACK >> 8) & 0xff,
 150                                 CTL_DEFAULT_SECTORS_PER_TRACK & 0xff},
 151         /*bytes_per_sector*/ {0, 0},
 152         /*interleave*/ {0, 0},
 153         /*track_skew*/ {0, 0},
 154         /*cylinder_skew*/ {0, 0},
 155         /*flags*/ SFP_HSEC,
 156         /*reserved*/ {0, 0, 0}
 157 };
 158
 159 const static struct scsi_format_page format_page_changeable = {
 160         /*page_code*/SMS_FORMAT_DEVICE_PAGE,
 161         /*page_length*/sizeof(struct scsi_format_page) - 2,
 162         /*tracks_per_zone*/ {0, 0},
 163         /*alt_sectors_per_zone*/ {0, 0},
 164         /*alt_tracks_per_zone*/ {0, 0},
 165         /*alt_tracks_per_lun*/ {0, 0},
 166         /*sectors_per_track*/ {0, 0},
 167         /*bytes_per_sector*/ {0, 0},
 168         /*interleave*/ {0, 0},
 169         /*track_skew*/ {0, 0},
 170         /*cylinder_skew*/ {0, 0},
 171         /*flags*/ 0,
 172         /*reserved*/ {0, 0, 0}
 173 };
 174
 175 const static struct scsi_rigid_disk_page rigid_disk_page_default = {
 176         /*page_code*/SMS_RIGID_DISK_PAGE,
 177         /*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
 178         /*cylinders*/ {0, 0, 0},
 179         /*heads*/ CTL_DEFAULT_HEADS,
 180         /*start_write_precomp*/ {0, 0, 0},
 181         /*start_reduced_current*/ {0, 0, 0},
 182         /*step_rate*/ {0, 0},
 183         /*landing_zone_cylinder*/ {0, 0, 0},
 184         /*rpl*/ SRDP_RPL_DISABLED,
 185         /*rotational_offset*/ 0,
 186         /*reserved1*/ 0,
 187         /*rotation_rate*/ {(CTL_DEFAULT_ROTATION_RATE >> 8) & 0xff,
 188                            CTL_DEFAULT_ROTATION_RATE & 0xff},
 189         /*reserved2*/ {0, 0}
 190 };
 191
 192 const static struct scsi_rigid_disk_page rigid_disk_page_changeable = {
 193         /*page_code*/SMS_RIGID_DISK_PAGE,
 194         /*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
 195         /*cylinders*/ {0, 0, 0},
 196         /*heads*/ 0,
 197         /*start_write_precomp*/ {0, 0, 0},
 198         /*start_reduced_current*/ {0, 0, 0},
 199         /*step_rate*/ {0, 0},
 200         /*landing_zone_cylinder*/ {0, 0, 0},
 201         /*rpl*/ 0,
 202         /*rotational_offset*/ 0,
 203         /*reserved1*/ 0,
 204         /*rotation_rate*/ {0, 0},
 205         /*reserved2*/ {0, 0}
 206 };
 207
 208 const static struct scsi_caching_page caching_page_default = {
 209         /*page_code*/SMS_CACHING_PAGE,
 210         /*page_length*/sizeof(struct scsi_caching_page) - 2,
 211         /*flags1*/ SCP_DISC | SCP_WCE,
 212         /*ret_priority*/ 0,
 213         /*disable_pf_transfer_len*/ {0xff, 0xff},
 214         /*min_prefetch*/ {0, 0},
 215         /*max_prefetch*/ {0xff, 0xff},
 216         /*max_pf_ceiling*/ {0xff, 0xff},
 217         /*flags2*/ 0,
 218         /*cache_segments*/ 0,
 219         /*cache_seg_size*/ {0, 0},
 220         /*reserved*/ 0,
 221         /*non_cache_seg_size*/ {0, 0, 0}
 222 };
 223
 224 const static struct scsi_caching_page caching_page_changeable = {
 225         /*page_code*/SMS_CACHING_PAGE,
 226         /*page_length*/sizeof(struct scsi_caching_page) - 2,
 227         /*flags1*/ SCP_WCE | SCP_RCD,
 228         /*ret_priority*/ 0,
 229         /*disable_pf_transfer_len*/ {0, 0},
 230         /*min_prefetch*/ {0, 0},
 231         /*max_prefetch*/ {0, 0},
 232         /*max_pf_ceiling*/ {0, 0},
 233         /*flags2*/ 0,
 234         /*cache_segments*/ 0,
 235         /*cache_seg_size*/ {0, 0},
 236         /*reserved*/ 0,
 237         /*non_cache_seg_size*/ {0, 0, 0}
 238 };
 239
 240 const static struct scsi_control_page control_page_default = {
 241         /*page_code*/SMS_CONTROL_MODE_PAGE,
 242         /*page_length*/sizeof(struct scsi_control_page) - 2,
 243         /*rlec*/0,
 244         /*queue_flags*/SCP_QUEUE_ALG_RESTRICTED,
 245         /*eca_and_aen*/0,
 246         /*flags4*/SCP_TAS,
 247         /*aen_holdoff_period*/{0, 0},
 248         /*busy_timeout_period*/{0, 0},
 249         /*extended_selftest_completion_time*/{0, 0}
 250 };
 251
 252 const static struct scsi_control_page control_page_changeable = {
 253         /*page_code*/SMS_CONTROL_MODE_PAGE,
 254         /*page_length*/sizeof(struct scsi_control_page) - 2,
 255         /*rlec*/SCP_DSENSE,
 256         /*queue_flags*/SCP_QUEUE_ALG_MASK,
 257         /*eca_and_aen*/SCP_SWP,
 258         /*flags4*/0,
 259         /*aen_holdoff_period*/{0, 0},
 260         /*busy_timeout_period*/{0, 0},
 261         /*extended_selftest_completion_time*/{0, 0}
 262 };
 263
 264 #define CTL_CEM_LEN     (sizeof(struct scsi_control_ext_page) - 4)
 265
 266 const static struct scsi_control_ext_page control_ext_page_default = {
 267         /*page_code*/SMS_CONTROL_MODE_PAGE | SMPH_SPF,
 268         /*subpage_code*/0x01,
 269         /*page_length*/{CTL_CEM_LEN >> 8, CTL_CEM_LEN},
 270         /*flags*/0,
 271         /*prio*/0,
 272         /*max_sense*/0
 273 };
 274
 275 const static struct scsi_control_ext_page control_ext_page_changeable = {
 276         /*page_code*/SMS_CONTROL_MODE_PAGE | SMPH_SPF,
 277         /*subpage_code*/0x01,
 278         /*page_length*/{CTL_CEM_LEN >> 8, CTL_CEM_LEN},
 279         /*flags*/0,
 280         /*prio*/0,
 281         /*max_sense*/0
 282 };
 283
 284 const static struct scsi_info_exceptions_page ie_page_default = {
 285         /*page_code*/SMS_INFO_EXCEPTIONS_PAGE,
 286         /*page_length*/sizeof(struct scsi_info_exceptions_page) - 2,
 287         /*info_flags*/SIEP_FLAGS_DEXCPT,
 288         /*mrie*/0,
 289         /*interval_timer*/{0, 0, 0, 0},
 290         /*report_count*/{0, 0, 0, 0}
 291 };
 292
 293 const static struct scsi_info_exceptions_page ie_page_changeable = {
 294         /*page_code*/SMS_INFO_EXCEPTIONS_PAGE,
 295         /*page_length*/sizeof(struct scsi_info_exceptions_page) - 2,
 296         /*info_flags*/0,
 297         /*mrie*/0,
 298         /*interval_timer*/{0, 0, 0, 0},
 299         /*report_count*/{0, 0, 0, 0}
 300 };
 301
 302 #define CTL_LBPM_LEN    (sizeof(struct ctl_logical_block_provisioning_page) - 4)
 303
 304 const static struct ctl_logical_block_provisioning_page lbp_page_default = {{
 305         /*page_code*/SMS_INFO_EXCEPTIONS_PAGE | SMPH_SPF,
 306         /*subpage_code*/0x02,
 307         /*page_length*/{CTL_LBPM_LEN >> 8, CTL_LBPM_LEN},
 308         /*flags*/0,
 309         /*reserved*/{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
 310         /*descr*/{}},
 311         {{/*flags*/0,
 312           /*resource*/0x01,
 313           /*reserved*/{0, 0},
 314           /*count*/{0, 0, 0, 0}},
 315          {/*flags*/0,
 316           /*resource*/0x02,
 317           /*reserved*/{0, 0},
 318           /*count*/{0, 0, 0, 0}},
 319          {/*flags*/0,
 320           /*resource*/0xf1,
 321           /*reserved*/{0, 0},
 322           /*count*/{0, 0, 0, 0}},
 323          {/*flags*/0,
 324           /*resource*/0xf2,
 325           /*reserved*/{0, 0},
 326           /*count*/{0, 0, 0, 0}}
 327         }
 328 };
 329
 330 const static struct ctl_logical_block_provisioning_page lbp_page_changeable = {{
 331         /*page_code*/SMS_INFO_EXCEPTIONS_PAGE | SMPH_SPF,
 332         /*subpage_code*/0x02,
 333         /*page_length*/{CTL_LBPM_LEN >> 8, CTL_LBPM_LEN},
 334         /*flags*/0,
 335         /*reserved*/{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
 336         /*descr*/{}},
 337         {{/*flags*/0,
 338           /*resource*/0,
 339           /*reserved*/{0, 0},
 340           /*count*/{0, 0, 0, 0}},
 341          {/*flags*/0,
 342           /*resource*/0,
 343           /*reserved*/{0, 0},
 344           /*count*/{0, 0, 0, 0}},
 345          {/*flags*/0,
 346           /*resource*/0,
 347           /*reserved*/{0, 0},
 348           /*count*/{0, 0, 0, 0}},
 349          {/*flags*/0,
 350           /*resource*/0,
 351           /*reserved*/{0, 0},
 352           /*count*/{0, 0, 0, 0}}
 353         }
 354 };
 355
 356 SYSCTL_NODE(_kern_cam, OID_AUTO, ctl, CTLFLAG_RD, 0, "CAM Target Layer");
 357 static int worker_threads = -1;
 358 TUNABLE_INT("kern.cam.ctl.worker_threads", &worker_threads);
 359 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, worker_threads, CTLFLAG_RDTUN,
 360     &worker_threads, 1, "Number of worker threads");
 361 static int ctl_debug = CTL_DEBUG_NONE;
 362 TUNABLE_INT("kern.cam.ctl.debug", &ctl_debug);
 363 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, debug, CTLFLAG_RWTUN,
 364     &ctl_debug, 0, "Enabled debug flags");
 365
 366 /*
 367  * Supported pages (0x00), Serial number (0x80), Device ID (0x83),
 368  * Extended INQUIRY Data (0x86), Mode Page Policy (0x87),
 369  * SCSI Ports (0x88), Third-party Copy (0x8F), Block limits (0xB0),
 370  * Block Device Characteristics (0xB1) and Logical Block Provisioning (0xB2)
 371  */
 372 #define SCSI_EVPD_NUM_SUPPORTED_PAGES   10
 373
 374 static void ctl_isc_event_handler(ctl_ha_channel chanel, ctl_ha_event event,
 375                                   int param);
 376 static void ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest);
 377 static void ctl_copy_sense_data_back(union ctl_io *src, union ctl_ha_msg *dest);
 378 static int ctl_init(void);
 379 void ctl_shutdown(void);
 380 static int ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td);
 381 static int ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td);
 382 static int ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio);
 383 static void ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
 384                               struct ctl_ooa *ooa_hdr,
 385                               struct ctl_ooa_entry *kern_entries);
 386 static int ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
 387                      struct thread *td);
 388 static int ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
 389                          struct ctl_be_lun *be_lun);
 390 static int ctl_free_lun(struct ctl_lun *lun);
 391 static void ctl_create_lun(struct ctl_be_lun *be_lun);
 392 static struct ctl_port * ctl_io_port(struct ctl_io_hdr *io_hdr);
 393
 394 static int ctl_do_mode_select(union ctl_io *io);
 395 static int ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun,
 396                            uint64_t res_key, uint64_t sa_res_key,
 397                            uint8_t type, uint32_t residx,
 398                            struct ctl_scsiio *ctsio,
 399                            struct scsi_per_res_out *cdb,
 400                            struct scsi_per_res_out_parms* param);
 401 static void ctl_pro_preempt_other(struct ctl_lun *lun,
 402                                   union ctl_ha_msg *msg);
 403 static void ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg);
 404 static int ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len);
 405 static int ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len);
 406 static int ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len);
 407 static int ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len);
 408 static int ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len);
 409 static int ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio,
 410                                          int alloc_len);
 411 static int ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio,
 412                                          int alloc_len);
 413 static int ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len);
 414 static int ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len);
 415 static int ctl_inquiry_evpd(struct ctl_scsiio *ctsio);
 416 static int ctl_inquiry_std(struct ctl_scsiio *ctsio);
 417 static int ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len);
 418 static ctl_action ctl_extent_check(union ctl_io *io1, union ctl_io *io2,
 419     bool seq);
 420 static ctl_action ctl_extent_check_seq(union ctl_io *io1, union ctl_io *io2);
 421 static ctl_action ctl_check_for_blockage(struct ctl_lun *lun,
 422     union ctl_io *pending_io, union ctl_io *ooa_io);
 423 static ctl_action ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
 424                                 union ctl_io *starting_io);
 425 static int ctl_check_blocked(struct ctl_lun *lun);
 426 static int ctl_scsiio_lun_check(struct ctl_lun *lun,
 427                                 const struct ctl_cmd_entry *entry,
 428                                 struct ctl_scsiio *ctsio);
 429 static void ctl_failover_lun(union ctl_io *io);
 430 static int ctl_scsiio_precheck(struct ctl_softc *ctl_softc,
 431                                struct ctl_scsiio *ctsio);
 432 static int ctl_scsiio(struct ctl_scsiio *ctsio);
 433
 434 static int ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io);
 435 static int ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
 436                             ctl_ua_type ua_type);
 437 static int ctl_do_lun_reset(struct ctl_lun *lun, union ctl_io *io,
 438                          ctl_ua_type ua_type);
 439 static int ctl_lun_reset(struct ctl_softc *ctl_softc, union ctl_io *io);
 440 static int ctl_abort_task(union ctl_io *io);
 441 static int ctl_abort_task_set(union ctl_io *io);
 442 static int ctl_query_task(union ctl_io *io, int task_set);
 443 static int ctl_i_t_nexus_reset(union ctl_io *io);
 444 static int ctl_query_async_event(union ctl_io *io);
 445 static void ctl_run_task(union ctl_io *io);
 446 #ifdef CTL_IO_DELAY
 447 static void ctl_datamove_timer_wakeup(void *arg);
 448 static void ctl_done_timer_wakeup(void *arg);
 449 #endif /* CTL_IO_DELAY */
 450
 451 static void ctl_send_datamove_done(union ctl_io *io, int have_lock);
 452 static void ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq);
 453 static int ctl_datamove_remote_dm_write_cb(union ctl_io *io);
 454 static void ctl_datamove_remote_write(union ctl_io *io);
 455 static int ctl_datamove_remote_dm_read_cb(union ctl_io *io);
 456 static void ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq);
 457 static int ctl_datamove_remote_sgl_setup(union ctl_io *io);
 458 static int ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
 459                                     ctl_ha_dt_cb callback);
 460 static void ctl_datamove_remote_read(union ctl_io *io);
 461 static void ctl_datamove_remote(union ctl_io *io);
 462 static void ctl_process_done(union ctl_io *io);
 463 static void ctl_lun_thread(void *arg);
 464 static void ctl_thresh_thread(void *arg);
 465 static void ctl_work_thread(void *arg);
 466 static void ctl_enqueue_incoming(union ctl_io *io);
 467 static void ctl_enqueue_rtr(union ctl_io *io);
 468 static void ctl_enqueue_done(union ctl_io *io);
 469 static void ctl_enqueue_isc(union ctl_io *io);
 470 static const struct ctl_cmd_entry *
 471     ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa);
 472 static const struct ctl_cmd_entry *
 473     ctl_validate_command(struct ctl_scsiio *ctsio);
 474 static int ctl_cmd_applicable(uint8_t lun_type,
 475     const struct ctl_cmd_entry *entry);
 476
 477 static uint64_t ctl_get_prkey(struct ctl_lun *lun, uint32_t residx);
 478 static void ctl_clr_prkey(struct ctl_lun *lun, uint32_t residx);
 479 static void ctl_alloc_prkey(struct ctl_lun *lun, uint32_t residx);
 480 static void ctl_set_prkey(struct ctl_lun *lun, uint32_t residx, uint64_t key);
 481
 482 /*
 483  * Load the serialization table.  This isn't very pretty, but is probably
 484  * the easiest way to do it.
 485  */
 486 #include "ctl_ser_table.c"
 487
 488 /*
 489  * We only need to define open, close and ioctl routines for this driver.
 490  */
 491 static struct cdevsw ctl_cdevsw = {
 492         .d_version =    D_VERSION,
 493         .d_flags =      0,
 494         .d_open =       ctl_open,
 495         .d_close =      ctl_close,
 496         .d_ioctl =      ctl_ioctl,
 497         .d_name =       "ctl",
 498 };
 499
 500
 501 MALLOC_DEFINE(M_CTL, "ctlmem", "Memory used for CTL");
 502
 503 static int ctl_module_event_handler(module_t, int /*modeventtype_t*/, void *);
 504
 505 static moduledata_t ctl_moduledata = {
 506         "ctl",
 507         ctl_module_event_handler,
 508         NULL
 509 };
 510
 511 DECLARE_MODULE(ctl, ctl_moduledata, SI_SUB_CONFIGURE, SI_ORDER_THIRD);
 512 MODULE_VERSION(ctl, 1);
 513
 514 static struct ctl_frontend ha_frontend =
 515 {
 516         .name = "ha",
 517 };
 518
 519 static void
 520 ctl_ha_datamove(union ctl_io *io)
 521 {
 522         struct ctl_lun *lun;
 523         struct ctl_sg_entry *sgl;
 524         union ctl_ha_msg msg;
 525         uint32_t sg_entries_sent;
 526         int do_sg_copy, i, j;
 527
 528         lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
 529         memset(&msg.dt, 0, sizeof(msg.dt));
 530         msg.hdr.msg_type = CTL_MSG_DATAMOVE;
 531         msg.hdr.original_sc = io->io_hdr.original_sc;
 532         msg.hdr.serializing_sc = io;
 533         msg.hdr.nexus = io->io_hdr.nexus;
 534         msg.hdr.status = io->io_hdr.status;
 535         msg.dt.flags = io->io_hdr.flags;
 536
 537         /*
 538          * We convert everything into a S/G list here.  We can't
 539          * pass by reference, only by value between controllers.
 540          * So we can't pass a pointer to the S/G list, only as many
 541          * S/G entries as we can fit in here.  If it's possible for
 542          * us to get more than CTL_HA_MAX_SG_ENTRIES S/G entries,
 543          * then we need to break this up into multiple transfers.
 544          */
 545         if (io->scsiio.kern_sg_entries == 0) {
 546                 msg.dt.kern_sg_entries = 1;
 547 #if 0
 548                 if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
 549                         msg.dt.sg_list[0].addr = io->scsiio.kern_data_ptr;
 550                 } else {
 551                         /* XXX KDM use busdma here! */
 552                         msg.dt.sg_list[0].addr =
 553                             (void *)vtophys(io->scsiio.kern_data_ptr);
 554                 }
 555 #else
 556                 KASSERT((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0,
 557                     ("HA does not support BUS_ADDR"));
 558                 msg.dt.sg_list[0].addr = io->scsiio.kern_data_ptr;
 559 #endif
 560                 msg.dt.sg_list[0].len = io->scsiio.kern_data_len;
 561                 do_sg_copy = 0;
 562         } else {
 563                 msg.dt.kern_sg_entries = io->scsiio.kern_sg_entries;
 564                 do_sg_copy = 1;
 565         }
 566
 567         msg.dt.kern_data_len = io->scsiio.kern_data_len;
 568         msg.dt.kern_total_len = io->scsiio.kern_total_len;
 569         msg.dt.kern_data_resid = io->scsiio.kern_data_resid;
 570         msg.dt.kern_rel_offset = io->scsiio.kern_rel_offset;
 571         msg.dt.sg_sequence = 0;
 572
 573         /*
 574          * Loop until we've sent all of the S/G entries.  On the
 575          * other end, we'll recompose these S/G entries into one
 576          * contiguous list before processing.
 577          */
 578         for (sg_entries_sent = 0; sg_entries_sent < msg.dt.kern_sg_entries;
 579             msg.dt.sg_sequence++) {
 580                 msg.dt.cur_sg_entries = MIN((sizeof(msg.dt.sg_list) /
 581                     sizeof(msg.dt.sg_list[0])),
 582                     msg.dt.kern_sg_entries - sg_entries_sent);
 583                 if (do_sg_copy != 0) {
 584                         sgl = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr;
 585                         for (i = sg_entries_sent, j = 0;
 586                              i < msg.dt.cur_sg_entries; i++, j++) {
 587 #if 0
 588                                 if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
 589                                         msg.dt.sg_list[j].addr = sgl[i].addr;
 590                                 } else {
 591                                         /* XXX KDM use busdma here! */
 592                                         msg.dt.sg_list[j].addr =
 593                                             (void *)vtophys(sgl[i].addr);
 594                                 }
 595 #else
 596                                 KASSERT((io->io_hdr.flags &
 597                                     CTL_FLAG_BUS_ADDR) == 0,
 598                                     ("HA does not support BUS_ADDR"));
 599                                 msg.dt.sg_list[j].addr = sgl[i].addr;
 600 #endif
 601                                 msg.dt.sg_list[j].len = sgl[i].len;
 602                         }
 603                 }
 604
 605                 sg_entries_sent += msg.dt.cur_sg_entries;
 606                 msg.dt.sg_last = (sg_entries_sent >= msg.dt.kern_sg_entries);
 607                 if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
 608                     sizeof(msg.dt) - sizeof(msg.dt.sg_list) +
 609                     sizeof(struct ctl_sg_entry) * msg.dt.cur_sg_entries,
 610                     M_WAITOK) > CTL_HA_STATUS_SUCCESS) {
 611                         io->io_hdr.port_status = 31341;
 612                         io->scsiio.be_move_done(io);
 613                         return;
 614                 }
 615                 msg.dt.sent_sg_entries = sg_entries_sent;
 616         }
 617
 618         /*
 619          * Officially handover the request from us to peer.
 620          * If failover has just happened, then we must return error.
 621          * If failover happen just after, then it is not our problem.
 622          */
 623         if (lun)
 624                 mtx_lock(&lun->lun_lock);
 625         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
 626                 if (lun)
 627                         mtx_unlock(&lun->lun_lock);
 628                 io->io_hdr.port_status = 31342;
 629                 io->scsiio.be_move_done(io);
 630                 return;
 631         }
 632         io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
 633         io->io_hdr.flags |= CTL_FLAG_DMA_INPROG;
 634         if (lun)
 635                 mtx_unlock(&lun->lun_lock);
 636 }
 637
 638 static void
 639 ctl_ha_done(union ctl_io *io)
 640 {
 641         union ctl_ha_msg msg;
 642
 643         if (io->io_hdr.io_type == CTL_IO_SCSI) {
 644                 memset(&msg, 0, sizeof(msg));
 645                 msg.hdr.msg_type = CTL_MSG_FINISH_IO;
 646                 msg.hdr.original_sc = io->io_hdr.original_sc;
 647                 msg.hdr.nexus = io->io_hdr.nexus;
 648                 msg.hdr.status = io->io_hdr.status;
 649                 msg.scsi.scsi_status = io->scsiio.scsi_status;
 650                 msg.scsi.tag_num = io->scsiio.tag_num;
 651                 msg.scsi.tag_type = io->scsiio.tag_type;
 652                 msg.scsi.sense_len = io->scsiio.sense_len;
 653                 msg.scsi.sense_residual = io->scsiio.sense_residual;
 654                 msg.scsi.residual = io->scsiio.residual;
 655                 memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
 656                     io->scsiio.sense_len);
 657                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
 658                     sizeof(msg.scsi) - sizeof(msg.scsi.sense_data) +
 659                     msg.scsi.sense_len, M_WAITOK);
 660         }
 661         ctl_free_io(io);
 662 }
 663
 664 static void
 665 ctl_isc_handler_finish_xfer(struct ctl_softc *ctl_softc,
 666                             union ctl_ha_msg *msg_info)
 667 {
 668         struct ctl_scsiio *ctsio;
 669
 670         if (msg_info->hdr.original_sc == NULL) {
 671                 printf("%s: original_sc == NULL!\n", __func__);
 672                 /* XXX KDM now what? */
 673                 return;
 674         }
 675
 676         ctsio = &msg_info->hdr.original_sc->scsiio;
 677         ctsio->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
 678         ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
 679         ctsio->io_hdr.status = msg_info->hdr.status;
 680         ctsio->scsi_status = msg_info->scsi.scsi_status;
 681         ctsio->sense_len = msg_info->scsi.sense_len;
 682         ctsio->sense_residual = msg_info->scsi.sense_residual;
 683         ctsio->residual = msg_info->scsi.residual;
 684         memcpy(&ctsio->sense_data, &msg_info->scsi.sense_data,
 685                msg_info->scsi.sense_len);
 686         ctl_enqueue_isc((union ctl_io *)ctsio);
 687 }
 688
 689 static void
 690 ctl_isc_handler_finish_ser_only(struct ctl_softc *ctl_softc,
 691                                 union ctl_ha_msg *msg_info)
 692 {
 693         struct ctl_scsiio *ctsio;
 694
 695         if (msg_info->hdr.serializing_sc == NULL) {
 696                 printf("%s: serializing_sc == NULL!\n", __func__);
 697                 /* XXX KDM now what? */
 698                 return;
 699         }
 700
 701         ctsio = &msg_info->hdr.serializing_sc->scsiio;
 702         ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
 703         ctl_enqueue_isc((union ctl_io *)ctsio);
 704 }
 705
 706 void
 707 ctl_isc_announce_lun(struct ctl_lun *lun)
 708 {
 709         struct ctl_softc *softc = lun->ctl_softc;
 710         union ctl_ha_msg *msg;
 711         struct ctl_ha_msg_lun_pr_key pr_key;
 712         int i, k;
 713
 714         if (softc->ha_link != CTL_HA_LINK_ONLINE)
 715                 return;
 716         mtx_lock(&lun->lun_lock);
 717         i = sizeof(msg->lun);
 718         if (lun->lun_devid)
 719                 i += lun->lun_devid->len;
 720         i += sizeof(pr_key) * lun->pr_key_count;
 721 alloc:
 722         mtx_unlock(&lun->lun_lock);
 723         msg = malloc(i, M_CTL, M_WAITOK);
 724         mtx_lock(&lun->lun_lock);
 725         k = sizeof(msg->lun);
 726         if (lun->lun_devid)
 727                 k += lun->lun_devid->len;
 728         k += sizeof(pr_key) * lun->pr_key_count;
 729         if (i < k) {
 730                 free(msg, M_CTL);
 731                 i = k;
 732                 goto alloc;
 733         }
 734         bzero(&msg->lun, sizeof(msg->lun));
 735         msg->hdr.msg_type = CTL_MSG_LUN_SYNC;
 736         msg->hdr.nexus.targ_lun = lun->lun;
 737         msg->hdr.nexus.targ_mapped_lun = lun->lun;
 738         msg->lun.flags = lun->flags;
 739         msg->lun.pr_generation = lun->PRGeneration;
 740         msg->lun.pr_res_idx = lun->pr_res_idx;
 741         msg->lun.pr_res_type = lun->res_type;
 742         msg->lun.pr_key_count = lun->pr_key_count;
 743         i = 0;
 744         if (lun->lun_devid) {
 745                 msg->lun.lun_devid_len = lun->lun_devid->len;
 746                 memcpy(&msg->lun.data[i], lun->lun_devid->data,
 747                     msg->lun.lun_devid_len);
 748                 i += msg->lun.lun_devid_len;
 749         }
 750         for (k = 0; k < CTL_MAX_INITIATORS; k++) {
 751                 if ((pr_key.pr_key = ctl_get_prkey(lun, k)) == 0)
 752                         continue;
 753                 pr_key.pr_iid = k;
 754                 memcpy(&msg->lun.data[i], &pr_key, sizeof(pr_key));
 755                 i += sizeof(pr_key);
 756         }
 757         mtx_unlock(&lun->lun_lock);
 758         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->port, sizeof(msg->port) + i,
 759             M_WAITOK);
 760         free(msg, M_CTL);
 761
 762         if (lun->flags & CTL_LUN_PRIMARY_SC) {
 763                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
 764                         ctl_isc_announce_mode(lun, -1,
 765                             lun->mode_pages.index[i].page_code & SMPH_PC_MASK,
 766                             lun->mode_pages.index[i].subpage);
 767                 }
 768         }
 769 }
 770
 771 void
 772 ctl_isc_announce_port(struct ctl_port *port)
 773 {
 774         struct ctl_softc *softc = port->ctl_softc;
 775         union ctl_ha_msg *msg;
 776         int i;
 777
 778         if (port->targ_port < softc->port_min ||
 779             port->targ_port >= softc->port_max ||
 780             softc->ha_link != CTL_HA_LINK_ONLINE)
 781                 return;
 782         i = sizeof(msg->port) + strlen(port->port_name) + 1;
 783         if (port->lun_map)
 784                 i += sizeof(uint32_t) * CTL_MAX_LUNS;
 785         if (port->port_devid)
 786                 i += port->port_devid->len;
 787         if (port->target_devid)
 788                 i += port->target_devid->len;
 789         if (port->init_devid)
 790                 i += port->init_devid->len;
 791         msg = malloc(i, M_CTL, M_WAITOK);
 792         bzero(&msg->port, sizeof(msg->port));
 793         msg->hdr.msg_type = CTL_MSG_PORT_SYNC;
 794         msg->hdr.nexus.targ_port = port->targ_port;
 795         msg->port.port_type = port->port_type;
 796         msg->port.physical_port = port->physical_port;
 797         msg->port.virtual_port = port->virtual_port;
 798         msg->port.status = port->status;
 799         i = 0;
 800         msg->port.name_len = sprintf(&msg->port.data[i],
 801             "%d:%s", softc->ha_id, port->port_name) + 1;
 802         i += msg->port.name_len;
 803         if (port->lun_map) {
 804                 msg->port.lun_map_len = sizeof(uint32_t) * CTL_MAX_LUNS;
 805                 memcpy(&msg->port.data[i], port->lun_map,
 806                     msg->port.lun_map_len);
 807                 i += msg->port.lun_map_len;
 808         }
 809         if (port->port_devid) {
 810                 msg->port.port_devid_len = port->port_devid->len;
 811                 memcpy(&msg->port.data[i], port->port_devid->data,
 812                     msg->port.port_devid_len);
 813                 i += msg->port.port_devid_len;
 814         }
 815         if (port->target_devid) {
 816                 msg->port.target_devid_len = port->target_devid->len;
 817                 memcpy(&msg->port.data[i], port->target_devid->data,
 818                     msg->port.target_devid_len);
 819                 i += msg->port.target_devid_len;
 820         }
 821         if (port->init_devid) {
 822                 msg->port.init_devid_len = port->init_devid->len;
 823                 memcpy(&msg->port.data[i], port->init_devid->data,
 824                     msg->port.init_devid_len);
 825                 i += msg->port.init_devid_len;
 826         }
 827         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->port, sizeof(msg->port) + i,
 828             M_WAITOK);
 829         free(msg, M_CTL);
 830 }
 831
 832 void
 833 ctl_isc_announce_iid(struct ctl_port *port, int iid)
 834 {
 835         struct ctl_softc *softc = port->ctl_softc;
 836         union ctl_ha_msg *msg;
 837         int i, l;
 838
 839         if (port->targ_port < softc->port_min ||
 840             port->targ_port >= softc->port_max ||
 841             softc->ha_link != CTL_HA_LINK_ONLINE)
 842                 return;
 843         mtx_lock(&softc->ctl_lock);
 844         i = sizeof(msg->iid);
 845         l = 0;
 846         if (port->wwpn_iid[iid].name)
 847                 l = strlen(port->wwpn_iid[iid].name) + 1;
 848         i += l;
 849         msg = malloc(i, M_CTL, M_NOWAIT);
 850         if (msg == NULL) {
 851                 mtx_unlock(&softc->ctl_lock);
 852                 return;
 853         }
 854         bzero(&msg->iid, sizeof(msg->iid));
 855         msg->hdr.msg_type = CTL_MSG_IID_SYNC;
 856         msg->hdr.nexus.targ_port = port->targ_port;
 857         msg->hdr.nexus.initid = iid;
 858         msg->iid.in_use = port->wwpn_iid[iid].in_use;
 859         msg->iid.name_len = l;
 860         msg->iid.wwpn = port->wwpn_iid[iid].wwpn;
 861         if (port->wwpn_iid[iid].name)
 862                 strlcpy(msg->iid.data, port->wwpn_iid[iid].name, l);
 863         mtx_unlock(&softc->ctl_lock);
 864         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->iid, i, M_NOWAIT);
 865         free(msg, M_CTL);
 866 }
 867
 868 void
 869 ctl_isc_announce_mode(struct ctl_lun *lun, uint32_t initidx,
 870     uint8_t page, uint8_t subpage)
 871 {
 872         struct ctl_softc *softc = lun->ctl_softc;
 873         union ctl_ha_msg msg;
 874         int i;
 875
 876         if (softc->ha_link != CTL_HA_LINK_ONLINE)
 877                 return;
 878         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
 879                 if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) ==
 880                     page && lun->mode_pages.index[i].subpage == subpage)
 881                         break;
 882         }
 883         if (i == CTL_NUM_MODE_PAGES)
 884                 return;
 885         bzero(&msg.mode, sizeof(msg.mode));
 886         msg.hdr.msg_type = CTL_MSG_MODE_SYNC;
 887         msg.hdr.nexus.targ_port = initidx / CTL_MAX_INIT_PER_PORT;
 888         msg.hdr.nexus.initid = initidx % CTL_MAX_INIT_PER_PORT;
 889         msg.hdr.nexus.targ_lun = lun->lun;
 890         msg.hdr.nexus.targ_mapped_lun = lun->lun;
 891         msg.mode.page_code = page;
 892         msg.mode.subpage = subpage;
 893         msg.mode.page_len = lun->mode_pages.index[i].page_len;
 894         memcpy(msg.mode.data, lun->mode_pages.index[i].page_data,
 895             msg.mode.page_len);
 896         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg.mode, sizeof(msg.mode),
 897             M_WAITOK);
 898 }
 899
 900 static void
 901 ctl_isc_ha_link_up(struct ctl_softc *softc)
 902 {
 903         struct ctl_port *port;
 904         struct ctl_lun *lun;
 905         union ctl_ha_msg msg;
 906         int i;
 907
 908         /* Announce this node parameters to peer for validation. */
 909         msg.login.msg_type = CTL_MSG_LOGIN;
 910         msg.login.version = CTL_HA_VERSION;
 911         msg.login.ha_mode = softc->ha_mode;
 912         msg.login.ha_id = softc->ha_id;
 913         msg.login.max_luns = CTL_MAX_LUNS;
 914         msg.login.max_ports = CTL_MAX_PORTS;
 915         msg.login.max_init_per_port = CTL_MAX_INIT_PER_PORT;
 916         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg.login, sizeof(msg.login),
 917             M_WAITOK);
 918
 919         STAILQ_FOREACH(port, &softc->port_list, links) {
 920                 ctl_isc_announce_port(port);
 921                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
 922                         if (port->wwpn_iid[i].in_use)
 923                                 ctl_isc_announce_iid(port, i);
 924                 }
 925         }
 926         STAILQ_FOREACH(lun, &softc->lun_list, links)
 927                 ctl_isc_announce_lun(lun);
 928 }
 929
 930 static void
 931 ctl_isc_ha_link_down(struct ctl_softc *softc)
 932 {
 933         struct ctl_port *port;
 934         struct ctl_lun *lun;
 935         union ctl_io *io;
 936         int i;
 937
 938         mtx_lock(&softc->ctl_lock);
 939         STAILQ_FOREACH(lun, &softc->lun_list, links) {
 940                 mtx_lock(&lun->lun_lock);
 941                 if (lun->flags & CTL_LUN_PEER_SC_PRIMARY) {
 942                         lun->flags &= ~CTL_LUN_PEER_SC_PRIMARY;
 943                         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
 944                 }
 945                 mtx_unlock(&lun->lun_lock);
 946
 947                 mtx_unlock(&softc->ctl_lock);
 948                 io = ctl_alloc_io(softc->othersc_pool);
 949                 mtx_lock(&softc->ctl_lock);
 950                 ctl_zero_io(io);
 951                 io->io_hdr.msg_type = CTL_MSG_FAILOVER;
 952                 io->io_hdr.nexus.targ_mapped_lun = lun->lun;
 953                 ctl_enqueue_isc(io);
 954         }
 955
 956         STAILQ_FOREACH(port, &softc->port_list, links) {
 957                 if (port->targ_port >= softc->port_min &&
 958                     port->targ_port < softc->port_max)
 959                         continue;
 960                 port->status &= ~CTL_PORT_STATUS_ONLINE;
 961                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
 962                         port->wwpn_iid[i].in_use = 0;
 963                         free(port->wwpn_iid[i].name, M_CTL);
 964                         port->wwpn_iid[i].name = NULL;
 965                 }
 966         }
 967         mtx_unlock(&softc->ctl_lock);
 968 }
 969
 970 static void
 971 ctl_isc_ua(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
 972 {
 973         struct ctl_lun *lun;
 974         uint32_t iid = ctl_get_initindex(&msg->hdr.nexus);
 975
 976         mtx_lock(&softc->ctl_lock);
 977         if (msg->hdr.nexus.targ_lun < CTL_MAX_LUNS &&
 978             (lun = softc->ctl_luns[msg->hdr.nexus.targ_mapped_lun]) != NULL) {
 979                 mtx_lock(&lun->lun_lock);
 980                 mtx_unlock(&softc->ctl_lock);
 981                 if (msg->ua.ua_type == CTL_UA_THIN_PROV_THRES &&
 982                     msg->ua.ua_set)
 983                         memcpy(lun->ua_tpt_info, msg->ua.ua_info, 8);
 984                 if (msg->ua.ua_all) {
 985                         if (msg->ua.ua_set)
 986                                 ctl_est_ua_all(lun, iid, msg->ua.ua_type);
 987                         else
 988                                 ctl_clr_ua_all(lun, iid, msg->ua.ua_type);
 989                 } else {
 990                         if (msg->ua.ua_set)
 991                                 ctl_est_ua(lun, iid, msg->ua.ua_type);
 992                         else
 993                                 ctl_clr_ua(lun, iid, msg->ua.ua_type);
 994                 }
 995                 mtx_unlock(&lun->lun_lock);
 996         } else
 997                 mtx_unlock(&softc->ctl_lock);
 998 }
 999
1000 static void
1001 ctl_isc_lun_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1002 {
1003         struct ctl_lun *lun;
1004         struct ctl_ha_msg_lun_pr_key pr_key;
1005         int i, k;
1006         ctl_lun_flags oflags;
1007         uint32_t targ_lun;
1008
1009         targ_lun = msg->hdr.nexus.targ_mapped_lun;
1010         mtx_lock(&softc->ctl_lock);
1011         if ((targ_lun >= CTL_MAX_LUNS) ||
1012             ((lun = softc->ctl_luns[targ_lun]) == NULL)) {
1013                 mtx_unlock(&softc->ctl_lock);
1014                 return;
1015         }
1016         mtx_lock(&lun->lun_lock);
1017         mtx_unlock(&softc->ctl_lock);
1018         if (lun->flags & CTL_LUN_DISABLED) {
1019                 mtx_unlock(&lun->lun_lock);
1020                 return;
1021         }
1022         i = (lun->lun_devid != NULL) ? lun->lun_devid->len : 0;
1023         if (msg->lun.lun_devid_len != i || (i > 0 &&
1024             memcmp(&msg->lun.data[0], lun->lun_devid->data, i) != 0)) {
1025                 mtx_unlock(&lun->lun_lock);
1026                 printf("%s: Received conflicting HA LUN %d\n",
1027                     __func__, msg->hdr.nexus.targ_lun);
1028                 return;
1029         } else {
1030                 /* Record whether peer is primary. */
1031                 oflags = lun->flags;
1032                 if ((msg->lun.flags & CTL_LUN_PRIMARY_SC) &&
1033                     (msg->lun.flags & CTL_LUN_DISABLED) == 0)
1034                         lun->flags |= CTL_LUN_PEER_SC_PRIMARY;
1035                 else
1036                         lun->flags &= ~CTL_LUN_PEER_SC_PRIMARY;
1037                 if (oflags != lun->flags)
1038                         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
1039
1040                 /* If peer is primary and we are not -- use data */
1041                 if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
1042                     (lun->flags & CTL_LUN_PEER_SC_PRIMARY)) {
1043                         lun->PRGeneration = msg->lun.pr_generation;
1044                         lun->pr_res_idx = msg->lun.pr_res_idx;
1045                         lun->res_type = msg->lun.pr_res_type;
1046                         lun->pr_key_count = msg->lun.pr_key_count;
1047                         for (k = 0; k < CTL_MAX_INITIATORS; k++)
1048                                 ctl_clr_prkey(lun, k);
1049                         for (k = 0; k < msg->lun.pr_key_count; k++) {
1050                                 memcpy(&pr_key, &msg->lun.data[i],
1051                                     sizeof(pr_key));
1052                                 ctl_alloc_prkey(lun, pr_key.pr_iid);
1053                                 ctl_set_prkey(lun, pr_key.pr_iid,
1054                                     pr_key.pr_key);
1055                                 i += sizeof(pr_key);
1056                         }
1057                 }
1058
1059                 mtx_unlock(&lun->lun_lock);
1060                 CTL_DEBUG_PRINT(("%s: Known LUN %d, peer is %s\n",
1061                     __func__, msg->hdr.nexus.targ_lun,
1062                     (msg->lun.flags & CTL_LUN_PRIMARY_SC) ?
1063                     "primary" : "secondary"));
1064
1065                 /* If we are primary but peer doesn't know -- notify */
1066                 if ((lun->flags & CTL_LUN_PRIMARY_SC) &&
1067                     (msg->lun.flags & CTL_LUN_PEER_SC_PRIMARY) == 0)
1068                         ctl_isc_announce_lun(lun);
1069         }
1070 }
1071
1072 static void
1073 ctl_isc_port_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1074 {
1075         struct ctl_port *port;
1076         struct ctl_lun *lun;
1077         int i, new;
1078
1079         port = softc->ctl_ports[msg->hdr.nexus.targ_port];
1080         if (port == NULL) {
1081                 CTL_DEBUG_PRINT(("%s: New port %d\n", __func__,
1082                     msg->hdr.nexus.targ_port));
1083                 new = 1;
1084                 port = malloc(sizeof(*port), M_CTL, M_WAITOK | M_ZERO);
1085                 port->frontend = &ha_frontend;
1086                 port->targ_port = msg->hdr.nexus.targ_port;
1087                 port->fe_datamove = ctl_ha_datamove;
1088                 port->fe_done = ctl_ha_done;
1089         } else if (port->frontend == &ha_frontend) {
1090                 CTL_DEBUG_PRINT(("%s: Updated port %d\n", __func__,
1091                     msg->hdr.nexus.targ_port));
1092                 new = 0;
1093         } else {
1094                 printf("%s: Received conflicting HA port %d\n",
1095                     __func__, msg->hdr.nexus.targ_port);
1096                 return;
1097         }
1098         port->port_type = msg->port.port_type;
1099         port->physical_port = msg->port.physical_port;
1100         port->virtual_port = msg->port.virtual_port;
1101         port->status = msg->port.status;
1102         i = 0;
1103         free(port->port_name, M_CTL);
1104         port->port_name = strndup(&msg->port.data[i], msg->port.name_len,
1105             M_CTL);
1106         i += msg->port.name_len;
1107         if (msg->port.lun_map_len != 0) {
1108                 if (port->lun_map == NULL)
1109                         port->lun_map = malloc(sizeof(uint32_t) * CTL_MAX_LUNS,
1110                             M_CTL, M_WAITOK);
1111                 memcpy(port->lun_map, &msg->port.data[i],
1112                     sizeof(uint32_t) * CTL_MAX_LUNS);
1113                 i += msg->port.lun_map_len;
1114         } else {
1115                 free(port->lun_map, M_CTL);
1116                 port->lun_map = NULL;
1117         }
1118         if (msg->port.port_devid_len != 0) {
1119                 if (port->port_devid == NULL ||
1120                     port->port_devid->len != msg->port.port_devid_len) {
1121                         free(port->port_devid, M_CTL);
1122                         port->port_devid = malloc(sizeof(struct ctl_devid) +
1123                             msg->port.port_devid_len, M_CTL, M_WAITOK);
1124                 }
1125                 memcpy(port->port_devid->data, &msg->port.data[i],
1126                     msg->port.port_devid_len);
1127                 port->port_devid->len = msg->port.port_devid_len;
1128                 i += msg->port.port_devid_len;
1129         } else {
1130                 free(port->port_devid, M_CTL);
1131                 port->port_devid = NULL;
1132         }
1133         if (msg->port.target_devid_len != 0) {
1134                 if (port->target_devid == NULL ||
1135                     port->target_devid->len != msg->port.target_devid_len) {
1136                         free(port->target_devid, M_CTL);
1137                         port->target_devid = malloc(sizeof(struct ctl_devid) +
1138                             msg->port.target_devid_len, M_CTL, M_WAITOK);
1139                 }
1140                 memcpy(port->target_devid->data, &msg->port.data[i],
1141                     msg->port.target_devid_len);
1142                 port->target_devid->len = msg->port.target_devid_len;
1143                 i += msg->port.target_devid_len;
1144         } else {
1145                 free(port->target_devid, M_CTL);
1146                 port->target_devid = NULL;
1147         }
1148         if (msg->port.init_devid_len != 0) {
1149                 if (port->init_devid == NULL ||
1150                     port->init_devid->len != msg->port.init_devid_len) {
1151                         free(port->init_devid, M_CTL);
1152                         port->init_devid = malloc(sizeof(struct ctl_devid) +
1153                             msg->port.init_devid_len, M_CTL, M_WAITOK);
1154                 }
1155                 memcpy(port->init_devid->data, &msg->port.data[i],
1156                     msg->port.init_devid_len);
1157                 port->init_devid->len = msg->port.init_devid_len;
1158                 i += msg->port.init_devid_len;
1159         } else {
1160                 free(port->init_devid, M_CTL);
1161                 port->init_devid = NULL;
1162         }
1163         if (new) {
1164                 if (ctl_port_register(port) != 0) {
1165                         printf("%s: ctl_port_register() failed with error\n",
1166                             __func__);
1167                 }
1168         }
1169         mtx_lock(&softc->ctl_lock);
1170         STAILQ_FOREACH(lun, &softc->lun_list, links) {
1171                 if (ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
1172                         continue;
1173                 mtx_lock(&lun->lun_lock);
1174                 ctl_est_ua_all(lun, -1, CTL_UA_INQ_CHANGE);
1175                 mtx_unlock(&lun->lun_lock);
1176         }
1177         mtx_unlock(&softc->ctl_lock);
1178 }
1179
1180 static void
1181 ctl_isc_iid_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1182 {
1183         struct ctl_port *port;
1184         int iid;
1185
1186         port = softc->ctl_ports[msg->hdr.nexus.targ_port];
1187         if (port == NULL) {
1188                 printf("%s: Received IID for unknown port %d\n",
1189                     __func__, msg->hdr.nexus.targ_port);
1190                 return;
1191         }
1192         iid = msg->hdr.nexus.initid;
1193         port->wwpn_iid[iid].in_use = msg->iid.in_use;
1194         port->wwpn_iid[iid].wwpn = msg->iid.wwpn;
1195         free(port->wwpn_iid[iid].name, M_CTL);
1196         if (msg->iid.name_len) {
1197                 port->wwpn_iid[iid].name = strndup(&msg->iid.data[0],
1198                     msg->iid.name_len, M_CTL);
1199         } else
1200                 port->wwpn_iid[iid].name = NULL;
1201 }
1202
1203 static void
1204 ctl_isc_login(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1205 {
1206
1207         if (msg->login.version != CTL_HA_VERSION) {
1208                 printf("CTL HA peers have different versions %d != %d\n",
1209                     msg->login.version, CTL_HA_VERSION);
1210                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1211                 return;
1212         }
1213         if (msg->login.ha_mode != softc->ha_mode) {
1214                 printf("CTL HA peers have different ha_mode %d != %d\n",
1215                     msg->login.ha_mode, softc->ha_mode);
1216                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1217                 return;
1218         }
1219         if (msg->login.ha_id == softc->ha_id) {
1220                 printf("CTL HA peers have same ha_id %d\n", msg->login.ha_id);
1221                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1222                 return;
1223         }
1224         if (msg->login.max_luns != CTL_MAX_LUNS ||
1225             msg->login.max_ports != CTL_MAX_PORTS ||
1226             msg->login.max_init_per_port != CTL_MAX_INIT_PER_PORT) {
1227                 printf("CTL HA peers have different limits\n");
1228                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1229                 return;
1230         }
1231 }
1232
1233 static void
1234 ctl_isc_mode_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1235 {
1236         struct ctl_lun *lun;
1237         int i;
1238         uint32_t initidx, targ_lun;
1239
1240         targ_lun = msg->hdr.nexus.targ_mapped_lun;
1241         mtx_lock(&softc->ctl_lock);
1242         if ((targ_lun >= CTL_MAX_LUNS) ||
1243             ((lun = softc->ctl_luns[targ_lun]) == NULL)) {
1244                 mtx_unlock(&softc->ctl_lock);
1245                 return;
1246         }
1247         mtx_lock(&lun->lun_lock);
1248         mtx_unlock(&softc->ctl_lock);
1249         if (lun->flags & CTL_LUN_DISABLED) {
1250                 mtx_unlock(&lun->lun_lock);
1251                 return;
1252         }
1253         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
1254                 if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) ==
1255                     msg->mode.page_code &&
1256                     lun->mode_pages.index[i].subpage == msg->mode.subpage)
1257                         break;
1258         }
1259         if (i == CTL_NUM_MODE_PAGES) {
1260                 mtx_unlock(&lun->lun_lock);
1261                 return;
1262         }
1263         memcpy(lun->mode_pages.index[i].page_data, msg->mode.data,
1264             lun->mode_pages.index[i].page_len);
1265         initidx = ctl_get_initindex(&msg->hdr.nexus);
1266         if (initidx != -1)
1267                 ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
1268         mtx_unlock(&lun->lun_lock);
1269 }
1270
1271 /*
1272  * ISC (Inter Shelf Communication) event handler.  Events from the HA
1273  * subsystem come in here.
1274  */
1275 static void
1276 ctl_isc_event_handler(ctl_ha_channel channel, ctl_ha_event event, int param)
1277 {
1278         struct ctl_softc *softc = control_softc;
1279         union ctl_io *io;
1280         struct ctl_prio *presio;
1281         ctl_ha_status isc_status;
1282
1283         CTL_DEBUG_PRINT(("CTL: Isc Msg event %d\n", event));
1284         if (event == CTL_HA_EVT_MSG_RECV) {
1285                 union ctl_ha_msg *msg, msgbuf;
1286
1287                 if (param > sizeof(msgbuf))
1288                         msg = malloc(param, M_CTL, M_WAITOK);
1289                 else
1290                         msg = &msgbuf;
1291                 isc_status = ctl_ha_msg_recv(CTL_HA_CHAN_CTL, msg, param,
1292                     M_WAITOK);
1293                 if (isc_status != CTL_HA_STATUS_SUCCESS) {
1294                         printf("%s: Error receiving message: %d\n",
1295                             __func__, isc_status);
1296                         if (msg != &msgbuf)
1297                                 free(msg, M_CTL);
1298                         return;
1299                 }
1300
1301                 CTL_DEBUG_PRINT(("CTL: msg_type %d\n", msg->msg_type));
1302                 switch (msg->hdr.msg_type) {
1303                 case CTL_MSG_SERIALIZE:
1304                         io = ctl_alloc_io(softc->othersc_pool);
1305                         ctl_zero_io(io);
1306                         // populate ctsio from msg
1307                         io->io_hdr.io_type = CTL_IO_SCSI;
1308                         io->io_hdr.msg_type = CTL_MSG_SERIALIZE;
1309                         io->io_hdr.original_sc = msg->hdr.original_sc;
1310                         io->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC |
1311                                             CTL_FLAG_IO_ACTIVE;
1312                         /*
1313                          * If we're in serialization-only mode, we don't
1314                          * want to go through full done processing.  Thus
1315                          * the COPY flag.
1316                          *
1317                          * XXX KDM add another flag that is more specific.
1318                          */
1319                         if (softc->ha_mode != CTL_HA_MODE_XFER)
1320                                 io->io_hdr.flags |= CTL_FLAG_INT_COPY;
1321                         io->io_hdr.nexus = msg->hdr.nexus;
1322 #if 0
1323                         printf("port %u, iid %u, lun %u\n",
1324                                io->io_hdr.nexus.targ_port,
1325                                io->io_hdr.nexus.initid,
1326                                io->io_hdr.nexus.targ_lun);
1327 #endif
1328                         io->scsiio.tag_num = msg->scsi.tag_num;
1329                         io->scsiio.tag_type = msg->scsi.tag_type;
1330 #ifdef CTL_TIME_IO
1331                         io->io_hdr.start_time = time_uptime;
1332                         getbinuptime(&io->io_hdr.start_bt);
1333 #endif /* CTL_TIME_IO */
1334                         io->scsiio.cdb_len = msg->scsi.cdb_len;
1335                         memcpy(io->scsiio.cdb, msg->scsi.cdb,
1336                                CTL_MAX_CDBLEN);
1337                         if (softc->ha_mode == CTL_HA_MODE_XFER) {
1338                                 const struct ctl_cmd_entry *entry;
1339
1340                                 entry = ctl_get_cmd_entry(&io->scsiio, NULL);
1341                                 io->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
1342                                 io->io_hdr.flags |=
1343                                         entry->flags & CTL_FLAG_DATA_MASK;
1344                         }
1345                         ctl_enqueue_isc(io);
1346                         break;
1347
1348                 /* Performed on the Originating SC, XFER mode only */
1349                 case CTL_MSG_DATAMOVE: {
1350                         struct ctl_sg_entry *sgl;
1351                         int i, j;
1352
1353                         io = msg->hdr.original_sc;
1354                         if (io == NULL) {
1355                                 printf("%s: original_sc == NULL!\n", __func__);
1356                                 /* XXX KDM do something here */
1357                                 break;
1358                         }
1359                         io->io_hdr.msg_type = CTL_MSG_DATAMOVE;
1360                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1361                         /*
1362                          * Keep track of this, we need to send it back over
1363                          * when the datamove is complete.
1364                          */
1365                         io->io_hdr.serializing_sc = msg->hdr.serializing_sc;
1366                         if (msg->hdr.status == CTL_SUCCESS)
1367                                 io->io_hdr.status = msg->hdr.status;
1368
1369                         if (msg->dt.sg_sequence == 0) {
1370 #ifdef CTL_TIME_IO
1371                                 getbinuptime(&io->io_hdr.dma_start_bt);
1372 #endif
1373                                 i = msg->dt.kern_sg_entries +
1374                                     msg->dt.kern_data_len /
1375                                     CTL_HA_DATAMOVE_SEGMENT + 1;
1376                                 sgl = malloc(sizeof(*sgl) * i, M_CTL,
1377                                     M_WAITOK | M_ZERO);
1378                                 io->io_hdr.remote_sglist = sgl;
1379                                 io->io_hdr.local_sglist =
1380                                     &sgl[msg->dt.kern_sg_entries];
1381
1382                                 io->scsiio.kern_data_ptr = (uint8_t *)sgl;
1383
1384                                 io->scsiio.kern_sg_entries =
1385                                         msg->dt.kern_sg_entries;
1386                                 io->scsiio.rem_sg_entries =
1387                                         msg->dt.kern_sg_entries;
1388                                 io->scsiio.kern_data_len =
1389                                         msg->dt.kern_data_len;
1390                                 io->scsiio.kern_total_len =
1391                                         msg->dt.kern_total_len;
1392                                 io->scsiio.kern_data_resid =
1393                                         msg->dt.kern_data_resid;
1394                                 io->scsiio.kern_rel_offset =
1395                                         msg->dt.kern_rel_offset;
1396                                 io->io_hdr.flags &= ~CTL_FLAG_BUS_ADDR;
1397                                 io->io_hdr.flags |= msg->dt.flags &
1398                                     CTL_FLAG_BUS_ADDR;
1399                         } else
1400                                 sgl = (struct ctl_sg_entry *)
1401                                         io->scsiio.kern_data_ptr;
1402
1403                         for (i = msg->dt.sent_sg_entries, j = 0;
1404                              i < (msg->dt.sent_sg_entries +
1405                              msg->dt.cur_sg_entries); i++, j++) {
1406                                 sgl[i].addr = msg->dt.sg_list[j].addr;
1407                                 sgl[i].len = msg->dt.sg_list[j].len;
1408
1409 #if 0
1410                                 printf("%s: DATAMOVE: %p,%lu j=%d, i=%d\n",
1411                                     __func__, sgl[i].addr, sgl[i].len, j, i);
1412 #endif
1413                         }
1414
1415                         /*
1416                          * If this is the last piece of the I/O, we've got
1417                          * the full S/G list.  Queue processing in the thread.
1418                          * Otherwise wait for the next piece.
1419                          */
1420                         if (msg->dt.sg_last != 0)
1421                                 ctl_enqueue_isc(io);
1422                         break;
1423                 }
1424                 /* Performed on the Serializing (primary) SC, XFER mode only */
1425                 case CTL_MSG_DATAMOVE_DONE: {
1426                         if (msg->hdr.serializing_sc == NULL) {
1427                                 printf("%s: serializing_sc == NULL!\n",
1428                                        __func__);
1429                                 /* XXX KDM now what? */
1430                                 break;
1431                         }
1432                         /*
1433                          * We grab the sense information here in case
1434                          * there was a failure, so we can return status
1435                          * back to the initiator.
1436                          */
1437                         io = msg->hdr.serializing_sc;
1438                         io->io_hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
1439                         io->io_hdr.flags &= ~CTL_FLAG_DMA_INPROG;
1440                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1441                         io->io_hdr.port_status = msg->scsi.fetd_status;
1442                         io->scsiio.residual = msg->scsi.residual;
1443                         if (msg->hdr.status != CTL_STATUS_NONE) {
1444                                 io->io_hdr.status = msg->hdr.status;
1445                                 io->scsiio.scsi_status = msg->scsi.scsi_status;
1446                                 io->scsiio.sense_len = msg->scsi.sense_len;
1447                                 io->scsiio.sense_residual =msg->scsi.sense_residual;
1448                                 memcpy(&io->scsiio.sense_data,
1449                                     &msg->scsi.sense_data,
1450                                     msg->scsi.sense_len);
1451                                 if (msg->hdr.status == CTL_SUCCESS)
1452                                         io->io_hdr.flags |= CTL_FLAG_STATUS_SENT;
1453                         }
1454                         ctl_enqueue_isc(io);
1455                         break;
1456                 }
1457
1458                 /* Preformed on Originating SC, SER_ONLY mode */
1459                 case CTL_MSG_R2R:
1460                         io = msg->hdr.original_sc;
1461                         if (io == NULL) {
1462                                 printf("%s: original_sc == NULL!\n",
1463                                     __func__);
1464                                 break;
1465                         }
1466                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1467                         io->io_hdr.msg_type = CTL_MSG_R2R;
1468                         io->io_hdr.serializing_sc = msg->hdr.serializing_sc;
1469                         ctl_enqueue_isc(io);
1470                         break;
1471
1472                 /*
1473                  * Performed on Serializing(i.e. primary SC) SC in SER_ONLY
1474                  * mode.
1475                  * Performed on the Originating (i.e. secondary) SC in XFER
1476                  * mode
1477                  */
1478                 case CTL_MSG_FINISH_IO:
1479                         if (softc->ha_mode == CTL_HA_MODE_XFER)
1480                                 ctl_isc_handler_finish_xfer(softc, msg);
1481                         else
1482                                 ctl_isc_handler_finish_ser_only(softc, msg);
1483                         break;
1484
1485                 /* Preformed on Originating SC */
1486                 case CTL_MSG_BAD_JUJU:
1487                         io = msg->hdr.original_sc;
1488                         if (io == NULL) {
1489                                 printf("%s: Bad JUJU!, original_sc is NULL!\n",
1490                                        __func__);
1491                                 break;
1492                         }
1493                         ctl_copy_sense_data(msg, io);
1494                         /*
1495                          * IO should have already been cleaned up on other
1496                          * SC so clear this flag so we won't send a message
1497                          * back to finish the IO there.
1498                          */
1499                         io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
1500                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1501
1502                         /* io = msg->hdr.serializing_sc; */
1503                         io->io_hdr.msg_type = CTL_MSG_BAD_JUJU;
1504                         ctl_enqueue_isc(io);
1505                         break;
1506
1507                 /* Handle resets sent from the other side */
1508                 case CTL_MSG_MANAGE_TASKS: {
1509                         struct ctl_taskio *taskio;
1510                         taskio = (struct ctl_taskio *)ctl_alloc_io(
1511                             softc->othersc_pool);
1512                         ctl_zero_io((union ctl_io *)taskio);
1513                         taskio->io_hdr.io_type = CTL_IO_TASK;
1514                         taskio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
1515                         taskio->io_hdr.nexus = msg->hdr.nexus;
1516                         taskio->task_action = msg->task.task_action;
1517                         taskio->tag_num = msg->task.tag_num;
1518                         taskio->tag_type = msg->task.tag_type;
1519 #ifdef CTL_TIME_IO
1520                         taskio->io_hdr.start_time = time_uptime;
1521                         getbinuptime(&taskio->io_hdr.start_bt);
1522 #endif /* CTL_TIME_IO */
1523                         ctl_run_task((union ctl_io *)taskio);
1524                         break;
1525                 }
1526                 /* Persistent Reserve action which needs attention */
1527                 case CTL_MSG_PERS_ACTION:
1528                         presio = (struct ctl_prio *)ctl_alloc_io(
1529                             softc->othersc_pool);
1530                         ctl_zero_io((union ctl_io *)presio);
1531                         presio->io_hdr.msg_type = CTL_MSG_PERS_ACTION;
1532                         presio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
1533                         presio->io_hdr.nexus = msg->hdr.nexus;
1534                         presio->pr_msg = msg->pr;
1535                         ctl_enqueue_isc((union ctl_io *)presio);
1536                         break;
1537                 case CTL_MSG_UA:
1538                         ctl_isc_ua(softc, msg, param);
1539                         break;
1540                 case CTL_MSG_PORT_SYNC:
1541                         ctl_isc_port_sync(softc, msg, param);
1542                         break;
1543                 case CTL_MSG_LUN_SYNC:
1544                         ctl_isc_lun_sync(softc, msg, param);
1545                         break;
1546                 case CTL_MSG_IID_SYNC:
1547                         ctl_isc_iid_sync(softc, msg, param);
1548                         break;
1549                 case CTL_MSG_LOGIN:
1550                         ctl_isc_login(softc, msg, param);
1551                         break;
1552                 case CTL_MSG_MODE_SYNC:
1553                         ctl_isc_mode_sync(softc, msg, param);
1554                         break;
1555                 default:
1556                         printf("Received HA message of unknown type %d\n",
1557                             msg->hdr.msg_type);
1558                         ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1559                         break;
1560                 }
1561                 if (msg != &msgbuf)
1562                         free(msg, M_CTL);
1563         } else if (event == CTL_HA_EVT_LINK_CHANGE) {
1564                 printf("CTL: HA link status changed from %d to %d\n",
1565                     softc->ha_link, param);
1566                 if (param == softc->ha_link)
1567                         return;
1568                 if (softc->ha_link == CTL_HA_LINK_ONLINE) {
1569                         softc->ha_link = param;
1570                         ctl_isc_ha_link_down(softc);
1571                 } else {
1572                         softc->ha_link = param;
1573                         if (softc->ha_link == CTL_HA_LINK_ONLINE)
1574                                 ctl_isc_ha_link_up(softc);
1575                 }
1576                 return;
1577         } else {
1578                 printf("ctl_isc_event_handler: Unknown event %d\n", event);
1579                 return;
1580         }
1581 }
1582
1583 static void
1584 ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest)
1585 {
1586
1587         memcpy(&dest->scsiio.sense_data, &src->scsi.sense_data,
1588             src->scsi.sense_len);
1589         dest->scsiio.scsi_status = src->scsi.scsi_status;
1590         dest->scsiio.sense_len = src->scsi.sense_len;
1591         dest->io_hdr.status = src->hdr.status;
1592 }
1593
1594 static void
1595 ctl_copy_sense_data_back(union ctl_io *src, union ctl_ha_msg *dest)
1596 {
1597
1598         memcpy(&dest->scsi.sense_data, &src->scsiio.sense_data,
1599             src->scsiio.sense_len);
1600         dest->scsi.scsi_status = src->scsiio.scsi_status;
1601         dest->scsi.sense_len = src->scsiio.sense_len;
1602         dest->hdr.status = src->io_hdr.status;
1603 }
1604
1605 void
1606 ctl_est_ua(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua)
1607 {
1608         struct ctl_softc *softc = lun->ctl_softc;
1609         ctl_ua_type *pu;
1610
1611         if (initidx < softc->init_min || initidx >= softc->init_max)
1612                 return;
1613         mtx_assert(&lun->lun_lock, MA_OWNED);
1614         pu = lun->pending_ua[initidx / CTL_MAX_INIT_PER_PORT];
1615         if (pu == NULL)
1616                 return;
1617         pu[initidx % CTL_MAX_INIT_PER_PORT] |= ua;
1618 }
1619
1620 void
1621 ctl_est_ua_port(struct ctl_lun *lun, int port, uint32_t except, ctl_ua_type ua)
1622 {
1623         int i;
1624
1625         mtx_assert(&lun->lun_lock, MA_OWNED);
1626         if (lun->pending_ua[port] == NULL)
1627                 return;
1628         for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1629                 if (port * CTL_MAX_INIT_PER_PORT + i == except)
1630                         continue;
1631                 lun->pending_ua[port][i] |= ua;
1632         }
1633 }
1634
1635 void
1636 ctl_est_ua_all(struct ctl_lun *lun, uint32_t except, ctl_ua_type ua)
1637 {
1638         struct ctl_softc *softc = lun->ctl_softc;
1639         int i;
1640
1641         mtx_assert(&lun->lun_lock, MA_OWNED);
1642         for (i = softc->port_min; i < softc->port_max; i++)
1643                 ctl_est_ua_port(lun, i, except, ua);
1644 }
1645
1646 void
1647 ctl_clr_ua(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua)
1648 {
1649         struct ctl_softc *softc = lun->ctl_softc;
1650         ctl_ua_type *pu;
1651
1652         if (initidx < softc->init_min || initidx >= softc->init_max)
1653                 return;
1654         mtx_assert(&lun->lun_lock, MA_OWNED);
1655         pu = lun->pending_ua[initidx / CTL_MAX_INIT_PER_PORT];
1656         if (pu == NULL)
1657                 return;
1658         pu[initidx % CTL_MAX_INIT_PER_PORT] &= ~ua;
1659 }
1660
1661 void
1662 ctl_clr_ua_all(struct ctl_lun *lun, uint32_t except, ctl_ua_type ua)
1663 {
1664         struct ctl_softc *softc = lun->ctl_softc;
1665         int i, j;
1666
1667         mtx_assert(&lun->lun_lock, MA_OWNED);
1668         for (i = softc->port_min; i < softc->port_max; i++) {
1669                 if (lun->pending_ua[i] == NULL)
1670                         continue;
1671                 for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
1672                         if (i * CTL_MAX_INIT_PER_PORT + j == except)
1673                                 continue;
1674                         lun->pending_ua[i][j] &= ~ua;
1675                 }
1676         }
1677 }
1678
1679 void
1680 ctl_clr_ua_allluns(struct ctl_softc *ctl_softc, uint32_t initidx,
1681     ctl_ua_type ua_type)
1682 {
1683         struct ctl_lun *lun;
1684
1685         mtx_assert(&ctl_softc->ctl_lock, MA_OWNED);
1686         STAILQ_FOREACH(lun, &ctl_softc->lun_list, links) {
1687                 mtx_lock(&lun->lun_lock);
1688                 ctl_clr_ua(lun, initidx, ua_type);
1689                 mtx_unlock(&lun->lun_lock);
1690         }
1691 }
1692
1693 static int
1694 ctl_ha_role_sysctl(SYSCTL_HANDLER_ARGS)
1695 {
1696         struct ctl_softc *softc = (struct ctl_softc *)arg1;
1697         struct ctl_lun *lun;
1698         struct ctl_lun_req ireq;
1699         int error, value;
1700
1701         value = (softc->flags & CTL_FLAG_ACTIVE_SHELF) ? 0 : 1;
1702         error = sysctl_handle_int(oidp, &value, 0, req);
1703         if ((error != 0) || (req->newptr == NULL))
1704                 return (error);
1705
1706         mtx_lock(&softc->ctl_lock);
1707         if (value == 0)
1708                 softc->flags |= CTL_FLAG_ACTIVE_SHELF;
1709         else
1710                 softc->flags &= ~CTL_FLAG_ACTIVE_SHELF;
1711         STAILQ_FOREACH(lun, &softc->lun_list, links) {
1712                 mtx_unlock(&softc->ctl_lock);
1713                 bzero(&ireq, sizeof(ireq));
1714                 ireq.reqtype = CTL_LUNREQ_MODIFY;
1715                 ireq.reqdata.modify.lun_id = lun->lun;
1716                 lun->backend->ioctl(NULL, CTL_LUN_REQ, (caddr_t)&ireq, 0,
1717                     curthread);
1718                 if (ireq.status != CTL_LUN_OK) {
1719                         printf("%s: CTL_LUNREQ_MODIFY returned %d '%s'\n",
1720                             __func__, ireq.status, ireq.error_str);
1721                 }
1722                 mtx_lock(&softc->ctl_lock);
1723         }
1724         mtx_unlock(&softc->ctl_lock);
1725         return (0);
1726 }
1727
1728 static int
1729 ctl_init(void)
1730 {
1731         struct ctl_softc *softc;
1732         void *other_pool;
1733         int i, error;
1734
1735         softc = control_softc = malloc(sizeof(*control_softc), M_DEVBUF,
1736                                M_WAITOK | M_ZERO);
1737
1738         softc->dev = make_dev(&ctl_cdevsw, 0, UID_ROOT, GID_OPERATOR, 0600,
1739                               "cam/ctl");
1740         softc->dev->si_drv1 = softc;
1741
1742         sysctl_ctx_init(&softc->sysctl_ctx);
1743         softc->sysctl_tree = SYSCTL_ADD_NODE(&softc->sysctl_ctx,
1744                 SYSCTL_STATIC_CHILDREN(_kern_cam), OID_AUTO, "ctl",
1745                 CTLFLAG_RD, 0, "CAM Target Layer");
1746
1747         if (softc->sysctl_tree == NULL) {
1748                 printf("%s: unable to allocate sysctl tree\n", __func__);
1749                 destroy_dev(softc->dev);
1750                 free(control_softc, M_DEVBUF);
1751                 control_softc = NULL;
1752                 return (ENOMEM);
1753         }
1754
1755         mtx_init(&softc->ctl_lock, "CTL mutex", NULL, MTX_DEF);
1756         softc->io_zone = uma_zcreate("CTL IO", sizeof(union ctl_io),
1757             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1758         softc->flags = 0;
1759
1760         SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1761             OID_AUTO, "ha_mode", CTLFLAG_RDTUN, (int *)&softc->ha_mode, 0,
1762             "HA mode (0 - act/stby, 1 - serialize only, 2 - xfer)");
1763
1764         /*
1765          * In Copan's HA scheme, the "master" and "slave" roles are
1766          * figured out through the slot the controller is in.  Although it
1767          * is an active/active system, someone has to be in charge.
1768          */
1769         SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1770             OID_AUTO, "ha_id", CTLFLAG_RDTUN, &softc->ha_id, 0,
1771             "HA head ID (0 - no HA)");
1772         if (softc->ha_id == 0 || softc->ha_id > NUM_TARGET_PORT_GROUPS) {
1773                 softc->flags |= CTL_FLAG_ACTIVE_SHELF;
1774                 softc->is_single = 1;
1775                 softc->port_cnt = CTL_MAX_PORTS;
1776                 softc->port_min = 0;
1777         } else {
1778                 softc->port_cnt = CTL_MAX_PORTS / NUM_TARGET_PORT_GROUPS;
1779                 softc->port_min = (softc->ha_id - 1) * softc->port_cnt;
1780         }
1781         softc->port_max = softc->port_min + softc->port_cnt;
1782         softc->init_min = softc->port_min * CTL_MAX_INIT_PER_PORT;
1783         softc->init_max = softc->port_max * CTL_MAX_INIT_PER_PORT;
1784
1785         SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1786             OID_AUTO, "ha_link", CTLFLAG_RD, (int *)&softc->ha_link, 0,
1787             "HA link state (0 - offline, 1 - unknown, 2 - online)");
1788
1789         STAILQ_INIT(&softc->lun_list);
1790         STAILQ_INIT(&softc->pending_lun_queue);
1791         STAILQ_INIT(&softc->fe_list);
1792         STAILQ_INIT(&softc->port_list);
1793         STAILQ_INIT(&softc->be_list);
1794         ctl_tpc_init(softc);
1795
1796         if (ctl_pool_create(softc, "othersc", CTL_POOL_ENTRIES_OTHER_SC,
1797                             &other_pool) != 0)
1798         {
1799                 printf("ctl: can't allocate %d entry other SC pool, "
1800                        "exiting\n", CTL_POOL_ENTRIES_OTHER_SC);
1801                 return (ENOMEM);
1802         }
1803         softc->othersc_pool = other_pool;
1804
1805         if (worker_threads <= 0)
1806                 worker_threads = max(1, mp_ncpus / 4);
1807         if (worker_threads > CTL_MAX_THREADS)
1808                 worker_threads = CTL_MAX_THREADS;
1809
1810         for (i = 0; i < worker_threads; i++) {
1811                 struct ctl_thread *thr = &softc->threads[i];
1812
1813                 mtx_init(&thr->queue_lock, "CTL queue mutex", NULL, MTX_DEF);
1814                 thr->ctl_softc = softc;
1815                 STAILQ_INIT(&thr->incoming_queue);
1816                 STAILQ_INIT(&thr->rtr_queue);
1817                 STAILQ_INIT(&thr->done_queue);
1818                 STAILQ_INIT(&thr->isc_queue);
1819
1820                 error = kproc_kthread_add(ctl_work_thread, thr,
1821                     &softc->ctl_proc, &thr->thread, 0, 0, "ctl", "work%d", i);
1822                 if (error != 0) {
1823                         printf("error creating CTL work thread!\n");
1824                         ctl_pool_free(other_pool);
1825                         return (error);
1826                 }
1827         }
1828         error = kproc_kthread_add(ctl_lun_thread, softc,
1829             &softc->ctl_proc, NULL, 0, 0, "ctl", "lun");
1830         if (error != 0) {
1831                 printf("error creating CTL lun thread!\n");
1832                 ctl_pool_free(other_pool);
1833                 return (error);
1834         }
1835         error = kproc_kthread_add(ctl_thresh_thread, softc,
1836             &softc->ctl_proc, NULL, 0, 0, "ctl", "thresh");
1837         if (error != 0) {
1838                 printf("error creating CTL threshold thread!\n");
1839                 ctl_pool_free(other_pool);
1840                 return (error);
1841         }
1842
1843         SYSCTL_ADD_PROC(&softc->sysctl_ctx,SYSCTL_CHILDREN(softc->sysctl_tree),
1844             OID_AUTO, "ha_role", CTLTYPE_INT | CTLFLAG_RWTUN,
1845             softc, 0, ctl_ha_role_sysctl, "I", "HA role for this head");
1846
1847         if (softc->is_single == 0) {
1848                 ctl_frontend_register(&ha_frontend);
1849                 if (ctl_ha_msg_init(softc) != CTL_HA_STATUS_SUCCESS) {
1850                         printf("ctl_init: ctl_ha_msg_init failed.\n");
1851                         softc->is_single = 1;
1852                 } else
1853                 if (ctl_ha_msg_register(CTL_HA_CHAN_CTL, ctl_isc_event_handler)
1854                     != CTL_HA_STATUS_SUCCESS) {
1855                         printf("ctl_init: ctl_ha_msg_register failed.\n");
1856                         softc->is_single = 1;
1857                 }
1858         }
1859         return (0);
1860 }
1861
1862 void
1863 ctl_shutdown(void)
1864 {
1865         struct ctl_softc *softc = control_softc;
1866         struct ctl_lun *lun, *next_lun;
1867
1868         if (softc->is_single == 0) {
1869                 ctl_ha_msg_shutdown(softc);
1870                 if (ctl_ha_msg_deregister(CTL_HA_CHAN_CTL)
1871                     != CTL_HA_STATUS_SUCCESS)
1872                         printf("%s: ctl_ha_msg_deregister failed.\n", __func__);
1873                 if (ctl_ha_msg_destroy(softc) != CTL_HA_STATUS_SUCCESS)
1874                         printf("%s: ctl_ha_msg_destroy failed.\n", __func__);
1875                 ctl_frontend_deregister(&ha_frontend);
1876         }
1877
1878         mtx_lock(&softc->ctl_lock);
1879
1880         /*
1881          * Free up each LUN.
1882          */
1883         for (lun = STAILQ_FIRST(&softc->lun_list); lun != NULL; lun = next_lun){
1884                 next_lun = STAILQ_NEXT(lun, links);
1885                 ctl_free_lun(lun);
1886         }
1887
1888         mtx_unlock(&softc->ctl_lock);
1889
1890 #if 0
1891         ctl_shutdown_thread(softc->work_thread);
1892         mtx_destroy(&softc->queue_lock);
1893 #endif
1894
1895         ctl_tpc_shutdown(softc);
1896         uma_zdestroy(softc->io_zone);
1897         mtx_destroy(&softc->ctl_lock);
1898
1899         destroy_dev(softc->dev);
1900
1901         sysctl_ctx_free(&softc->sysctl_ctx);
1902
1903         free(control_softc, M_DEVBUF);
1904         control_softc = NULL;
1905 }
1906
1907 static int
1908 ctl_module_event_handler(module_t mod, int what, void *arg)
1909 {
1910
1911         switch (what) {
1912         case MOD_LOAD:
1913                 return (ctl_init());
1914         case MOD_UNLOAD:
1915                 return (EBUSY);
1916         default:
1917                 return (EOPNOTSUPP);
1918         }
1919 }
1920
1921 /*
1922  * XXX KDM should we do some access checks here?  Bump a reference count to
1923  * prevent a CTL module from being unloaded while someone has it open?
1924  */
1925 static int
1926 ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td)
1927 {
1928         return (0);
1929 }
1930
1931 static int
1932 ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td)
1933 {
1934         return (0);
1935 }
1936
1937 /*
1938  * Remove an initiator by port number and initiator ID.
1939  * Returns 0 for success, -1 for failure.
1940  */
1941 int
1942 ctl_remove_initiator(struct ctl_port *port, int iid)
1943 {
1944         struct ctl_softc *softc = port->ctl_softc;
1945
1946         mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1947
1948         if (iid > CTL_MAX_INIT_PER_PORT) {
1949                 printf("%s: initiator ID %u > maximun %u!\n",
1950                        __func__, iid, CTL_MAX_INIT_PER_PORT);
1951                 return (-1);
1952         }
1953
1954         mtx_lock(&softc->ctl_lock);
1955         port->wwpn_iid[iid].in_use--;
1956         port->wwpn_iid[iid].last_use = time_uptime;
1957         mtx_unlock(&softc->ctl_lock);
1958         ctl_isc_announce_iid(port, iid);
1959
1960         return (0);
1961 }
1962
1963 /*
1964  * Add an initiator to the initiator map.
1965  * Returns iid for success, < 0 for failure.
1966  */
1967 int
1968 ctl_add_initiator(struct ctl_port *port, int iid, uint64_t wwpn, char *name)
1969 {
1970         struct ctl_softc *softc = port->ctl_softc;
1971         time_t best_time;
1972         int i, best;
1973
1974         mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1975
1976         if (iid >= CTL_MAX_INIT_PER_PORT) {
1977                 printf("%s: WWPN %#jx initiator ID %u > maximum %u!\n",
1978                        __func__, wwpn, iid, CTL_MAX_INIT_PER_PORT);
1979                 free(name, M_CTL);
1980                 return (-1);
1981         }
1982
1983         mtx_lock(&softc->ctl_lock);
1984
1985         if (iid < 0 && (wwpn != 0 || name != NULL)) {
1986                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1987                         if (wwpn != 0 && wwpn == port->wwpn_iid[i].wwpn) {
1988                                 iid = i;
1989                                 break;
1990                         }
1991                         if (name != NULL && port->wwpn_iid[i].name != NULL &&
1992                             strcmp(name, port->wwpn_iid[i].name) == 0) {
1993                                 iid = i;
1994                                 break;
1995                         }
1996                 }
1997         }
1998
1999         if (iid < 0) {
2000                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
2001                         if (port->wwpn_iid[i].in_use == 0 &&
2002                             port->wwpn_iid[i].wwpn == 0 &&
2003                             port->wwpn_iid[i].name == NULL) {
2004                                 iid = i;
2005                                 break;
2006                         }
2007                 }
2008         }
2009
2010         if (iid < 0) {
2011                 best = -1;
2012                 best_time = INT32_MAX;
2013                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
2014                         if (port->wwpn_iid[i].in_use == 0) {
2015                                 if (port->wwpn_iid[i].last_use < best_time) {
2016                                         best = i;
2017                                         best_time = port->wwpn_iid[i].last_use;
2018                                 }
2019                         }
2020                 }
2021                 iid = best;
2022         }
2023
2024         if (iid < 0) {
2025                 mtx_unlock(&softc->ctl_lock);
2026                 free(name, M_CTL);
2027                 return (-2);
2028         }
2029
2030         if (port->wwpn_iid[iid].in_use > 0 && (wwpn != 0 || name != NULL)) {
2031                 /*
2032                  * This is not an error yet.
2033                  */
2034                 if (wwpn != 0 && wwpn == port->wwpn_iid[iid].wwpn) {
2035 #if 0
2036                         printf("%s: port %d iid %u WWPN %#jx arrived"
2037                             " again\n", __func__, port->targ_port,
2038                             iid, (uintmax_t)wwpn);
2039 #endif
2040                         goto take;
2041                 }
2042                 if (name != NULL && port->wwpn_iid[iid].name != NULL &&
2043                     strcmp(name, port->wwpn_iid[iid].name) == 0) {
2044 #if 0
2045                         printf("%s: port %d iid %u name '%s' arrived"
2046                             " again\n", __func__, port->targ_port,
2047                             iid, name);
2048 #endif
2049                         goto take;
2050                 }
2051
2052                 /*
2053                  * This is an error, but what do we do about it?  The
2054                  * driver is telling us we have a new WWPN for this
2055                  * initiator ID, so we pretty much need to use it.
2056                  */
2057                 printf("%s: port %d iid %u WWPN %#jx '%s' arrived,"
2058                     " but WWPN %#jx '%s' is still at that address\n",
2059                     __func__, port->targ_port, iid, wwpn, name,
2060                     (uintmax_t)port->wwpn_iid[iid].wwpn,
2061                     port->wwpn_iid[iid].name);
2062
2063                 /*
2064                  * XXX KDM clear have_ca and ua_pending on each LUN for
2065                  * this initiator.
2066                  */
2067         }
2068 take:
2069         free(port->wwpn_iid[iid].name, M_CTL);
2070         port->wwpn_iid[iid].name = name;
2071         port->wwpn_iid[iid].wwpn = wwpn;
2072         port->wwpn_iid[iid].in_use++;
2073         mtx_unlock(&softc->ctl_lock);
2074         ctl_isc_announce_iid(port, iid);
2075
2076         return (iid);
2077 }
2078
2079 static int
2080 ctl_create_iid(struct ctl_port *port, int iid, uint8_t *buf)
2081 {
2082         int len;
2083
2084         switch (port->port_type) {
2085         case CTL_PORT_FC:
2086         {
2087                 struct scsi_transportid_fcp *id =
2088                     (struct scsi_transportid_fcp *)buf;
2089                 if (port->wwpn_iid[iid].wwpn == 0)
2090                         return (0);
2091                 memset(id, 0, sizeof(*id));
2092                 id->format_protocol = SCSI_PROTO_FC;
2093                 scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->n_port_name);
2094                 return (sizeof(*id));
2095         }
2096         case CTL_PORT_ISCSI:
2097         {
2098                 struct scsi_transportid_iscsi_port *id =
2099                     (struct scsi_transportid_iscsi_port *)buf;
2100                 if (port->wwpn_iid[iid].name == NULL)
2101                         return (0);
2102                 memset(id, 0, 256);
2103                 id->format_protocol = SCSI_TRN_ISCSI_FORMAT_PORT |
2104                     SCSI_PROTO_ISCSI;
2105                 len = strlcpy(id->iscsi_name, port->wwpn_iid[iid].name, 252) + 1;
2106                 len = roundup2(min(len, 252), 4);
2107                 scsi_ulto2b(len, id->additional_length);
2108                 return (sizeof(*id) + len);
2109         }
2110         case CTL_PORT_SAS:
2111         {
2112                 struct scsi_transportid_sas *id =
2113                     (struct scsi_transportid_sas *)buf;
2114                 if (port->wwpn_iid[iid].wwpn == 0)
2115                         return (0);
2116                 memset(id, 0, sizeof(*id));
2117                 id->format_protocol = SCSI_PROTO_SAS;
2118                 scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->sas_address);
2119                 return (sizeof(*id));
2120         }
2121         default:
2122         {
2123                 struct scsi_transportid_spi *id =
2124                     (struct scsi_transportid_spi *)buf;
2125                 memset(id, 0, sizeof(*id));
2126                 id->format_protocol = SCSI_PROTO_SPI;
2127                 scsi_ulto2b(iid, id->scsi_addr);
2128                 scsi_ulto2b(port->targ_port, id->rel_trgt_port_id);
2129                 return (sizeof(*id));
2130         }
2131         }
2132 }
2133
2134 /*
2135  * Serialize a command that went down the "wrong" side, and so was sent to
2136  * this controller for execution.  The logic is a little different than the
2137  * standard case in ctl_scsiio_precheck().  Errors in this case need to get
2138  * sent back to the other side, but in the success case, we execute the
2139  * command on this side (XFER mode) or tell the other side to execute it
2140  * (SER_ONLY mode).
2141  */
2142 static int
2143 ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio)
2144 {
2145         struct ctl_softc *softc = control_softc;
2146         union ctl_ha_msg msg_info;
2147         struct ctl_port *port;
2148         struct ctl_lun *lun;
2149         const struct ctl_cmd_entry *entry;
2150         int retval = 0;
2151         uint32_t targ_lun;
2152
2153         targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
2154         mtx_lock(&softc->ctl_lock);
2155
2156         /* Make sure that we know about this port. */
2157         port = ctl_io_port(&ctsio->io_hdr);
2158         if (port == NULL || (port->status & CTL_PORT_STATUS_ONLINE) == 0) {
2159                 ctl_set_internal_failure(ctsio, /*sks_valid*/ 0,
2160                                          /*retry_count*/ 1);
2161                 goto badjuju;
2162         }
2163
2164         /* Make sure that we know about this LUN. */
2165         if ((targ_lun < CTL_MAX_LUNS) &&
2166             ((lun = softc->ctl_luns[targ_lun]) != NULL)) {
2167                 mtx_lock(&lun->lun_lock);
2168                 mtx_unlock(&softc->ctl_lock);
2169                 /*
2170                  * If the LUN is invalid, pretend that it doesn't exist.
2171                  * It will go away as soon as all pending I/O has been
2172                  * completed.
2173                  */
2174                 if (lun->flags & CTL_LUN_DISABLED) {
2175                         mtx_unlock(&lun->lun_lock);
2176                         lun = NULL;
2177                 }
2178         } else {
2179                 mtx_unlock(&softc->ctl_lock);
2180                 lun = NULL;
2181         }
2182         if (lun == NULL) {
2183                 /*
2184                  * The other node would not send this request to us unless
2185                  * received announce that we are primary node for this LUN.
2186                  * If this LUN does not exist now, it is probably result of
2187                  * a race, so respond to initiator in the most opaque way.
2188                  */
2189                 ctl_set_busy(ctsio);
2190                 goto badjuju;
2191         }
2192
2193         entry = ctl_get_cmd_entry(ctsio, NULL);
2194         if (ctl_scsiio_lun_check(lun, entry, ctsio) != 0) {
2195                 mtx_unlock(&lun->lun_lock);
2196                 goto badjuju;
2197         }
2198
2199         ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = lun;
2200         ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = lun->be_lun;
2201
2202         /*
2203          * Every I/O goes into the OOA queue for a
2204          * particular LUN, and stays there until completion.
2205          */
2206 #ifdef CTL_TIME_IO
2207         if (TAILQ_EMPTY(&lun->ooa_queue))
2208                 lun->idle_time += getsbinuptime() - lun->last_busy;
2209 #endif
2210         TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2211
2212         switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
2213                 (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr, ctl_ooaq,
2214                  ooa_links))) {
2215         case CTL_ACTION_BLOCK:
2216                 ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
2217                 TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
2218                                   blocked_links);
2219                 mtx_unlock(&lun->lun_lock);
2220                 break;
2221         case CTL_ACTION_PASS:
2222         case CTL_ACTION_SKIP:
2223                 if (softc->ha_mode == CTL_HA_MODE_XFER) {
2224                         ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
2225                         ctl_enqueue_rtr((union ctl_io *)ctsio);
2226                         mtx_unlock(&lun->lun_lock);
2227                 } else {
2228                         ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
2229                         mtx_unlock(&lun->lun_lock);
2230
2231                         /* send msg back to other side */
2232                         msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
2233                         msg_info.hdr.serializing_sc = (union ctl_io *)ctsio;
2234                         msg_info.hdr.msg_type = CTL_MSG_R2R;
2235                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
2236                             sizeof(msg_info.hdr), M_WAITOK);
2237                 }
2238                 break;
2239         case CTL_ACTION_OVERLAP:
2240                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2241                 mtx_unlock(&lun->lun_lock);
2242                 ctl_set_overlapped_cmd(ctsio);
2243                 goto badjuju;
2244         case CTL_ACTION_OVERLAP_TAG:
2245                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2246                 mtx_unlock(&lun->lun_lock);
2247                 ctl_set_overlapped_tag(ctsio, ctsio->tag_num);
2248                 goto badjuju;
2249         case CTL_ACTION_ERROR:
2250         default:
2251                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2252                 mtx_unlock(&lun->lun_lock);
2253
2254                 ctl_set_internal_failure(ctsio, /*sks_valid*/ 0,
2255                                          /*retry_count*/ 0);
2256 badjuju:
2257                 ctl_copy_sense_data_back((union ctl_io *)ctsio, &msg_info);
2258                 msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
2259                 msg_info.hdr.serializing_sc = NULL;
2260                 msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
2261                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
2262                     sizeof(msg_info.scsi), M_WAITOK);
2263                 retval = 1;
2264                 break;
2265         }
2266         return (retval);
2267 }
2268
2269 /*
2270  * Returns 0 for success, errno for failure.
2271  */
2272 static void
2273 ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
2274                    struct ctl_ooa *ooa_hdr, struct ctl_ooa_entry *kern_entries)
2275 {
2276         union ctl_io *io;
2277
2278         mtx_lock(&lun->lun_lock);
2279         for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); (io != NULL);
2280              (*cur_fill_num)++, io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2281              ooa_links)) {
2282                 struct ctl_ooa_entry *entry;
2283
2284                 /*
2285                  * If we've got more than we can fit, just count the
2286                  * remaining entries.
2287                  */
2288                 if (*cur_fill_num >= ooa_hdr->alloc_num)
2289                         continue;
2290
2291                 entry = &kern_entries[*cur_fill_num];
2292
2293                 entry->tag_num = io->scsiio.tag_num;
2294                 entry->lun_num = lun->lun;
2295 #ifdef CTL_TIME_IO
2296                 entry->start_bt = io->io_hdr.start_bt;
2297 #endif
2298                 bcopy(io->scsiio.cdb, entry->cdb, io->scsiio.cdb_len);
2299                 entry->cdb_len = io->scsiio.cdb_len;
2300                 if (io->io_hdr.flags & CTL_FLAG_BLOCKED)
2301                         entry->cmd_flags |= CTL_OOACMD_FLAG_BLOCKED;
2302
2303                 if (io->io_hdr.flags & CTL_FLAG_DMA_INPROG)
2304                         entry->cmd_flags |= CTL_OOACMD_FLAG_DMA;
2305
2306                 if (io->io_hdr.flags & CTL_FLAG_ABORT)
2307                         entry->cmd_flags |= CTL_OOACMD_FLAG_ABORT;
2308
2309                 if (io->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR)
2310                         entry->cmd_flags |= CTL_OOACMD_FLAG_RTR;
2311
2312                 if (io->io_hdr.flags & CTL_FLAG_DMA_QUEUED)
2313                         entry->cmd_flags |= CTL_OOACMD_FLAG_DMA_QUEUED;
2314         }
2315         mtx_unlock(&lun->lun_lock);
2316 }
2317
2318 static void *
2319 ctl_copyin_alloc(void *user_addr, int len, char *error_str,
2320                  size_t error_str_len)
2321 {
2322         void *kptr;
2323
2324         kptr = malloc(len, M_CTL, M_WAITOK | M_ZERO);
2325
2326         if (copyin(user_addr, kptr, len) != 0) {
2327                 snprintf(error_str, error_str_len, "Error copying %d bytes "
2328                          "from user address %p to kernel address %p", len,
2329                          user_addr, kptr);
2330                 free(kptr, M_CTL);
2331                 return (NULL);
2332         }
2333
2334         return (kptr);
2335 }
2336
2337 static void
2338 ctl_free_args(int num_args, struct ctl_be_arg *args)
2339 {
2340         int i;
2341
2342         if (args == NULL)
2343                 return;
2344
2345         for (i = 0; i < num_args; i++) {
2346                 free(args[i].kname, M_CTL);
2347                 free(args[i].kvalue, M_CTL);
2348         }
2349
2350         free(args, M_CTL);
2351 }
2352
2353 static struct ctl_be_arg *
2354 ctl_copyin_args(int num_args, struct ctl_be_arg *uargs,
2355                 char *error_str, size_t error_str_len)
2356 {
2357         struct ctl_be_arg *args;
2358         int i;
2359
2360         args = ctl_copyin_alloc(uargs, num_args * sizeof(*args),
2361                                 error_str, error_str_len);
2362
2363         if (args == NULL)
2364                 goto bailout;
2365
2366         for (i = 0; i < num_args; i++) {
2367                 args[i].kname = NULL;
2368                 args[i].kvalue = NULL;
2369         }
2370
2371         for (i = 0; i < num_args; i++) {
2372                 uint8_t *tmpptr;
2373
2374                 args[i].kname = ctl_copyin_alloc(args[i].name,
2375                         args[i].namelen, error_str, error_str_len);
2376                 if (args[i].kname == NULL)
2377                         goto bailout;
2378
2379                 if (args[i].kname[args[i].namelen - 1] != '\0') {
2380                         snprintf(error_str, error_str_len, "Argument %d "
2381                                  "name is not NUL-terminated", i);
2382                         goto bailout;
2383                 }
2384
2385                 if (args[i].flags & CTL_BEARG_RD) {
2386                         tmpptr = ctl_copyin_alloc(args[i].value,
2387                                 args[i].vallen, error_str, error_str_len);
2388                         if (tmpptr == NULL)
2389                                 goto bailout;
2390                         if ((args[i].flags & CTL_BEARG_ASCII)
2391                          && (tmpptr[args[i].vallen - 1] != '\0')) {
2392                                 snprintf(error_str, error_str_len, "Argument "
2393                                     "%d value is not NUL-terminated", i);
2394                                 goto bailout;
2395                         }
2396                         args[i].kvalue = tmpptr;
2397                 } else {
2398                         args[i].kvalue = malloc(args[i].vallen,
2399                             M_CTL, M_WAITOK | M_ZERO);
2400                 }
2401         }
2402
2403         return (args);
2404 bailout:
2405
2406         ctl_free_args(num_args, args);
2407
2408         return (NULL);
2409 }
2410
2411 static void
2412 ctl_copyout_args(int num_args, struct ctl_be_arg *args)
2413 {
2414         int i;
2415
2416         for (i = 0; i < num_args; i++) {
2417                 if (args[i].flags & CTL_BEARG_WR)
2418                         copyout(args[i].kvalue, args[i].value, args[i].vallen);
2419         }
2420 }
2421
2422 /*
2423  * Escape characters that are illegal or not recommended in XML.
2424  */
2425 int
2426 ctl_sbuf_printf_esc(struct sbuf *sb, char *str, int size)
2427 {
2428         char *end = str + size;
2429         int retval;
2430
2431         retval = 0;
2432
2433         for (; *str && str < end; str++) {
2434                 switch (*str) {
2435                 case '&':
2436                         retval = sbuf_printf(sb, "&amp;");
2437                         break;
2438                 case '>':
2439                         retval = sbuf_printf(sb, "&gt;");
2440                         break;
2441                 case '<':
2442                         retval = sbuf_printf(sb, "&lt;");
2443                         break;
2444                 default:
2445                         retval = sbuf_putc(sb, *str);
2446                         break;
2447                 }
2448
2449                 if (retval != 0)
2450                         break;
2451
2452         }
2453
2454         return (retval);
2455 }
2456
2457 static void
2458 ctl_id_sbuf(struct ctl_devid *id, struct sbuf *sb)
2459 {
2460         struct scsi_vpd_id_descriptor *desc;
2461         int i;
2462
2463         if (id == NULL || id->len < 4)
2464                 return;
2465         desc = (struct scsi_vpd_id_descriptor *)id->data;
2466         switch (desc->id_type & SVPD_ID_TYPE_MASK) {
2467         case SVPD_ID_TYPE_T10:
2468                 sbuf_printf(sb, "t10.");
2469                 break;
2470         case SVPD_ID_TYPE_EUI64:
2471                 sbuf_printf(sb, "eui.");
2472                 break;
2473         case SVPD_ID_TYPE_NAA:
2474                 sbuf_printf(sb, "naa.");
2475                 break;
2476         case SVPD_ID_TYPE_SCSI_NAME:
2477                 break;
2478         }
2479         switch (desc->proto_codeset & SVPD_ID_CODESET_MASK) {
2480         case SVPD_ID_CODESET_BINARY:
2481                 for (i = 0; i < desc->length; i++)
2482                         sbuf_printf(sb, "%02x", desc->identifier[i]);
2483                 break;
2484         case SVPD_ID_CODESET_ASCII:
2485                 sbuf_printf(sb, "%.*s", (int)desc->length,
2486                     (char *)desc->identifier);
2487                 break;
2488         case SVPD_ID_CODESET_UTF8:
2489                 sbuf_printf(sb, "%s", (char *)desc->identifier);
2490                 break;
2491         }
2492 }
2493
2494 static int
2495 ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
2496           struct thread *td)
2497 {
2498         struct ctl_softc *softc = dev->si_drv1;
2499         struct ctl_lun *lun;
2500         int retval;
2501
2502         retval = 0;
2503
2504         switch (cmd) {
2505         case CTL_IO:
2506                 retval = ctl_ioctl_io(dev, cmd, addr, flag, td);
2507                 break;
2508         case CTL_ENABLE_PORT:
2509         case CTL_DISABLE_PORT:
2510         case CTL_SET_PORT_WWNS: {
2511                 struct ctl_port *port;
2512                 struct ctl_port_entry *entry;
2513
2514                 entry = (struct ctl_port_entry *)addr;
2515
2516                 mtx_lock(&softc->ctl_lock);
2517                 STAILQ_FOREACH(port, &softc->port_list, links) {
2518                         int action, done;
2519
2520                         if (port->targ_port < softc->port_min ||
2521                             port->targ_port >= softc->port_max)
2522                                 continue;
2523
2524                         action = 0;
2525                         done = 0;
2526                         if ((entry->port_type == CTL_PORT_NONE)
2527                          && (entry->targ_port == port->targ_port)) {
2528                                 /*
2529                                  * If the user only wants to enable or
2530                                  * disable or set WWNs on a specific port,
2531                                  * do the operation and we're done.
2532                                  */
2533                                 action = 1;
2534                                 done = 1;
2535                         } else if (entry->port_type & port->port_type) {
2536                                 /*
2537                                  * Compare the user's type mask with the
2538                                  * particular frontend type to see if we
2539                                  * have a match.
2540                                  */
2541                                 action = 1;
2542                                 done = 0;
2543
2544                                 /*
2545                                  * Make sure the user isn't trying to set
2546                                  * WWNs on multiple ports at the same time.
2547                                  */
2548                                 if (cmd == CTL_SET_PORT_WWNS) {
2549                                         printf("%s: Can't set WWNs on "
2550                                                "multiple ports\n", __func__);
2551                                         retval = EINVAL;
2552                                         break;
2553                                 }
2554                         }
2555                         if (action == 0)
2556                                 continue;
2557
2558                         /*
2559                          * XXX KDM we have to drop the lock here, because
2560                          * the online/offline operations can potentially
2561                          * block.  We need to reference count the frontends
2562                          * so they can't go away,
2563                          */
2564                         if (cmd == CTL_ENABLE_PORT) {
2565                                 mtx_unlock(&softc->ctl_lock);
2566                                 ctl_port_online(port);
2567                                 mtx_lock(&softc->ctl_lock);
2568                         } else if (cmd == CTL_DISABLE_PORT) {
2569                                 mtx_unlock(&softc->ctl_lock);
2570                                 ctl_port_offline(port);
2571                                 mtx_lock(&softc->ctl_lock);
2572                         } else if (cmd == CTL_SET_PORT_WWNS) {
2573                                 ctl_port_set_wwns(port,
2574                                     (entry->flags & CTL_PORT_WWNN_VALID) ?
2575                                     1 : 0, entry->wwnn,
2576                                     (entry->flags & CTL_PORT_WWPN_VALID) ?
2577                                     1 : 0, entry->wwpn);
2578                         }
2579                         if (done != 0)
2580                                 break;
2581                 }
2582                 mtx_unlock(&softc->ctl_lock);
2583                 break;
2584         }
2585         case CTL_GET_OOA: {
2586                 struct ctl_ooa *ooa_hdr;
2587                 struct ctl_ooa_entry *entries;
2588                 uint32_t cur_fill_num;
2589
2590                 ooa_hdr = (struct ctl_ooa *)addr;
2591
2592                 if ((ooa_hdr->alloc_len == 0)
2593                  || (ooa_hdr->alloc_num == 0)) {
2594                         printf("%s: CTL_GET_OOA: alloc len %u and alloc num %u "
2595                                "must be non-zero\n", __func__,
2596                                ooa_hdr->alloc_len, ooa_hdr->alloc_num);
2597                         retval = EINVAL;
2598                         break;
2599                 }
2600
2601                 if (ooa_hdr->alloc_len != (ooa_hdr->alloc_num *
2602                     sizeof(struct ctl_ooa_entry))) {
2603                         printf("%s: CTL_GET_OOA: alloc len %u must be alloc "
2604                                "num %d * sizeof(struct ctl_ooa_entry) %zd\n",
2605                                __func__, ooa_hdr->alloc_len,
2606                                ooa_hdr->alloc_num,sizeof(struct ctl_ooa_entry));
2607                         retval = EINVAL;
2608                         break;
2609                 }
2610
2611                 entries = malloc(ooa_hdr->alloc_len, M_CTL, M_WAITOK | M_ZERO);
2612                 if (entries == NULL) {
2613                         printf("%s: could not allocate %d bytes for OOA "
2614                                "dump\n", __func__, ooa_hdr->alloc_len);
2615                         retval = ENOMEM;
2616                         break;
2617                 }
2618
2619                 mtx_lock(&softc->ctl_lock);
2620                 if (((ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) == 0)
2621                  && ((ooa_hdr->lun_num >= CTL_MAX_LUNS)
2622                   || (softc->ctl_luns[ooa_hdr->lun_num] == NULL))) {
2623                         mtx_unlock(&softc->ctl_lock);
2624                         free(entries, M_CTL);
2625                         printf("%s: CTL_GET_OOA: invalid LUN %ju\n",
2626                                __func__, (uintmax_t)ooa_hdr->lun_num);
2627                         retval = EINVAL;
2628                         break;
2629                 }
2630
2631                 cur_fill_num = 0;
2632
2633                 if (ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) {
2634                         STAILQ_FOREACH(lun, &softc->lun_list, links) {
2635                                 ctl_ioctl_fill_ooa(lun, &cur_fill_num,
2636                                     ooa_hdr, entries);
2637                         }
2638                 } else {
2639                         lun = softc->ctl_luns[ooa_hdr->lun_num];
2640                         ctl_ioctl_fill_ooa(lun, &cur_fill_num, ooa_hdr,
2641                             entries);
2642                 }
2643                 mtx_unlock(&softc->ctl_lock);
2644
2645                 ooa_hdr->fill_num = min(cur_fill_num, ooa_hdr->alloc_num);
2646                 ooa_hdr->fill_len = ooa_hdr->fill_num *
2647                         sizeof(struct ctl_ooa_entry);
2648                 retval = copyout(entries, ooa_hdr->entries, ooa_hdr->fill_len);
2649                 if (retval != 0) {
2650                         printf("%s: error copying out %d bytes for OOA dump\n",
2651                                __func__, ooa_hdr->fill_len);
2652                 }
2653
2654                 getbinuptime(&ooa_hdr->cur_bt);
2655
2656                 if (cur_fill_num > ooa_hdr->alloc_num) {
2657                         ooa_hdr->dropped_num = cur_fill_num -ooa_hdr->alloc_num;
2658                         ooa_hdr->status = CTL_OOA_NEED_MORE_SPACE;
2659                 } else {
2660                         ooa_hdr->dropped_num = 0;
2661                         ooa_hdr->status = CTL_OOA_OK;
2662                 }
2663
2664                 free(entries, M_CTL);
2665                 break;
2666         }
2667         case CTL_DELAY_IO: {
2668                 struct ctl_io_delay_info *delay_info;
2669
2670                 delay_info = (struct ctl_io_delay_info *)addr;
2671
2672 #ifdef CTL_IO_DELAY
2673                 mtx_lock(&softc->ctl_lock);
2674
2675                 if ((delay_info->lun_id >= CTL_MAX_LUNS)
2676                  || (softc->ctl_luns[delay_info->lun_id] == NULL)) {
2677                         delay_info->status = CTL_DELAY_STATUS_INVALID_LUN;
2678                 } else {
2679                         lun = softc->ctl_luns[delay_info->lun_id];
2680                         mtx_lock(&lun->lun_lock);
2681
2682                         delay_info->status = CTL_DELAY_STATUS_OK;
2683
2684                         switch (delay_info->delay_type) {
2685                         case CTL_DELAY_TYPE_CONT:
2686                                 break;
2687                         case CTL_DELAY_TYPE_ONESHOT:
2688                                 break;
2689                         default:
2690                                 delay_info->status =
2691                                         CTL_DELAY_STATUS_INVALID_TYPE;
2692                                 break;
2693                         }
2694
2695                         switch (delay_info->delay_loc) {
2696                         case CTL_DELAY_LOC_DATAMOVE:
2697                                 lun->delay_info.datamove_type =
2698                                         delay_info->delay_type;
2699                                 lun->delay_info.datamove_delay =
2700                                         delay_info->delay_secs;
2701                                 break;
2702                         case CTL_DELAY_LOC_DONE:
2703                                 lun->delay_info.done_type =
2704                                         delay_info->delay_type;
2705                                 lun->delay_info.done_delay =
2706                                         delay_info->delay_secs;
2707                                 break;
2708                         default:
2709                                 delay_info->status =
2710                                         CTL_DELAY_STATUS_INVALID_LOC;
2711                                 break;
2712                         }
2713                         mtx_unlock(&lun->lun_lock);
2714                 }
2715
2716                 mtx_unlock(&softc->ctl_lock);
2717 #else
2718                 delay_info->status = CTL_DELAY_STATUS_NOT_IMPLEMENTED;
2719 #endif /* CTL_IO_DELAY */
2720                 break;
2721         }
2722         case CTL_GETSTATS: {
2723                 struct ctl_stats *stats;
2724                 int i;
2725
2726                 stats = (struct ctl_stats *)addr;
2727
2728                 if ((sizeof(struct ctl_lun_io_stats) * softc->num_luns) >
2729                      stats->alloc_len) {
2730                         stats->status = CTL_SS_NEED_MORE_SPACE;
2731                         stats->num_luns = softc->num_luns;
2732                         break;
2733                 }
2734                 /*
2735                  * XXX KDM no locking here.  If the LUN list changes,
2736                  * things can blow up.
2737                  */
2738                 for (i = 0, lun = STAILQ_FIRST(&softc->lun_list); lun != NULL;
2739                      i++, lun = STAILQ_NEXT(lun, links)) {
2740                         retval = copyout(&lun->stats, &stats->lun_stats[i],
2741                                          sizeof(lun->stats));
2742                         if (retval != 0)
2743                                 break;
2744                 }
2745                 stats->num_luns = softc->num_luns;
2746                 stats->fill_len = sizeof(struct ctl_lun_io_stats) *
2747                                  softc->num_luns;
2748                 stats->status = CTL_SS_OK;
2749 #ifdef CTL_TIME_IO
2750                 stats->flags = CTL_STATS_FLAG_TIME_VALID;
2751 #else
2752                 stats->flags = CTL_STATS_FLAG_NONE;
2753 #endif
2754                 getnanouptime(&stats->timestamp);
2755                 break;
2756         }
2757         case CTL_ERROR_INJECT: {
2758                 struct ctl_error_desc *err_desc, *new_err_desc;
2759
2760                 err_desc = (struct ctl_error_desc *)addr;
2761
2762                 new_err_desc = malloc(sizeof(*new_err_desc), M_CTL,
2763                                       M_WAITOK | M_ZERO);
2764                 bcopy(err_desc, new_err_desc, sizeof(*new_err_desc));
2765
2766                 mtx_lock(&softc->ctl_lock);
2767                 lun = softc->ctl_luns[err_desc->lun_id];
2768                 if (lun == NULL) {
2769                         mtx_unlock(&softc->ctl_lock);
2770                         free(new_err_desc, M_CTL);
2771                         printf("%s: CTL_ERROR_INJECT: invalid LUN %ju\n",
2772                                __func__, (uintmax_t)err_desc->lun_id);
2773                         retval = EINVAL;
2774                         break;
2775                 }
2776                 mtx_lock(&lun->lun_lock);
2777                 mtx_unlock(&softc->ctl_lock);
2778
2779                 /*
2780                  * We could do some checking here to verify the validity
2781                  * of the request, but given the complexity of error
2782                  * injection requests, the checking logic would be fairly
2783                  * complex.
2784                  *
2785                  * For now, if the request is invalid, it just won't get
2786                  * executed and might get deleted.
2787                  */
2788                 STAILQ_INSERT_TAIL(&lun->error_list, new_err_desc, links);
2789
2790                 /*
2791                  * XXX KDM check to make sure the serial number is unique,
2792                  * in case we somehow manage to wrap.  That shouldn't
2793                  * happen for a very long time, but it's the right thing to
2794                  * do.
2795                  */
2796                 new_err_desc->serial = lun->error_serial;
2797                 err_desc->serial = lun->error_serial;
2798                 lun->error_serial++;
2799
2800                 mtx_unlock(&lun->lun_lock);
2801                 break;
2802         }
2803         case CTL_ERROR_INJECT_DELETE: {
2804                 struct ctl_error_desc *delete_desc, *desc, *desc2;
2805                 int delete_done;
2806
2807                 delete_desc = (struct ctl_error_desc *)addr;
2808                 delete_done = 0;
2809
2810                 mtx_lock(&softc->ctl_lock);
2811                 lun = softc->ctl_luns[delete_desc->lun_id];
2812                 if (lun == NULL) {
2813                         mtx_unlock(&softc->ctl_lock);
2814                         printf("%s: CTL_ERROR_INJECT_DELETE: invalid LUN %ju\n",
2815                                __func__, (uintmax_t)delete_desc->lun_id);
2816                         retval = EINVAL;
2817                         break;
2818                 }
2819                 mtx_lock(&lun->lun_lock);
2820                 mtx_unlock(&softc->ctl_lock);
2821                 STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
2822                         if (desc->serial != delete_desc->serial)
2823                                 continue;
2824
2825                         STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc,
2826                                       links);
2827                         free(desc, M_CTL);
2828                         delete_done = 1;
2829                 }
2830                 mtx_unlock(&lun->lun_lock);
2831                 if (delete_done == 0) {
2832                         printf("%s: CTL_ERROR_INJECT_DELETE: can't find "
2833                                "error serial %ju on LUN %u\n", __func__,
2834                                delete_desc->serial, delete_desc->lun_id);
2835                         retval = EINVAL;
2836                         break;
2837                 }
2838                 break;
2839         }
2840         case CTL_DUMP_STRUCTS: {
2841                 int i, j, k;
2842                 struct ctl_port *port;
2843                 struct ctl_frontend *fe;
2844
2845                 mtx_lock(&softc->ctl_lock);
2846                 printf("CTL Persistent Reservation information start:\n");
2847                 for (i = 0; i < CTL_MAX_LUNS; i++) {
2848                         lun = softc->ctl_luns[i];
2849
2850                         if ((lun == NULL)
2851                          || ((lun->flags & CTL_LUN_DISABLED) != 0))
2852                                 continue;
2853
2854                         for (j = 0; j < CTL_MAX_PORTS; j++) {
2855                                 if (lun->pr_keys[j] == NULL)
2856                                         continue;
2857                                 for (k = 0; k < CTL_MAX_INIT_PER_PORT; k++){
2858                                         if (lun->pr_keys[j][k] == 0)
2859                                                 continue;
2860                                         printf("  LUN %d port %d iid %d key "
2861                                                "%#jx\n", i, j, k,
2862                                                (uintmax_t)lun->pr_keys[j][k]);
2863                                 }
2864                         }
2865                 }
2866                 printf("CTL Persistent Reservation information end\n");
2867                 printf("CTL Ports:\n");
2868                 STAILQ_FOREACH(port, &softc->port_list, links) {
2869                         printf("  Port %d '%s' Frontend '%s' Type %u pp %d vp %d WWNN "
2870                                "%#jx WWPN %#jx\n", port->targ_port, port->port_name,
2871                                port->frontend->name, port->port_type,
2872                                port->physical_port, port->virtual_port,
2873                                (uintmax_t)port->wwnn, (uintmax_t)port->wwpn);
2874                         for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
2875                                 if (port->wwpn_iid[j].in_use == 0 &&
2876                                     port->wwpn_iid[j].wwpn == 0 &&
2877                                     port->wwpn_iid[j].name == NULL)
2878                                         continue;
2879
2880                                 printf("    iid %u use %d WWPN %#jx '%s'\n",
2881                                     j, port->wwpn_iid[j].in_use,
2882                                     (uintmax_t)port->wwpn_iid[j].wwpn,
2883                                     port->wwpn_iid[j].name);
2884                         }
2885                 }
2886                 printf("CTL Port information end\n");
2887                 mtx_unlock(&softc->ctl_lock);
2888                 /*
2889                  * XXX KDM calling this without a lock.  We'd likely want
2890                  * to drop the lock before calling the frontend's dump
2891                  * routine anyway.
2892                  */
2893                 printf("CTL Frontends:\n");
2894                 STAILQ_FOREACH(fe, &softc->fe_list, links) {
2895                         printf("  Frontend '%s'\n", fe->name);
2896                         if (fe->fe_dump != NULL)
2897                                 fe->fe_dump();
2898                 }
2899                 printf("CTL Frontend information end\n");
2900                 break;
2901         }
2902         case CTL_LUN_REQ: {
2903                 struct ctl_lun_req *lun_req;
2904                 struct ctl_backend_driver *backend;
2905
2906                 lun_req = (struct ctl_lun_req *)addr;
2907
2908                 backend = ctl_backend_find(lun_req->backend);
2909                 if (backend == NULL) {
2910                         lun_req->status = CTL_LUN_ERROR;
2911                         snprintf(lun_req->error_str,
2912                                  sizeof(lun_req->error_str),
2913                                  "Backend \"%s\" not found.",
2914                                  lun_req->backend);
2915                         break;
2916                 }
2917                 if (lun_req->num_be_args > 0) {
2918                         lun_req->kern_be_args = ctl_copyin_args(
2919                                 lun_req->num_be_args,
2920                                 lun_req->be_args,
2921                                 lun_req->error_str,
2922                                 sizeof(lun_req->error_str));
2923                         if (lun_req->kern_be_args == NULL) {
2924                                 lun_req->status = CTL_LUN_ERROR;
2925                                 break;
2926                         }
2927                 }
2928
2929                 retval = backend->ioctl(dev, cmd, addr, flag, td);
2930
2931                 if (lun_req->num_be_args > 0) {
2932                         ctl_copyout_args(lun_req->num_be_args,
2933                                       lun_req->kern_be_args);
2934                         ctl_free_args(lun_req->num_be_args,
2935                                       lun_req->kern_be_args);
2936                 }
2937                 break;
2938         }
2939         case CTL_LUN_LIST: {
2940                 struct sbuf *sb;
2941                 struct ctl_lun_list *list;
2942                 struct ctl_option *opt;
2943
2944                 list = (struct ctl_lun_list *)addr;
2945
2946                 /*
2947                  * Allocate a fixed length sbuf here, based on the length
2948                  * of the user's buffer.  We could allocate an auto-extending
2949                  * buffer, and then tell the user how much larger our
2950                  * amount of data is than his buffer, but that presents
2951                  * some problems:
2952                  *
2953                  * 1.  The sbuf(9) routines use a blocking malloc, and so
2954                  *     we can't hold a lock while calling them with an
2955                  *     auto-extending buffer.
2956                  *
2957                  * 2.  There is not currently a LUN reference counting
2958                  *     mechanism, outside of outstanding transactions on
2959                  *     the LUN's OOA queue.  So a LUN could go away on us
2960                  *     while we're getting the LUN number, backend-specific
2961                  *     information, etc.  Thus, given the way things
2962                  *     currently work, we need to hold the CTL lock while
2963                  *     grabbing LUN information.
2964                  *
2965                  * So, from the user's standpoint, the best thing to do is
2966                  * allocate what he thinks is a reasonable buffer length,
2967                  * and then if he gets a CTL_LUN_LIST_NEED_MORE_SPACE error,
2968                  * double the buffer length and try again.  (And repeat
2969                  * that until he succeeds.)
2970                  */
2971                 sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
2972                 if (sb == NULL) {
2973                         list->status = CTL_LUN_LIST_ERROR;
2974                         snprintf(list->error_str, sizeof(list->error_str),
2975                                  "Unable to allocate %d bytes for LUN list",
2976                                  list->alloc_len);
2977                         break;
2978                 }
2979
2980                 sbuf_printf(sb, "<ctllunlist>\n");
2981
2982                 mtx_lock(&softc->ctl_lock);
2983                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
2984                         mtx_lock(&lun->lun_lock);
2985                         retval = sbuf_printf(sb, "<lun id=\"%ju\">\n",
2986                                              (uintmax_t)lun->lun);
2987
2988                         /*
2989                          * Bail out as soon as we see that we've overfilled
2990                          * the buffer.
2991                          */
2992                         if (retval != 0)
2993                                 break;
2994
2995                         retval = sbuf_printf(sb, "\t<backend_type>%s"
2996                                              "</backend_type>\n",
2997                                              (lun->backend == NULL) ?  "none" :
2998                                              lun->backend->name);
2999
3000                         if (retval != 0)
3001                                 break;
3002
3003                         retval = sbuf_printf(sb, "\t<lun_type>%d</lun_type>\n",
3004                                              lun->be_lun->lun_type);
3005
3006                         if (retval != 0)
3007                                 break;
3008
3009                         if (lun->backend == NULL) {
3010                                 retval = sbuf_printf(sb, "</lun>\n");
3011                                 if (retval != 0)
3012                                         break;
3013                                 continue;
3014                         }
3015
3016                         retval = sbuf_printf(sb, "\t<size>%ju</size>\n",
3017                                              (lun->be_lun->maxlba > 0) ?
3018                                              lun->be_lun->maxlba + 1 : 0);
3019
3020                         if (retval != 0)
3021                                 break;
3022
3023                         retval = sbuf_printf(sb, "\t<blocksize>%u</blocksize>\n",
3024                                              lun->be_lun->blocksize);
3025
3026                         if (retval != 0)
3027                                 break;
3028
3029                         retval = sbuf_printf(sb, "\t<serial_number>");
3030
3031                         if (retval != 0)
3032                                 break;
3033
3034                         retval = ctl_sbuf_printf_esc(sb,
3035                             lun->be_lun->serial_num,
3036                             sizeof(lun->be_lun->serial_num));
3037
3038                         if (retval != 0)
3039                                 break;
3040
3041                         retval = sbuf_printf(sb, "</serial_number>\n");
3042
3043                         if (retval != 0)
3044                                 break;
3045
3046                         retval = sbuf_printf(sb, "\t<device_id>");
3047
3048                         if (retval != 0)
3049                                 break;
3050
3051                         retval = ctl_sbuf_printf_esc(sb,
3052                             lun->be_lun->device_id,
3053                             sizeof(lun->be_lun->device_id));
3054
3055                         if (retval != 0)
3056                                 break;
3057
3058                         retval = sbuf_printf(sb, "</device_id>\n");
3059
3060                         if (retval != 0)
3061                                 break;
3062
3063                         if (lun->backend->lun_info != NULL) {
3064                                 retval = lun->backend->lun_info(lun->be_lun->be_lun, sb);
3065                                 if (retval != 0)
3066                                         break;
3067                         }
3068                         STAILQ_FOREACH(opt, &lun->be_lun->options, links) {
3069                                 retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3070                                     opt->name, opt->value, opt->name);
3071                                 if (retval != 0)
3072                                         break;
3073                         }
3074
3075                         retval = sbuf_printf(sb, "</lun>\n");
3076
3077                         if (retval != 0)
3078                                 break;
3079                         mtx_unlock(&lun->lun_lock);
3080                 }
3081                 if (lun != NULL)
3082                         mtx_unlock(&lun->lun_lock);
3083                 mtx_unlock(&softc->ctl_lock);
3084
3085                 if ((retval != 0)
3086                  || ((retval = sbuf_printf(sb, "</ctllunlist>\n")) != 0)) {
3087                         retval = 0;
3088                         sbuf_delete(sb);
3089                         list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3090                         snprintf(list->error_str, sizeof(list->error_str),
3091                                  "Out of space, %d bytes is too small",
3092                                  list->alloc_len);
3093                         break;
3094                 }
3095
3096                 sbuf_finish(sb);
3097
3098                 retval = copyout(sbuf_data(sb), list->lun_xml,
3099                                  sbuf_len(sb) + 1);
3100
3101                 list->fill_len = sbuf_len(sb) + 1;
3102                 list->status = CTL_LUN_LIST_OK;
3103                 sbuf_delete(sb);
3104                 break;
3105         }
3106         case CTL_ISCSI: {
3107                 struct ctl_iscsi *ci;
3108                 struct ctl_frontend *fe;
3109
3110                 ci = (struct ctl_iscsi *)addr;
3111
3112                 fe = ctl_frontend_find("iscsi");
3113                 if (fe == NULL) {
3114                         ci->status = CTL_ISCSI_ERROR;
3115                         snprintf(ci->error_str, sizeof(ci->error_str),
3116                             "Frontend \"iscsi\" not found.");
3117                         break;
3118                 }
3119
3120                 retval = fe->ioctl(dev, cmd, addr, flag, td);
3121                 break;
3122         }
3123         case CTL_PORT_REQ: {
3124                 struct ctl_req *req;
3125                 struct ctl_frontend *fe;
3126
3127                 req = (struct ctl_req *)addr;
3128
3129                 fe = ctl_frontend_find(req->driver);
3130                 if (fe == NULL) {
3131                         req->status = CTL_LUN_ERROR;
3132                         snprintf(req->error_str, sizeof(req->error_str),
3133                             "Frontend \"%s\" not found.", req->driver);
3134                         break;
3135                 }
3136                 if (req->num_args > 0) {
3137                         req->kern_args = ctl_copyin_args(req->num_args,
3138                             req->args, req->error_str, sizeof(req->error_str));
3139                         if (req->kern_args == NULL) {
3140                                 req->status = CTL_LUN_ERROR;
3141                                 break;
3142                         }
3143                 }
3144
3145                 if (fe->ioctl)
3146                         retval = fe->ioctl(dev, cmd, addr, flag, td);
3147                 else
3148                         retval = ENODEV;
3149
3150                 if (req->num_args > 0) {
3151                         ctl_copyout_args(req->num_args, req->kern_args);
3152                         ctl_free_args(req->num_args, req->kern_args);
3153                 }
3154                 break;
3155         }
3156         case CTL_PORT_LIST: {
3157                 struct sbuf *sb;
3158                 struct ctl_port *port;
3159                 struct ctl_lun_list *list;
3160                 struct ctl_option *opt;
3161                 int j;
3162                 uint32_t plun;
3163
3164                 list = (struct ctl_lun_list *)addr;
3165
3166                 sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3167                 if (sb == NULL) {
3168                         list->status = CTL_LUN_LIST_ERROR;
3169                         snprintf(list->error_str, sizeof(list->error_str),
3170                                  "Unable to allocate %d bytes for LUN list",
3171                                  list->alloc_len);
3172                         break;
3173                 }
3174
3175                 sbuf_printf(sb, "<ctlportlist>\n");
3176
3177                 mtx_lock(&softc->ctl_lock);
3178                 STAILQ_FOREACH(port, &softc->port_list, links) {
3179                         retval = sbuf_printf(sb, "<targ_port id=\"%ju\">\n",
3180                                              (uintmax_t)port->targ_port);
3181
3182                         /*
3183                          * Bail out as soon as we see that we've overfilled
3184                          * the buffer.
3185                          */
3186                         if (retval != 0)
3187                                 break;
3188
3189                         retval = sbuf_printf(sb, "\t<frontend_type>%s"
3190                             "</frontend_type>\n", port->frontend->name);
3191                         if (retval != 0)
3192                                 break;
3193
3194                         retval = sbuf_printf(sb, "\t<port_type>%d</port_type>\n",
3195                                              port->port_type);
3196                         if (retval != 0)
3197                                 break;
3198
3199                         retval = sbuf_printf(sb, "\t<online>%s</online>\n",
3200                             (port->status & CTL_PORT_STATUS_ONLINE) ? "YES" : "NO");
3201                         if (retval != 0)
3202                                 break;
3203
3204                         retval = sbuf_printf(sb, "\t<port_name>%s</port_name>\n",
3205                             port->port_name);
3206                         if (retval != 0)
3207                                 break;
3208
3209                         retval = sbuf_printf(sb, "\t<physical_port>%d</physical_port>\n",
3210                             port->physical_port);
3211                         if (retval != 0)
3212                                 break;
3213
3214                         retval = sbuf_printf(sb, "\t<virtual_port>%d</virtual_port>\n",
3215                             port->virtual_port);
3216                         if (retval != 0)
3217                                 break;
3218
3219                         if (port->target_devid != NULL) {
3220                                 sbuf_printf(sb, "\t<target>");
3221                                 ctl_id_sbuf(port->target_devid, sb);
3222                                 sbuf_printf(sb, "</target>\n");
3223                         }
3224
3225                         if (port->port_devid != NULL) {
3226                                 sbuf_printf(sb, "\t<port>");
3227                                 ctl_id_sbuf(port->port_devid, sb);
3228                                 sbuf_printf(sb, "</port>\n");
3229                         }
3230
3231                         if (port->port_info != NULL) {
3232                                 retval = port->port_info(port->onoff_arg, sb);
3233                                 if (retval != 0)
3234                                         break;
3235                         }
3236                         STAILQ_FOREACH(opt, &port->options, links) {
3237                                 retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3238                                     opt->name, opt->value, opt->name);
3239                                 if (retval != 0)
3240                                         break;
3241                         }
3242
3243                         if (port->lun_map != NULL) {
3244                                 sbuf_printf(sb, "\t<lun_map>on</lun_map>\n");
3245                                 for (j = 0; j < CTL_MAX_LUNS; j++) {
3246                                         plun = ctl_lun_map_from_port(port, j);
3247                                         if (plun >= CTL_MAX_LUNS)
3248                                                 continue;
3249                                         sbuf_printf(sb,
3250                                             "\t<lun id=\"%u\">%u</lun>\n",
3251                                             j, plun);
3252                                 }
3253                         }
3254
3255                         for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
3256                                 if (port->wwpn_iid[j].in_use == 0 ||
3257                                     (port->wwpn_iid[j].wwpn == 0 &&
3258                                      port->wwpn_iid[j].name == NULL))
3259                                         continue;
3260
3261                                 if (port->wwpn_iid[j].name != NULL)
3262                                         retval = sbuf_printf(sb,
3263                                             "\t<initiator id=\"%u\">%s</initiator>\n",
3264                                             j, port->wwpn_iid[j].name);
3265                                 else
3266                                         retval = sbuf_printf(sb,
3267                                             "\t<initiator id=\"%u\">naa.%08jx</initiator>\n",
3268                                             j, port->wwpn_iid[j].wwpn);
3269                                 if (retval != 0)
3270                                         break;
3271                         }
3272                         if (retval != 0)
3273                                 break;
3274
3275                         retval = sbuf_printf(sb, "</targ_port>\n");
3276                         if (retval != 0)
3277                                 break;
3278                 }
3279                 mtx_unlock(&softc->ctl_lock);
3280
3281                 if ((retval != 0)
3282                  || ((retval = sbuf_printf(sb, "</ctlportlist>\n")) != 0)) {
3283                         retval = 0;
3284                         sbuf_delete(sb);
3285                         list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3286                         snprintf(list->error_str, sizeof(list->error_str),
3287                                  "Out of space, %d bytes is too small",
3288                                  list->alloc_len);
3289                         break;
3290                 }
3291
3292                 sbuf_finish(sb);
3293
3294                 retval = copyout(sbuf_data(sb), list->lun_xml,
3295                                  sbuf_len(sb) + 1);
3296
3297                 list->fill_len = sbuf_len(sb) + 1;
3298                 list->status = CTL_LUN_LIST_OK;
3299                 sbuf_delete(sb);
3300                 break;
3301         }
3302         case CTL_LUN_MAP: {
3303                 struct ctl_lun_map *lm  = (struct ctl_lun_map *)addr;
3304                 struct ctl_port *port;
3305
3306                 mtx_lock(&softc->ctl_lock);
3307                 if (lm->port < softc->port_min ||
3308                     lm->port >= softc->port_max ||
3309                     (port = softc->ctl_ports[lm->port]) == NULL) {
3310                         mtx_unlock(&softc->ctl_lock);
3311                         return (ENXIO);
3312                 }
3313                 if (port->status & CTL_PORT_STATUS_ONLINE) {
3314                         STAILQ_FOREACH(lun, &softc->lun_list, links) {
3315                                 if (ctl_lun_map_to_port(port, lun->lun) >=
3316                                     CTL_MAX_LUNS)
3317                                         continue;
3318                                 mtx_lock(&lun->lun_lock);
3319                                 ctl_est_ua_port(lun, lm->port, -1,
3320                                     CTL_UA_LUN_CHANGE);
3321                                 mtx_unlock(&lun->lun_lock);
3322                         }
3323                 }
3324                 mtx_unlock(&softc->ctl_lock); // XXX: port_enable sleeps
3325                 if (lm->plun < CTL_MAX_LUNS) {
3326                         if (lm->lun == UINT32_MAX)
3327                                 retval = ctl_lun_map_unset(port, lm->plun);
3328                         else if (lm->lun < CTL_MAX_LUNS &&
3329                             softc->ctl_luns[lm->lun] != NULL)
3330                                 retval = ctl_lun_map_set(port, lm->plun, lm->lun);
3331                         else
3332                                 return (ENXIO);
3333                 } else if (lm->plun == UINT32_MAX) {
3334                         if (lm->lun == UINT32_MAX)
3335                                 retval = ctl_lun_map_deinit(port);
3336                         else
3337                                 retval = ctl_lun_map_init(port);
3338                 } else
3339                         return (ENXIO);
3340                 if (port->status & CTL_PORT_STATUS_ONLINE)
3341                         ctl_isc_announce_port(port);
3342                 break;
3343         }
3344         default: {
3345                 /* XXX KDM should we fix this? */
3346 #if 0
3347                 struct ctl_backend_driver *backend;
3348                 unsigned int type;
3349                 int found;
3350
3351                 found = 0;
3352
3353                 /*
3354                  * We encode the backend type as the ioctl type for backend
3355                  * ioctls.  So parse it out here, and then search for a
3356                  * backend of this type.
3357                  */
3358                 type = _IOC_TYPE(cmd);
3359
3360                 STAILQ_FOREACH(backend, &softc->be_list, links) {
3361                         if (backend->type == type) {
3362                                 found = 1;
3363                                 break;
3364                         }
3365                 }
3366                 if (found == 0) {
3367                         printf("ctl: unknown ioctl command %#lx or backend "
3368                                "%d\n", cmd, type);
3369                         retval = EINVAL;
3370                         break;
3371                 }
3372                 retval = backend->ioctl(dev, cmd, addr, flag, td);
3373 #endif
3374                 retval = ENOTTY;
3375                 break;
3376         }
3377         }
3378         return (retval);
3379 }
3380
3381 uint32_t
3382 ctl_get_initindex(struct ctl_nexus *nexus)
3383 {
3384         return (nexus->initid + (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3385 }
3386
3387 int
3388 ctl_lun_map_init(struct ctl_port *port)
3389 {
3390         struct ctl_softc *softc = port->ctl_softc;
3391         struct ctl_lun *lun;
3392         uint32_t i;
3393
3394         if (port->lun_map == NULL)
3395                 port->lun_map = malloc(sizeof(uint32_t) * CTL_MAX_LUNS,
3396                     M_CTL, M_NOWAIT);
3397         if (port->lun_map == NULL)
3398                 return (ENOMEM);
3399         for (i = 0; i < CTL_MAX_LUNS; i++)
3400                 port->lun_map[i] = UINT32_MAX;
3401         if (port->status & CTL_PORT_STATUS_ONLINE) {
3402                 if (port->lun_disable != NULL) {
3403                         STAILQ_FOREACH(lun, &softc->lun_list, links)
3404                                 port->lun_disable(port->targ_lun_arg, lun->lun);
3405                 }
3406                 ctl_isc_announce_port(port);
3407         }
3408         return (0);
3409 }
3410
3411 int
3412 ctl_lun_map_deinit(struct ctl_port *port)
3413 {
3414         struct ctl_softc *softc = port->ctl_softc;
3415         struct ctl_lun *lun;
3416
3417         if (port->lun_map == NULL)
3418                 return (0);
3419         free(port->lun_map, M_CTL);
3420         port->lun_map = NULL;
3421         if (port->status & CTL_PORT_STATUS_ONLINE) {
3422                 if (port->lun_enable != NULL) {
3423                         STAILQ_FOREACH(lun, &softc->lun_list, links)
3424                                 port->lun_enable(port->targ_lun_arg, lun->lun);
3425                 }
3426                 ctl_isc_announce_port(port);
3427         }
3428         return (0);
3429 }
3430
3431 int
3432 ctl_lun_map_set(struct ctl_port *port, uint32_t plun, uint32_t glun)
3433 {
3434         int status;
3435         uint32_t old;
3436
3437         if (port->lun_map == NULL) {
3438                 status = ctl_lun_map_init(port);
3439                 if (status != 0)
3440                         return (status);
3441         }
3442         old = port->lun_map[plun];
3443         port->lun_map[plun] = glun;
3444         if ((port->status & CTL_PORT_STATUS_ONLINE) && old >= CTL_MAX_LUNS) {
3445                 if (port->lun_enable != NULL)
3446                         port->lun_enable(port->targ_lun_arg, plun);
3447                 ctl_isc_announce_port(port);
3448         }
3449         return (0);
3450 }
3451
3452 int
3453 ctl_lun_map_unset(struct ctl_port *port, uint32_t plun)
3454 {
3455         uint32_t old;
3456
3457         if (port->lun_map == NULL)
3458                 return (0);
3459         old = port->lun_map[plun];
3460         port->lun_map[plun] = UINT32_MAX;
3461         if ((port->status & CTL_PORT_STATUS_ONLINE) && old < CTL_MAX_LUNS) {
3462                 if (port->lun_disable != NULL)
3463                         port->lun_disable(port->targ_lun_arg, plun);
3464                 ctl_isc_announce_port(port);
3465         }
3466         return (0);
3467 }
3468
3469 uint32_t
3470 ctl_lun_map_from_port(struct ctl_port *port, uint32_t lun_id)
3471 {
3472
3473         if (port == NULL)
3474                 return (UINT32_MAX);
3475         if (port->lun_map == NULL || lun_id >= CTL_MAX_LUNS)
3476                 return (lun_id);
3477         return (port->lun_map[lun_id]);
3478 }
3479
3480 uint32_t
3481 ctl_lun_map_to_port(struct ctl_port *port, uint32_t lun_id)
3482 {
3483         uint32_t i;
3484
3485         if (port == NULL)
3486                 return (UINT32_MAX);
3487         if (port->lun_map == NULL)
3488                 return (lun_id);
3489         for (i = 0; i < CTL_MAX_LUNS; i++) {
3490                 if (port->lun_map[i] == lun_id)
3491                         return (i);
3492         }
3493         return (UINT32_MAX);
3494 }
3495
3496 static struct ctl_port *
3497 ctl_io_port(struct ctl_io_hdr *io_hdr)
3498 {
3499
3500         return (control_softc->ctl_ports[io_hdr->nexus.targ_port]);
3501 }
3502
3503 int
3504 ctl_ffz(uint32_t *mask, uint32_t first, uint32_t last)
3505 {
3506         int i;
3507
3508         for (i = first; i < last; i++) {
3509                 if ((mask[i / 32] & (1 << (i % 32))) == 0)
3510                         return (i);
3511         }
3512         return (-1);
3513 }
3514
3515 int
3516 ctl_set_mask(uint32_t *mask, uint32_t bit)
3517 {
3518         uint32_t chunk, piece;
3519
3520         chunk = bit >> 5;
3521         piece = bit % (sizeof(uint32_t) * 8);
3522
3523         if ((mask[chunk] & (1 << piece)) != 0)
3524                 return (-1);
3525         else
3526                 mask[chunk] |= (1 << piece);
3527
3528         return (0);
3529 }
3530
3531 int
3532 ctl_clear_mask(uint32_t *mask, uint32_t bit)
3533 {
3534         uint32_t chunk, piece;
3535
3536         chunk = bit >> 5;
3537         piece = bit % (sizeof(uint32_t) * 8);
3538
3539         if ((mask[chunk] & (1 << piece)) == 0)
3540                 return (-1);
3541         else
3542                 mask[chunk] &= ~(1 << piece);
3543
3544         return (0);
3545 }
3546
3547 int
3548 ctl_is_set(uint32_t *mask, uint32_t bit)
3549 {
3550         uint32_t chunk, piece;
3551
3552         chunk = bit >> 5;
3553         piece = bit % (sizeof(uint32_t) * 8);
3554
3555         if ((mask[chunk] & (1 << piece)) == 0)
3556                 return (0);
3557         else
3558                 return (1);
3559 }
3560
3561 static uint64_t
3562 ctl_get_prkey(struct ctl_lun *lun, uint32_t residx)
3563 {
3564         uint64_t *t;
3565
3566         t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3567         if (t == NULL)
3568                 return (0);
3569         return (t[residx % CTL_MAX_INIT_PER_PORT]);
3570 }
3571
3572 static void
3573 ctl_clr_prkey(struct ctl_lun *lun, uint32_t residx)
3574 {
3575         uint64_t *t;
3576
3577         t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3578         if (t == NULL)
3579                 return;
3580         t[residx % CTL_MAX_INIT_PER_PORT] = 0;
3581 }
3582
3583 static void
3584 ctl_alloc_prkey(struct ctl_lun *lun, uint32_t residx)
3585 {
3586         uint64_t *p;
3587         u_int i;
3588
3589         i = residx/CTL_MAX_INIT_PER_PORT;
3590         if (lun->pr_keys[i] != NULL)
3591                 return;
3592         mtx_unlock(&lun->lun_lock);
3593         p = malloc(sizeof(uint64_t) * CTL_MAX_INIT_PER_PORT, M_CTL,
3594             M_WAITOK | M_ZERO);
3595         mtx_lock(&lun->lun_lock);
3596         if (lun->pr_keys[i] == NULL)
3597                 lun->pr_keys[i] = p;
3598         else
3599                 free(p, M_CTL);
3600 }
3601
3602 static void
3603 ctl_set_prkey(struct ctl_lun *lun, uint32_t residx, uint64_t key)
3604 {
3605         uint64_t *t;
3606
3607         t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3608         KASSERT(t != NULL, ("prkey %d is not allocated", residx));
3609         t[residx % CTL_MAX_INIT_PER_PORT] = key;
3610 }
3611
3612 /*
3613  * ctl_softc, pool_name, total_ctl_io are passed in.
3614  * npool is passed out.
3615  */
3616 int
3617 ctl_pool_create(struct ctl_softc *ctl_softc, const char *pool_name,
3618                 uint32_t total_ctl_io, void **npool)
3619 {
3620 #ifdef IO_POOLS
3621         struct ctl_io_pool *pool;
3622
3623         pool = (struct ctl_io_pool *)malloc(sizeof(*pool), M_CTL,
3624                                             M_NOWAIT | M_ZERO);
3625         if (pool == NULL)
3626                 return (ENOMEM);
3627
3628         snprintf(pool->name, sizeof(pool->name), "CTL IO %s", pool_name);
3629         pool->ctl_softc = ctl_softc;
3630         pool->zone = uma_zsecond_create(pool->name, NULL,
3631             NULL, NULL, NULL, ctl_softc->io_zone);
3632         /* uma_prealloc(pool->zone, total_ctl_io); */
3633
3634         *npool = pool;
3635 #else
3636         *npool = ctl_softc->io_zone;
3637 #endif
3638         return (0);
3639 }
3640
3641 void
3642 ctl_pool_free(struct ctl_io_pool *pool)
3643 {
3644
3645         if (pool == NULL)
3646                 return;
3647
3648 #ifdef IO_POOLS
3649         uma_zdestroy(pool->zone);
3650         free(pool, M_CTL);
3651 #endif
3652 }
3653
3654 union ctl_io *
3655 ctl_alloc_io(void *pool_ref)
3656 {
3657         union ctl_io *io;
3658 #ifdef IO_POOLS
3659         struct ctl_io_pool *pool = (struct ctl_io_pool *)pool_ref;
3660
3661         io = uma_zalloc(pool->zone, M_WAITOK);
3662 #else
3663         io = uma_zalloc((uma_zone_t)pool_ref, M_WAITOK);
3664 #endif
3665         if (io != NULL)
3666                 io->io_hdr.pool = pool_ref;
3667         return (io);
3668 }
3669
3670 union ctl_io *
3671 ctl_alloc_io_nowait(void *pool_ref)
3672 {
3673         union ctl_io *io;
3674 #ifdef IO_POOLS
3675         struct ctl_io_pool *pool = (struct ctl_io_pool *)pool_ref;
3676
3677         io = uma_zalloc(pool->zone, M_NOWAIT);
3678 #else
3679         io = uma_zalloc((uma_zone_t)pool_ref, M_NOWAIT);
3680 #endif
3681         if (io != NULL)
3682                 io->io_hdr.pool = pool_ref;
3683         return (io);
3684 }
3685
3686 void
3687 ctl_free_io(union ctl_io *io)
3688 {
3689 #ifdef IO_POOLS
3690         struct ctl_io_pool *pool;
3691 #endif
3692
3693         if (io == NULL)
3694                 return;
3695
3696 #ifdef IO_POOLS
3697         pool = (struct ctl_io_pool *)io->io_hdr.pool;
3698         uma_zfree(pool->zone, io);
3699 #else
3700         uma_zfree((uma_zone_t)io->io_hdr.pool, io);
3701 #endif
3702 }
3703
3704 void
3705 ctl_zero_io(union ctl_io *io)
3706 {
3707         void *pool_ref;
3708
3709         if (io == NULL)
3710                 return;
3711
3712         /*
3713          * May need to preserve linked list pointers at some point too.
3714          */
3715         pool_ref = io->io_hdr.pool;
3716         memset(io, 0, sizeof(*io));
3717         io->io_hdr.pool = pool_ref;
3718 }
3719
3720 /*
3721  * This routine is currently used for internal copies of ctl_ios that need
3722  * to persist for some reason after we've already returned status to the
3723  * FETD.  (Thus the flag set.)
3724  *
3725  * XXX XXX
3726  * Note that this makes a blind copy of all fields in the ctl_io, except
3727  * for the pool reference.  This includes any memory that has been
3728  * allocated!  That memory will no longer be valid after done has been
3729  * called, so this would be VERY DANGEROUS for command that actually does
3730  * any reads or writes.  Right now (11/7/2005), this is only used for immediate
3731  * start and stop commands, which don't transfer any data, so this is not a
3732  * problem.  If it is used for anything else, the caller would also need to
3733  * allocate data buffer space and this routine would need to be modified to
3734  * copy the data buffer(s) as well.
3735  */
3736 void
3737 ctl_copy_io(union ctl_io *src, union ctl_io *dest)
3738 {
3739         void *pool_ref;
3740
3741         if ((src == NULL)
3742          || (dest == NULL))
3743                 return;
3744
3745         /*
3746          * May need to preserve linked list pointers at some point too.
3747          */
3748         pool_ref = dest->io_hdr.pool;
3749
3750         memcpy(dest, src, MIN(sizeof(*src), sizeof(*dest)));
3751
3752         dest->io_hdr.pool = pool_ref;
3753         /*
3754          * We need to know that this is an internal copy, and doesn't need
3755          * to get passed back to the FETD that allocated it.
3756          */
3757         dest->io_hdr.flags |= CTL_FLAG_INT_COPY;
3758 }
3759
3760 int
3761 ctl_expand_number(const char *buf, uint64_t *num)
3762 {
3763         char *endptr;
3764         uint64_t number;
3765         unsigned shift;
3766
3767         number = strtoq(buf, &endptr, 0);
3768
3769         switch (tolower((unsigned char)*endptr)) {
3770         case 'e':
3771                 shift = 60;
3772                 break;
3773         case 'p':
3774                 shift = 50;
3775                 break;
3776         case 't':
3777                 shift = 40;
3778                 break;
3779         case 'g':
3780                 shift = 30;
3781                 break;
3782         case 'm':
3783                 shift = 20;
3784                 break;
3785         case 'k':
3786                 shift = 10;
3787                 break;
3788         case 'b':
3789         case '\0': /* No unit. */
3790                 *num = number;
3791                 return (0);
3792         default:
3793                 /* Unrecognized unit. */
3794                 return (-1);
3795         }
3796
3797         if ((number << shift) >> shift != number) {
3798                 /* Overflow */
3799                 return (-1);
3800         }
3801         *num = number << shift;
3802         return (0);
3803 }
3804
3805
3806 /*
3807  * This routine could be used in the future to load default and/or saved
3808  * mode page parameters for a particuar lun.
3809  */
3810 static int
3811 ctl_init_page_index(struct ctl_lun *lun)
3812 {
3813         int i;
3814         struct ctl_page_index *page_index;
3815         const char *value;
3816         uint64_t ival;
3817
3818         memcpy(&lun->mode_pages.index, page_index_template,
3819                sizeof(page_index_template));
3820
3821         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
3822
3823                 page_index = &lun->mode_pages.index[i];
3824                 /*
3825                  * If this is a disk-only mode page, there's no point in
3826                  * setting it up.  For some pages, we have to have some
3827                  * basic information about the disk in order to calculate the
3828                  * mode page data.
3829                  */
3830                 if ((lun->be_lun->lun_type != T_DIRECT)
3831                  && (page_index->page_flags & CTL_PAGE_FLAG_DISK_ONLY))
3832                         continue;
3833
3834                 switch (page_index->page_code & SMPH_PC_MASK) {
3835                 case SMS_RW_ERROR_RECOVERY_PAGE: {
3836                         if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3837                                 panic("subpage is incorrect!");
3838                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_CURRENT],
3839                                &rw_er_page_default,
3840                                sizeof(rw_er_page_default));
3841                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_CHANGEABLE],
3842                                &rw_er_page_changeable,
3843                                sizeof(rw_er_page_changeable));
3844                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_DEFAULT],
3845                                &rw_er_page_default,
3846                                sizeof(rw_er_page_default));
3847                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_SAVED],
3848                                &rw_er_page_default,
3849                                sizeof(rw_er_page_default));
3850                         page_index->page_data =
3851                                 (uint8_t *)lun->mode_pages.rw_er_page;
3852                         break;
3853                 }
3854                 case SMS_FORMAT_DEVICE_PAGE: {
3855                         struct scsi_format_page *format_page;
3856
3857                         if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3858                                 panic("subpage is incorrect!");
3859
3860                         /*
3861                          * Sectors per track are set above.  Bytes per
3862                          * sector need to be set here on a per-LUN basis.
3863                          */
3864                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_CURRENT],
3865                                &format_page_default,
3866                                sizeof(format_page_default));
3867                         memcpy(&lun->mode_pages.format_page[
3868                                CTL_PAGE_CHANGEABLE], &format_page_changeable,
3869                                sizeof(format_page_changeable));
3870                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_DEFAULT],
3871                                &format_page_default,
3872                                sizeof(format_page_default));
3873                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_SAVED],
3874                                &format_page_default,
3875                                sizeof(format_page_default));
3876
3877                         format_page = &lun->mode_pages.format_page[
3878                                 CTL_PAGE_CURRENT];
3879                         scsi_ulto2b(lun->be_lun->blocksize,
3880                                     format_page->bytes_per_sector);
3881
3882                         format_page = &lun->mode_pages.format_page[
3883                                 CTL_PAGE_DEFAULT];
3884                         scsi_ulto2b(lun->be_lun->blocksize,
3885                                     format_page->bytes_per_sector);
3886
3887                         format_page = &lun->mode_pages.format_page[
3888                                 CTL_PAGE_SAVED];
3889                         scsi_ulto2b(lun->be_lun->blocksize,
3890                                     format_page->bytes_per_sector);
3891
3892                         page_index->page_data =
3893                                 (uint8_t *)lun->mode_pages.format_page;
3894                         break;
3895                 }
3896                 case SMS_RIGID_DISK_PAGE: {
3897                         struct scsi_rigid_disk_page *rigid_disk_page;
3898                         uint32_t sectors_per_cylinder;
3899                         uint64_t cylinders;
3900 #ifndef __XSCALE__
3901                         int shift;
3902 #endif /* !__XSCALE__ */
3903
3904                         if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3905                                 panic("invalid subpage value %d",
3906                                       page_index->subpage);
3907
3908                         /*
3909                          * Rotation rate and sectors per track are set
3910                          * above.  We calculate the cylinders here based on
3911                          * capacity.  Due to the number of heads and
3912                          * sectors per track we're using, smaller arrays
3913                          * may turn out to have 0 cylinders.  Linux and
3914                          * FreeBSD don't pay attention to these mode pages
3915                          * to figure out capacity, but Solaris does.  It
3916                          * seems to deal with 0 cylinders just fine, and
3917                          * works out a fake geometry based on the capacity.
3918                          */
3919                         memcpy(&lun->mode_pages.rigid_disk_page[
3920                                CTL_PAGE_DEFAULT], &rigid_disk_page_default,
3921                                sizeof(rigid_disk_page_default));
3922                         memcpy(&lun->mode_pages.rigid_disk_page[
3923                                CTL_PAGE_CHANGEABLE],&rigid_disk_page_changeable,
3924                                sizeof(rigid_disk_page_changeable));
3925
3926                         sectors_per_cylinder = CTL_DEFAULT_SECTORS_PER_TRACK *
3927                                 CTL_DEFAULT_HEADS;
3928
3929                         /*
3930                          * The divide method here will be more accurate,
3931                          * probably, but results in floating point being
3932                          * used in the kernel on i386 (__udivdi3()).  On the
3933                          * XScale, though, __udivdi3() is implemented in
3934                          * software.
3935                          *
3936                          * The shift method for cylinder calculation is
3937                          * accurate if sectors_per_cylinder is a power of
3938                          * 2.  Otherwise it might be slightly off -- you
3939                          * might have a bit of a truncation problem.
3940                          */
3941 #ifdef  __XSCALE__
3942                         cylinders = (lun->be_lun->maxlba + 1) /
3943                                 sectors_per_cylinder;
3944 #else
3945                         for (shift = 31; shift > 0; shift--) {
3946                                 if (sectors_per_cylinder & (1 << shift))
3947                                         break;
3948                         }
3949                         cylinders = (lun->be_lun->maxlba + 1) >> shift;
3950 #endif
3951
3952                         /*
3953                          * We've basically got 3 bytes, or 24 bits for the
3954                          * cylinder size in the mode page.  If we're over,
3955                          * just round down to 2^24.
3956                          */
3957                         if (cylinders > 0xffffff)
3958                                 cylinders = 0xffffff;
3959
3960                         rigid_disk_page = &lun->mode_pages.rigid_disk_page[
3961                                 CTL_PAGE_DEFAULT];
3962                         scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
3963
3964                         if ((value = ctl_get_opt(&lun->be_lun->options,
3965                             "rpm")) != NULL) {
3966                                 scsi_ulto2b(strtol(value, NULL, 0),
3967                                      rigid_disk_page->rotation_rate);
3968                         }
3969
3970                         memcpy(&lun->mode_pages.rigid_disk_page[CTL_PAGE_CURRENT],
3971                                &lun->mode_pages.rigid_disk_page[CTL_PAGE_DEFAULT],
3972                                sizeof(rigid_disk_page_default));
3973                         memcpy(&lun->mode_pages.rigid_disk_page[CTL_PAGE_SAVED],
3974                                &lun->mode_pages.rigid_disk_page[CTL_PAGE_DEFAULT],
3975                                sizeof(rigid_disk_page_default));
3976
3977                         page_index->page_data =
3978                                 (uint8_t *)lun->mode_pages.rigid_disk_page;
3979                         break;
3980                 }
3981                 case SMS_CACHING_PAGE: {
3982                         struct scsi_caching_page *caching_page;
3983
3984                         if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3985                                 panic("invalid subpage value %d",
3986                                       page_index->subpage);
3987                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_DEFAULT],
3988                                &caching_page_default,
3989                                sizeof(caching_page_default));
3990                         memcpy(&lun->mode_pages.caching_page[
3991                                CTL_PAGE_CHANGEABLE], &caching_page_changeable,
3992                                sizeof(caching_page_changeable));
3993                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_SAVED],
3994                                &caching_page_default,
3995                                sizeof(caching_page_default));
3996                         caching_page = &lun->mode_pages.caching_page[
3997                             CTL_PAGE_SAVED];
3998                         value = ctl_get_opt(&lun->be_lun->options, "writecache");
3999                         if (value != NULL && strcmp(value, "off") == 0)
4000                                 caching_page->flags1 &= ~SCP_WCE;
4001                         value = ctl_get_opt(&lun->be_lun->options, "readcache");
4002                         if (value != NULL && strcmp(value, "off") == 0)
4003                                 caching_page->flags1 |= SCP_RCD;
4004                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_CURRENT],
4005                                &lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4006                                sizeof(caching_page_default));
4007                         page_index->page_data =
4008                                 (uint8_t *)lun->mode_pages.caching_page;
4009                         break;
4010                 }
4011                 case SMS_CONTROL_MODE_PAGE: {
4012                         switch (page_index->subpage) {
4013                         case SMS_SUBPAGE_PAGE_0: {
4014                                 struct scsi_control_page *control_page;
4015
4016                                 memcpy(&lun->mode_pages.control_page[
4017                                     CTL_PAGE_DEFAULT],
4018                                        &control_page_default,
4019                                        sizeof(control_page_default));
4020                                 memcpy(&lun->mode_pages.control_page[
4021                                     CTL_PAGE_CHANGEABLE],
4022                                        &control_page_changeable,
4023                                        sizeof(control_page_changeable));
4024                                 memcpy(&lun->mode_pages.control_page[
4025                                     CTL_PAGE_SAVED],
4026                                        &control_page_default,
4027                                        sizeof(control_page_default));
4028                                 control_page = &lun->mode_pages.control_page[
4029                                     CTL_PAGE_SAVED];
4030                                 value = ctl_get_opt(&lun->be_lun->options,
4031                                     "reordering");
4032                                 if (value != NULL &&
4033                                     strcmp(value, "unrestricted") == 0) {
4034                                         control_page->queue_flags &=
4035                                             ~SCP_QUEUE_ALG_MASK;
4036                                         control_page->queue_flags |=
4037                                             SCP_QUEUE_ALG_UNRESTRICTED;
4038                                 }
4039                                 memcpy(&lun->mode_pages.control_page[
4040                                     CTL_PAGE_CURRENT],
4041                                        &lun->mode_pages.control_page[
4042                                     CTL_PAGE_SAVED],
4043                                        sizeof(control_page_default));
4044                                 page_index->page_data =
4045                                     (uint8_t *)lun->mode_pages.control_page;
4046                                 break;
4047                         }
4048                         case 0x01:
4049                                 memcpy(&lun->mode_pages.control_ext_page[
4050                                     CTL_PAGE_DEFAULT],
4051                                        &control_ext_page_default,
4052                                        sizeof(control_ext_page_default));
4053                                 memcpy(&lun->mode_pages.control_ext_page[
4054                                     CTL_PAGE_CHANGEABLE],
4055                                        &control_ext_page_changeable,
4056                                        sizeof(control_ext_page_changeable));
4057                                 memcpy(&lun->mode_pages.control_ext_page[
4058                                     CTL_PAGE_SAVED],
4059                                        &control_ext_page_default,
4060                                        sizeof(control_ext_page_default));
4061                                 memcpy(&lun->mode_pages.control_ext_page[
4062                                     CTL_PAGE_CURRENT],
4063                                        &lun->mode_pages.control_ext_page[
4064                                     CTL_PAGE_SAVED],
4065                                        sizeof(control_ext_page_default));
4066                                 page_index->page_data =
4067                                     (uint8_t *)lun->mode_pages.control_ext_page;
4068                                 break;
4069                         }
4070                         break;
4071                 }
4072                 case SMS_INFO_EXCEPTIONS_PAGE: {
4073                         switch (page_index->subpage) {
4074                         case SMS_SUBPAGE_PAGE_0:
4075                                 memcpy(&lun->mode_pages.ie_page[CTL_PAGE_CURRENT],
4076                                        &ie_page_default,
4077                                        sizeof(ie_page_default));
4078                                 memcpy(&lun->mode_pages.ie_page[
4079                                        CTL_PAGE_CHANGEABLE], &ie_page_changeable,
4080                                        sizeof(ie_page_changeable));
4081                                 memcpy(&lun->mode_pages.ie_page[CTL_PAGE_DEFAULT],
4082                                        &ie_page_default,
4083                                        sizeof(ie_page_default));
4084                                 memcpy(&lun->mode_pages.ie_page[CTL_PAGE_SAVED],
4085                                        &ie_page_default,
4086                                        sizeof(ie_page_default));
4087                                 page_index->page_data =
4088                                         (uint8_t *)lun->mode_pages.ie_page;
4089                                 break;
4090                         case 0x02: {
4091                                 struct ctl_logical_block_provisioning_page *page;
4092
4093                                 memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_DEFAULT],
4094                                        &lbp_page_default,
4095                                        sizeof(lbp_page_default));
4096                                 memcpy(&lun->mode_pages.lbp_page[
4097                                        CTL_PAGE_CHANGEABLE], &lbp_page_changeable,
4098                                        sizeof(lbp_page_changeable));
4099                                 memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_SAVED],
4100                                        &lbp_page_default,
4101                                        sizeof(lbp_page_default));
4102                                 page = &lun->mode_pages.lbp_page[CTL_PAGE_SAVED];
4103                                 value = ctl_get_opt(&lun->be_lun->options,
4104                                     "avail-threshold");
4105                                 if (value != NULL &&
4106                                     ctl_expand_number(value, &ival) == 0) {
4107                                         page->descr[0].flags |= SLBPPD_ENABLED |
4108                                             SLBPPD_ARMING_DEC;
4109                                         if (lun->be_lun->blocksize)
4110                                                 ival /= lun->be_lun->blocksize;
4111                                         else
4112                                                 ival /= 512;
4113                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4114                                             page->descr[0].count);
4115                                 }
4116                                 value = ctl_get_opt(&lun->be_lun->options,
4117                                     "used-threshold");
4118                                 if (value != NULL &&
4119                                     ctl_expand_number(value, &ival) == 0) {
4120                                         page->descr[1].flags |= SLBPPD_ENABLED |
4121                                             SLBPPD_ARMING_INC;
4122                                         if (lun->be_lun->blocksize)
4123                                                 ival /= lun->be_lun->blocksize;
4124                                         else
4125                                                 ival /= 512;
4126                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4127                                             page->descr[1].count);
4128                                 }
4129                                 value = ctl_get_opt(&lun->be_lun->options,
4130                                     "pool-avail-threshold");
4131                                 if (value != NULL &&
4132                                     ctl_expand_number(value, &ival) == 0) {
4133                                         page->descr[2].flags |= SLBPPD_ENABLED |
4134                                             SLBPPD_ARMING_DEC;
4135                                         if (lun->be_lun->blocksize)
4136                                                 ival /= lun->be_lun->blocksize;
4137                                         else
4138                                                 ival /= 512;
4139                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4140                                             page->descr[2].count);
4141                                 }
4142                                 value = ctl_get_opt(&lun->be_lun->options,
4143                                     "pool-used-threshold");
4144                                 if (value != NULL &&
4145                                     ctl_expand_number(value, &ival) == 0) {
4146                                         page->descr[3].flags |= SLBPPD_ENABLED |
4147                                             SLBPPD_ARMING_INC;
4148                                         if (lun->be_lun->blocksize)
4149                                                 ival /= lun->be_lun->blocksize;
4150                                         else
4151                                                 ival /= 512;
4152                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4153                                             page->descr[3].count);
4154                                 }
4155                                 memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_CURRENT],
4156                                        &lun->mode_pages.lbp_page[CTL_PAGE_SAVED],
4157                                        sizeof(lbp_page_default));
4158                                 page_index->page_data =
4159                                         (uint8_t *)lun->mode_pages.lbp_page;
4160                         }}
4161                         break;
4162                 }
4163                 case SMS_VENDOR_SPECIFIC_PAGE:{
4164                         switch (page_index->subpage) {
4165                         case DBGCNF_SUBPAGE_CODE: {
4166                                 memcpy(&lun->mode_pages.debugconf_subpage[
4167                                        CTL_PAGE_CURRENT],
4168                                        &debugconf_page_default,
4169                                        sizeof(debugconf_page_default));
4170                                 memcpy(&lun->mode_pages.debugconf_subpage[
4171                                        CTL_PAGE_CHANGEABLE],
4172                                        &debugconf_page_changeable,
4173                                        sizeof(debugconf_page_changeable));
4174                                 memcpy(&lun->mode_pages.debugconf_subpage[
4175                                        CTL_PAGE_DEFAULT],
4176                                        &debugconf_page_default,
4177                                        sizeof(debugconf_page_default));
4178                                 memcpy(&lun->mode_pages.debugconf_subpage[
4179                                        CTL_PAGE_SAVED],
4180                                        &debugconf_page_default,
4181                                        sizeof(debugconf_page_default));
4182                                 page_index->page_data =
4183                                     (uint8_t *)lun->mode_pages.debugconf_subpage;
4184                                 break;
4185                         }
4186                         default:
4187                                 panic("invalid subpage value %d",
4188                                       page_index->subpage);
4189                                 break;
4190                         }
4191                         break;
4192                 }
4193                 default:
4194                         panic("invalid page value %d",
4195                               page_index->page_code & SMPH_PC_MASK);
4196                         break;
4197         }
4198         }
4199
4200         return (CTL_RETVAL_COMPLETE);
4201 }
4202
4203 static int
4204 ctl_init_log_page_index(struct ctl_lun *lun)
4205 {
4206         struct ctl_page_index *page_index;
4207         int i, j, k, prev;
4208
4209         memcpy(&lun->log_pages.index, log_page_index_template,
4210                sizeof(log_page_index_template));
4211
4212         prev = -1;
4213         for (i = 0, j = 0, k = 0; i < CTL_NUM_LOG_PAGES; i++) {
4214
4215                 page_index = &lun->log_pages.index[i];
4216                 /*
4217                  * If this is a disk-only mode page, there's no point in
4218                  * setting it up.  For some pages, we have to have some
4219                  * basic information about the disk in order to calculate the
4220                  * mode page data.
4221                  */
4222                 if ((lun->be_lun->lun_type != T_DIRECT)
4223                  && (page_index->page_flags & CTL_PAGE_FLAG_DISK_ONLY))
4224                         continue;
4225
4226                 if (page_index->page_code == SLS_LOGICAL_BLOCK_PROVISIONING &&
4227                      lun->backend->lun_attr == NULL)
4228                         continue;
4229
4230                 if (page_index->page_code != prev) {
4231                         lun->log_pages.pages_page[j] = page_index->page_code;
4232                         prev = page_index->page_code;
4233                         j++;
4234                 }
4235                 lun->log_pages.subpages_page[k*2] = page_index->page_code;
4236                 lun->log_pages.subpages_page[k*2+1] = page_index->subpage;
4237                 k++;
4238         }
4239         lun->log_pages.index[0].page_data = &lun->log_pages.pages_page[0];
4240         lun->log_pages.index[0].page_len = j;
4241         lun->log_pages.index[1].page_data = &lun->log_pages.subpages_page[0];
4242         lun->log_pages.index[1].page_len = k * 2;
4243         lun->log_pages.index[2].page_data = &lun->log_pages.lbp_page[0];
4244         lun->log_pages.index[2].page_len = 12*CTL_NUM_LBP_PARAMS;
4245         lun->log_pages.index[3].page_data = (uint8_t *)&lun->log_pages.stat_page;
4246         lun->log_pages.index[3].page_len = sizeof(lun->log_pages.stat_page);
4247
4248         return (CTL_RETVAL_COMPLETE);
4249 }
4250
4251 static int
4252 hex2bin(const char *str, uint8_t *buf, int buf_size)
4253 {
4254         int i;
4255         u_char c;
4256
4257         memset(buf, 0, buf_size);
4258         while (isspace(str[0]))
4259                 str++;
4260         if (str[0] == '0' && (str[1] == 'x' || str[1] == 'X'))
4261                 str += 2;
4262         buf_size *= 2;
4263         for (i = 0; str[i] != 0 && i < buf_size; i++) {
4264                 c = str[i];
4265                 if (isdigit(c))
4266                         c -= '0';
4267                 else if (isalpha(c))
4268                         c -= isupper(c) ? 'A' - 10 : 'a' - 10;
4269                 else
4270                         break;
4271                 if (c >= 16)
4272                         break;
4273                 if ((i & 1) == 0)
4274                         buf[i / 2] |= (c << 4);
4275                 else
4276                         buf[i / 2] |= c;
4277         }
4278         return ((i + 1) / 2);
4279 }
4280
4281 /*
4282  * LUN allocation.
4283  *
4284  * Requirements:
4285  * - caller allocates and zeros LUN storage, or passes in a NULL LUN if he
4286  *   wants us to allocate the LUN and he can block.
4287  * - ctl_softc is always set
4288  * - be_lun is set if the LUN has a backend (needed for disk LUNs)
4289  *
4290  * Returns 0 for success, non-zero (errno) for failure.
4291  */
4292 static int
4293 ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *ctl_lun,
4294               struct ctl_be_lun *const be_lun)
4295 {
4296         struct ctl_lun *nlun, *lun;
4297         struct scsi_vpd_id_descriptor *desc;
4298         struct scsi_vpd_id_t10 *t10id;
4299         const char *eui, *naa, *scsiname, *vendor;
4300         int lun_number, i, lun_malloced;
4301         int devidlen, idlen1, idlen2 = 0, len;
4302
4303         if (be_lun == NULL)
4304                 return (EINVAL);
4305
4306         /*
4307          * We currently only support Direct Access or Processor LUN types.
4308          */
4309         switch (be_lun->lun_type) {
4310         case T_DIRECT:
4311                 break;
4312         case T_PROCESSOR:
4313                 break;
4314         case T_SEQUENTIAL:
4315         case T_CHANGER:
4316         default:
4317                 be_lun->lun_config_status(be_lun->be_lun,
4318                                           CTL_LUN_CONFIG_FAILURE);
4319                 break;
4320         }
4321         if (ctl_lun == NULL) {
4322                 lun = malloc(sizeof(*lun), M_CTL, M_WAITOK);
4323                 lun_malloced = 1;
4324         } else {
4325                 lun_malloced = 0;
4326                 lun = ctl_lun;
4327         }
4328
4329         memset(lun, 0, sizeof(*lun));
4330         if (lun_malloced)
4331                 lun->flags = CTL_LUN_MALLOCED;
4332
4333         /* Generate LUN ID. */
4334         devidlen = max(CTL_DEVID_MIN_LEN,
4335             strnlen(be_lun->device_id, CTL_DEVID_LEN));
4336         idlen1 = sizeof(*t10id) + devidlen;
4337         len = sizeof(struct scsi_vpd_id_descriptor) + idlen1;
4338         scsiname = ctl_get_opt(&be_lun->options, "scsiname");
4339         if (scsiname != NULL) {
4340                 idlen2 = roundup2(strlen(scsiname) + 1, 4);
4341                 len += sizeof(struct scsi_vpd_id_descriptor) + idlen2;
4342         }
4343         eui = ctl_get_opt(&be_lun->options, "eui");
4344         if (eui != NULL) {
4345                 len += sizeof(struct scsi_vpd_id_descriptor) + 16;
4346         }
4347         naa = ctl_get_opt(&be_lun->options, "naa");
4348         if (naa != NULL) {
4349                 len += sizeof(struct scsi_vpd_id_descriptor) + 16;
4350         }
4351         lun->lun_devid = malloc(sizeof(struct ctl_devid) + len,
4352             M_CTL, M_WAITOK | M_ZERO);
4353         desc = (struct scsi_vpd_id_descriptor *)lun->lun_devid->data;
4354         desc->proto_codeset = SVPD_ID_CODESET_ASCII;
4355         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_T10;
4356         desc->length = idlen1;
4357         t10id = (struct scsi_vpd_id_t10 *)&desc->identifier[0];
4358         memset(t10id->vendor, ' ', sizeof(t10id->vendor));
4359         if ((vendor = ctl_get_opt(&be_lun->options, "vendor")) == NULL) {
4360                 strncpy((char *)t10id->vendor, CTL_VENDOR, sizeof(t10id->vendor));
4361         } else {
4362                 strncpy(t10id->vendor, vendor,
4363                     min(sizeof(t10id->vendor), strlen(vendor)));
4364         }
4365         strncpy((char *)t10id->vendor_spec_id,
4366             (char *)be_lun->device_id, devidlen);
4367         if (scsiname != NULL) {
4368                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4369                     desc->length);
4370                 desc->proto_codeset = SVPD_ID_CODESET_UTF8;
4371                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4372                     SVPD_ID_TYPE_SCSI_NAME;
4373                 desc->length = idlen2;
4374                 strlcpy(desc->identifier, scsiname, idlen2);
4375         }
4376         if (eui != NULL) {
4377                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4378                     desc->length);
4379                 desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4380                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4381                     SVPD_ID_TYPE_EUI64;
4382                 desc->length = hex2bin(eui, desc->identifier, 16);
4383                 desc->length = desc->length > 12 ? 16 :
4384                     (desc->length > 8 ? 12 : 8);
4385                 len -= 16 - desc->length;
4386         }
4387         if (naa != NULL) {
4388                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4389                     desc->length);
4390                 desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4391                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4392                     SVPD_ID_TYPE_NAA;
4393                 desc->length = hex2bin(naa, desc->identifier, 16);
4394                 desc->length = desc->length > 8 ? 16 : 8;
4395                 len -= 16 - desc->length;
4396         }
4397         lun->lun_devid->len = len;
4398
4399         mtx_lock(&ctl_softc->ctl_lock);
4400         /*
4401          * See if the caller requested a particular LUN number.  If so, see
4402          * if it is available.  Otherwise, allocate the first available LUN.
4403          */
4404         if (be_lun->flags & CTL_LUN_FLAG_ID_REQ) {
4405                 if ((be_lun->req_lun_id > (CTL_MAX_LUNS - 1))
4406                  || (ctl_is_set(ctl_softc->ctl_lun_mask, be_lun->req_lun_id))) {
4407                         mtx_unlock(&ctl_softc->ctl_lock);
4408                         if (be_lun->req_lun_id > (CTL_MAX_LUNS - 1)) {
4409                                 printf("ctl: requested LUN ID %d is higher "
4410                                        "than CTL_MAX_LUNS - 1 (%d)\n",
4411                                        be_lun->req_lun_id, CTL_MAX_LUNS - 1);
4412                         } else {
4413                                 /*
4414                                  * XXX KDM return an error, or just assign
4415                                  * another LUN ID in this case??
4416                                  */
4417                                 printf("ctl: requested LUN ID %d is already "
4418                                        "in use\n", be_lun->req_lun_id);
4419                         }
4420                         if (lun->flags & CTL_LUN_MALLOCED)
4421                                 free(lun, M_CTL);
4422                         be_lun->lun_config_status(be_lun->be_lun,
4423                                                   CTL_LUN_CONFIG_FAILURE);
4424                         return (ENOSPC);
4425                 }
4426                 lun_number = be_lun->req_lun_id;
4427         } else {
4428                 lun_number = ctl_ffz(ctl_softc->ctl_lun_mask, 0, CTL_MAX_LUNS);
4429                 if (lun_number == -1) {
4430                         mtx_unlock(&ctl_softc->ctl_lock);
4431                         printf("ctl: can't allocate LUN, out of LUNs\n");
4432                         if (lun->flags & CTL_LUN_MALLOCED)
4433                                 free(lun, M_CTL);
4434                         be_lun->lun_config_status(be_lun->be_lun,
4435                                                   CTL_LUN_CONFIG_FAILURE);
4436                         return (ENOSPC);
4437                 }
4438         }
4439         ctl_set_mask(ctl_softc->ctl_lun_mask, lun_number);
4440
4441         mtx_init(&lun->lun_lock, "CTL LUN", NULL, MTX_DEF);
4442         lun->lun = lun_number;
4443         lun->be_lun = be_lun;
4444         /*
4445          * The processor LUN is always enabled.  Disk LUNs come on line
4446          * disabled, and must be enabled by the backend.
4447          */
4448         lun->flags |= CTL_LUN_DISABLED;
4449         lun->backend = be_lun->be;
4450         be_lun->ctl_lun = lun;
4451         be_lun->lun_id = lun_number;
4452         atomic_add_int(&be_lun->be->num_luns, 1);
4453         if (be_lun->flags & CTL_LUN_FLAG_OFFLINE)
4454                 lun->flags |= CTL_LUN_OFFLINE;
4455
4456         if (be_lun->flags & CTL_LUN_FLAG_POWERED_OFF)
4457                 lun->flags |= CTL_LUN_STOPPED;
4458
4459         if (be_lun->flags & CTL_LUN_FLAG_INOPERABLE)
4460                 lun->flags |= CTL_LUN_INOPERABLE;
4461
4462         if (be_lun->flags & CTL_LUN_FLAG_PRIMARY)
4463                 lun->flags |= CTL_LUN_PRIMARY_SC;
4464
4465         lun->ctl_softc = ctl_softc;
4466 #ifdef CTL_TIME_IO
4467         lun->last_busy = getsbinuptime();
4468 #endif
4469         TAILQ_INIT(&lun->ooa_queue);
4470         TAILQ_INIT(&lun->blocked_queue);
4471         STAILQ_INIT(&lun->error_list);
4472         ctl_tpc_lun_init(lun);
4473
4474         /*
4475          * Initialize the mode and log page index.
4476          */
4477         ctl_init_page_index(lun);
4478         ctl_init_log_page_index(lun);
4479
4480         /*
4481          * Now, before we insert this lun on the lun list, set the lun
4482          * inventory changed UA for all other luns.
4483          */
4484         STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) {
4485                 mtx_lock(&nlun->lun_lock);
4486                 ctl_est_ua_all(nlun, -1, CTL_UA_LUN_CHANGE);
4487                 mtx_unlock(&nlun->lun_lock);
4488         }
4489
4490         STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links);
4491
4492         ctl_softc->ctl_luns[lun_number] = lun;
4493
4494         ctl_softc->num_luns++;
4495
4496         /* Setup statistics gathering */
4497         lun->stats.device_type = be_lun->lun_type;
4498         lun->stats.lun_number = lun_number;
4499         if (lun->stats.device_type == T_DIRECT)
4500                 lun->stats.blocksize = be_lun->blocksize;
4501         else
4502                 lun->stats.flags = CTL_LUN_STATS_NO_BLOCKSIZE;
4503         for (i = 0;i < CTL_MAX_PORTS;i++)
4504                 lun->stats.ports[i].targ_port = i;
4505
4506         mtx_unlock(&ctl_softc->ctl_lock);
4507
4508         lun->be_lun->lun_config_status(lun->be_lun->be_lun, CTL_LUN_CONFIG_OK);
4509         return (0);
4510 }
4511
4512 /*
4513  * Delete a LUN.
4514  * Assumptions:
4515  * - LUN has already been marked invalid and any pending I/O has been taken
4516  *   care of.
4517  */
4518 static int
4519 ctl_free_lun(struct ctl_lun *lun)
4520 {
4521         struct ctl_softc *softc;
4522         struct ctl_lun *nlun;
4523         int i;
4524
4525         softc = lun->ctl_softc;
4526
4527         mtx_assert(&softc->ctl_lock, MA_OWNED);
4528
4529         STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links);
4530
4531         ctl_clear_mask(softc->ctl_lun_mask, lun->lun);
4532
4533         softc->ctl_luns[lun->lun] = NULL;
4534
4535         if (!TAILQ_EMPTY(&lun->ooa_queue))
4536                 panic("Freeing a LUN %p with outstanding I/O!!\n", lun);
4537
4538         softc->num_luns--;
4539
4540         /*
4541          * Tell the backend to free resources, if this LUN has a backend.
4542          */
4543         atomic_subtract_int(&lun->be_lun->be->num_luns, 1);
4544         lun->be_lun->lun_shutdown(lun->be_lun->be_lun);
4545
4546         ctl_tpc_lun_shutdown(lun);
4547         mtx_destroy(&lun->lun_lock);
4548         free(lun->lun_devid, M_CTL);
4549         for (i = 0; i < CTL_MAX_PORTS; i++)
4550                 free(lun->pending_ua[i], M_CTL);
4551         for (i = 0; i < CTL_MAX_PORTS; i++)
4552                 free(lun->pr_keys[i], M_CTL);
4553         free(lun->write_buffer, M_CTL);
4554         if (lun->flags & CTL_LUN_MALLOCED)
4555                 free(lun, M_CTL);
4556
4557         STAILQ_FOREACH(nlun, &softc->lun_list, links) {
4558                 mtx_lock(&nlun->lun_lock);
4559                 ctl_est_ua_all(nlun, -1, CTL_UA_LUN_CHANGE);
4560                 mtx_unlock(&nlun->lun_lock);
4561         }
4562
4563         return (0);
4564 }
4565
4566 static void
4567 ctl_create_lun(struct ctl_be_lun *be_lun)
4568 {
4569
4570         /*
4571          * ctl_alloc_lun() should handle all potential failure cases.
4572          */
4573         ctl_alloc_lun(control_softc, NULL, be_lun);
4574 }
4575
4576 int
4577 ctl_add_lun(struct ctl_be_lun *be_lun)
4578 {
4579         struct ctl_softc *softc = control_softc;
4580
4581         mtx_lock(&softc->ctl_lock);
4582         STAILQ_INSERT_TAIL(&softc->pending_lun_queue, be_lun, links);
4583         mtx_unlock(&softc->ctl_lock);
4584         wakeup(&softc->pending_lun_queue);
4585
4586         return (0);
4587 }
4588
4589 int
4590 ctl_enable_lun(struct ctl_be_lun *be_lun)
4591 {
4592         struct ctl_softc *softc;
4593         struct ctl_port *port, *nport;
4594         struct ctl_lun *lun;
4595         int retval;
4596
4597         lun = (struct ctl_lun *)be_lun->ctl_lun;
4598         softc = lun->ctl_softc;
4599
4600         mtx_lock(&softc->ctl_lock);
4601         mtx_lock(&lun->lun_lock);
4602         if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4603                 /*
4604                  * eh?  Why did we get called if the LUN is already
4605                  * enabled?
4606                  */
4607                 mtx_unlock(&lun->lun_lock);
4608                 mtx_unlock(&softc->ctl_lock);
4609                 return (0);
4610         }
4611         lun->flags &= ~CTL_LUN_DISABLED;
4612         mtx_unlock(&lun->lun_lock);
4613
4614         for (port = STAILQ_FIRST(&softc->port_list); port != NULL; port = nport) {
4615                 nport = STAILQ_NEXT(port, links);
4616                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0 ||
4617                     port->lun_map != NULL || port->lun_enable == NULL)
4618                         continue;
4619
4620                 /*
4621                  * Drop the lock while we call the FETD's enable routine.
4622                  * This can lead to a callback into CTL (at least in the
4623                  * case of the internal initiator frontend.
4624                  */
4625                 mtx_unlock(&softc->ctl_lock);
4626                 retval = port->lun_enable(port->targ_lun_arg, lun->lun);
4627                 mtx_lock(&softc->ctl_lock);
4628                 if (retval != 0) {
4629                         printf("%s: FETD %s port %d returned error "
4630                                "%d for lun_enable on lun %jd\n",
4631                                __func__, port->port_name, port->targ_port,
4632                                retval, (intmax_t)lun->lun);
4633                 }
4634         }
4635
4636         mtx_unlock(&softc->ctl_lock);
4637         ctl_isc_announce_lun(lun);
4638
4639         return (0);
4640 }
4641
4642 int
4643 ctl_disable_lun(struct ctl_be_lun *be_lun)
4644 {
4645         struct ctl_softc *softc;
4646         struct ctl_port *port;
4647         struct ctl_lun *lun;
4648         int retval;
4649
4650         lun = (struct ctl_lun *)be_lun->ctl_lun;
4651         softc = lun->ctl_softc;
4652
4653         mtx_lock(&softc->ctl_lock);
4654         mtx_lock(&lun->lun_lock);
4655         if (lun->flags & CTL_LUN_DISABLED) {
4656                 mtx_unlock(&lun->lun_lock);
4657                 mtx_unlock(&softc->ctl_lock);
4658                 return (0);
4659         }
4660         lun->flags |= CTL_LUN_DISABLED;
4661         mtx_unlock(&lun->lun_lock);
4662
4663         STAILQ_FOREACH(port, &softc->port_list, links) {
4664                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0 ||
4665                     port->lun_map != NULL || port->lun_disable == NULL)
4666                         continue;
4667
4668                 /*
4669                  * Drop the lock before we call the frontend's disable
4670                  * routine, to avoid lock order reversals.
4671                  *
4672                  * XXX KDM what happens if the frontend list changes while
4673                  * we're traversing it?  It's unlikely, but should be handled.
4674                  */
4675                 mtx_unlock(&softc->ctl_lock);
4676                 retval = port->lun_disable(port->targ_lun_arg, lun->lun);
4677                 mtx_lock(&softc->ctl_lock);
4678                 if (retval != 0) {
4679                         printf("%s: FETD %s port %d returned error "
4680                                "%d for lun_disable on lun %jd\n",
4681                                __func__, port->port_name, port->targ_port,
4682                                retval, (intmax_t)lun->lun);
4683                 }
4684         }
4685
4686         mtx_unlock(&softc->ctl_lock);
4687         ctl_isc_announce_lun(lun);
4688
4689         return (0);
4690 }
4691
4692 int
4693 ctl_start_lun(struct ctl_be_lun *be_lun)
4694 {
4695         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4696
4697         mtx_lock(&lun->lun_lock);
4698         lun->flags &= ~CTL_LUN_STOPPED;
4699         mtx_unlock(&lun->lun_lock);
4700         return (0);
4701 }
4702
4703 int
4704 ctl_stop_lun(struct ctl_be_lun *be_lun)
4705 {
4706         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4707
4708         mtx_lock(&lun->lun_lock);
4709         lun->flags |= CTL_LUN_STOPPED;
4710         mtx_unlock(&lun->lun_lock);
4711         return (0);
4712 }
4713
4714 int
4715 ctl_lun_offline(struct ctl_be_lun *be_lun)
4716 {
4717         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4718
4719         mtx_lock(&lun->lun_lock);
4720         lun->flags |= CTL_LUN_OFFLINE;
4721         mtx_unlock(&lun->lun_lock);
4722         return (0);
4723 }
4724
4725 int
4726 ctl_lun_online(struct ctl_be_lun *be_lun)
4727 {
4728         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4729
4730         mtx_lock(&lun->lun_lock);
4731         lun->flags &= ~CTL_LUN_OFFLINE;
4732         mtx_unlock(&lun->lun_lock);
4733         return (0);
4734 }
4735
4736 int
4737 ctl_lun_primary(struct ctl_be_lun *be_lun)
4738 {
4739         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4740
4741         mtx_lock(&lun->lun_lock);
4742         lun->flags |= CTL_LUN_PRIMARY_SC;
4743         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
4744         mtx_unlock(&lun->lun_lock);
4745         ctl_isc_announce_lun(lun);
4746         return (0);
4747 }
4748
4749 int
4750 ctl_lun_secondary(struct ctl_be_lun *be_lun)
4751 {
4752         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4753
4754         mtx_lock(&lun->lun_lock);
4755         lun->flags &= ~CTL_LUN_PRIMARY_SC;
4756         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
4757         mtx_unlock(&lun->lun_lock);
4758         ctl_isc_announce_lun(lun);
4759         return (0);
4760 }
4761
4762 int
4763 ctl_invalidate_lun(struct ctl_be_lun *be_lun)
4764 {
4765         struct ctl_softc *softc;
4766         struct ctl_lun *lun;
4767
4768         lun = (struct ctl_lun *)be_lun->ctl_lun;
4769         softc = lun->ctl_softc;
4770
4771         mtx_lock(&lun->lun_lock);
4772
4773         /*
4774          * The LUN needs to be disabled before it can be marked invalid.
4775          */
4776         if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4777                 mtx_unlock(&lun->lun_lock);
4778                 return (-1);
4779         }
4780         /*
4781          * Mark the LUN invalid.
4782          */
4783         lun->flags |= CTL_LUN_INVALID;
4784
4785         /*
4786          * If there is nothing in the OOA queue, go ahead and free the LUN.
4787          * If we have something in the OOA queue, we'll free it when the
4788          * last I/O completes.
4789          */
4790         if (TAILQ_EMPTY(&lun->ooa_queue)) {
4791                 mtx_unlock(&lun->lun_lock);
4792                 mtx_lock(&softc->ctl_lock);
4793                 ctl_free_lun(lun);
4794                 mtx_unlock(&softc->ctl_lock);
4795         } else
4796                 mtx_unlock(&lun->lun_lock);
4797
4798         return (0);
4799 }
4800
4801 int
4802 ctl_lun_inoperable(struct ctl_be_lun *be_lun)
4803 {
4804         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4805
4806         mtx_lock(&lun->lun_lock);
4807         lun->flags |= CTL_LUN_INOPERABLE;
4808         mtx_unlock(&lun->lun_lock);
4809         return (0);
4810 }
4811
4812 int
4813 ctl_lun_operable(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_INOPERABLE;
4819         mtx_unlock(&lun->lun_lock);
4820         return (0);
4821 }
4822
4823 void
4824 ctl_lun_capacity_changed(struct ctl_be_lun *be_lun)
4825 {
4826         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4827         union ctl_ha_msg msg;
4828
4829         mtx_lock(&lun->lun_lock);
4830         ctl_est_ua_all(lun, -1, CTL_UA_CAPACITY_CHANGED);
4831         mtx_unlock(&lun->lun_lock);
4832         if (lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
4833                 /* Send msg to other side. */
4834                 bzero(&msg.ua, sizeof(msg.ua));
4835                 msg.hdr.msg_type = CTL_MSG_UA;
4836                 msg.hdr.nexus.initid = -1;
4837                 msg.hdr.nexus.targ_port = -1;
4838                 msg.hdr.nexus.targ_lun = lun->lun;
4839                 msg.hdr.nexus.targ_mapped_lun = lun->lun;
4840                 msg.ua.ua_all = 1;
4841                 msg.ua.ua_set = 1;
4842                 msg.ua.ua_type = CTL_UA_CAPACITY_CHANGED;
4843                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg.ua),
4844                     M_WAITOK);
4845         }
4846 }
4847
4848 /*
4849  * Backend "memory move is complete" callback for requests that never
4850  * make it down to say RAIDCore's configuration code.
4851  */
4852 int
4853 ctl_config_move_done(union ctl_io *io)
4854 {
4855         int retval;
4856
4857         CTL_DEBUG_PRINT(("ctl_config_move_done\n"));
4858         KASSERT(io->io_hdr.io_type == CTL_IO_SCSI,
4859             ("Config I/O type isn't CTL_IO_SCSI (%d)!", io->io_hdr.io_type));
4860
4861         if ((io->io_hdr.port_status != 0) &&
4862             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
4863              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
4864                 /*
4865                  * For hardware error sense keys, the sense key
4866                  * specific value is defined to be a retry count,
4867                  * but we use it to pass back an internal FETD
4868                  * error code.  XXX KDM  Hopefully the FETD is only
4869                  * using 16 bits for an error code, since that's
4870                  * all the space we have in the sks field.
4871                  */
4872                 ctl_set_internal_failure(&io->scsiio,
4873                                          /*sks_valid*/ 1,
4874                                          /*retry_count*/
4875                                          io->io_hdr.port_status);
4876         }
4877
4878         if (ctl_debug & CTL_DEBUG_CDB_DATA)
4879                 ctl_data_print(io);
4880         if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN) ||
4881             ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
4882              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS) ||
4883             ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) {
4884                 /*
4885                  * XXX KDM just assuming a single pointer here, and not a
4886                  * S/G list.  If we start using S/G lists for config data,
4887                  * we'll need to know how to clean them up here as well.
4888                  */
4889                 if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
4890                         free(io->scsiio.kern_data_ptr, M_CTL);
4891                 ctl_done(io);
4892                 retval = CTL_RETVAL_COMPLETE;
4893         } else {
4894                 /*
4895                  * XXX KDM now we need to continue data movement.  Some
4896                  * options:
4897                  * - call ctl_scsiio() again?  We don't do this for data
4898                  *   writes, because for those at least we know ahead of
4899                  *   time where the write will go and how long it is.  For
4900                  *   config writes, though, that information is largely
4901                  *   contained within the write itself, thus we need to
4902                  *   parse out the data again.
4903                  *
4904                  * - Call some other function once the data is in?
4905                  */
4906
4907                 /*
4908                  * XXX KDM call ctl_scsiio() again for now, and check flag
4909                  * bits to see whether we're allocated or not.
4910                  */
4911                 retval = ctl_scsiio(&io->scsiio);
4912         }
4913         return (retval);
4914 }
4915
4916 /*
4917  * This gets called by a backend driver when it is done with a
4918  * data_submit method.
4919  */
4920 void
4921 ctl_data_submit_done(union ctl_io *io)
4922 {
4923         /*
4924          * If the IO_CONT flag is set, we need to call the supplied
4925          * function to continue processing the I/O, instead of completing
4926          * the I/O just yet.
4927          *
4928          * If there is an error, though, we don't want to keep processing.
4929          * Instead, just send status back to the initiator.
4930          */
4931         if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
4932             (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
4933             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
4934              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
4935                 io->scsiio.io_cont(io);
4936                 return;
4937         }
4938         ctl_done(io);
4939 }
4940
4941 /*
4942  * This gets called by a backend driver when it is done with a
4943  * configuration write.
4944  */
4945 void
4946 ctl_config_write_done(union ctl_io *io)
4947 {
4948         uint8_t *buf;
4949
4950         /*
4951          * If the IO_CONT flag is set, we need to call the supplied
4952          * function to continue processing the I/O, instead of completing
4953          * the I/O just yet.
4954          *
4955          * If there is an error, though, we don't want to keep processing.
4956          * Instead, just send status back to the initiator.
4957          */
4958         if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
4959             (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
4960             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
4961              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
4962                 io->scsiio.io_cont(io);
4963                 return;
4964         }
4965         /*
4966          * Since a configuration write can be done for commands that actually
4967          * have data allocated, like write buffer, and commands that have
4968          * no data, like start/stop unit, we need to check here.
4969          */
4970         if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
4971                 buf = io->scsiio.kern_data_ptr;
4972         else
4973                 buf = NULL;
4974         ctl_done(io);
4975         if (buf)
4976                 free(buf, M_CTL);
4977 }
4978
4979 void
4980 ctl_config_read_done(union ctl_io *io)
4981 {
4982         uint8_t *buf;
4983
4984         /*
4985          * If there is some error -- we are done, skip data transfer.
4986          */
4987         if ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0 ||
4988             ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
4989              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) {
4990                 if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
4991                         buf = io->scsiio.kern_data_ptr;
4992                 else
4993                         buf = NULL;
4994                 ctl_done(io);
4995                 if (buf)
4996                         free(buf, M_CTL);
4997                 return;
4998         }
4999
5000         /*
5001          * If the IO_CONT flag is set, we need to call the supplied
5002          * function to continue processing the I/O, instead of completing
5003          * the I/O just yet.
5004          */
5005         if (io->io_hdr.flags & CTL_FLAG_IO_CONT) {
5006                 io->scsiio.io_cont(io);
5007                 return;
5008         }
5009
5010         ctl_datamove(io);
5011 }
5012
5013 /*
5014  * SCSI release command.
5015  */
5016 int
5017 ctl_scsi_release(struct ctl_scsiio *ctsio)
5018 {
5019         struct ctl_lun *lun;
5020         uint32_t residx;
5021
5022         CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
5023
5024         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5025         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5026
5027         /*
5028          * XXX KDM right now, we only support LUN reservation.  We don't
5029          * support 3rd party reservations, or extent reservations, which
5030          * might actually need the parameter list.  If we've gotten this
5031          * far, we've got a LUN reservation.  Anything else got kicked out
5032          * above.  So, according to SPC, ignore the length.
5033          */
5034
5035         mtx_lock(&lun->lun_lock);
5036
5037         /*
5038          * According to SPC, it is not an error for an intiator to attempt
5039          * to release a reservation on a LUN that isn't reserved, or that
5040          * is reserved by another initiator.  The reservation can only be
5041          * released, though, by the initiator who made it or by one of
5042          * several reset type events.
5043          */
5044         if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == residx))
5045                         lun->flags &= ~CTL_LUN_RESERVED;
5046
5047         mtx_unlock(&lun->lun_lock);
5048
5049         ctl_set_success(ctsio);
5050         ctl_done((union ctl_io *)ctsio);
5051         return (CTL_RETVAL_COMPLETE);
5052 }
5053
5054 int
5055 ctl_scsi_reserve(struct ctl_scsiio *ctsio)
5056 {
5057         struct ctl_lun *lun;
5058         uint32_t residx;
5059
5060         CTL_DEBUG_PRINT(("ctl_reserve\n"));
5061
5062         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5063         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5064
5065         /*
5066          * XXX KDM right now, we only support LUN reservation.  We don't
5067          * support 3rd party reservations, or extent reservations, which
5068          * might actually need the parameter list.  If we've gotten this
5069          * far, we've got a LUN reservation.  Anything else got kicked out
5070          * above.  So, according to SPC, ignore the length.
5071          */
5072
5073         mtx_lock(&lun->lun_lock);
5074         if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx != residx)) {
5075                 ctl_set_reservation_conflict(ctsio);
5076                 goto bailout;
5077         }
5078         lun->flags |= CTL_LUN_RESERVED;
5079         lun->res_idx = residx;
5080         ctl_set_success(ctsio);
5081
5082 bailout:
5083         mtx_unlock(&lun->lun_lock);
5084         ctl_done((union ctl_io *)ctsio);
5085         return (CTL_RETVAL_COMPLETE);
5086 }
5087
5088 int
5089 ctl_start_stop(struct ctl_scsiio *ctsio)
5090 {
5091         struct scsi_start_stop_unit *cdb;
5092         struct ctl_lun *lun;
5093         int retval;
5094
5095         CTL_DEBUG_PRINT(("ctl_start_stop\n"));
5096
5097         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5098         cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
5099
5100         /*
5101          * XXX KDM
5102          * We don't support the immediate bit on a stop unit.  In order to
5103          * do that, we would need to code up a way to know that a stop is
5104          * pending, and hold off any new commands until it completes, one
5105          * way or another.  Then we could accept or reject those commands
5106          * depending on its status.  We would almost need to do the reverse
5107          * of what we do below for an immediate start -- return the copy of
5108          * the ctl_io to the FETD with status to send to the host (and to
5109          * free the copy!) and then free the original I/O once the stop
5110          * actually completes.  That way, the OOA queue mechanism can work
5111          * to block commands that shouldn't proceed.  Another alternative
5112          * would be to put the copy in the queue in place of the original,
5113          * and return the original back to the caller.  That could be
5114          * slightly safer..
5115          */
5116         if ((cdb->byte2 & SSS_IMMED)
5117          && ((cdb->how & SSS_START) == 0)) {
5118                 ctl_set_invalid_field(ctsio,
5119                                       /*sks_valid*/ 1,
5120                                       /*command*/ 1,
5121                                       /*field*/ 1,
5122                                       /*bit_valid*/ 1,
5123                                       /*bit*/ 0);
5124                 ctl_done((union ctl_io *)ctsio);
5125                 return (CTL_RETVAL_COMPLETE);
5126         }
5127
5128         if ((lun->flags & CTL_LUN_PR_RESERVED)
5129          && ((cdb->how & SSS_START)==0)) {
5130                 uint32_t residx;
5131
5132                 residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5133                 if (ctl_get_prkey(lun, residx) == 0
5134                  || (lun->pr_res_idx!=residx && lun->res_type < 4)) {
5135
5136                         ctl_set_reservation_conflict(ctsio);
5137                         ctl_done((union ctl_io *)ctsio);
5138                         return (CTL_RETVAL_COMPLETE);
5139                 }
5140         }
5141
5142         /*
5143          * If there is no backend on this device, we can't start or stop
5144          * it.  In theory we shouldn't get any start/stop commands in the
5145          * first place at this level if the LUN doesn't have a backend.
5146          * That should get stopped by the command decode code.
5147          */
5148         if (lun->backend == NULL) {
5149                 ctl_set_invalid_opcode(ctsio);
5150                 ctl_done((union ctl_io *)ctsio);
5151                 return (CTL_RETVAL_COMPLETE);
5152         }
5153
5154         /*
5155          * In the non-immediate case, we send the request to
5156          * the backend and return status to the user when
5157          * it is done.
5158          *
5159          * In the immediate case, we allocate a new ctl_io
5160          * to hold a copy of the request, and send that to
5161          * the backend.  We then set good status on the
5162          * user's request and return it immediately.
5163          */
5164         if (cdb->byte2 & SSS_IMMED) {
5165                 union ctl_io *new_io;
5166
5167                 new_io = ctl_alloc_io(ctsio->io_hdr.pool);
5168                 ctl_copy_io((union ctl_io *)ctsio, new_io);
5169                 retval = lun->backend->config_write(new_io);
5170                 ctl_set_success(ctsio);
5171                 ctl_done((union ctl_io *)ctsio);
5172         } else {
5173                 retval = lun->backend->config_write(
5174                         (union ctl_io *)ctsio);
5175         }
5176         return (retval);
5177 }
5178
5179 /*
5180  * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
5181  * we don't really do anything with the LBA and length fields if the user
5182  * passes them in.  Instead we'll just flush out the cache for the entire
5183  * LUN.
5184  */
5185 int
5186 ctl_sync_cache(struct ctl_scsiio *ctsio)
5187 {
5188         struct ctl_lun *lun;
5189         struct ctl_softc *softc;
5190         struct ctl_lba_len_flags *lbalen;
5191         uint64_t starting_lba;
5192         uint32_t block_count;
5193         int retval;
5194         uint8_t byte2;
5195
5196         CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
5197
5198         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5199         softc = lun->ctl_softc;
5200         retval = 0;
5201
5202         switch (ctsio->cdb[0]) {
5203         case SYNCHRONIZE_CACHE: {
5204                 struct scsi_sync_cache *cdb;
5205                 cdb = (struct scsi_sync_cache *)ctsio->cdb;
5206
5207                 starting_lba = scsi_4btoul(cdb->begin_lba);
5208                 block_count = scsi_2btoul(cdb->lb_count);
5209                 byte2 = cdb->byte2;
5210                 break;
5211         }
5212         case SYNCHRONIZE_CACHE_16: {
5213                 struct scsi_sync_cache_16 *cdb;
5214                 cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5215
5216                 starting_lba = scsi_8btou64(cdb->begin_lba);
5217                 block_count = scsi_4btoul(cdb->lb_count);
5218                 byte2 = cdb->byte2;
5219                 break;
5220         }
5221         default:
5222                 ctl_set_invalid_opcode(ctsio);
5223                 ctl_done((union ctl_io *)ctsio);
5224                 goto bailout;
5225                 break; /* NOTREACHED */
5226         }
5227
5228         /*
5229          * We check the LBA and length, but don't do anything with them.
5230          * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5231          * get flushed.  This check will just help satisfy anyone who wants
5232          * to see an error for an out of range LBA.
5233          */
5234         if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5235                 ctl_set_lba_out_of_range(ctsio);
5236                 ctl_done((union ctl_io *)ctsio);
5237                 goto bailout;
5238         }
5239
5240         /*
5241          * If this LUN has no backend, we can't flush the cache anyway.
5242          */
5243         if (lun->backend == NULL) {
5244                 ctl_set_invalid_opcode(ctsio);
5245                 ctl_done((union ctl_io *)ctsio);
5246                 goto bailout;
5247         }
5248
5249         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5250         lbalen->lba = starting_lba;
5251         lbalen->len = block_count;
5252         lbalen->flags = byte2;
5253         retval = lun->backend->config_write((union ctl_io *)ctsio);
5254
5255 bailout:
5256         return (retval);
5257 }
5258
5259 int
5260 ctl_format(struct ctl_scsiio *ctsio)
5261 {
5262         struct scsi_format *cdb;
5263         struct ctl_lun *lun;
5264         int length, defect_list_len;
5265
5266         CTL_DEBUG_PRINT(("ctl_format\n"));
5267
5268         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5269
5270         cdb = (struct scsi_format *)ctsio->cdb;
5271
5272         length = 0;
5273         if (cdb->byte2 & SF_FMTDATA) {
5274                 if (cdb->byte2 & SF_LONGLIST)
5275                         length = sizeof(struct scsi_format_header_long);
5276                 else
5277                         length = sizeof(struct scsi_format_header_short);
5278         }
5279
5280         if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5281          && (length > 0)) {
5282                 ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5283                 ctsio->kern_data_len = length;
5284                 ctsio->kern_total_len = length;
5285                 ctsio->kern_data_resid = 0;
5286                 ctsio->kern_rel_offset = 0;
5287                 ctsio->kern_sg_entries = 0;
5288                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5289                 ctsio->be_move_done = ctl_config_move_done;
5290                 ctl_datamove((union ctl_io *)ctsio);
5291
5292                 return (CTL_RETVAL_COMPLETE);
5293         }
5294
5295         defect_list_len = 0;
5296
5297         if (cdb->byte2 & SF_FMTDATA) {
5298                 if (cdb->byte2 & SF_LONGLIST) {
5299                         struct scsi_format_header_long *header;
5300
5301                         header = (struct scsi_format_header_long *)
5302                                 ctsio->kern_data_ptr;
5303
5304                         defect_list_len = scsi_4btoul(header->defect_list_len);
5305                         if (defect_list_len != 0) {
5306                                 ctl_set_invalid_field(ctsio,
5307                                                       /*sks_valid*/ 1,
5308                                                       /*command*/ 0,
5309                                                       /*field*/ 2,
5310                                                       /*bit_valid*/ 0,
5311                                                       /*bit*/ 0);
5312                                 goto bailout;
5313                         }
5314                 } else {
5315                         struct scsi_format_header_short *header;
5316
5317                         header = (struct scsi_format_header_short *)
5318                                 ctsio->kern_data_ptr;
5319
5320                         defect_list_len = scsi_2btoul(header->defect_list_len);
5321                         if (defect_list_len != 0) {
5322                                 ctl_set_invalid_field(ctsio,
5323                                                       /*sks_valid*/ 1,
5324                                                       /*command*/ 0,
5325                                                       /*field*/ 2,
5326                                                       /*bit_valid*/ 0,
5327                                                       /*bit*/ 0);
5328                                 goto bailout;
5329                         }
5330                 }
5331         }
5332
5333         /*
5334          * The format command will clear out the "Medium format corrupted"
5335          * status if set by the configuration code.  That status is really
5336          * just a way to notify the host that we have lost the media, and
5337          * get them to issue a command that will basically make them think
5338          * they're blowing away the media.
5339          */
5340         mtx_lock(&lun->lun_lock);
5341         lun->flags &= ~CTL_LUN_INOPERABLE;
5342         mtx_unlock(&lun->lun_lock);
5343
5344         ctl_set_success(ctsio);
5345 bailout:
5346
5347         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5348                 free(ctsio->kern_data_ptr, M_CTL);
5349                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5350         }
5351
5352         ctl_done((union ctl_io *)ctsio);
5353         return (CTL_RETVAL_COMPLETE);
5354 }
5355
5356 int
5357 ctl_read_buffer(struct ctl_scsiio *ctsio)
5358 {
5359         struct ctl_lun *lun;
5360         uint64_t buffer_offset;
5361         uint32_t len;
5362         uint8_t byte2;
5363         static uint8_t descr[4];
5364         static uint8_t echo_descr[4] = { 0 };
5365
5366         CTL_DEBUG_PRINT(("ctl_read_buffer\n"));
5367         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5368         switch (ctsio->cdb[0]) {
5369         case READ_BUFFER: {
5370                 struct scsi_read_buffer *cdb;
5371
5372                 cdb = (struct scsi_read_buffer *)ctsio->cdb;
5373                 buffer_offset = scsi_3btoul(cdb->offset);
5374                 len = scsi_3btoul(cdb->length);
5375                 byte2 = cdb->byte2;
5376                 break;
5377         }
5378         case READ_BUFFER_16: {
5379                 struct scsi_read_buffer_16 *cdb;
5380
5381                 cdb = (struct scsi_read_buffer_16 *)ctsio->cdb;
5382                 buffer_offset = scsi_8btou64(cdb->offset);
5383                 len = scsi_4btoul(cdb->length);
5384                 byte2 = cdb->byte2;
5385                 break;
5386         }
5387         default: /* This shouldn't happen. */
5388                 ctl_set_invalid_opcode(ctsio);
5389                 ctl_done((union ctl_io *)ctsio);
5390                 return (CTL_RETVAL_COMPLETE);
5391         }
5392
5393         if ((byte2 & RWB_MODE) != RWB_MODE_DATA &&
5394             (byte2 & RWB_MODE) != RWB_MODE_ECHO_DESCR &&
5395             (byte2 & RWB_MODE) != RWB_MODE_DESCR) {
5396                 ctl_set_invalid_field(ctsio,
5397                                       /*sks_valid*/ 1,
5398                                       /*command*/ 1,
5399                                       /*field*/ 1,
5400                                       /*bit_valid*/ 1,
5401                                       /*bit*/ 4);
5402                 ctl_done((union ctl_io *)ctsio);
5403                 return (CTL_RETVAL_COMPLETE);
5404         }
5405
5406         if (buffer_offset > CTL_WRITE_BUFFER_SIZE ||
5407             buffer_offset + len > CTL_WRITE_BUFFER_SIZE) {
5408                 ctl_set_invalid_field(ctsio,
5409                                       /*sks_valid*/ 1,
5410                                       /*command*/ 1,
5411                                       /*field*/ 6,
5412                                       /*bit_valid*/ 0,
5413                                       /*bit*/ 0);
5414                 ctl_done((union ctl_io *)ctsio);
5415                 return (CTL_RETVAL_COMPLETE);
5416         }
5417
5418         if ((byte2 & RWB_MODE) == RWB_MODE_DESCR) {
5419                 descr[0] = 0;
5420                 scsi_ulto3b(CTL_WRITE_BUFFER_SIZE, &descr[1]);
5421                 ctsio->kern_data_ptr = descr;
5422                 len = min(len, sizeof(descr));
5423         } else if ((byte2 & RWB_MODE) == RWB_MODE_ECHO_DESCR) {
5424                 ctsio->kern_data_ptr = echo_descr;
5425                 len = min(len, sizeof(echo_descr));
5426         } else {
5427                 if (lun->write_buffer == NULL) {
5428                         lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE,
5429                             M_CTL, M_WAITOK);
5430                 }
5431                 ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5432         }
5433         ctsio->kern_data_len = len;
5434         ctsio->kern_total_len = len;
5435         ctsio->kern_data_resid = 0;
5436         ctsio->kern_rel_offset = 0;
5437         ctsio->kern_sg_entries = 0;
5438         ctl_set_success(ctsio);
5439         ctsio->be_move_done = ctl_config_move_done;
5440         ctl_datamove((union ctl_io *)ctsio);
5441         return (CTL_RETVAL_COMPLETE);
5442 }
5443
5444 int
5445 ctl_write_buffer(struct ctl_scsiio *ctsio)
5446 {
5447         struct scsi_write_buffer *cdb;
5448         struct ctl_lun *lun;
5449         int buffer_offset, len;
5450
5451         CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5452
5453         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5454         cdb = (struct scsi_write_buffer *)ctsio->cdb;
5455
5456         if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA) {
5457                 ctl_set_invalid_field(ctsio,
5458                                       /*sks_valid*/ 1,
5459                                       /*command*/ 1,
5460                                       /*field*/ 1,
5461                                       /*bit_valid*/ 1,
5462                                       /*bit*/ 4);
5463                 ctl_done((union ctl_io *)ctsio);
5464                 return (CTL_RETVAL_COMPLETE);
5465         }
5466
5467         len = scsi_3btoul(cdb->length);
5468         buffer_offset = scsi_3btoul(cdb->offset);
5469
5470         if (buffer_offset + len > CTL_WRITE_BUFFER_SIZE) {
5471                 ctl_set_invalid_field(ctsio,
5472                                       /*sks_valid*/ 1,
5473                                       /*command*/ 1,
5474                                       /*field*/ 6,
5475                                       /*bit_valid*/ 0,
5476                                       /*bit*/ 0);
5477                 ctl_done((union ctl_io *)ctsio);
5478                 return (CTL_RETVAL_COMPLETE);
5479         }
5480
5481         /*
5482          * If we've got a kernel request that hasn't been malloced yet,
5483          * malloc it and tell the caller the data buffer is here.
5484          */
5485         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5486                 if (lun->write_buffer == NULL) {
5487                         lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE,
5488                             M_CTL, M_WAITOK);
5489                 }
5490                 ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5491                 ctsio->kern_data_len = len;
5492                 ctsio->kern_total_len = len;
5493                 ctsio->kern_data_resid = 0;
5494                 ctsio->kern_rel_offset = 0;
5495                 ctsio->kern_sg_entries = 0;
5496                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5497                 ctsio->be_move_done = ctl_config_move_done;
5498                 ctl_datamove((union ctl_io *)ctsio);
5499
5500                 return (CTL_RETVAL_COMPLETE);
5501         }
5502
5503         ctl_set_success(ctsio);
5504         ctl_done((union ctl_io *)ctsio);
5505         return (CTL_RETVAL_COMPLETE);
5506 }
5507
5508 int
5509 ctl_write_same(struct ctl_scsiio *ctsio)
5510 {
5511         struct ctl_lun *lun;
5512         struct ctl_lba_len_flags *lbalen;
5513         uint64_t lba;
5514         uint32_t num_blocks;
5515         int len, retval;
5516         uint8_t byte2;
5517
5518         CTL_DEBUG_PRINT(("ctl_write_same\n"));
5519
5520         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5521
5522         switch (ctsio->cdb[0]) {
5523         case WRITE_SAME_10: {
5524                 struct scsi_write_same_10 *cdb;
5525
5526                 cdb = (struct scsi_write_same_10 *)ctsio->cdb;
5527
5528                 lba = scsi_4btoul(cdb->addr);
5529                 num_blocks = scsi_2btoul(cdb->length);
5530                 byte2 = cdb->byte2;
5531                 break;
5532         }
5533         case WRITE_SAME_16: {
5534                 struct scsi_write_same_16 *cdb;
5535
5536                 cdb = (struct scsi_write_same_16 *)ctsio->cdb;
5537
5538                 lba = scsi_8btou64(cdb->addr);
5539                 num_blocks = scsi_4btoul(cdb->length);
5540                 byte2 = cdb->byte2;
5541                 break;
5542         }
5543         default:
5544                 /*
5545                  * We got a command we don't support.  This shouldn't
5546                  * happen, commands should be filtered out above us.
5547                  */
5548                 ctl_set_invalid_opcode(ctsio);
5549                 ctl_done((union ctl_io *)ctsio);
5550
5551                 return (CTL_RETVAL_COMPLETE);
5552                 break; /* NOTREACHED */
5553         }
5554
5555         /* ANCHOR flag can be used only together with UNMAP */
5556         if ((byte2 & SWS_UNMAP) == 0 && (byte2 & SWS_ANCHOR) != 0) {
5557                 ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
5558                     /*command*/ 1, /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 0);
5559                 ctl_done((union ctl_io *)ctsio);
5560                 return (CTL_RETVAL_COMPLETE);
5561         }
5562
5563         /*
5564          * The first check is to make sure we're in bounds, the second
5565          * check is to catch wrap-around problems.  If the lba + num blocks
5566          * is less than the lba, then we've wrapped around and the block
5567          * range is invalid anyway.
5568          */
5569         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5570          || ((lba + num_blocks) < lba)) {
5571                 ctl_set_lba_out_of_range(ctsio);
5572                 ctl_done((union ctl_io *)ctsio);
5573                 return (CTL_RETVAL_COMPLETE);
5574         }
5575
5576         /* Zero number of blocks means "to the last logical block" */
5577         if (num_blocks == 0) {
5578                 if ((lun->be_lun->maxlba + 1) - lba > UINT32_MAX) {
5579                         ctl_set_invalid_field(ctsio,
5580                                               /*sks_valid*/ 0,
5581                                               /*command*/ 1,
5582                                               /*field*/ 0,
5583                                               /*bit_valid*/ 0,
5584                                               /*bit*/ 0);
5585                         ctl_done((union ctl_io *)ctsio);
5586                         return (CTL_RETVAL_COMPLETE);
5587                 }
5588                 num_blocks = (lun->be_lun->maxlba + 1) - lba;
5589         }
5590
5591         len = lun->be_lun->blocksize;
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 ((byte2 & SWS_NDOB) == 0 &&
5598             (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5599                 ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
5600                 ctsio->kern_data_len = len;
5601                 ctsio->kern_total_len = len;
5602                 ctsio->kern_data_resid = 0;
5603                 ctsio->kern_rel_offset = 0;
5604                 ctsio->kern_sg_entries = 0;
5605                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5606                 ctsio->be_move_done = ctl_config_move_done;
5607                 ctl_datamove((union ctl_io *)ctsio);
5608
5609                 return (CTL_RETVAL_COMPLETE);
5610         }
5611
5612         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5613         lbalen->lba = lba;
5614         lbalen->len = num_blocks;
5615         lbalen->flags = byte2;
5616         retval = lun->backend->config_write((union ctl_io *)ctsio);
5617
5618         return (retval);
5619 }
5620
5621 int
5622 ctl_unmap(struct ctl_scsiio *ctsio)
5623 {
5624         struct ctl_lun *lun;
5625         struct scsi_unmap *cdb;
5626         struct ctl_ptr_len_flags *ptrlen;
5627         struct scsi_unmap_header *hdr;
5628         struct scsi_unmap_desc *buf, *end, *endnz, *range;
5629         uint64_t lba;
5630         uint32_t num_blocks;
5631         int len, retval;
5632         uint8_t byte2;
5633
5634         CTL_DEBUG_PRINT(("ctl_unmap\n"));
5635
5636         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5637         cdb = (struct scsi_unmap *)ctsio->cdb;
5638
5639         len = scsi_2btoul(cdb->length);
5640         byte2 = cdb->byte2;
5641
5642         /*
5643          * If we've got a kernel request that hasn't been malloced yet,
5644          * malloc it and tell the caller the data buffer is here.
5645          */
5646         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5647                 ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
5648                 ctsio->kern_data_len = len;
5649                 ctsio->kern_total_len = len;
5650                 ctsio->kern_data_resid = 0;
5651                 ctsio->kern_rel_offset = 0;
5652                 ctsio->kern_sg_entries = 0;
5653                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5654                 ctsio->be_move_done = ctl_config_move_done;
5655                 ctl_datamove((union ctl_io *)ctsio);
5656
5657                 return (CTL_RETVAL_COMPLETE);
5658         }
5659
5660         len = ctsio->kern_total_len - ctsio->kern_data_resid;
5661         hdr = (struct scsi_unmap_header *)ctsio->kern_data_ptr;
5662         if (len < sizeof (*hdr) ||
5663             len < (scsi_2btoul(hdr->length) + sizeof(hdr->length)) ||
5664             len < (scsi_2btoul(hdr->desc_length) + sizeof (*hdr)) ||
5665             scsi_2btoul(hdr->desc_length) % sizeof(*buf) != 0) {
5666                 ctl_set_invalid_field(ctsio,
5667                                       /*sks_valid*/ 0,
5668                                       /*command*/ 0,
5669                                       /*field*/ 0,
5670                                       /*bit_valid*/ 0,
5671                                       /*bit*/ 0);
5672                 goto done;
5673         }
5674         len = scsi_2btoul(hdr->desc_length);
5675         buf = (struct scsi_unmap_desc *)(hdr + 1);
5676         end = buf + len / sizeof(*buf);
5677
5678         endnz = buf;
5679         for (range = buf; range < end; range++) {
5680                 lba = scsi_8btou64(range->lba);
5681                 num_blocks = scsi_4btoul(range->length);
5682                 if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5683                  || ((lba + num_blocks) < lba)) {
5684                         ctl_set_lba_out_of_range(ctsio);
5685                         ctl_done((union ctl_io *)ctsio);
5686                         return (CTL_RETVAL_COMPLETE);
5687                 }
5688                 if (num_blocks != 0)
5689                         endnz = range + 1;
5690         }
5691
5692         /*
5693          * Block backend can not handle zero last range.
5694          * Filter it out and return if there is nothing left.
5695          */
5696         len = (uint8_t *)endnz - (uint8_t *)buf;
5697         if (len == 0) {
5698                 ctl_set_success(ctsio);
5699                 goto done;
5700         }
5701
5702         mtx_lock(&lun->lun_lock);
5703         ptrlen = (struct ctl_ptr_len_flags *)
5704             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5705         ptrlen->ptr = (void *)buf;
5706         ptrlen->len = len;
5707         ptrlen->flags = byte2;
5708         ctl_check_blocked(lun);
5709         mtx_unlock(&lun->lun_lock);
5710
5711         retval = lun->backend->config_write((union ctl_io *)ctsio);
5712         return (retval);
5713
5714 done:
5715         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5716                 free(ctsio->kern_data_ptr, M_CTL);
5717                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5718         }
5719         ctl_done((union ctl_io *)ctsio);
5720         return (CTL_RETVAL_COMPLETE);
5721 }
5722
5723 /*
5724  * Note that this function currently doesn't actually do anything inside
5725  * CTL to enforce things if the DQue bit is turned on.
5726  *
5727  * Also note that this function can't be used in the default case, because
5728  * the DQue bit isn't set in the changeable mask for the control mode page
5729  * anyway.  This is just here as an example for how to implement a page
5730  * handler, and a placeholder in case we want to allow the user to turn
5731  * tagged queueing on and off.
5732  *
5733  * The D_SENSE bit handling is functional, however, and will turn
5734  * descriptor sense on and off for a given LUN.
5735  */
5736 int
5737 ctl_control_page_handler(struct ctl_scsiio *ctsio,
5738                          struct ctl_page_index *page_index, uint8_t *page_ptr)
5739 {
5740         struct scsi_control_page *current_cp, *saved_cp, *user_cp;
5741         struct ctl_lun *lun;
5742         int set_ua;
5743         uint32_t initidx;
5744
5745         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5746         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5747         set_ua = 0;
5748
5749         user_cp = (struct scsi_control_page *)page_ptr;
5750         current_cp = (struct scsi_control_page *)
5751                 (page_index->page_data + (page_index->page_len *
5752                 CTL_PAGE_CURRENT));
5753         saved_cp = (struct scsi_control_page *)
5754                 (page_index->page_data + (page_index->page_len *
5755                 CTL_PAGE_SAVED));
5756
5757         mtx_lock(&lun->lun_lock);
5758         if (((current_cp->rlec & SCP_DSENSE) == 0)
5759          && ((user_cp->rlec & SCP_DSENSE) != 0)) {
5760                 /*
5761                  * Descriptor sense is currently turned off and the user
5762                  * wants to turn it on.
5763                  */
5764                 current_cp->rlec |= SCP_DSENSE;
5765                 saved_cp->rlec |= SCP_DSENSE;
5766                 lun->flags |= CTL_LUN_SENSE_DESC;
5767                 set_ua = 1;
5768         } else if (((current_cp->rlec & SCP_DSENSE) != 0)
5769                 && ((user_cp->rlec & SCP_DSENSE) == 0)) {
5770                 /*
5771                  * Descriptor sense is currently turned on, and the user
5772                  * wants to turn it off.
5773                  */
5774                 current_cp->rlec &= ~SCP_DSENSE;
5775                 saved_cp->rlec &= ~SCP_DSENSE;
5776                 lun->flags &= ~CTL_LUN_SENSE_DESC;
5777                 set_ua = 1;
5778         }
5779         if ((current_cp->queue_flags & SCP_QUEUE_ALG_MASK) !=
5780             (user_cp->queue_flags & SCP_QUEUE_ALG_MASK)) {
5781                 current_cp->queue_flags &= ~SCP_QUEUE_ALG_MASK;
5782                 current_cp->queue_flags |= user_cp->queue_flags & SCP_QUEUE_ALG_MASK;
5783                 saved_cp->queue_flags &= ~SCP_QUEUE_ALG_MASK;
5784                 saved_cp->queue_flags |= user_cp->queue_flags & SCP_QUEUE_ALG_MASK;
5785                 set_ua = 1;
5786         }
5787         if ((current_cp->eca_and_aen & SCP_SWP) !=
5788             (user_cp->eca_and_aen & SCP_SWP)) {
5789                 current_cp->eca_and_aen &= ~SCP_SWP;
5790                 current_cp->eca_and_aen |= user_cp->eca_and_aen & SCP_SWP;
5791                 saved_cp->eca_and_aen &= ~SCP_SWP;
5792                 saved_cp->eca_and_aen |= user_cp->eca_and_aen & SCP_SWP;
5793                 set_ua = 1;
5794         }
5795         if (set_ua != 0)
5796                 ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
5797         mtx_unlock(&lun->lun_lock);
5798         if (set_ua) {
5799                 ctl_isc_announce_mode(lun,
5800                     ctl_get_initindex(&ctsio->io_hdr.nexus),
5801                     page_index->page_code, page_index->subpage);
5802         }
5803         return (0);
5804 }
5805
5806 int
5807 ctl_caching_sp_handler(struct ctl_scsiio *ctsio,
5808                      struct ctl_page_index *page_index, uint8_t *page_ptr)
5809 {
5810         struct scsi_caching_page *current_cp, *saved_cp, *user_cp;
5811         struct ctl_lun *lun;
5812         int set_ua;
5813         uint32_t initidx;
5814
5815         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5816         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5817         set_ua = 0;
5818
5819         user_cp = (struct scsi_caching_page *)page_ptr;
5820         current_cp = (struct scsi_caching_page *)
5821                 (page_index->page_data + (page_index->page_len *
5822                 CTL_PAGE_CURRENT));
5823         saved_cp = (struct scsi_caching_page *)
5824                 (page_index->page_data + (page_index->page_len *
5825                 CTL_PAGE_SAVED));
5826
5827         mtx_lock(&lun->lun_lock);
5828         if ((current_cp->flags1 & (SCP_WCE | SCP_RCD)) !=
5829             (user_cp->flags1 & (SCP_WCE | SCP_RCD))) {
5830                 current_cp->flags1 &= ~(SCP_WCE | SCP_RCD);
5831                 current_cp->flags1 |= user_cp->flags1 & (SCP_WCE | SCP_RCD);
5832                 saved_cp->flags1 &= ~(SCP_WCE | SCP_RCD);
5833                 saved_cp->flags1 |= user_cp->flags1 & (SCP_WCE | SCP_RCD);
5834                 set_ua = 1;
5835         }
5836         if (set_ua != 0)
5837                 ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
5838         mtx_unlock(&lun->lun_lock);
5839         if (set_ua) {
5840                 ctl_isc_announce_mode(lun,
5841                     ctl_get_initindex(&ctsio->io_hdr.nexus),
5842                     page_index->page_code, page_index->subpage);
5843         }
5844         return (0);
5845 }
5846
5847 int
5848 ctl_debugconf_sp_select_handler(struct ctl_scsiio *ctsio,
5849                                 struct ctl_page_index *page_index,
5850                                 uint8_t *page_ptr)
5851 {
5852         uint8_t *c;
5853         int i;
5854
5855         c = ((struct copan_debugconf_subpage *)page_ptr)->ctl_time_io_secs;
5856         ctl_time_io_secs =
5857                 (c[0] << 8) |
5858                 (c[1] << 0) |
5859                 0;
5860         CTL_DEBUG_PRINT(("set ctl_time_io_secs to %d\n", ctl_time_io_secs));
5861         printf("set ctl_time_io_secs to %d\n", ctl_time_io_secs);
5862         printf("page data:");
5863         for (i=0; i<8; i++)
5864                 printf(" %.2x",page_ptr[i]);
5865         printf("\n");
5866         return (0);
5867 }
5868
5869 int
5870 ctl_debugconf_sp_sense_handler(struct ctl_scsiio *ctsio,
5871                                struct ctl_page_index *page_index,
5872                                int pc)
5873 {
5874         struct copan_debugconf_subpage *page;
5875
5876         page = (struct copan_debugconf_subpage *)page_index->page_data +
5877                 (page_index->page_len * pc);
5878
5879         switch (pc) {
5880         case SMS_PAGE_CTRL_CHANGEABLE >> 6:
5881         case SMS_PAGE_CTRL_DEFAULT >> 6:
5882         case SMS_PAGE_CTRL_SAVED >> 6:
5883                 /*
5884                  * We don't update the changable or default bits for this page.
5885                  */
5886                 break;
5887         case SMS_PAGE_CTRL_CURRENT >> 6:
5888                 page->ctl_time_io_secs[0] = ctl_time_io_secs >> 8;
5889                 page->ctl_time_io_secs[1] = ctl_time_io_secs >> 0;
5890                 break;
5891         default:
5892                 break;
5893         }
5894         return (0);
5895 }
5896
5897
5898 static int
5899 ctl_do_mode_select(union ctl_io *io)
5900 {
5901         struct scsi_mode_page_header *page_header;
5902         struct ctl_page_index *page_index;
5903         struct ctl_scsiio *ctsio;
5904         int control_dev, page_len;
5905         int page_len_offset, page_len_size;
5906         union ctl_modepage_info *modepage_info;
5907         struct ctl_lun *lun;
5908         int *len_left, *len_used;
5909         int retval, i;
5910
5911         ctsio = &io->scsiio;
5912         page_index = NULL;
5913         page_len = 0;
5914         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5915
5916         if (lun->be_lun->lun_type != T_DIRECT)
5917                 control_dev = 1;
5918         else
5919                 control_dev = 0;
5920
5921         modepage_info = (union ctl_modepage_info *)
5922                 ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
5923         len_left = &modepage_info->header.len_left;
5924         len_used = &modepage_info->header.len_used;
5925
5926 do_next_page:
5927
5928         page_header = (struct scsi_mode_page_header *)
5929                 (ctsio->kern_data_ptr + *len_used);
5930
5931         if (*len_left == 0) {
5932                 free(ctsio->kern_data_ptr, M_CTL);
5933                 ctl_set_success(ctsio);
5934                 ctl_done((union ctl_io *)ctsio);
5935                 return (CTL_RETVAL_COMPLETE);
5936         } else if (*len_left < sizeof(struct scsi_mode_page_header)) {
5937
5938                 free(ctsio->kern_data_ptr, M_CTL);
5939                 ctl_set_param_len_error(ctsio);
5940                 ctl_done((union ctl_io *)ctsio);
5941                 return (CTL_RETVAL_COMPLETE);
5942
5943         } else if ((page_header->page_code & SMPH_SPF)
5944                 && (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
5945
5946                 free(ctsio->kern_data_ptr, M_CTL);
5947                 ctl_set_param_len_error(ctsio);
5948                 ctl_done((union ctl_io *)ctsio);
5949                 return (CTL_RETVAL_COMPLETE);
5950         }
5951
5952
5953         /*
5954          * XXX KDM should we do something with the block descriptor?
5955          */
5956         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
5957
5958                 if ((control_dev != 0)
5959                  && (lun->mode_pages.index[i].page_flags &
5960                      CTL_PAGE_FLAG_DISK_ONLY))
5961                         continue;
5962
5963                 if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
5964                     (page_header->page_code & SMPH_PC_MASK))
5965                         continue;
5966
5967                 /*
5968                  * If neither page has a subpage code, then we've got a
5969                  * match.
5970                  */
5971                 if (((lun->mode_pages.index[i].page_code & SMPH_SPF) == 0)
5972                  && ((page_header->page_code & SMPH_SPF) == 0)) {
5973                         page_index = &lun->mode_pages.index[i];
5974                         page_len = page_header->page_length;
5975                         break;
5976                 }
5977
5978                 /*
5979                  * If both pages have subpages, then the subpage numbers
5980                  * have to match.
5981                  */
5982                 if ((lun->mode_pages.index[i].page_code & SMPH_SPF)
5983                   && (page_header->page_code & SMPH_SPF)) {
5984                         struct scsi_mode_page_header_sp *sph;
5985
5986                         sph = (struct scsi_mode_page_header_sp *)page_header;
5987
5988                         if (lun->mode_pages.index[i].subpage ==
5989                             sph->subpage) {
5990                                 page_index = &lun->mode_pages.index[i];
5991                                 page_len = scsi_2btoul(sph->page_length);
5992                                 break;
5993                         }
5994                 }
5995         }
5996
5997         /*
5998          * If we couldn't find the page, or if we don't have a mode select
5999          * handler for it, send back an error to the user.
6000          */
6001         if ((page_index == NULL)
6002          || (page_index->select_handler == NULL)) {
6003                 ctl_set_invalid_field(ctsio,
6004                                       /*sks_valid*/ 1,
6005                                       /*command*/ 0,
6006                                       /*field*/ *len_used,
6007                                       /*bit_valid*/ 0,
6008                                       /*bit*/ 0);
6009                 free(ctsio->kern_data_ptr, M_CTL);
6010                 ctl_done((union ctl_io *)ctsio);
6011                 return (CTL_RETVAL_COMPLETE);
6012         }
6013
6014         if (page_index->page_code & SMPH_SPF) {
6015                 page_len_offset = 2;
6016                 page_len_size = 2;
6017         } else {
6018                 page_len_size = 1;
6019                 page_len_offset = 1;
6020         }
6021
6022         /*
6023          * If the length the initiator gives us isn't the one we specify in
6024          * the mode page header, or if they didn't specify enough data in
6025          * the CDB to avoid truncating this page, kick out the request.
6026          */
6027         if ((page_len != (page_index->page_len - page_len_offset -
6028                           page_len_size))
6029          || (*len_left < page_index->page_len)) {
6030
6031
6032                 ctl_set_invalid_field(ctsio,
6033                                       /*sks_valid*/ 1,
6034                                       /*command*/ 0,
6035                                       /*field*/ *len_used + page_len_offset,
6036                                       /*bit_valid*/ 0,
6037                                       /*bit*/ 0);
6038                 free(ctsio->kern_data_ptr, M_CTL);
6039                 ctl_done((union ctl_io *)ctsio);
6040                 return (CTL_RETVAL_COMPLETE);
6041         }
6042
6043         /*
6044          * Run through the mode page, checking to make sure that the bits
6045          * the user changed are actually legal for him to change.
6046          */
6047         for (i = 0; i < page_index->page_len; i++) {
6048                 uint8_t *user_byte, *change_mask, *current_byte;
6049                 int bad_bit;
6050                 int j;
6051
6052                 user_byte = (uint8_t *)page_header + i;
6053                 change_mask = page_index->page_data +
6054                               (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
6055                 current_byte = page_index->page_data +
6056                                (page_index->page_len * CTL_PAGE_CURRENT) + i;
6057
6058                 /*
6059                  * Check to see whether the user set any bits in this byte
6060                  * that he is not allowed to set.
6061                  */
6062                 if ((*user_byte & ~(*change_mask)) ==
6063                     (*current_byte & ~(*change_mask)))
6064                         continue;
6065
6066                 /*
6067                  * Go through bit by bit to determine which one is illegal.
6068                  */
6069                 bad_bit = 0;
6070                 for (j = 7; j >= 0; j--) {
6071                         if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
6072                             (((1 << i) & ~(*change_mask)) & *current_byte)) {
6073                                 bad_bit = i;
6074                                 break;
6075                         }
6076                 }
6077                 ctl_set_invalid_field(ctsio,
6078                                       /*sks_valid*/ 1,
6079                                       /*command*/ 0,
6080                                       /*field*/ *len_used + i,
6081                                       /*bit_valid*/ 1,
6082                                       /*bit*/ bad_bit);
6083                 free(ctsio->kern_data_ptr, M_CTL);
6084                 ctl_done((union ctl_io *)ctsio);
6085                 return (CTL_RETVAL_COMPLETE);
6086         }
6087
6088         /*
6089          * Decrement these before we call the page handler, since we may
6090          * end up getting called back one way or another before the handler
6091          * returns to this context.
6092          */
6093         *len_left -= page_index->page_len;
6094         *len_used += page_index->page_len;
6095
6096         retval = page_index->select_handler(ctsio, page_index,
6097                                             (uint8_t *)page_header);
6098
6099         /*
6100          * If the page handler returns CTL_RETVAL_QUEUED, then we need to
6101          * wait until this queued command completes to finish processing
6102          * the mode page.  If it returns anything other than
6103          * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
6104          * already set the sense information, freed the data pointer, and
6105          * completed the io for us.
6106          */
6107         if (retval != CTL_RETVAL_COMPLETE)
6108                 goto bailout_no_done;
6109
6110         /*
6111          * If the initiator sent us more than one page, parse the next one.
6112          */
6113         if (*len_left > 0)
6114                 goto do_next_page;
6115
6116         ctl_set_success(ctsio);
6117         free(ctsio->kern_data_ptr, M_CTL);
6118         ctl_done((union ctl_io *)ctsio);
6119
6120 bailout_no_done:
6121
6122         return (CTL_RETVAL_COMPLETE);
6123
6124 }
6125
6126 int
6127 ctl_mode_select(struct ctl_scsiio *ctsio)
6128 {
6129         int param_len, pf, sp;
6130         int header_size, bd_len;
6131         union ctl_modepage_info *modepage_info;
6132
6133         switch (ctsio->cdb[0]) {
6134         case MODE_SELECT_6: {
6135                 struct scsi_mode_select_6 *cdb;
6136
6137                 cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
6138
6139                 pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6140                 sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6141                 param_len = cdb->length;
6142                 header_size = sizeof(struct scsi_mode_header_6);
6143                 break;
6144         }
6145         case MODE_SELECT_10: {
6146                 struct scsi_mode_select_10 *cdb;
6147
6148                 cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
6149
6150                 pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6151                 sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6152                 param_len = scsi_2btoul(cdb->length);
6153                 header_size = sizeof(struct scsi_mode_header_10);
6154                 break;
6155         }
6156         default:
6157                 ctl_set_invalid_opcode(ctsio);
6158                 ctl_done((union ctl_io *)ctsio);
6159                 return (CTL_RETVAL_COMPLETE);
6160         }
6161
6162         /*
6163          * From SPC-3:
6164          * "A parameter list length of zero indicates that the Data-Out Buffer
6165          * shall be empty. This condition shall not be considered as an error."
6166          */
6167         if (param_len == 0) {
6168                 ctl_set_success(ctsio);
6169                 ctl_done((union ctl_io *)ctsio);
6170                 return (CTL_RETVAL_COMPLETE);
6171         }
6172
6173         /*
6174          * Since we'll hit this the first time through, prior to
6175          * allocation, we don't need to free a data buffer here.
6176          */
6177         if (param_len < header_size) {
6178                 ctl_set_param_len_error(ctsio);
6179                 ctl_done((union ctl_io *)ctsio);
6180                 return (CTL_RETVAL_COMPLETE);
6181         }
6182
6183         /*
6184          * Allocate the data buffer and grab the user's data.  In theory,
6185          * we shouldn't have to sanity check the parameter list length here
6186          * because the maximum size is 64K.  We should be able to malloc
6187          * that much without too many problems.
6188          */
6189         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6190                 ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6191                 ctsio->kern_data_len = param_len;
6192                 ctsio->kern_total_len = param_len;
6193                 ctsio->kern_data_resid = 0;
6194                 ctsio->kern_rel_offset = 0;
6195                 ctsio->kern_sg_entries = 0;
6196                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6197                 ctsio->be_move_done = ctl_config_move_done;
6198                 ctl_datamove((union ctl_io *)ctsio);
6199
6200                 return (CTL_RETVAL_COMPLETE);
6201         }
6202
6203         switch (ctsio->cdb[0]) {
6204         case MODE_SELECT_6: {
6205                 struct scsi_mode_header_6 *mh6;
6206
6207                 mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6208                 bd_len = mh6->blk_desc_len;
6209                 break;
6210         }
6211         case MODE_SELECT_10: {
6212                 struct scsi_mode_header_10 *mh10;
6213
6214                 mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6215                 bd_len = scsi_2btoul(mh10->blk_desc_len);
6216                 break;
6217         }
6218         default:
6219                 panic("Invalid CDB type %#x", ctsio->cdb[0]);
6220                 break;
6221         }
6222
6223         if (param_len < (header_size + bd_len)) {
6224                 free(ctsio->kern_data_ptr, M_CTL);
6225                 ctl_set_param_len_error(ctsio);
6226                 ctl_done((union ctl_io *)ctsio);
6227                 return (CTL_RETVAL_COMPLETE);
6228         }
6229
6230         /*
6231          * Set the IO_CONT flag, so that if this I/O gets passed to
6232          * ctl_config_write_done(), it'll get passed back to
6233          * ctl_do_mode_select() for further processing, or completion if
6234          * we're all done.
6235          */
6236         ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6237         ctsio->io_cont = ctl_do_mode_select;
6238
6239         modepage_info = (union ctl_modepage_info *)
6240                 ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6241         memset(modepage_info, 0, sizeof(*modepage_info));
6242         modepage_info->header.len_left = param_len - header_size - bd_len;
6243         modepage_info->header.len_used = header_size + bd_len;
6244
6245         return (ctl_do_mode_select((union ctl_io *)ctsio));
6246 }
6247
6248 int
6249 ctl_mode_sense(struct ctl_scsiio *ctsio)
6250 {
6251         struct ctl_lun *lun;
6252         int pc, page_code, dbd, llba, subpage;
6253         int alloc_len, page_len, header_len, total_len;
6254         struct scsi_mode_block_descr *block_desc;
6255         struct ctl_page_index *page_index;
6256         int control_dev;
6257
6258         dbd = 0;
6259         llba = 0;
6260         block_desc = NULL;
6261         page_index = NULL;
6262
6263         CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6264
6265         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6266
6267         if (lun->be_lun->lun_type != T_DIRECT)
6268                 control_dev = 1;
6269         else
6270                 control_dev = 0;
6271
6272         switch (ctsio->cdb[0]) {
6273         case MODE_SENSE_6: {
6274                 struct scsi_mode_sense_6 *cdb;
6275
6276                 cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6277
6278                 header_len = sizeof(struct scsi_mode_hdr_6);
6279                 if (cdb->byte2 & SMS_DBD)
6280                         dbd = 1;
6281                 else
6282                         header_len += sizeof(struct scsi_mode_block_descr);
6283
6284                 pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6285                 page_code = cdb->page & SMS_PAGE_CODE;
6286                 subpage = cdb->subpage;
6287                 alloc_len = cdb->length;
6288                 break;
6289         }
6290         case MODE_SENSE_10: {
6291                 struct scsi_mode_sense_10 *cdb;
6292
6293                 cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6294
6295                 header_len = sizeof(struct scsi_mode_hdr_10);
6296
6297                 if (cdb->byte2 & SMS_DBD)
6298                         dbd = 1;
6299                 else
6300                         header_len += sizeof(struct scsi_mode_block_descr);
6301                 if (cdb->byte2 & SMS10_LLBAA)
6302                         llba = 1;
6303                 pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6304                 page_code = cdb->page & SMS_PAGE_CODE;
6305                 subpage = cdb->subpage;
6306                 alloc_len = scsi_2btoul(cdb->length);
6307                 break;
6308         }
6309         default:
6310                 ctl_set_invalid_opcode(ctsio);
6311                 ctl_done((union ctl_io *)ctsio);
6312                 return (CTL_RETVAL_COMPLETE);
6313                 break; /* NOTREACHED */
6314         }
6315
6316         /*
6317          * We have to make a first pass through to calculate the size of
6318          * the pages that match the user's query.  Then we allocate enough
6319          * memory to hold it, and actually copy the data into the buffer.
6320          */
6321         switch (page_code) {
6322         case SMS_ALL_PAGES_PAGE: {
6323                 int i;
6324
6325                 page_len = 0;
6326
6327                 /*
6328                  * At the moment, values other than 0 and 0xff here are
6329                  * reserved according to SPC-3.
6330                  */
6331                 if ((subpage != SMS_SUBPAGE_PAGE_0)
6332                  && (subpage != SMS_SUBPAGE_ALL)) {
6333                         ctl_set_invalid_field(ctsio,
6334                                               /*sks_valid*/ 1,
6335                                               /*command*/ 1,
6336                                               /*field*/ 3,
6337                                               /*bit_valid*/ 0,
6338                                               /*bit*/ 0);
6339                         ctl_done((union ctl_io *)ctsio);
6340                         return (CTL_RETVAL_COMPLETE);
6341                 }
6342
6343                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6344                         if ((control_dev != 0)
6345                          && (lun->mode_pages.index[i].page_flags &
6346                              CTL_PAGE_FLAG_DISK_ONLY))
6347                                 continue;
6348
6349                         /*
6350                          * We don't use this subpage if the user didn't
6351                          * request all subpages.
6352                          */
6353                         if ((lun->mode_pages.index[i].subpage != 0)
6354                          && (subpage == SMS_SUBPAGE_PAGE_0))
6355                                 continue;
6356
6357 #if 0
6358                         printf("found page %#x len %d\n",
6359                                lun->mode_pages.index[i].page_code &
6360                                SMPH_PC_MASK,
6361                                lun->mode_pages.index[i].page_len);
6362 #endif
6363                         page_len += lun->mode_pages.index[i].page_len;
6364                 }
6365                 break;
6366         }
6367         default: {
6368                 int i;
6369
6370                 page_len = 0;
6371
6372                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6373                         /* Look for the right page code */
6374                         if ((lun->mode_pages.index[i].page_code &
6375                              SMPH_PC_MASK) != page_code)
6376                                 continue;
6377
6378                         /* Look for the right subpage or the subpage wildcard*/
6379                         if ((lun->mode_pages.index[i].subpage != subpage)
6380                          && (subpage != SMS_SUBPAGE_ALL))
6381                                 continue;
6382
6383                         /* Make sure the page is supported for this dev type */
6384                         if ((control_dev != 0)
6385                          && (lun->mode_pages.index[i].page_flags &
6386                              CTL_PAGE_FLAG_DISK_ONLY))
6387                                 continue;
6388
6389 #if 0
6390                         printf("found page %#x len %d\n",
6391                                lun->mode_pages.index[i].page_code &
6392                                SMPH_PC_MASK,
6393                                lun->mode_pages.index[i].page_len);
6394 #endif
6395
6396                         page_len += lun->mode_pages.index[i].page_len;
6397                 }
6398
6399                 if (page_len == 0) {
6400                         ctl_set_invalid_field(ctsio,
6401                                               /*sks_valid*/ 1,
6402                                               /*command*/ 1,
6403                                               /*field*/ 2,
6404                                               /*bit_valid*/ 1,
6405                                               /*bit*/ 5);
6406                         ctl_done((union ctl_io *)ctsio);
6407                         return (CTL_RETVAL_COMPLETE);
6408                 }
6409                 break;
6410         }
6411         }
6412
6413         total_len = header_len + page_len;
6414 #if 0
6415         printf("header_len = %d, page_len = %d, total_len = %d\n",
6416                header_len, page_len, total_len);
6417 #endif
6418
6419         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6420         ctsio->kern_sg_entries = 0;
6421         ctsio->kern_data_resid = 0;
6422         ctsio->kern_rel_offset = 0;
6423         if (total_len < alloc_len) {
6424                 ctsio->residual = alloc_len - total_len;
6425                 ctsio->kern_data_len = total_len;
6426                 ctsio->kern_total_len = total_len;
6427         } else {
6428                 ctsio->residual = 0;
6429                 ctsio->kern_data_len = alloc_len;
6430                 ctsio->kern_total_len = alloc_len;
6431         }
6432
6433         switch (ctsio->cdb[0]) {
6434         case MODE_SENSE_6: {
6435                 struct scsi_mode_hdr_6 *header;
6436
6437                 header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
6438
6439                 header->datalen = MIN(total_len - 1, 254);
6440                 if (control_dev == 0) {
6441                         header->dev_specific = 0x10; /* DPOFUA */
6442                         if ((lun->be_lun->flags & CTL_LUN_FLAG_READONLY) ||
6443                             (lun->mode_pages.control_page[CTL_PAGE_CURRENT]
6444                             .eca_and_aen & SCP_SWP) != 0)
6445                                     header->dev_specific |= 0x80; /* WP */
6446                 }
6447                 if (dbd)
6448                         header->block_descr_len = 0;
6449                 else
6450                         header->block_descr_len =
6451                                 sizeof(struct scsi_mode_block_descr);
6452                 block_desc = (struct scsi_mode_block_descr *)&header[1];
6453                 break;
6454         }
6455         case MODE_SENSE_10: {
6456                 struct scsi_mode_hdr_10 *header;
6457                 int datalen;
6458
6459                 header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
6460
6461                 datalen = MIN(total_len - 2, 65533);
6462                 scsi_ulto2b(datalen, header->datalen);
6463                 if (control_dev == 0) {
6464                         header->dev_specific = 0x10; /* DPOFUA */
6465                         if ((lun->be_lun->flags & CTL_LUN_FLAG_READONLY) ||
6466                             (lun->mode_pages.control_page[CTL_PAGE_CURRENT]
6467                             .eca_and_aen & SCP_SWP) != 0)
6468                                     header->dev_specific |= 0x80; /* WP */
6469                 }
6470                 if (dbd)
6471                         scsi_ulto2b(0, header->block_descr_len);
6472                 else
6473                         scsi_ulto2b(sizeof(struct scsi_mode_block_descr),
6474                                     header->block_descr_len);
6475                 block_desc = (struct scsi_mode_block_descr *)&header[1];
6476                 break;
6477         }
6478         default:
6479                 panic("invalid CDB type %#x", ctsio->cdb[0]);
6480                 break; /* NOTREACHED */
6481         }
6482
6483         /*
6484          * If we've got a disk, use its blocksize in the block
6485          * descriptor.  Otherwise, just set it to 0.
6486          */
6487         if (dbd == 0) {
6488                 if (control_dev == 0)
6489                         scsi_ulto3b(lun->be_lun->blocksize,
6490                                     block_desc->block_len);
6491                 else
6492                         scsi_ulto3b(0, block_desc->block_len);
6493         }
6494
6495         switch (page_code) {
6496         case SMS_ALL_PAGES_PAGE: {
6497                 int i, data_used;
6498
6499                 data_used = header_len;
6500                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6501                         struct ctl_page_index *page_index;
6502
6503                         page_index = &lun->mode_pages.index[i];
6504
6505                         if ((control_dev != 0)
6506                          && (page_index->page_flags &
6507                             CTL_PAGE_FLAG_DISK_ONLY))
6508                                 continue;
6509
6510                         /*
6511                          * We don't use this subpage if the user didn't
6512                          * request all subpages.  We already checked (above)
6513                          * to make sure the user only specified a subpage
6514                          * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case.
6515                          */
6516                         if ((page_index->subpage != 0)
6517                          && (subpage == SMS_SUBPAGE_PAGE_0))
6518                                 continue;
6519
6520                         /*
6521                          * Call the handler, if it exists, to update the
6522                          * page to the latest values.
6523                          */
6524                         if (page_index->sense_handler != NULL)
6525                                 page_index->sense_handler(ctsio, page_index,pc);
6526
6527                         memcpy(ctsio->kern_data_ptr + data_used,
6528                                page_index->page_data +
6529                                (page_index->page_len * pc),
6530                                page_index->page_len);
6531                         data_used += page_index->page_len;
6532                 }
6533                 break;
6534         }
6535         default: {
6536                 int i, data_used;
6537
6538                 data_used = header_len;
6539
6540                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6541                         struct ctl_page_index *page_index;
6542
6543                         page_index = &lun->mode_pages.index[i];
6544
6545                         /* Look for the right page code */
6546                         if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6547                                 continue;
6548
6549                         /* Look for the right subpage or the subpage wildcard*/
6550                         if ((page_index->subpage != subpage)
6551                          && (subpage != SMS_SUBPAGE_ALL))
6552                                 continue;
6553
6554                         /* Make sure the page is supported for this dev type */
6555                         if ((control_dev != 0)
6556                          && (page_index->page_flags &
6557                              CTL_PAGE_FLAG_DISK_ONLY))
6558                                 continue;
6559
6560                         /*
6561                          * Call the handler, if it exists, to update the
6562                          * page to the latest values.
6563                          */
6564                         if (page_index->sense_handler != NULL)
6565                                 page_index->sense_handler(ctsio, page_index,pc);
6566
6567                         memcpy(ctsio->kern_data_ptr + data_used,
6568                                page_index->page_data +
6569                                (page_index->page_len * pc),
6570                                page_index->page_len);
6571                         data_used += page_index->page_len;
6572                 }
6573                 break;
6574         }
6575         }
6576
6577         ctl_set_success(ctsio);
6578         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6579         ctsio->be_move_done = ctl_config_move_done;
6580         ctl_datamove((union ctl_io *)ctsio);
6581         return (CTL_RETVAL_COMPLETE);
6582 }
6583
6584 int
6585 ctl_lbp_log_sense_handler(struct ctl_scsiio *ctsio,
6586                                struct ctl_page_index *page_index,
6587                                int pc)
6588 {
6589         struct ctl_lun *lun;
6590         struct scsi_log_param_header *phdr;
6591         uint8_t *data;
6592         uint64_t val;
6593
6594         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6595         data = page_index->page_data;
6596
6597         if (lun->backend->lun_attr != NULL &&
6598             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "blocksavail"))
6599              != UINT64_MAX) {
6600                 phdr = (struct scsi_log_param_header *)data;
6601                 scsi_ulto2b(0x0001, phdr->param_code);
6602                 phdr->param_control = SLP_LBIN | SLP_LP;
6603                 phdr->param_len = 8;
6604                 data = (uint8_t *)(phdr + 1);
6605                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6606                 data[4] = 0x02; /* per-pool */
6607                 data += phdr->param_len;
6608         }
6609
6610         if (lun->backend->lun_attr != NULL &&
6611             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "blocksused"))
6612              != UINT64_MAX) {
6613                 phdr = (struct scsi_log_param_header *)data;
6614                 scsi_ulto2b(0x0002, phdr->param_code);
6615                 phdr->param_control = SLP_LBIN | SLP_LP;
6616                 phdr->param_len = 8;
6617                 data = (uint8_t *)(phdr + 1);
6618                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6619                 data[4] = 0x01; /* per-LUN */
6620                 data += phdr->param_len;
6621         }
6622
6623         if (lun->backend->lun_attr != NULL &&
6624             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "poolblocksavail"))
6625              != UINT64_MAX) {
6626                 phdr = (struct scsi_log_param_header *)data;
6627                 scsi_ulto2b(0x00f1, phdr->param_code);
6628                 phdr->param_control = SLP_LBIN | SLP_LP;
6629                 phdr->param_len = 8;
6630                 data = (uint8_t *)(phdr + 1);
6631                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6632                 data[4] = 0x02; /* per-pool */
6633                 data += phdr->param_len;
6634         }
6635
6636         if (lun->backend->lun_attr != NULL &&
6637             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "poolblocksused"))
6638              != UINT64_MAX) {
6639                 phdr = (struct scsi_log_param_header *)data;
6640                 scsi_ulto2b(0x00f2, phdr->param_code);
6641                 phdr->param_control = SLP_LBIN | SLP_LP;
6642                 phdr->param_len = 8;
6643                 data = (uint8_t *)(phdr + 1);
6644                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6645                 data[4] = 0x02; /* per-pool */
6646                 data += phdr->param_len;
6647         }
6648
6649         page_index->page_len = data - page_index->page_data;
6650         return (0);
6651 }
6652
6653 int
6654 ctl_sap_log_sense_handler(struct ctl_scsiio *ctsio,
6655                                struct ctl_page_index *page_index,
6656                                int pc)
6657 {
6658         struct ctl_lun *lun;
6659         struct stat_page *data;
6660         uint64_t rn, wn, rb, wb;
6661         struct bintime rt, wt;
6662         int i;
6663
6664         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6665         data = (struct stat_page *)page_index->page_data;
6666
6667         scsi_ulto2b(SLP_SAP, data->sap.hdr.param_code);
6668         data->sap.hdr.param_control = SLP_LBIN;
6669         data->sap.hdr.param_len = sizeof(struct scsi_log_stat_and_perf) -
6670             sizeof(struct scsi_log_param_header);
6671         rn = wn = rb = wb = 0;
6672         bintime_clear(&rt);
6673         bintime_clear(&wt);
6674         for (i = 0; i < CTL_MAX_PORTS; i++) {
6675                 rn += lun->stats.ports[i].operations[CTL_STATS_READ];
6676                 wn += lun->stats.ports[i].operations[CTL_STATS_WRITE];
6677                 rb += lun->stats.ports[i].bytes[CTL_STATS_READ];
6678                 wb += lun->stats.ports[i].bytes[CTL_STATS_WRITE];
6679                 bintime_add(&rt, &lun->stats.ports[i].time[CTL_STATS_READ]);
6680                 bintime_add(&wt, &lun->stats.ports[i].time[CTL_STATS_WRITE]);
6681         }
6682         scsi_u64to8b(rn, data->sap.read_num);
6683         scsi_u64to8b(wn, data->sap.write_num);
6684         if (lun->stats.blocksize > 0) {
6685                 scsi_u64to8b(wb / lun->stats.blocksize,
6686                     data->sap.recvieved_lba);
6687                 scsi_u64to8b(rb / lun->stats.blocksize,
6688                     data->sap.transmitted_lba);
6689         }
6690         scsi_u64to8b((uint64_t)rt.sec * 1000 + rt.frac / (UINT64_MAX / 1000),
6691             data->sap.read_int);
6692         scsi_u64to8b((uint64_t)wt.sec * 1000 + wt.frac / (UINT64_MAX / 1000),
6693             data->sap.write_int);
6694         scsi_u64to8b(0, data->sap.weighted_num);
6695         scsi_u64to8b(0, data->sap.weighted_int);
6696         scsi_ulto2b(SLP_IT, data->it.hdr.param_code);
6697         data->it.hdr.param_control = SLP_LBIN;
6698         data->it.hdr.param_len = sizeof(struct scsi_log_idle_time) -
6699             sizeof(struct scsi_log_param_header);
6700 #ifdef CTL_TIME_IO
6701         scsi_u64to8b(lun->idle_time / SBT_1MS, data->it.idle_int);
6702 #endif
6703         scsi_ulto2b(SLP_TI, data->ti.hdr.param_code);
6704         data->it.hdr.param_control = SLP_LBIN;
6705         data->ti.hdr.param_len = sizeof(struct scsi_log_time_interval) -
6706             sizeof(struct scsi_log_param_header);
6707         scsi_ulto4b(3, data->ti.exponent);
6708         scsi_ulto4b(1, data->ti.integer);
6709
6710         page_index->page_len = sizeof(*data);
6711         return (0);
6712 }
6713
6714 int
6715 ctl_log_sense(struct ctl_scsiio *ctsio)
6716 {
6717         struct ctl_lun *lun;
6718         int i, pc, page_code, subpage;
6719         int alloc_len, total_len;
6720         struct ctl_page_index *page_index;
6721         struct scsi_log_sense *cdb;
6722         struct scsi_log_header *header;
6723
6724         CTL_DEBUG_PRINT(("ctl_log_sense\n"));
6725
6726         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6727         cdb = (struct scsi_log_sense *)ctsio->cdb;
6728         pc = (cdb->page & SLS_PAGE_CTRL_MASK) >> 6;
6729         page_code = cdb->page & SLS_PAGE_CODE;
6730         subpage = cdb->subpage;
6731         alloc_len = scsi_2btoul(cdb->length);
6732
6733         page_index = NULL;
6734         for (i = 0; i < CTL_NUM_LOG_PAGES; i++) {
6735                 page_index = &lun->log_pages.index[i];
6736
6737                 /* Look for the right page code */
6738                 if ((page_index->page_code & SL_PAGE_CODE) != page_code)
6739                         continue;
6740
6741                 /* Look for the right subpage or the subpage wildcard*/
6742                 if (page_index->subpage != subpage)
6743                         continue;
6744
6745                 break;
6746         }
6747         if (i >= CTL_NUM_LOG_PAGES) {
6748                 ctl_set_invalid_field(ctsio,
6749                                       /*sks_valid*/ 1,
6750                                       /*command*/ 1,
6751                                       /*field*/ 2,
6752                                       /*bit_valid*/ 0,
6753                                       /*bit*/ 0);
6754                 ctl_done((union ctl_io *)ctsio);
6755                 return (CTL_RETVAL_COMPLETE);
6756         }
6757
6758         total_len = sizeof(struct scsi_log_header) + page_index->page_len;
6759
6760         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6761         ctsio->kern_sg_entries = 0;
6762         ctsio->kern_data_resid = 0;
6763         ctsio->kern_rel_offset = 0;
6764         if (total_len < alloc_len) {
6765                 ctsio->residual = alloc_len - total_len;
6766                 ctsio->kern_data_len = total_len;
6767                 ctsio->kern_total_len = total_len;
6768         } else {
6769                 ctsio->residual = 0;
6770                 ctsio->kern_data_len = alloc_len;
6771                 ctsio->kern_total_len = alloc_len;
6772         }
6773
6774         header = (struct scsi_log_header *)ctsio->kern_data_ptr;
6775         header->page = page_index->page_code;
6776         if (page_index->subpage) {
6777                 header->page |= SL_SPF;
6778                 header->subpage = page_index->subpage;
6779         }
6780         scsi_ulto2b(page_index->page_len, header->datalen);
6781
6782         /*
6783          * Call the handler, if it exists, to update the
6784          * page to the latest values.
6785          */
6786         if (page_index->sense_handler != NULL)
6787                 page_index->sense_handler(ctsio, page_index, pc);
6788
6789         memcpy(header + 1, page_index->page_data, page_index->page_len);
6790
6791         ctl_set_success(ctsio);
6792         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6793         ctsio->be_move_done = ctl_config_move_done;
6794         ctl_datamove((union ctl_io *)ctsio);
6795         return (CTL_RETVAL_COMPLETE);
6796 }
6797
6798 int
6799 ctl_read_capacity(struct ctl_scsiio *ctsio)
6800 {
6801         struct scsi_read_capacity *cdb;
6802         struct scsi_read_capacity_data *data;
6803         struct ctl_lun *lun;
6804         uint32_t lba;
6805
6806         CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
6807
6808         cdb = (struct scsi_read_capacity *)ctsio->cdb;
6809
6810         lba = scsi_4btoul(cdb->addr);
6811         if (((cdb->pmi & SRC_PMI) == 0)
6812          && (lba != 0)) {
6813                 ctl_set_invalid_field(/*ctsio*/ ctsio,
6814                                       /*sks_valid*/ 1,
6815                                       /*command*/ 1,
6816                                       /*field*/ 2,
6817                                       /*bit_valid*/ 0,
6818                                       /*bit*/ 0);
6819                 ctl_done((union ctl_io *)ctsio);
6820                 return (CTL_RETVAL_COMPLETE);
6821         }
6822
6823         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6824
6825         ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
6826         data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
6827         ctsio->residual = 0;
6828         ctsio->kern_data_len = sizeof(*data);
6829         ctsio->kern_total_len = sizeof(*data);
6830         ctsio->kern_data_resid = 0;
6831         ctsio->kern_rel_offset = 0;
6832         ctsio->kern_sg_entries = 0;
6833
6834         /*
6835          * If the maximum LBA is greater than 0xfffffffe, the user must
6836          * issue a SERVICE ACTION IN (16) command, with the read capacity
6837          * serivce action set.
6838          */
6839         if (lun->be_lun->maxlba > 0xfffffffe)
6840                 scsi_ulto4b(0xffffffff, data->addr);
6841         else
6842                 scsi_ulto4b(lun->be_lun->maxlba, data->addr);
6843
6844         /*
6845          * XXX KDM this may not be 512 bytes...
6846          */
6847         scsi_ulto4b(lun->be_lun->blocksize, data->length);
6848
6849         ctl_set_success(ctsio);
6850         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6851         ctsio->be_move_done = ctl_config_move_done;
6852         ctl_datamove((union ctl_io *)ctsio);
6853         return (CTL_RETVAL_COMPLETE);
6854 }
6855
6856 int
6857 ctl_read_capacity_16(struct ctl_scsiio *ctsio)
6858 {
6859         struct scsi_read_capacity_16 *cdb;
6860         struct scsi_read_capacity_data_long *data;
6861         struct ctl_lun *lun;
6862         uint64_t lba;
6863         uint32_t alloc_len;
6864
6865         CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
6866
6867         cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
6868
6869         alloc_len = scsi_4btoul(cdb->alloc_len);
6870         lba = scsi_8btou64(cdb->addr);
6871
6872         if ((cdb->reladr & SRC16_PMI)
6873          && (lba != 0)) {
6874                 ctl_set_invalid_field(/*ctsio*/ ctsio,
6875                                       /*sks_valid*/ 1,
6876                                       /*command*/ 1,
6877                                       /*field*/ 2,
6878                                       /*bit_valid*/ 0,
6879                                       /*bit*/ 0);
6880                 ctl_done((union ctl_io *)ctsio);
6881                 return (CTL_RETVAL_COMPLETE);
6882         }
6883
6884         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6885
6886         ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
6887         data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
6888
6889         if (sizeof(*data) < alloc_len) {
6890                 ctsio->residual = alloc_len - sizeof(*data);
6891                 ctsio->kern_data_len = sizeof(*data);
6892                 ctsio->kern_total_len = sizeof(*data);
6893         } else {
6894                 ctsio->residual = 0;
6895                 ctsio->kern_data_len = alloc_len;
6896                 ctsio->kern_total_len = alloc_len;
6897         }
6898         ctsio->kern_data_resid = 0;
6899         ctsio->kern_rel_offset = 0;
6900         ctsio->kern_sg_entries = 0;
6901
6902         scsi_u64to8b(lun->be_lun->maxlba, data->addr);
6903         /* XXX KDM this may not be 512 bytes... */
6904         scsi_ulto4b(lun->be_lun->blocksize, data->length);
6905         data->prot_lbppbe = lun->be_lun->pblockexp & SRC16_LBPPBE;
6906         scsi_ulto2b(lun->be_lun->pblockoff & SRC16_LALBA_A, data->lalba_lbp);
6907         if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
6908                 data->lalba_lbp[0] |= SRC16_LBPME | SRC16_LBPRZ;
6909
6910         ctl_set_success(ctsio);
6911         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6912         ctsio->be_move_done = ctl_config_move_done;
6913         ctl_datamove((union ctl_io *)ctsio);
6914         return (CTL_RETVAL_COMPLETE);
6915 }
6916
6917 int
6918 ctl_get_lba_status(struct ctl_scsiio *ctsio)
6919 {
6920         struct scsi_get_lba_status *cdb;
6921         struct scsi_get_lba_status_data *data;
6922         struct ctl_lun *lun;
6923         struct ctl_lba_len_flags *lbalen;
6924         uint64_t lba;
6925         uint32_t alloc_len, total_len;
6926         int retval;
6927
6928         CTL_DEBUG_PRINT(("ctl_get_lba_status\n"));
6929
6930         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6931         cdb = (struct scsi_get_lba_status *)ctsio->cdb;
6932         lba = scsi_8btou64(cdb->addr);
6933         alloc_len = scsi_4btoul(cdb->alloc_len);
6934
6935         if (lba > lun->be_lun->maxlba) {
6936                 ctl_set_lba_out_of_range(ctsio);
6937                 ctl_done((union ctl_io *)ctsio);
6938                 return (CTL_RETVAL_COMPLETE);
6939         }
6940
6941         total_len = sizeof(*data) + sizeof(data->descr[0]);
6942         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6943         data = (struct scsi_get_lba_status_data *)ctsio->kern_data_ptr;
6944
6945         if (total_len < alloc_len) {
6946                 ctsio->residual = alloc_len - total_len;
6947                 ctsio->kern_data_len = total_len;
6948                 ctsio->kern_total_len = total_len;
6949         } else {
6950                 ctsio->residual = 0;
6951                 ctsio->kern_data_len = alloc_len;
6952                 ctsio->kern_total_len = alloc_len;
6953         }
6954         ctsio->kern_data_resid = 0;
6955         ctsio->kern_rel_offset = 0;
6956         ctsio->kern_sg_entries = 0;
6957
6958         /* Fill dummy data in case backend can't tell anything. */
6959         scsi_ulto4b(4 + sizeof(data->descr[0]), data->length);
6960         scsi_u64to8b(lba, data->descr[0].addr);
6961         scsi_ulto4b(MIN(UINT32_MAX, lun->be_lun->maxlba + 1 - lba),
6962             data->descr[0].length);
6963         data->descr[0].status = 0; /* Mapped or unknown. */
6964
6965         ctl_set_success(ctsio);
6966         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6967         ctsio->be_move_done = ctl_config_move_done;
6968
6969         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
6970         lbalen->lba = lba;
6971         lbalen->len = total_len;
6972         lbalen->flags = 0;
6973         retval = lun->backend->config_read((union ctl_io *)ctsio);
6974         return (CTL_RETVAL_COMPLETE);
6975 }
6976
6977 int
6978 ctl_read_defect(struct ctl_scsiio *ctsio)
6979 {
6980         struct scsi_read_defect_data_10 *ccb10;
6981         struct scsi_read_defect_data_12 *ccb12;
6982         struct scsi_read_defect_data_hdr_10 *data10;
6983         struct scsi_read_defect_data_hdr_12 *data12;
6984         uint32_t alloc_len, data_len;
6985         uint8_t format;
6986
6987         CTL_DEBUG_PRINT(("ctl_read_defect\n"));
6988
6989         if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
6990                 ccb10 = (struct scsi_read_defect_data_10 *)&ctsio->cdb;
6991                 format = ccb10->format;
6992                 alloc_len = scsi_2btoul(ccb10->alloc_length);
6993                 data_len = sizeof(*data10);
6994         } else {
6995                 ccb12 = (struct scsi_read_defect_data_12 *)&ctsio->cdb;
6996                 format = ccb12->format;
6997                 alloc_len = scsi_4btoul(ccb12->alloc_length);
6998                 data_len = sizeof(*data12);
6999         }
7000         if (alloc_len == 0) {
7001                 ctl_set_success(ctsio);
7002                 ctl_done((union ctl_io *)ctsio);
7003                 return (CTL_RETVAL_COMPLETE);
7004         }
7005
7006         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
7007         if (data_len < alloc_len) {
7008                 ctsio->residual = alloc_len - data_len;
7009                 ctsio->kern_data_len = data_len;
7010                 ctsio->kern_total_len = data_len;
7011         } else {
7012                 ctsio->residual = 0;
7013                 ctsio->kern_data_len = alloc_len;
7014                 ctsio->kern_total_len = alloc_len;
7015         }
7016         ctsio->kern_data_resid = 0;
7017         ctsio->kern_rel_offset = 0;
7018         ctsio->kern_sg_entries = 0;
7019
7020         if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
7021                 data10 = (struct scsi_read_defect_data_hdr_10 *)
7022                     ctsio->kern_data_ptr;
7023                 data10->format = format;
7024                 scsi_ulto2b(0, data10->length);
7025         } else {
7026                 data12 = (struct scsi_read_defect_data_hdr_12 *)
7027                     ctsio->kern_data_ptr;
7028                 data12->format = format;
7029                 scsi_ulto2b(0, data12->generation);
7030                 scsi_ulto4b(0, data12->length);
7031         }
7032
7033         ctl_set_success(ctsio);
7034         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7035         ctsio->be_move_done = ctl_config_move_done;
7036         ctl_datamove((union ctl_io *)ctsio);
7037         return (CTL_RETVAL_COMPLETE);
7038 }
7039
7040 int
7041 ctl_report_tagret_port_groups(struct ctl_scsiio *ctsio)
7042 {
7043         struct scsi_maintenance_in *cdb;
7044         int retval;
7045         int alloc_len, ext, total_len = 0, g, pc, pg, gs, os;
7046         int num_target_port_groups, num_target_ports;
7047         struct ctl_lun *lun;
7048         struct ctl_softc *softc;
7049         struct ctl_port *port;
7050         struct scsi_target_group_data *rtg_ptr;
7051         struct scsi_target_group_data_extended *rtg_ext_ptr;
7052         struct scsi_target_port_group_descriptor *tpg_desc;
7053
7054         CTL_DEBUG_PRINT(("ctl_report_tagret_port_groups\n"));
7055
7056         cdb = (struct scsi_maintenance_in *)ctsio->cdb;
7057         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7058         softc = lun->ctl_softc;
7059
7060         retval = CTL_RETVAL_COMPLETE;
7061
7062         switch (cdb->byte2 & STG_PDF_MASK) {
7063         case STG_PDF_LENGTH:
7064                 ext = 0;
7065                 break;
7066         case STG_PDF_EXTENDED:
7067                 ext = 1;
7068                 break;
7069         default:
7070                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7071                                       /*sks_valid*/ 1,
7072                                       /*command*/ 1,
7073                                       /*field*/ 2,
7074                                       /*bit_valid*/ 1,
7075                                       /*bit*/ 5);
7076                 ctl_done((union ctl_io *)ctsio);
7077                 return(retval);
7078         }
7079
7080         if (softc->is_single)
7081                 num_target_port_groups = 1;
7082         else
7083                 num_target_port_groups = NUM_TARGET_PORT_GROUPS;
7084         num_target_ports = 0;
7085         mtx_lock(&softc->ctl_lock);
7086         STAILQ_FOREACH(port, &softc->port_list, links) {
7087                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7088                         continue;
7089                 if (ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
7090                         continue;
7091                 num_target_ports++;
7092         }
7093         mtx_unlock(&softc->ctl_lock);
7094
7095         if (ext)
7096                 total_len = sizeof(struct scsi_target_group_data_extended);
7097         else
7098                 total_len = sizeof(struct scsi_target_group_data);
7099         total_len += sizeof(struct scsi_target_port_group_descriptor) *
7100                 num_target_port_groups +
7101             sizeof(struct scsi_target_port_descriptor) * num_target_ports;
7102
7103         alloc_len = scsi_4btoul(cdb->length);
7104
7105         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7106
7107         ctsio->kern_sg_entries = 0;
7108
7109         if (total_len < alloc_len) {
7110                 ctsio->residual = alloc_len - total_len;
7111                 ctsio->kern_data_len = total_len;
7112                 ctsio->kern_total_len = total_len;
7113         } else {
7114                 ctsio->residual = 0;
7115                 ctsio->kern_data_len = alloc_len;
7116                 ctsio->kern_total_len = alloc_len;
7117         }
7118         ctsio->kern_data_resid = 0;
7119         ctsio->kern_rel_offset = 0;
7120
7121         if (ext) {
7122                 rtg_ext_ptr = (struct scsi_target_group_data_extended *)
7123                     ctsio->kern_data_ptr;
7124                 scsi_ulto4b(total_len - 4, rtg_ext_ptr->length);
7125                 rtg_ext_ptr->format_type = 0x10;
7126                 rtg_ext_ptr->implicit_transition_time = 0;
7127                 tpg_desc = &rtg_ext_ptr->groups[0];
7128         } else {
7129                 rtg_ptr = (struct scsi_target_group_data *)
7130                     ctsio->kern_data_ptr;
7131                 scsi_ulto4b(total_len - 4, rtg_ptr->length);
7132                 tpg_desc = &rtg_ptr->groups[0];
7133         }
7134
7135         mtx_lock(&softc->ctl_lock);
7136         pg = softc->port_min / softc->port_cnt;
7137         if (softc->ha_link == CTL_HA_LINK_OFFLINE)
7138                 gs = TPG_ASYMMETRIC_ACCESS_UNAVAILABLE;
7139         else if (softc->ha_link == CTL_HA_LINK_UNKNOWN)
7140                 gs = TPG_ASYMMETRIC_ACCESS_TRANSITIONING;
7141         else if (softc->ha_mode == CTL_HA_MODE_ACT_STBY)
7142                 gs = TPG_ASYMMETRIC_ACCESS_STANDBY;
7143         else
7144                 gs = TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7145         if (lun->flags & CTL_LUN_PRIMARY_SC) {
7146                 os = gs;
7147                 gs = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7148         } else
7149                 os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7150         for (g = 0; g < num_target_port_groups; g++) {
7151                 tpg_desc->pref_state = (g == pg) ? gs : os;
7152                 tpg_desc->support = TPG_AO_SUP | TPG_AN_SUP | TPG_S_SUP |
7153                     TPG_U_SUP | TPG_T_SUP;
7154                 scsi_ulto2b(g + 1, tpg_desc->target_port_group);
7155                 tpg_desc->status = TPG_IMPLICIT;
7156                 pc = 0;
7157                 STAILQ_FOREACH(port, &softc->port_list, links) {
7158                         if (port->targ_port < g * softc->port_cnt ||
7159                             port->targ_port >= (g + 1) * softc->port_cnt)
7160                                 continue;
7161                         if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7162                                 continue;
7163                         if (ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
7164                                 continue;
7165                         scsi_ulto2b(port->targ_port, tpg_desc->descriptors[pc].
7166                             relative_target_port_identifier);
7167                         pc++;
7168                 }
7169                 tpg_desc->target_port_count = pc;
7170                 tpg_desc = (struct scsi_target_port_group_descriptor *)
7171                     &tpg_desc->descriptors[pc];
7172         }
7173         mtx_unlock(&softc->ctl_lock);
7174
7175         ctl_set_success(ctsio);
7176         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7177         ctsio->be_move_done = ctl_config_move_done;
7178         ctl_datamove((union ctl_io *)ctsio);
7179         return(retval);
7180 }
7181
7182 int
7183 ctl_report_supported_opcodes(struct ctl_scsiio *ctsio)
7184 {
7185         struct ctl_lun *lun;
7186         struct scsi_report_supported_opcodes *cdb;
7187         const struct ctl_cmd_entry *entry, *sentry;
7188         struct scsi_report_supported_opcodes_all *all;
7189         struct scsi_report_supported_opcodes_descr *descr;
7190         struct scsi_report_supported_opcodes_one *one;
7191         int retval;
7192         int alloc_len, total_len;
7193         int opcode, service_action, i, j, num;
7194
7195         CTL_DEBUG_PRINT(("ctl_report_supported_opcodes\n"));
7196
7197         cdb = (struct scsi_report_supported_opcodes *)ctsio->cdb;
7198         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7199
7200         retval = CTL_RETVAL_COMPLETE;
7201
7202         opcode = cdb->requested_opcode;
7203         service_action = scsi_2btoul(cdb->requested_service_action);
7204         switch (cdb->options & RSO_OPTIONS_MASK) {
7205         case RSO_OPTIONS_ALL:
7206                 num = 0;
7207                 for (i = 0; i < 256; i++) {
7208                         entry = &ctl_cmd_table[i];
7209                         if (entry->flags & CTL_CMD_FLAG_SA5) {
7210                                 for (j = 0; j < 32; j++) {
7211                                         sentry = &((const struct ctl_cmd_entry *)
7212                                             entry->execute)[j];
7213                                         if (ctl_cmd_applicable(
7214                                             lun->be_lun->lun_type, sentry))
7215                                                 num++;
7216                                 }
7217                         } else {
7218                                 if (ctl_cmd_applicable(lun->be_lun->lun_type,
7219                                     entry))
7220                                         num++;
7221                         }
7222                 }
7223                 total_len = sizeof(struct scsi_report_supported_opcodes_all) +
7224                     num * sizeof(struct scsi_report_supported_opcodes_descr);
7225                 break;
7226         case RSO_OPTIONS_OC:
7227                 if (ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) {
7228                         ctl_set_invalid_field(/*ctsio*/ ctsio,
7229                                               /*sks_valid*/ 1,
7230                                               /*command*/ 1,
7231                                               /*field*/ 2,
7232                                               /*bit_valid*/ 1,
7233                                               /*bit*/ 2);
7234                         ctl_done((union ctl_io *)ctsio);
7235                         return (CTL_RETVAL_COMPLETE);
7236                 }
7237                 total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7238                 break;
7239         case RSO_OPTIONS_OC_SA:
7240                 if ((ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) == 0 ||
7241                     service_action >= 32) {
7242                         ctl_set_invalid_field(/*ctsio*/ ctsio,
7243                                               /*sks_valid*/ 1,
7244                                               /*command*/ 1,
7245                                               /*field*/ 2,
7246                                               /*bit_valid*/ 1,
7247                                               /*bit*/ 2);
7248                         ctl_done((union ctl_io *)ctsio);
7249                         return (CTL_RETVAL_COMPLETE);
7250                 }
7251                 total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7252                 break;
7253         default:
7254                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7255                                       /*sks_valid*/ 1,
7256                                       /*command*/ 1,
7257                                       /*field*/ 2,
7258                                       /*bit_valid*/ 1,
7259                                       /*bit*/ 2);
7260                 ctl_done((union ctl_io *)ctsio);
7261                 return (CTL_RETVAL_COMPLETE);
7262         }
7263
7264         alloc_len = scsi_4btoul(cdb->length);
7265
7266         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7267
7268         ctsio->kern_sg_entries = 0;
7269
7270         if (total_len < alloc_len) {
7271                 ctsio->residual = alloc_len - total_len;
7272                 ctsio->kern_data_len = total_len;
7273                 ctsio->kern_total_len = total_len;
7274         } else {
7275                 ctsio->residual = 0;
7276                 ctsio->kern_data_len = alloc_len;
7277                 ctsio->kern_total_len = alloc_len;
7278         }
7279         ctsio->kern_data_resid = 0;
7280         ctsio->kern_rel_offset = 0;
7281
7282         switch (cdb->options & RSO_OPTIONS_MASK) {
7283         case RSO_OPTIONS_ALL:
7284                 all = (struct scsi_report_supported_opcodes_all *)
7285                     ctsio->kern_data_ptr;
7286                 num = 0;
7287                 for (i = 0; i < 256; i++) {
7288                         entry = &ctl_cmd_table[i];
7289                         if (entry->flags & CTL_CMD_FLAG_SA5) {
7290                                 for (j = 0; j < 32; j++) {
7291                                         sentry = &((const struct ctl_cmd_entry *)
7292                                             entry->execute)[j];
7293                                         if (!ctl_cmd_applicable(
7294                                             lun->be_lun->lun_type, sentry))
7295                                                 continue;
7296                                         descr = &all->descr[num++];
7297                                         descr->opcode = i;
7298                                         scsi_ulto2b(j, descr->service_action);
7299                                         descr->flags = RSO_SERVACTV;
7300                                         scsi_ulto2b(sentry->length,
7301                                             descr->cdb_length);
7302                                 }
7303                         } else {
7304                                 if (!ctl_cmd_applicable(lun->be_lun->lun_type,
7305                                     entry))
7306                                         continue;
7307                                 descr = &all->descr[num++];
7308                                 descr->opcode = i;
7309                                 scsi_ulto2b(0, descr->service_action);
7310                                 descr->flags = 0;
7311                                 scsi_ulto2b(entry->length, descr->cdb_length);
7312                         }
7313                 }
7314                 scsi_ulto4b(
7315                     num * sizeof(struct scsi_report_supported_opcodes_descr),
7316                     all->length);
7317                 break;
7318         case RSO_OPTIONS_OC:
7319                 one = (struct scsi_report_supported_opcodes_one *)
7320                     ctsio->kern_data_ptr;
7321                 entry = &ctl_cmd_table[opcode];
7322                 goto fill_one;
7323         case RSO_OPTIONS_OC_SA:
7324                 one = (struct scsi_report_supported_opcodes_one *)
7325                     ctsio->kern_data_ptr;
7326                 entry = &ctl_cmd_table[opcode];
7327                 entry = &((const struct ctl_cmd_entry *)
7328                     entry->execute)[service_action];
7329 fill_one:
7330                 if (ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
7331                         one->support = 3;
7332                         scsi_ulto2b(entry->length, one->cdb_length);
7333                         one->cdb_usage[0] = opcode;
7334                         memcpy(&one->cdb_usage[1], entry->usage,
7335                             entry->length - 1);
7336                 } else
7337                         one->support = 1;
7338                 break;
7339         }
7340
7341         ctl_set_success(ctsio);
7342         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7343         ctsio->be_move_done = ctl_config_move_done;
7344         ctl_datamove((union ctl_io *)ctsio);
7345         return(retval);
7346 }
7347
7348 int
7349 ctl_report_supported_tmf(struct ctl_scsiio *ctsio)
7350 {
7351         struct scsi_report_supported_tmf *cdb;
7352         struct scsi_report_supported_tmf_data *data;
7353         int retval;
7354         int alloc_len, total_len;
7355
7356         CTL_DEBUG_PRINT(("ctl_report_supported_tmf\n"));
7357
7358         cdb = (struct scsi_report_supported_tmf *)ctsio->cdb;
7359
7360         retval = CTL_RETVAL_COMPLETE;
7361
7362         total_len = sizeof(struct scsi_report_supported_tmf_data);
7363         alloc_len = scsi_4btoul(cdb->length);
7364
7365         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7366
7367         ctsio->kern_sg_entries = 0;
7368
7369         if (total_len < alloc_len) {
7370                 ctsio->residual = alloc_len - total_len;
7371                 ctsio->kern_data_len = total_len;
7372                 ctsio->kern_total_len = total_len;
7373         } else {
7374                 ctsio->residual = 0;
7375                 ctsio->kern_data_len = alloc_len;
7376                 ctsio->kern_total_len = alloc_len;
7377         }
7378         ctsio->kern_data_resid = 0;
7379         ctsio->kern_rel_offset = 0;
7380
7381         data = (struct scsi_report_supported_tmf_data *)ctsio->kern_data_ptr;
7382         data->byte1 |= RST_ATS | RST_ATSS | RST_CTSS | RST_LURS | RST_QTS |
7383             RST_TRS;
7384         data->byte2 |= RST_QAES | RST_QTSS | RST_ITNRS;
7385
7386         ctl_set_success(ctsio);
7387         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7388         ctsio->be_move_done = ctl_config_move_done;
7389         ctl_datamove((union ctl_io *)ctsio);
7390         return (retval);
7391 }
7392
7393 int
7394 ctl_report_timestamp(struct ctl_scsiio *ctsio)
7395 {
7396         struct scsi_report_timestamp *cdb;
7397         struct scsi_report_timestamp_data *data;
7398         struct timeval tv;
7399         int64_t timestamp;
7400         int retval;
7401         int alloc_len, total_len;
7402
7403         CTL_DEBUG_PRINT(("ctl_report_timestamp\n"));
7404
7405         cdb = (struct scsi_report_timestamp *)ctsio->cdb;
7406
7407         retval = CTL_RETVAL_COMPLETE;
7408
7409         total_len = sizeof(struct scsi_report_timestamp_data);
7410         alloc_len = scsi_4btoul(cdb->length);
7411
7412         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7413
7414         ctsio->kern_sg_entries = 0;
7415
7416         if (total_len < alloc_len) {
7417                 ctsio->residual = alloc_len - total_len;
7418                 ctsio->kern_data_len = total_len;
7419                 ctsio->kern_total_len = total_len;
7420         } else {
7421                 ctsio->residual = 0;
7422                 ctsio->kern_data_len = alloc_len;
7423                 ctsio->kern_total_len = alloc_len;
7424         }
7425         ctsio->kern_data_resid = 0;
7426         ctsio->kern_rel_offset = 0;
7427
7428         data = (struct scsi_report_timestamp_data *)ctsio->kern_data_ptr;
7429         scsi_ulto2b(sizeof(*data) - 2, data->length);
7430         data->origin = RTS_ORIG_OUTSIDE;
7431         getmicrotime(&tv);
7432         timestamp = (int64_t)tv.tv_sec * 1000 + tv.tv_usec / 1000;
7433         scsi_ulto4b(timestamp >> 16, data->timestamp);
7434         scsi_ulto2b(timestamp & 0xffff, &data->timestamp[4]);
7435
7436         ctl_set_success(ctsio);
7437         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7438         ctsio->be_move_done = ctl_config_move_done;
7439         ctl_datamove((union ctl_io *)ctsio);
7440         return (retval);
7441 }
7442
7443 int
7444 ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7445 {
7446         struct scsi_per_res_in *cdb;
7447         int alloc_len, total_len = 0;
7448         /* struct scsi_per_res_in_rsrv in_data; */
7449         struct ctl_lun *lun;
7450         struct ctl_softc *softc;
7451         uint64_t key;
7452
7453         CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7454
7455         cdb = (struct scsi_per_res_in *)ctsio->cdb;
7456
7457         alloc_len = scsi_2btoul(cdb->length);
7458
7459         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7460         softc = lun->ctl_softc;
7461
7462 retry:
7463         mtx_lock(&lun->lun_lock);
7464         switch (cdb->action) {
7465         case SPRI_RK: /* read keys */
7466                 total_len = sizeof(struct scsi_per_res_in_keys) +
7467                         lun->pr_key_count *
7468                         sizeof(struct scsi_per_res_key);
7469                 break;
7470         case SPRI_RR: /* read reservation */
7471                 if (lun->flags & CTL_LUN_PR_RESERVED)
7472                         total_len = sizeof(struct scsi_per_res_in_rsrv);
7473                 else
7474                         total_len = sizeof(struct scsi_per_res_in_header);
7475                 break;
7476         case SPRI_RC: /* report capabilities */
7477                 total_len = sizeof(struct scsi_per_res_cap);
7478                 break;
7479         case SPRI_RS: /* read full status */
7480                 total_len = sizeof(struct scsi_per_res_in_header) +
7481                     (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7482                     lun->pr_key_count;
7483                 break;
7484         default:
7485                 panic("Invalid PR type %x", cdb->action);
7486         }
7487         mtx_unlock(&lun->lun_lock);
7488
7489         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7490
7491         if (total_len < alloc_len) {
7492                 ctsio->residual = alloc_len - total_len;
7493                 ctsio->kern_data_len = total_len;
7494                 ctsio->kern_total_len = total_len;
7495         } else {
7496                 ctsio->residual = 0;
7497                 ctsio->kern_data_len = alloc_len;
7498                 ctsio->kern_total_len = alloc_len;
7499         }
7500
7501         ctsio->kern_data_resid = 0;
7502         ctsio->kern_rel_offset = 0;
7503         ctsio->kern_sg_entries = 0;
7504
7505         mtx_lock(&lun->lun_lock);
7506         switch (cdb->action) {
7507         case SPRI_RK: { // read keys
7508         struct scsi_per_res_in_keys *res_keys;
7509                 int i, key_count;
7510
7511                 res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7512
7513                 /*
7514                  * We had to drop the lock to allocate our buffer, which
7515                  * leaves time for someone to come in with another
7516                  * persistent reservation.  (That is unlikely, though,
7517                  * since this should be the only persistent reservation
7518                  * command active right now.)
7519                  */
7520                 if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7521                     (lun->pr_key_count *
7522                      sizeof(struct scsi_per_res_key)))){
7523                         mtx_unlock(&lun->lun_lock);
7524                         free(ctsio->kern_data_ptr, M_CTL);
7525                         printf("%s: reservation length changed, retrying\n",
7526                                __func__);
7527                         goto retry;
7528                 }
7529
7530                 scsi_ulto4b(lun->PRGeneration, res_keys->header.generation);
7531
7532                 scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7533                              lun->pr_key_count, res_keys->header.length);
7534
7535                 for (i = 0, key_count = 0; i < CTL_MAX_INITIATORS; i++) {
7536                         if ((key = ctl_get_prkey(lun, i)) == 0)
7537                                 continue;
7538
7539                         /*
7540                          * We used lun->pr_key_count to calculate the
7541                          * size to allocate.  If it turns out the number of
7542                          * initiators with the registered flag set is
7543                          * larger than that (i.e. they haven't been kept in
7544                          * sync), we've got a problem.
7545                          */
7546                         if (key_count >= lun->pr_key_count) {
7547                                 key_count++;
7548                                 continue;
7549                         }
7550                         scsi_u64to8b(key, res_keys->keys[key_count].key);
7551                         key_count++;
7552                 }
7553                 break;
7554         }
7555         case SPRI_RR: { // read reservation
7556                 struct scsi_per_res_in_rsrv *res;
7557                 int tmp_len, header_only;
7558
7559                 res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7560
7561                 scsi_ulto4b(lun->PRGeneration, res->header.generation);
7562
7563                 if (lun->flags & CTL_LUN_PR_RESERVED)
7564                 {
7565                         tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7566                         scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7567                                     res->header.length);
7568                         header_only = 0;
7569                 } else {
7570                         tmp_len = sizeof(struct scsi_per_res_in_header);
7571                         scsi_ulto4b(0, res->header.length);
7572                         header_only = 1;
7573                 }
7574
7575                 /*
7576                  * We had to drop the lock to allocate our buffer, which
7577                  * leaves time for someone to come in with another
7578                  * persistent reservation.  (That is unlikely, though,
7579                  * since this should be the only persistent reservation
7580                  * command active right now.)
7581                  */
7582                 if (tmp_len != total_len) {
7583                         mtx_unlock(&lun->lun_lock);
7584                         free(ctsio->kern_data_ptr, M_CTL);
7585                         printf("%s: reservation status changed, retrying\n",
7586                                __func__);
7587                         goto retry;
7588                 }
7589
7590                 /*
7591                  * No reservation held, so we're done.
7592                  */
7593                 if (header_only != 0)
7594                         break;
7595
7596                 /*
7597                  * If the registration is an All Registrants type, the key
7598                  * is 0, since it doesn't really matter.
7599                  */
7600                 if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7601                         scsi_u64to8b(ctl_get_prkey(lun, lun->pr_res_idx),
7602                             res->data.reservation);
7603                 }
7604                 res->data.scopetype = lun->res_type;
7605                 break;
7606         }
7607         case SPRI_RC:     //report capabilities
7608         {
7609                 struct scsi_per_res_cap *res_cap;
7610                 uint16_t type_mask;
7611
7612                 res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7613                 scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7614                 res_cap->flags2 |= SPRI_TMV | SPRI_ALLOW_5;
7615                 type_mask = SPRI_TM_WR_EX_AR |
7616                             SPRI_TM_EX_AC_RO |
7617                             SPRI_TM_WR_EX_RO |
7618                             SPRI_TM_EX_AC |
7619                             SPRI_TM_WR_EX |
7620                             SPRI_TM_EX_AC_AR;
7621                 scsi_ulto2b(type_mask, res_cap->type_mask);
7622                 break;
7623         }
7624         case SPRI_RS: { // read full status
7625                 struct scsi_per_res_in_full *res_status;
7626                 struct scsi_per_res_in_full_desc *res_desc;
7627                 struct ctl_port *port;
7628                 int i, len;
7629
7630                 res_status = (struct scsi_per_res_in_full*)ctsio->kern_data_ptr;
7631
7632                 /*
7633                  * We had to drop the lock to allocate our buffer, which
7634                  * leaves time for someone to come in with another
7635                  * persistent reservation.  (That is unlikely, though,
7636                  * since this should be the only persistent reservation
7637                  * command active right now.)
7638                  */
7639                 if (total_len < (sizeof(struct scsi_per_res_in_header) +
7640                     (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7641                      lun->pr_key_count)){
7642                         mtx_unlock(&lun->lun_lock);
7643                         free(ctsio->kern_data_ptr, M_CTL);
7644                         printf("%s: reservation length changed, retrying\n",
7645                                __func__);
7646                         goto retry;
7647                 }
7648
7649                 scsi_ulto4b(lun->PRGeneration, res_status->header.generation);
7650
7651                 res_desc = &res_status->desc[0];
7652                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
7653                         if ((key = ctl_get_prkey(lun, i)) == 0)
7654                                 continue;
7655
7656                         scsi_u64to8b(key, res_desc->res_key.key);
7657                         if ((lun->flags & CTL_LUN_PR_RESERVED) &&
7658                             (lun->pr_res_idx == i ||
7659                              lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS)) {
7660                                 res_desc->flags = SPRI_FULL_R_HOLDER;
7661                                 res_desc->scopetype = lun->res_type;
7662                         }
7663                         scsi_ulto2b(i / CTL_MAX_INIT_PER_PORT,
7664                             res_desc->rel_trgt_port_id);
7665                         len = 0;
7666                         port = softc->ctl_ports[i / CTL_MAX_INIT_PER_PORT];
7667                         if (port != NULL)
7668                                 len = ctl_create_iid(port,
7669                                     i % CTL_MAX_INIT_PER_PORT,
7670                                     res_desc->transport_id);
7671                         scsi_ulto4b(len, res_desc->additional_length);
7672                         res_desc = (struct scsi_per_res_in_full_desc *)
7673                             &res_desc->transport_id[len];
7674                 }
7675                 scsi_ulto4b((uint8_t *)res_desc - (uint8_t *)&res_status->desc[0],
7676                     res_status->header.length);
7677                 break;
7678         }
7679         default:
7680                 /*
7681                  * This is a bug, because we just checked for this above,
7682                  * and should have returned an error.
7683                  */
7684                 panic("Invalid PR type %x", cdb->action);
7685                 break; /* NOTREACHED */
7686         }
7687         mtx_unlock(&lun->lun_lock);
7688
7689         ctl_set_success(ctsio);
7690         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7691         ctsio->be_move_done = ctl_config_move_done;
7692         ctl_datamove((union ctl_io *)ctsio);
7693         return (CTL_RETVAL_COMPLETE);
7694 }
7695
7696 /*
7697  * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
7698  * it should return.
7699  */
7700 static int
7701 ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
7702                 uint64_t sa_res_key, uint8_t type, uint32_t residx,
7703                 struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
7704                 struct scsi_per_res_out_parms* param)
7705 {
7706         union ctl_ha_msg persis_io;
7707         int i;
7708
7709         mtx_lock(&lun->lun_lock);
7710         if (sa_res_key == 0) {
7711                 if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
7712                         /* validate scope and type */
7713                         if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7714                              SPR_LU_SCOPE) {
7715                                 mtx_unlock(&lun->lun_lock);
7716                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7717                                                       /*sks_valid*/ 1,
7718                                                       /*command*/ 1,
7719                                                       /*field*/ 2,
7720                                                       /*bit_valid*/ 1,
7721                                                       /*bit*/ 4);
7722                                 ctl_done((union ctl_io *)ctsio);
7723                                 return (1);
7724                         }
7725
7726                         if (type>8 || type==2 || type==4 || type==0) {
7727                                 mtx_unlock(&lun->lun_lock);
7728                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7729                                                       /*sks_valid*/ 1,
7730                                                       /*command*/ 1,
7731                                                       /*field*/ 2,
7732                                                       /*bit_valid*/ 1,
7733                                                       /*bit*/ 0);
7734                                 ctl_done((union ctl_io *)ctsio);
7735                                 return (1);
7736                         }
7737
7738                         /*
7739                          * Unregister everybody else and build UA for
7740                          * them
7741                          */
7742                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
7743                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
7744                                         continue;
7745
7746                                 ctl_clr_prkey(lun, i);
7747                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7748                         }
7749                         lun->pr_key_count = 1;
7750                         lun->res_type = type;
7751                         if (lun->res_type != SPR_TYPE_WR_EX_AR
7752                          && lun->res_type != SPR_TYPE_EX_AC_AR)
7753                                 lun->pr_res_idx = residx;
7754                         lun->PRGeneration++;
7755                         mtx_unlock(&lun->lun_lock);
7756
7757                         /* send msg to other side */
7758                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7759                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7760                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7761                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
7762                         persis_io.pr.pr_info.res_type = type;
7763                         memcpy(persis_io.pr.pr_info.sa_res_key,
7764                                param->serv_act_res_key,
7765                                sizeof(param->serv_act_res_key));
7766                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
7767                             sizeof(persis_io.pr), M_WAITOK);
7768                 } else {
7769                         /* not all registrants */
7770                         mtx_unlock(&lun->lun_lock);
7771                         free(ctsio->kern_data_ptr, M_CTL);
7772                         ctl_set_invalid_field(ctsio,
7773                                               /*sks_valid*/ 1,
7774                                               /*command*/ 0,
7775                                               /*field*/ 8,
7776                                               /*bit_valid*/ 0,
7777                                               /*bit*/ 0);
7778                         ctl_done((union ctl_io *)ctsio);
7779                         return (1);
7780                 }
7781         } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
7782                 || !(lun->flags & CTL_LUN_PR_RESERVED)) {
7783                 int found = 0;
7784
7785                 if (res_key == sa_res_key) {
7786                         /* special case */
7787                         /*
7788                          * The spec implies this is not good but doesn't
7789                          * say what to do. There are two choices either
7790                          * generate a res conflict or check condition
7791                          * with illegal field in parameter data. Since
7792                          * that is what is done when the sa_res_key is
7793                          * zero I'll take that approach since this has
7794                          * to do with the sa_res_key.
7795                          */
7796                         mtx_unlock(&lun->lun_lock);
7797                         free(ctsio->kern_data_ptr, M_CTL);
7798                         ctl_set_invalid_field(ctsio,
7799                                               /*sks_valid*/ 1,
7800                                               /*command*/ 0,
7801                                               /*field*/ 8,
7802                                               /*bit_valid*/ 0,
7803                                               /*bit*/ 0);
7804                         ctl_done((union ctl_io *)ctsio);
7805                         return (1);
7806                 }
7807
7808                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
7809                         if (ctl_get_prkey(lun, i) != sa_res_key)
7810                                 continue;
7811
7812                         found = 1;
7813                         ctl_clr_prkey(lun, i);
7814                         lun->pr_key_count--;
7815                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7816                 }
7817                 if (!found) {
7818                         mtx_unlock(&lun->lun_lock);
7819                         free(ctsio->kern_data_ptr, M_CTL);
7820                         ctl_set_reservation_conflict(ctsio);
7821                         ctl_done((union ctl_io *)ctsio);
7822                         return (CTL_RETVAL_COMPLETE);
7823                 }
7824                 lun->PRGeneration++;
7825                 mtx_unlock(&lun->lun_lock);
7826
7827                 /* send msg to other side */
7828                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7829                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7830                 persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7831                 persis_io.pr.pr_info.residx = lun->pr_res_idx;
7832                 persis_io.pr.pr_info.res_type = type;
7833                 memcpy(persis_io.pr.pr_info.sa_res_key,
7834                        param->serv_act_res_key,
7835                        sizeof(param->serv_act_res_key));
7836                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
7837                     sizeof(persis_io.pr), M_WAITOK);
7838         } else {
7839                 /* Reserved but not all registrants */
7840                 /* sa_res_key is res holder */
7841                 if (sa_res_key == ctl_get_prkey(lun, lun->pr_res_idx)) {
7842                         /* validate scope and type */
7843                         if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7844                              SPR_LU_SCOPE) {
7845                                 mtx_unlock(&lun->lun_lock);
7846                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7847                                                       /*sks_valid*/ 1,
7848                                                       /*command*/ 1,
7849                                                       /*field*/ 2,
7850                                                       /*bit_valid*/ 1,
7851                                                       /*bit*/ 4);
7852                                 ctl_done((union ctl_io *)ctsio);
7853                                 return (1);
7854                         }
7855
7856                         if (type>8 || type==2 || type==4 || type==0) {
7857                                 mtx_unlock(&lun->lun_lock);
7858                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7859                                                       /*sks_valid*/ 1,
7860                                                       /*command*/ 1,
7861                                                       /*field*/ 2,
7862                                                       /*bit_valid*/ 1,
7863                                                       /*bit*/ 0);
7864                                 ctl_done((union ctl_io *)ctsio);
7865                                 return (1);
7866                         }
7867
7868                         /*
7869                          * Do the following:
7870                          * if sa_res_key != res_key remove all
7871                          * registrants w/sa_res_key and generate UA
7872                          * for these registrants(Registrations
7873                          * Preempted) if it wasn't an exclusive
7874                          * reservation generate UA(Reservations
7875                          * Preempted) for all other registered nexuses
7876                          * if the type has changed. Establish the new
7877                          * reservation and holder. If res_key and
7878                          * sa_res_key are the same do the above
7879                          * except don't unregister the res holder.
7880                          */
7881
7882                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
7883                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
7884                                         continue;
7885
7886                                 if (sa_res_key == ctl_get_prkey(lun, i)) {
7887                                         ctl_clr_prkey(lun, i);
7888                                         lun->pr_key_count--;
7889                                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7890                                 } else if (type != lun->res_type
7891                                         && (lun->res_type == SPR_TYPE_WR_EX_RO
7892                                          || lun->res_type ==SPR_TYPE_EX_AC_RO)){
7893                                         ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
7894                                 }
7895                         }
7896                         lun->res_type = type;
7897                         if (lun->res_type != SPR_TYPE_WR_EX_AR
7898                          && lun->res_type != SPR_TYPE_EX_AC_AR)
7899                                 lun->pr_res_idx = residx;
7900                         else
7901                                 lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
7902                         lun->PRGeneration++;
7903                         mtx_unlock(&lun->lun_lock);
7904
7905                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7906                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7907                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7908                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
7909                         persis_io.pr.pr_info.res_type = type;
7910                         memcpy(persis_io.pr.pr_info.sa_res_key,
7911                                param->serv_act_res_key,
7912                                sizeof(param->serv_act_res_key));
7913                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
7914                             sizeof(persis_io.pr), M_WAITOK);
7915                 } else {
7916                         /*
7917                          * sa_res_key is not the res holder just
7918                          * remove registrants
7919                          */
7920                         int found=0;
7921
7922                         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
7923                                 if (sa_res_key != ctl_get_prkey(lun, i))
7924                                         continue;
7925
7926                                 found = 1;
7927                                 ctl_clr_prkey(lun, i);
7928                                 lun->pr_key_count--;
7929                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7930                         }
7931
7932                         if (!found) {
7933                                 mtx_unlock(&lun->lun_lock);
7934                                 free(ctsio->kern_data_ptr, M_CTL);
7935                                 ctl_set_reservation_conflict(ctsio);
7936                                 ctl_done((union ctl_io *)ctsio);
7937                                 return (1);
7938                         }
7939                         lun->PRGeneration++;
7940                         mtx_unlock(&lun->lun_lock);
7941
7942                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7943                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7944                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7945                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
7946                         persis_io.pr.pr_info.res_type = type;
7947                         memcpy(persis_io.pr.pr_info.sa_res_key,
7948                                param->serv_act_res_key,
7949                                sizeof(param->serv_act_res_key));
7950                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
7951                             sizeof(persis_io.pr), M_WAITOK);
7952                 }
7953         }
7954         return (0);
7955 }
7956
7957 static void
7958 ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
7959 {
7960         uint64_t sa_res_key;
7961         int i;
7962
7963         sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
7964
7965         if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
7966          || lun->pr_res_idx == CTL_PR_NO_RESERVATION
7967          || sa_res_key != ctl_get_prkey(lun, lun->pr_res_idx)) {
7968                 if (sa_res_key == 0) {
7969                         /*
7970                          * Unregister everybody else and build UA for
7971                          * them
7972                          */
7973                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
7974                                 if (i == msg->pr.pr_info.residx ||
7975                                     ctl_get_prkey(lun, i) == 0)
7976                                         continue;
7977
7978                                 ctl_clr_prkey(lun, i);
7979                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7980                         }
7981
7982                         lun->pr_key_count = 1;
7983                         lun->res_type = msg->pr.pr_info.res_type;
7984                         if (lun->res_type != SPR_TYPE_WR_EX_AR
7985                          && lun->res_type != SPR_TYPE_EX_AC_AR)
7986                                 lun->pr_res_idx = msg->pr.pr_info.residx;
7987                 } else {
7988                         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
7989                                 if (sa_res_key == ctl_get_prkey(lun, i))
7990                                         continue;
7991
7992                                 ctl_clr_prkey(lun, i);
7993                                 lun->pr_key_count--;
7994                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7995                         }
7996                 }
7997         } else {
7998                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
7999                         if (i == msg->pr.pr_info.residx ||
8000                             ctl_get_prkey(lun, i) == 0)
8001                                 continue;
8002
8003                         if (sa_res_key == ctl_get_prkey(lun, i)) {
8004                                 ctl_clr_prkey(lun, i);
8005                                 lun->pr_key_count--;
8006                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8007                         } else if (msg->pr.pr_info.res_type != lun->res_type
8008                                 && (lun->res_type == SPR_TYPE_WR_EX_RO
8009                                  || lun->res_type == SPR_TYPE_EX_AC_RO)) {
8010                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8011                         }
8012                 }
8013                 lun->res_type = msg->pr.pr_info.res_type;
8014                 if (lun->res_type != SPR_TYPE_WR_EX_AR
8015                  && lun->res_type != SPR_TYPE_EX_AC_AR)
8016                         lun->pr_res_idx = msg->pr.pr_info.residx;
8017                 else
8018                         lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8019         }
8020         lun->PRGeneration++;
8021
8022 }
8023
8024
8025 int
8026 ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
8027 {
8028         int retval;
8029         u_int32_t param_len;
8030         struct scsi_per_res_out *cdb;
8031         struct ctl_lun *lun;
8032         struct scsi_per_res_out_parms* param;
8033         struct ctl_softc *softc;
8034         uint32_t residx;
8035         uint64_t res_key, sa_res_key, key;
8036         uint8_t type;
8037         union ctl_ha_msg persis_io;
8038         int    i;
8039
8040         CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n"));
8041
8042         retval = CTL_RETVAL_COMPLETE;
8043
8044         cdb = (struct scsi_per_res_out *)ctsio->cdb;
8045         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8046         softc = lun->ctl_softc;
8047
8048         /*
8049          * We only support whole-LUN scope.  The scope & type are ignored for
8050          * register, register and ignore existing key and clear.
8051          * We sometimes ignore scope and type on preempts too!!
8052          * Verify reservation type here as well.
8053          */
8054         type = cdb->scope_type & SPR_TYPE_MASK;
8055         if ((cdb->action == SPRO_RESERVE)
8056          || (cdb->action == SPRO_RELEASE)) {
8057                 if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
8058                         ctl_set_invalid_field(/*ctsio*/ ctsio,
8059                                               /*sks_valid*/ 1,
8060                                               /*command*/ 1,
8061                                               /*field*/ 2,
8062                                               /*bit_valid*/ 1,
8063                                               /*bit*/ 4);
8064                         ctl_done((union ctl_io *)ctsio);
8065                         return (CTL_RETVAL_COMPLETE);
8066                 }
8067
8068                 if (type>8 || type==2 || type==4 || type==0) {
8069                         ctl_set_invalid_field(/*ctsio*/ ctsio,
8070                                               /*sks_valid*/ 1,
8071                                               /*command*/ 1,
8072                                               /*field*/ 2,
8073                                               /*bit_valid*/ 1,
8074                                               /*bit*/ 0);
8075                         ctl_done((union ctl_io *)ctsio);
8076                         return (CTL_RETVAL_COMPLETE);
8077                 }
8078         }
8079
8080         param_len = scsi_4btoul(cdb->length);
8081
8082         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
8083                 ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
8084                 ctsio->kern_data_len = param_len;
8085                 ctsio->kern_total_len = param_len;
8086                 ctsio->kern_data_resid = 0;
8087                 ctsio->kern_rel_offset = 0;
8088                 ctsio->kern_sg_entries = 0;
8089                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
8090                 ctsio->be_move_done = ctl_config_move_done;
8091                 ctl_datamove((union ctl_io *)ctsio);
8092
8093                 return (CTL_RETVAL_COMPLETE);
8094         }
8095
8096         param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
8097
8098         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
8099         res_key = scsi_8btou64(param->res_key.key);
8100         sa_res_key = scsi_8btou64(param->serv_act_res_key);
8101
8102         /*
8103          * Validate the reservation key here except for SPRO_REG_IGNO
8104          * This must be done for all other service actions
8105          */
8106         if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
8107                 mtx_lock(&lun->lun_lock);
8108                 if ((key = ctl_get_prkey(lun, residx)) != 0) {
8109                         if (res_key != key) {
8110                                 /*
8111                                  * The current key passed in doesn't match
8112                                  * the one the initiator previously
8113                                  * registered.
8114                                  */
8115                                 mtx_unlock(&lun->lun_lock);
8116                                 free(ctsio->kern_data_ptr, M_CTL);
8117                                 ctl_set_reservation_conflict(ctsio);
8118                                 ctl_done((union ctl_io *)ctsio);
8119                                 return (CTL_RETVAL_COMPLETE);
8120                         }
8121                 } else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
8122                         /*
8123                          * We are not registered
8124                          */
8125                         mtx_unlock(&lun->lun_lock);
8126                         free(ctsio->kern_data_ptr, M_CTL);
8127                         ctl_set_reservation_conflict(ctsio);
8128                         ctl_done((union ctl_io *)ctsio);
8129                         return (CTL_RETVAL_COMPLETE);
8130                 } else if (res_key != 0) {
8131                         /*
8132                          * We are not registered and trying to register but
8133                          * the register key isn't zero.
8134                          */
8135                         mtx_unlock(&lun->lun_lock);
8136                         free(ctsio->kern_data_ptr, M_CTL);
8137                         ctl_set_reservation_conflict(ctsio);
8138                         ctl_done((union ctl_io *)ctsio);
8139                         return (CTL_RETVAL_COMPLETE);
8140                 }
8141                 mtx_unlock(&lun->lun_lock);
8142         }
8143
8144         switch (cdb->action & SPRO_ACTION_MASK) {
8145         case SPRO_REGISTER:
8146         case SPRO_REG_IGNO: {
8147
8148 #if 0
8149                 printf("Registration received\n");
8150 #endif
8151
8152                 /*
8153                  * We don't support any of these options, as we report in
8154                  * the read capabilities request (see
8155                  * ctl_persistent_reserve_in(), above).
8156                  */
8157                 if ((param->flags & SPR_SPEC_I_PT)
8158                  || (param->flags & SPR_ALL_TG_PT)
8159                  || (param->flags & SPR_APTPL)) {
8160                         int bit_ptr;
8161
8162                         if (param->flags & SPR_APTPL)
8163                                 bit_ptr = 0;
8164                         else if (param->flags & SPR_ALL_TG_PT)
8165                                 bit_ptr = 2;
8166                         else /* SPR_SPEC_I_PT */
8167                                 bit_ptr = 3;
8168
8169                         free(ctsio->kern_data_ptr, M_CTL);
8170                         ctl_set_invalid_field(ctsio,
8171                                               /*sks_valid*/ 1,
8172                                               /*command*/ 0,
8173                                               /*field*/ 20,
8174                                               /*bit_valid*/ 1,
8175                                               /*bit*/ bit_ptr);
8176                         ctl_done((union ctl_io *)ctsio);
8177                         return (CTL_RETVAL_COMPLETE);
8178                 }
8179
8180                 mtx_lock(&lun->lun_lock);
8181
8182                 /*
8183                  * The initiator wants to clear the
8184                  * key/unregister.
8185                  */
8186                 if (sa_res_key == 0) {
8187                         if ((res_key == 0
8188                           && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
8189                          || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
8190                           && ctl_get_prkey(lun, residx) == 0)) {
8191                                 mtx_unlock(&lun->lun_lock);
8192                                 goto done;
8193                         }
8194
8195                         ctl_clr_prkey(lun, residx);
8196                         lun->pr_key_count--;
8197
8198                         if (residx == lun->pr_res_idx) {
8199                                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8200                                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8201
8202                                 if ((lun->res_type == SPR_TYPE_WR_EX_RO
8203                                   || lun->res_type == SPR_TYPE_EX_AC_RO)
8204                                  && lun->pr_key_count) {
8205                                         /*
8206                                          * If the reservation is a registrants
8207                                          * only type we need to generate a UA
8208                                          * for other registered inits.  The
8209                                          * sense code should be RESERVATIONS
8210                                          * RELEASED
8211                                          */
8212
8213                                         for (i = softc->init_min; i < softc->init_max; i++){
8214                                                 if (ctl_get_prkey(lun, i) == 0)
8215                                                         continue;
8216                                                 ctl_est_ua(lun, i,
8217                                                     CTL_UA_RES_RELEASE);
8218                                         }
8219                                 }
8220                                 lun->res_type = 0;
8221                         } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8222                                 if (lun->pr_key_count==0) {
8223                                         lun->flags &= ~CTL_LUN_PR_RESERVED;
8224                                         lun->res_type = 0;
8225                                         lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8226                                 }
8227                         }
8228                         lun->PRGeneration++;
8229                         mtx_unlock(&lun->lun_lock);
8230
8231                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8232                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8233                         persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
8234                         persis_io.pr.pr_info.residx = residx;
8235                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8236                             sizeof(persis_io.pr), M_WAITOK);
8237                 } else /* sa_res_key != 0 */ {
8238
8239                         /*
8240                          * If we aren't registered currently then increment
8241                          * the key count and set the registered flag.
8242                          */
8243                         ctl_alloc_prkey(lun, residx);
8244                         if (ctl_get_prkey(lun, residx) == 0)
8245                                 lun->pr_key_count++;
8246                         ctl_set_prkey(lun, residx, sa_res_key);
8247                         lun->PRGeneration++;
8248                         mtx_unlock(&lun->lun_lock);
8249
8250                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8251                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8252                         persis_io.pr.pr_info.action = CTL_PR_REG_KEY;
8253                         persis_io.pr.pr_info.residx = residx;
8254                         memcpy(persis_io.pr.pr_info.sa_res_key,
8255                                param->serv_act_res_key,
8256                                sizeof(param->serv_act_res_key));
8257                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8258                             sizeof(persis_io.pr), M_WAITOK);
8259                 }
8260
8261                 break;
8262         }
8263         case SPRO_RESERVE:
8264 #if 0
8265                 printf("Reserve executed type %d\n", type);
8266 #endif
8267                 mtx_lock(&lun->lun_lock);
8268                 if (lun->flags & CTL_LUN_PR_RESERVED) {
8269                         /*
8270                          * if this isn't the reservation holder and it's
8271                          * not a "all registrants" type or if the type is
8272                          * different then we have a conflict
8273                          */
8274                         if ((lun->pr_res_idx != residx
8275                           && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8276                          || lun->res_type != type) {
8277                                 mtx_unlock(&lun->lun_lock);
8278                                 free(ctsio->kern_data_ptr, M_CTL);
8279                                 ctl_set_reservation_conflict(ctsio);
8280                                 ctl_done((union ctl_io *)ctsio);
8281                                 return (CTL_RETVAL_COMPLETE);
8282                         }
8283                         mtx_unlock(&lun->lun_lock);
8284                 } else /* create a reservation */ {
8285                         /*
8286                          * If it's not an "all registrants" type record
8287                          * reservation holder
8288                          */
8289                         if (type != SPR_TYPE_WR_EX_AR
8290                          && type != SPR_TYPE_EX_AC_AR)
8291                                 lun->pr_res_idx = residx; /* Res holder */
8292                         else
8293                                 lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8294
8295                         lun->flags |= CTL_LUN_PR_RESERVED;
8296                         lun->res_type = type;
8297
8298                         mtx_unlock(&lun->lun_lock);
8299
8300                         /* send msg to other side */
8301                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8302                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8303                         persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8304                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8305                         persis_io.pr.pr_info.res_type = type;
8306                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8307                             sizeof(persis_io.pr), M_WAITOK);
8308                 }
8309                 break;
8310
8311         case SPRO_RELEASE:
8312                 mtx_lock(&lun->lun_lock);
8313                 if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8314                         /* No reservation exists return good status */
8315                         mtx_unlock(&lun->lun_lock);
8316                         goto done;
8317                 }
8318                 /*
8319                  * Is this nexus a reservation holder?
8320                  */
8321                 if (lun->pr_res_idx != residx
8322                  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8323                         /*
8324                          * not a res holder return good status but
8325                          * do nothing
8326                          */
8327                         mtx_unlock(&lun->lun_lock);
8328                         goto done;
8329                 }
8330
8331                 if (lun->res_type != type) {
8332                         mtx_unlock(&lun->lun_lock);
8333                         free(ctsio->kern_data_ptr, M_CTL);
8334                         ctl_set_illegal_pr_release(ctsio);
8335                         ctl_done((union ctl_io *)ctsio);
8336                         return (CTL_RETVAL_COMPLETE);
8337                 }
8338
8339                 /* okay to release */
8340                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8341                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8342                 lun->res_type = 0;
8343
8344                 /*
8345                  * if this isn't an exclusive access
8346                  * res generate UA for all other
8347                  * registrants.
8348                  */
8349                 if (type != SPR_TYPE_EX_AC
8350                  && type != SPR_TYPE_WR_EX) {
8351                         for (i = softc->init_min; i < softc->init_max; i++) {
8352                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
8353                                         continue;
8354                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8355                         }
8356                 }
8357                 mtx_unlock(&lun->lun_lock);
8358
8359                 /* Send msg to other side */
8360                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8361                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8362                 persis_io.pr.pr_info.action = CTL_PR_RELEASE;
8363                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8364                      sizeof(persis_io.pr), M_WAITOK);
8365                 break;
8366
8367         case SPRO_CLEAR:
8368                 /* send msg to other side */
8369
8370                 mtx_lock(&lun->lun_lock);
8371                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8372                 lun->res_type = 0;
8373                 lun->pr_key_count = 0;
8374                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8375
8376                 ctl_clr_prkey(lun, residx);
8377                 for (i = 0; i < CTL_MAX_INITIATORS; i++)
8378                         if (ctl_get_prkey(lun, i) != 0) {
8379                                 ctl_clr_prkey(lun, i);
8380                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8381                         }
8382                 lun->PRGeneration++;
8383                 mtx_unlock(&lun->lun_lock);
8384
8385                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8386                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8387                 persis_io.pr.pr_info.action = CTL_PR_CLEAR;
8388                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8389                      sizeof(persis_io.pr), M_WAITOK);
8390                 break;
8391
8392         case SPRO_PREEMPT:
8393         case SPRO_PRE_ABO: {
8394                 int nretval;
8395
8396                 nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8397                                           residx, ctsio, cdb, param);
8398                 if (nretval != 0)
8399                         return (CTL_RETVAL_COMPLETE);
8400                 break;
8401         }
8402         default:
8403                 panic("Invalid PR type %x", cdb->action);
8404         }
8405
8406 done:
8407         free(ctsio->kern_data_ptr, M_CTL);
8408         ctl_set_success(ctsio);
8409         ctl_done((union ctl_io *)ctsio);
8410
8411         return (retval);
8412 }
8413
8414 /*
8415  * This routine is for handling a message from the other SC pertaining to
8416  * persistent reserve out. All the error checking will have been done
8417  * so only perorming the action need be done here to keep the two
8418  * in sync.
8419  */
8420 static void
8421 ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg)
8422 {
8423         struct ctl_softc *softc = control_softc;
8424         struct ctl_lun *lun;
8425         int i;
8426         uint32_t residx, targ_lun;
8427
8428         targ_lun = msg->hdr.nexus.targ_mapped_lun;
8429         mtx_lock(&softc->ctl_lock);
8430         if ((targ_lun >= CTL_MAX_LUNS) ||
8431             ((lun = softc->ctl_luns[targ_lun]) == NULL)) {
8432                 mtx_unlock(&softc->ctl_lock);
8433                 return;
8434         }
8435         mtx_lock(&lun->lun_lock);
8436         mtx_unlock(&softc->ctl_lock);
8437         if (lun->flags & CTL_LUN_DISABLED) {
8438                 mtx_unlock(&lun->lun_lock);
8439                 return;
8440         }
8441         residx = ctl_get_initindex(&msg->hdr.nexus);
8442         switch(msg->pr.pr_info.action) {
8443         case CTL_PR_REG_KEY:
8444                 ctl_alloc_prkey(lun, msg->pr.pr_info.residx);
8445                 if (ctl_get_prkey(lun, msg->pr.pr_info.residx) == 0)
8446                         lun->pr_key_count++;
8447                 ctl_set_prkey(lun, msg->pr.pr_info.residx,
8448                     scsi_8btou64(msg->pr.pr_info.sa_res_key));
8449                 lun->PRGeneration++;
8450                 break;
8451
8452         case CTL_PR_UNREG_KEY:
8453                 ctl_clr_prkey(lun, msg->pr.pr_info.residx);
8454                 lun->pr_key_count--;
8455
8456                 /* XXX Need to see if the reservation has been released */
8457                 /* if so do we need to generate UA? */
8458                 if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8459                         lun->flags &= ~CTL_LUN_PR_RESERVED;
8460                         lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8461
8462                         if ((lun->res_type == SPR_TYPE_WR_EX_RO
8463                           || lun->res_type == SPR_TYPE_EX_AC_RO)
8464                          && lun->pr_key_count) {
8465                                 /*
8466                                  * If the reservation is a registrants
8467                                  * only type we need to generate a UA
8468                                  * for other registered inits.  The
8469                                  * sense code should be RESERVATIONS
8470                                  * RELEASED
8471                                  */
8472
8473                                 for (i = softc->init_min; i < softc->init_max; i++) {
8474                                         if (ctl_get_prkey(lun, i) == 0)
8475                                                 continue;
8476
8477                                         ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8478                                 }
8479                         }
8480                         lun->res_type = 0;
8481                 } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8482                         if (lun->pr_key_count==0) {
8483                                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8484                                 lun->res_type = 0;
8485                                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8486                         }
8487                 }
8488                 lun->PRGeneration++;
8489                 break;
8490
8491         case CTL_PR_RESERVE:
8492                 lun->flags |= CTL_LUN_PR_RESERVED;
8493                 lun->res_type = msg->pr.pr_info.res_type;
8494                 lun->pr_res_idx = msg->pr.pr_info.residx;
8495
8496                 break;
8497
8498         case CTL_PR_RELEASE:
8499                 /*
8500                  * if this isn't an exclusive access res generate UA for all
8501                  * other registrants.
8502                  */
8503                 if (lun->res_type != SPR_TYPE_EX_AC
8504                  && lun->res_type != SPR_TYPE_WR_EX) {
8505                         for (i = softc->init_min; i < softc->init_max; i++)
8506                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
8507                                         continue;
8508                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8509                 }
8510
8511                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8512                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8513                 lun->res_type = 0;
8514                 break;
8515
8516         case CTL_PR_PREEMPT:
8517                 ctl_pro_preempt_other(lun, msg);
8518                 break;
8519         case CTL_PR_CLEAR:
8520                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8521                 lun->res_type = 0;
8522                 lun->pr_key_count = 0;
8523                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8524
8525                 for (i=0; i < CTL_MAX_INITIATORS; i++) {
8526                         if (ctl_get_prkey(lun, i) == 0)
8527                                 continue;
8528                         ctl_clr_prkey(lun, i);
8529                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8530                 }
8531                 lun->PRGeneration++;
8532                 break;
8533         }
8534
8535         mtx_unlock(&lun->lun_lock);
8536 }
8537
8538 int
8539 ctl_read_write(struct ctl_scsiio *ctsio)
8540 {
8541         struct ctl_lun *lun;
8542         struct ctl_lba_len_flags *lbalen;
8543         uint64_t lba;
8544         uint32_t num_blocks;
8545         int flags, retval;
8546         int isread;
8547
8548         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8549
8550         CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
8551
8552         flags = 0;
8553         isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
8554               || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
8555         switch (ctsio->cdb[0]) {
8556         case READ_6:
8557         case WRITE_6: {
8558                 struct scsi_rw_6 *cdb;
8559
8560                 cdb = (struct scsi_rw_6 *)ctsio->cdb;
8561
8562                 lba = scsi_3btoul(cdb->addr);
8563                 /* only 5 bits are valid in the most significant address byte */
8564                 lba &= 0x1fffff;
8565                 num_blocks = cdb->length;
8566                 /*
8567                  * This is correct according to SBC-2.
8568                  */
8569                 if (num_blocks == 0)
8570                         num_blocks = 256;
8571                 break;
8572         }
8573         case READ_10:
8574         case WRITE_10: {
8575                 struct scsi_rw_10 *cdb;
8576
8577                 cdb = (struct scsi_rw_10 *)ctsio->cdb;
8578                 if (cdb->byte2 & SRW10_FUA)
8579                         flags |= CTL_LLF_FUA;
8580                 if (cdb->byte2 & SRW10_DPO)
8581                         flags |= CTL_LLF_DPO;
8582                 lba = scsi_4btoul(cdb->addr);
8583                 num_blocks = scsi_2btoul(cdb->length);
8584                 break;
8585         }
8586         case WRITE_VERIFY_10: {
8587                 struct scsi_write_verify_10 *cdb;
8588
8589                 cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
8590                 flags |= CTL_LLF_FUA;
8591                 if (cdb->byte2 & SWV_DPO)
8592                         flags |= CTL_LLF_DPO;
8593                 lba = scsi_4btoul(cdb->addr);
8594                 num_blocks = scsi_2btoul(cdb->length);
8595                 break;
8596         }
8597         case READ_12:
8598         case WRITE_12: {
8599                 struct scsi_rw_12 *cdb;
8600
8601                 cdb = (struct scsi_rw_12 *)ctsio->cdb;
8602                 if (cdb->byte2 & SRW12_FUA)
8603                         flags |= CTL_LLF_FUA;
8604                 if (cdb->byte2 & SRW12_DPO)
8605                         flags |= CTL_LLF_DPO;
8606                 lba = scsi_4btoul(cdb->addr);
8607                 num_blocks = scsi_4btoul(cdb->length);
8608                 break;
8609         }
8610         case WRITE_VERIFY_12: {
8611                 struct scsi_write_verify_12 *cdb;
8612
8613                 cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
8614                 flags |= CTL_LLF_FUA;
8615                 if (cdb->byte2 & SWV_DPO)
8616                         flags |= CTL_LLF_DPO;
8617                 lba = scsi_4btoul(cdb->addr);
8618                 num_blocks = scsi_4btoul(cdb->length);
8619                 break;
8620         }
8621         case READ_16:
8622         case WRITE_16: {
8623                 struct scsi_rw_16 *cdb;
8624
8625                 cdb = (struct scsi_rw_16 *)ctsio->cdb;
8626                 if (cdb->byte2 & SRW12_FUA)
8627                         flags |= CTL_LLF_FUA;
8628                 if (cdb->byte2 & SRW12_DPO)
8629                         flags |= CTL_LLF_DPO;
8630                 lba = scsi_8btou64(cdb->addr);
8631                 num_blocks = scsi_4btoul(cdb->length);
8632                 break;
8633         }
8634         case WRITE_ATOMIC_16: {
8635                 struct scsi_write_atomic_16 *cdb;
8636
8637                 if (lun->be_lun->atomicblock == 0) {
8638                         ctl_set_invalid_opcode(ctsio);
8639                         ctl_done((union ctl_io *)ctsio);
8640                         return (CTL_RETVAL_COMPLETE);
8641                 }
8642
8643                 cdb = (struct scsi_write_atomic_16 *)ctsio->cdb;
8644                 if (cdb->byte2 & SRW12_FUA)
8645                         flags |= CTL_LLF_FUA;
8646                 if (cdb->byte2 & SRW12_DPO)
8647                         flags |= CTL_LLF_DPO;
8648                 lba = scsi_8btou64(cdb->addr);
8649                 num_blocks = scsi_2btoul(cdb->length);
8650                 if (num_blocks > lun->be_lun->atomicblock) {
8651                         ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
8652                             /*command*/ 1, /*field*/ 12, /*bit_valid*/ 0,
8653                             /*bit*/ 0);
8654                         ctl_done((union ctl_io *)ctsio);
8655                         return (CTL_RETVAL_COMPLETE);
8656                 }
8657                 break;
8658         }
8659         case WRITE_VERIFY_16: {
8660                 struct scsi_write_verify_16 *cdb;
8661
8662                 cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
8663                 flags |= CTL_LLF_FUA;
8664                 if (cdb->byte2 & SWV_DPO)
8665                         flags |= CTL_LLF_DPO;
8666                 lba = scsi_8btou64(cdb->addr);
8667                 num_blocks = scsi_4btoul(cdb->length);
8668                 break;
8669         }
8670         default:
8671                 /*
8672                  * We got a command we don't support.  This shouldn't
8673                  * happen, commands should be filtered out above us.
8674                  */
8675                 ctl_set_invalid_opcode(ctsio);
8676                 ctl_done((union ctl_io *)ctsio);
8677
8678                 return (CTL_RETVAL_COMPLETE);
8679                 break; /* NOTREACHED */
8680         }
8681
8682         /*
8683          * The first check is to make sure we're in bounds, the second
8684          * check is to catch wrap-around problems.  If the lba + num blocks
8685          * is less than the lba, then we've wrapped around and the block
8686          * range is invalid anyway.
8687          */
8688         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8689          || ((lba + num_blocks) < lba)) {
8690                 ctl_set_lba_out_of_range(ctsio);
8691                 ctl_done((union ctl_io *)ctsio);
8692                 return (CTL_RETVAL_COMPLETE);
8693         }
8694
8695         /*
8696          * According to SBC-3, a transfer length of 0 is not an error.
8697          * Note that this cannot happen with WRITE(6) or READ(6), since 0
8698          * translates to 256 blocks for those commands.
8699          */
8700         if (num_blocks == 0) {
8701                 ctl_set_success(ctsio);
8702                 ctl_done((union ctl_io *)ctsio);
8703                 return (CTL_RETVAL_COMPLETE);
8704         }
8705
8706         /* Set FUA and/or DPO if caches are disabled. */
8707         if (isread) {
8708                 if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
8709                     SCP_RCD) != 0)
8710                         flags |= CTL_LLF_FUA | CTL_LLF_DPO;
8711         } else {
8712                 if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
8713                     SCP_WCE) == 0)
8714                         flags |= CTL_LLF_FUA;
8715         }
8716
8717         lbalen = (struct ctl_lba_len_flags *)
8718             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8719         lbalen->lba = lba;
8720         lbalen->len = num_blocks;
8721         lbalen->flags = (isread ? CTL_LLF_READ : CTL_LLF_WRITE) | flags;
8722
8723         ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
8724         ctsio->kern_rel_offset = 0;
8725
8726         CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
8727
8728         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8729         return (retval);
8730 }
8731
8732 static int
8733 ctl_cnw_cont(union ctl_io *io)
8734 {
8735         struct ctl_scsiio *ctsio;
8736         struct ctl_lun *lun;
8737         struct ctl_lba_len_flags *lbalen;
8738         int retval;
8739
8740         ctsio = &io->scsiio;
8741         ctsio->io_hdr.status = CTL_STATUS_NONE;
8742         ctsio->io_hdr.flags &= ~CTL_FLAG_IO_CONT;
8743         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8744         lbalen = (struct ctl_lba_len_flags *)
8745             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8746         lbalen->flags &= ~CTL_LLF_COMPARE;
8747         lbalen->flags |= CTL_LLF_WRITE;
8748
8749         CTL_DEBUG_PRINT(("ctl_cnw_cont: calling data_submit()\n"));
8750         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8751         return (retval);
8752 }
8753
8754 int
8755 ctl_cnw(struct ctl_scsiio *ctsio)
8756 {
8757         struct ctl_lun *lun;
8758         struct ctl_lba_len_flags *lbalen;
8759         uint64_t lba;
8760         uint32_t num_blocks;
8761         int flags, retval;
8762
8763         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8764
8765         CTL_DEBUG_PRINT(("ctl_cnw: command: %#x\n", ctsio->cdb[0]));
8766
8767         flags = 0;
8768         switch (ctsio->cdb[0]) {
8769         case COMPARE_AND_WRITE: {
8770                 struct scsi_compare_and_write *cdb;
8771
8772                 cdb = (struct scsi_compare_and_write *)ctsio->cdb;
8773                 if (cdb->byte2 & SRW10_FUA)
8774                         flags |= CTL_LLF_FUA;
8775                 if (cdb->byte2 & SRW10_DPO)
8776                         flags |= CTL_LLF_DPO;
8777                 lba = scsi_8btou64(cdb->addr);
8778                 num_blocks = cdb->length;
8779                 break;
8780         }
8781         default:
8782                 /*
8783                  * We got a command we don't support.  This shouldn't
8784                  * happen, commands should be filtered out above us.
8785                  */
8786                 ctl_set_invalid_opcode(ctsio);
8787                 ctl_done((union ctl_io *)ctsio);
8788
8789                 return (CTL_RETVAL_COMPLETE);
8790                 break; /* NOTREACHED */
8791         }
8792
8793         /*
8794          * The first check is to make sure we're in bounds, the second
8795          * check is to catch wrap-around problems.  If the lba + num blocks
8796          * is less than the lba, then we've wrapped around and the block
8797          * range is invalid anyway.
8798          */
8799         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8800          || ((lba + num_blocks) < lba)) {
8801                 ctl_set_lba_out_of_range(ctsio);
8802                 ctl_done((union ctl_io *)ctsio);
8803                 return (CTL_RETVAL_COMPLETE);
8804         }
8805
8806         /*
8807          * According to SBC-3, a transfer length of 0 is not an error.
8808          */
8809         if (num_blocks == 0) {
8810                 ctl_set_success(ctsio);
8811                 ctl_done((union ctl_io *)ctsio);
8812                 return (CTL_RETVAL_COMPLETE);
8813         }
8814
8815         /* Set FUA if write cache is disabled. */
8816         if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
8817             SCP_WCE) == 0)
8818                 flags |= CTL_LLF_FUA;
8819
8820         ctsio->kern_total_len = 2 * num_blocks * lun->be_lun->blocksize;
8821         ctsio->kern_rel_offset = 0;
8822
8823         /*
8824          * Set the IO_CONT flag, so that if this I/O gets passed to
8825          * ctl_data_submit_done(), it'll get passed back to
8826          * ctl_ctl_cnw_cont() for further processing.
8827          */
8828         ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
8829         ctsio->io_cont = ctl_cnw_cont;
8830
8831         lbalen = (struct ctl_lba_len_flags *)
8832             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8833         lbalen->lba = lba;
8834         lbalen->len = num_blocks;
8835         lbalen->flags = CTL_LLF_COMPARE | flags;
8836
8837         CTL_DEBUG_PRINT(("ctl_cnw: calling data_submit()\n"));
8838         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8839         return (retval);
8840 }
8841
8842 int
8843 ctl_verify(struct ctl_scsiio *ctsio)
8844 {
8845         struct ctl_lun *lun;
8846         struct ctl_lba_len_flags *lbalen;
8847         uint64_t lba;
8848         uint32_t num_blocks;
8849         int bytchk, flags;
8850         int retval;
8851
8852         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8853
8854         CTL_DEBUG_PRINT(("ctl_verify: command: %#x\n", ctsio->cdb[0]));
8855
8856         bytchk = 0;
8857         flags = CTL_LLF_FUA;
8858         switch (ctsio->cdb[0]) {
8859         case VERIFY_10: {
8860                 struct scsi_verify_10 *cdb;
8861
8862                 cdb = (struct scsi_verify_10 *)ctsio->cdb;
8863                 if (cdb->byte2 & SVFY_BYTCHK)
8864                         bytchk = 1;
8865                 if (cdb->byte2 & SVFY_DPO)
8866                         flags |= CTL_LLF_DPO;
8867                 lba = scsi_4btoul(cdb->addr);
8868                 num_blocks = scsi_2btoul(cdb->length);
8869                 break;
8870         }
8871         case VERIFY_12: {
8872                 struct scsi_verify_12 *cdb;
8873
8874                 cdb = (struct scsi_verify_12 *)ctsio->cdb;
8875                 if (cdb->byte2 & SVFY_BYTCHK)
8876                         bytchk = 1;
8877                 if (cdb->byte2 & SVFY_DPO)
8878                         flags |= CTL_LLF_DPO;
8879                 lba = scsi_4btoul(cdb->addr);
8880                 num_blocks = scsi_4btoul(cdb->length);
8881                 break;
8882         }
8883         case VERIFY_16: {
8884                 struct scsi_rw_16 *cdb;
8885
8886                 cdb = (struct scsi_rw_16 *)ctsio->cdb;
8887                 if (cdb->byte2 & SVFY_BYTCHK)
8888                         bytchk = 1;
8889                 if (cdb->byte2 & SVFY_DPO)
8890                         flags |= CTL_LLF_DPO;
8891                 lba = scsi_8btou64(cdb->addr);
8892                 num_blocks = scsi_4btoul(cdb->length);
8893                 break;
8894         }
8895         default:
8896                 /*
8897                  * We got a command we don't support.  This shouldn't
8898                  * happen, commands should be filtered out above us.
8899                  */
8900                 ctl_set_invalid_opcode(ctsio);
8901                 ctl_done((union ctl_io *)ctsio);
8902                 return (CTL_RETVAL_COMPLETE);
8903         }
8904
8905         /*
8906          * The first check is to make sure we're in bounds, the second
8907          * check is to catch wrap-around problems.  If the lba + num blocks
8908          * is less than the lba, then we've wrapped around and the block
8909          * range is invalid anyway.
8910          */
8911         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8912          || ((lba + num_blocks) < lba)) {
8913                 ctl_set_lba_out_of_range(ctsio);
8914                 ctl_done((union ctl_io *)ctsio);
8915                 return (CTL_RETVAL_COMPLETE);
8916         }
8917
8918         /*
8919          * According to SBC-3, a transfer length of 0 is not an error.
8920          */
8921         if (num_blocks == 0) {
8922                 ctl_set_success(ctsio);
8923                 ctl_done((union ctl_io *)ctsio);
8924                 return (CTL_RETVAL_COMPLETE);
8925         }
8926
8927         lbalen = (struct ctl_lba_len_flags *)
8928             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8929         lbalen->lba = lba;
8930         lbalen->len = num_blocks;
8931         if (bytchk) {
8932                 lbalen->flags = CTL_LLF_COMPARE | flags;
8933                 ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
8934         } else {
8935                 lbalen->flags = CTL_LLF_VERIFY | flags;
8936                 ctsio->kern_total_len = 0;
8937         }
8938         ctsio->kern_rel_offset = 0;
8939
8940         CTL_DEBUG_PRINT(("ctl_verify: calling data_submit()\n"));
8941         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8942         return (retval);
8943 }
8944
8945 int
8946 ctl_report_luns(struct ctl_scsiio *ctsio)
8947 {
8948         struct ctl_softc *softc;
8949         struct scsi_report_luns *cdb;
8950         struct scsi_report_luns_data *lun_data;
8951         struct ctl_lun *lun, *request_lun;
8952         struct ctl_port *port;
8953         int num_luns, retval;
8954         uint32_t alloc_len, lun_datalen;
8955         int num_filled;
8956         uint32_t initidx, targ_lun_id, lun_id;
8957
8958         retval = CTL_RETVAL_COMPLETE;
8959         cdb = (struct scsi_report_luns *)ctsio->cdb;
8960         port = ctl_io_port(&ctsio->io_hdr);
8961         softc = port->ctl_softc;
8962
8963         CTL_DEBUG_PRINT(("ctl_report_luns\n"));
8964
8965         mtx_lock(&softc->ctl_lock);
8966         num_luns = 0;
8967         for (targ_lun_id = 0; targ_lun_id < CTL_MAX_LUNS; targ_lun_id++) {
8968                 if (ctl_lun_map_from_port(port, targ_lun_id) < CTL_MAX_LUNS)
8969                         num_luns++;
8970         }
8971         mtx_unlock(&softc->ctl_lock);
8972
8973         switch (cdb->select_report) {
8974         case RPL_REPORT_DEFAULT:
8975         case RPL_REPORT_ALL:
8976         case RPL_REPORT_NONSUBSID:
8977                 break;
8978         case RPL_REPORT_WELLKNOWN:
8979         case RPL_REPORT_ADMIN:
8980         case RPL_REPORT_CONGLOM:
8981                 num_luns = 0;
8982                 break;
8983         default:
8984                 ctl_set_invalid_field(ctsio,
8985                                       /*sks_valid*/ 1,
8986                                       /*command*/ 1,
8987                                       /*field*/ 2,
8988                                       /*bit_valid*/ 0,
8989                                       /*bit*/ 0);
8990                 ctl_done((union ctl_io *)ctsio);
8991                 return (retval);
8992                 break; /* NOTREACHED */
8993         }
8994
8995         alloc_len = scsi_4btoul(cdb->length);
8996         /*
8997          * The initiator has to allocate at least 16 bytes for this request,
8998          * so he can at least get the header and the first LUN.  Otherwise
8999          * we reject the request (per SPC-3 rev 14, section 6.21).
9000          */
9001         if (alloc_len < (sizeof(struct scsi_report_luns_data) +
9002             sizeof(struct scsi_report_luns_lundata))) {
9003                 ctl_set_invalid_field(ctsio,
9004                                       /*sks_valid*/ 1,
9005                                       /*command*/ 1,
9006                                       /*field*/ 6,
9007                                       /*bit_valid*/ 0,
9008                                       /*bit*/ 0);
9009                 ctl_done((union ctl_io *)ctsio);
9010                 return (retval);
9011         }
9012
9013         request_lun = (struct ctl_lun *)
9014                 ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9015
9016         lun_datalen = sizeof(*lun_data) +
9017                 (num_luns * sizeof(struct scsi_report_luns_lundata));
9018
9019         ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO);
9020         lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
9021         ctsio->kern_sg_entries = 0;
9022
9023         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9024
9025         mtx_lock(&softc->ctl_lock);
9026         for (targ_lun_id = 0, num_filled = 0; targ_lun_id < CTL_MAX_LUNS && num_filled < num_luns; targ_lun_id++) {
9027                 lun_id = ctl_lun_map_from_port(port, targ_lun_id);
9028                 if (lun_id >= CTL_MAX_LUNS)
9029                         continue;
9030                 lun = softc->ctl_luns[lun_id];
9031                 if (lun == NULL)
9032                         continue;
9033
9034                 if (targ_lun_id <= 0xff) {
9035                         /*
9036                          * Peripheral addressing method, bus number 0.
9037                          */
9038                         lun_data->luns[num_filled].lundata[0] =
9039                                 RPL_LUNDATA_ATYP_PERIPH;
9040                         lun_data->luns[num_filled].lundata[1] = targ_lun_id;
9041                         num_filled++;
9042                 } else if (targ_lun_id <= 0x3fff) {
9043                         /*
9044                          * Flat addressing method.
9045                          */
9046                         lun_data->luns[num_filled].lundata[0] =
9047                                 RPL_LUNDATA_ATYP_FLAT | (targ_lun_id >> 8);
9048                         lun_data->luns[num_filled].lundata[1] =
9049                                 (targ_lun_id & 0xff);
9050                         num_filled++;
9051                 } else if (targ_lun_id <= 0xffffff) {
9052                         /*
9053                          * Extended flat addressing method.
9054                          */
9055                         lun_data->luns[num_filled].lundata[0] =
9056                             RPL_LUNDATA_ATYP_EXTLUN | 0x12;
9057                         scsi_ulto3b(targ_lun_id,
9058                             &lun_data->luns[num_filled].lundata[1]);
9059                         num_filled++;
9060                 } else {
9061                         printf("ctl_report_luns: bogus LUN number %jd, "
9062                                "skipping\n", (intmax_t)targ_lun_id);
9063                 }
9064                 /*
9065                  * According to SPC-3, rev 14 section 6.21:
9066                  *
9067                  * "The execution of a REPORT LUNS command to any valid and
9068                  * installed logical unit shall clear the REPORTED LUNS DATA
9069                  * HAS CHANGED unit attention condition for all logical
9070                  * units of that target with respect to the requesting
9071                  * initiator. A valid and installed logical unit is one
9072                  * having a PERIPHERAL QUALIFIER of 000b in the standard
9073                  * INQUIRY data (see 6.4.2)."
9074                  *
9075                  * If request_lun is NULL, the LUN this report luns command
9076                  * was issued to is either disabled or doesn't exist. In that
9077                  * case, we shouldn't clear any pending lun change unit
9078                  * attention.
9079                  */
9080                 if (request_lun != NULL) {
9081                         mtx_lock(&lun->lun_lock);
9082                         ctl_clr_ua(lun, initidx, CTL_UA_LUN_CHANGE);
9083                         mtx_unlock(&lun->lun_lock);
9084                 }
9085         }
9086         mtx_unlock(&softc->ctl_lock);
9087
9088         /*
9089          * It's quite possible that we've returned fewer LUNs than we allocated
9090          * space for.  Trim it.
9091          */
9092         lun_datalen = sizeof(*lun_data) +
9093                 (num_filled * sizeof(struct scsi_report_luns_lundata));
9094
9095         if (lun_datalen < alloc_len) {
9096                 ctsio->residual = alloc_len - lun_datalen;
9097                 ctsio->kern_data_len = lun_datalen;
9098                 ctsio->kern_total_len = lun_datalen;
9099         } else {
9100                 ctsio->residual = 0;
9101                 ctsio->kern_data_len = alloc_len;
9102                 ctsio->kern_total_len = alloc_len;
9103         }
9104         ctsio->kern_data_resid = 0;
9105         ctsio->kern_rel_offset = 0;
9106         ctsio->kern_sg_entries = 0;
9107
9108         /*
9109          * We set this to the actual data length, regardless of how much
9110          * space we actually have to return results.  If the user looks at
9111          * this value, he'll know whether or not he allocated enough space
9112          * and reissue the command if necessary.  We don't support well
9113          * known logical units, so if the user asks for that, return none.
9114          */
9115         scsi_ulto4b(lun_datalen - 8, lun_data->length);
9116
9117         /*
9118          * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
9119          * this request.
9120          */
9121         ctl_set_success(ctsio);
9122         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9123         ctsio->be_move_done = ctl_config_move_done;
9124         ctl_datamove((union ctl_io *)ctsio);
9125         return (retval);
9126 }
9127
9128 int
9129 ctl_request_sense(struct ctl_scsiio *ctsio)
9130 {
9131         struct scsi_request_sense *cdb;
9132         struct scsi_sense_data *sense_ptr;
9133         struct ctl_softc *ctl_softc;
9134         struct ctl_lun *lun;
9135         uint32_t initidx;
9136         int have_error;
9137         scsi_sense_data_type sense_format;
9138         ctl_ua_type ua_type;
9139
9140         cdb = (struct scsi_request_sense *)ctsio->cdb;
9141
9142         ctl_softc = control_softc;
9143         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9144
9145         CTL_DEBUG_PRINT(("ctl_request_sense\n"));
9146
9147         /*
9148          * Determine which sense format the user wants.
9149          */
9150         if (cdb->byte2 & SRS_DESC)
9151                 sense_format = SSD_TYPE_DESC;
9152         else
9153                 sense_format = SSD_TYPE_FIXED;
9154
9155         ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
9156         sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
9157         ctsio->kern_sg_entries = 0;
9158
9159         /*
9160          * struct scsi_sense_data, which is currently set to 256 bytes, is
9161          * larger than the largest allowed value for the length field in the
9162          * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
9163          */
9164         ctsio->residual = 0;
9165         ctsio->kern_data_len = cdb->length;
9166         ctsio->kern_total_len = cdb->length;
9167
9168         ctsio->kern_data_resid = 0;
9169         ctsio->kern_rel_offset = 0;
9170         ctsio->kern_sg_entries = 0;
9171
9172         /*
9173          * If we don't have a LUN, we don't have any pending sense.
9174          */
9175         if (lun == NULL)
9176                 goto no_sense;
9177
9178         have_error = 0;
9179         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9180         /*
9181          * Check for pending sense, and then for pending unit attentions.
9182          * Pending sense gets returned first, then pending unit attentions.
9183          */
9184         mtx_lock(&lun->lun_lock);
9185 #ifdef CTL_WITH_CA
9186         if (ctl_is_set(lun->have_ca, initidx)) {
9187                 scsi_sense_data_type stored_format;
9188
9189                 /*
9190                  * Check to see which sense format was used for the stored
9191                  * sense data.
9192                  */
9193                 stored_format = scsi_sense_type(&lun->pending_sense[initidx]);
9194
9195                 /*
9196                  * If the user requested a different sense format than the
9197                  * one we stored, then we need to convert it to the other
9198                  * format.  If we're going from descriptor to fixed format
9199                  * sense data, we may lose things in translation, depending
9200                  * on what options were used.
9201                  *
9202                  * If the stored format is SSD_TYPE_NONE (i.e. invalid),
9203                  * for some reason we'll just copy it out as-is.
9204                  */
9205                 if ((stored_format == SSD_TYPE_FIXED)
9206                  && (sense_format == SSD_TYPE_DESC))
9207                         ctl_sense_to_desc((struct scsi_sense_data_fixed *)
9208                             &lun->pending_sense[initidx],
9209                             (struct scsi_sense_data_desc *)sense_ptr);
9210                 else if ((stored_format == SSD_TYPE_DESC)
9211                       && (sense_format == SSD_TYPE_FIXED))
9212                         ctl_sense_to_fixed((struct scsi_sense_data_desc *)
9213                             &lun->pending_sense[initidx],
9214                             (struct scsi_sense_data_fixed *)sense_ptr);
9215                 else
9216                         memcpy(sense_ptr, &lun->pending_sense[initidx],
9217                                MIN(sizeof(*sense_ptr),
9218                                sizeof(lun->pending_sense[initidx])));
9219
9220                 ctl_clear_mask(lun->have_ca, initidx);
9221                 have_error = 1;
9222         } else
9223 #endif
9224         {
9225                 ua_type = ctl_build_ua(lun, initidx, sense_ptr, sense_format);
9226                 if (ua_type != CTL_UA_NONE)
9227                         have_error = 1;
9228                 if (ua_type == CTL_UA_LUN_CHANGE) {
9229                         mtx_unlock(&lun->lun_lock);
9230                         mtx_lock(&ctl_softc->ctl_lock);
9231                         ctl_clr_ua_allluns(ctl_softc, initidx, ua_type);
9232                         mtx_unlock(&ctl_softc->ctl_lock);
9233                         mtx_lock(&lun->lun_lock);
9234                 }
9235
9236         }
9237         mtx_unlock(&lun->lun_lock);
9238
9239         /*
9240          * We already have a pending error, return it.
9241          */
9242         if (have_error != 0) {
9243                 /*
9244                  * We report the SCSI status as OK, since the status of the
9245                  * request sense command itself is OK.
9246                  * We report 0 for the sense length, because we aren't doing
9247                  * autosense in this case.  We're reporting sense as
9248                  * parameter data.
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 (CTL_RETVAL_COMPLETE);
9255         }
9256
9257 no_sense:
9258
9259         /*
9260          * No sense information to report, so we report that everything is
9261          * okay.
9262          */
9263         ctl_set_sense_data(sense_ptr,
9264                            lun,
9265                            sense_format,
9266                            /*current_error*/ 1,
9267                            /*sense_key*/ SSD_KEY_NO_SENSE,
9268                            /*asc*/ 0x00,
9269                            /*ascq*/ 0x00,
9270                            SSD_ELEM_NONE);
9271
9272         /*
9273          * We report 0 for the sense length, because we aren't doing
9274          * autosense in this case.  We're reporting sense as parameter data.
9275          */
9276         ctl_set_success(ctsio);
9277         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9278         ctsio->be_move_done = ctl_config_move_done;
9279         ctl_datamove((union ctl_io *)ctsio);
9280         return (CTL_RETVAL_COMPLETE);
9281 }
9282
9283 int
9284 ctl_tur(struct ctl_scsiio *ctsio)
9285 {
9286
9287         CTL_DEBUG_PRINT(("ctl_tur\n"));
9288
9289         ctl_set_success(ctsio);
9290         ctl_done((union ctl_io *)ctsio);
9291
9292         return (CTL_RETVAL_COMPLETE);
9293 }
9294
9295 /*
9296  * SCSI VPD page 0x00, the Supported VPD Pages page.
9297  */
9298 static int
9299 ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
9300 {
9301         struct scsi_vpd_supported_pages *pages;
9302         int sup_page_size;
9303         struct ctl_lun *lun;
9304         int p;
9305
9306         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9307
9308         sup_page_size = sizeof(struct scsi_vpd_supported_pages) *
9309             SCSI_EVPD_NUM_SUPPORTED_PAGES;
9310         ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO);
9311         pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
9312         ctsio->kern_sg_entries = 0;
9313
9314         if (sup_page_size < alloc_len) {
9315                 ctsio->residual = alloc_len - sup_page_size;
9316                 ctsio->kern_data_len = sup_page_size;
9317                 ctsio->kern_total_len = sup_page_size;
9318         } else {
9319                 ctsio->residual = 0;
9320                 ctsio->kern_data_len = alloc_len;
9321                 ctsio->kern_total_len = alloc_len;
9322         }
9323         ctsio->kern_data_resid = 0;
9324         ctsio->kern_rel_offset = 0;
9325         ctsio->kern_sg_entries = 0;
9326
9327         /*
9328          * The control device is always connected.  The disk device, on the
9329          * other hand, may not be online all the time.  Need to change this
9330          * to figure out whether the disk device is actually online or not.
9331          */
9332         if (lun != NULL)
9333                 pages->device = (SID_QUAL_LU_CONNECTED << 5) |
9334                                 lun->be_lun->lun_type;
9335         else
9336                 pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9337
9338         p = 0;
9339         /* Supported VPD pages */
9340         pages->page_list[p++] = SVPD_SUPPORTED_PAGES;
9341         /* Serial Number */
9342         pages->page_list[p++] = SVPD_UNIT_SERIAL_NUMBER;
9343         /* Device Identification */
9344         pages->page_list[p++] = SVPD_DEVICE_ID;
9345         /* Extended INQUIRY Data */
9346         pages->page_list[p++] = SVPD_EXTENDED_INQUIRY_DATA;
9347         /* Mode Page Policy */
9348         pages->page_list[p++] = SVPD_MODE_PAGE_POLICY;
9349         /* SCSI Ports */
9350         pages->page_list[p++] = SVPD_SCSI_PORTS;
9351         /* Third-party Copy */
9352         pages->page_list[p++] = SVPD_SCSI_TPC;
9353         if (lun != NULL && lun->be_lun->lun_type == T_DIRECT) {
9354                 /* Block limits */
9355                 pages->page_list[p++] = SVPD_BLOCK_LIMITS;
9356                 /* Block Device Characteristics */
9357                 pages->page_list[p++] = SVPD_BDC;
9358                 /* Logical Block Provisioning */
9359                 pages->page_list[p++] = SVPD_LBP;
9360         }
9361         pages->length = p;
9362
9363         ctl_set_success(ctsio);
9364         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9365         ctsio->be_move_done = ctl_config_move_done;
9366         ctl_datamove((union ctl_io *)ctsio);
9367         return (CTL_RETVAL_COMPLETE);
9368 }
9369
9370 /*
9371  * SCSI VPD page 0x80, the Unit Serial Number page.
9372  */
9373 static int
9374 ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
9375 {
9376         struct scsi_vpd_unit_serial_number *sn_ptr;
9377         struct ctl_lun *lun;
9378         int data_len;
9379
9380         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9381
9382         data_len = 4 + CTL_SN_LEN;
9383         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9384         sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
9385         if (data_len < alloc_len) {
9386                 ctsio->residual = alloc_len - data_len;
9387                 ctsio->kern_data_len = data_len;
9388                 ctsio->kern_total_len = data_len;
9389         } else {
9390                 ctsio->residual = 0;
9391                 ctsio->kern_data_len = alloc_len;
9392                 ctsio->kern_total_len = alloc_len;
9393         }
9394         ctsio->kern_data_resid = 0;
9395         ctsio->kern_rel_offset = 0;
9396         ctsio->kern_sg_entries = 0;
9397
9398         /*
9399          * The control device is always connected.  The disk device, on the
9400          * other hand, may not be online all the time.  Need to change this
9401          * to figure out whether the disk device is actually online or not.
9402          */
9403         if (lun != NULL)
9404                 sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9405                                   lun->be_lun->lun_type;
9406         else
9407                 sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9408
9409         sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9410         sn_ptr->length = CTL_SN_LEN;
9411         /*
9412          * If we don't have a LUN, we just leave the serial number as
9413          * all spaces.
9414          */
9415         if (lun != NULL) {
9416                 strncpy((char *)sn_ptr->serial_num,
9417                         (char *)lun->be_lun->serial_num, CTL_SN_LEN);
9418         } else
9419                 memset(sn_ptr->serial_num, 0x20, CTL_SN_LEN);
9420
9421         ctl_set_success(ctsio);
9422         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9423         ctsio->be_move_done = ctl_config_move_done;
9424         ctl_datamove((union ctl_io *)ctsio);
9425         return (CTL_RETVAL_COMPLETE);
9426 }
9427
9428
9429 /*
9430  * SCSI VPD page 0x86, the Extended INQUIRY Data page.
9431  */
9432 static int
9433 ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len)
9434 {
9435         struct scsi_vpd_extended_inquiry_data *eid_ptr;
9436         struct ctl_lun *lun;
9437         int data_len;
9438
9439         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9440
9441         data_len = sizeof(struct scsi_vpd_extended_inquiry_data);
9442         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9443         eid_ptr = (struct scsi_vpd_extended_inquiry_data *)ctsio->kern_data_ptr;
9444         ctsio->kern_sg_entries = 0;
9445
9446         if (data_len < alloc_len) {
9447                 ctsio->residual = alloc_len - data_len;
9448                 ctsio->kern_data_len = data_len;
9449                 ctsio->kern_total_len = data_len;
9450         } else {
9451                 ctsio->residual = 0;
9452                 ctsio->kern_data_len = alloc_len;
9453                 ctsio->kern_total_len = alloc_len;
9454         }
9455         ctsio->kern_data_resid = 0;
9456         ctsio->kern_rel_offset = 0;
9457         ctsio->kern_sg_entries = 0;
9458
9459         /*
9460          * The control device is always connected.  The disk device, on the
9461          * other hand, may not be online all the time.
9462          */
9463         if (lun != NULL)
9464                 eid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9465                                      lun->be_lun->lun_type;
9466         else
9467                 eid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9468         eid_ptr->page_code = SVPD_EXTENDED_INQUIRY_DATA;
9469         scsi_ulto2b(data_len - 4, eid_ptr->page_length);
9470         /*
9471          * We support head of queue, ordered and simple tags.
9472          */
9473         eid_ptr->flags2 = SVPD_EID_HEADSUP | SVPD_EID_ORDSUP | SVPD_EID_SIMPSUP;
9474         /*
9475          * Volatile cache supported.
9476          */
9477         eid_ptr->flags3 = SVPD_EID_V_SUP;
9478
9479         /*
9480          * This means that we clear the REPORTED LUNS DATA HAS CHANGED unit
9481          * attention for a particular IT nexus on all LUNs once we report
9482          * it to that nexus once.  This bit is required as of SPC-4.
9483          */
9484         eid_ptr->flags4 = SVPD_EID_LUICLT;
9485
9486         /*
9487          * XXX KDM in order to correctly answer this, we would need
9488          * information from the SIM to determine how much sense data it
9489          * can send.  So this would really be a path inquiry field, most
9490          * likely.  This can be set to a maximum of 252 according to SPC-4,
9491          * but the hardware may or may not be able to support that much.
9492          * 0 just means that the maximum sense data length is not reported.
9493          */
9494         eid_ptr->max_sense_length = 0;
9495
9496         ctl_set_success(ctsio);
9497         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9498         ctsio->be_move_done = ctl_config_move_done;
9499         ctl_datamove((union ctl_io *)ctsio);
9500         return (CTL_RETVAL_COMPLETE);
9501 }
9502
9503 static int
9504 ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len)
9505 {
9506         struct scsi_vpd_mode_page_policy *mpp_ptr;
9507         struct ctl_lun *lun;
9508         int data_len;
9509
9510         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9511
9512         data_len = sizeof(struct scsi_vpd_mode_page_policy) +
9513             sizeof(struct scsi_vpd_mode_page_policy_descr);
9514
9515         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9516         mpp_ptr = (struct scsi_vpd_mode_page_policy *)ctsio->kern_data_ptr;
9517         ctsio->kern_sg_entries = 0;
9518
9519         if (data_len < alloc_len) {
9520                 ctsio->residual = alloc_len - data_len;
9521                 ctsio->kern_data_len = data_len;
9522                 ctsio->kern_total_len = data_len;
9523         } else {
9524                 ctsio->residual = 0;
9525                 ctsio->kern_data_len = alloc_len;
9526                 ctsio->kern_total_len = alloc_len;
9527         }
9528         ctsio->kern_data_resid = 0;
9529         ctsio->kern_rel_offset = 0;
9530         ctsio->kern_sg_entries = 0;
9531
9532         /*
9533          * The control device is always connected.  The disk device, on the
9534          * other hand, may not be online all the time.
9535          */
9536         if (lun != NULL)
9537                 mpp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9538                                      lun->be_lun->lun_type;
9539         else
9540                 mpp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9541         mpp_ptr->page_code = SVPD_MODE_PAGE_POLICY;
9542         scsi_ulto2b(data_len - 4, mpp_ptr->page_length);
9543         mpp_ptr->descr[0].page_code = 0x3f;
9544         mpp_ptr->descr[0].subpage_code = 0xff;
9545         mpp_ptr->descr[0].policy = SVPD_MPP_SHARED;
9546
9547         ctl_set_success(ctsio);
9548         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9549         ctsio->be_move_done = ctl_config_move_done;
9550         ctl_datamove((union ctl_io *)ctsio);
9551         return (CTL_RETVAL_COMPLETE);
9552 }
9553
9554 /*
9555  * SCSI VPD page 0x83, the Device Identification page.
9556  */
9557 static int
9558 ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
9559 {
9560         struct scsi_vpd_device_id *devid_ptr;
9561         struct scsi_vpd_id_descriptor *desc;
9562         struct ctl_softc *softc;
9563         struct ctl_lun *lun;
9564         struct ctl_port *port;
9565         int data_len;
9566         uint8_t proto;
9567
9568         softc = control_softc;
9569
9570         port = ctl_io_port(&ctsio->io_hdr);
9571         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9572
9573         data_len = sizeof(struct scsi_vpd_device_id) +
9574             sizeof(struct scsi_vpd_id_descriptor) +
9575                 sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
9576             sizeof(struct scsi_vpd_id_descriptor) +
9577                 sizeof(struct scsi_vpd_id_trgt_port_grp_id);
9578         if (lun && lun->lun_devid)
9579                 data_len += lun->lun_devid->len;
9580         if (port && port->port_devid)
9581                 data_len += port->port_devid->len;
9582         if (port && port->target_devid)
9583                 data_len += port->target_devid->len;
9584
9585         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9586         devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
9587         ctsio->kern_sg_entries = 0;
9588
9589         if (data_len < alloc_len) {
9590                 ctsio->residual = alloc_len - data_len;
9591                 ctsio->kern_data_len = data_len;
9592                 ctsio->kern_total_len = data_len;
9593         } else {
9594                 ctsio->residual = 0;
9595                 ctsio->kern_data_len = alloc_len;
9596                 ctsio->kern_total_len = alloc_len;
9597         }
9598         ctsio->kern_data_resid = 0;
9599         ctsio->kern_rel_offset = 0;
9600         ctsio->kern_sg_entries = 0;
9601
9602         /*
9603          * The control device is always connected.  The disk device, on the
9604          * other hand, may not be online all the time.
9605          */
9606         if (lun != NULL)
9607                 devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9608                                      lun->be_lun->lun_type;
9609         else
9610                 devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9611         devid_ptr->page_code = SVPD_DEVICE_ID;
9612         scsi_ulto2b(data_len - 4, devid_ptr->length);
9613
9614         if (port && port->port_type == CTL_PORT_FC)
9615                 proto = SCSI_PROTO_FC << 4;
9616         else if (port && port->port_type == CTL_PORT_ISCSI)
9617                 proto = SCSI_PROTO_ISCSI << 4;
9618         else
9619                 proto = SCSI_PROTO_SPI << 4;
9620         desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
9621
9622         /*
9623          * We're using a LUN association here.  i.e., this device ID is a
9624          * per-LUN identifier.
9625          */
9626         if (lun && lun->lun_devid) {
9627                 memcpy(desc, lun->lun_devid->data, lun->lun_devid->len);
9628                 desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9629                     lun->lun_devid->len);
9630         }
9631
9632         /*
9633          * This is for the WWPN which is a port association.
9634          */
9635         if (port && port->port_devid) {
9636                 memcpy(desc, port->port_devid->data, port->port_devid->len);
9637                 desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9638                     port->port_devid->len);
9639         }
9640
9641         /*
9642          * This is for the Relative Target Port(type 4h) identifier
9643          */
9644         desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9645         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9646             SVPD_ID_TYPE_RELTARG;
9647         desc->length = 4;
9648         scsi_ulto2b(ctsio->io_hdr.nexus.targ_port, &desc->identifier[2]);
9649         desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9650             sizeof(struct scsi_vpd_id_rel_trgt_port_id));
9651
9652         /*
9653          * This is for the Target Port Group(type 5h) identifier
9654          */
9655         desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9656         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9657             SVPD_ID_TYPE_TPORTGRP;
9658         desc->length = 4;
9659         scsi_ulto2b(ctsio->io_hdr.nexus.targ_port / softc->port_cnt + 1,
9660             &desc->identifier[2]);
9661         desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9662             sizeof(struct scsi_vpd_id_trgt_port_grp_id));
9663
9664         /*
9665          * This is for the Target identifier
9666          */
9667         if (port && port->target_devid) {
9668                 memcpy(desc, port->target_devid->data, port->target_devid->len);
9669         }
9670
9671         ctl_set_success(ctsio);
9672         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9673         ctsio->be_move_done = ctl_config_move_done;
9674         ctl_datamove((union ctl_io *)ctsio);
9675         return (CTL_RETVAL_COMPLETE);
9676 }
9677
9678 static int
9679 ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio, int alloc_len)
9680 {
9681         struct ctl_softc *softc = control_softc;
9682         struct scsi_vpd_scsi_ports *sp;
9683         struct scsi_vpd_port_designation *pd;
9684         struct scsi_vpd_port_designation_cont *pdc;
9685         struct ctl_lun *lun;
9686         struct ctl_port *port;
9687         int data_len, num_target_ports, iid_len, id_len;
9688
9689         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9690
9691         num_target_ports = 0;
9692         iid_len = 0;
9693         id_len = 0;
9694         mtx_lock(&softc->ctl_lock);
9695         STAILQ_FOREACH(port, &softc->port_list, links) {
9696                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
9697                         continue;
9698                 if (lun != NULL &&
9699                     ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
9700                         continue;
9701                 num_target_ports++;
9702                 if (port->init_devid)
9703                         iid_len += port->init_devid->len;
9704                 if (port->port_devid)
9705                         id_len += port->port_devid->len;
9706         }
9707         mtx_unlock(&softc->ctl_lock);
9708
9709         data_len = sizeof(struct scsi_vpd_scsi_ports) +
9710             num_target_ports * (sizeof(struct scsi_vpd_port_designation) +
9711              sizeof(struct scsi_vpd_port_designation_cont)) + iid_len + id_len;
9712         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9713         sp = (struct scsi_vpd_scsi_ports *)ctsio->kern_data_ptr;
9714         ctsio->kern_sg_entries = 0;
9715
9716         if (data_len < alloc_len) {
9717                 ctsio->residual = alloc_len - data_len;
9718                 ctsio->kern_data_len = data_len;
9719                 ctsio->kern_total_len = data_len;
9720         } else {
9721                 ctsio->residual = 0;
9722                 ctsio->kern_data_len = alloc_len;
9723                 ctsio->kern_total_len = alloc_len;
9724         }
9725         ctsio->kern_data_resid = 0;
9726         ctsio->kern_rel_offset = 0;
9727         ctsio->kern_sg_entries = 0;
9728
9729         /*
9730          * The control device is always connected.  The disk device, on the
9731          * other hand, may not be online all the time.  Need to change this
9732          * to figure out whether the disk device is actually online or not.
9733          */
9734         if (lun != NULL)
9735                 sp->device = (SID_QUAL_LU_CONNECTED << 5) |
9736                                   lun->be_lun->lun_type;
9737         else
9738                 sp->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9739
9740         sp->page_code = SVPD_SCSI_PORTS;
9741         scsi_ulto2b(data_len - sizeof(struct scsi_vpd_scsi_ports),
9742             sp->page_length);
9743         pd = &sp->design[0];
9744
9745         mtx_lock(&softc->ctl_lock);
9746         STAILQ_FOREACH(port, &softc->port_list, links) {
9747                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
9748                         continue;
9749                 if (lun != NULL &&
9750                     ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
9751                         continue;
9752                 scsi_ulto2b(port->targ_port, pd->relative_port_id);
9753                 if (port->init_devid) {
9754                         iid_len = port->init_devid->len;
9755                         memcpy(pd->initiator_transportid,
9756                             port->init_devid->data, port->init_devid->len);
9757                 } else
9758                         iid_len = 0;
9759                 scsi_ulto2b(iid_len, pd->initiator_transportid_length);
9760                 pdc = (struct scsi_vpd_port_designation_cont *)
9761                     (&pd->initiator_transportid[iid_len]);
9762                 if (port->port_devid) {
9763                         id_len = port->port_devid->len;
9764                         memcpy(pdc->target_port_descriptors,
9765                             port->port_devid->data, port->port_devid->len);
9766                 } else
9767                         id_len = 0;
9768                 scsi_ulto2b(id_len, pdc->target_port_descriptors_length);
9769                 pd = (struct scsi_vpd_port_designation *)
9770                     ((uint8_t *)pdc->target_port_descriptors + id_len);
9771         }
9772         mtx_unlock(&softc->ctl_lock);
9773
9774         ctl_set_success(ctsio);
9775         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9776         ctsio->be_move_done = ctl_config_move_done;
9777         ctl_datamove((union ctl_io *)ctsio);
9778         return (CTL_RETVAL_COMPLETE);
9779 }
9780
9781 static int
9782 ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio, int alloc_len)
9783 {
9784         struct scsi_vpd_block_limits *bl_ptr;
9785         struct ctl_lun *lun;
9786
9787         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9788
9789         ctsio->kern_data_ptr = malloc(sizeof(*bl_ptr), M_CTL, M_WAITOK | M_ZERO);
9790         bl_ptr = (struct scsi_vpd_block_limits *)ctsio->kern_data_ptr;
9791         ctsio->kern_sg_entries = 0;
9792
9793         if (sizeof(*bl_ptr) < alloc_len) {
9794                 ctsio->residual = alloc_len - sizeof(*bl_ptr);
9795                 ctsio->kern_data_len = sizeof(*bl_ptr);
9796                 ctsio->kern_total_len = sizeof(*bl_ptr);
9797         } else {
9798                 ctsio->residual = 0;
9799                 ctsio->kern_data_len = alloc_len;
9800                 ctsio->kern_total_len = alloc_len;
9801         }
9802         ctsio->kern_data_resid = 0;
9803         ctsio->kern_rel_offset = 0;
9804         ctsio->kern_sg_entries = 0;
9805
9806         /*
9807          * The control device is always connected.  The disk device, on the
9808          * other hand, may not be online all the time.  Need to change this
9809          * to figure out whether the disk device is actually online or not.
9810          */
9811         if (lun != NULL)
9812                 bl_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9813                                   lun->be_lun->lun_type;
9814         else
9815                 bl_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9816
9817         bl_ptr->page_code = SVPD_BLOCK_LIMITS;
9818         scsi_ulto2b(sizeof(*bl_ptr) - 4, bl_ptr->page_length);
9819         bl_ptr->max_cmp_write_len = 0xff;
9820         scsi_ulto4b(0xffffffff, bl_ptr->max_txfer_len);
9821         if (lun != NULL) {
9822                 scsi_ulto4b(lun->be_lun->opttxferlen, bl_ptr->opt_txfer_len);
9823                 if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
9824                         scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_lba_cnt);
9825                         scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_blk_cnt);
9826                         if (lun->be_lun->ublockexp != 0) {
9827                                 scsi_ulto4b((1 << lun->be_lun->ublockexp),
9828                                     bl_ptr->opt_unmap_grain);
9829                                 scsi_ulto4b(0x80000000 | lun->be_lun->ublockoff,
9830                                     bl_ptr->unmap_grain_align);
9831                         }
9832                 }
9833                 scsi_ulto4b(lun->be_lun->atomicblock,
9834                     bl_ptr->max_atomic_transfer_length);
9835                 scsi_ulto4b(0, bl_ptr->atomic_alignment);
9836                 scsi_ulto4b(0, bl_ptr->atomic_transfer_length_granularity);
9837                 scsi_ulto4b(0, bl_ptr->max_atomic_transfer_length_with_atomic_boundary);
9838                 scsi_ulto4b(0, bl_ptr->max_atomic_boundary_size);
9839         }
9840         scsi_u64to8b(UINT64_MAX, bl_ptr->max_write_same_length);
9841
9842         ctl_set_success(ctsio);
9843         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9844         ctsio->be_move_done = ctl_config_move_done;
9845         ctl_datamove((union ctl_io *)ctsio);
9846         return (CTL_RETVAL_COMPLETE);
9847 }
9848
9849 static int
9850 ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len)
9851 {
9852         struct scsi_vpd_block_device_characteristics *bdc_ptr;
9853         struct ctl_lun *lun;
9854         const char *value;
9855         u_int i;
9856
9857         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9858
9859         ctsio->kern_data_ptr = malloc(sizeof(*bdc_ptr), M_CTL, M_WAITOK | M_ZERO);
9860         bdc_ptr = (struct scsi_vpd_block_device_characteristics *)ctsio->kern_data_ptr;
9861         ctsio->kern_sg_entries = 0;
9862
9863         if (sizeof(*bdc_ptr) < alloc_len) {
9864                 ctsio->residual = alloc_len - sizeof(*bdc_ptr);
9865                 ctsio->kern_data_len = sizeof(*bdc_ptr);
9866                 ctsio->kern_total_len = sizeof(*bdc_ptr);
9867         } else {
9868                 ctsio->residual = 0;
9869                 ctsio->kern_data_len = alloc_len;
9870                 ctsio->kern_total_len = alloc_len;
9871         }
9872         ctsio->kern_data_resid = 0;
9873         ctsio->kern_rel_offset = 0;
9874         ctsio->kern_sg_entries = 0;
9875
9876         /*
9877          * The control device is always connected.  The disk device, on the
9878          * other hand, may not be online all the time.  Need to change this
9879          * to figure out whether the disk device is actually online or not.
9880          */
9881         if (lun != NULL)
9882                 bdc_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9883                                   lun->be_lun->lun_type;
9884         else
9885                 bdc_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9886         bdc_ptr->page_code = SVPD_BDC;
9887         scsi_ulto2b(sizeof(*bdc_ptr) - 4, bdc_ptr->page_length);
9888         if (lun != NULL &&
9889             (value = ctl_get_opt(&lun->be_lun->options, "rpm")) != NULL)
9890                 i = strtol(value, NULL, 0);
9891         else
9892                 i = CTL_DEFAULT_ROTATION_RATE;
9893         scsi_ulto2b(i, bdc_ptr->medium_rotation_rate);
9894         if (lun != NULL &&
9895             (value = ctl_get_opt(&lun->be_lun->options, "formfactor")) != NULL)
9896                 i = strtol(value, NULL, 0);
9897         else
9898                 i = 0;
9899         bdc_ptr->wab_wac_ff = (i & 0x0f);
9900         bdc_ptr->flags = SVPD_FUAB | SVPD_VBULS;
9901
9902         ctl_set_success(ctsio);
9903         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9904         ctsio->be_move_done = ctl_config_move_done;
9905         ctl_datamove((union ctl_io *)ctsio);
9906         return (CTL_RETVAL_COMPLETE);
9907 }
9908
9909 static int
9910 ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len)
9911 {
9912         struct scsi_vpd_logical_block_prov *lbp_ptr;
9913         struct ctl_lun *lun;
9914
9915         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9916
9917         ctsio->kern_data_ptr = malloc(sizeof(*lbp_ptr), M_CTL, M_WAITOK | M_ZERO);
9918         lbp_ptr = (struct scsi_vpd_logical_block_prov *)ctsio->kern_data_ptr;
9919         ctsio->kern_sg_entries = 0;
9920
9921         if (sizeof(*lbp_ptr) < alloc_len) {
9922                 ctsio->residual = alloc_len - sizeof(*lbp_ptr);
9923                 ctsio->kern_data_len = sizeof(*lbp_ptr);
9924                 ctsio->kern_total_len = sizeof(*lbp_ptr);
9925         } else {
9926                 ctsio->residual = 0;
9927                 ctsio->kern_data_len = alloc_len;
9928                 ctsio->kern_total_len = alloc_len;
9929         }
9930         ctsio->kern_data_resid = 0;
9931         ctsio->kern_rel_offset = 0;
9932         ctsio->kern_sg_entries = 0;
9933
9934         /*
9935          * The control device is always connected.  The disk device, on the
9936          * other hand, may not be online all the time.  Need to change this
9937          * to figure out whether the disk device is actually online or not.
9938          */
9939         if (lun != NULL)
9940                 lbp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9941                                   lun->be_lun->lun_type;
9942         else
9943                 lbp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9944
9945         lbp_ptr->page_code = SVPD_LBP;
9946         scsi_ulto2b(sizeof(*lbp_ptr) - 4, lbp_ptr->page_length);
9947         lbp_ptr->threshold_exponent = CTL_LBP_EXPONENT;
9948         if (lun != NULL && lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
9949                 lbp_ptr->flags = SVPD_LBP_UNMAP | SVPD_LBP_WS16 |
9950                     SVPD_LBP_WS10 | SVPD_LBP_RZ | SVPD_LBP_ANC_SUP;
9951                 lbp_ptr->prov_type = SVPD_LBP_THIN;
9952         }
9953
9954         ctl_set_success(ctsio);
9955         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9956         ctsio->be_move_done = ctl_config_move_done;
9957         ctl_datamove((union ctl_io *)ctsio);
9958         return (CTL_RETVAL_COMPLETE);
9959 }
9960
9961 /*
9962  * INQUIRY with the EVPD bit set.
9963  */
9964 static int
9965 ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
9966 {
9967         struct ctl_lun *lun;
9968         struct scsi_inquiry *cdb;
9969         int alloc_len, retval;
9970
9971         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9972         cdb = (struct scsi_inquiry *)ctsio->cdb;
9973         alloc_len = scsi_2btoul(cdb->length);
9974
9975         switch (cdb->page_code) {
9976         case SVPD_SUPPORTED_PAGES:
9977                 retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
9978                 break;
9979         case SVPD_UNIT_SERIAL_NUMBER:
9980                 retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
9981                 break;
9982         case SVPD_DEVICE_ID:
9983                 retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
9984                 break;
9985         case SVPD_EXTENDED_INQUIRY_DATA:
9986                 retval = ctl_inquiry_evpd_eid(ctsio, alloc_len);
9987                 break;
9988         case SVPD_MODE_PAGE_POLICY:
9989                 retval = ctl_inquiry_evpd_mpp(ctsio, alloc_len);
9990                 break;
9991         case SVPD_SCSI_PORTS:
9992                 retval = ctl_inquiry_evpd_scsi_ports(ctsio, alloc_len);
9993                 break;
9994         case SVPD_SCSI_TPC:
9995                 retval = ctl_inquiry_evpd_tpc(ctsio, alloc_len);
9996                 break;
9997         case SVPD_BLOCK_LIMITS:
9998                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
9999                         goto err;
10000                 retval = ctl_inquiry_evpd_block_limits(ctsio, alloc_len);
10001                 break;
10002         case SVPD_BDC:
10003                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
10004                         goto err;
10005                 retval = ctl_inquiry_evpd_bdc(ctsio, alloc_len);
10006                 break;
10007         case SVPD_LBP:
10008                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
10009                         goto err;
10010                 retval = ctl_inquiry_evpd_lbp(ctsio, alloc_len);
10011                 break;
10012         default:
10013 err:
10014                 ctl_set_invalid_field(ctsio,
10015                                       /*sks_valid*/ 1,
10016                                       /*command*/ 1,
10017                                       /*field*/ 2,
10018                                       /*bit_valid*/ 0,
10019                                       /*bit*/ 0);
10020                 ctl_done((union ctl_io *)ctsio);
10021                 retval = CTL_RETVAL_COMPLETE;
10022                 break;
10023         }
10024
10025         return (retval);
10026 }
10027
10028 /*
10029  * Standard INQUIRY data.
10030  */
10031 static int
10032 ctl_inquiry_std(struct ctl_scsiio *ctsio)
10033 {
10034         struct scsi_inquiry_data *inq_ptr;
10035         struct scsi_inquiry *cdb;
10036         struct ctl_softc *softc = control_softc;
10037         struct ctl_port *port;
10038         struct ctl_lun *lun;
10039         char *val;
10040         uint32_t alloc_len, data_len;
10041         ctl_port_type port_type;
10042
10043         port = ctl_io_port(&ctsio->io_hdr);
10044         port_type = port->port_type;
10045         if (port_type == CTL_PORT_IOCTL || port_type == CTL_PORT_INTERNAL)
10046                 port_type = CTL_PORT_SCSI;
10047
10048         lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10049         cdb = (struct scsi_inquiry *)ctsio->cdb;
10050         alloc_len = scsi_2btoul(cdb->length);
10051
10052         /*
10053          * We malloc the full inquiry data size here and fill it
10054          * in.  If the user only asks for less, we'll give him
10055          * that much.
10056          */
10057         data_len = offsetof(struct scsi_inquiry_data, vendor_specific1);
10058         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10059         inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
10060         ctsio->kern_sg_entries = 0;
10061         ctsio->kern_data_resid = 0;
10062         ctsio->kern_rel_offset = 0;
10063
10064         if (data_len < alloc_len) {
10065                 ctsio->residual = alloc_len - data_len;
10066                 ctsio->kern_data_len = data_len;
10067                 ctsio->kern_total_len = data_len;
10068         } else {
10069                 ctsio->residual = 0;
10070                 ctsio->kern_data_len = alloc_len;
10071                 ctsio->kern_total_len = alloc_len;
10072         }
10073
10074         if (lun != NULL) {
10075                 if ((lun->flags & CTL_LUN_PRIMARY_SC) ||
10076                     softc->ha_link >= CTL_HA_LINK_UNKNOWN) {
10077                         inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10078                             lun->be_lun->lun_type;
10079                 } else {
10080                         inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) |
10081                             lun->be_lun->lun_type;
10082                 }
10083         } else
10084                 inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE;
10085
10086         /* RMB in byte 2 is 0 */
10087         inq_ptr->version = SCSI_REV_SPC4;
10088
10089         /*
10090          * According to SAM-3, even if a device only supports a single
10091          * level of LUN addressing, it should still set the HISUP bit:
10092          *
10093          * 4.9.1 Logical unit numbers overview
10094          *
10095          * All logical unit number formats described in this standard are
10096          * hierarchical in structure even when only a single level in that
10097          * hierarchy is used. The HISUP bit shall be set to one in the
10098          * standard INQUIRY data (see SPC-2) when any logical unit number
10099          * format described in this standard is used.  Non-hierarchical
10100          * formats are outside the scope of this standard.
10101          *
10102          * Therefore we set the HiSup bit here.
10103          *
10104          * The reponse format is 2, per SPC-3.
10105          */
10106         inq_ptr->response_format = SID_HiSup | 2;
10107
10108         inq_ptr->additional_length = data_len -
10109             (offsetof(struct scsi_inquiry_data, additional_length) + 1);
10110         CTL_DEBUG_PRINT(("additional_length = %d\n",
10111                          inq_ptr->additional_length));
10112
10113         inq_ptr->spc3_flags = SPC3_SID_3PC | SPC3_SID_TPGS_IMPLICIT;
10114         /* 16 bit addressing */
10115         if (port_type == CTL_PORT_SCSI)
10116                 inq_ptr->spc2_flags = SPC2_SID_ADDR16;
10117         /* XXX set the SID_MultiP bit here if we're actually going to
10118            respond on multiple ports */
10119         inq_ptr->spc2_flags |= SPC2_SID_MultiP;
10120
10121         /* 16 bit data bus, synchronous transfers */
10122         if (port_type == CTL_PORT_SCSI)
10123                 inq_ptr->flags = SID_WBus16 | SID_Sync;
10124         /*
10125          * XXX KDM do we want to support tagged queueing on the control
10126          * device at all?
10127          */
10128         if ((lun == NULL)
10129          || (lun->be_lun->lun_type != T_PROCESSOR))
10130                 inq_ptr->flags |= SID_CmdQue;
10131         /*
10132          * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
10133          * We have 8 bytes for the vendor name, and 16 bytes for the device
10134          * name and 4 bytes for the revision.
10135          */
10136         if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10137             "vendor")) == NULL) {
10138                 strncpy(inq_ptr->vendor, CTL_VENDOR, sizeof(inq_ptr->vendor));
10139         } else {
10140                 memset(inq_ptr->vendor, ' ', sizeof(inq_ptr->vendor));
10141                 strncpy(inq_ptr->vendor, val,
10142                     min(sizeof(inq_ptr->vendor), strlen(val)));
10143         }
10144         if (lun == NULL) {
10145                 strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10146                     sizeof(inq_ptr->product));
10147         } else if ((val = ctl_get_opt(&lun->be_lun->options, "product")) == NULL) {
10148                 switch (lun->be_lun->lun_type) {
10149                 case T_DIRECT:
10150                         strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10151                             sizeof(inq_ptr->product));
10152                         break;
10153                 case T_PROCESSOR:
10154                         strncpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT,
10155                             sizeof(inq_ptr->product));
10156                         break;
10157                 default:
10158                         strncpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT,
10159                             sizeof(inq_ptr->product));
10160                         break;
10161                 }
10162         } else {
10163                 memset(inq_ptr->product, ' ', sizeof(inq_ptr->product));
10164                 strncpy(inq_ptr->product, val,
10165                     min(sizeof(inq_ptr->product), strlen(val)));
10166         }
10167
10168         /*
10169          * XXX make this a macro somewhere so it automatically gets
10170          * incremented when we make changes.
10171          */
10172         if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10173             "revision")) == NULL) {
10174                 strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
10175         } else {
10176                 memset(inq_ptr->revision, ' ', sizeof(inq_ptr->revision));
10177                 strncpy(inq_ptr->revision, val,
10178                     min(sizeof(inq_ptr->revision), strlen(val)));
10179         }
10180
10181         /*
10182          * For parallel SCSI, we support double transition and single
10183          * transition clocking.  We also support QAS (Quick Arbitration
10184          * and Selection) and Information Unit transfers on both the
10185          * control and array devices.
10186          */
10187         if (port_type == CTL_PORT_SCSI)
10188                 inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
10189                                     SID_SPI_IUS;
10190
10191         /* SAM-5 (no version claimed) */
10192         scsi_ulto2b(0x00A0, inq_ptr->version1);
10193         /* SPC-4 (no version claimed) */
10194         scsi_ulto2b(0x0460, inq_ptr->version2);
10195         if (port_type == CTL_PORT_FC) {
10196                 /* FCP-2 ANSI INCITS.350:2003 */
10197                 scsi_ulto2b(0x0917, inq_ptr->version3);
10198         } else if (port_type == CTL_PORT_SCSI) {
10199                 /* SPI-4 ANSI INCITS.362:200x */
10200                 scsi_ulto2b(0x0B56, inq_ptr->version3);
10201         } else if (port_type == CTL_PORT_ISCSI) {
10202                 /* iSCSI (no version claimed) */
10203                 scsi_ulto2b(0x0960, inq_ptr->version3);
10204         } else if (port_type == CTL_PORT_SAS) {
10205                 /* SAS (no version claimed) */
10206                 scsi_ulto2b(0x0BE0, inq_ptr->version3);
10207         }
10208
10209         if (lun == NULL) {
10210                 /* SBC-4 (no version claimed) */
10211                 scsi_ulto2b(0x0600, inq_ptr->version4);
10212         } else {
10213                 switch (lun->be_lun->lun_type) {
10214                 case T_DIRECT:
10215                         /* SBC-4 (no version claimed) */
10216                         scsi_ulto2b(0x0600, inq_ptr->version4);
10217                         break;
10218                 case T_PROCESSOR:
10219                 default:
10220                         break;
10221                 }
10222         }
10223
10224         ctl_set_success(ctsio);
10225         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10226         ctsio->be_move_done = ctl_config_move_done;
10227         ctl_datamove((union ctl_io *)ctsio);
10228         return (CTL_RETVAL_COMPLETE);
10229 }
10230
10231 int
10232 ctl_inquiry(struct ctl_scsiio *ctsio)
10233 {
10234         struct scsi_inquiry *cdb;
10235         int retval;
10236
10237         CTL_DEBUG_PRINT(("ctl_inquiry\n"));
10238
10239         cdb = (struct scsi_inquiry *)ctsio->cdb;
10240         if (cdb->byte2 & SI_EVPD)
10241                 retval = ctl_inquiry_evpd(ctsio);
10242         else if (cdb->page_code == 0)
10243                 retval = ctl_inquiry_std(ctsio);
10244         else {
10245                 ctl_set_invalid_field(ctsio,
10246                                       /*sks_valid*/ 1,
10247                                       /*command*/ 1,
10248                                       /*field*/ 2,
10249                                       /*bit_valid*/ 0,
10250                                       /*bit*/ 0);
10251                 ctl_done((union ctl_io *)ctsio);
10252                 return (CTL_RETVAL_COMPLETE);
10253         }
10254
10255         return (retval);
10256 }
10257
10258 /*
10259  * For known CDB types, parse the LBA and length.
10260  */
10261 static int
10262 ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len)
10263 {
10264         if (io->io_hdr.io_type != CTL_IO_SCSI)
10265                 return (1);
10266
10267         switch (io->scsiio.cdb[0]) {
10268         case COMPARE_AND_WRITE: {
10269                 struct scsi_compare_and_write *cdb;
10270
10271                 cdb = (struct scsi_compare_and_write *)io->scsiio.cdb;
10272
10273                 *lba = scsi_8btou64(cdb->addr);
10274                 *len = cdb->length;
10275                 break;
10276         }
10277         case READ_6:
10278         case WRITE_6: {
10279                 struct scsi_rw_6 *cdb;
10280
10281                 cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
10282
10283                 *lba = scsi_3btoul(cdb->addr);
10284                 /* only 5 bits are valid in the most significant address byte */
10285                 *lba &= 0x1fffff;
10286                 *len = cdb->length;
10287                 break;
10288         }
10289         case READ_10:
10290         case WRITE_10: {
10291                 struct scsi_rw_10 *cdb;
10292
10293                 cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
10294
10295                 *lba = scsi_4btoul(cdb->addr);
10296                 *len = scsi_2btoul(cdb->length);
10297                 break;
10298         }
10299         case WRITE_VERIFY_10: {
10300                 struct scsi_write_verify_10 *cdb;
10301
10302                 cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
10303
10304                 *lba = scsi_4btoul(cdb->addr);
10305                 *len = scsi_2btoul(cdb->length);
10306                 break;
10307         }
10308         case READ_12:
10309         case WRITE_12: {
10310                 struct scsi_rw_12 *cdb;
10311
10312                 cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
10313
10314                 *lba = scsi_4btoul(cdb->addr);
10315                 *len = scsi_4btoul(cdb->length);
10316                 break;
10317         }
10318         case WRITE_VERIFY_12: {
10319                 struct scsi_write_verify_12 *cdb;
10320
10321                 cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
10322
10323                 *lba = scsi_4btoul(cdb->addr);
10324                 *len = scsi_4btoul(cdb->length);
10325                 break;
10326         }
10327         case READ_16:
10328         case WRITE_16: {
10329                 struct scsi_rw_16 *cdb;
10330
10331                 cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
10332
10333                 *lba = scsi_8btou64(cdb->addr);
10334                 *len = scsi_4btoul(cdb->length);
10335                 break;
10336         }
10337         case WRITE_ATOMIC_16: {
10338                 struct scsi_write_atomic_16 *cdb;
10339
10340                 cdb = (struct scsi_write_atomic_16 *)io->scsiio.cdb;
10341
10342                 *lba = scsi_8btou64(cdb->addr);
10343                 *len = scsi_2btoul(cdb->length);
10344                 break;
10345         }
10346         case WRITE_VERIFY_16: {
10347                 struct scsi_write_verify_16 *cdb;
10348
10349                 cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
10350
10351                 *lba = scsi_8btou64(cdb->addr);
10352                 *len = scsi_4btoul(cdb->length);
10353                 break;
10354         }
10355         case WRITE_SAME_10: {
10356                 struct scsi_write_same_10 *cdb;
10357
10358                 cdb = (struct scsi_write_same_10 *)io->scsiio.cdb;
10359
10360                 *lba = scsi_4btoul(cdb->addr);
10361                 *len = scsi_2btoul(cdb->length);
10362                 break;
10363         }
10364         case WRITE_SAME_16: {
10365                 struct scsi_write_same_16 *cdb;
10366
10367                 cdb = (struct scsi_write_same_16 *)io->scsiio.cdb;
10368
10369                 *lba = scsi_8btou64(cdb->addr);
10370                 *len = scsi_4btoul(cdb->length);
10371                 break;
10372         }
10373         case VERIFY_10: {
10374                 struct scsi_verify_10 *cdb;
10375
10376                 cdb = (struct scsi_verify_10 *)io->scsiio.cdb;
10377
10378                 *lba = scsi_4btoul(cdb->addr);
10379                 *len = scsi_2btoul(cdb->length);
10380                 break;
10381         }
10382         case VERIFY_12: {
10383                 struct scsi_verify_12 *cdb;
10384
10385                 cdb = (struct scsi_verify_12 *)io->scsiio.cdb;
10386
10387                 *lba = scsi_4btoul(cdb->addr);
10388                 *len = scsi_4btoul(cdb->length);
10389                 break;
10390         }
10391         case VERIFY_16: {
10392                 struct scsi_verify_16 *cdb;
10393
10394                 cdb = (struct scsi_verify_16 *)io->scsiio.cdb;
10395
10396                 *lba = scsi_8btou64(cdb->addr);
10397                 *len = scsi_4btoul(cdb->length);
10398                 break;
10399         }
10400         case UNMAP: {
10401                 *lba = 0;
10402                 *len = UINT64_MAX;
10403                 break;
10404         }
10405         case SERVICE_ACTION_IN: {       /* GET LBA STATUS */
10406                 struct scsi_get_lba_status *cdb;
10407
10408                 cdb = (struct scsi_get_lba_status *)io->scsiio.cdb;
10409                 *lba = scsi_8btou64(cdb->addr);
10410                 *len = UINT32_MAX;
10411                 break;
10412         }
10413         default:
10414                 return (1);
10415                 break; /* NOTREACHED */
10416         }
10417
10418         return (0);
10419 }
10420
10421 static ctl_action
10422 ctl_extent_check_lba(uint64_t lba1, uint64_t len1, uint64_t lba2, uint64_t len2,
10423     bool seq)
10424 {
10425         uint64_t endlba1, endlba2;
10426
10427         endlba1 = lba1 + len1 - (seq ? 0 : 1);
10428         endlba2 = lba2 + len2 - 1;
10429
10430         if ((endlba1 < lba2) || (endlba2 < lba1))
10431                 return (CTL_ACTION_PASS);
10432         else
10433                 return (CTL_ACTION_BLOCK);
10434 }
10435
10436 static int
10437 ctl_extent_check_unmap(union ctl_io *io, uint64_t lba2, uint64_t len2)
10438 {
10439         struct ctl_ptr_len_flags *ptrlen;
10440         struct scsi_unmap_desc *buf, *end, *range;
10441         uint64_t lba;
10442         uint32_t len;
10443
10444         /* If not UNMAP -- go other way. */
10445         if (io->io_hdr.io_type != CTL_IO_SCSI ||
10446             io->scsiio.cdb[0] != UNMAP)
10447                 return (CTL_ACTION_ERROR);
10448
10449         /* If UNMAP without data -- block and wait for data. */
10450         ptrlen = (struct ctl_ptr_len_flags *)
10451             &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
10452         if ((io->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0 ||
10453             ptrlen->ptr == NULL)
10454                 return (CTL_ACTION_BLOCK);
10455
10456         /* UNMAP with data -- check for collision. */
10457         buf = (struct scsi_unmap_desc *)ptrlen->ptr;
10458         end = buf + ptrlen->len / sizeof(*buf);
10459         for (range = buf; range < end; range++) {
10460                 lba = scsi_8btou64(range->lba);
10461                 len = scsi_4btoul(range->length);
10462                 if ((lba < lba2 + len2) && (lba + len > lba2))
10463                         return (CTL_ACTION_BLOCK);
10464         }
10465         return (CTL_ACTION_PASS);
10466 }
10467
10468 static ctl_action
10469 ctl_extent_check(union ctl_io *io1, union ctl_io *io2, bool seq)
10470 {
10471         uint64_t lba1, lba2;
10472         uint64_t len1, len2;
10473         int retval;
10474
10475         if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
10476                 return (CTL_ACTION_ERROR);
10477
10478         retval = ctl_extent_check_unmap(io1, lba2, len2);
10479         if (retval != CTL_ACTION_ERROR)
10480                 return (retval);
10481
10482         if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
10483                 return (CTL_ACTION_ERROR);
10484
10485         if (io1->io_hdr.flags & CTL_FLAG_SERSEQ_DONE)
10486                 seq = FALSE;
10487         return (ctl_extent_check_lba(lba1, len1, lba2, len2, seq));
10488 }
10489
10490 static ctl_action
10491 ctl_extent_check_seq(union ctl_io *io1, union ctl_io *io2)
10492 {
10493         uint64_t lba1, lba2;
10494         uint64_t len1, len2;
10495
10496         if (io1->io_hdr.flags & CTL_FLAG_SERSEQ_DONE)
10497                 return (CTL_ACTION_PASS);
10498         if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
10499                 return (CTL_ACTION_ERROR);
10500         if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
10501                 return (CTL_ACTION_ERROR);
10502
10503         if (lba1 + len1 == lba2)
10504                 return (CTL_ACTION_BLOCK);
10505         return (CTL_ACTION_PASS);
10506 }
10507
10508 static ctl_action
10509 ctl_check_for_blockage(struct ctl_lun *lun, union ctl_io *pending_io,
10510     union ctl_io *ooa_io)
10511 {
10512         const struct ctl_cmd_entry *pending_entry, *ooa_entry;
10513         const ctl_serialize_action *serialize_row;
10514
10515         /*
10516          * The initiator attempted multiple untagged commands at the same
10517          * time.  Can't do that.
10518          */
10519         if ((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10520          && (ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10521          && ((pending_io->io_hdr.nexus.targ_port ==
10522               ooa_io->io_hdr.nexus.targ_port)
10523           && (pending_io->io_hdr.nexus.initid ==
10524               ooa_io->io_hdr.nexus.initid))
10525          && ((ooa_io->io_hdr.flags & (CTL_FLAG_ABORT |
10526               CTL_FLAG_STATUS_SENT)) == 0))
10527                 return (CTL_ACTION_OVERLAP);
10528
10529         /*
10530          * The initiator attempted to send multiple tagged commands with
10531          * the same ID.  (It's fine if different initiators have the same
10532          * tag ID.)
10533          *
10534          * Even if all of those conditions are true, we don't kill the I/O
10535          * if the command ahead of us has been aborted.  We won't end up
10536          * sending it to the FETD, and it's perfectly legal to resend a
10537          * command with the same tag number as long as the previous
10538          * instance of this tag number has been aborted somehow.
10539          */
10540         if ((pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10541          && (ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10542          && (pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
10543          && ((pending_io->io_hdr.nexus.targ_port ==
10544               ooa_io->io_hdr.nexus.targ_port)
10545           && (pending_io->io_hdr.nexus.initid ==
10546               ooa_io->io_hdr.nexus.initid))
10547          && ((ooa_io->io_hdr.flags & (CTL_FLAG_ABORT |
10548               CTL_FLAG_STATUS_SENT)) == 0))
10549                 return (CTL_ACTION_OVERLAP_TAG);
10550
10551         /*
10552          * If we get a head of queue tag, SAM-3 says that we should
10553          * immediately execute it.
10554          *
10555          * What happens if this command would normally block for some other
10556          * reason?  e.g. a request sense with a head of queue tag
10557          * immediately after a write.  Normally that would block, but this
10558          * will result in its getting executed immediately...
10559          *
10560          * We currently return "pass" instead of "skip", so we'll end up
10561          * going through the rest of the queue to check for overlapped tags.
10562          *
10563          * XXX KDM check for other types of blockage first??
10564          */
10565         if (pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10566                 return (CTL_ACTION_PASS);
10567
10568         /*
10569          * Ordered tags have to block until all items ahead of them
10570          * have completed.  If we get called with an ordered tag, we always
10571          * block, if something else is ahead of us in the queue.
10572          */
10573         if (pending_io->scsiio.tag_type == CTL_TAG_ORDERED)
10574                 return (CTL_ACTION_BLOCK);
10575
10576         /*
10577          * Simple tags get blocked until all head of queue and ordered tags
10578          * ahead of them have completed.  I'm lumping untagged commands in
10579          * with simple tags here.  XXX KDM is that the right thing to do?
10580          */
10581         if (((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10582           || (pending_io->scsiio.tag_type == CTL_TAG_SIMPLE))
10583          && ((ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10584           || (ooa_io->scsiio.tag_type == CTL_TAG_ORDERED)))
10585                 return (CTL_ACTION_BLOCK);
10586
10587         pending_entry = ctl_get_cmd_entry(&pending_io->scsiio, NULL);
10588         ooa_entry = ctl_get_cmd_entry(&ooa_io->scsiio, NULL);
10589
10590         serialize_row = ctl_serialize_table[ooa_entry->seridx];
10591
10592         switch (serialize_row[pending_entry->seridx]) {
10593         case CTL_SER_BLOCK:
10594                 return (CTL_ACTION_BLOCK);
10595         case CTL_SER_EXTENT:
10596                 return (ctl_extent_check(ooa_io, pending_io,
10597                     (lun->be_lun && lun->be_lun->serseq == CTL_LUN_SERSEQ_ON)));
10598         case CTL_SER_EXTENTOPT:
10599                 if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT].queue_flags
10600                     & SCP_QUEUE_ALG_MASK) != SCP_QUEUE_ALG_UNRESTRICTED)
10601                         return (ctl_extent_check(ooa_io, pending_io,
10602                             (lun->be_lun &&
10603                              lun->be_lun->serseq == CTL_LUN_SERSEQ_ON)));
10604                 return (CTL_ACTION_PASS);
10605         case CTL_SER_EXTENTSEQ:
10606                 if (lun->be_lun && lun->be_lun->serseq != CTL_LUN_SERSEQ_OFF)
10607                         return (ctl_extent_check_seq(ooa_io, pending_io));
10608                 return (CTL_ACTION_PASS);
10609         case CTL_SER_PASS:
10610                 return (CTL_ACTION_PASS);
10611         case CTL_SER_BLOCKOPT:
10612                 if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT].queue_flags
10613                     & SCP_QUEUE_ALG_MASK) != SCP_QUEUE_ALG_UNRESTRICTED)
10614                         return (CTL_ACTION_BLOCK);
10615                 return (CTL_ACTION_PASS);
10616         case CTL_SER_SKIP:
10617                 return (CTL_ACTION_SKIP);
10618         default:
10619                 panic("invalid serialization value %d",
10620                       serialize_row[pending_entry->seridx]);
10621         }
10622
10623         return (CTL_ACTION_ERROR);
10624 }
10625
10626 /*
10627  * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
10628  * Assumptions:
10629  * - pending_io is generally either incoming, or on the blocked queue
10630  * - starting I/O is the I/O we want to start the check with.
10631  */
10632 static ctl_action
10633 ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
10634               union ctl_io *starting_io)
10635 {
10636         union ctl_io *ooa_io;
10637         ctl_action action;
10638
10639         mtx_assert(&lun->lun_lock, MA_OWNED);
10640
10641         /*
10642          * Run back along the OOA queue, starting with the current
10643          * blocked I/O and going through every I/O before it on the
10644          * queue.  If starting_io is NULL, we'll just end up returning
10645          * CTL_ACTION_PASS.
10646          */
10647         for (ooa_io = starting_io; ooa_io != NULL;
10648              ooa_io = (union ctl_io *)TAILQ_PREV(&ooa_io->io_hdr, ctl_ooaq,
10649              ooa_links)){
10650
10651                 /*
10652                  * This routine just checks to see whether
10653                  * cur_blocked is blocked by ooa_io, which is ahead
10654                  * of it in the queue.  It doesn't queue/dequeue
10655                  * cur_blocked.
10656                  */
10657                 action = ctl_check_for_blockage(lun, pending_io, ooa_io);
10658                 switch (action) {
10659                 case CTL_ACTION_BLOCK:
10660                 case CTL_ACTION_OVERLAP:
10661                 case CTL_ACTION_OVERLAP_TAG:
10662                 case CTL_ACTION_SKIP:
10663                 case CTL_ACTION_ERROR:
10664                         return (action);
10665                         break; /* NOTREACHED */
10666                 case CTL_ACTION_PASS:
10667                         break;
10668                 default:
10669                         panic("invalid action %d", action);
10670                         break;  /* NOTREACHED */
10671                 }
10672         }
10673
10674         return (CTL_ACTION_PASS);
10675 }
10676
10677 /*
10678  * Assumptions:
10679  * - An I/O has just completed, and has been removed from the per-LUN OOA
10680  *   queue, so some items on the blocked queue may now be unblocked.
10681  */
10682 static int
10683 ctl_check_blocked(struct ctl_lun *lun)
10684 {
10685         struct ctl_softc *softc = lun->ctl_softc;
10686         union ctl_io *cur_blocked, *next_blocked;
10687
10688         mtx_assert(&lun->lun_lock, MA_OWNED);
10689
10690         /*
10691          * Run forward from the head of the blocked queue, checking each
10692          * entry against the I/Os prior to it on the OOA queue to see if
10693          * there is still any blockage.
10694          *
10695          * We cannot use the TAILQ_FOREACH() macro, because it can't deal
10696          * with our removing a variable on it while it is traversing the
10697          * list.
10698          */
10699         for (cur_blocked = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue);
10700              cur_blocked != NULL; cur_blocked = next_blocked) {
10701                 union ctl_io *prev_ooa;
10702                 ctl_action action;
10703
10704                 next_blocked = (union ctl_io *)TAILQ_NEXT(&cur_blocked->io_hdr,
10705                                                           blocked_links);
10706
10707                 prev_ooa = (union ctl_io *)TAILQ_PREV(&cur_blocked->io_hdr,
10708                                                       ctl_ooaq, ooa_links);
10709
10710                 /*
10711                  * If cur_blocked happens to be the first item in the OOA
10712                  * queue now, prev_ooa will be NULL, and the action
10713                  * returned will just be CTL_ACTION_PASS.
10714                  */
10715                 action = ctl_check_ooa(lun, cur_blocked, prev_ooa);
10716
10717                 switch (action) {
10718                 case CTL_ACTION_BLOCK:
10719                         /* Nothing to do here, still blocked */
10720                         break;
10721                 case CTL_ACTION_OVERLAP:
10722                 case CTL_ACTION_OVERLAP_TAG:
10723                         /*
10724                          * This shouldn't happen!  In theory we've already
10725                          * checked this command for overlap...
10726                          */
10727                         break;
10728                 case CTL_ACTION_PASS:
10729                 case CTL_ACTION_SKIP: {
10730                         const struct ctl_cmd_entry *entry;
10731
10732                         /*
10733                          * The skip case shouldn't happen, this transaction
10734                          * should have never made it onto the blocked queue.
10735                          */
10736                         /*
10737                          * This I/O is no longer blocked, we can remove it
10738                          * from the blocked queue.  Since this is a TAILQ
10739                          * (doubly linked list), we can do O(1) removals
10740                          * from any place on the list.
10741                          */
10742                         TAILQ_REMOVE(&lun->blocked_queue, &cur_blocked->io_hdr,
10743                                      blocked_links);
10744                         cur_blocked->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
10745
10746                         if ((softc->ha_mode != CTL_HA_MODE_XFER) &&
10747                             (cur_blocked->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)){
10748                                 /*
10749                                  * Need to send IO back to original side to
10750                                  * run
10751                                  */
10752                                 union ctl_ha_msg msg_info;
10753
10754                                 cur_blocked->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
10755                                 msg_info.hdr.original_sc =
10756                                         cur_blocked->io_hdr.original_sc;
10757                                 msg_info.hdr.serializing_sc = cur_blocked;
10758                                 msg_info.hdr.msg_type = CTL_MSG_R2R;
10759                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
10760                                     sizeof(msg_info.hdr), M_NOWAIT);
10761                                 break;
10762                         }
10763                         entry = ctl_get_cmd_entry(&cur_blocked->scsiio, NULL);
10764
10765                         /*
10766                          * Check this I/O for LUN state changes that may
10767                          * have happened while this command was blocked.
10768                          * The LUN state may have been changed by a command
10769                          * ahead of us in the queue, so we need to re-check
10770                          * for any states that can be caused by SCSI
10771                          * commands.
10772                          */
10773                         if (ctl_scsiio_lun_check(lun, entry,
10774                                                  &cur_blocked->scsiio) == 0) {
10775                                 cur_blocked->io_hdr.flags |=
10776                                                       CTL_FLAG_IS_WAS_ON_RTR;
10777                                 ctl_enqueue_rtr(cur_blocked);
10778                         } else
10779                                 ctl_done(cur_blocked);
10780                         break;
10781                 }
10782                 default:
10783                         /*
10784                          * This probably shouldn't happen -- we shouldn't
10785                          * get CTL_ACTION_ERROR, or anything else.
10786                          */
10787                         break;
10788                 }
10789         }
10790
10791         return (CTL_RETVAL_COMPLETE);
10792 }
10793
10794 /*
10795  * This routine (with one exception) checks LUN flags that can be set by
10796  * commands ahead of us in the OOA queue.  These flags have to be checked
10797  * when a command initially comes in, and when we pull a command off the
10798  * blocked queue and are preparing to execute it.  The reason we have to
10799  * check these flags for commands on the blocked queue is that the LUN
10800  * state may have been changed by a command ahead of us while we're on the
10801  * blocked queue.
10802  *
10803  * Ordering is somewhat important with these checks, so please pay
10804  * careful attention to the placement of any new checks.
10805  */
10806 static int
10807 ctl_scsiio_lun_check(struct ctl_lun *lun,
10808     const struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
10809 {
10810         struct ctl_softc *softc = lun->ctl_softc;
10811         int retval;
10812         uint32_t residx;
10813
10814         retval = 0;
10815
10816         mtx_assert(&lun->lun_lock, MA_OWNED);
10817
10818         /*
10819          * If this shelf is a secondary shelf controller, we may have to
10820          * reject some commands disallowed by HA mode and link state.
10821          */
10822         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
10823                 if (softc->ha_link == CTL_HA_LINK_OFFLINE &&
10824                     (entry->flags & CTL_CMD_FLAG_OK_ON_UNAVAIL) == 0) {
10825                         ctl_set_lun_unavail(ctsio);
10826                         retval = 1;
10827                         goto bailout;
10828                 }
10829                 if ((lun->flags & CTL_LUN_PEER_SC_PRIMARY) == 0 &&
10830                     (entry->flags & CTL_CMD_FLAG_OK_ON_UNAVAIL) == 0) {
10831                         ctl_set_lun_transit(ctsio);
10832                         retval = 1;
10833                         goto bailout;
10834                 }
10835                 if (softc->ha_mode == CTL_HA_MODE_ACT_STBY &&
10836                     (entry->flags & CTL_CMD_FLAG_OK_ON_STANDBY) == 0) {
10837                         ctl_set_lun_standby(ctsio);
10838                         retval = 1;
10839                         goto bailout;
10840                 }
10841
10842                 /* The rest of checks are only done on executing side */
10843                 if (softc->ha_mode == CTL_HA_MODE_XFER)
10844                         goto bailout;
10845         }
10846
10847         if (entry->pattern & CTL_LUN_PAT_WRITE) {
10848                 if (lun->be_lun &&
10849                     lun->be_lun->flags & CTL_LUN_FLAG_READONLY) {
10850                         ctl_set_hw_write_protected(ctsio);
10851                         retval = 1;
10852                         goto bailout;
10853                 }
10854                 if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT]
10855                     .eca_and_aen & SCP_SWP) != 0) {
10856                         ctl_set_sense(ctsio, /*current_error*/ 1,
10857                             /*sense_key*/ SSD_KEY_DATA_PROTECT,
10858                             /*asc*/ 0x27, /*ascq*/ 0x02, SSD_ELEM_NONE);
10859                         retval = 1;
10860                         goto bailout;
10861                 }
10862         }
10863
10864         /*
10865          * Check for a reservation conflict.  If this command isn't allowed
10866          * even on reserved LUNs, and if this initiator isn't the one who
10867          * reserved us, reject the command with a reservation conflict.
10868          */
10869         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
10870         if ((lun->flags & CTL_LUN_RESERVED)
10871          && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
10872                 if (lun->res_idx != residx) {
10873                         ctl_set_reservation_conflict(ctsio);
10874                         retval = 1;
10875                         goto bailout;
10876                 }
10877         }
10878
10879         if ((lun->flags & CTL_LUN_PR_RESERVED) == 0 ||
10880             (entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV)) {
10881                 /* No reservation or command is allowed. */;
10882         } else if ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_WRESV) &&
10883             (lun->res_type == SPR_TYPE_WR_EX ||
10884              lun->res_type == SPR_TYPE_WR_EX_RO ||
10885              lun->res_type == SPR_TYPE_WR_EX_AR)) {
10886                 /* The command is allowed for Write Exclusive resv. */;
10887         } else {
10888                 /*
10889                  * if we aren't registered or it's a res holder type
10890                  * reservation and this isn't the res holder then set a
10891                  * conflict.
10892                  */
10893                 if (ctl_get_prkey(lun, residx) == 0
10894                  || (residx != lun->pr_res_idx && lun->res_type < 4)) {
10895                         ctl_set_reservation_conflict(ctsio);
10896                         retval = 1;
10897                         goto bailout;
10898                 }
10899         }
10900
10901         if ((lun->flags & CTL_LUN_OFFLINE)
10902          && ((entry->flags & CTL_CMD_FLAG_OK_ON_STANDBY) == 0)) {
10903                 ctl_set_lun_not_ready(ctsio);
10904                 retval = 1;
10905                 goto bailout;
10906         }
10907
10908         if ((lun->flags & CTL_LUN_STOPPED)
10909          && ((entry->flags & CTL_CMD_FLAG_OK_ON_STOPPED) == 0)) {
10910                 /* "Logical unit not ready, initializing cmd. required" */
10911                 ctl_set_lun_stopped(ctsio);
10912                 retval = 1;
10913                 goto bailout;
10914         }
10915
10916         if ((lun->flags & CTL_LUN_INOPERABLE)
10917          && ((entry->flags & CTL_CMD_FLAG_OK_ON_INOPERABLE) == 0)) {
10918                 /* "Medium format corrupted" */
10919                 ctl_set_medium_format_corrupted(ctsio);
10920                 retval = 1;
10921                 goto bailout;
10922         }
10923
10924 bailout:
10925         return (retval);
10926 }
10927
10928 static void
10929 ctl_failover_io(union ctl_io *io, int have_lock)
10930 {
10931         ctl_set_busy(&io->scsiio);
10932         ctl_done(io);
10933 }
10934
10935 static void
10936 ctl_failover_lun(union ctl_io *rio)
10937 {
10938         struct ctl_softc *softc = control_softc;
10939         struct ctl_lun *lun;
10940         struct ctl_io_hdr *io, *next_io;
10941         uint32_t targ_lun;
10942
10943         targ_lun = rio->io_hdr.nexus.targ_mapped_lun;
10944         CTL_DEBUG_PRINT(("FAILOVER for lun %ju\n", targ_lun));
10945
10946         /* Find and lock the LUN. */
10947         mtx_lock(&softc->ctl_lock);
10948         if ((targ_lun < CTL_MAX_LUNS) &&
10949             ((lun = softc->ctl_luns[targ_lun]) != NULL)) {
10950                 mtx_lock(&lun->lun_lock);
10951                 mtx_unlock(&softc->ctl_lock);
10952                 if (lun->flags & CTL_LUN_DISABLED) {
10953                         mtx_unlock(&lun->lun_lock);
10954                         return;
10955                 }
10956         } else {
10957                 mtx_unlock(&softc->ctl_lock);
10958                 return;
10959         }
10960
10961         if (softc->ha_mode == CTL_HA_MODE_XFER) {
10962                 TAILQ_FOREACH_SAFE(io, &lun->ooa_queue, ooa_links, next_io) {
10963                         /* We are master */
10964                         if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
10965                                 if (io->flags & CTL_FLAG_IO_ACTIVE) {
10966                                         io->flags |= CTL_FLAG_ABORT;
10967                                         io->flags |= CTL_FLAG_FAILOVER;
10968                                 } else { /* This can be only due to DATAMOVE */
10969                                         io->msg_type = CTL_MSG_DATAMOVE_DONE;
10970                                         io->flags &= ~CTL_FLAG_DMA_INPROG;
10971                                         io->flags |= CTL_FLAG_IO_ACTIVE;
10972                                         io->port_status = 31340;
10973                                         ctl_enqueue_isc((union ctl_io *)io);
10974                                 }
10975                         }
10976                         /* We are slave */
10977                         if (io->flags & CTL_FLAG_SENT_2OTHER_SC) {
10978                                 io->flags &= ~CTL_FLAG_SENT_2OTHER_SC;
10979                                 if (io->flags & CTL_FLAG_IO_ACTIVE) {
10980                                         io->flags |= CTL_FLAG_FAILOVER;
10981                                 } else {
10982                                         ctl_set_busy(&((union ctl_io *)io)->
10983                                             scsiio);
10984                                         ctl_done((union ctl_io *)io);
10985                                 }
10986                         }
10987                 }
10988         } else { /* SERIALIZE modes */
10989                 TAILQ_FOREACH_SAFE(io, &lun->blocked_queue, blocked_links,
10990                     next_io) {
10991                         /* We are master */
10992                         if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
10993                                 TAILQ_REMOVE(&lun->blocked_queue, io,
10994                                     blocked_links);
10995                                 io->flags &= ~CTL_FLAG_BLOCKED;
10996                                 TAILQ_REMOVE(&lun->ooa_queue, io, ooa_links);
10997                                 ctl_free_io((union ctl_io *)io);
10998                         }
10999                 }
11000                 TAILQ_FOREACH_SAFE(io, &lun->ooa_queue, ooa_links, next_io) {
11001                         /* We are master */
11002                         if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
11003                                 TAILQ_REMOVE(&lun->ooa_queue, io, ooa_links);
11004                                 ctl_free_io((union ctl_io *)io);
11005                         }
11006                         /* We are slave */
11007                         if (io->flags & CTL_FLAG_SENT_2OTHER_SC) {
11008                                 io->flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11009                                 if (!(io->flags & CTL_FLAG_IO_ACTIVE)) {
11010                                         ctl_set_busy(&((union ctl_io *)io)->
11011                                             scsiio);
11012                                         ctl_done((union ctl_io *)io);
11013                                 }
11014                         }
11015                 }
11016                 ctl_check_blocked(lun);
11017         }
11018         mtx_unlock(&lun->lun_lock);
11019 }
11020
11021 static int
11022 ctl_scsiio_precheck(struct ctl_softc *softc, struct ctl_scsiio *ctsio)
11023 {
11024         struct ctl_lun *lun;
11025         const struct ctl_cmd_entry *entry;
11026         uint32_t initidx, targ_lun;
11027         int retval;
11028
11029         retval = 0;
11030
11031         lun = NULL;
11032
11033         targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
11034         if ((targ_lun < CTL_MAX_LUNS)
11035          && ((lun = softc->ctl_luns[targ_lun]) != NULL)) {
11036                 /*
11037                  * If the LUN is invalid, pretend that it doesn't exist.
11038                  * It will go away as soon as all pending I/O has been
11039                  * completed.
11040                  */
11041                 mtx_lock(&lun->lun_lock);
11042                 if (lun->flags & CTL_LUN_DISABLED) {
11043                         mtx_unlock(&lun->lun_lock);
11044                         lun = NULL;
11045                         ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
11046                         ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
11047                 } else {
11048                         ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = lun;
11049                         ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr =
11050                                 lun->be_lun;
11051
11052                         /*
11053                          * Every I/O goes into the OOA queue for a
11054                          * particular LUN, and stays there until completion.
11055                          */
11056 #ifdef CTL_TIME_IO
11057                         if (TAILQ_EMPTY(&lun->ooa_queue)) {
11058                                 lun->idle_time += getsbinuptime() -
11059                                     lun->last_busy;
11060                         }
11061 #endif
11062                         TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr,
11063                             ooa_links);
11064                 }
11065         } else {
11066                 ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
11067                 ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
11068         }
11069
11070         /* Get command entry and return error if it is unsuppotyed. */
11071         entry = ctl_validate_command(ctsio);
11072         if (entry == NULL) {
11073                 if (lun)
11074                         mtx_unlock(&lun->lun_lock);
11075                 return (retval);
11076         }
11077
11078         ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
11079         ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
11080
11081         /*
11082          * Check to see whether we can send this command to LUNs that don't
11083          * exist.  This should pretty much only be the case for inquiry
11084          * and request sense.  Further checks, below, really require having
11085          * a LUN, so we can't really check the command anymore.  Just put
11086          * it on the rtr queue.
11087          */
11088         if (lun == NULL) {
11089                 if (entry->flags & CTL_CMD_FLAG_OK_ON_NO_LUN) {
11090                         ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11091                         ctl_enqueue_rtr((union ctl_io *)ctsio);
11092                         return (retval);
11093                 }
11094
11095                 ctl_set_unsupported_lun(ctsio);
11096                 ctl_done((union ctl_io *)ctsio);
11097                 CTL_DEBUG_PRINT(("ctl_scsiio_precheck: bailing out due to invalid LUN\n"));
11098                 return (retval);
11099         } else {
11100                 /*
11101                  * Make sure we support this particular command on this LUN.
11102                  * e.g., we don't support writes to the control LUN.
11103                  */
11104                 if (!ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
11105                         mtx_unlock(&lun->lun_lock);
11106                         ctl_set_invalid_opcode(ctsio);
11107                         ctl_done((union ctl_io *)ctsio);
11108                         return (retval);
11109                 }
11110         }
11111
11112         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11113
11114 #ifdef CTL_WITH_CA
11115         /*
11116          * If we've got a request sense, it'll clear the contingent
11117          * allegiance condition.  Otherwise, if we have a CA condition for
11118          * this initiator, clear it, because it sent down a command other
11119          * than request sense.
11120          */
11121         if ((ctsio->cdb[0] != REQUEST_SENSE)
11122          && (ctl_is_set(lun->have_ca, initidx)))
11123                 ctl_clear_mask(lun->have_ca, initidx);
11124 #endif
11125
11126         /*
11127          * If the command has this flag set, it handles its own unit
11128          * attention reporting, we shouldn't do anything.  Otherwise we
11129          * check for any pending unit attentions, and send them back to the
11130          * initiator.  We only do this when a command initially comes in,
11131          * not when we pull it off the blocked queue.
11132          *
11133          * According to SAM-3, section 5.3.2, the order that things get
11134          * presented back to the host is basically unit attentions caused
11135          * by some sort of reset event, busy status, reservation conflicts
11136          * or task set full, and finally any other status.
11137          *
11138          * One issue here is that some of the unit attentions we report
11139          * don't fall into the "reset" category (e.g. "reported luns data
11140          * has changed").  So reporting it here, before the reservation
11141          * check, may be technically wrong.  I guess the only thing to do
11142          * would be to check for and report the reset events here, and then
11143          * check for the other unit attention types after we check for a
11144          * reservation conflict.
11145          *
11146          * XXX KDM need to fix this
11147          */
11148         if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
11149                 ctl_ua_type ua_type;
11150
11151                 ua_type = ctl_build_ua(lun, initidx, &ctsio->sense_data,
11152                     SSD_TYPE_NONE);
11153                 if (ua_type != CTL_UA_NONE) {
11154                         mtx_unlock(&lun->lun_lock);
11155                         ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
11156                         ctsio->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
11157                         ctsio->sense_len = SSD_FULL_SIZE;
11158                         ctl_done((union ctl_io *)ctsio);
11159                         return (retval);
11160                 }
11161         }
11162
11163
11164         if (ctl_scsiio_lun_check(lun, entry, ctsio) != 0) {
11165                 mtx_unlock(&lun->lun_lock);
11166                 ctl_done((union ctl_io *)ctsio);
11167                 return (retval);
11168         }
11169
11170         /*
11171          * XXX CHD this is where we want to send IO to other side if
11172          * this LUN is secondary on this SC. We will need to make a copy
11173          * of the IO and flag the IO on this side as SENT_2OTHER and the flag
11174          * the copy we send as FROM_OTHER.
11175          * We also need to stuff the address of the original IO so we can
11176          * find it easily. Something similar will need be done on the other
11177          * side so when we are done we can find the copy.
11178          */
11179         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
11180             (lun->flags & CTL_LUN_PEER_SC_PRIMARY) != 0 &&
11181             (entry->flags & CTL_CMD_FLAG_RUN_HERE) == 0) {
11182                 union ctl_ha_msg msg_info;
11183                 int isc_retval;
11184
11185                 ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11186                 ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11187                 mtx_unlock(&lun->lun_lock);
11188
11189                 msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
11190                 msg_info.hdr.original_sc = (union ctl_io *)ctsio;
11191                 msg_info.hdr.serializing_sc = NULL;
11192                 msg_info.hdr.nexus = ctsio->io_hdr.nexus;
11193                 msg_info.scsi.tag_num = ctsio->tag_num;
11194                 msg_info.scsi.tag_type = ctsio->tag_type;
11195                 msg_info.scsi.cdb_len = ctsio->cdb_len;
11196                 memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
11197
11198                 if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11199                     sizeof(msg_info.scsi) - sizeof(msg_info.scsi.sense_data),
11200                     M_WAITOK)) > CTL_HA_STATUS_SUCCESS) {
11201                         ctl_set_busy(ctsio);
11202                         ctl_done((union ctl_io *)ctsio);
11203                         return (retval);
11204                 }
11205                 return (retval);
11206         }
11207
11208         switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
11209                               (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr,
11210                               ctl_ooaq, ooa_links))) {
11211         case CTL_ACTION_BLOCK:
11212                 ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
11213                 TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
11214                                   blocked_links);
11215                 mtx_unlock(&lun->lun_lock);
11216                 return (retval);
11217         case CTL_ACTION_PASS:
11218         case CTL_ACTION_SKIP:
11219                 ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11220                 mtx_unlock(&lun->lun_lock);
11221                 ctl_enqueue_rtr((union ctl_io *)ctsio);
11222                 break;
11223         case CTL_ACTION_OVERLAP:
11224                 mtx_unlock(&lun->lun_lock);
11225                 ctl_set_overlapped_cmd(ctsio);
11226                 ctl_done((union ctl_io *)ctsio);
11227                 break;
11228         case CTL_ACTION_OVERLAP_TAG:
11229                 mtx_unlock(&lun->lun_lock);
11230                 ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
11231                 ctl_done((union ctl_io *)ctsio);
11232                 break;
11233         case CTL_ACTION_ERROR:
11234         default:
11235                 mtx_unlock(&lun->lun_lock);
11236                 ctl_set_internal_failure(ctsio,
11237                                          /*sks_valid*/ 0,
11238                                          /*retry_count*/ 0);
11239                 ctl_done((union ctl_io *)ctsio);
11240                 break;
11241         }
11242         return (retval);
11243 }
11244
11245 const struct ctl_cmd_entry *
11246 ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa)
11247 {
11248         const struct ctl_cmd_entry *entry;
11249         int service_action;
11250
11251         entry = &ctl_cmd_table[ctsio->cdb[0]];
11252         if (sa)
11253                 *sa = ((entry->flags & CTL_CMD_FLAG_SA5) != 0);
11254         if (entry->flags & CTL_CMD_FLAG_SA5) {
11255                 service_action = ctsio->cdb[1] & SERVICE_ACTION_MASK;
11256                 entry = &((const struct ctl_cmd_entry *)
11257                     entry->execute)[service_action];
11258         }
11259         return (entry);
11260 }
11261
11262 const struct ctl_cmd_entry *
11263 ctl_validate_command(struct ctl_scsiio *ctsio)
11264 {
11265         const struct ctl_cmd_entry *entry;
11266         int i, sa;
11267         uint8_t diff;
11268
11269         entry = ctl_get_cmd_entry(ctsio, &sa);
11270         if (entry->execute == NULL) {
11271                 if (sa)
11272                         ctl_set_invalid_field(ctsio,
11273                                               /*sks_valid*/ 1,
11274                                               /*command*/ 1,
11275                                               /*field*/ 1,
11276                                               /*bit_valid*/ 1,
11277                                               /*bit*/ 4);
11278                 else
11279                         ctl_set_invalid_opcode(ctsio);
11280                 ctl_done((union ctl_io *)ctsio);
11281                 return (NULL);
11282         }
11283         KASSERT(entry->length > 0,
11284             ("Not defined length for command 0x%02x/0x%02x",
11285              ctsio->cdb[0], ctsio->cdb[1]));
11286         for (i = 1; i < entry->length; i++) {
11287                 diff = ctsio->cdb[i] & ~entry->usage[i - 1];
11288                 if (diff == 0)
11289                         continue;
11290                 ctl_set_invalid_field(ctsio,
11291                                       /*sks_valid*/ 1,
11292                                       /*command*/ 1,
11293                                       /*field*/ i,
11294                                       /*bit_valid*/ 1,
11295                                       /*bit*/ fls(diff) - 1);
11296                 ctl_done((union ctl_io *)ctsio);
11297                 return (NULL);
11298         }
11299         return (entry);
11300 }
11301
11302 static int
11303 ctl_cmd_applicable(uint8_t lun_type, const struct ctl_cmd_entry *entry)
11304 {
11305
11306         switch (lun_type) {
11307         case T_PROCESSOR:
11308                 if ((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0)
11309                         return (0);
11310                 break;
11311         case T_DIRECT:
11312                 if ((entry->flags & CTL_CMD_FLAG_OK_ON_SLUN) == 0)
11313                         return (0);
11314                 break;
11315         default:
11316                 return (0);
11317         }
11318         return (1);
11319 }
11320
11321 static int
11322 ctl_scsiio(struct ctl_scsiio *ctsio)
11323 {
11324         int retval;
11325         const struct ctl_cmd_entry *entry;
11326
11327         retval = CTL_RETVAL_COMPLETE;
11328
11329         CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
11330
11331         entry = ctl_get_cmd_entry(ctsio, NULL);
11332
11333         /*
11334          * If this I/O has been aborted, just send it straight to
11335          * ctl_done() without executing it.
11336          */
11337         if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
11338                 ctl_done((union ctl_io *)ctsio);
11339                 goto bailout;
11340         }
11341
11342         /*
11343          * All the checks should have been handled by ctl_scsiio_precheck().
11344          * We should be clear now to just execute the I/O.
11345          */
11346         retval = entry->execute(ctsio);
11347
11348 bailout:
11349         return (retval);
11350 }
11351
11352 /*
11353  * Since we only implement one target right now, a bus reset simply resets
11354  * our single target.
11355  */
11356 static int
11357 ctl_bus_reset(struct ctl_softc *softc, union ctl_io *io)
11358 {
11359         return(ctl_target_reset(softc, io, CTL_UA_BUS_RESET));
11360 }
11361
11362 static int
11363 ctl_target_reset(struct ctl_softc *softc, union ctl_io *io,
11364                  ctl_ua_type ua_type)
11365 {
11366         struct ctl_port *port;
11367         struct ctl_lun *lun;
11368         int retval;
11369
11370         if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11371                 union ctl_ha_msg msg_info;
11372
11373                 msg_info.hdr.nexus = io->io_hdr.nexus;
11374                 if (ua_type==CTL_UA_TARG_RESET)
11375                         msg_info.task.task_action = CTL_TASK_TARGET_RESET;
11376                 else
11377                         msg_info.task.task_action = CTL_TASK_BUS_RESET;
11378                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11379                 msg_info.hdr.original_sc = NULL;
11380                 msg_info.hdr.serializing_sc = NULL;
11381                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11382                     sizeof(msg_info.task), M_WAITOK);
11383         }
11384         retval = 0;
11385
11386         mtx_lock(&softc->ctl_lock);
11387         port = ctl_io_port(&io->io_hdr);
11388         STAILQ_FOREACH(lun, &softc->lun_list, links) {
11389                 if (port != NULL &&
11390                     ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
11391                         continue;
11392                 retval += ctl_do_lun_reset(lun, io, ua_type);
11393         }
11394         mtx_unlock(&softc->ctl_lock);
11395         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11396         return (retval);
11397 }
11398
11399 /*
11400  * The LUN should always be set.  The I/O is optional, and is used to
11401  * distinguish between I/Os sent by this initiator, and by other
11402  * initiators.  We set unit attention for initiators other than this one.
11403  * SAM-3 is vague on this point.  It does say that a unit attention should
11404  * be established for other initiators when a LUN is reset (see section
11405  * 5.7.3), but it doesn't specifically say that the unit attention should
11406  * be established for this particular initiator when a LUN is reset.  Here
11407  * is the relevant text, from SAM-3 rev 8:
11408  *
11409  * 5.7.2 When a SCSI initiator port aborts its own tasks
11410  *
11411  * When a SCSI initiator port causes its own task(s) to be aborted, no
11412  * notification that the task(s) have been aborted shall be returned to
11413  * the SCSI initiator port other than the completion response for the
11414  * command or task management function action that caused the task(s) to
11415  * be aborted and notification(s) associated with related effects of the
11416  * action (e.g., a reset unit attention condition).
11417  *
11418  * XXX KDM for now, we're setting unit attention for all initiators.
11419  */
11420 static int
11421 ctl_do_lun_reset(struct ctl_lun *lun, union ctl_io *io, ctl_ua_type ua_type)
11422 {
11423         union ctl_io *xio;
11424 #if 0
11425         uint32_t initidx;
11426 #endif
11427 #ifdef CTL_WITH_CA
11428         int i;
11429 #endif
11430
11431         mtx_lock(&lun->lun_lock);
11432         /*
11433          * Run through the OOA queue and abort each I/O.
11434          */
11435         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11436              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11437                 xio->io_hdr.flags |= CTL_FLAG_ABORT | CTL_FLAG_ABORT_STATUS;
11438         }
11439
11440         /*
11441          * This version sets unit attention for every
11442          */
11443 #if 0
11444         initidx = ctl_get_initindex(&io->io_hdr.nexus);
11445         ctl_est_ua_all(lun, initidx, ua_type);
11446 #else
11447         ctl_est_ua_all(lun, -1, ua_type);
11448 #endif
11449
11450         /*
11451          * A reset (any kind, really) clears reservations established with
11452          * RESERVE/RELEASE.  It does not clear reservations established
11453          * with PERSISTENT RESERVE OUT, but we don't support that at the
11454          * moment anyway.  See SPC-2, section 5.6.  SPC-3 doesn't address
11455          * reservations made with the RESERVE/RELEASE commands, because
11456          * those commands are obsolete in SPC-3.
11457          */
11458         lun->flags &= ~CTL_LUN_RESERVED;
11459
11460 #ifdef CTL_WITH_CA
11461         for (i = 0; i < CTL_MAX_INITIATORS; i++)
11462                 ctl_clear_mask(lun->have_ca, i);
11463 #endif
11464         mtx_unlock(&lun->lun_lock);
11465
11466         return (0);
11467 }
11468
11469 static int
11470 ctl_lun_reset(struct ctl_softc *softc, union ctl_io *io)
11471 {
11472         struct ctl_lun *lun;
11473         uint32_t targ_lun;
11474         int retval;
11475
11476         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11477         mtx_lock(&softc->ctl_lock);
11478         if ((targ_lun >= CTL_MAX_LUNS) ||
11479             (lun = softc->ctl_luns[targ_lun]) == NULL) {
11480                 mtx_unlock(&softc->ctl_lock);
11481                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
11482                 return (1);
11483         }
11484         retval = ctl_do_lun_reset(lun, io, CTL_UA_LUN_RESET);
11485         mtx_unlock(&softc->ctl_lock);
11486         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11487
11488         if ((io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) == 0) {
11489                 union ctl_ha_msg msg_info;
11490
11491                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11492                 msg_info.hdr.nexus = io->io_hdr.nexus;
11493                 msg_info.task.task_action = CTL_TASK_LUN_RESET;
11494                 msg_info.hdr.original_sc = NULL;
11495                 msg_info.hdr.serializing_sc = NULL;
11496                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11497                     sizeof(msg_info.task), M_WAITOK);
11498         }
11499         return (retval);
11500 }
11501
11502 static void
11503 ctl_abort_tasks_lun(struct ctl_lun *lun, uint32_t targ_port, uint32_t init_id,
11504     int other_sc)
11505 {
11506         union ctl_io *xio;
11507
11508         mtx_assert(&lun->lun_lock, MA_OWNED);
11509
11510         /*
11511          * Run through the OOA queue and attempt to find the given I/O.
11512          * The target port, initiator ID, tag type and tag number have to
11513          * match the values that we got from the initiator.  If we have an
11514          * untagged command to abort, simply abort the first untagged command
11515          * we come to.  We only allow one untagged command at a time of course.
11516          */
11517         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11518              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11519
11520                 if ((targ_port == UINT32_MAX ||
11521                      targ_port == xio->io_hdr.nexus.targ_port) &&
11522                     (init_id == UINT32_MAX ||
11523                      init_id == xio->io_hdr.nexus.initid)) {
11524                         if (targ_port != xio->io_hdr.nexus.targ_port ||
11525                             init_id != xio->io_hdr.nexus.initid)
11526                                 xio->io_hdr.flags |= CTL_FLAG_ABORT_STATUS;
11527                         xio->io_hdr.flags |= CTL_FLAG_ABORT;
11528                         if (!other_sc && !(lun->flags & CTL_LUN_PRIMARY_SC)) {
11529                                 union ctl_ha_msg msg_info;
11530
11531                                 msg_info.hdr.nexus = xio->io_hdr.nexus;
11532                                 msg_info.task.task_action = CTL_TASK_ABORT_TASK;
11533                                 msg_info.task.tag_num = xio->scsiio.tag_num;
11534                                 msg_info.task.tag_type = xio->scsiio.tag_type;
11535                                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11536                                 msg_info.hdr.original_sc = NULL;
11537                                 msg_info.hdr.serializing_sc = NULL;
11538                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11539                                     sizeof(msg_info.task), M_NOWAIT);
11540                         }
11541                 }
11542         }
11543 }
11544
11545 static int
11546 ctl_abort_task_set(union ctl_io *io)
11547 {
11548         struct ctl_softc *softc = control_softc;
11549         struct ctl_lun *lun;
11550         uint32_t targ_lun;
11551
11552         /*
11553          * Look up the LUN.
11554          */
11555         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11556         mtx_lock(&softc->ctl_lock);
11557         if ((targ_lun >= CTL_MAX_LUNS) ||
11558             (lun = softc->ctl_luns[targ_lun]) == NULL) {
11559                 mtx_unlock(&softc->ctl_lock);
11560                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
11561                 return (1);
11562         }
11563
11564         mtx_lock(&lun->lun_lock);
11565         mtx_unlock(&softc->ctl_lock);
11566         if (io->taskio.task_action == CTL_TASK_ABORT_TASK_SET) {
11567                 ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
11568                     io->io_hdr.nexus.initid,
11569                     (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
11570         } else { /* CTL_TASK_CLEAR_TASK_SET */
11571                 ctl_abort_tasks_lun(lun, UINT32_MAX, UINT32_MAX,
11572                     (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
11573         }
11574         mtx_unlock(&lun->lun_lock);
11575         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11576         return (0);
11577 }
11578
11579 static int
11580 ctl_i_t_nexus_reset(union ctl_io *io)
11581 {
11582         struct ctl_softc *softc = control_softc;
11583         struct ctl_lun *lun;
11584         uint32_t initidx;
11585
11586         if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11587                 union ctl_ha_msg msg_info;
11588
11589                 msg_info.hdr.nexus = io->io_hdr.nexus;
11590                 msg_info.task.task_action = CTL_TASK_I_T_NEXUS_RESET;
11591                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11592                 msg_info.hdr.original_sc = NULL;
11593                 msg_info.hdr.serializing_sc = NULL;
11594                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11595                     sizeof(msg_info.task), M_WAITOK);
11596         }
11597
11598         initidx = ctl_get_initindex(&io->io_hdr.nexus);
11599         mtx_lock(&softc->ctl_lock);
11600         STAILQ_FOREACH(lun, &softc->lun_list, links) {
11601                 mtx_lock(&lun->lun_lock);
11602                 ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
11603                     io->io_hdr.nexus.initid, 1);
11604 #ifdef CTL_WITH_CA
11605                 ctl_clear_mask(lun->have_ca, initidx);
11606 #endif
11607                 if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == initidx))
11608                         lun->flags &= ~CTL_LUN_RESERVED;
11609                 ctl_est_ua(lun, initidx, CTL_UA_I_T_NEXUS_LOSS);
11610                 mtx_unlock(&lun->lun_lock);
11611         }
11612         mtx_unlock(&softc->ctl_lock);
11613         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11614         return (0);
11615 }
11616
11617 static int
11618 ctl_abort_task(union ctl_io *io)
11619 {
11620         union ctl_io *xio;
11621         struct ctl_lun *lun;
11622         struct ctl_softc *softc;
11623 #if 0
11624         struct sbuf sb;
11625         char printbuf[128];
11626 #endif
11627         int found;
11628         uint32_t targ_lun;
11629
11630         softc = control_softc;
11631         found = 0;
11632
11633         /*
11634          * Look up the LUN.
11635          */
11636         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11637         mtx_lock(&softc->ctl_lock);
11638         if ((targ_lun >= CTL_MAX_LUNS) ||
11639             (lun = softc->ctl_luns[targ_lun]) == NULL) {
11640                 mtx_unlock(&softc->ctl_lock);
11641                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
11642                 return (1);
11643         }
11644
11645 #if 0
11646         printf("ctl_abort_task: called for lun %lld, tag %d type %d\n",
11647                lun->lun, io->taskio.tag_num, io->taskio.tag_type);
11648 #endif
11649
11650         mtx_lock(&lun->lun_lock);
11651         mtx_unlock(&softc->ctl_lock);
11652         /*
11653          * Run through the OOA queue and attempt to find the given I/O.
11654          * The target port, initiator ID, tag type and tag number have to
11655          * match the values that we got from the initiator.  If we have an
11656          * untagged command to abort, simply abort the first untagged command
11657          * we come to.  We only allow one untagged command at a time of course.
11658          */
11659         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11660              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11661 #if 0
11662                 sbuf_new(&sb, printbuf, sizeof(printbuf), SBUF_FIXEDLEN);
11663
11664                 sbuf_printf(&sb, "LUN %lld tag %d type %d%s%s%s%s: ",
11665                             lun->lun, xio->scsiio.tag_num,
11666                             xio->scsiio.tag_type,
11667                             (xio->io_hdr.blocked_links.tqe_prev
11668                             == NULL) ? "" : " BLOCKED",
11669                             (xio->io_hdr.flags &
11670                             CTL_FLAG_DMA_INPROG) ? " DMA" : "",
11671                             (xio->io_hdr.flags &
11672                             CTL_FLAG_ABORT) ? " ABORT" : "",
11673                             (xio->io_hdr.flags &
11674                             CTL_FLAG_IS_WAS_ON_RTR ? " RTR" : ""));
11675                 ctl_scsi_command_string(&xio->scsiio, NULL, &sb);
11676                 sbuf_finish(&sb);
11677                 printf("%s\n", sbuf_data(&sb));
11678 #endif
11679
11680                 if ((xio->io_hdr.nexus.targ_port != io->io_hdr.nexus.targ_port)
11681                  || (xio->io_hdr.nexus.initid != io->io_hdr.nexus.initid)
11682                  || (xio->io_hdr.flags & CTL_FLAG_ABORT))
11683                         continue;
11684
11685                 /*
11686                  * If the abort says that the task is untagged, the
11687                  * task in the queue must be untagged.  Otherwise,
11688                  * we just check to see whether the tag numbers
11689                  * match.  This is because the QLogic firmware
11690                  * doesn't pass back the tag type in an abort
11691                  * request.
11692                  */
11693 #if 0
11694                 if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
11695                   && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
11696                  || (xio->scsiio.tag_num == io->taskio.tag_num))
11697 #endif
11698                 /*
11699                  * XXX KDM we've got problems with FC, because it
11700                  * doesn't send down a tag type with aborts.  So we
11701                  * can only really go by the tag number...
11702                  * This may cause problems with parallel SCSI.
11703                  * Need to figure that out!!
11704                  */
11705                 if (xio->scsiio.tag_num == io->taskio.tag_num) {
11706                         xio->io_hdr.flags |= CTL_FLAG_ABORT;
11707                         found = 1;
11708                         if ((io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) == 0 &&
11709                             !(lun->flags & CTL_LUN_PRIMARY_SC)) {
11710                                 union ctl_ha_msg msg_info;
11711
11712                                 msg_info.hdr.nexus = io->io_hdr.nexus;
11713                                 msg_info.task.task_action = CTL_TASK_ABORT_TASK;
11714                                 msg_info.task.tag_num = io->taskio.tag_num;
11715                                 msg_info.task.tag_type = io->taskio.tag_type;
11716                                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11717                                 msg_info.hdr.original_sc = NULL;
11718                                 msg_info.hdr.serializing_sc = NULL;
11719 #if 0
11720                                 printf("Sent Abort to other side\n");
11721 #endif
11722                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11723                                     sizeof(msg_info.task), M_NOWAIT);
11724                         }
11725 #if 0
11726                         printf("ctl_abort_task: found I/O to abort\n");
11727 #endif
11728                 }
11729         }
11730         mtx_unlock(&lun->lun_lock);
11731
11732         if (found == 0) {
11733                 /*
11734                  * This isn't really an error.  It's entirely possible for
11735                  * the abort and command completion to cross on the wire.
11736                  * This is more of an informative/diagnostic error.
11737                  */
11738 #if 0
11739                 printf("ctl_abort_task: ABORT sent for nonexistent I/O: "
11740                        "%u:%u:%u tag %d type %d\n",
11741                        io->io_hdr.nexus.initid,
11742                        io->io_hdr.nexus.targ_port,
11743                        io->io_hdr.nexus.targ_lun, io->taskio.tag_num,
11744                        io->taskio.tag_type);
11745 #endif
11746         }
11747         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11748         return (0);
11749 }
11750
11751 static int
11752 ctl_query_task(union ctl_io *io, int task_set)
11753 {
11754         union ctl_io *xio;
11755         struct ctl_lun *lun;
11756         struct ctl_softc *softc;
11757         int found = 0;
11758         uint32_t targ_lun;
11759
11760         softc = control_softc;
11761         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11762         mtx_lock(&softc->ctl_lock);
11763         if ((targ_lun >= CTL_MAX_LUNS) ||
11764             (lun = softc->ctl_luns[targ_lun]) == NULL) {
11765                 mtx_unlock(&softc->ctl_lock);
11766                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
11767                 return (1);
11768         }
11769         mtx_lock(&lun->lun_lock);
11770         mtx_unlock(&softc->ctl_lock);
11771         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11772              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11773
11774                 if ((xio->io_hdr.nexus.targ_port != io->io_hdr.nexus.targ_port)
11775                  || (xio->io_hdr.nexus.initid != io->io_hdr.nexus.initid)
11776                  || (xio->io_hdr.flags & CTL_FLAG_ABORT))
11777                         continue;
11778
11779                 if (task_set || xio->scsiio.tag_num == io->taskio.tag_num) {
11780                         found = 1;
11781                         break;
11782                 }
11783         }
11784         mtx_unlock(&lun->lun_lock);
11785         if (found)
11786                 io->taskio.task_status = CTL_TASK_FUNCTION_SUCCEEDED;
11787         else
11788                 io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11789         return (0);
11790 }
11791
11792 static int
11793 ctl_query_async_event(union ctl_io *io)
11794 {
11795         struct ctl_lun *lun;
11796         struct ctl_softc *softc;
11797         ctl_ua_type ua;
11798         uint32_t targ_lun, initidx;
11799
11800         softc = control_softc;
11801         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11802         mtx_lock(&softc->ctl_lock);
11803         if ((targ_lun >= CTL_MAX_LUNS) ||
11804             (lun = softc->ctl_luns[targ_lun]) == NULL) {
11805                 mtx_unlock(&softc->ctl_lock);
11806                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
11807                 return (1);
11808         }
11809         mtx_lock(&lun->lun_lock);
11810         mtx_unlock(&softc->ctl_lock);
11811         initidx = ctl_get_initindex(&io->io_hdr.nexus);
11812         ua = ctl_build_qae(lun, initidx, io->taskio.task_resp);
11813         mtx_unlock(&lun->lun_lock);
11814         if (ua != CTL_UA_NONE)
11815                 io->taskio.task_status = CTL_TASK_FUNCTION_SUCCEEDED;
11816         else
11817                 io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11818         return (0);
11819 }
11820
11821 static void
11822 ctl_run_task(union ctl_io *io)
11823 {
11824         struct ctl_softc *softc = control_softc;
11825         int retval = 1;
11826
11827         CTL_DEBUG_PRINT(("ctl_run_task\n"));
11828         KASSERT(io->io_hdr.io_type == CTL_IO_TASK,
11829             ("ctl_run_task: Unextected io_type %d\n", io->io_hdr.io_type));
11830         io->taskio.task_status = CTL_TASK_FUNCTION_NOT_SUPPORTED;
11831         bzero(io->taskio.task_resp, sizeof(io->taskio.task_resp));
11832         switch (io->taskio.task_action) {
11833         case CTL_TASK_ABORT_TASK:
11834                 retval = ctl_abort_task(io);
11835                 break;
11836         case CTL_TASK_ABORT_TASK_SET:
11837         case CTL_TASK_CLEAR_TASK_SET:
11838                 retval = ctl_abort_task_set(io);
11839                 break;
11840         case CTL_TASK_CLEAR_ACA:
11841                 break;
11842         case CTL_TASK_I_T_NEXUS_RESET:
11843                 retval = ctl_i_t_nexus_reset(io);
11844                 break;
11845         case CTL_TASK_LUN_RESET:
11846                 retval = ctl_lun_reset(softc, io);
11847                 break;
11848         case CTL_TASK_TARGET_RESET:
11849                 retval = ctl_target_reset(softc, io, CTL_UA_TARG_RESET);
11850                 break;
11851         case CTL_TASK_BUS_RESET:
11852                 retval = ctl_bus_reset(softc, io);
11853                 break;
11854         case CTL_TASK_PORT_LOGIN:
11855                 break;
11856         case CTL_TASK_PORT_LOGOUT:
11857                 break;
11858         case CTL_TASK_QUERY_TASK:
11859                 retval = ctl_query_task(io, 0);
11860                 break;
11861         case CTL_TASK_QUERY_TASK_SET:
11862                 retval = ctl_query_task(io, 1);
11863                 break;
11864         case CTL_TASK_QUERY_ASYNC_EVENT:
11865                 retval = ctl_query_async_event(io);
11866                 break;
11867         default:
11868                 printf("%s: got unknown task management event %d\n",
11869                        __func__, io->taskio.task_action);
11870                 break;
11871         }
11872         if (retval == 0)
11873                 io->io_hdr.status = CTL_SUCCESS;
11874         else
11875                 io->io_hdr.status = CTL_ERROR;
11876         ctl_done(io);
11877 }
11878
11879 /*
11880  * For HA operation.  Handle commands that come in from the other
11881  * controller.
11882  */
11883 static void
11884 ctl_handle_isc(union ctl_io *io)
11885 {
11886         int free_io;
11887         struct ctl_lun *lun;
11888         struct ctl_softc *softc = control_softc;
11889         uint32_t targ_lun;
11890
11891         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11892         lun = softc->ctl_luns[targ_lun];
11893
11894         switch (io->io_hdr.msg_type) {
11895         case CTL_MSG_SERIALIZE:
11896                 free_io = ctl_serialize_other_sc_cmd(&io->scsiio);
11897                 break;
11898         case CTL_MSG_R2R: {
11899                 const struct ctl_cmd_entry *entry;
11900
11901                 /*
11902                  * This is only used in SER_ONLY mode.
11903                  */
11904                 free_io = 0;
11905                 entry = ctl_get_cmd_entry(&io->scsiio, NULL);
11906                 mtx_lock(&lun->lun_lock);
11907                 if (ctl_scsiio_lun_check(lun,
11908                     entry, (struct ctl_scsiio *)io) != 0) {
11909                         mtx_unlock(&lun->lun_lock);
11910                         ctl_done(io);
11911                         break;
11912                 }
11913                 io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11914                 mtx_unlock(&lun->lun_lock);
11915                 ctl_enqueue_rtr(io);
11916                 break;
11917         }
11918         case CTL_MSG_FINISH_IO:
11919                 if (softc->ha_mode == CTL_HA_MODE_XFER) {
11920                         free_io = 0;
11921                         ctl_done(io);
11922                 } else {
11923                         free_io = 1;
11924                         mtx_lock(&lun->lun_lock);
11925                         TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr,
11926                                      ooa_links);
11927                         ctl_check_blocked(lun);
11928                         mtx_unlock(&lun->lun_lock);
11929                 }
11930                 break;
11931         case CTL_MSG_PERS_ACTION:
11932                 ctl_hndl_per_res_out_on_other_sc(
11933                         (union ctl_ha_msg *)&io->presio.pr_msg);
11934                 free_io = 1;
11935                 break;
11936         case CTL_MSG_BAD_JUJU:
11937                 free_io = 0;
11938                 ctl_done(io);
11939                 break;
11940         case CTL_MSG_DATAMOVE:
11941                 /* Only used in XFER mode */
11942                 free_io = 0;
11943                 ctl_datamove_remote(io);
11944                 break;
11945         case CTL_MSG_DATAMOVE_DONE:
11946                 /* Only used in XFER mode */
11947                 free_io = 0;
11948                 io->scsiio.be_move_done(io);
11949                 break;
11950         case CTL_MSG_FAILOVER:
11951                 ctl_failover_lun(io);
11952                 free_io = 1;
11953                 break;
11954         default:
11955                 free_io = 1;
11956                 printf("%s: Invalid message type %d\n",
11957                        __func__, io->io_hdr.msg_type);
11958                 break;
11959         }
11960         if (free_io)
11961                 ctl_free_io(io);
11962
11963 }
11964
11965
11966 /*
11967  * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
11968  * there is no match.
11969  */
11970 static ctl_lun_error_pattern
11971 ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
11972 {
11973         const struct ctl_cmd_entry *entry;
11974         ctl_lun_error_pattern filtered_pattern, pattern;
11975
11976         pattern = desc->error_pattern;
11977
11978         /*
11979          * XXX KDM we need more data passed into this function to match a
11980          * custom pattern, and we actually need to implement custom pattern
11981          * matching.
11982          */
11983         if (pattern & CTL_LUN_PAT_CMD)
11984                 return (CTL_LUN_PAT_CMD);
11985
11986         if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
11987                 return (CTL_LUN_PAT_ANY);
11988
11989         entry = ctl_get_cmd_entry(ctsio, NULL);
11990
11991         filtered_pattern = entry->pattern & pattern;
11992
11993         /*
11994          * If the user requested specific flags in the pattern (e.g.
11995          * CTL_LUN_PAT_RANGE), make sure the command supports all of those
11996          * flags.
11997          *
11998          * If the user did not specify any flags, it doesn't matter whether
11999          * or not the command supports the flags.
12000          */
12001         if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
12002              (pattern & ~CTL_LUN_PAT_MASK))
12003                 return (CTL_LUN_PAT_NONE);
12004
12005         /*
12006          * If the user asked for a range check, see if the requested LBA
12007          * range overlaps with this command's LBA range.
12008          */
12009         if (filtered_pattern & CTL_LUN_PAT_RANGE) {
12010                 uint64_t lba1;
12011                 uint64_t len1;
12012                 ctl_action action;
12013                 int retval;
12014
12015                 retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
12016                 if (retval != 0)
12017                         return (CTL_LUN_PAT_NONE);
12018
12019                 action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
12020                                               desc->lba_range.len, FALSE);
12021                 /*
12022                  * A "pass" means that the LBA ranges don't overlap, so
12023                  * this doesn't match the user's range criteria.
12024                  */
12025                 if (action == CTL_ACTION_PASS)
12026                         return (CTL_LUN_PAT_NONE);
12027         }
12028
12029         return (filtered_pattern);
12030 }
12031
12032 static void
12033 ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
12034 {
12035         struct ctl_error_desc *desc, *desc2;
12036
12037         mtx_assert(&lun->lun_lock, MA_OWNED);
12038
12039         STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
12040                 ctl_lun_error_pattern pattern;
12041                 /*
12042                  * Check to see whether this particular command matches
12043                  * the pattern in the descriptor.
12044                  */
12045                 pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
12046                 if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
12047                         continue;
12048
12049                 switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
12050                 case CTL_LUN_INJ_ABORTED:
12051                         ctl_set_aborted(&io->scsiio);
12052                         break;
12053                 case CTL_LUN_INJ_MEDIUM_ERR:
12054                         ctl_set_medium_error(&io->scsiio,
12055                             (io->io_hdr.flags & CTL_FLAG_DATA_MASK) !=
12056                              CTL_FLAG_DATA_OUT);
12057                         break;
12058                 case CTL_LUN_INJ_UA:
12059                         /* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
12060                          * OCCURRED */
12061                         ctl_set_ua(&io->scsiio, 0x29, 0x00);
12062                         break;
12063                 case CTL_LUN_INJ_CUSTOM:
12064                         /*
12065                          * We're assuming the user knows what he is doing.
12066                          * Just copy the sense information without doing
12067                          * checks.
12068                          */
12069                         bcopy(&desc->custom_sense, &io->scsiio.sense_data,
12070                               MIN(sizeof(desc->custom_sense),
12071                                   sizeof(io->scsiio.sense_data)));
12072                         io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
12073                         io->scsiio.sense_len = SSD_FULL_SIZE;
12074                         io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
12075                         break;
12076                 case CTL_LUN_INJ_NONE:
12077                 default:
12078                         /*
12079                          * If this is an error injection type we don't know
12080                          * about, clear the continuous flag (if it is set)
12081                          * so it will get deleted below.
12082                          */
12083                         desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
12084                         break;
12085                 }
12086                 /*
12087                  * By default, each error injection action is a one-shot
12088                  */
12089                 if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
12090                         continue;
12091
12092                 STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
12093
12094                 free(desc, M_CTL);
12095         }
12096 }
12097
12098 #ifdef CTL_IO_DELAY
12099 static void
12100 ctl_datamove_timer_wakeup(void *arg)
12101 {
12102         union ctl_io *io;
12103
12104         io = (union ctl_io *)arg;
12105
12106         ctl_datamove(io);
12107 }
12108 #endif /* CTL_IO_DELAY */
12109
12110 void
12111 ctl_datamove(union ctl_io *io)
12112 {
12113         struct ctl_lun *lun;
12114         void (*fe_datamove)(union ctl_io *io);
12115
12116         mtx_assert(&control_softc->ctl_lock, MA_NOTOWNED);
12117
12118         CTL_DEBUG_PRINT(("ctl_datamove\n"));
12119
12120         lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12121 #ifdef CTL_TIME_IO
12122         if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12123                 char str[256];
12124                 char path_str[64];
12125                 struct sbuf sb;
12126
12127                 ctl_scsi_path_string(io, path_str, sizeof(path_str));
12128                 sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12129
12130                 sbuf_cat(&sb, path_str);
12131                 switch (io->io_hdr.io_type) {
12132                 case CTL_IO_SCSI:
12133                         ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12134                         sbuf_printf(&sb, "\n");
12135                         sbuf_cat(&sb, path_str);
12136                         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12137                                     io->scsiio.tag_num, io->scsiio.tag_type);
12138                         break;
12139                 case CTL_IO_TASK:
12140                         sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12141                                     "Tag Type: %d\n", io->taskio.task_action,
12142                                     io->taskio.tag_num, io->taskio.tag_type);
12143                         break;
12144                 default:
12145                         printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12146                         panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12147                         break;
12148                 }
12149                 sbuf_cat(&sb, path_str);
12150                 sbuf_printf(&sb, "ctl_datamove: %jd seconds\n",
12151                             (intmax_t)time_uptime - io->io_hdr.start_time);
12152                 sbuf_finish(&sb);
12153                 printf("%s", sbuf_data(&sb));
12154         }
12155 #endif /* CTL_TIME_IO */
12156
12157 #ifdef CTL_IO_DELAY
12158         if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
12159                 io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
12160         } else {
12161                 if ((lun != NULL)
12162                  && (lun->delay_info.datamove_delay > 0)) {
12163
12164                         callout_init(&io->io_hdr.delay_callout, /*mpsafe*/ 1);
12165                         io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
12166                         callout_reset(&io->io_hdr.delay_callout,
12167                                       lun->delay_info.datamove_delay * hz,
12168                                       ctl_datamove_timer_wakeup, io);
12169                         if (lun->delay_info.datamove_type ==
12170                             CTL_DELAY_TYPE_ONESHOT)
12171                                 lun->delay_info.datamove_delay = 0;
12172                         return;
12173                 }
12174         }
12175 #endif
12176
12177         /*
12178          * This command has been aborted.  Set the port status, so we fail
12179          * the data move.
12180          */
12181         if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12182                 printf("ctl_datamove: tag 0x%04x on (%u:%u:%u) aborted\n",
12183                        io->scsiio.tag_num, io->io_hdr.nexus.initid,
12184                        io->io_hdr.nexus.targ_port,
12185                        io->io_hdr.nexus.targ_lun);
12186                 io->io_hdr.port_status = 31337;
12187                 /*
12188                  * Note that the backend, in this case, will get the
12189                  * callback in its context.  In other cases it may get
12190                  * called in the frontend's interrupt thread context.
12191                  */
12192                 io->scsiio.be_move_done(io);
12193                 return;
12194         }
12195
12196         /* Don't confuse frontend with zero length data move. */
12197         if (io->scsiio.kern_data_len == 0) {
12198                 io->scsiio.be_move_done(io);
12199                 return;
12200         }
12201
12202         fe_datamove = ctl_io_port(&io->io_hdr)->fe_datamove;
12203         fe_datamove(io);
12204 }
12205
12206 static void
12207 ctl_send_datamove_done(union ctl_io *io, int have_lock)
12208 {
12209         union ctl_ha_msg msg;
12210 #ifdef CTL_TIME_IO
12211         struct bintime cur_bt;
12212 #endif
12213
12214         memset(&msg, 0, sizeof(msg));
12215         msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
12216         msg.hdr.original_sc = io;
12217         msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
12218         msg.hdr.nexus = io->io_hdr.nexus;
12219         msg.hdr.status = io->io_hdr.status;
12220         msg.scsi.tag_num = io->scsiio.tag_num;
12221         msg.scsi.tag_type = io->scsiio.tag_type;
12222         msg.scsi.scsi_status = io->scsiio.scsi_status;
12223         memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
12224                io->scsiio.sense_len);
12225         msg.scsi.sense_len = io->scsiio.sense_len;
12226         msg.scsi.sense_residual = io->scsiio.sense_residual;
12227         msg.scsi.fetd_status = io->io_hdr.port_status;
12228         msg.scsi.residual = io->scsiio.residual;
12229         io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12230         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12231                 ctl_failover_io(io, /*have_lock*/ have_lock);
12232                 return;
12233         }
12234         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12235             sizeof(msg.scsi) - sizeof(msg.scsi.sense_data) +
12236             msg.scsi.sense_len, M_WAITOK);
12237
12238 #ifdef CTL_TIME_IO
12239         getbinuptime(&cur_bt);
12240         bintime_sub(&cur_bt, &io->io_hdr.dma_start_bt);
12241         bintime_add(&io->io_hdr.dma_bt, &cur_bt);
12242 #endif
12243         io->io_hdr.num_dmas++;
12244 }
12245
12246 /*
12247  * The DMA to the remote side is done, now we need to tell the other side
12248  * we're done so it can continue with its data movement.
12249  */
12250 static void
12251 ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
12252 {
12253         union ctl_io *io;
12254         int i;
12255
12256         io = rq->context;
12257
12258         if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12259                 printf("%s: ISC DMA write failed with error %d", __func__,
12260                        rq->ret);
12261                 ctl_set_internal_failure(&io->scsiio,
12262                                          /*sks_valid*/ 1,
12263                                          /*retry_count*/ rq->ret);
12264         }
12265
12266         ctl_dt_req_free(rq);
12267
12268         for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12269                 free(io->io_hdr.local_sglist[i].addr, M_CTL);
12270         free(io->io_hdr.remote_sglist, M_CTL);
12271         io->io_hdr.remote_sglist = NULL;
12272         io->io_hdr.local_sglist = NULL;
12273
12274         /*
12275          * The data is in local and remote memory, so now we need to send
12276          * status (good or back) back to the other side.
12277          */
12278         ctl_send_datamove_done(io, /*have_lock*/ 0);
12279 }
12280
12281 /*
12282  * We've moved the data from the host/controller into local memory.  Now we
12283  * need to push it over to the remote controller's memory.
12284  */
12285 static int
12286 ctl_datamove_remote_dm_write_cb(union ctl_io *io)
12287 {
12288         int retval;
12289
12290         retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
12291                                           ctl_datamove_remote_write_cb);
12292         return (retval);
12293 }
12294
12295 static void
12296 ctl_datamove_remote_write(union ctl_io *io)
12297 {
12298         int retval;
12299         void (*fe_datamove)(union ctl_io *io);
12300
12301         /*
12302          * - Get the data from the host/HBA into local memory.
12303          * - DMA memory from the local controller to the remote controller.
12304          * - Send status back to the remote controller.
12305          */
12306
12307         retval = ctl_datamove_remote_sgl_setup(io);
12308         if (retval != 0)
12309                 return;
12310
12311         /* Switch the pointer over so the FETD knows what to do */
12312         io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12313
12314         /*
12315          * Use a custom move done callback, since we need to send completion
12316          * back to the other controller, not to the backend on this side.
12317          */
12318         io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
12319
12320         fe_datamove = ctl_io_port(&io->io_hdr)->fe_datamove;
12321         fe_datamove(io);
12322 }
12323
12324 static int
12325 ctl_datamove_remote_dm_read_cb(union ctl_io *io)
12326 {
12327 #if 0
12328         char str[256];
12329         char path_str[64];
12330         struct sbuf sb;
12331 #endif
12332         int i;
12333
12334         for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12335                 free(io->io_hdr.local_sglist[i].addr, M_CTL);
12336         free(io->io_hdr.remote_sglist, M_CTL);
12337         io->io_hdr.remote_sglist = NULL;
12338         io->io_hdr.local_sglist = NULL;
12339
12340 #if 0
12341         scsi_path_string(io, path_str, sizeof(path_str));
12342         sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12343         sbuf_cat(&sb, path_str);
12344         scsi_command_string(&io->scsiio, NULL, &sb);
12345         sbuf_printf(&sb, "\n");
12346         sbuf_cat(&sb, path_str);
12347         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12348                     io->scsiio.tag_num, io->scsiio.tag_type);
12349         sbuf_cat(&sb, path_str);
12350         sbuf_printf(&sb, "%s: flags %#x, status %#x\n", __func__,
12351                     io->io_hdr.flags, io->io_hdr.status);
12352         sbuf_finish(&sb);
12353         printk("%s", sbuf_data(&sb));
12354 #endif
12355
12356
12357         /*
12358          * The read is done, now we need to send status (good or bad) back
12359          * to the other side.
12360          */
12361         ctl_send_datamove_done(io, /*have_lock*/ 0);
12362
12363         return (0);
12364 }
12365
12366 static void
12367 ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
12368 {
12369         union ctl_io *io;
12370         void (*fe_datamove)(union ctl_io *io);
12371
12372         io = rq->context;
12373
12374         if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12375                 printf("%s: ISC DMA read failed with error %d\n", __func__,
12376                        rq->ret);
12377                 ctl_set_internal_failure(&io->scsiio,
12378                                          /*sks_valid*/ 1,
12379                                          /*retry_count*/ rq->ret);
12380         }
12381
12382         ctl_dt_req_free(rq);
12383
12384         /* Switch the pointer over so the FETD knows what to do */
12385         io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12386
12387         /*
12388          * Use a custom move done callback, since we need to send completion
12389          * back to the other controller, not to the backend on this side.
12390          */
12391         io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
12392
12393         /* XXX KDM add checks like the ones in ctl_datamove? */
12394
12395         fe_datamove = ctl_io_port(&io->io_hdr)->fe_datamove;
12396         fe_datamove(io);
12397 }
12398
12399 static int
12400 ctl_datamove_remote_sgl_setup(union ctl_io *io)
12401 {
12402         struct ctl_sg_entry *local_sglist;
12403         uint32_t len_to_go;
12404         int retval;
12405         int i;
12406
12407         retval = 0;
12408         local_sglist = io->io_hdr.local_sglist;
12409         len_to_go = io->scsiio.kern_data_len;
12410
12411         /*
12412          * The difficult thing here is that the size of the various
12413          * S/G segments may be different than the size from the
12414          * remote controller.  That'll make it harder when DMAing
12415          * the data back to the other side.
12416          */
12417         for (i = 0; len_to_go > 0; i++) {
12418                 local_sglist[i].len = MIN(len_to_go, CTL_HA_DATAMOVE_SEGMENT);
12419                 local_sglist[i].addr =
12420                     malloc(local_sglist[i].len, M_CTL, M_WAITOK);
12421
12422                 len_to_go -= local_sglist[i].len;
12423         }
12424         /*
12425          * Reset the number of S/G entries accordingly.  The original
12426          * number of S/G entries is available in rem_sg_entries.
12427          */
12428         io->scsiio.kern_sg_entries = i;
12429
12430 #if 0
12431         printf("%s: kern_sg_entries = %d\n", __func__,
12432                io->scsiio.kern_sg_entries);
12433         for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12434                 printf("%s: sg[%d] = %p, %lu\n", __func__, i,
12435                        local_sglist[i].addr, local_sglist[i].len);
12436 #endif
12437
12438         return (retval);
12439 }
12440
12441 static int
12442 ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
12443                          ctl_ha_dt_cb callback)
12444 {
12445         struct ctl_ha_dt_req *rq;
12446         struct ctl_sg_entry *remote_sglist, *local_sglist;
12447         uint32_t local_used, remote_used, total_used;
12448         int i, j, isc_ret;
12449
12450         rq = ctl_dt_req_alloc();
12451
12452         /*
12453          * If we failed to allocate the request, and if the DMA didn't fail
12454          * anyway, set busy status.  This is just a resource allocation
12455          * failure.
12456          */
12457         if ((rq == NULL)
12458          && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
12459              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS))
12460                 ctl_set_busy(&io->scsiio);
12461
12462         if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
12463             (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS) {
12464
12465                 if (rq != NULL)
12466                         ctl_dt_req_free(rq);
12467
12468                 /*
12469                  * The data move failed.  We need to return status back
12470                  * to the other controller.  No point in trying to DMA
12471                  * data to the remote controller.
12472                  */
12473
12474                 ctl_send_datamove_done(io, /*have_lock*/ 0);
12475
12476                 return (1);
12477         }
12478
12479         local_sglist = io->io_hdr.local_sglist;
12480         remote_sglist = io->io_hdr.remote_sglist;
12481         local_used = 0;
12482         remote_used = 0;
12483         total_used = 0;
12484
12485         /*
12486          * Pull/push the data over the wire from/to the other controller.
12487          * This takes into account the possibility that the local and
12488          * remote sglists may not be identical in terms of the size of
12489          * the elements and the number of elements.
12490          *
12491          * One fundamental assumption here is that the length allocated for
12492          * both the local and remote sglists is identical.  Otherwise, we've
12493          * essentially got a coding error of some sort.
12494          */
12495         isc_ret = CTL_HA_STATUS_SUCCESS;
12496         for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
12497                 uint32_t cur_len;
12498                 uint8_t *tmp_ptr;
12499
12500                 rq->command = command;
12501                 rq->context = io;
12502
12503                 /*
12504                  * Both pointers should be aligned.  But it is possible
12505                  * that the allocation length is not.  They should both
12506                  * also have enough slack left over at the end, though,
12507                  * to round up to the next 8 byte boundary.
12508                  */
12509                 cur_len = MIN(local_sglist[i].len - local_used,
12510                               remote_sglist[j].len - remote_used);
12511                 rq->size = cur_len;
12512
12513                 tmp_ptr = (uint8_t *)local_sglist[i].addr;
12514                 tmp_ptr += local_used;
12515
12516 #if 0
12517                 /* Use physical addresses when talking to ISC hardware */
12518                 if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
12519                         /* XXX KDM use busdma */
12520                         rq->local = vtophys(tmp_ptr);
12521                 } else
12522                         rq->local = tmp_ptr;
12523 #else
12524                 KASSERT((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0,
12525                     ("HA does not support BUS_ADDR"));
12526                 rq->local = tmp_ptr;
12527 #endif
12528
12529                 tmp_ptr = (uint8_t *)remote_sglist[j].addr;
12530                 tmp_ptr += remote_used;
12531                 rq->remote = tmp_ptr;
12532
12533                 rq->callback = NULL;
12534
12535                 local_used += cur_len;
12536                 if (local_used >= local_sglist[i].len) {
12537                         i++;
12538                         local_used = 0;
12539                 }
12540
12541                 remote_used += cur_len;
12542                 if (remote_used >= remote_sglist[j].len) {
12543                         j++;
12544                         remote_used = 0;
12545                 }
12546                 total_used += cur_len;
12547
12548                 if (total_used >= io->scsiio.kern_data_len)
12549                         rq->callback = callback;
12550
12551 #if 0
12552                 printf("%s: %s: local %p remote %p size %d\n", __func__,
12553                        (command == CTL_HA_DT_CMD_WRITE) ? "WRITE" : "READ",
12554                        rq->local, rq->remote, rq->size);
12555 #endif
12556
12557                 isc_ret = ctl_dt_single(rq);
12558                 if (isc_ret > CTL_HA_STATUS_SUCCESS)
12559                         break;
12560         }
12561         if (isc_ret != CTL_HA_STATUS_WAIT) {
12562                 rq->ret = isc_ret;
12563                 callback(rq);
12564         }
12565
12566         return (0);
12567 }
12568
12569 static void
12570 ctl_datamove_remote_read(union ctl_io *io)
12571 {
12572         int retval;
12573         int i;
12574
12575         /*
12576          * This will send an error to the other controller in the case of a
12577          * failure.
12578          */
12579         retval = ctl_datamove_remote_sgl_setup(io);
12580         if (retval != 0)
12581                 return;
12582
12583         retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
12584                                           ctl_datamove_remote_read_cb);
12585         if (retval != 0) {
12586                 /*
12587                  * Make sure we free memory if there was an error..  The
12588                  * ctl_datamove_remote_xfer() function will send the
12589                  * datamove done message, or call the callback with an
12590                  * error if there is a problem.
12591                  */
12592                 for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12593                         free(io->io_hdr.local_sglist[i].addr, M_CTL);
12594                 free(io->io_hdr.remote_sglist, M_CTL);
12595                 io->io_hdr.remote_sglist = NULL;
12596                 io->io_hdr.local_sglist = NULL;
12597         }
12598 }
12599
12600 /*
12601  * Process a datamove request from the other controller.  This is used for
12602  * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
12603  * first.  Once that is complete, the data gets DMAed into the remote
12604  * controller's memory.  For reads, we DMA from the remote controller's
12605  * memory into our memory first, and then move it out to the FETD.
12606  */
12607 static void
12608 ctl_datamove_remote(union ctl_io *io)
12609 {
12610
12611         mtx_assert(&control_softc->ctl_lock, MA_NOTOWNED);
12612
12613         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12614                 ctl_failover_io(io, /*have_lock*/ 0);
12615                 return;
12616         }
12617
12618         /*
12619          * Note that we look for an aborted I/O here, but don't do some of
12620          * the other checks that ctl_datamove() normally does.
12621          * We don't need to run the datamove delay code, since that should
12622          * have been done if need be on the other controller.
12623          */
12624         if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12625                 printf("%s: tag 0x%04x on (%u:%u:%u) aborted\n", __func__,
12626                        io->scsiio.tag_num, io->io_hdr.nexus.initid,
12627                        io->io_hdr.nexus.targ_port,
12628                        io->io_hdr.nexus.targ_lun);
12629                 io->io_hdr.port_status = 31338;
12630                 ctl_send_datamove_done(io, /*have_lock*/ 0);
12631                 return;
12632         }
12633
12634         if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT)
12635                 ctl_datamove_remote_write(io);
12636         else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN)
12637                 ctl_datamove_remote_read(io);
12638         else {
12639                 io->io_hdr.port_status = 31339;
12640                 ctl_send_datamove_done(io, /*have_lock*/ 0);
12641         }
12642 }
12643
12644 static void
12645 ctl_process_done(union ctl_io *io)
12646 {
12647         struct ctl_lun *lun;
12648         struct ctl_softc *softc = control_softc;
12649         void (*fe_done)(union ctl_io *io);
12650         union ctl_ha_msg msg;
12651         uint32_t targ_port = io->io_hdr.nexus.targ_port;
12652
12653         CTL_DEBUG_PRINT(("ctl_process_done\n"));
12654         fe_done = softc->ctl_ports[targ_port]->fe_done;
12655
12656 #ifdef CTL_TIME_IO
12657         if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12658                 char str[256];
12659                 char path_str[64];
12660                 struct sbuf sb;
12661
12662                 ctl_scsi_path_string(io, path_str, sizeof(path_str));
12663                 sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12664
12665                 sbuf_cat(&sb, path_str);
12666                 switch (io->io_hdr.io_type) {
12667                 case CTL_IO_SCSI:
12668                         ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12669                         sbuf_printf(&sb, "\n");
12670                         sbuf_cat(&sb, path_str);
12671                         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12672                                     io->scsiio.tag_num, io->scsiio.tag_type);
12673                         break;
12674                 case CTL_IO_TASK:
12675                         sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12676                                     "Tag Type: %d\n", io->taskio.task_action,
12677                                     io->taskio.tag_num, io->taskio.tag_type);
12678                         break;
12679                 default:
12680                         printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12681                         panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12682                         break;
12683                 }
12684                 sbuf_cat(&sb, path_str);
12685                 sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
12686                             (intmax_t)time_uptime - io->io_hdr.start_time);
12687                 sbuf_finish(&sb);
12688                 printf("%s", sbuf_data(&sb));
12689         }
12690 #endif /* CTL_TIME_IO */
12691
12692         switch (io->io_hdr.io_type) {
12693         case CTL_IO_SCSI:
12694                 break;
12695         case CTL_IO_TASK:
12696                 if (ctl_debug & CTL_DEBUG_INFO)
12697                         ctl_io_error_print(io, NULL);
12698                 fe_done(io);
12699                 return;
12700         default:
12701                 panic("ctl_process_done: invalid io type %d\n",
12702                       io->io_hdr.io_type);
12703                 break; /* NOTREACHED */
12704         }
12705
12706         lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12707         if (lun == NULL) {
12708                 CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
12709                                  io->io_hdr.nexus.targ_mapped_lun));
12710                 goto bailout;
12711         }
12712
12713         mtx_lock(&lun->lun_lock);
12714
12715         /*
12716          * Check to see if we have any errors to inject here.  We only
12717          * inject errors for commands that don't already have errors set.
12718          */
12719         if ((STAILQ_FIRST(&lun->error_list) != NULL) &&
12720             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS) &&
12721             ((io->io_hdr.flags & CTL_FLAG_STATUS_SENT) == 0))
12722                 ctl_inject_error(lun, io);
12723
12724         /*
12725          * XXX KDM how do we treat commands that aren't completed
12726          * successfully?
12727          *
12728          * XXX KDM should we also track I/O latency?
12729          */
12730         if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS &&
12731             io->io_hdr.io_type == CTL_IO_SCSI) {
12732 #ifdef CTL_TIME_IO
12733                 struct bintime cur_bt;
12734 #endif
12735                 int type;
12736
12737                 if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
12738                     CTL_FLAG_DATA_IN)
12739                         type = CTL_STATS_READ;
12740                 else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
12741                     CTL_FLAG_DATA_OUT)
12742                         type = CTL_STATS_WRITE;
12743                 else
12744                         type = CTL_STATS_NO_IO;
12745
12746                 lun->stats.ports[targ_port].bytes[type] +=
12747                     io->scsiio.kern_total_len;
12748                 lun->stats.ports[targ_port].operations[type]++;
12749 #ifdef CTL_TIME_IO
12750                 bintime_add(&lun->stats.ports[targ_port].dma_time[type],
12751                    &io->io_hdr.dma_bt);
12752                 getbinuptime(&cur_bt);
12753                 bintime_sub(&cur_bt, &io->io_hdr.start_bt);
12754                 bintime_add(&lun->stats.ports[targ_port].time[type], &cur_bt);
12755 #endif
12756                 lun->stats.ports[targ_port].num_dmas[type] +=
12757                     io->io_hdr.num_dmas;
12758         }
12759
12760         /*
12761          * Remove this from the OOA queue.
12762          */
12763         TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
12764 #ifdef CTL_TIME_IO
12765         if (TAILQ_EMPTY(&lun->ooa_queue))
12766                 lun->last_busy = getsbinuptime();
12767 #endif
12768
12769         /*
12770          * Run through the blocked queue on this LUN and see if anything
12771          * has become unblocked, now that this transaction is done.
12772          */
12773         ctl_check_blocked(lun);
12774
12775         /*
12776          * If the LUN has been invalidated, free it if there is nothing
12777          * left on its OOA queue.
12778          */
12779         if ((lun->flags & CTL_LUN_INVALID)
12780          && TAILQ_EMPTY(&lun->ooa_queue)) {
12781                 mtx_unlock(&lun->lun_lock);
12782                 mtx_lock(&softc->ctl_lock);
12783                 ctl_free_lun(lun);
12784                 mtx_unlock(&softc->ctl_lock);
12785         } else
12786                 mtx_unlock(&lun->lun_lock);
12787
12788 bailout:
12789
12790         /*
12791          * If this command has been aborted, make sure we set the status
12792          * properly.  The FETD is responsible for freeing the I/O and doing
12793          * whatever it needs to do to clean up its state.
12794          */
12795         if (io->io_hdr.flags & CTL_FLAG_ABORT)
12796                 ctl_set_task_aborted(&io->scsiio);
12797
12798         /*
12799          * If enabled, print command error status.
12800          */
12801         if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS &&
12802             (ctl_debug & CTL_DEBUG_INFO) != 0)
12803                 ctl_io_error_print(io, NULL);
12804
12805         /*
12806          * Tell the FETD or the other shelf controller we're done with this
12807          * command.  Note that only SCSI commands get to this point.  Task
12808          * management commands are completed above.
12809          */
12810         if ((softc->ha_mode != CTL_HA_MODE_XFER) &&
12811             (io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)) {
12812                 memset(&msg, 0, sizeof(msg));
12813                 msg.hdr.msg_type = CTL_MSG_FINISH_IO;
12814                 msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
12815                 msg.hdr.nexus = io->io_hdr.nexus;
12816                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12817                     sizeof(msg.scsi) - sizeof(msg.scsi.sense_data),
12818                     M_WAITOK);
12819         }
12820
12821         fe_done(io);
12822 }
12823
12824 #ifdef CTL_WITH_CA
12825 /*
12826  * Front end should call this if it doesn't do autosense.  When the request
12827  * sense comes back in from the initiator, we'll dequeue this and send it.
12828  */
12829 int
12830 ctl_queue_sense(union ctl_io *io)
12831 {
12832         struct ctl_lun *lun;
12833         struct ctl_port *port;
12834         struct ctl_softc *softc;
12835         uint32_t initidx, targ_lun;
12836
12837         softc = control_softc;
12838
12839         CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
12840
12841         /*
12842          * LUN lookup will likely move to the ctl_work_thread() once we
12843          * have our new queueing infrastructure (that doesn't put things on
12844          * a per-LUN queue initially).  That is so that we can handle
12845          * things like an INQUIRY to a LUN that we don't have enabled.  We
12846          * can't deal with that right now.
12847          */
12848         mtx_lock(&softc->ctl_lock);
12849
12850         /*
12851          * If we don't have a LUN for this, just toss the sense
12852          * information.
12853          */
12854         port = ctl_io_port(&ctsio->io_hdr);
12855         targ_lun = ctl_lun_map_from_port(port, io->io_hdr.nexus.targ_lun);
12856         if ((targ_lun < CTL_MAX_LUNS)
12857          && (softc->ctl_luns[targ_lun] != NULL))
12858                 lun = softc->ctl_luns[targ_lun];
12859         else
12860                 goto bailout;
12861
12862         initidx = ctl_get_initindex(&io->io_hdr.nexus);
12863
12864         mtx_lock(&lun->lun_lock);
12865         /*
12866          * Already have CA set for this LUN...toss the sense information.
12867          */
12868         if (ctl_is_set(lun->have_ca, initidx)) {
12869                 mtx_unlock(&lun->lun_lock);
12870                 goto bailout;
12871         }
12872
12873         memcpy(&lun->pending_sense[initidx], &io->scsiio.sense_data,
12874                MIN(sizeof(lun->pending_sense[initidx]),
12875                sizeof(io->scsiio.sense_data)));
12876         ctl_set_mask(lun->have_ca, initidx);
12877         mtx_unlock(&lun->lun_lock);
12878
12879 bailout:
12880         mtx_unlock(&softc->ctl_lock);
12881
12882         ctl_free_io(io);
12883
12884         return (CTL_RETVAL_COMPLETE);
12885 }
12886 #endif
12887
12888 /*
12889  * Primary command inlet from frontend ports.  All SCSI and task I/O
12890  * requests must go through this function.
12891  */
12892 int
12893 ctl_queue(union ctl_io *io)
12894 {
12895         struct ctl_port *port;
12896
12897         CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
12898
12899 #ifdef CTL_TIME_IO
12900         io->io_hdr.start_time = time_uptime;
12901         getbinuptime(&io->io_hdr.start_bt);
12902 #endif /* CTL_TIME_IO */
12903
12904         /* Map FE-specific LUN ID into global one. */
12905         port = ctl_io_port(&io->io_hdr);
12906         io->io_hdr.nexus.targ_mapped_lun =
12907             ctl_lun_map_from_port(port, io->io_hdr.nexus.targ_lun);
12908
12909         switch (io->io_hdr.io_type) {
12910         case CTL_IO_SCSI:
12911         case CTL_IO_TASK:
12912                 if (ctl_debug & CTL_DEBUG_CDB)
12913                         ctl_io_print(io);
12914                 ctl_enqueue_incoming(io);
12915                 break;
12916         default:
12917                 printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
12918                 return (EINVAL);
12919         }
12920
12921         return (CTL_RETVAL_COMPLETE);
12922 }
12923
12924 #ifdef CTL_IO_DELAY
12925 static void
12926 ctl_done_timer_wakeup(void *arg)
12927 {
12928         union ctl_io *io;
12929
12930         io = (union ctl_io *)arg;
12931         ctl_done(io);
12932 }
12933 #endif /* CTL_IO_DELAY */
12934
12935 void
12936 ctl_serseq_done(union ctl_io *io)
12937 {
12938         struct ctl_lun *lun;
12939
12940         lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12941         if (lun->be_lun == NULL ||
12942             lun->be_lun->serseq == CTL_LUN_SERSEQ_OFF)
12943                 return;
12944         mtx_lock(&lun->lun_lock);
12945         io->io_hdr.flags |= CTL_FLAG_SERSEQ_DONE;
12946         ctl_check_blocked(lun);
12947         mtx_unlock(&lun->lun_lock);
12948 }
12949
12950 void
12951 ctl_done(union ctl_io *io)
12952 {
12953
12954         /*
12955          * Enable this to catch duplicate completion issues.
12956          */
12957 #if 0
12958         if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
12959                 printf("%s: type %d msg %d cdb %x iptl: "
12960                        "%u:%u:%u tag 0x%04x "
12961                        "flag %#x status %x\n",
12962                         __func__,
12963                         io->io_hdr.io_type,
12964                         io->io_hdr.msg_type,
12965                         io->scsiio.cdb[0],
12966                         io->io_hdr.nexus.initid,
12967                         io->io_hdr.nexus.targ_port,
12968                         io->io_hdr.nexus.targ_lun,
12969                         (io->io_hdr.io_type ==
12970                         CTL_IO_TASK) ?
12971                         io->taskio.tag_num :
12972                         io->scsiio.tag_num,
12973                         io->io_hdr.flags,
12974                         io->io_hdr.status);
12975         } else
12976                 io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
12977 #endif
12978
12979         /*
12980          * This is an internal copy of an I/O, and should not go through
12981          * the normal done processing logic.
12982          */
12983         if (io->io_hdr.flags & CTL_FLAG_INT_COPY)
12984                 return;
12985
12986 #ifdef CTL_IO_DELAY
12987         if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
12988                 struct ctl_lun *lun;
12989
12990                 lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12991
12992                 io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
12993         } else {
12994                 struct ctl_lun *lun;
12995
12996                 lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12997
12998                 if ((lun != NULL)
12999                  && (lun->delay_info.done_delay > 0)) {
13000
13001                         callout_init(&io->io_hdr.delay_callout, /*mpsafe*/ 1);
13002                         io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
13003                         callout_reset(&io->io_hdr.delay_callout,
13004                                       lun->delay_info.done_delay * hz,
13005                                       ctl_done_timer_wakeup, io);
13006                         if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
13007                                 lun->delay_info.done_delay = 0;
13008                         return;
13009                 }
13010         }
13011 #endif /* CTL_IO_DELAY */
13012
13013         ctl_enqueue_done(io);
13014 }
13015
13016 static void
13017 ctl_work_thread(void *arg)
13018 {
13019         struct ctl_thread *thr = (struct ctl_thread *)arg;
13020         struct ctl_softc *softc = thr->ctl_softc;
13021         union ctl_io *io;
13022         int retval;
13023
13024         CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
13025
13026         for (;;) {
13027                 /*
13028                  * We handle the queues in this order:
13029                  * - ISC
13030                  * - done queue (to free up resources, unblock other commands)
13031                  * - RtR queue
13032                  * - incoming queue
13033                  *
13034                  * If those queues are empty, we break out of the loop and
13035                  * go to sleep.
13036                  */
13037                 mtx_lock(&thr->queue_lock);
13038                 io = (union ctl_io *)STAILQ_FIRST(&thr->isc_queue);
13039                 if (io != NULL) {
13040                         STAILQ_REMOVE_HEAD(&thr->isc_queue, links);
13041                         mtx_unlock(&thr->queue_lock);
13042                         ctl_handle_isc(io);
13043                         continue;
13044                 }
13045                 io = (union ctl_io *)STAILQ_FIRST(&thr->done_queue);
13046                 if (io != NULL) {
13047                         STAILQ_REMOVE_HEAD(&thr->done_queue, links);
13048                         /* clear any blocked commands, call fe_done */
13049                         mtx_unlock(&thr->queue_lock);
13050                         ctl_process_done(io);
13051                         continue;
13052                 }
13053                 io = (union ctl_io *)STAILQ_FIRST(&thr->incoming_queue);
13054                 if (io != NULL) {
13055                         STAILQ_REMOVE_HEAD(&thr->incoming_queue, links);
13056                         mtx_unlock(&thr->queue_lock);
13057                         if (io->io_hdr.io_type == CTL_IO_TASK)
13058                                 ctl_run_task(io);
13059                         else
13060                                 ctl_scsiio_precheck(softc, &io->scsiio);
13061                         continue;
13062                 }
13063                 io = (union ctl_io *)STAILQ_FIRST(&thr->rtr_queue);
13064                 if (io != NULL) {
13065                         STAILQ_REMOVE_HEAD(&thr->rtr_queue, links);
13066                         mtx_unlock(&thr->queue_lock);
13067                         retval = ctl_scsiio(&io->scsiio);
13068                         if (retval != CTL_RETVAL_COMPLETE)
13069                                 CTL_DEBUG_PRINT(("ctl_scsiio failed\n"));
13070                         continue;
13071                 }
13072
13073                 /* Sleep until we have something to do. */
13074                 mtx_sleep(thr, &thr->queue_lock, PDROP | PRIBIO, "-", 0);
13075         }
13076 }
13077
13078 static void
13079 ctl_lun_thread(void *arg)
13080 {
13081         struct ctl_softc *softc = (struct ctl_softc *)arg;
13082         struct ctl_be_lun *be_lun;
13083
13084         CTL_DEBUG_PRINT(("ctl_lun_thread starting\n"));
13085
13086         for (;;) {
13087                 mtx_lock(&softc->ctl_lock);
13088                 be_lun = STAILQ_FIRST(&softc->pending_lun_queue);
13089                 if (be_lun != NULL) {
13090                         STAILQ_REMOVE_HEAD(&softc->pending_lun_queue, links);
13091                         mtx_unlock(&softc->ctl_lock);
13092                         ctl_create_lun(be_lun);
13093                         continue;
13094                 }
13095
13096                 /* Sleep until we have something to do. */
13097                 mtx_sleep(&softc->pending_lun_queue, &softc->ctl_lock,
13098                     PDROP | PRIBIO, "-", 0);
13099         }
13100 }
13101
13102 static void
13103 ctl_thresh_thread(void *arg)
13104 {
13105         struct ctl_softc *softc = (struct ctl_softc *)arg;
13106         struct ctl_lun *lun;
13107         struct scsi_da_rw_recovery_page *rwpage;
13108         struct ctl_logical_block_provisioning_page *page;
13109         const char *attr;
13110         union ctl_ha_msg msg;
13111         uint64_t thres, val;
13112         int i, e, set;
13113
13114         CTL_DEBUG_PRINT(("ctl_thresh_thread starting\n"));
13115
13116         for (;;) {
13117                 mtx_lock(&softc->ctl_lock);
13118                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
13119                         if ((lun->flags & CTL_LUN_DISABLED) ||
13120                             (lun->flags & CTL_LUN_OFFLINE) ||
13121                             lun->backend->lun_attr == NULL)
13122                                 continue;
13123                         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
13124                             softc->ha_mode == CTL_HA_MODE_XFER)
13125                                 continue;
13126                         rwpage = &lun->mode_pages.rw_er_page[CTL_PAGE_CURRENT];
13127                         if ((rwpage->byte8 & SMS_RWER_LBPERE) == 0)
13128                                 continue;
13129                         e = 0;
13130                         page = &lun->mode_pages.lbp_page[CTL_PAGE_CURRENT];
13131                         for (i = 0; i < CTL_NUM_LBP_THRESH; i++) {
13132                                 if ((page->descr[i].flags & SLBPPD_ENABLED) == 0)
13133                                         continue;
13134                                 thres = scsi_4btoul(page->descr[i].count);
13135                                 thres <<= CTL_LBP_EXPONENT;
13136                                 switch (page->descr[i].resource) {
13137                                 case 0x01:
13138                                         attr = "blocksavail";
13139                                         break;
13140                                 case 0x02:
13141                                         attr = "blocksused";
13142                                         break;
13143                                 case 0xf1:
13144                                         attr = "poolblocksavail";
13145                                         break;
13146                                 case 0xf2:
13147                                         attr = "poolblocksused";
13148                                         break;
13149                                 default:
13150                                         continue;
13151                                 }
13152                                 mtx_unlock(&softc->ctl_lock); // XXX
13153                                 val = lun->backend->lun_attr(
13154                                     lun->be_lun->be_lun, attr);
13155                                 mtx_lock(&softc->ctl_lock);
13156                                 if (val == UINT64_MAX)
13157                                         continue;
13158                                 if ((page->descr[i].flags & SLBPPD_ARMING_MASK)
13159                                     == SLBPPD_ARMING_INC)
13160                                         e = (val >= thres);
13161                                 else
13162                                         e = (val <= thres);
13163                                 if (e)
13164                                         break;
13165                         }
13166                         mtx_lock(&lun->lun_lock);
13167                         if (e) {
13168                                 scsi_u64to8b((uint8_t *)&page->descr[i] -
13169                                     (uint8_t *)page, lun->ua_tpt_info);
13170                                 if (lun->lasttpt == 0 ||
13171                                     time_uptime - lun->lasttpt >= CTL_LBP_UA_PERIOD) {
13172                                         lun->lasttpt = time_uptime;
13173                                         ctl_est_ua_all(lun, -1, CTL_UA_THIN_PROV_THRES);
13174                                         set = 1;
13175                                 } else
13176                                         set = 0;
13177                         } else {
13178                                 lun->lasttpt = 0;
13179                                 ctl_clr_ua_all(lun, -1, CTL_UA_THIN_PROV_THRES);
13180                                 set = -1;
13181                         }
13182                         mtx_unlock(&lun->lun_lock);
13183                         if (set != 0 &&
13184                             lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
13185                                 /* Send msg to other side. */
13186                                 bzero(&msg.ua, sizeof(msg.ua));
13187                                 msg.hdr.msg_type = CTL_MSG_UA;
13188                                 msg.hdr.nexus.initid = -1;
13189                                 msg.hdr.nexus.targ_port = -1;
13190                                 msg.hdr.nexus.targ_lun = lun->lun;
13191                                 msg.hdr.nexus.targ_mapped_lun = lun->lun;
13192                                 msg.ua.ua_all = 1;
13193                                 msg.ua.ua_set = (set > 0);
13194                                 msg.ua.ua_type = CTL_UA_THIN_PROV_THRES;
13195                                 memcpy(msg.ua.ua_info, lun->ua_tpt_info, 8);
13196                                 mtx_unlock(&softc->ctl_lock); // XXX
13197                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13198                                     sizeof(msg.ua), M_WAITOK);
13199                                 mtx_lock(&softc->ctl_lock);
13200                         }
13201                 }
13202                 mtx_unlock(&softc->ctl_lock);
13203                 pause("-", CTL_LBP_PERIOD * hz);
13204         }
13205 }
13206
13207 static void
13208 ctl_enqueue_incoming(union ctl_io *io)
13209 {
13210         struct ctl_softc *softc = control_softc;
13211         struct ctl_thread *thr;
13212         u_int idx;
13213
13214         idx = (io->io_hdr.nexus.targ_port * 127 +
13215                io->io_hdr.nexus.initid) % worker_threads;
13216         thr = &softc->threads[idx];
13217         mtx_lock(&thr->queue_lock);
13218         STAILQ_INSERT_TAIL(&thr->incoming_queue, &io->io_hdr, links);
13219         mtx_unlock(&thr->queue_lock);
13220         wakeup(thr);
13221 }
13222
13223 static void
13224 ctl_enqueue_rtr(union ctl_io *io)
13225 {
13226         struct ctl_softc *softc = control_softc;
13227         struct ctl_thread *thr;
13228
13229         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13230         mtx_lock(&thr->queue_lock);
13231         STAILQ_INSERT_TAIL(&thr->rtr_queue, &io->io_hdr, links);
13232         mtx_unlock(&thr->queue_lock);
13233         wakeup(thr);
13234 }
13235
13236 static void
13237 ctl_enqueue_done(union ctl_io *io)
13238 {
13239         struct ctl_softc *softc = control_softc;
13240         struct ctl_thread *thr;
13241
13242         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13243         mtx_lock(&thr->queue_lock);
13244         STAILQ_INSERT_TAIL(&thr->done_queue, &io->io_hdr, links);
13245         mtx_unlock(&thr->queue_lock);
13246         wakeup(thr);
13247 }
13248
13249 static void
13250 ctl_enqueue_isc(union ctl_io *io)
13251 {
13252         struct ctl_softc *softc = control_softc;
13253         struct ctl_thread *thr;
13254
13255         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13256         mtx_lock(&thr->queue_lock);
13257         STAILQ_INSERT_TAIL(&thr->isc_queue, &io->io_hdr, links);
13258         mtx_unlock(&thr->queue_lock);
13259         wakeup(thr);
13260 }
13261
13262 /*
13263  *  vim: ts=8
13264  */