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 const static struct scsi_info_exceptions_page ie_page_default = {
 265         /*page_code*/SMS_INFO_EXCEPTIONS_PAGE,
 266         /*page_length*/sizeof(struct scsi_info_exceptions_page) - 2,
 267         /*info_flags*/SIEP_FLAGS_DEXCPT,
 268         /*mrie*/0,
 269         /*interval_timer*/{0, 0, 0, 0},
 270         /*report_count*/{0, 0, 0, 0}
 271 };
 272
 273 const static struct scsi_info_exceptions_page ie_page_changeable = {
 274         /*page_code*/SMS_INFO_EXCEPTIONS_PAGE,
 275         /*page_length*/sizeof(struct scsi_info_exceptions_page) - 2,
 276         /*info_flags*/0,
 277         /*mrie*/0,
 278         /*interval_timer*/{0, 0, 0, 0},
 279         /*report_count*/{0, 0, 0, 0}
 280 };
 281
 282 #define CTL_LBPM_LEN    (sizeof(struct ctl_logical_block_provisioning_page) - 4)
 283
 284 const static struct ctl_logical_block_provisioning_page lbp_page_default = {{
 285         /*page_code*/SMS_INFO_EXCEPTIONS_PAGE | SMPH_SPF,
 286         /*subpage_code*/0x02,
 287         /*page_length*/{CTL_LBPM_LEN >> 8, CTL_LBPM_LEN},
 288         /*flags*/0,
 289         /*reserved*/{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
 290         /*descr*/{}},
 291         {{/*flags*/0,
 292           /*resource*/0x01,
 293           /*reserved*/{0, 0},
 294           /*count*/{0, 0, 0, 0}},
 295          {/*flags*/0,
 296           /*resource*/0x02,
 297           /*reserved*/{0, 0},
 298           /*count*/{0, 0, 0, 0}},
 299          {/*flags*/0,
 300           /*resource*/0xf1,
 301           /*reserved*/{0, 0},
 302           /*count*/{0, 0, 0, 0}},
 303          {/*flags*/0,
 304           /*resource*/0xf2,
 305           /*reserved*/{0, 0},
 306           /*count*/{0, 0, 0, 0}}
 307         }
 308 };
 309
 310 const static struct ctl_logical_block_provisioning_page lbp_page_changeable = {{
 311         /*page_code*/SMS_INFO_EXCEPTIONS_PAGE | SMPH_SPF,
 312         /*subpage_code*/0x02,
 313         /*page_length*/{CTL_LBPM_LEN >> 8, CTL_LBPM_LEN},
 314         /*flags*/0,
 315         /*reserved*/{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
 316         /*descr*/{}},
 317         {{/*flags*/0,
 318           /*resource*/0,
 319           /*reserved*/{0, 0},
 320           /*count*/{0, 0, 0, 0}},
 321          {/*flags*/0,
 322           /*resource*/0,
 323           /*reserved*/{0, 0},
 324           /*count*/{0, 0, 0, 0}},
 325          {/*flags*/0,
 326           /*resource*/0,
 327           /*reserved*/{0, 0},
 328           /*count*/{0, 0, 0, 0}},
 329          {/*flags*/0,
 330           /*resource*/0,
 331           /*reserved*/{0, 0},
 332           /*count*/{0, 0, 0, 0}}
 333         }
 334 };
 335
 336 SYSCTL_NODE(_kern_cam, OID_AUTO, ctl, CTLFLAG_RD, 0, "CAM Target Layer");
 337 static int worker_threads = -1;
 338 TUNABLE_INT("kern.cam.ctl.worker_threads", &worker_threads);
 339 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, worker_threads, CTLFLAG_RDTUN,
 340     &worker_threads, 1, "Number of worker threads");
 341 static int ctl_debug = CTL_DEBUG_NONE;
 342 TUNABLE_INT("kern.cam.ctl.debug", &ctl_debug);
 343 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, debug, CTLFLAG_RWTUN,
 344     &ctl_debug, 0, "Enabled debug flags");
 345
 346 /*
 347  * Supported pages (0x00), Serial number (0x80), Device ID (0x83),
 348  * Extended INQUIRY Data (0x86), Mode Page Policy (0x87),
 349  * SCSI Ports (0x88), Third-party Copy (0x8F), Block limits (0xB0),
 350  * Block Device Characteristics (0xB1) and Logical Block Provisioning (0xB2)
 351  */
 352 #define SCSI_EVPD_NUM_SUPPORTED_PAGES   10
 353
 354 static void ctl_isc_event_handler(ctl_ha_channel chanel, ctl_ha_event event,
 355                                   int param);
 356 static void ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest);
 357 static void ctl_copy_sense_data_back(union ctl_io *src, union ctl_ha_msg *dest);
 358 static int ctl_init(void);
 359 void ctl_shutdown(void);
 360 static int ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td);
 361 static int ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td);
 362 static int ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio);
 363 static int ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
 364                               struct ctl_ooa *ooa_hdr,
 365                               struct ctl_ooa_entry *kern_entries);
 366 static int ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
 367                      struct thread *td);
 368 static int ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
 369                          struct ctl_be_lun *be_lun);
 370 static int ctl_free_lun(struct ctl_lun *lun);
 371 static void ctl_create_lun(struct ctl_be_lun *be_lun);
 372 static struct ctl_port * ctl_io_port(struct ctl_io_hdr *io_hdr);
 373
 374 static int ctl_do_mode_select(union ctl_io *io);
 375 static int ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun,
 376                            uint64_t res_key, uint64_t sa_res_key,
 377                            uint8_t type, uint32_t residx,
 378                            struct ctl_scsiio *ctsio,
 379                            struct scsi_per_res_out *cdb,
 380                            struct scsi_per_res_out_parms* param);
 381 static void ctl_pro_preempt_other(struct ctl_lun *lun,
 382                                   union ctl_ha_msg *msg);
 383 static void ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg);
 384 static int ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len);
 385 static int ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len);
 386 static int ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len);
 387 static int ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len);
 388 static int ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len);
 389 static int ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio,
 390                                          int alloc_len);
 391 static int ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio,
 392                                          int alloc_len);
 393 static int ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len);
 394 static int ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len);
 395 static int ctl_inquiry_evpd(struct ctl_scsiio *ctsio);
 396 static int ctl_inquiry_std(struct ctl_scsiio *ctsio);
 397 static int ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len);
 398 static ctl_action ctl_extent_check(union ctl_io *io1, union ctl_io *io2,
 399     bool seq);
 400 static ctl_action ctl_extent_check_seq(union ctl_io *io1, union ctl_io *io2);
 401 static ctl_action ctl_check_for_blockage(struct ctl_lun *lun,
 402     union ctl_io *pending_io, union ctl_io *ooa_io);
 403 static ctl_action ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
 404                                 union ctl_io *starting_io);
 405 static int ctl_check_blocked(struct ctl_lun *lun);
 406 static int ctl_scsiio_lun_check(struct ctl_lun *lun,
 407                                 const struct ctl_cmd_entry *entry,
 408                                 struct ctl_scsiio *ctsio);
 409 static void ctl_failover_lun(struct ctl_lun *lun);
 410 static int ctl_scsiio_precheck(struct ctl_softc *ctl_softc,
 411                                struct ctl_scsiio *ctsio);
 412 static int ctl_scsiio(struct ctl_scsiio *ctsio);
 413
 414 static int ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io);
 415 static int ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
 416                             ctl_ua_type ua_type);
 417 static int ctl_do_lun_reset(struct ctl_lun *lun, union ctl_io *io,
 418                          ctl_ua_type ua_type);
 419 static int ctl_lun_reset(struct ctl_softc *ctl_softc, union ctl_io *io);
 420 static int ctl_abort_task(union ctl_io *io);
 421 static int ctl_abort_task_set(union ctl_io *io);
 422 static int ctl_query_task(union ctl_io *io, int task_set);
 423 static int ctl_i_t_nexus_reset(union ctl_io *io);
 424 static int ctl_query_async_event(union ctl_io *io);
 425 static void ctl_run_task(union ctl_io *io);
 426 #ifdef CTL_IO_DELAY
 427 static void ctl_datamove_timer_wakeup(void *arg);
 428 static void ctl_done_timer_wakeup(void *arg);
 429 #endif /* CTL_IO_DELAY */
 430
 431 static void ctl_send_datamove_done(union ctl_io *io, int have_lock);
 432 static void ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq);
 433 static int ctl_datamove_remote_dm_write_cb(union ctl_io *io);
 434 static void ctl_datamove_remote_write(union ctl_io *io);
 435 static int ctl_datamove_remote_dm_read_cb(union ctl_io *io);
 436 static void ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq);
 437 static int ctl_datamove_remote_sgl_setup(union ctl_io *io);
 438 static int ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
 439                                     ctl_ha_dt_cb callback);
 440 static void ctl_datamove_remote_read(union ctl_io *io);
 441 static void ctl_datamove_remote(union ctl_io *io);
 442 static int ctl_process_done(union ctl_io *io);
 443 static void ctl_lun_thread(void *arg);
 444 static void ctl_thresh_thread(void *arg);
 445 static void ctl_work_thread(void *arg);
 446 static void ctl_enqueue_incoming(union ctl_io *io);
 447 static void ctl_enqueue_rtr(union ctl_io *io);
 448 static void ctl_enqueue_done(union ctl_io *io);
 449 static void ctl_enqueue_isc(union ctl_io *io);
 450 static const struct ctl_cmd_entry *
 451     ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa);
 452 static const struct ctl_cmd_entry *
 453     ctl_validate_command(struct ctl_scsiio *ctsio);
 454 static int ctl_cmd_applicable(uint8_t lun_type,
 455     const struct ctl_cmd_entry *entry);
 456
 457 static uint64_t ctl_get_prkey(struct ctl_lun *lun, uint32_t residx);
 458 static void ctl_clr_prkey(struct ctl_lun *lun, uint32_t residx);
 459 static void ctl_alloc_prkey(struct ctl_lun *lun, uint32_t residx);
 460 static void ctl_set_prkey(struct ctl_lun *lun, uint32_t residx, uint64_t key);
 461
 462 /*
 463  * Load the serialization table.  This isn't very pretty, but is probably
 464  * the easiest way to do it.
 465  */
 466 #include "ctl_ser_table.c"
 467
 468 /*
 469  * We only need to define open, close and ioctl routines for this driver.
 470  */
 471 static struct cdevsw ctl_cdevsw = {
 472         .d_version =    D_VERSION,
 473         .d_flags =      0,
 474         .d_open =       ctl_open,
 475         .d_close =      ctl_close,
 476         .d_ioctl =      ctl_ioctl,
 477         .d_name =       "ctl",
 478 };
 479
 480
 481 MALLOC_DEFINE(M_CTL, "ctlmem", "Memory used for CTL");
 482
 483 static int ctl_module_event_handler(module_t, int /*modeventtype_t*/, void *);
 484
 485 static moduledata_t ctl_moduledata = {
 486         "ctl",
 487         ctl_module_event_handler,
 488         NULL
 489 };
 490
 491 DECLARE_MODULE(ctl, ctl_moduledata, SI_SUB_CONFIGURE, SI_ORDER_THIRD);
 492 MODULE_VERSION(ctl, 1);
 493
 494 static struct ctl_frontend ha_frontend =
 495 {
 496         .name = "ha",
 497 };
 498
 499 static void
 500 ctl_isc_handler_finish_xfer(struct ctl_softc *ctl_softc,
 501                             union ctl_ha_msg *msg_info)
 502 {
 503         struct ctl_scsiio *ctsio;
 504
 505         if (msg_info->hdr.original_sc == NULL) {
 506                 printf("%s: original_sc == NULL!\n", __func__);
 507                 /* XXX KDM now what? */
 508                 return;
 509         }
 510
 511         ctsio = &msg_info->hdr.original_sc->scsiio;
 512         ctsio->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
 513         ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
 514         ctsio->io_hdr.status = msg_info->hdr.status;
 515         ctsio->scsi_status = msg_info->scsi.scsi_status;
 516         ctsio->sense_len = msg_info->scsi.sense_len;
 517         ctsio->sense_residual = msg_info->scsi.sense_residual;
 518         ctsio->residual = msg_info->scsi.residual;
 519         memcpy(&ctsio->sense_data, &msg_info->scsi.sense_data,
 520                msg_info->scsi.sense_len);
 521         ctl_enqueue_isc((union ctl_io *)ctsio);
 522 }
 523
 524 static void
 525 ctl_isc_handler_finish_ser_only(struct ctl_softc *ctl_softc,
 526                                 union ctl_ha_msg *msg_info)
 527 {
 528         struct ctl_scsiio *ctsio;
 529
 530         if (msg_info->hdr.serializing_sc == NULL) {
 531                 printf("%s: serializing_sc == NULL!\n", __func__);
 532                 /* XXX KDM now what? */
 533                 return;
 534         }
 535
 536         ctsio = &msg_info->hdr.serializing_sc->scsiio;
 537         ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
 538         ctl_enqueue_isc((union ctl_io *)ctsio);
 539 }
 540
 541 void
 542 ctl_isc_announce_lun(struct ctl_lun *lun)
 543 {
 544         struct ctl_softc *softc = lun->ctl_softc;
 545         union ctl_ha_msg *msg;
 546         struct ctl_ha_msg_lun_pr_key pr_key;
 547         int i, k;
 548
 549         if (softc->ha_link != CTL_HA_LINK_ONLINE)
 550                 return;
 551         mtx_lock(&lun->lun_lock);
 552         i = sizeof(msg->lun);
 553         if (lun->lun_devid)
 554                 i += lun->lun_devid->len;
 555         i += sizeof(pr_key) * lun->pr_key_count;
 556 alloc:
 557         mtx_unlock(&lun->lun_lock);
 558         msg = malloc(i, M_CTL, M_WAITOK);
 559         mtx_lock(&lun->lun_lock);
 560         k = sizeof(msg->lun);
 561         if (lun->lun_devid)
 562                 k += lun->lun_devid->len;
 563         k += sizeof(pr_key) * lun->pr_key_count;
 564         if (i < k) {
 565                 free(msg, M_CTL);
 566                 i = k;
 567                 goto alloc;
 568         }
 569         bzero(&msg->lun, sizeof(msg->lun));
 570         msg->hdr.msg_type = CTL_MSG_LUN_SYNC;
 571         msg->hdr.nexus.targ_lun = lun->lun;
 572         msg->hdr.nexus.targ_mapped_lun = lun->lun;
 573         msg->lun.flags = lun->flags;
 574         msg->lun.pr_generation = lun->PRGeneration;
 575         msg->lun.pr_res_idx = lun->pr_res_idx;
 576         msg->lun.pr_res_type = lun->res_type;
 577         msg->lun.pr_key_count = lun->pr_key_count;
 578         i = 0;
 579         if (lun->lun_devid) {
 580                 msg->lun.lun_devid_len = lun->lun_devid->len;
 581                 memcpy(&msg->lun.data[i], lun->lun_devid->data,
 582                     msg->lun.lun_devid_len);
 583                 i += msg->lun.lun_devid_len;
 584         }
 585         for (k = 0; k < CTL_MAX_INITIATORS; k++) {
 586                 if ((pr_key.pr_key = ctl_get_prkey(lun, k)) == 0)
 587                         continue;
 588                 pr_key.pr_iid = k;
 589                 memcpy(&msg->lun.data[i], &pr_key, sizeof(pr_key));
 590                 i += sizeof(pr_key);
 591         }
 592         mtx_unlock(&lun->lun_lock);
 593         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->port, sizeof(msg->port) + i,
 594             M_WAITOK);
 595         free(msg, M_CTL);
 596 }
 597
 598 void
 599 ctl_isc_announce_port(struct ctl_port *port)
 600 {
 601         struct ctl_softc *softc = control_softc;
 602         union ctl_ha_msg *msg;
 603         int i;
 604
 605         if (port->targ_port < softc->port_min ||
 606             port->targ_port >= softc->port_max ||
 607             softc->ha_link != CTL_HA_LINK_ONLINE)
 608                 return;
 609         i = sizeof(msg->port) + strlen(port->port_name) + 1;
 610         if (port->lun_map)
 611                 i += sizeof(uint32_t) * CTL_MAX_LUNS;
 612         if (port->port_devid)
 613                 i += port->port_devid->len;
 614         if (port->target_devid)
 615                 i += port->target_devid->len;
 616         if (port->init_devid)
 617                 i += port->init_devid->len;
 618         msg = malloc(i, M_CTL, M_WAITOK);
 619         bzero(&msg->port, sizeof(msg->port));
 620         msg->hdr.msg_type = CTL_MSG_PORT_SYNC;
 621         msg->hdr.nexus.targ_port = port->targ_port;
 622         msg->port.port_type = port->port_type;
 623         msg->port.physical_port = port->physical_port;
 624         msg->port.virtual_port = port->virtual_port;
 625         msg->port.status = port->status;
 626         i = 0;
 627         msg->port.name_len = sprintf(&msg->port.data[i],
 628             "%d:%s", softc->ha_id, port->port_name) + 1;
 629         i += msg->port.name_len;
 630         if (port->lun_map) {
 631                 msg->port.lun_map_len = sizeof(uint32_t) * CTL_MAX_LUNS;
 632                 memcpy(&msg->port.data[i], port->lun_map,
 633                     msg->port.lun_map_len);
 634                 i += msg->port.lun_map_len;
 635         }
 636         if (port->port_devid) {
 637                 msg->port.port_devid_len = port->port_devid->len;
 638                 memcpy(&msg->port.data[i], port->port_devid->data,
 639                     msg->port.port_devid_len);
 640                 i += msg->port.port_devid_len;
 641         }
 642         if (port->target_devid) {
 643                 msg->port.target_devid_len = port->target_devid->len;
 644                 memcpy(&msg->port.data[i], port->target_devid->data,
 645                     msg->port.target_devid_len);
 646                 i += msg->port.target_devid_len;
 647         }
 648         if (port->init_devid) {
 649                 msg->port.init_devid_len = port->init_devid->len;
 650                 memcpy(&msg->port.data[i], port->init_devid->data,
 651                     msg->port.init_devid_len);
 652                 i += msg->port.init_devid_len;
 653         }
 654         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->port, sizeof(msg->port) + i,
 655             M_WAITOK);
 656         free(msg, M_CTL);
 657 }
 658
 659 void
 660 ctl_isc_announce_iid(struct ctl_port *port, int iid)
 661 {
 662         struct ctl_softc *softc = control_softc;
 663         union ctl_ha_msg *msg;
 664         int i, l;
 665
 666         if (port->targ_port < softc->port_min ||
 667             port->targ_port >= softc->port_max ||
 668             softc->ha_link != CTL_HA_LINK_ONLINE)
 669                 return;
 670         mtx_lock(&softc->ctl_lock);
 671         i = sizeof(msg->iid);
 672         l = 0;
 673         if (port->wwpn_iid[iid].name)
 674                 l = strlen(port->wwpn_iid[iid].name) + 1;
 675         i += l;
 676         msg = malloc(i, M_CTL, M_NOWAIT);
 677         if (msg == NULL) {
 678                 mtx_unlock(&softc->ctl_lock);
 679                 return;
 680         }
 681         bzero(&msg->iid, sizeof(msg->iid));
 682         msg->hdr.msg_type = CTL_MSG_IID_SYNC;
 683         msg->hdr.nexus.targ_port = port->targ_port;
 684         msg->hdr.nexus.initid = iid;
 685         msg->iid.in_use = port->wwpn_iid[iid].in_use;
 686         msg->iid.name_len = l;
 687         msg->iid.wwpn = port->wwpn_iid[iid].wwpn;
 688         if (port->wwpn_iid[iid].name)
 689                 strlcpy(msg->iid.data, port->wwpn_iid[iid].name, l);
 690         mtx_unlock(&softc->ctl_lock);
 691         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->iid, i, M_NOWAIT);
 692         free(msg, M_CTL);
 693 }
 694
 695 static void
 696 ctl_isc_ha_link_up(struct ctl_softc *softc)
 697 {
 698         struct ctl_port *port;
 699         struct ctl_lun *lun;
 700         int i;
 701
 702         STAILQ_FOREACH(port, &softc->port_list, links) {
 703                 ctl_isc_announce_port(port);
 704                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
 705                         if (port->wwpn_iid[i].in_use)
 706                                 ctl_isc_announce_iid(port, i);
 707                 }
 708         }
 709         STAILQ_FOREACH(lun, &softc->lun_list, links)
 710                 ctl_isc_announce_lun(lun);
 711 }
 712
 713 static void
 714 ctl_isc_ha_link_down(struct ctl_softc *softc)
 715 {
 716         struct ctl_port *port;
 717         struct ctl_lun *lun;
 718         union ctl_io *io;
 719         int i;
 720
 721         mtx_lock(&softc->ctl_lock);
 722         STAILQ_FOREACH(lun, &softc->lun_list, links) {
 723                 mtx_lock(&lun->lun_lock);
 724                 if (lun->flags & CTL_LUN_PEER_SC_PRIMARY) {
 725                         lun->flags &= ~CTL_LUN_PEER_SC_PRIMARY;
 726                         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
 727                 }
 728                 mtx_unlock(&lun->lun_lock);
 729
 730                 mtx_unlock(&softc->ctl_lock);
 731                 io = ctl_alloc_io(softc->othersc_pool);
 732                 mtx_lock(&softc->ctl_lock);
 733                 ctl_zero_io(io);
 734                 io->io_hdr.msg_type = CTL_MSG_FAILOVER;
 735                 io->io_hdr.nexus.targ_mapped_lun = lun->lun;
 736                 ctl_enqueue_isc(io);
 737         }
 738
 739         STAILQ_FOREACH(port, &softc->port_list, links) {
 740                 if (port->targ_port >= softc->port_min &&
 741                     port->targ_port < softc->port_max)
 742                         continue;
 743                 port->status &= ~CTL_PORT_STATUS_ONLINE;
 744                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
 745                         port->wwpn_iid[i].in_use = 0;
 746                         free(port->wwpn_iid[i].name, M_CTL);
 747                         port->wwpn_iid[i].name = NULL;
 748                 }
 749         }
 750         mtx_unlock(&softc->ctl_lock);
 751 }
 752
 753 static void
 754 ctl_isc_ua(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
 755 {
 756         struct ctl_lun *lun;
 757         uint32_t iid = ctl_get_initindex(&msg->hdr.nexus);
 758
 759         mtx_lock(&softc->ctl_lock);
 760         if (msg->hdr.nexus.targ_lun < CTL_MAX_LUNS &&
 761             (lun = softc->ctl_luns[msg->hdr.nexus.targ_mapped_lun]) != NULL) {
 762                 mtx_lock(&lun->lun_lock);
 763                 mtx_unlock(&softc->ctl_lock);
 764                 if (msg->ua.ua_type == CTL_UA_THIN_PROV_THRES &&
 765                     msg->ua.ua_set)
 766                         memcpy(lun->ua_tpt_info, msg->ua.ua_info, 8);
 767                 if (msg->ua.ua_all) {
 768                         if (msg->ua.ua_set)
 769                                 ctl_est_ua_all(lun, iid, msg->ua.ua_type);
 770                         else
 771                                 ctl_clr_ua_all(lun, iid, msg->ua.ua_type);
 772                 } else {
 773                         if (msg->ua.ua_set)
 774                                 ctl_est_ua(lun, iid, msg->ua.ua_type);
 775                         else
 776                                 ctl_clr_ua(lun, iid, msg->ua.ua_type);
 777                 }
 778                 mtx_unlock(&lun->lun_lock);
 779         } else
 780                 mtx_unlock(&softc->ctl_lock);
 781 }
 782
 783 static void
 784 ctl_isc_lun_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
 785 {
 786         struct ctl_lun *lun;
 787         struct ctl_ha_msg_lun_pr_key pr_key;
 788         int i, k;
 789         ctl_lun_flags oflags;
 790         uint32_t targ_lun;
 791
 792         targ_lun = msg->hdr.nexus.targ_mapped_lun;
 793         mtx_lock(&softc->ctl_lock);
 794         if ((targ_lun >= CTL_MAX_LUNS) ||
 795             ((lun = softc->ctl_luns[targ_lun]) == NULL)) {
 796                 mtx_unlock(&softc->ctl_lock);
 797                 return;
 798         }
 799         mtx_lock(&lun->lun_lock);
 800         mtx_unlock(&softc->ctl_lock);
 801         if (lun->flags & CTL_LUN_DISABLED) {
 802                 mtx_unlock(&lun->lun_lock);
 803                 return;
 804         }
 805         i = (lun->lun_devid != NULL) ? lun->lun_devid->len : 0;
 806         if (msg->lun.lun_devid_len != i || (i > 0 &&
 807             memcmp(&msg->lun.data[0], lun->lun_devid->data, i) != 0)) {
 808                 mtx_unlock(&lun->lun_lock);
 809                 printf("%s: Received conflicting HA LUN %d\n",
 810                     __func__, msg->hdr.nexus.targ_lun);
 811                 return;
 812         } else {
 813                 /* Record whether peer is primary. */
 814                 oflags = lun->flags;
 815                 if ((msg->lun.flags & CTL_LUN_PRIMARY_SC) &&
 816                     (msg->lun.flags & CTL_LUN_DISABLED) == 0)
 817                         lun->flags |= CTL_LUN_PEER_SC_PRIMARY;
 818                 else
 819                         lun->flags &= ~CTL_LUN_PEER_SC_PRIMARY;
 820                 if (oflags != lun->flags)
 821                         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
 822
 823                 /* If peer is primary and we are not -- use data */
 824                 if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
 825                     (lun->flags & CTL_LUN_PEER_SC_PRIMARY)) {
 826                         lun->PRGeneration = msg->lun.pr_generation;
 827                         lun->pr_res_idx = msg->lun.pr_res_idx;
 828                         lun->res_type = msg->lun.pr_res_type;
 829                         lun->pr_key_count = msg->lun.pr_key_count;
 830                         for (k = 0; k < CTL_MAX_INITIATORS; k++)
 831                                 ctl_clr_prkey(lun, k);
 832                         for (k = 0; k < msg->lun.pr_key_count; k++) {
 833                                 memcpy(&pr_key, &msg->lun.data[i],
 834                                     sizeof(pr_key));
 835                                 ctl_alloc_prkey(lun, pr_key.pr_iid);
 836                                 ctl_set_prkey(lun, pr_key.pr_iid,
 837                                     pr_key.pr_key);
 838                                 i += sizeof(pr_key);
 839                         }
 840                 }
 841
 842                 mtx_unlock(&lun->lun_lock);
 843                 CTL_DEBUG_PRINT(("%s: Known LUN %d, peer is %s\n",
 844                     __func__, msg->hdr.nexus.targ_lun,
 845                     (msg->lun.flags & CTL_LUN_PRIMARY_SC) ?
 846                     "primary" : "secondary"));
 847
 848                 /* If we are primary but peer doesn't know -- notify */
 849                 if ((lun->flags & CTL_LUN_PRIMARY_SC) &&
 850                     (msg->lun.flags & CTL_LUN_PEER_SC_PRIMARY) == 0)
 851                         ctl_isc_announce_lun(lun);
 852         }
 853 }
 854
 855 static void
 856 ctl_isc_port_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
 857 {
 858         struct ctl_port *port;
 859         struct ctl_lun *lun;
 860         int i, new;
 861
 862         port = softc->ctl_ports[msg->hdr.nexus.targ_port];
 863         if (port == NULL) {
 864                 CTL_DEBUG_PRINT(("%s: New port %d\n", __func__,
 865                     msg->hdr.nexus.targ_port));
 866                 new = 1;
 867                 port = malloc(sizeof(*port), M_CTL, M_WAITOK | M_ZERO);
 868                 port->frontend = &ha_frontend;
 869                 port->targ_port = msg->hdr.nexus.targ_port;
 870         } else if (port->frontend == &ha_frontend) {
 871                 CTL_DEBUG_PRINT(("%s: Updated port %d\n", __func__,
 872                     msg->hdr.nexus.targ_port));
 873                 new = 0;
 874         } else {
 875                 printf("%s: Received conflicting HA port %d\n",
 876                     __func__, msg->hdr.nexus.targ_port);
 877                 return;
 878         }
 879         port->port_type = msg->port.port_type;
 880         port->physical_port = msg->port.physical_port;
 881         port->virtual_port = msg->port.virtual_port;
 882         port->status = msg->port.status;
 883         i = 0;
 884         free(port->port_name, M_CTL);
 885         port->port_name = strndup(&msg->port.data[i], msg->port.name_len,
 886             M_CTL);
 887         i += msg->port.name_len;
 888         if (msg->port.lun_map_len != 0) {
 889                 if (port->lun_map == NULL)
 890                         port->lun_map = malloc(sizeof(uint32_t) * CTL_MAX_LUNS,
 891                             M_CTL, M_WAITOK);
 892                 memcpy(port->lun_map, &msg->port.data[i],
 893                     sizeof(uint32_t) * CTL_MAX_LUNS);
 894                 i += msg->port.lun_map_len;
 895         } else {
 896                 free(port->lun_map, M_CTL);
 897                 port->lun_map = NULL;
 898         }
 899         if (msg->port.port_devid_len != 0) {
 900                 if (port->port_devid == NULL ||
 901                     port->port_devid->len != msg->port.port_devid_len) {
 902                         free(port->port_devid, M_CTL);
 903                         port->port_devid = malloc(sizeof(struct ctl_devid) +
 904                             msg->port.port_devid_len, M_CTL, M_WAITOK);
 905                 }
 906                 memcpy(port->port_devid->data, &msg->port.data[i],
 907                     msg->port.port_devid_len);
 908                 port->port_devid->len = msg->port.port_devid_len;
 909                 i += msg->port.port_devid_len;
 910         } else {
 911                 free(port->port_devid, M_CTL);
 912                 port->port_devid = NULL;
 913         }
 914         if (msg->port.target_devid_len != 0) {
 915                 if (port->target_devid == NULL ||
 916                     port->target_devid->len != msg->port.target_devid_len) {
 917                         free(port->target_devid, M_CTL);
 918                         port->target_devid = malloc(sizeof(struct ctl_devid) +
 919                             msg->port.target_devid_len, M_CTL, M_WAITOK);
 920                 }
 921                 memcpy(port->target_devid->data, &msg->port.data[i],
 922                     msg->port.target_devid_len);
 923                 port->target_devid->len = msg->port.target_devid_len;
 924                 i += msg->port.target_devid_len;
 925         } else {
 926                 free(port->target_devid, M_CTL);
 927                 port->target_devid = NULL;
 928         }
 929         if (msg->port.init_devid_len != 0) {
 930                 if (port->init_devid == NULL ||
 931                     port->init_devid->len != msg->port.init_devid_len) {
 932                         free(port->init_devid, M_CTL);
 933                         port->init_devid = malloc(sizeof(struct ctl_devid) +
 934                             msg->port.init_devid_len, M_CTL, M_WAITOK);
 935                 }
 936                 memcpy(port->init_devid->data, &msg->port.data[i],
 937                     msg->port.init_devid_len);
 938                 port->init_devid->len = msg->port.init_devid_len;
 939                 i += msg->port.init_devid_len;
 940         } else {
 941                 free(port->init_devid, M_CTL);
 942                 port->init_devid = NULL;
 943         }
 944         if (new) {
 945                 if (ctl_port_register(port) != 0) {
 946                         printf("%s: ctl_port_register() failed with error\n",
 947                             __func__);
 948                 }
 949         }
 950         mtx_lock(&softc->ctl_lock);
 951         STAILQ_FOREACH(lun, &softc->lun_list, links) {
 952                 if (ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
 953                         continue;
 954                 mtx_lock(&lun->lun_lock);
 955                 ctl_est_ua_all(lun, -1, CTL_UA_INQ_CHANGE);
 956                 mtx_unlock(&lun->lun_lock);
 957         }
 958         mtx_unlock(&softc->ctl_lock);
 959 }
 960
 961 static void
 962 ctl_isc_iid_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
 963 {
 964         struct ctl_port *port;
 965         int iid;
 966
 967         port = softc->ctl_ports[msg->hdr.nexus.targ_port];
 968         if (port == NULL) {
 969                 printf("%s: Received IID for unknown port %d\n",
 970                     __func__, msg->hdr.nexus.targ_port);
 971                 return;
 972         }
 973         iid = msg->hdr.nexus.initid;
 974         port->wwpn_iid[iid].in_use = msg->iid.in_use;
 975         port->wwpn_iid[iid].wwpn = msg->iid.wwpn;
 976         free(port->wwpn_iid[iid].name, M_CTL);
 977         if (msg->iid.name_len) {
 978                 port->wwpn_iid[iid].name = strndup(&msg->iid.data[0],
 979                     msg->iid.name_len, M_CTL);
 980         } else
 981                 port->wwpn_iid[iid].name = NULL;
 982 }
 983
 984 /*
 985  * ISC (Inter Shelf Communication) event handler.  Events from the HA
 986  * subsystem come in here.
 987  */
 988 static void
 989 ctl_isc_event_handler(ctl_ha_channel channel, ctl_ha_event event, int param)
 990 {
 991         struct ctl_softc *softc;
 992         union ctl_io *io;
 993         struct ctl_prio *presio;
 994         ctl_ha_status isc_status;
 995
 996         softc = control_softc;
 997         CTL_DEBUG_PRINT(("CTL: Isc Msg event %d\n", event));
 998         if (event == CTL_HA_EVT_MSG_RECV) {
 999                 union ctl_ha_msg *msg, msgbuf;
1000
1001                 if (param > sizeof(msgbuf))
1002                         msg = malloc(param, M_CTL, M_WAITOK);
1003                 else
1004                         msg = &msgbuf;
1005                 isc_status = ctl_ha_msg_recv(CTL_HA_CHAN_CTL, msg, param,
1006                     M_WAITOK);
1007                 if (isc_status != CTL_HA_STATUS_SUCCESS) {
1008                         printf("%s: Error receiving message: %d\n",
1009                             __func__, isc_status);
1010                         if (msg != &msgbuf)
1011                                 free(msg, M_CTL);
1012                         return;
1013                 }
1014
1015                 CTL_DEBUG_PRINT(("CTL: msg_type %d\n", msg->msg_type));
1016                 switch (msg->hdr.msg_type) {
1017                 case CTL_MSG_SERIALIZE:
1018                         io = ctl_alloc_io(softc->othersc_pool);
1019                         ctl_zero_io(io);
1020                         // populate ctsio from msg
1021                         io->io_hdr.io_type = CTL_IO_SCSI;
1022                         io->io_hdr.msg_type = CTL_MSG_SERIALIZE;
1023                         io->io_hdr.original_sc = msg->hdr.original_sc;
1024                         io->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC |
1025                                             CTL_FLAG_IO_ACTIVE;
1026                         /*
1027                          * If we're in serialization-only mode, we don't
1028                          * want to go through full done processing.  Thus
1029                          * the COPY flag.
1030                          *
1031                          * XXX KDM add another flag that is more specific.
1032                          */
1033                         if (softc->ha_mode != CTL_HA_MODE_XFER)
1034                                 io->io_hdr.flags |= CTL_FLAG_INT_COPY;
1035                         io->io_hdr.nexus = msg->hdr.nexus;
1036 #if 0
1037                         printf("port %u, iid %u, lun %u\n",
1038                                io->io_hdr.nexus.targ_port,
1039                                io->io_hdr.nexus.initid,
1040                                io->io_hdr.nexus.targ_lun);
1041 #endif
1042                         io->scsiio.tag_num = msg->scsi.tag_num;
1043                         io->scsiio.tag_type = msg->scsi.tag_type;
1044 #ifdef CTL_TIME_IO
1045                         io->io_hdr.start_time = time_uptime;
1046                         getbintime(&io->io_hdr.start_bt);
1047 #endif /* CTL_TIME_IO */
1048                         io->scsiio.cdb_len = msg->scsi.cdb_len;
1049                         memcpy(io->scsiio.cdb, msg->scsi.cdb,
1050                                CTL_MAX_CDBLEN);
1051                         if (softc->ha_mode == CTL_HA_MODE_XFER) {
1052                                 const struct ctl_cmd_entry *entry;
1053
1054                                 entry = ctl_get_cmd_entry(&io->scsiio, NULL);
1055                                 io->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
1056                                 io->io_hdr.flags |=
1057                                         entry->flags & CTL_FLAG_DATA_MASK;
1058                         }
1059                         ctl_enqueue_isc(io);
1060                         break;
1061
1062                 /* Performed on the Originating SC, XFER mode only */
1063                 case CTL_MSG_DATAMOVE: {
1064                         struct ctl_sg_entry *sgl;
1065                         int i, j;
1066
1067                         io = msg->hdr.original_sc;
1068                         if (io == NULL) {
1069                                 printf("%s: original_sc == NULL!\n", __func__);
1070                                 /* XXX KDM do something here */
1071                                 break;
1072                         }
1073                         io->io_hdr.msg_type = CTL_MSG_DATAMOVE;
1074                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1075                         /*
1076                          * Keep track of this, we need to send it back over
1077                          * when the datamove is complete.
1078                          */
1079                         io->io_hdr.serializing_sc = msg->hdr.serializing_sc;
1080                         if (msg->hdr.status == CTL_SUCCESS)
1081                                 io->io_hdr.status = msg->hdr.status;
1082
1083                         if (msg->dt.sg_sequence == 0) {
1084                                 i = msg->dt.kern_sg_entries +
1085                                     msg->dt.kern_data_len /
1086                                     CTL_HA_DATAMOVE_SEGMENT + 1;
1087                                 sgl = malloc(sizeof(*sgl) * i, M_CTL,
1088                                     M_WAITOK | M_ZERO);
1089                                 io->io_hdr.remote_sglist = sgl;
1090                                 io->io_hdr.local_sglist =
1091                                     &sgl[msg->dt.kern_sg_entries];
1092
1093                                 io->scsiio.kern_data_ptr = (uint8_t *)sgl;
1094
1095                                 io->scsiio.kern_sg_entries =
1096                                         msg->dt.kern_sg_entries;
1097                                 io->scsiio.rem_sg_entries =
1098                                         msg->dt.kern_sg_entries;
1099                                 io->scsiio.kern_data_len =
1100                                         msg->dt.kern_data_len;
1101                                 io->scsiio.kern_total_len =
1102                                         msg->dt.kern_total_len;
1103                                 io->scsiio.kern_data_resid =
1104                                         msg->dt.kern_data_resid;
1105                                 io->scsiio.kern_rel_offset =
1106                                         msg->dt.kern_rel_offset;
1107                                 io->io_hdr.flags &= ~CTL_FLAG_BUS_ADDR;
1108                                 io->io_hdr.flags |= msg->dt.flags &
1109                                     CTL_FLAG_BUS_ADDR;
1110                         } else
1111                                 sgl = (struct ctl_sg_entry *)
1112                                         io->scsiio.kern_data_ptr;
1113
1114                         for (i = msg->dt.sent_sg_entries, j = 0;
1115                              i < (msg->dt.sent_sg_entries +
1116                              msg->dt.cur_sg_entries); i++, j++) {
1117                                 sgl[i].addr = msg->dt.sg_list[j].addr;
1118                                 sgl[i].len = msg->dt.sg_list[j].len;
1119
1120 #if 0
1121                                 printf("%s: DATAMOVE: %p,%lu j=%d, i=%d\n",
1122                                     __func__, sgl[i].addr, sgl[i].len, j, i);
1123 #endif
1124                         }
1125
1126                         /*
1127                          * If this is the last piece of the I/O, we've got
1128                          * the full S/G list.  Queue processing in the thread.
1129                          * Otherwise wait for the next piece.
1130                          */
1131                         if (msg->dt.sg_last != 0)
1132                                 ctl_enqueue_isc(io);
1133                         break;
1134                 }
1135                 /* Performed on the Serializing (primary) SC, XFER mode only */
1136                 case CTL_MSG_DATAMOVE_DONE: {
1137                         if (msg->hdr.serializing_sc == NULL) {
1138                                 printf("%s: serializing_sc == NULL!\n",
1139                                        __func__);
1140                                 /* XXX KDM now what? */
1141                                 break;
1142                         }
1143                         /*
1144                          * We grab the sense information here in case
1145                          * there was a failure, so we can return status
1146                          * back to the initiator.
1147                          */
1148                         io = msg->hdr.serializing_sc;
1149                         io->io_hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
1150                         io->io_hdr.flags &= ~CTL_FLAG_DMA_INPROG;
1151                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1152                         io->io_hdr.port_status = msg->scsi.fetd_status;
1153                         io->scsiio.residual = msg->scsi.residual;
1154                         if (msg->hdr.status != CTL_STATUS_NONE) {
1155                                 io->io_hdr.status = msg->hdr.status;
1156                                 io->scsiio.scsi_status = msg->scsi.scsi_status;
1157                                 io->scsiio.sense_len = msg->scsi.sense_len;
1158                                 io->scsiio.sense_residual =msg->scsi.sense_residual;
1159                                 memcpy(&io->scsiio.sense_data,
1160                                     &msg->scsi.sense_data,
1161                                     msg->scsi.sense_len);
1162                                 if (msg->hdr.status == CTL_SUCCESS)
1163                                         io->io_hdr.flags |= CTL_FLAG_STATUS_SENT;
1164                         }
1165                         ctl_enqueue_isc(io);
1166                         break;
1167                 }
1168
1169                 /* Preformed on Originating SC, SER_ONLY mode */
1170                 case CTL_MSG_R2R:
1171                         io = msg->hdr.original_sc;
1172                         if (io == NULL) {
1173                                 printf("%s: original_sc == NULL!\n",
1174                                     __func__);
1175                                 break;
1176                         }
1177                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1178                         io->io_hdr.msg_type = CTL_MSG_R2R;
1179                         io->io_hdr.serializing_sc = msg->hdr.serializing_sc;
1180                         ctl_enqueue_isc(io);
1181                         break;
1182
1183                 /*
1184                  * Performed on Serializing(i.e. primary SC) SC in SER_ONLY
1185                  * mode.
1186                  * Performed on the Originating (i.e. secondary) SC in XFER
1187                  * mode
1188                  */
1189                 case CTL_MSG_FINISH_IO:
1190                         if (softc->ha_mode == CTL_HA_MODE_XFER)
1191                                 ctl_isc_handler_finish_xfer(softc, msg);
1192                         else
1193                                 ctl_isc_handler_finish_ser_only(softc, msg);
1194                         break;
1195
1196                 /* Preformed on Originating SC */
1197                 case CTL_MSG_BAD_JUJU:
1198                         io = msg->hdr.original_sc;
1199                         if (io == NULL) {
1200                                 printf("%s: Bad JUJU!, original_sc is NULL!\n",
1201                                        __func__);
1202                                 break;
1203                         }
1204                         ctl_copy_sense_data(msg, io);
1205                         /*
1206                          * IO should have already been cleaned up on other
1207                          * SC so clear this flag so we won't send a message
1208                          * back to finish the IO there.
1209                          */
1210                         io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
1211                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1212
1213                         /* io = msg->hdr.serializing_sc; */
1214                         io->io_hdr.msg_type = CTL_MSG_BAD_JUJU;
1215                         ctl_enqueue_isc(io);
1216                         break;
1217
1218                 /* Handle resets sent from the other side */
1219                 case CTL_MSG_MANAGE_TASKS: {
1220                         struct ctl_taskio *taskio;
1221                         taskio = (struct ctl_taskio *)ctl_alloc_io(
1222                             softc->othersc_pool);
1223                         ctl_zero_io((union ctl_io *)taskio);
1224                         taskio->io_hdr.io_type = CTL_IO_TASK;
1225                         taskio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
1226                         taskio->io_hdr.nexus = msg->hdr.nexus;
1227                         taskio->task_action = msg->task.task_action;
1228                         taskio->tag_num = msg->task.tag_num;
1229                         taskio->tag_type = msg->task.tag_type;
1230 #ifdef CTL_TIME_IO
1231                         taskio->io_hdr.start_time = time_uptime;
1232                         getbintime(&taskio->io_hdr.start_bt);
1233 #endif /* CTL_TIME_IO */
1234                         ctl_run_task((union ctl_io *)taskio);
1235                         break;
1236                 }
1237                 /* Persistent Reserve action which needs attention */
1238                 case CTL_MSG_PERS_ACTION:
1239                         presio = (struct ctl_prio *)ctl_alloc_io(
1240                             softc->othersc_pool);
1241                         ctl_zero_io((union ctl_io *)presio);
1242                         presio->io_hdr.msg_type = CTL_MSG_PERS_ACTION;
1243                         presio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
1244                         presio->io_hdr.nexus = msg->hdr.nexus;
1245                         presio->pr_msg = msg->pr;
1246                         ctl_enqueue_isc((union ctl_io *)presio);
1247                         break;
1248                 case CTL_MSG_UA:
1249                         ctl_isc_ua(softc, msg, param);
1250                         break;
1251                 case CTL_MSG_PORT_SYNC:
1252                         ctl_isc_port_sync(softc, msg, param);
1253                         break;
1254                 case CTL_MSG_LUN_SYNC:
1255                         ctl_isc_lun_sync(softc, msg, param);
1256                         break;
1257                 case CTL_MSG_IID_SYNC:
1258                         ctl_isc_iid_sync(softc, msg, param);
1259                         break;
1260                 default:
1261                         printf("Received HA message of unknown type %d\n",
1262                             msg->hdr.msg_type);
1263                         break;
1264                 }
1265                 if (msg != &msgbuf)
1266                         free(msg, M_CTL);
1267         } else if (event == CTL_HA_EVT_LINK_CHANGE) {
1268                 printf("CTL: HA link status changed from %d to %d\n",
1269                     softc->ha_link, param);
1270                 if (param == softc->ha_link)
1271                         return;
1272                 if (softc->ha_link == CTL_HA_LINK_ONLINE) {
1273                         softc->ha_link = param;
1274                         ctl_isc_ha_link_down(softc);
1275                 } else {
1276                         softc->ha_link = param;
1277                         if (softc->ha_link == CTL_HA_LINK_ONLINE)
1278                                 ctl_isc_ha_link_up(softc);
1279                 }
1280                 return;
1281         } else {
1282                 printf("ctl_isc_event_handler: Unknown event %d\n", event);
1283                 return;
1284         }
1285 }
1286
1287 static void
1288 ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest)
1289 {
1290
1291         memcpy(&dest->scsiio.sense_data, &src->scsi.sense_data,
1292             src->scsi.sense_len);
1293         dest->scsiio.scsi_status = src->scsi.scsi_status;
1294         dest->scsiio.sense_len = src->scsi.sense_len;
1295         dest->io_hdr.status = src->hdr.status;
1296 }
1297
1298 static void
1299 ctl_copy_sense_data_back(union ctl_io *src, union ctl_ha_msg *dest)
1300 {
1301
1302         memcpy(&dest->scsi.sense_data, &src->scsiio.sense_data,
1303             src->scsiio.sense_len);
1304         dest->scsi.scsi_status = src->scsiio.scsi_status;
1305         dest->scsi.sense_len = src->scsiio.sense_len;
1306         dest->hdr.status = src->io_hdr.status;
1307 }
1308
1309 void
1310 ctl_est_ua(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua)
1311 {
1312         struct ctl_softc *softc = lun->ctl_softc;
1313         ctl_ua_type *pu;
1314
1315         if (initidx < softc->init_min || initidx >= softc->init_max)
1316                 return;
1317         mtx_assert(&lun->lun_lock, MA_OWNED);
1318         pu = lun->pending_ua[initidx / CTL_MAX_INIT_PER_PORT];
1319         if (pu == NULL)
1320                 return;
1321         pu[initidx % CTL_MAX_INIT_PER_PORT] |= ua;
1322 }
1323
1324 void
1325 ctl_est_ua_port(struct ctl_lun *lun, int port, uint32_t except, ctl_ua_type ua)
1326 {
1327         int i;
1328
1329         mtx_assert(&lun->lun_lock, MA_OWNED);
1330         if (lun->pending_ua[port] == NULL)
1331                 return;
1332         for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1333                 if (port * CTL_MAX_INIT_PER_PORT + i == except)
1334                         continue;
1335                 lun->pending_ua[port][i] |= ua;
1336         }
1337 }
1338
1339 void
1340 ctl_est_ua_all(struct ctl_lun *lun, uint32_t except, ctl_ua_type ua)
1341 {
1342         struct ctl_softc *softc = lun->ctl_softc;
1343         int i;
1344
1345         mtx_assert(&lun->lun_lock, MA_OWNED);
1346         for (i = softc->port_min; i < softc->port_max; i++)
1347                 ctl_est_ua_port(lun, i, except, ua);
1348 }
1349
1350 void
1351 ctl_clr_ua(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua)
1352 {
1353         struct ctl_softc *softc = lun->ctl_softc;
1354         ctl_ua_type *pu;
1355
1356         if (initidx < softc->init_min || initidx >= softc->init_max)
1357                 return;
1358         mtx_assert(&lun->lun_lock, MA_OWNED);
1359         pu = lun->pending_ua[initidx / CTL_MAX_INIT_PER_PORT];
1360         if (pu == NULL)
1361                 return;
1362         pu[initidx % CTL_MAX_INIT_PER_PORT] &= ~ua;
1363 }
1364
1365 void
1366 ctl_clr_ua_all(struct ctl_lun *lun, uint32_t except, ctl_ua_type ua)
1367 {
1368         struct ctl_softc *softc = lun->ctl_softc;
1369         int i, j;
1370
1371         mtx_assert(&lun->lun_lock, MA_OWNED);
1372         for (i = softc->port_min; i < softc->port_max; i++) {
1373                 if (lun->pending_ua[i] == NULL)
1374                         continue;
1375                 for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
1376                         if (i * CTL_MAX_INIT_PER_PORT + j == except)
1377                                 continue;
1378                         lun->pending_ua[i][j] &= ~ua;
1379                 }
1380         }
1381 }
1382
1383 void
1384 ctl_clr_ua_allluns(struct ctl_softc *ctl_softc, uint32_t initidx,
1385     ctl_ua_type ua_type)
1386 {
1387         struct ctl_lun *lun;
1388
1389         mtx_assert(&ctl_softc->ctl_lock, MA_OWNED);
1390         STAILQ_FOREACH(lun, &ctl_softc->lun_list, links) {
1391                 mtx_lock(&lun->lun_lock);
1392                 ctl_clr_ua(lun, initidx, ua_type);
1393                 mtx_unlock(&lun->lun_lock);
1394         }
1395 }
1396
1397 static int
1398 ctl_ha_role_sysctl(SYSCTL_HANDLER_ARGS)
1399 {
1400         struct ctl_softc *softc = (struct ctl_softc *)arg1;
1401         struct ctl_lun *lun;
1402         struct ctl_lun_req ireq;
1403         int error, value;
1404
1405         value = (softc->flags & CTL_FLAG_ACTIVE_SHELF) ? 0 : 1;
1406         error = sysctl_handle_int(oidp, &value, 0, req);
1407         if ((error != 0) || (req->newptr == NULL))
1408                 return (error);
1409
1410         mtx_lock(&softc->ctl_lock);
1411         if (value == 0)
1412                 softc->flags |= CTL_FLAG_ACTIVE_SHELF;
1413         else
1414                 softc->flags &= ~CTL_FLAG_ACTIVE_SHELF;
1415         STAILQ_FOREACH(lun, &softc->lun_list, links) {
1416                 mtx_unlock(&softc->ctl_lock);
1417                 bzero(&ireq, sizeof(ireq));
1418                 ireq.reqtype = CTL_LUNREQ_MODIFY;
1419                 ireq.reqdata.modify.lun_id = lun->lun;
1420                 lun->backend->ioctl(NULL, CTL_LUN_REQ, (caddr_t)&ireq, 0,
1421                     curthread);
1422                 if (ireq.status != CTL_LUN_OK) {
1423                         printf("%s: CTL_LUNREQ_MODIFY returned %d '%s'\n",
1424                             __func__, ireq.status, ireq.error_str);
1425                 }
1426                 mtx_lock(&softc->ctl_lock);
1427         }
1428         mtx_unlock(&softc->ctl_lock);
1429         return (0);
1430 }
1431
1432 static int
1433 ctl_init(void)
1434 {
1435         struct ctl_softc *softc;
1436         void *other_pool;
1437         int i, error, retval;
1438
1439         retval = 0;
1440         control_softc = malloc(sizeof(*control_softc), M_DEVBUF,
1441                                M_WAITOK | M_ZERO);
1442         softc = control_softc;
1443
1444         softc->dev = make_dev(&ctl_cdevsw, 0, UID_ROOT, GID_OPERATOR, 0600,
1445                               "cam/ctl");
1446
1447         softc->dev->si_drv1 = softc;
1448
1449         sysctl_ctx_init(&softc->sysctl_ctx);
1450         softc->sysctl_tree = SYSCTL_ADD_NODE(&softc->sysctl_ctx,
1451                 SYSCTL_STATIC_CHILDREN(_kern_cam), OID_AUTO, "ctl",
1452                 CTLFLAG_RD, 0, "CAM Target Layer");
1453
1454         if (softc->sysctl_tree == NULL) {
1455                 printf("%s: unable to allocate sysctl tree\n", __func__);
1456                 destroy_dev(softc->dev);
1457                 free(control_softc, M_DEVBUF);
1458                 control_softc = NULL;
1459                 return (ENOMEM);
1460         }
1461
1462         mtx_init(&softc->ctl_lock, "CTL mutex", NULL, MTX_DEF);
1463         softc->io_zone = uma_zcreate("CTL IO", sizeof(union ctl_io),
1464             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1465         softc->open_count = 0;
1466
1467         /*
1468          * Default to actually sending a SYNCHRONIZE CACHE command down to
1469          * the drive.
1470          */
1471         softc->flags = CTL_FLAG_REAL_SYNC;
1472
1473         SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1474             OID_AUTO, "ha_mode", CTLFLAG_RDTUN, (int *)&softc->ha_mode, 0,
1475             "HA mode (0 - act/stby, 1 - serialize only, 2 - xfer)");
1476
1477         /*
1478          * In Copan's HA scheme, the "master" and "slave" roles are
1479          * figured out through the slot the controller is in.  Although it
1480          * is an active/active system, someone has to be in charge.
1481          */
1482         SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1483             OID_AUTO, "ha_id", CTLFLAG_RDTUN, &softc->ha_id, 0,
1484             "HA head ID (0 - no HA)");
1485         if (softc->ha_id == 0 || softc->ha_id > NUM_TARGET_PORT_GROUPS) {
1486                 softc->flags |= CTL_FLAG_ACTIVE_SHELF;
1487                 softc->is_single = 1;
1488                 softc->port_cnt = CTL_MAX_PORTS;
1489                 softc->port_min = 0;
1490         } else {
1491                 softc->port_cnt = CTL_MAX_PORTS / NUM_TARGET_PORT_GROUPS;
1492                 softc->port_min = (softc->ha_id - 1) * softc->port_cnt;
1493         }
1494         softc->port_max = softc->port_min + softc->port_cnt;
1495         softc->init_min = softc->port_min * CTL_MAX_INIT_PER_PORT;
1496         softc->init_max = softc->port_max * CTL_MAX_INIT_PER_PORT;
1497
1498         SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1499             OID_AUTO, "ha_link", CTLFLAG_RD, (int *)&softc->ha_link, 0,
1500             "HA link state (0 - offline, 1 - unknown, 2 - online)");
1501
1502         STAILQ_INIT(&softc->lun_list);
1503         STAILQ_INIT(&softc->pending_lun_queue);
1504         STAILQ_INIT(&softc->fe_list);
1505         STAILQ_INIT(&softc->port_list);
1506         STAILQ_INIT(&softc->be_list);
1507         ctl_tpc_init(softc);
1508
1509         if (ctl_pool_create(softc, "othersc", CTL_POOL_ENTRIES_OTHER_SC,
1510                             &other_pool) != 0)
1511         {
1512                 printf("ctl: can't allocate %d entry other SC pool, "
1513                        "exiting\n", CTL_POOL_ENTRIES_OTHER_SC);
1514                 return (ENOMEM);
1515         }
1516         softc->othersc_pool = other_pool;
1517
1518         if (worker_threads <= 0)
1519                 worker_threads = max(1, mp_ncpus / 4);
1520         if (worker_threads > CTL_MAX_THREADS)
1521                 worker_threads = CTL_MAX_THREADS;
1522
1523         for (i = 0; i < worker_threads; i++) {
1524                 struct ctl_thread *thr = &softc->threads[i];
1525
1526                 mtx_init(&thr->queue_lock, "CTL queue mutex", NULL, MTX_DEF);
1527                 thr->ctl_softc = softc;
1528                 STAILQ_INIT(&thr->incoming_queue);
1529                 STAILQ_INIT(&thr->rtr_queue);
1530                 STAILQ_INIT(&thr->done_queue);
1531                 STAILQ_INIT(&thr->isc_queue);
1532
1533                 error = kproc_kthread_add(ctl_work_thread, thr,
1534                     &softc->ctl_proc, &thr->thread, 0, 0, "ctl", "work%d", i);
1535                 if (error != 0) {
1536                         printf("error creating CTL work thread!\n");
1537                         ctl_pool_free(other_pool);
1538                         return (error);
1539                 }
1540         }
1541         error = kproc_kthread_add(ctl_lun_thread, softc,
1542             &softc->ctl_proc, NULL, 0, 0, "ctl", "lun");
1543         if (error != 0) {
1544                 printf("error creating CTL lun thread!\n");
1545                 ctl_pool_free(other_pool);
1546                 return (error);
1547         }
1548         error = kproc_kthread_add(ctl_thresh_thread, softc,
1549             &softc->ctl_proc, NULL, 0, 0, "ctl", "thresh");
1550         if (error != 0) {
1551                 printf("error creating CTL threshold thread!\n");
1552                 ctl_pool_free(other_pool);
1553                 return (error);
1554         }
1555
1556         SYSCTL_ADD_PROC(&softc->sysctl_ctx,SYSCTL_CHILDREN(softc->sysctl_tree),
1557             OID_AUTO, "ha_role", CTLTYPE_INT | CTLFLAG_RWTUN,
1558             softc, 0, ctl_ha_role_sysctl, "I", "HA role for this head");
1559
1560         if (softc->is_single == 0) {
1561                 ctl_frontend_register(&ha_frontend);
1562                 if (ctl_ha_msg_init(softc) != CTL_HA_STATUS_SUCCESS) {
1563                         printf("ctl_init: ctl_ha_msg_init failed.\n");
1564                         softc->is_single = 1;
1565                 } else
1566                 if (ctl_ha_msg_register(CTL_HA_CHAN_CTL, ctl_isc_event_handler)
1567                     != CTL_HA_STATUS_SUCCESS) {
1568                         printf("ctl_init: ctl_ha_msg_register failed.\n");
1569                         softc->is_single = 1;
1570                 }
1571         }
1572         return (0);
1573 }
1574
1575 void
1576 ctl_shutdown(void)
1577 {
1578         struct ctl_softc *softc;
1579         struct ctl_lun *lun, *next_lun;
1580
1581         softc = (struct ctl_softc *)control_softc;
1582
1583         if (softc->is_single == 0) {
1584                 ctl_ha_msg_shutdown(softc);
1585                 if (ctl_ha_msg_deregister(CTL_HA_CHAN_CTL)
1586                     != CTL_HA_STATUS_SUCCESS)
1587                         printf("%s: ctl_ha_msg_deregister failed.\n", __func__);
1588                 if (ctl_ha_msg_destroy(softc) != CTL_HA_STATUS_SUCCESS)
1589                         printf("%s: ctl_ha_msg_destroy failed.\n", __func__);
1590                 ctl_frontend_deregister(&ha_frontend);
1591         }
1592
1593         mtx_lock(&softc->ctl_lock);
1594
1595         /*
1596          * Free up each LUN.
1597          */
1598         for (lun = STAILQ_FIRST(&softc->lun_list); lun != NULL; lun = next_lun){
1599                 next_lun = STAILQ_NEXT(lun, links);
1600                 ctl_free_lun(lun);
1601         }
1602
1603         mtx_unlock(&softc->ctl_lock);
1604
1605 #if 0
1606         ctl_shutdown_thread(softc->work_thread);
1607         mtx_destroy(&softc->queue_lock);
1608 #endif
1609
1610         ctl_tpc_shutdown(softc);
1611         uma_zdestroy(softc->io_zone);
1612         mtx_destroy(&softc->ctl_lock);
1613
1614         destroy_dev(softc->dev);
1615
1616         sysctl_ctx_free(&softc->sysctl_ctx);
1617
1618         free(control_softc, M_DEVBUF);
1619         control_softc = NULL;
1620 }
1621
1622 static int
1623 ctl_module_event_handler(module_t mod, int what, void *arg)
1624 {
1625
1626         switch (what) {
1627         case MOD_LOAD:
1628                 return (ctl_init());
1629         case MOD_UNLOAD:
1630                 return (EBUSY);
1631         default:
1632                 return (EOPNOTSUPP);
1633         }
1634 }
1635
1636 /*
1637  * XXX KDM should we do some access checks here?  Bump a reference count to
1638  * prevent a CTL module from being unloaded while someone has it open?
1639  */
1640 static int
1641 ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td)
1642 {
1643         return (0);
1644 }
1645
1646 static int
1647 ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td)
1648 {
1649         return (0);
1650 }
1651
1652 /*
1653  * Remove an initiator by port number and initiator ID.
1654  * Returns 0 for success, -1 for failure.
1655  */
1656 int
1657 ctl_remove_initiator(struct ctl_port *port, int iid)
1658 {
1659         struct ctl_softc *softc = control_softc;
1660
1661         mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1662
1663         if (iid > CTL_MAX_INIT_PER_PORT) {
1664                 printf("%s: initiator ID %u > maximun %u!\n",
1665                        __func__, iid, CTL_MAX_INIT_PER_PORT);
1666                 return (-1);
1667         }
1668
1669         mtx_lock(&softc->ctl_lock);
1670         port->wwpn_iid[iid].in_use--;
1671         port->wwpn_iid[iid].last_use = time_uptime;
1672         mtx_unlock(&softc->ctl_lock);
1673         ctl_isc_announce_iid(port, iid);
1674
1675         return (0);
1676 }
1677
1678 /*
1679  * Add an initiator to the initiator map.
1680  * Returns iid for success, < 0 for failure.
1681  */
1682 int
1683 ctl_add_initiator(struct ctl_port *port, int iid, uint64_t wwpn, char *name)
1684 {
1685         struct ctl_softc *softc = control_softc;
1686         time_t best_time;
1687         int i, best;
1688
1689         mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1690
1691         if (iid >= CTL_MAX_INIT_PER_PORT) {
1692                 printf("%s: WWPN %#jx initiator ID %u > maximum %u!\n",
1693                        __func__, wwpn, iid, CTL_MAX_INIT_PER_PORT);
1694                 free(name, M_CTL);
1695                 return (-1);
1696         }
1697
1698         mtx_lock(&softc->ctl_lock);
1699
1700         if (iid < 0 && (wwpn != 0 || name != NULL)) {
1701                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1702                         if (wwpn != 0 && wwpn == port->wwpn_iid[i].wwpn) {
1703                                 iid = i;
1704                                 break;
1705                         }
1706                         if (name != NULL && port->wwpn_iid[i].name != NULL &&
1707                             strcmp(name, port->wwpn_iid[i].name) == 0) {
1708                                 iid = i;
1709                                 break;
1710                         }
1711                 }
1712         }
1713
1714         if (iid < 0) {
1715                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1716                         if (port->wwpn_iid[i].in_use == 0 &&
1717                             port->wwpn_iid[i].wwpn == 0 &&
1718                             port->wwpn_iid[i].name == NULL) {
1719                                 iid = i;
1720                                 break;
1721                         }
1722                 }
1723         }
1724
1725         if (iid < 0) {
1726                 best = -1;
1727                 best_time = INT32_MAX;
1728                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1729                         if (port->wwpn_iid[i].in_use == 0) {
1730                                 if (port->wwpn_iid[i].last_use < best_time) {
1731                                         best = i;
1732                                         best_time = port->wwpn_iid[i].last_use;
1733                                 }
1734                         }
1735                 }
1736                 iid = best;
1737         }
1738
1739         if (iid < 0) {
1740                 mtx_unlock(&softc->ctl_lock);
1741                 free(name, M_CTL);
1742                 return (-2);
1743         }
1744
1745         if (port->wwpn_iid[iid].in_use > 0 && (wwpn != 0 || name != NULL)) {
1746                 /*
1747                  * This is not an error yet.
1748                  */
1749                 if (wwpn != 0 && wwpn == port->wwpn_iid[iid].wwpn) {
1750 #if 0
1751                         printf("%s: port %d iid %u WWPN %#jx arrived"
1752                             " again\n", __func__, port->targ_port,
1753                             iid, (uintmax_t)wwpn);
1754 #endif
1755                         goto take;
1756                 }
1757                 if (name != NULL && port->wwpn_iid[iid].name != NULL &&
1758                     strcmp(name, port->wwpn_iid[iid].name) == 0) {
1759 #if 0
1760                         printf("%s: port %d iid %u name '%s' arrived"
1761                             " again\n", __func__, port->targ_port,
1762                             iid, name);
1763 #endif
1764                         goto take;
1765                 }
1766
1767                 /*
1768                  * This is an error, but what do we do about it?  The
1769                  * driver is telling us we have a new WWPN for this
1770                  * initiator ID, so we pretty much need to use it.
1771                  */
1772                 printf("%s: port %d iid %u WWPN %#jx '%s' arrived,"
1773                     " but WWPN %#jx '%s' is still at that address\n",
1774                     __func__, port->targ_port, iid, wwpn, name,
1775                     (uintmax_t)port->wwpn_iid[iid].wwpn,
1776                     port->wwpn_iid[iid].name);
1777
1778                 /*
1779                  * XXX KDM clear have_ca and ua_pending on each LUN for
1780                  * this initiator.
1781                  */
1782         }
1783 take:
1784         free(port->wwpn_iid[iid].name, M_CTL);
1785         port->wwpn_iid[iid].name = name;
1786         port->wwpn_iid[iid].wwpn = wwpn;
1787         port->wwpn_iid[iid].in_use++;
1788         mtx_unlock(&softc->ctl_lock);
1789         ctl_isc_announce_iid(port, iid);
1790
1791         return (iid);
1792 }
1793
1794 static int
1795 ctl_create_iid(struct ctl_port *port, int iid, uint8_t *buf)
1796 {
1797         int len;
1798
1799         switch (port->port_type) {
1800         case CTL_PORT_FC:
1801         {
1802                 struct scsi_transportid_fcp *id =
1803                     (struct scsi_transportid_fcp *)buf;
1804                 if (port->wwpn_iid[iid].wwpn == 0)
1805                         return (0);
1806                 memset(id, 0, sizeof(*id));
1807                 id->format_protocol = SCSI_PROTO_FC;
1808                 scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->n_port_name);
1809                 return (sizeof(*id));
1810         }
1811         case CTL_PORT_ISCSI:
1812         {
1813                 struct scsi_transportid_iscsi_port *id =
1814                     (struct scsi_transportid_iscsi_port *)buf;
1815                 if (port->wwpn_iid[iid].name == NULL)
1816                         return (0);
1817                 memset(id, 0, 256);
1818                 id->format_protocol = SCSI_TRN_ISCSI_FORMAT_PORT |
1819                     SCSI_PROTO_ISCSI;
1820                 len = strlcpy(id->iscsi_name, port->wwpn_iid[iid].name, 252) + 1;
1821                 len = roundup2(min(len, 252), 4);
1822                 scsi_ulto2b(len, id->additional_length);
1823                 return (sizeof(*id) + len);
1824         }
1825         case CTL_PORT_SAS:
1826         {
1827                 struct scsi_transportid_sas *id =
1828                     (struct scsi_transportid_sas *)buf;
1829                 if (port->wwpn_iid[iid].wwpn == 0)
1830                         return (0);
1831                 memset(id, 0, sizeof(*id));
1832                 id->format_protocol = SCSI_PROTO_SAS;
1833                 scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->sas_address);
1834                 return (sizeof(*id));
1835         }
1836         default:
1837         {
1838                 struct scsi_transportid_spi *id =
1839                     (struct scsi_transportid_spi *)buf;
1840                 memset(id, 0, sizeof(*id));
1841                 id->format_protocol = SCSI_PROTO_SPI;
1842                 scsi_ulto2b(iid, id->scsi_addr);
1843                 scsi_ulto2b(port->targ_port, id->rel_trgt_port_id);
1844                 return (sizeof(*id));
1845         }
1846         }
1847 }
1848
1849 /*
1850  * Serialize a command that went down the "wrong" side, and so was sent to
1851  * this controller for execution.  The logic is a little different than the
1852  * standard case in ctl_scsiio_precheck().  Errors in this case need to get
1853  * sent back to the other side, but in the success case, we execute the
1854  * command on this side (XFER mode) or tell the other side to execute it
1855  * (SER_ONLY mode).
1856  */
1857 static int
1858 ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio)
1859 {
1860         struct ctl_softc *softc;
1861         union ctl_ha_msg msg_info;
1862         struct ctl_lun *lun;
1863         const struct ctl_cmd_entry *entry;
1864         int retval = 0;
1865         uint32_t targ_lun;
1866
1867         softc = control_softc;
1868
1869         targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
1870         mtx_lock(&softc->ctl_lock);
1871         if ((targ_lun < CTL_MAX_LUNS) &&
1872             ((lun = softc->ctl_luns[targ_lun]) != NULL)) {
1873                 mtx_lock(&lun->lun_lock);
1874                 mtx_unlock(&softc->ctl_lock);
1875                 /*
1876                  * If the LUN is invalid, pretend that it doesn't exist.
1877                  * It will go away as soon as all pending I/O has been
1878                  * completed.
1879                  */
1880                 if (lun->flags & CTL_LUN_DISABLED) {
1881                         mtx_unlock(&lun->lun_lock);
1882                         lun = NULL;
1883                 }
1884         } else {
1885                 mtx_unlock(&softc->ctl_lock);
1886                 lun = NULL;
1887         }
1888         if (lun == NULL) {
1889                 /*
1890                  * The other node would not send this request to us unless
1891                  * received announce that we are primary node for this LUN.
1892                  * If this LUN does not exist now, it is probably result of
1893                  * a race, so respond to initiator in the most opaque way.
1894                  */
1895                 ctl_set_busy(ctsio);
1896                 ctl_copy_sense_data_back((union ctl_io *)ctsio, &msg_info);
1897                 msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1898                 msg_info.hdr.serializing_sc = NULL;
1899                 msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1900                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1901                     sizeof(msg_info.scsi), M_WAITOK);
1902                 return(1);
1903         }
1904
1905         entry = ctl_get_cmd_entry(ctsio, NULL);
1906         if (ctl_scsiio_lun_check(lun, entry, ctsio) != 0) {
1907                 mtx_unlock(&lun->lun_lock);
1908                 ctl_copy_sense_data_back((union ctl_io *)ctsio, &msg_info);
1909                 msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1910                 msg_info.hdr.serializing_sc = NULL;
1911                 msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1912                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1913                     sizeof(msg_info.scsi), M_WAITOK);
1914                 return(1);
1915         }
1916
1917         ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = lun;
1918         ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = lun->be_lun;
1919
1920         /*
1921          * Every I/O goes into the OOA queue for a
1922          * particular LUN, and stays there until completion.
1923          */
1924 #ifdef CTL_TIME_IO
1925         if (TAILQ_EMPTY(&lun->ooa_queue))
1926                 lun->idle_time += getsbinuptime() - lun->last_busy;
1927 #endif
1928         TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1929
1930         switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
1931                 (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr, ctl_ooaq,
1932                  ooa_links))) {
1933         case CTL_ACTION_BLOCK:
1934                 ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
1935                 TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
1936                                   blocked_links);
1937                 mtx_unlock(&lun->lun_lock);
1938                 break;
1939         case CTL_ACTION_PASS:
1940         case CTL_ACTION_SKIP:
1941                 if (softc->ha_mode == CTL_HA_MODE_XFER) {
1942                         ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
1943                         ctl_enqueue_rtr((union ctl_io *)ctsio);
1944                         mtx_unlock(&lun->lun_lock);
1945                 } else {
1946                         ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
1947                         mtx_unlock(&lun->lun_lock);
1948
1949                         /* send msg back to other side */
1950                         msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1951                         msg_info.hdr.serializing_sc = (union ctl_io *)ctsio;
1952                         msg_info.hdr.msg_type = CTL_MSG_R2R;
1953                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1954                             sizeof(msg_info.hdr), M_WAITOK);
1955                 }
1956                 break;
1957         case CTL_ACTION_OVERLAP:
1958                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1959                 mtx_unlock(&lun->lun_lock);
1960                 retval = 1;
1961
1962                 ctl_set_overlapped_cmd(ctsio);
1963                 ctl_copy_sense_data_back((union ctl_io *)ctsio, &msg_info);
1964                 msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1965                 msg_info.hdr.serializing_sc = NULL;
1966                 msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1967                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1968                     sizeof(msg_info.scsi), M_WAITOK);
1969                 break;
1970         case CTL_ACTION_OVERLAP_TAG:
1971                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1972                 mtx_unlock(&lun->lun_lock);
1973                 retval = 1;
1974                 ctl_set_overlapped_tag(ctsio, ctsio->tag_num);
1975                 ctl_copy_sense_data_back((union ctl_io *)ctsio, &msg_info);
1976                 msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1977                 msg_info.hdr.serializing_sc = NULL;
1978                 msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1979                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1980                     sizeof(msg_info.scsi), M_WAITOK);
1981                 break;
1982         case CTL_ACTION_ERROR:
1983         default:
1984                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1985                 mtx_unlock(&lun->lun_lock);
1986                 retval = 1;
1987
1988                 ctl_set_internal_failure(ctsio, /*sks_valid*/ 0,
1989                                          /*retry_count*/ 0);
1990                 ctl_copy_sense_data_back((union ctl_io *)ctsio, &msg_info);
1991                 msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1992                 msg_info.hdr.serializing_sc = NULL;
1993                 msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1994                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1995                     sizeof(msg_info.scsi), M_WAITOK);
1996                 break;
1997         }
1998         return (retval);
1999 }
2000
2001 /*
2002  * Returns 0 for success, errno for failure.
2003  */
2004 static int
2005 ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
2006                    struct ctl_ooa *ooa_hdr, struct ctl_ooa_entry *kern_entries)
2007 {
2008         union ctl_io *io;
2009         int retval;
2010
2011         retval = 0;
2012
2013         mtx_lock(&lun->lun_lock);
2014         for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); (io != NULL);
2015              (*cur_fill_num)++, io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2016              ooa_links)) {
2017                 struct ctl_ooa_entry *entry;
2018
2019                 /*
2020                  * If we've got more than we can fit, just count the
2021                  * remaining entries.
2022                  */
2023                 if (*cur_fill_num >= ooa_hdr->alloc_num)
2024                         continue;
2025
2026                 entry = &kern_entries[*cur_fill_num];
2027
2028                 entry->tag_num = io->scsiio.tag_num;
2029                 entry->lun_num = lun->lun;
2030 #ifdef CTL_TIME_IO
2031                 entry->start_bt = io->io_hdr.start_bt;
2032 #endif
2033                 bcopy(io->scsiio.cdb, entry->cdb, io->scsiio.cdb_len);
2034                 entry->cdb_len = io->scsiio.cdb_len;
2035                 if (io->io_hdr.flags & CTL_FLAG_BLOCKED)
2036                         entry->cmd_flags |= CTL_OOACMD_FLAG_BLOCKED;
2037
2038                 if (io->io_hdr.flags & CTL_FLAG_DMA_INPROG)
2039                         entry->cmd_flags |= CTL_OOACMD_FLAG_DMA;
2040
2041                 if (io->io_hdr.flags & CTL_FLAG_ABORT)
2042                         entry->cmd_flags |= CTL_OOACMD_FLAG_ABORT;
2043
2044                 if (io->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR)
2045                         entry->cmd_flags |= CTL_OOACMD_FLAG_RTR;
2046
2047                 if (io->io_hdr.flags & CTL_FLAG_DMA_QUEUED)
2048                         entry->cmd_flags |= CTL_OOACMD_FLAG_DMA_QUEUED;
2049         }
2050         mtx_unlock(&lun->lun_lock);
2051
2052         return (retval);
2053 }
2054
2055 static void *
2056 ctl_copyin_alloc(void *user_addr, int len, char *error_str,
2057                  size_t error_str_len)
2058 {
2059         void *kptr;
2060
2061         kptr = malloc(len, M_CTL, M_WAITOK | M_ZERO);
2062
2063         if (copyin(user_addr, kptr, len) != 0) {
2064                 snprintf(error_str, error_str_len, "Error copying %d bytes "
2065                          "from user address %p to kernel address %p", len,
2066                          user_addr, kptr);
2067                 free(kptr, M_CTL);
2068                 return (NULL);
2069         }
2070
2071         return (kptr);
2072 }
2073
2074 static void
2075 ctl_free_args(int num_args, struct ctl_be_arg *args)
2076 {
2077         int i;
2078
2079         if (args == NULL)
2080                 return;
2081
2082         for (i = 0; i < num_args; i++) {
2083                 free(args[i].kname, M_CTL);
2084                 free(args[i].kvalue, M_CTL);
2085         }
2086
2087         free(args, M_CTL);
2088 }
2089
2090 static struct ctl_be_arg *
2091 ctl_copyin_args(int num_args, struct ctl_be_arg *uargs,
2092                 char *error_str, size_t error_str_len)
2093 {
2094         struct ctl_be_arg *args;
2095         int i;
2096
2097         args = ctl_copyin_alloc(uargs, num_args * sizeof(*args),
2098                                 error_str, error_str_len);
2099
2100         if (args == NULL)
2101                 goto bailout;
2102
2103         for (i = 0; i < num_args; i++) {
2104                 args[i].kname = NULL;
2105                 args[i].kvalue = NULL;
2106         }
2107
2108         for (i = 0; i < num_args; i++) {
2109                 uint8_t *tmpptr;
2110
2111                 args[i].kname = ctl_copyin_alloc(args[i].name,
2112                         args[i].namelen, error_str, error_str_len);
2113                 if (args[i].kname == NULL)
2114                         goto bailout;
2115
2116                 if (args[i].kname[args[i].namelen - 1] != '\0') {
2117                         snprintf(error_str, error_str_len, "Argument %d "
2118                                  "name is not NUL-terminated", i);
2119                         goto bailout;
2120                 }
2121
2122                 if (args[i].flags & CTL_BEARG_RD) {
2123                         tmpptr = ctl_copyin_alloc(args[i].value,
2124                                 args[i].vallen, error_str, error_str_len);
2125                         if (tmpptr == NULL)
2126                                 goto bailout;
2127                         if ((args[i].flags & CTL_BEARG_ASCII)
2128                          && (tmpptr[args[i].vallen - 1] != '\0')) {
2129                                 snprintf(error_str, error_str_len, "Argument "
2130                                     "%d value is not NUL-terminated", i);
2131                                 goto bailout;
2132                         }
2133                         args[i].kvalue = tmpptr;
2134                 } else {
2135                         args[i].kvalue = malloc(args[i].vallen,
2136                             M_CTL, M_WAITOK | M_ZERO);
2137                 }
2138         }
2139
2140         return (args);
2141 bailout:
2142
2143         ctl_free_args(num_args, args);
2144
2145         return (NULL);
2146 }
2147
2148 static void
2149 ctl_copyout_args(int num_args, struct ctl_be_arg *args)
2150 {
2151         int i;
2152
2153         for (i = 0; i < num_args; i++) {
2154                 if (args[i].flags & CTL_BEARG_WR)
2155                         copyout(args[i].kvalue, args[i].value, args[i].vallen);
2156         }
2157 }
2158
2159 /*
2160  * Escape characters that are illegal or not recommended in XML.
2161  */
2162 int
2163 ctl_sbuf_printf_esc(struct sbuf *sb, char *str, int size)
2164 {
2165         char *end = str + size;
2166         int retval;
2167
2168         retval = 0;
2169
2170         for (; *str && str < end; str++) {
2171                 switch (*str) {
2172                 case '&':
2173                         retval = sbuf_printf(sb, "&amp;");
2174                         break;
2175                 case '>':
2176                         retval = sbuf_printf(sb, "&gt;");
2177                         break;
2178                 case '<':
2179                         retval = sbuf_printf(sb, "&lt;");
2180                         break;
2181                 default:
2182                         retval = sbuf_putc(sb, *str);
2183                         break;
2184                 }
2185
2186                 if (retval != 0)
2187                         break;
2188
2189         }
2190
2191         return (retval);
2192 }
2193
2194 static void
2195 ctl_id_sbuf(struct ctl_devid *id, struct sbuf *sb)
2196 {
2197         struct scsi_vpd_id_descriptor *desc;
2198         int i;
2199
2200         if (id == NULL || id->len < 4)
2201                 return;
2202         desc = (struct scsi_vpd_id_descriptor *)id->data;
2203         switch (desc->id_type & SVPD_ID_TYPE_MASK) {
2204         case SVPD_ID_TYPE_T10:
2205                 sbuf_printf(sb, "t10.");
2206                 break;
2207         case SVPD_ID_TYPE_EUI64:
2208                 sbuf_printf(sb, "eui.");
2209                 break;
2210         case SVPD_ID_TYPE_NAA:
2211                 sbuf_printf(sb, "naa.");
2212                 break;
2213         case SVPD_ID_TYPE_SCSI_NAME:
2214                 break;
2215         }
2216         switch (desc->proto_codeset & SVPD_ID_CODESET_MASK) {
2217         case SVPD_ID_CODESET_BINARY:
2218                 for (i = 0; i < desc->length; i++)
2219                         sbuf_printf(sb, "%02x", desc->identifier[i]);
2220                 break;
2221         case SVPD_ID_CODESET_ASCII:
2222                 sbuf_printf(sb, "%.*s", (int)desc->length,
2223                     (char *)desc->identifier);
2224                 break;
2225         case SVPD_ID_CODESET_UTF8:
2226                 sbuf_printf(sb, "%s", (char *)desc->identifier);
2227                 break;
2228         }
2229 }
2230
2231 static int
2232 ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
2233           struct thread *td)
2234 {
2235         struct ctl_softc *softc;
2236         struct ctl_lun *lun;
2237         int retval;
2238
2239         softc = control_softc;
2240
2241         retval = 0;
2242
2243         switch (cmd) {
2244         case CTL_IO:
2245                 retval = ctl_ioctl_io(dev, cmd, addr, flag, td);
2246                 break;
2247         case CTL_ENABLE_PORT:
2248         case CTL_DISABLE_PORT:
2249         case CTL_SET_PORT_WWNS: {
2250                 struct ctl_port *port;
2251                 struct ctl_port_entry *entry;
2252
2253                 entry = (struct ctl_port_entry *)addr;
2254
2255                 mtx_lock(&softc->ctl_lock);
2256                 STAILQ_FOREACH(port, &softc->port_list, links) {
2257                         int action, done;
2258
2259                         if (port->targ_port < softc->port_min ||
2260                             port->targ_port >= softc->port_max)
2261                                 continue;
2262
2263                         action = 0;
2264                         done = 0;
2265                         if ((entry->port_type == CTL_PORT_NONE)
2266                          && (entry->targ_port == port->targ_port)) {
2267                                 /*
2268                                  * If the user only wants to enable or
2269                                  * disable or set WWNs on a specific port,
2270                                  * do the operation and we're done.
2271                                  */
2272                                 action = 1;
2273                                 done = 1;
2274                         } else if (entry->port_type & port->port_type) {
2275                                 /*
2276                                  * Compare the user's type mask with the
2277                                  * particular frontend type to see if we
2278                                  * have a match.
2279                                  */
2280                                 action = 1;
2281                                 done = 0;
2282
2283                                 /*
2284                                  * Make sure the user isn't trying to set
2285                                  * WWNs on multiple ports at the same time.
2286                                  */
2287                                 if (cmd == CTL_SET_PORT_WWNS) {
2288                                         printf("%s: Can't set WWNs on "
2289                                                "multiple ports\n", __func__);
2290                                         retval = EINVAL;
2291                                         break;
2292                                 }
2293                         }
2294                         if (action == 0)
2295                                 continue;
2296
2297                         /*
2298                          * XXX KDM we have to drop the lock here, because
2299                          * the online/offline operations can potentially
2300                          * block.  We need to reference count the frontends
2301                          * so they can't go away,
2302                          */
2303                         if (cmd == CTL_ENABLE_PORT) {
2304                                 mtx_unlock(&softc->ctl_lock);
2305                                 ctl_port_online(port);
2306                                 mtx_lock(&softc->ctl_lock);
2307                         } else if (cmd == CTL_DISABLE_PORT) {
2308                                 mtx_unlock(&softc->ctl_lock);
2309                                 ctl_port_offline(port);
2310                                 mtx_lock(&softc->ctl_lock);
2311                         } else if (cmd == CTL_SET_PORT_WWNS) {
2312                                 ctl_port_set_wwns(port,
2313                                     (entry->flags & CTL_PORT_WWNN_VALID) ?
2314                                     1 : 0, entry->wwnn,
2315                                     (entry->flags & CTL_PORT_WWPN_VALID) ?
2316                                     1 : 0, entry->wwpn);
2317                         }
2318                         if (done != 0)
2319                                 break;
2320                 }
2321                 mtx_unlock(&softc->ctl_lock);
2322                 break;
2323         }
2324         case CTL_GET_PORT_LIST: {
2325                 struct ctl_port *port;
2326                 struct ctl_port_list *list;
2327                 int i;
2328
2329                 list = (struct ctl_port_list *)addr;
2330
2331                 if (list->alloc_len != (list->alloc_num *
2332                     sizeof(struct ctl_port_entry))) {
2333                         printf("%s: CTL_GET_PORT_LIST: alloc_len %u != "
2334                                "alloc_num %u * sizeof(struct ctl_port_entry) "
2335                                "%zu\n", __func__, list->alloc_len,
2336                                list->alloc_num, sizeof(struct ctl_port_entry));
2337                         retval = EINVAL;
2338                         break;
2339                 }
2340                 list->fill_len = 0;
2341                 list->fill_num = 0;
2342                 list->dropped_num = 0;
2343                 i = 0;
2344                 mtx_lock(&softc->ctl_lock);
2345                 STAILQ_FOREACH(port, &softc->port_list, links) {
2346                         struct ctl_port_entry entry, *list_entry;
2347
2348                         if (list->fill_num >= list->alloc_num) {
2349                                 list->dropped_num++;
2350                                 continue;
2351                         }
2352
2353                         entry.port_type = port->port_type;
2354                         strlcpy(entry.port_name, port->port_name,
2355                                 sizeof(entry.port_name));
2356                         entry.targ_port = port->targ_port;
2357                         entry.physical_port = port->physical_port;
2358                         entry.virtual_port = port->virtual_port;
2359                         entry.wwnn = port->wwnn;
2360                         entry.wwpn = port->wwpn;
2361                         if (port->status & CTL_PORT_STATUS_ONLINE)
2362                                 entry.online = 1;
2363                         else
2364                                 entry.online = 0;
2365
2366                         list_entry = &list->entries[i];
2367
2368                         retval = copyout(&entry, list_entry, sizeof(entry));
2369                         if (retval != 0) {
2370                                 printf("%s: CTL_GET_PORT_LIST: copyout "
2371                                        "returned %d\n", __func__, retval);
2372                                 break;
2373                         }
2374                         i++;
2375                         list->fill_num++;
2376                         list->fill_len += sizeof(entry);
2377                 }
2378                 mtx_unlock(&softc->ctl_lock);
2379
2380                 /*
2381                  * If this is non-zero, we had a copyout fault, so there's
2382                  * probably no point in attempting to set the status inside
2383                  * the structure.
2384                  */
2385                 if (retval != 0)
2386                         break;
2387
2388                 if (list->dropped_num > 0)
2389                         list->status = CTL_PORT_LIST_NEED_MORE_SPACE;
2390                 else
2391                         list->status = CTL_PORT_LIST_OK;
2392                 break;
2393         }
2394         case CTL_DUMP_OOA: {
2395                 union ctl_io *io;
2396                 char printbuf[128];
2397                 struct sbuf sb;
2398
2399                 mtx_lock(&softc->ctl_lock);
2400                 printf("Dumping OOA queues:\n");
2401                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
2402                         mtx_lock(&lun->lun_lock);
2403                         for (io = (union ctl_io *)TAILQ_FIRST(
2404                              &lun->ooa_queue); io != NULL;
2405                              io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2406                              ooa_links)) {
2407                                 sbuf_new(&sb, printbuf, sizeof(printbuf),
2408                                          SBUF_FIXEDLEN);
2409                                 sbuf_printf(&sb, "LUN %jd tag 0x%04x%s%s%s%s: ",
2410                                             (intmax_t)lun->lun,
2411                                             io->scsiio.tag_num,
2412                                             (io->io_hdr.flags &
2413                                             CTL_FLAG_BLOCKED) ? "" : " BLOCKED",
2414                                             (io->io_hdr.flags &
2415                                             CTL_FLAG_DMA_INPROG) ? " DMA" : "",
2416                                             (io->io_hdr.flags &
2417                                             CTL_FLAG_ABORT) ? " ABORT" : "",
2418                                             (io->io_hdr.flags &
2419                                         CTL_FLAG_IS_WAS_ON_RTR) ? " RTR" : "");
2420                                 ctl_scsi_command_string(&io->scsiio, NULL, &sb);
2421                                 sbuf_finish(&sb);
2422                                 printf("%s\n", sbuf_data(&sb));
2423                         }
2424                         mtx_unlock(&lun->lun_lock);
2425                 }
2426                 printf("OOA queues dump done\n");
2427                 mtx_unlock(&softc->ctl_lock);
2428                 break;
2429         }
2430         case CTL_GET_OOA: {
2431                 struct ctl_ooa *ooa_hdr;
2432                 struct ctl_ooa_entry *entries;
2433                 uint32_t cur_fill_num;
2434
2435                 ooa_hdr = (struct ctl_ooa *)addr;
2436
2437                 if ((ooa_hdr->alloc_len == 0)
2438                  || (ooa_hdr->alloc_num == 0)) {
2439                         printf("%s: CTL_GET_OOA: alloc len %u and alloc num %u "
2440                                "must be non-zero\n", __func__,
2441                                ooa_hdr->alloc_len, ooa_hdr->alloc_num);
2442                         retval = EINVAL;
2443                         break;
2444                 }
2445
2446                 if (ooa_hdr->alloc_len != (ooa_hdr->alloc_num *
2447                     sizeof(struct ctl_ooa_entry))) {
2448                         printf("%s: CTL_GET_OOA: alloc len %u must be alloc "
2449                                "num %d * sizeof(struct ctl_ooa_entry) %zd\n",
2450                                __func__, ooa_hdr->alloc_len,
2451                                ooa_hdr->alloc_num,sizeof(struct ctl_ooa_entry));
2452                         retval = EINVAL;
2453                         break;
2454                 }
2455
2456                 entries = malloc(ooa_hdr->alloc_len, M_CTL, M_WAITOK | M_ZERO);
2457                 if (entries == NULL) {
2458                         printf("%s: could not allocate %d bytes for OOA "
2459                                "dump\n", __func__, ooa_hdr->alloc_len);
2460                         retval = ENOMEM;
2461                         break;
2462                 }
2463
2464                 mtx_lock(&softc->ctl_lock);
2465                 if (((ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) == 0)
2466                  && ((ooa_hdr->lun_num >= CTL_MAX_LUNS)
2467                   || (softc->ctl_luns[ooa_hdr->lun_num] == NULL))) {
2468                         mtx_unlock(&softc->ctl_lock);
2469                         free(entries, M_CTL);
2470                         printf("%s: CTL_GET_OOA: invalid LUN %ju\n",
2471                                __func__, (uintmax_t)ooa_hdr->lun_num);
2472                         retval = EINVAL;
2473                         break;
2474                 }
2475
2476                 cur_fill_num = 0;
2477
2478                 if (ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) {
2479                         STAILQ_FOREACH(lun, &softc->lun_list, links) {
2480                                 retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,
2481                                         ooa_hdr, entries);
2482                                 if (retval != 0)
2483                                         break;
2484                         }
2485                         if (retval != 0) {
2486                                 mtx_unlock(&softc->ctl_lock);
2487                                 free(entries, M_CTL);
2488                                 break;
2489                         }
2490                 } else {
2491                         lun = softc->ctl_luns[ooa_hdr->lun_num];
2492
2493                         retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,ooa_hdr,
2494                                                     entries);
2495                 }
2496                 mtx_unlock(&softc->ctl_lock);
2497
2498                 ooa_hdr->fill_num = min(cur_fill_num, ooa_hdr->alloc_num);
2499                 ooa_hdr->fill_len = ooa_hdr->fill_num *
2500                         sizeof(struct ctl_ooa_entry);
2501                 retval = copyout(entries, ooa_hdr->entries, ooa_hdr->fill_len);
2502                 if (retval != 0) {
2503                         printf("%s: error copying out %d bytes for OOA dump\n",
2504                                __func__, ooa_hdr->fill_len);
2505                 }
2506
2507                 getbintime(&ooa_hdr->cur_bt);
2508
2509                 if (cur_fill_num > ooa_hdr->alloc_num) {
2510                         ooa_hdr->dropped_num = cur_fill_num -ooa_hdr->alloc_num;
2511                         ooa_hdr->status = CTL_OOA_NEED_MORE_SPACE;
2512                 } else {
2513                         ooa_hdr->dropped_num = 0;
2514                         ooa_hdr->status = CTL_OOA_OK;
2515                 }
2516
2517                 free(entries, M_CTL);
2518                 break;
2519         }
2520         case CTL_CHECK_OOA: {
2521                 union ctl_io *io;
2522                 struct ctl_ooa_info *ooa_info;
2523
2524
2525                 ooa_info = (struct ctl_ooa_info *)addr;
2526
2527                 if (ooa_info->lun_id >= CTL_MAX_LUNS) {
2528                         ooa_info->status = CTL_OOA_INVALID_LUN;
2529                         break;
2530                 }
2531                 mtx_lock(&softc->ctl_lock);
2532                 lun = softc->ctl_luns[ooa_info->lun_id];
2533                 if (lun == NULL) {
2534                         mtx_unlock(&softc->ctl_lock);
2535                         ooa_info->status = CTL_OOA_INVALID_LUN;
2536                         break;
2537                 }
2538                 mtx_lock(&lun->lun_lock);
2539                 mtx_unlock(&softc->ctl_lock);
2540                 ooa_info->num_entries = 0;
2541                 for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
2542                      io != NULL; io = (union ctl_io *)TAILQ_NEXT(
2543                      &io->io_hdr, ooa_links)) {
2544                         ooa_info->num_entries++;
2545                 }
2546                 mtx_unlock(&lun->lun_lock);
2547
2548                 ooa_info->status = CTL_OOA_SUCCESS;
2549
2550                 break;
2551         }
2552         case CTL_DELAY_IO: {
2553                 struct ctl_io_delay_info *delay_info;
2554
2555                 delay_info = (struct ctl_io_delay_info *)addr;
2556
2557 #ifdef CTL_IO_DELAY
2558                 mtx_lock(&softc->ctl_lock);
2559
2560                 if ((delay_info->lun_id >= CTL_MAX_LUNS)
2561                  || (softc->ctl_luns[delay_info->lun_id] == NULL)) {
2562                         delay_info->status = CTL_DELAY_STATUS_INVALID_LUN;
2563                 } else {
2564                         lun = softc->ctl_luns[delay_info->lun_id];
2565                         mtx_lock(&lun->lun_lock);
2566
2567                         delay_info->status = CTL_DELAY_STATUS_OK;
2568
2569                         switch (delay_info->delay_type) {
2570                         case CTL_DELAY_TYPE_CONT:
2571                                 break;
2572                         case CTL_DELAY_TYPE_ONESHOT:
2573                                 break;
2574                         default:
2575                                 delay_info->status =
2576                                         CTL_DELAY_STATUS_INVALID_TYPE;
2577                                 break;
2578                         }
2579
2580                         switch (delay_info->delay_loc) {
2581                         case CTL_DELAY_LOC_DATAMOVE:
2582                                 lun->delay_info.datamove_type =
2583                                         delay_info->delay_type;
2584                                 lun->delay_info.datamove_delay =
2585                                         delay_info->delay_secs;
2586                                 break;
2587                         case CTL_DELAY_LOC_DONE:
2588                                 lun->delay_info.done_type =
2589                                         delay_info->delay_type;
2590                                 lun->delay_info.done_delay =
2591                                         delay_info->delay_secs;
2592                                 break;
2593                         default:
2594                                 delay_info->status =
2595                                         CTL_DELAY_STATUS_INVALID_LOC;
2596                                 break;
2597                         }
2598                         mtx_unlock(&lun->lun_lock);
2599                 }
2600
2601                 mtx_unlock(&softc->ctl_lock);
2602 #else
2603                 delay_info->status = CTL_DELAY_STATUS_NOT_IMPLEMENTED;
2604 #endif /* CTL_IO_DELAY */
2605                 break;
2606         }
2607         case CTL_REALSYNC_SET: {
2608                 int *syncstate;
2609
2610                 syncstate = (int *)addr;
2611
2612                 mtx_lock(&softc->ctl_lock);
2613                 switch (*syncstate) {
2614                 case 0:
2615                         softc->flags &= ~CTL_FLAG_REAL_SYNC;
2616                         break;
2617                 case 1:
2618                         softc->flags |= CTL_FLAG_REAL_SYNC;
2619                         break;
2620                 default:
2621                         retval = EINVAL;
2622                         break;
2623                 }
2624                 mtx_unlock(&softc->ctl_lock);
2625                 break;
2626         }
2627         case CTL_REALSYNC_GET: {
2628                 int *syncstate;
2629
2630                 syncstate = (int*)addr;
2631
2632                 mtx_lock(&softc->ctl_lock);
2633                 if (softc->flags & CTL_FLAG_REAL_SYNC)
2634                         *syncstate = 1;
2635                 else
2636                         *syncstate = 0;
2637                 mtx_unlock(&softc->ctl_lock);
2638
2639                 break;
2640         }
2641         case CTL_SETSYNC:
2642         case CTL_GETSYNC: {
2643                 struct ctl_sync_info *sync_info;
2644
2645                 sync_info = (struct ctl_sync_info *)addr;
2646
2647                 mtx_lock(&softc->ctl_lock);
2648                 lun = softc->ctl_luns[sync_info->lun_id];
2649                 if (lun == NULL) {
2650                         mtx_unlock(&softc->ctl_lock);
2651                         sync_info->status = CTL_GS_SYNC_NO_LUN;
2652                         break;
2653                 }
2654                 /*
2655                  * Get or set the sync interval.  We're not bounds checking
2656                  * in the set case, hopefully the user won't do something
2657                  * silly.
2658                  */
2659                 mtx_lock(&lun->lun_lock);
2660                 mtx_unlock(&softc->ctl_lock);
2661                 if (cmd == CTL_GETSYNC)
2662                         sync_info->sync_interval = lun->sync_interval;
2663                 else
2664                         lun->sync_interval = sync_info->sync_interval;
2665                 mtx_unlock(&lun->lun_lock);
2666
2667                 sync_info->status = CTL_GS_SYNC_OK;
2668
2669                 break;
2670         }
2671         case CTL_GETSTATS: {
2672                 struct ctl_stats *stats;
2673                 int i;
2674
2675                 stats = (struct ctl_stats *)addr;
2676
2677                 if ((sizeof(struct ctl_lun_io_stats) * softc->num_luns) >
2678                      stats->alloc_len) {
2679                         stats->status = CTL_SS_NEED_MORE_SPACE;
2680                         stats->num_luns = softc->num_luns;
2681                         break;
2682                 }
2683                 /*
2684                  * XXX KDM no locking here.  If the LUN list changes,
2685                  * things can blow up.
2686                  */
2687                 for (i = 0, lun = STAILQ_FIRST(&softc->lun_list); lun != NULL;
2688                      i++, lun = STAILQ_NEXT(lun, links)) {
2689                         retval = copyout(&lun->stats, &stats->lun_stats[i],
2690                                          sizeof(lun->stats));
2691                         if (retval != 0)
2692                                 break;
2693                 }
2694                 stats->num_luns = softc->num_luns;
2695                 stats->fill_len = sizeof(struct ctl_lun_io_stats) *
2696                                  softc->num_luns;
2697                 stats->status = CTL_SS_OK;
2698 #ifdef CTL_TIME_IO
2699                 stats->flags = CTL_STATS_FLAG_TIME_VALID;
2700 #else
2701                 stats->flags = CTL_STATS_FLAG_NONE;
2702 #endif
2703                 getnanouptime(&stats->timestamp);
2704                 break;
2705         }
2706         case CTL_ERROR_INJECT: {
2707                 struct ctl_error_desc *err_desc, *new_err_desc;
2708
2709                 err_desc = (struct ctl_error_desc *)addr;
2710
2711                 new_err_desc = malloc(sizeof(*new_err_desc), M_CTL,
2712                                       M_WAITOK | M_ZERO);
2713                 bcopy(err_desc, new_err_desc, sizeof(*new_err_desc));
2714
2715                 mtx_lock(&softc->ctl_lock);
2716                 lun = softc->ctl_luns[err_desc->lun_id];
2717                 if (lun == NULL) {
2718                         mtx_unlock(&softc->ctl_lock);
2719                         free(new_err_desc, M_CTL);
2720                         printf("%s: CTL_ERROR_INJECT: invalid LUN %ju\n",
2721                                __func__, (uintmax_t)err_desc->lun_id);
2722                         retval = EINVAL;
2723                         break;
2724                 }
2725                 mtx_lock(&lun->lun_lock);
2726                 mtx_unlock(&softc->ctl_lock);
2727
2728                 /*
2729                  * We could do some checking here to verify the validity
2730                  * of the request, but given the complexity of error
2731                  * injection requests, the checking logic would be fairly
2732                  * complex.
2733                  *
2734                  * For now, if the request is invalid, it just won't get
2735                  * executed and might get deleted.
2736                  */
2737                 STAILQ_INSERT_TAIL(&lun->error_list, new_err_desc, links);
2738
2739                 /*
2740                  * XXX KDM check to make sure the serial number is unique,
2741                  * in case we somehow manage to wrap.  That shouldn't
2742                  * happen for a very long time, but it's the right thing to
2743                  * do.
2744                  */
2745                 new_err_desc->serial = lun->error_serial;
2746                 err_desc->serial = lun->error_serial;
2747                 lun->error_serial++;
2748
2749                 mtx_unlock(&lun->lun_lock);
2750                 break;
2751         }
2752         case CTL_ERROR_INJECT_DELETE: {
2753                 struct ctl_error_desc *delete_desc, *desc, *desc2;
2754                 int delete_done;
2755
2756                 delete_desc = (struct ctl_error_desc *)addr;
2757                 delete_done = 0;
2758
2759                 mtx_lock(&softc->ctl_lock);
2760                 lun = softc->ctl_luns[delete_desc->lun_id];
2761                 if (lun == NULL) {
2762                         mtx_unlock(&softc->ctl_lock);
2763                         printf("%s: CTL_ERROR_INJECT_DELETE: invalid LUN %ju\n",
2764                                __func__, (uintmax_t)delete_desc->lun_id);
2765                         retval = EINVAL;
2766                         break;
2767                 }
2768                 mtx_lock(&lun->lun_lock);
2769                 mtx_unlock(&softc->ctl_lock);
2770                 STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
2771                         if (desc->serial != delete_desc->serial)
2772                                 continue;
2773
2774                         STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc,
2775                                       links);
2776                         free(desc, M_CTL);
2777                         delete_done = 1;
2778                 }
2779                 mtx_unlock(&lun->lun_lock);
2780                 if (delete_done == 0) {
2781                         printf("%s: CTL_ERROR_INJECT_DELETE: can't find "
2782                                "error serial %ju on LUN %u\n", __func__,
2783                                delete_desc->serial, delete_desc->lun_id);
2784                         retval = EINVAL;
2785                         break;
2786                 }
2787                 break;
2788         }
2789         case CTL_DUMP_STRUCTS: {
2790                 int i, j, k;
2791                 struct ctl_port *port;
2792                 struct ctl_frontend *fe;
2793
2794                 mtx_lock(&softc->ctl_lock);
2795                 printf("CTL Persistent Reservation information start:\n");
2796                 for (i = 0; i < CTL_MAX_LUNS; i++) {
2797                         lun = softc->ctl_luns[i];
2798
2799                         if ((lun == NULL)
2800                          || ((lun->flags & CTL_LUN_DISABLED) != 0))
2801                                 continue;
2802
2803                         for (j = 0; j < CTL_MAX_PORTS; j++) {
2804                                 if (lun->pr_keys[j] == NULL)
2805                                         continue;
2806                                 for (k = 0; k < CTL_MAX_INIT_PER_PORT; k++){
2807                                         if (lun->pr_keys[j][k] == 0)
2808                                                 continue;
2809                                         printf("  LUN %d port %d iid %d key "
2810                                                "%#jx\n", i, j, k,
2811                                                (uintmax_t)lun->pr_keys[j][k]);
2812                                 }
2813                         }
2814                 }
2815                 printf("CTL Persistent Reservation information end\n");
2816                 printf("CTL Ports:\n");
2817                 STAILQ_FOREACH(port, &softc->port_list, links) {
2818                         printf("  Port %d '%s' Frontend '%s' Type %u pp %d vp %d WWNN "
2819                                "%#jx WWPN %#jx\n", port->targ_port, port->port_name,
2820                                port->frontend->name, port->port_type,
2821                                port->physical_port, port->virtual_port,
2822                                (uintmax_t)port->wwnn, (uintmax_t)port->wwpn);
2823                         for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
2824                                 if (port->wwpn_iid[j].in_use == 0 &&
2825                                     port->wwpn_iid[j].wwpn == 0 &&
2826                                     port->wwpn_iid[j].name == NULL)
2827                                         continue;
2828
2829                                 printf("    iid %u use %d WWPN %#jx '%s'\n",
2830                                     j, port->wwpn_iid[j].in_use,
2831                                     (uintmax_t)port->wwpn_iid[j].wwpn,
2832                                     port->wwpn_iid[j].name);
2833                         }
2834                 }
2835                 printf("CTL Port information end\n");
2836                 mtx_unlock(&softc->ctl_lock);
2837                 /*
2838                  * XXX KDM calling this without a lock.  We'd likely want
2839                  * to drop the lock before calling the frontend's dump
2840                  * routine anyway.
2841                  */
2842                 printf("CTL Frontends:\n");
2843                 STAILQ_FOREACH(fe, &softc->fe_list, links) {
2844                         printf("  Frontend '%s'\n", fe->name);
2845                         if (fe->fe_dump != NULL)
2846                                 fe->fe_dump();
2847                 }
2848                 printf("CTL Frontend information end\n");
2849                 break;
2850         }
2851         case CTL_LUN_REQ: {
2852                 struct ctl_lun_req *lun_req;
2853                 struct ctl_backend_driver *backend;
2854
2855                 lun_req = (struct ctl_lun_req *)addr;
2856
2857                 backend = ctl_backend_find(lun_req->backend);
2858                 if (backend == NULL) {
2859                         lun_req->status = CTL_LUN_ERROR;
2860                         snprintf(lun_req->error_str,
2861                                  sizeof(lun_req->error_str),
2862                                  "Backend \"%s\" not found.",
2863                                  lun_req->backend);
2864                         break;
2865                 }
2866                 if (lun_req->num_be_args > 0) {
2867                         lun_req->kern_be_args = ctl_copyin_args(
2868                                 lun_req->num_be_args,
2869                                 lun_req->be_args,
2870                                 lun_req->error_str,
2871                                 sizeof(lun_req->error_str));
2872                         if (lun_req->kern_be_args == NULL) {
2873                                 lun_req->status = CTL_LUN_ERROR;
2874                                 break;
2875                         }
2876                 }
2877
2878                 retval = backend->ioctl(dev, cmd, addr, flag, td);
2879
2880                 if (lun_req->num_be_args > 0) {
2881                         ctl_copyout_args(lun_req->num_be_args,
2882                                       lun_req->kern_be_args);
2883                         ctl_free_args(lun_req->num_be_args,
2884                                       lun_req->kern_be_args);
2885                 }
2886                 break;
2887         }
2888         case CTL_LUN_LIST: {
2889                 struct sbuf *sb;
2890                 struct ctl_lun_list *list;
2891                 struct ctl_option *opt;
2892
2893                 list = (struct ctl_lun_list *)addr;
2894
2895                 /*
2896                  * Allocate a fixed length sbuf here, based on the length
2897                  * of the user's buffer.  We could allocate an auto-extending
2898                  * buffer, and then tell the user how much larger our
2899                  * amount of data is than his buffer, but that presents
2900                  * some problems:
2901                  *
2902                  * 1.  The sbuf(9) routines use a blocking malloc, and so
2903                  *     we can't hold a lock while calling them with an
2904                  *     auto-extending buffer.
2905                  *
2906                  * 2.  There is not currently a LUN reference counting
2907                  *     mechanism, outside of outstanding transactions on
2908                  *     the LUN's OOA queue.  So a LUN could go away on us
2909                  *     while we're getting the LUN number, backend-specific
2910                  *     information, etc.  Thus, given the way things
2911                  *     currently work, we need to hold the CTL lock while
2912                  *     grabbing LUN information.
2913                  *
2914                  * So, from the user's standpoint, the best thing to do is
2915                  * allocate what he thinks is a reasonable buffer length,
2916                  * and then if he gets a CTL_LUN_LIST_NEED_MORE_SPACE error,
2917                  * double the buffer length and try again.  (And repeat
2918                  * that until he succeeds.)
2919                  */
2920                 sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
2921                 if (sb == NULL) {
2922                         list->status = CTL_LUN_LIST_ERROR;
2923                         snprintf(list->error_str, sizeof(list->error_str),
2924                                  "Unable to allocate %d bytes for LUN list",
2925                                  list->alloc_len);
2926                         break;
2927                 }
2928
2929                 sbuf_printf(sb, "<ctllunlist>\n");
2930
2931                 mtx_lock(&softc->ctl_lock);
2932                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
2933                         mtx_lock(&lun->lun_lock);
2934                         retval = sbuf_printf(sb, "<lun id=\"%ju\">\n",
2935                                              (uintmax_t)lun->lun);
2936
2937                         /*
2938                          * Bail out as soon as we see that we've overfilled
2939                          * the buffer.
2940                          */
2941                         if (retval != 0)
2942                                 break;
2943
2944                         retval = sbuf_printf(sb, "\t<backend_type>%s"
2945                                              "</backend_type>\n",
2946                                              (lun->backend == NULL) ?  "none" :
2947                                              lun->backend->name);
2948
2949                         if (retval != 0)
2950                                 break;
2951
2952                         retval = sbuf_printf(sb, "\t<lun_type>%d</lun_type>\n",
2953                                              lun->be_lun->lun_type);
2954
2955                         if (retval != 0)
2956                                 break;
2957
2958                         if (lun->backend == NULL) {
2959                                 retval = sbuf_printf(sb, "</lun>\n");
2960                                 if (retval != 0)
2961                                         break;
2962                                 continue;
2963                         }
2964
2965                         retval = sbuf_printf(sb, "\t<size>%ju</size>\n",
2966                                              (lun->be_lun->maxlba > 0) ?
2967                                              lun->be_lun->maxlba + 1 : 0);
2968
2969                         if (retval != 0)
2970                                 break;
2971
2972                         retval = sbuf_printf(sb, "\t<blocksize>%u</blocksize>\n",
2973                                              lun->be_lun->blocksize);
2974
2975                         if (retval != 0)
2976                                 break;
2977
2978                         retval = sbuf_printf(sb, "\t<serial_number>");
2979
2980                         if (retval != 0)
2981                                 break;
2982
2983                         retval = ctl_sbuf_printf_esc(sb,
2984                             lun->be_lun->serial_num,
2985                             sizeof(lun->be_lun->serial_num));
2986
2987                         if (retval != 0)
2988                                 break;
2989
2990                         retval = sbuf_printf(sb, "</serial_number>\n");
2991
2992                         if (retval != 0)
2993                                 break;
2994
2995                         retval = sbuf_printf(sb, "\t<device_id>");
2996
2997                         if (retval != 0)
2998                                 break;
2999
3000                         retval = ctl_sbuf_printf_esc(sb,
3001                             lun->be_lun->device_id,
3002                             sizeof(lun->be_lun->device_id));
3003
3004                         if (retval != 0)
3005                                 break;
3006
3007                         retval = sbuf_printf(sb, "</device_id>\n");
3008
3009                         if (retval != 0)
3010                                 break;
3011
3012                         if (lun->backend->lun_info != NULL) {
3013                                 retval = lun->backend->lun_info(lun->be_lun->be_lun, sb);
3014                                 if (retval != 0)
3015                                         break;
3016                         }
3017                         STAILQ_FOREACH(opt, &lun->be_lun->options, links) {
3018                                 retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3019                                     opt->name, opt->value, opt->name);
3020                                 if (retval != 0)
3021                                         break;
3022                         }
3023
3024                         retval = sbuf_printf(sb, "</lun>\n");
3025
3026                         if (retval != 0)
3027                                 break;
3028                         mtx_unlock(&lun->lun_lock);
3029                 }
3030                 if (lun != NULL)
3031                         mtx_unlock(&lun->lun_lock);
3032                 mtx_unlock(&softc->ctl_lock);
3033
3034                 if ((retval != 0)
3035                  || ((retval = sbuf_printf(sb, "</ctllunlist>\n")) != 0)) {
3036                         retval = 0;
3037                         sbuf_delete(sb);
3038                         list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3039                         snprintf(list->error_str, sizeof(list->error_str),
3040                                  "Out of space, %d bytes is too small",
3041                                  list->alloc_len);
3042                         break;
3043                 }
3044
3045                 sbuf_finish(sb);
3046
3047                 retval = copyout(sbuf_data(sb), list->lun_xml,
3048                                  sbuf_len(sb) + 1);
3049
3050                 list->fill_len = sbuf_len(sb) + 1;
3051                 list->status = CTL_LUN_LIST_OK;
3052                 sbuf_delete(sb);
3053                 break;
3054         }
3055         case CTL_ISCSI: {
3056                 struct ctl_iscsi *ci;
3057                 struct ctl_frontend *fe;
3058
3059                 ci = (struct ctl_iscsi *)addr;
3060
3061                 fe = ctl_frontend_find("iscsi");
3062                 if (fe == NULL) {
3063                         ci->status = CTL_ISCSI_ERROR;
3064                         snprintf(ci->error_str, sizeof(ci->error_str),
3065                             "Frontend \"iscsi\" not found.");
3066                         break;
3067                 }
3068
3069                 retval = fe->ioctl(dev, cmd, addr, flag, td);
3070                 break;
3071         }
3072         case CTL_PORT_REQ: {
3073                 struct ctl_req *req;
3074                 struct ctl_frontend *fe;
3075
3076                 req = (struct ctl_req *)addr;
3077
3078                 fe = ctl_frontend_find(req->driver);
3079                 if (fe == NULL) {
3080                         req->status = CTL_LUN_ERROR;
3081                         snprintf(req->error_str, sizeof(req->error_str),
3082                             "Frontend \"%s\" not found.", req->driver);
3083                         break;
3084                 }
3085                 if (req->num_args > 0) {
3086                         req->kern_args = ctl_copyin_args(req->num_args,
3087                             req->args, req->error_str, sizeof(req->error_str));
3088                         if (req->kern_args == NULL) {
3089                                 req->status = CTL_LUN_ERROR;
3090                                 break;
3091                         }
3092                 }
3093
3094                 if (fe->ioctl)
3095                         retval = fe->ioctl(dev, cmd, addr, flag, td);
3096                 else
3097                         retval = ENODEV;
3098
3099                 if (req->num_args > 0) {
3100                         ctl_copyout_args(req->num_args, req->kern_args);
3101                         ctl_free_args(req->num_args, req->kern_args);
3102                 }
3103                 break;
3104         }
3105         case CTL_PORT_LIST: {
3106                 struct sbuf *sb;
3107                 struct ctl_port *port;
3108                 struct ctl_lun_list *list;
3109                 struct ctl_option *opt;
3110                 int j;
3111                 uint32_t plun;
3112
3113                 list = (struct ctl_lun_list *)addr;
3114
3115                 sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3116                 if (sb == NULL) {
3117                         list->status = CTL_LUN_LIST_ERROR;
3118                         snprintf(list->error_str, sizeof(list->error_str),
3119                                  "Unable to allocate %d bytes for LUN list",
3120                                  list->alloc_len);
3121                         break;
3122                 }
3123
3124                 sbuf_printf(sb, "<ctlportlist>\n");
3125
3126                 mtx_lock(&softc->ctl_lock);
3127                 STAILQ_FOREACH(port, &softc->port_list, links) {
3128                         retval = sbuf_printf(sb, "<targ_port id=\"%ju\">\n",
3129                                              (uintmax_t)port->targ_port);
3130
3131                         /*
3132                          * Bail out as soon as we see that we've overfilled
3133                          * the buffer.
3134                          */
3135                         if (retval != 0)
3136                                 break;
3137
3138                         retval = sbuf_printf(sb, "\t<frontend_type>%s"
3139                             "</frontend_type>\n", port->frontend->name);
3140                         if (retval != 0)
3141                                 break;
3142
3143                         retval = sbuf_printf(sb, "\t<port_type>%d</port_type>\n",
3144                                              port->port_type);
3145                         if (retval != 0)
3146                                 break;
3147
3148                         retval = sbuf_printf(sb, "\t<online>%s</online>\n",
3149                             (port->status & CTL_PORT_STATUS_ONLINE) ? "YES" : "NO");
3150                         if (retval != 0)
3151                                 break;
3152
3153                         retval = sbuf_printf(sb, "\t<port_name>%s</port_name>\n",
3154                             port->port_name);
3155                         if (retval != 0)
3156                                 break;
3157
3158                         retval = sbuf_printf(sb, "\t<physical_port>%d</physical_port>\n",
3159                             port->physical_port);
3160                         if (retval != 0)
3161                                 break;
3162
3163                         retval = sbuf_printf(sb, "\t<virtual_port>%d</virtual_port>\n",
3164                             port->virtual_port);
3165                         if (retval != 0)
3166                                 break;
3167
3168                         if (port->target_devid != NULL) {
3169                                 sbuf_printf(sb, "\t<target>");
3170                                 ctl_id_sbuf(port->target_devid, sb);
3171                                 sbuf_printf(sb, "</target>\n");
3172                         }
3173
3174                         if (port->port_devid != NULL) {
3175                                 sbuf_printf(sb, "\t<port>");
3176                                 ctl_id_sbuf(port->port_devid, sb);
3177                                 sbuf_printf(sb, "</port>\n");
3178                         }
3179
3180                         if (port->port_info != NULL) {
3181                                 retval = port->port_info(port->onoff_arg, sb);
3182                                 if (retval != 0)
3183                                         break;
3184                         }
3185                         STAILQ_FOREACH(opt, &port->options, links) {
3186                                 retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3187                                     opt->name, opt->value, opt->name);
3188                                 if (retval != 0)
3189                                         break;
3190                         }
3191
3192                         if (port->lun_map != NULL) {
3193                                 sbuf_printf(sb, "\t<lun_map>on</lun_map>\n");
3194                                 for (j = 0; j < CTL_MAX_LUNS; j++) {
3195                                         plun = ctl_lun_map_from_port(port, j);
3196                                         if (plun >= CTL_MAX_LUNS)
3197                                                 continue;
3198                                         sbuf_printf(sb,
3199                                             "\t<lun id=\"%u\">%u</lun>\n",
3200                                             j, plun);
3201                                 }
3202                         }
3203
3204                         for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
3205                                 if (port->wwpn_iid[j].in_use == 0 ||
3206                                     (port->wwpn_iid[j].wwpn == 0 &&
3207                                      port->wwpn_iid[j].name == NULL))
3208                                         continue;
3209
3210                                 if (port->wwpn_iid[j].name != NULL)
3211                                         retval = sbuf_printf(sb,
3212                                             "\t<initiator id=\"%u\">%s</initiator>\n",
3213                                             j, port->wwpn_iid[j].name);
3214                                 else
3215                                         retval = sbuf_printf(sb,
3216                                             "\t<initiator id=\"%u\">naa.%08jx</initiator>\n",
3217                                             j, port->wwpn_iid[j].wwpn);
3218                                 if (retval != 0)
3219                                         break;
3220                         }
3221                         if (retval != 0)
3222                                 break;
3223
3224                         retval = sbuf_printf(sb, "</targ_port>\n");
3225                         if (retval != 0)
3226                                 break;
3227                 }
3228                 mtx_unlock(&softc->ctl_lock);
3229
3230                 if ((retval != 0)
3231                  || ((retval = sbuf_printf(sb, "</ctlportlist>\n")) != 0)) {
3232                         retval = 0;
3233                         sbuf_delete(sb);
3234                         list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3235                         snprintf(list->error_str, sizeof(list->error_str),
3236                                  "Out of space, %d bytes is too small",
3237                                  list->alloc_len);
3238                         break;
3239                 }
3240
3241                 sbuf_finish(sb);
3242
3243                 retval = copyout(sbuf_data(sb), list->lun_xml,
3244                                  sbuf_len(sb) + 1);
3245
3246                 list->fill_len = sbuf_len(sb) + 1;
3247                 list->status = CTL_LUN_LIST_OK;
3248                 sbuf_delete(sb);
3249                 break;
3250         }
3251         case CTL_LUN_MAP: {
3252                 struct ctl_lun_map *lm  = (struct ctl_lun_map *)addr;
3253                 struct ctl_port *port;
3254
3255                 mtx_lock(&softc->ctl_lock);
3256                 if (lm->port < softc->port_min ||
3257                     lm->port >= softc->port_max ||
3258                     (port = softc->ctl_ports[lm->port]) == NULL) {
3259                         mtx_unlock(&softc->ctl_lock);
3260                         return (ENXIO);
3261                 }
3262                 if (port->status & CTL_PORT_STATUS_ONLINE) {
3263                         STAILQ_FOREACH(lun, &softc->lun_list, links) {
3264                                 if (ctl_lun_map_to_port(port, lun->lun) >=
3265                                     CTL_MAX_LUNS)
3266                                         continue;
3267                                 mtx_lock(&lun->lun_lock);
3268                                 ctl_est_ua_port(lun, lm->port, -1,
3269                                     CTL_UA_LUN_CHANGE);
3270                                 mtx_unlock(&lun->lun_lock);
3271                         }
3272                 }
3273                 mtx_unlock(&softc->ctl_lock); // XXX: port_enable sleeps
3274                 if (lm->plun < CTL_MAX_LUNS) {
3275                         if (lm->lun == UINT32_MAX)
3276                                 retval = ctl_lun_map_unset(port, lm->plun);
3277                         else if (lm->lun < CTL_MAX_LUNS &&
3278                             softc->ctl_luns[lm->lun] != NULL)
3279                                 retval = ctl_lun_map_set(port, lm->plun, lm->lun);
3280                         else
3281                                 return (ENXIO);
3282                 } else if (lm->plun == UINT32_MAX) {
3283                         if (lm->lun == UINT32_MAX)
3284                                 retval = ctl_lun_map_deinit(port);
3285                         else
3286                                 retval = ctl_lun_map_init(port);
3287                 } else
3288                         return (ENXIO);
3289                 if (port->status & CTL_PORT_STATUS_ONLINE)
3290                         ctl_isc_announce_port(port);
3291                 break;
3292         }
3293         default: {
3294                 /* XXX KDM should we fix this? */
3295 #if 0
3296                 struct ctl_backend_driver *backend;
3297                 unsigned int type;
3298                 int found;
3299
3300                 found = 0;
3301
3302                 /*
3303                  * We encode the backend type as the ioctl type for backend
3304                  * ioctls.  So parse it out here, and then search for a
3305                  * backend of this type.
3306                  */
3307                 type = _IOC_TYPE(cmd);
3308
3309                 STAILQ_FOREACH(backend, &softc->be_list, links) {
3310                         if (backend->type == type) {
3311                                 found = 1;
3312                                 break;
3313                         }
3314                 }
3315                 if (found == 0) {
3316                         printf("ctl: unknown ioctl command %#lx or backend "
3317                                "%d\n", cmd, type);
3318                         retval = EINVAL;
3319                         break;
3320                 }
3321                 retval = backend->ioctl(dev, cmd, addr, flag, td);
3322 #endif
3323                 retval = ENOTTY;
3324                 break;
3325         }
3326         }
3327         return (retval);
3328 }
3329
3330 uint32_t
3331 ctl_get_initindex(struct ctl_nexus *nexus)
3332 {
3333         return (nexus->initid + (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3334 }
3335
3336 int
3337 ctl_lun_map_init(struct ctl_port *port)
3338 {
3339         struct ctl_softc *softc = control_softc;
3340         struct ctl_lun *lun;
3341         uint32_t i;
3342
3343         if (port->lun_map == NULL)
3344                 port->lun_map = malloc(sizeof(uint32_t) * CTL_MAX_LUNS,
3345                     M_CTL, M_NOWAIT);
3346         if (port->lun_map == NULL)
3347                 return (ENOMEM);
3348         for (i = 0; i < CTL_MAX_LUNS; i++)
3349                 port->lun_map[i] = UINT32_MAX;
3350         if (port->status & CTL_PORT_STATUS_ONLINE) {
3351                 if (port->lun_disable != NULL) {
3352                         STAILQ_FOREACH(lun, &softc->lun_list, links)
3353                                 port->lun_disable(port->targ_lun_arg, lun->lun);
3354                 }
3355                 ctl_isc_announce_port(port);
3356         }
3357         return (0);
3358 }
3359
3360 int
3361 ctl_lun_map_deinit(struct ctl_port *port)
3362 {
3363         struct ctl_softc *softc = control_softc;
3364         struct ctl_lun *lun;
3365
3366         if (port->lun_map == NULL)
3367                 return (0);
3368         free(port->lun_map, M_CTL);
3369         port->lun_map = NULL;
3370         if (port->status & CTL_PORT_STATUS_ONLINE) {
3371                 if (port->lun_enable != NULL) {
3372                         STAILQ_FOREACH(lun, &softc->lun_list, links)
3373                                 port->lun_enable(port->targ_lun_arg, lun->lun);
3374                 }
3375                 ctl_isc_announce_port(port);
3376         }
3377         return (0);
3378 }
3379
3380 int
3381 ctl_lun_map_set(struct ctl_port *port, uint32_t plun, uint32_t glun)
3382 {
3383         int status;
3384         uint32_t old;
3385
3386         if (port->lun_map == NULL) {
3387                 status = ctl_lun_map_init(port);
3388                 if (status != 0)
3389                         return (status);
3390         }
3391         old = port->lun_map[plun];
3392         port->lun_map[plun] = glun;
3393         if ((port->status & CTL_PORT_STATUS_ONLINE) && old >= CTL_MAX_LUNS) {
3394                 if (port->lun_enable != NULL)
3395                         port->lun_enable(port->targ_lun_arg, plun);
3396                 ctl_isc_announce_port(port);
3397         }
3398         return (0);
3399 }
3400
3401 int
3402 ctl_lun_map_unset(struct ctl_port *port, uint32_t plun)
3403 {
3404         uint32_t old;
3405
3406         if (port->lun_map == NULL)
3407                 return (0);
3408         old = port->lun_map[plun];
3409         port->lun_map[plun] = UINT32_MAX;
3410         if ((port->status & CTL_PORT_STATUS_ONLINE) && old < CTL_MAX_LUNS) {
3411                 if (port->lun_disable != NULL)
3412                         port->lun_disable(port->targ_lun_arg, plun);
3413                 ctl_isc_announce_port(port);
3414         }
3415         return (0);
3416 }
3417
3418 uint32_t
3419 ctl_lun_map_from_port(struct ctl_port *port, uint32_t lun_id)
3420 {
3421
3422         if (port == NULL)
3423                 return (UINT32_MAX);
3424         if (port->lun_map == NULL || lun_id >= CTL_MAX_LUNS)
3425                 return (lun_id);
3426         return (port->lun_map[lun_id]);
3427 }
3428
3429 uint32_t
3430 ctl_lun_map_to_port(struct ctl_port *port, uint32_t lun_id)
3431 {
3432         uint32_t i;
3433
3434         if (port == NULL)
3435                 return (UINT32_MAX);
3436         if (port->lun_map == NULL)
3437                 return (lun_id);
3438         for (i = 0; i < CTL_MAX_LUNS; i++) {
3439                 if (port->lun_map[i] == lun_id)
3440                         return (i);
3441         }
3442         return (UINT32_MAX);
3443 }
3444
3445 static struct ctl_port *
3446 ctl_io_port(struct ctl_io_hdr *io_hdr)
3447 {
3448
3449         return (control_softc->ctl_ports[io_hdr->nexus.targ_port]);
3450 }
3451
3452 int
3453 ctl_ffz(uint32_t *mask, uint32_t first, uint32_t last)
3454 {
3455         int i;
3456
3457         for (i = first; i < last; i++) {
3458                 if ((mask[i / 32] & (1 << (i % 32))) == 0)
3459                         return (i);
3460         }
3461         return (-1);
3462 }
3463
3464 int
3465 ctl_set_mask(uint32_t *mask, uint32_t bit)
3466 {
3467         uint32_t chunk, piece;
3468
3469         chunk = bit >> 5;
3470         piece = bit % (sizeof(uint32_t) * 8);
3471
3472         if ((mask[chunk] & (1 << piece)) != 0)
3473                 return (-1);
3474         else
3475                 mask[chunk] |= (1 << piece);
3476
3477         return (0);
3478 }
3479
3480 int
3481 ctl_clear_mask(uint32_t *mask, uint32_t bit)
3482 {
3483         uint32_t chunk, piece;
3484
3485         chunk = bit >> 5;
3486         piece = bit % (sizeof(uint32_t) * 8);
3487
3488         if ((mask[chunk] & (1 << piece)) == 0)
3489                 return (-1);
3490         else
3491                 mask[chunk] &= ~(1 << piece);
3492
3493         return (0);
3494 }
3495
3496 int
3497 ctl_is_set(uint32_t *mask, uint32_t bit)
3498 {
3499         uint32_t chunk, piece;
3500
3501         chunk = bit >> 5;
3502         piece = bit % (sizeof(uint32_t) * 8);
3503
3504         if ((mask[chunk] & (1 << piece)) == 0)
3505                 return (0);
3506         else
3507                 return (1);
3508 }
3509
3510 static uint64_t
3511 ctl_get_prkey(struct ctl_lun *lun, uint32_t residx)
3512 {
3513         uint64_t *t;
3514
3515         t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3516         if (t == NULL)
3517                 return (0);
3518         return (t[residx % CTL_MAX_INIT_PER_PORT]);
3519 }
3520
3521 static void
3522 ctl_clr_prkey(struct ctl_lun *lun, uint32_t residx)
3523 {
3524         uint64_t *t;
3525
3526         t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3527         if (t == NULL)
3528                 return;
3529         t[residx % CTL_MAX_INIT_PER_PORT] = 0;
3530 }
3531
3532 static void
3533 ctl_alloc_prkey(struct ctl_lun *lun, uint32_t residx)
3534 {
3535         uint64_t *p;
3536         u_int i;
3537
3538         i = residx/CTL_MAX_INIT_PER_PORT;
3539         if (lun->pr_keys[i] != NULL)
3540                 return;
3541         mtx_unlock(&lun->lun_lock);
3542         p = malloc(sizeof(uint64_t) * CTL_MAX_INIT_PER_PORT, M_CTL,
3543             M_WAITOK | M_ZERO);
3544         mtx_lock(&lun->lun_lock);
3545         if (lun->pr_keys[i] == NULL)
3546                 lun->pr_keys[i] = p;
3547         else
3548                 free(p, M_CTL);
3549 }
3550
3551 static void
3552 ctl_set_prkey(struct ctl_lun *lun, uint32_t residx, uint64_t key)
3553 {
3554         uint64_t *t;
3555
3556         t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3557         KASSERT(t != NULL, ("prkey %d is not allocated", residx));
3558         t[residx % CTL_MAX_INIT_PER_PORT] = key;
3559 }
3560
3561 /*
3562  * ctl_softc, pool_name, total_ctl_io are passed in.
3563  * npool is passed out.
3564  */
3565 int
3566 ctl_pool_create(struct ctl_softc *ctl_softc, const char *pool_name,
3567                 uint32_t total_ctl_io, void **npool)
3568 {
3569 #ifdef IO_POOLS
3570         struct ctl_io_pool *pool;
3571
3572         pool = (struct ctl_io_pool *)malloc(sizeof(*pool), M_CTL,
3573                                             M_NOWAIT | M_ZERO);
3574         if (pool == NULL)
3575                 return (ENOMEM);
3576
3577         snprintf(pool->name, sizeof(pool->name), "CTL IO %s", pool_name);
3578         pool->ctl_softc = ctl_softc;
3579         pool->zone = uma_zsecond_create(pool->name, NULL,
3580             NULL, NULL, NULL, ctl_softc->io_zone);
3581         /* uma_prealloc(pool->zone, total_ctl_io); */
3582
3583         *npool = pool;
3584 #else
3585         *npool = ctl_softc->io_zone;
3586 #endif
3587         return (0);
3588 }
3589
3590 void
3591 ctl_pool_free(struct ctl_io_pool *pool)
3592 {
3593
3594         if (pool == NULL)
3595                 return;
3596
3597 #ifdef IO_POOLS
3598         uma_zdestroy(pool->zone);
3599         free(pool, M_CTL);
3600 #endif
3601 }
3602
3603 union ctl_io *
3604 ctl_alloc_io(void *pool_ref)
3605 {
3606         union ctl_io *io;
3607 #ifdef IO_POOLS
3608         struct ctl_io_pool *pool = (struct ctl_io_pool *)pool_ref;
3609
3610         io = uma_zalloc(pool->zone, M_WAITOK);
3611 #else
3612         io = uma_zalloc((uma_zone_t)pool_ref, M_WAITOK);
3613 #endif
3614         if (io != NULL)
3615                 io->io_hdr.pool = pool_ref;
3616         return (io);
3617 }
3618
3619 union ctl_io *
3620 ctl_alloc_io_nowait(void *pool_ref)
3621 {
3622         union ctl_io *io;
3623 #ifdef IO_POOLS
3624         struct ctl_io_pool *pool = (struct ctl_io_pool *)pool_ref;
3625
3626         io = uma_zalloc(pool->zone, M_NOWAIT);
3627 #else
3628         io = uma_zalloc((uma_zone_t)pool_ref, M_NOWAIT);
3629 #endif
3630         if (io != NULL)
3631                 io->io_hdr.pool = pool_ref;
3632         return (io);
3633 }
3634
3635 void
3636 ctl_free_io(union ctl_io *io)
3637 {
3638 #ifdef IO_POOLS
3639         struct ctl_io_pool *pool;
3640 #endif
3641
3642         if (io == NULL)
3643                 return;
3644
3645 #ifdef IO_POOLS
3646         pool = (struct ctl_io_pool *)io->io_hdr.pool;
3647         uma_zfree(pool->zone, io);
3648 #else
3649         uma_zfree((uma_zone_t)io->io_hdr.pool, io);
3650 #endif
3651 }
3652
3653 void
3654 ctl_zero_io(union ctl_io *io)
3655 {
3656         void *pool_ref;
3657
3658         if (io == NULL)
3659                 return;
3660
3661         /*
3662          * May need to preserve linked list pointers at some point too.
3663          */
3664         pool_ref = io->io_hdr.pool;
3665         memset(io, 0, sizeof(*io));
3666         io->io_hdr.pool = pool_ref;
3667 }
3668
3669 /*
3670  * This routine is currently used for internal copies of ctl_ios that need
3671  * to persist for some reason after we've already returned status to the
3672  * FETD.  (Thus the flag set.)
3673  *
3674  * XXX XXX
3675  * Note that this makes a blind copy of all fields in the ctl_io, except
3676  * for the pool reference.  This includes any memory that has been
3677  * allocated!  That memory will no longer be valid after done has been
3678  * called, so this would be VERY DANGEROUS for command that actually does
3679  * any reads or writes.  Right now (11/7/2005), this is only used for immediate
3680  * start and stop commands, which don't transfer any data, so this is not a
3681  * problem.  If it is used for anything else, the caller would also need to
3682  * allocate data buffer space and this routine would need to be modified to
3683  * copy the data buffer(s) as well.
3684  */
3685 void
3686 ctl_copy_io(union ctl_io *src, union ctl_io *dest)
3687 {
3688         void *pool_ref;
3689
3690         if ((src == NULL)
3691          || (dest == NULL))
3692                 return;
3693
3694         /*
3695          * May need to preserve linked list pointers at some point too.
3696          */
3697         pool_ref = dest->io_hdr.pool;
3698
3699         memcpy(dest, src, MIN(sizeof(*src), sizeof(*dest)));
3700
3701         dest->io_hdr.pool = pool_ref;
3702         /*
3703          * We need to know that this is an internal copy, and doesn't need
3704          * to get passed back to the FETD that allocated it.
3705          */
3706         dest->io_hdr.flags |= CTL_FLAG_INT_COPY;
3707 }
3708
3709 int
3710 ctl_expand_number(const char *buf, uint64_t *num)
3711 {
3712         char *endptr;
3713         uint64_t number;
3714         unsigned shift;
3715
3716         number = strtoq(buf, &endptr, 0);
3717
3718         switch (tolower((unsigned char)*endptr)) {
3719         case 'e':
3720                 shift = 60;
3721                 break;
3722         case 'p':
3723                 shift = 50;
3724                 break;
3725         case 't':
3726                 shift = 40;
3727                 break;
3728         case 'g':
3729                 shift = 30;
3730                 break;
3731         case 'm':
3732                 shift = 20;
3733                 break;
3734         case 'k':
3735                 shift = 10;
3736                 break;
3737         case 'b':
3738         case '\0': /* No unit. */
3739                 *num = number;
3740                 return (0);
3741         default:
3742                 /* Unrecognized unit. */
3743                 return (-1);
3744         }
3745
3746         if ((number << shift) >> shift != number) {
3747                 /* Overflow */
3748                 return (-1);
3749         }
3750         *num = number << shift;
3751         return (0);
3752 }
3753
3754
3755 /*
3756  * This routine could be used in the future to load default and/or saved
3757  * mode page parameters for a particuar lun.
3758  */
3759 static int
3760 ctl_init_page_index(struct ctl_lun *lun)
3761 {
3762         int i;
3763         struct ctl_page_index *page_index;
3764         const char *value;
3765         uint64_t ival;
3766
3767         memcpy(&lun->mode_pages.index, page_index_template,
3768                sizeof(page_index_template));
3769
3770         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
3771
3772                 page_index = &lun->mode_pages.index[i];
3773                 /*
3774                  * If this is a disk-only mode page, there's no point in
3775                  * setting it up.  For some pages, we have to have some
3776                  * basic information about the disk in order to calculate the
3777                  * mode page data.
3778                  */
3779                 if ((lun->be_lun->lun_type != T_DIRECT)
3780                  && (page_index->page_flags & CTL_PAGE_FLAG_DISK_ONLY))
3781                         continue;
3782
3783                 switch (page_index->page_code & SMPH_PC_MASK) {
3784                 case SMS_RW_ERROR_RECOVERY_PAGE: {
3785                         if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3786                                 panic("subpage is incorrect!");
3787                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_CURRENT],
3788                                &rw_er_page_default,
3789                                sizeof(rw_er_page_default));
3790                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_CHANGEABLE],
3791                                &rw_er_page_changeable,
3792                                sizeof(rw_er_page_changeable));
3793                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_DEFAULT],
3794                                &rw_er_page_default,
3795                                sizeof(rw_er_page_default));
3796                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_SAVED],
3797                                &rw_er_page_default,
3798                                sizeof(rw_er_page_default));
3799                         page_index->page_data =
3800                                 (uint8_t *)lun->mode_pages.rw_er_page;
3801                         break;
3802                 }
3803                 case SMS_FORMAT_DEVICE_PAGE: {
3804                         struct scsi_format_page *format_page;
3805
3806                         if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3807                                 panic("subpage is incorrect!");
3808
3809                         /*
3810                          * Sectors per track are set above.  Bytes per
3811                          * sector need to be set here on a per-LUN basis.
3812                          */
3813                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_CURRENT],
3814                                &format_page_default,
3815                                sizeof(format_page_default));
3816                         memcpy(&lun->mode_pages.format_page[
3817                                CTL_PAGE_CHANGEABLE], &format_page_changeable,
3818                                sizeof(format_page_changeable));
3819                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_DEFAULT],
3820                                &format_page_default,
3821                                sizeof(format_page_default));
3822                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_SAVED],
3823                                &format_page_default,
3824                                sizeof(format_page_default));
3825
3826                         format_page = &lun->mode_pages.format_page[
3827                                 CTL_PAGE_CURRENT];
3828                         scsi_ulto2b(lun->be_lun->blocksize,
3829                                     format_page->bytes_per_sector);
3830
3831                         format_page = &lun->mode_pages.format_page[
3832                                 CTL_PAGE_DEFAULT];
3833                         scsi_ulto2b(lun->be_lun->blocksize,
3834                                     format_page->bytes_per_sector);
3835
3836                         format_page = &lun->mode_pages.format_page[
3837                                 CTL_PAGE_SAVED];
3838                         scsi_ulto2b(lun->be_lun->blocksize,
3839                                     format_page->bytes_per_sector);
3840
3841                         page_index->page_data =
3842                                 (uint8_t *)lun->mode_pages.format_page;
3843                         break;
3844                 }
3845                 case SMS_RIGID_DISK_PAGE: {
3846                         struct scsi_rigid_disk_page *rigid_disk_page;
3847                         uint32_t sectors_per_cylinder;
3848                         uint64_t cylinders;
3849 #ifndef __XSCALE__
3850                         int shift;
3851 #endif /* !__XSCALE__ */
3852
3853                         if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3854                                 panic("invalid subpage value %d",
3855                                       page_index->subpage);
3856
3857                         /*
3858                          * Rotation rate and sectors per track are set
3859                          * above.  We calculate the cylinders here based on
3860                          * capacity.  Due to the number of heads and
3861                          * sectors per track we're using, smaller arrays
3862                          * may turn out to have 0 cylinders.  Linux and
3863                          * FreeBSD don't pay attention to these mode pages
3864                          * to figure out capacity, but Solaris does.  It
3865                          * seems to deal with 0 cylinders just fine, and
3866                          * works out a fake geometry based on the capacity.
3867                          */
3868                         memcpy(&lun->mode_pages.rigid_disk_page[
3869                                CTL_PAGE_DEFAULT], &rigid_disk_page_default,
3870                                sizeof(rigid_disk_page_default));
3871                         memcpy(&lun->mode_pages.rigid_disk_page[
3872                                CTL_PAGE_CHANGEABLE],&rigid_disk_page_changeable,
3873                                sizeof(rigid_disk_page_changeable));
3874
3875                         sectors_per_cylinder = CTL_DEFAULT_SECTORS_PER_TRACK *
3876                                 CTL_DEFAULT_HEADS;
3877
3878                         /*
3879                          * The divide method here will be more accurate,
3880                          * probably, but results in floating point being
3881                          * used in the kernel on i386 (__udivdi3()).  On the
3882                          * XScale, though, __udivdi3() is implemented in
3883                          * software.
3884                          *
3885                          * The shift method for cylinder calculation is
3886                          * accurate if sectors_per_cylinder is a power of
3887                          * 2.  Otherwise it might be slightly off -- you
3888                          * might have a bit of a truncation problem.
3889                          */
3890 #ifdef  __XSCALE__
3891                         cylinders = (lun->be_lun->maxlba + 1) /
3892                                 sectors_per_cylinder;
3893 #else
3894                         for (shift = 31; shift > 0; shift--) {
3895                                 if (sectors_per_cylinder & (1 << shift))
3896                                         break;
3897                         }
3898                         cylinders = (lun->be_lun->maxlba + 1) >> shift;
3899 #endif
3900
3901                         /*
3902                          * We've basically got 3 bytes, or 24 bits for the
3903                          * cylinder size in the mode page.  If we're over,
3904                          * just round down to 2^24.
3905                          */
3906                         if (cylinders > 0xffffff)
3907                                 cylinders = 0xffffff;
3908
3909                         rigid_disk_page = &lun->mode_pages.rigid_disk_page[
3910                                 CTL_PAGE_DEFAULT];
3911                         scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
3912
3913                         if ((value = ctl_get_opt(&lun->be_lun->options,
3914                             "rpm")) != NULL) {
3915                                 scsi_ulto2b(strtol(value, NULL, 0),
3916                                      rigid_disk_page->rotation_rate);
3917                         }
3918
3919                         memcpy(&lun->mode_pages.rigid_disk_page[CTL_PAGE_CURRENT],
3920                                &lun->mode_pages.rigid_disk_page[CTL_PAGE_DEFAULT],
3921                                sizeof(rigid_disk_page_default));
3922                         memcpy(&lun->mode_pages.rigid_disk_page[CTL_PAGE_SAVED],
3923                                &lun->mode_pages.rigid_disk_page[CTL_PAGE_DEFAULT],
3924                                sizeof(rigid_disk_page_default));
3925
3926                         page_index->page_data =
3927                                 (uint8_t *)lun->mode_pages.rigid_disk_page;
3928                         break;
3929                 }
3930                 case SMS_CACHING_PAGE: {
3931                         struct scsi_caching_page *caching_page;
3932
3933                         if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3934                                 panic("invalid subpage value %d",
3935                                       page_index->subpage);
3936                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_DEFAULT],
3937                                &caching_page_default,
3938                                sizeof(caching_page_default));
3939                         memcpy(&lun->mode_pages.caching_page[
3940                                CTL_PAGE_CHANGEABLE], &caching_page_changeable,
3941                                sizeof(caching_page_changeable));
3942                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_SAVED],
3943                                &caching_page_default,
3944                                sizeof(caching_page_default));
3945                         caching_page = &lun->mode_pages.caching_page[
3946                             CTL_PAGE_SAVED];
3947                         value = ctl_get_opt(&lun->be_lun->options, "writecache");
3948                         if (value != NULL && strcmp(value, "off") == 0)
3949                                 caching_page->flags1 &= ~SCP_WCE;
3950                         value = ctl_get_opt(&lun->be_lun->options, "readcache");
3951                         if (value != NULL && strcmp(value, "off") == 0)
3952                                 caching_page->flags1 |= SCP_RCD;
3953                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_CURRENT],
3954                                &lun->mode_pages.caching_page[CTL_PAGE_SAVED],
3955                                sizeof(caching_page_default));
3956                         page_index->page_data =
3957                                 (uint8_t *)lun->mode_pages.caching_page;
3958                         break;
3959                 }
3960                 case SMS_CONTROL_MODE_PAGE: {
3961                         struct scsi_control_page *control_page;
3962
3963                         if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3964                                 panic("invalid subpage value %d",
3965                                       page_index->subpage);
3966
3967                         memcpy(&lun->mode_pages.control_page[CTL_PAGE_DEFAULT],
3968                                &control_page_default,
3969                                sizeof(control_page_default));
3970                         memcpy(&lun->mode_pages.control_page[
3971                                CTL_PAGE_CHANGEABLE], &control_page_changeable,
3972                                sizeof(control_page_changeable));
3973                         memcpy(&lun->mode_pages.control_page[CTL_PAGE_SAVED],
3974                                &control_page_default,
3975                                sizeof(control_page_default));
3976                         control_page = &lun->mode_pages.control_page[
3977                             CTL_PAGE_SAVED];
3978                         value = ctl_get_opt(&lun->be_lun->options, "reordering");
3979                         if (value != NULL && strcmp(value, "unrestricted") == 0) {
3980                                 control_page->queue_flags &= ~SCP_QUEUE_ALG_MASK;
3981                                 control_page->queue_flags |= SCP_QUEUE_ALG_UNRESTRICTED;
3982                         }
3983                         memcpy(&lun->mode_pages.control_page[CTL_PAGE_CURRENT],
3984                                &lun->mode_pages.control_page[CTL_PAGE_SAVED],
3985                                sizeof(control_page_default));
3986                         page_index->page_data =
3987                                 (uint8_t *)lun->mode_pages.control_page;
3988                         break;
3989
3990                 }
3991                 case SMS_INFO_EXCEPTIONS_PAGE: {
3992                         switch (page_index->subpage) {
3993                         case SMS_SUBPAGE_PAGE_0:
3994                                 memcpy(&lun->mode_pages.ie_page[CTL_PAGE_CURRENT],
3995                                        &ie_page_default,
3996                                        sizeof(ie_page_default));
3997                                 memcpy(&lun->mode_pages.ie_page[
3998                                        CTL_PAGE_CHANGEABLE], &ie_page_changeable,
3999                                        sizeof(ie_page_changeable));
4000                                 memcpy(&lun->mode_pages.ie_page[CTL_PAGE_DEFAULT],
4001                                        &ie_page_default,
4002                                        sizeof(ie_page_default));
4003                                 memcpy(&lun->mode_pages.ie_page[CTL_PAGE_SAVED],
4004                                        &ie_page_default,
4005                                        sizeof(ie_page_default));
4006                                 page_index->page_data =
4007                                         (uint8_t *)lun->mode_pages.ie_page;
4008                                 break;
4009                         case 0x02: {
4010                                 struct ctl_logical_block_provisioning_page *page;
4011
4012                                 memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_DEFAULT],
4013                                        &lbp_page_default,
4014                                        sizeof(lbp_page_default));
4015                                 memcpy(&lun->mode_pages.lbp_page[
4016                                        CTL_PAGE_CHANGEABLE], &lbp_page_changeable,
4017                                        sizeof(lbp_page_changeable));
4018                                 memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_SAVED],
4019                                        &lbp_page_default,
4020                                        sizeof(lbp_page_default));
4021                                 page = &lun->mode_pages.lbp_page[CTL_PAGE_SAVED];
4022                                 value = ctl_get_opt(&lun->be_lun->options,
4023                                     "avail-threshold");
4024                                 if (value != NULL &&
4025                                     ctl_expand_number(value, &ival) == 0) {
4026                                         page->descr[0].flags |= SLBPPD_ENABLED |
4027                                             SLBPPD_ARMING_DEC;
4028                                         if (lun->be_lun->blocksize)
4029                                                 ival /= lun->be_lun->blocksize;
4030                                         else
4031                                                 ival /= 512;
4032                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4033                                             page->descr[0].count);
4034                                 }
4035                                 value = ctl_get_opt(&lun->be_lun->options,
4036                                     "used-threshold");
4037                                 if (value != NULL &&
4038                                     ctl_expand_number(value, &ival) == 0) {
4039                                         page->descr[1].flags |= SLBPPD_ENABLED |
4040                                             SLBPPD_ARMING_INC;
4041                                         if (lun->be_lun->blocksize)
4042                                                 ival /= lun->be_lun->blocksize;
4043                                         else
4044                                                 ival /= 512;
4045                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4046                                             page->descr[1].count);
4047                                 }
4048                                 value = ctl_get_opt(&lun->be_lun->options,
4049                                     "pool-avail-threshold");
4050                                 if (value != NULL &&
4051                                     ctl_expand_number(value, &ival) == 0) {
4052                                         page->descr[2].flags |= SLBPPD_ENABLED |
4053                                             SLBPPD_ARMING_DEC;
4054                                         if (lun->be_lun->blocksize)
4055                                                 ival /= lun->be_lun->blocksize;
4056                                         else
4057                                                 ival /= 512;
4058                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4059                                             page->descr[2].count);
4060                                 }
4061                                 value = ctl_get_opt(&lun->be_lun->options,
4062                                     "pool-used-threshold");
4063                                 if (value != NULL &&
4064                                     ctl_expand_number(value, &ival) == 0) {
4065                                         page->descr[3].flags |= SLBPPD_ENABLED |
4066                                             SLBPPD_ARMING_INC;
4067                                         if (lun->be_lun->blocksize)
4068                                                 ival /= lun->be_lun->blocksize;
4069                                         else
4070                                                 ival /= 512;
4071                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4072                                             page->descr[3].count);
4073                                 }
4074                                 memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_CURRENT],
4075                                        &lun->mode_pages.lbp_page[CTL_PAGE_SAVED],
4076                                        sizeof(lbp_page_default));
4077                                 page_index->page_data =
4078                                         (uint8_t *)lun->mode_pages.lbp_page;
4079                         }}
4080                         break;
4081                 }
4082                 case SMS_VENDOR_SPECIFIC_PAGE:{
4083                         switch (page_index->subpage) {
4084                         case DBGCNF_SUBPAGE_CODE: {
4085                                 struct copan_debugconf_subpage *current_page,
4086                                                                *saved_page;
4087
4088                                 memcpy(&lun->mode_pages.debugconf_subpage[
4089                                        CTL_PAGE_CURRENT],
4090                                        &debugconf_page_default,
4091                                        sizeof(debugconf_page_default));
4092                                 memcpy(&lun->mode_pages.debugconf_subpage[
4093                                        CTL_PAGE_CHANGEABLE],
4094                                        &debugconf_page_changeable,
4095                                        sizeof(debugconf_page_changeable));
4096                                 memcpy(&lun->mode_pages.debugconf_subpage[
4097                                        CTL_PAGE_DEFAULT],
4098                                        &debugconf_page_default,
4099                                        sizeof(debugconf_page_default));
4100                                 memcpy(&lun->mode_pages.debugconf_subpage[
4101                                        CTL_PAGE_SAVED],
4102                                        &debugconf_page_default,
4103                                        sizeof(debugconf_page_default));
4104                                 page_index->page_data =
4105                                         (uint8_t *)lun->mode_pages.debugconf_subpage;
4106
4107                                 current_page = (struct copan_debugconf_subpage *)
4108                                         (page_index->page_data +
4109                                          (page_index->page_len *
4110                                           CTL_PAGE_CURRENT));
4111                                 saved_page = (struct copan_debugconf_subpage *)
4112                                         (page_index->page_data +
4113                                          (page_index->page_len *
4114                                           CTL_PAGE_SAVED));
4115                                 break;
4116                         }
4117                         default:
4118                                 panic("invalid subpage value %d",
4119                                       page_index->subpage);
4120                                 break;
4121                         }
4122                         break;
4123                 }
4124                 default:
4125                         panic("invalid page value %d",
4126                               page_index->page_code & SMPH_PC_MASK);
4127                         break;
4128         }
4129         }
4130
4131         return (CTL_RETVAL_COMPLETE);
4132 }
4133
4134 static int
4135 ctl_init_log_page_index(struct ctl_lun *lun)
4136 {
4137         struct ctl_page_index *page_index;
4138         int i, j, k, prev;
4139
4140         memcpy(&lun->log_pages.index, log_page_index_template,
4141                sizeof(log_page_index_template));
4142
4143         prev = -1;
4144         for (i = 0, j = 0, k = 0; i < CTL_NUM_LOG_PAGES; i++) {
4145
4146                 page_index = &lun->log_pages.index[i];
4147                 /*
4148                  * If this is a disk-only mode page, there's no point in
4149                  * setting it up.  For some pages, we have to have some
4150                  * basic information about the disk in order to calculate the
4151                  * mode page data.
4152                  */
4153                 if ((lun->be_lun->lun_type != T_DIRECT)
4154                  && (page_index->page_flags & CTL_PAGE_FLAG_DISK_ONLY))
4155                         continue;
4156
4157                 if (page_index->page_code == SLS_LOGICAL_BLOCK_PROVISIONING &&
4158                      lun->backend->lun_attr == NULL)
4159                         continue;
4160
4161                 if (page_index->page_code != prev) {
4162                         lun->log_pages.pages_page[j] = page_index->page_code;
4163                         prev = page_index->page_code;
4164                         j++;
4165                 }
4166                 lun->log_pages.subpages_page[k*2] = page_index->page_code;
4167                 lun->log_pages.subpages_page[k*2+1] = page_index->subpage;
4168                 k++;
4169         }
4170         lun->log_pages.index[0].page_data = &lun->log_pages.pages_page[0];
4171         lun->log_pages.index[0].page_len = j;
4172         lun->log_pages.index[1].page_data = &lun->log_pages.subpages_page[0];
4173         lun->log_pages.index[1].page_len = k * 2;
4174         lun->log_pages.index[2].page_data = &lun->log_pages.lbp_page[0];
4175         lun->log_pages.index[2].page_len = 12*CTL_NUM_LBP_PARAMS;
4176         lun->log_pages.index[3].page_data = (uint8_t *)&lun->log_pages.stat_page;
4177         lun->log_pages.index[3].page_len = sizeof(lun->log_pages.stat_page);
4178
4179         return (CTL_RETVAL_COMPLETE);
4180 }
4181
4182 static int
4183 hex2bin(const char *str, uint8_t *buf, int buf_size)
4184 {
4185         int i;
4186         u_char c;
4187
4188         memset(buf, 0, buf_size);
4189         while (isspace(str[0]))
4190                 str++;
4191         if (str[0] == '0' && (str[1] == 'x' || str[1] == 'X'))
4192                 str += 2;
4193         buf_size *= 2;
4194         for (i = 0; str[i] != 0 && i < buf_size; i++) {
4195                 c = str[i];
4196                 if (isdigit(c))
4197                         c -= '0';
4198                 else if (isalpha(c))
4199                         c -= isupper(c) ? 'A' - 10 : 'a' - 10;
4200                 else
4201                         break;
4202                 if (c >= 16)
4203                         break;
4204                 if ((i & 1) == 0)
4205                         buf[i / 2] |= (c << 4);
4206                 else
4207                         buf[i / 2] |= c;
4208         }
4209         return ((i + 1) / 2);
4210 }
4211
4212 /*
4213  * LUN allocation.
4214  *
4215  * Requirements:
4216  * - caller allocates and zeros LUN storage, or passes in a NULL LUN if he
4217  *   wants us to allocate the LUN and he can block.
4218  * - ctl_softc is always set
4219  * - be_lun is set if the LUN has a backend (needed for disk LUNs)
4220  *
4221  * Returns 0 for success, non-zero (errno) for failure.
4222  */
4223 static int
4224 ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *ctl_lun,
4225               struct ctl_be_lun *const be_lun)
4226 {
4227         struct ctl_lun *nlun, *lun;
4228         struct scsi_vpd_id_descriptor *desc;
4229         struct scsi_vpd_id_t10 *t10id;
4230         const char *eui, *naa, *scsiname, *vendor;
4231         int lun_number, i, lun_malloced;
4232         int devidlen, idlen1, idlen2 = 0, len;
4233
4234         if (be_lun == NULL)
4235                 return (EINVAL);
4236
4237         /*
4238          * We currently only support Direct Access or Processor LUN types.
4239          */
4240         switch (be_lun->lun_type) {
4241         case T_DIRECT:
4242                 break;
4243         case T_PROCESSOR:
4244                 break;
4245         case T_SEQUENTIAL:
4246         case T_CHANGER:
4247         default:
4248                 be_lun->lun_config_status(be_lun->be_lun,
4249                                           CTL_LUN_CONFIG_FAILURE);
4250                 break;
4251         }
4252         if (ctl_lun == NULL) {
4253                 lun = malloc(sizeof(*lun), M_CTL, M_WAITOK);
4254                 lun_malloced = 1;
4255         } else {
4256                 lun_malloced = 0;
4257                 lun = ctl_lun;
4258         }
4259
4260         memset(lun, 0, sizeof(*lun));
4261         if (lun_malloced)
4262                 lun->flags = CTL_LUN_MALLOCED;
4263
4264         /* Generate LUN ID. */
4265         devidlen = max(CTL_DEVID_MIN_LEN,
4266             strnlen(be_lun->device_id, CTL_DEVID_LEN));
4267         idlen1 = sizeof(*t10id) + devidlen;
4268         len = sizeof(struct scsi_vpd_id_descriptor) + idlen1;
4269         scsiname = ctl_get_opt(&be_lun->options, "scsiname");
4270         if (scsiname != NULL) {
4271                 idlen2 = roundup2(strlen(scsiname) + 1, 4);
4272                 len += sizeof(struct scsi_vpd_id_descriptor) + idlen2;
4273         }
4274         eui = ctl_get_opt(&be_lun->options, "eui");
4275         if (eui != NULL) {
4276                 len += sizeof(struct scsi_vpd_id_descriptor) + 16;
4277         }
4278         naa = ctl_get_opt(&be_lun->options, "naa");
4279         if (naa != NULL) {
4280                 len += sizeof(struct scsi_vpd_id_descriptor) + 16;
4281         }
4282         lun->lun_devid = malloc(sizeof(struct ctl_devid) + len,
4283             M_CTL, M_WAITOK | M_ZERO);
4284         desc = (struct scsi_vpd_id_descriptor *)lun->lun_devid->data;
4285         desc->proto_codeset = SVPD_ID_CODESET_ASCII;
4286         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_T10;
4287         desc->length = idlen1;
4288         t10id = (struct scsi_vpd_id_t10 *)&desc->identifier[0];
4289         memset(t10id->vendor, ' ', sizeof(t10id->vendor));
4290         if ((vendor = ctl_get_opt(&be_lun->options, "vendor")) == NULL) {
4291                 strncpy((char *)t10id->vendor, CTL_VENDOR, sizeof(t10id->vendor));
4292         } else {
4293                 strncpy(t10id->vendor, vendor,
4294                     min(sizeof(t10id->vendor), strlen(vendor)));
4295         }
4296         strncpy((char *)t10id->vendor_spec_id,
4297             (char *)be_lun->device_id, devidlen);
4298         if (scsiname != NULL) {
4299                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4300                     desc->length);
4301                 desc->proto_codeset = SVPD_ID_CODESET_UTF8;
4302                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4303                     SVPD_ID_TYPE_SCSI_NAME;
4304                 desc->length = idlen2;
4305                 strlcpy(desc->identifier, scsiname, idlen2);
4306         }
4307         if (eui != NULL) {
4308                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4309                     desc->length);
4310                 desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4311                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4312                     SVPD_ID_TYPE_EUI64;
4313                 desc->length = hex2bin(eui, desc->identifier, 16);
4314                 desc->length = desc->length > 12 ? 16 :
4315                     (desc->length > 8 ? 12 : 8);
4316                 len -= 16 - desc->length;
4317         }
4318         if (naa != NULL) {
4319                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4320                     desc->length);
4321                 desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4322                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4323                     SVPD_ID_TYPE_NAA;
4324                 desc->length = hex2bin(naa, desc->identifier, 16);
4325                 desc->length = desc->length > 8 ? 16 : 8;
4326                 len -= 16 - desc->length;
4327         }
4328         lun->lun_devid->len = len;
4329
4330         mtx_lock(&ctl_softc->ctl_lock);
4331         /*
4332          * See if the caller requested a particular LUN number.  If so, see
4333          * if it is available.  Otherwise, allocate the first available LUN.
4334          */
4335         if (be_lun->flags & CTL_LUN_FLAG_ID_REQ) {
4336                 if ((be_lun->req_lun_id > (CTL_MAX_LUNS - 1))
4337                  || (ctl_is_set(ctl_softc->ctl_lun_mask, be_lun->req_lun_id))) {
4338                         mtx_unlock(&ctl_softc->ctl_lock);
4339                         if (be_lun->req_lun_id > (CTL_MAX_LUNS - 1)) {
4340                                 printf("ctl: requested LUN ID %d is higher "
4341                                        "than CTL_MAX_LUNS - 1 (%d)\n",
4342                                        be_lun->req_lun_id, CTL_MAX_LUNS - 1);
4343                         } else {
4344                                 /*
4345                                  * XXX KDM return an error, or just assign
4346                                  * another LUN ID in this case??
4347                                  */
4348                                 printf("ctl: requested LUN ID %d is already "
4349                                        "in use\n", be_lun->req_lun_id);
4350                         }
4351                         if (lun->flags & CTL_LUN_MALLOCED)
4352                                 free(lun, M_CTL);
4353                         be_lun->lun_config_status(be_lun->be_lun,
4354                                                   CTL_LUN_CONFIG_FAILURE);
4355                         return (ENOSPC);
4356                 }
4357                 lun_number = be_lun->req_lun_id;
4358         } else {
4359                 lun_number = ctl_ffz(ctl_softc->ctl_lun_mask, 0, CTL_MAX_LUNS);
4360                 if (lun_number == -1) {
4361                         mtx_unlock(&ctl_softc->ctl_lock);
4362                         printf("ctl: can't allocate LUN, out of LUNs\n");
4363                         if (lun->flags & CTL_LUN_MALLOCED)
4364                                 free(lun, M_CTL);
4365                         be_lun->lun_config_status(be_lun->be_lun,
4366                                                   CTL_LUN_CONFIG_FAILURE);
4367                         return (ENOSPC);
4368                 }
4369         }
4370         ctl_set_mask(ctl_softc->ctl_lun_mask, lun_number);
4371
4372         mtx_init(&lun->lun_lock, "CTL LUN", NULL, MTX_DEF);
4373         lun->lun = lun_number;
4374         lun->be_lun = be_lun;
4375         /*
4376          * The processor LUN is always enabled.  Disk LUNs come on line
4377          * disabled, and must be enabled by the backend.
4378          */
4379         lun->flags |= CTL_LUN_DISABLED;
4380         lun->backend = be_lun->be;
4381         be_lun->ctl_lun = lun;
4382         be_lun->lun_id = lun_number;
4383         atomic_add_int(&be_lun->be->num_luns, 1);
4384         if (be_lun->flags & CTL_LUN_FLAG_OFFLINE)
4385                 lun->flags |= CTL_LUN_OFFLINE;
4386
4387         if (be_lun->flags & CTL_LUN_FLAG_POWERED_OFF)
4388                 lun->flags |= CTL_LUN_STOPPED;
4389
4390         if (be_lun->flags & CTL_LUN_FLAG_INOPERABLE)
4391                 lun->flags |= CTL_LUN_INOPERABLE;
4392
4393         if (be_lun->flags & CTL_LUN_FLAG_PRIMARY)
4394                 lun->flags |= CTL_LUN_PRIMARY_SC;
4395
4396         lun->ctl_softc = ctl_softc;
4397 #ifdef CTL_TIME_IO
4398         lun->last_busy = getsbinuptime();
4399 #endif
4400         TAILQ_INIT(&lun->ooa_queue);
4401         TAILQ_INIT(&lun->blocked_queue);
4402         STAILQ_INIT(&lun->error_list);
4403         ctl_tpc_lun_init(lun);
4404
4405         /*
4406          * Initialize the mode and log page index.
4407          */
4408         ctl_init_page_index(lun);
4409         ctl_init_log_page_index(lun);
4410
4411         /*
4412          * Now, before we insert this lun on the lun list, set the lun
4413          * inventory changed UA for all other luns.
4414          */
4415         STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) {
4416                 mtx_lock(&nlun->lun_lock);
4417                 ctl_est_ua_all(nlun, -1, CTL_UA_LUN_CHANGE);
4418                 mtx_unlock(&nlun->lun_lock);
4419         }
4420
4421         STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links);
4422
4423         ctl_softc->ctl_luns[lun_number] = lun;
4424
4425         ctl_softc->num_luns++;
4426
4427         /* Setup statistics gathering */
4428         lun->stats.device_type = be_lun->lun_type;
4429         lun->stats.lun_number = lun_number;
4430         if (lun->stats.device_type == T_DIRECT)
4431                 lun->stats.blocksize = be_lun->blocksize;
4432         else
4433                 lun->stats.flags = CTL_LUN_STATS_NO_BLOCKSIZE;
4434         for (i = 0;i < CTL_MAX_PORTS;i++)
4435                 lun->stats.ports[i].targ_port = i;
4436
4437         mtx_unlock(&ctl_softc->ctl_lock);
4438
4439         lun->be_lun->lun_config_status(lun->be_lun->be_lun, CTL_LUN_CONFIG_OK);
4440         return (0);
4441 }
4442
4443 /*
4444  * Delete a LUN.
4445  * Assumptions:
4446  * - LUN has already been marked invalid and any pending I/O has been taken
4447  *   care of.
4448  */
4449 static int
4450 ctl_free_lun(struct ctl_lun *lun)
4451 {
4452         struct ctl_softc *softc;
4453         struct ctl_lun *nlun;
4454         int i;
4455
4456         softc = lun->ctl_softc;
4457
4458         mtx_assert(&softc->ctl_lock, MA_OWNED);
4459
4460         STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links);
4461
4462         ctl_clear_mask(softc->ctl_lun_mask, lun->lun);
4463
4464         softc->ctl_luns[lun->lun] = NULL;
4465
4466         if (!TAILQ_EMPTY(&lun->ooa_queue))
4467                 panic("Freeing a LUN %p with outstanding I/O!!\n", lun);
4468
4469         softc->num_luns--;
4470
4471         /*
4472          * Tell the backend to free resources, if this LUN has a backend.
4473          */
4474         atomic_subtract_int(&lun->be_lun->be->num_luns, 1);
4475         lun->be_lun->lun_shutdown(lun->be_lun->be_lun);
4476
4477         ctl_tpc_lun_shutdown(lun);
4478         mtx_destroy(&lun->lun_lock);
4479         free(lun->lun_devid, M_CTL);
4480         for (i = 0; i < CTL_MAX_PORTS; i++)
4481                 free(lun->pending_ua[i], M_CTL);
4482         for (i = 0; i < CTL_MAX_PORTS; i++)
4483                 free(lun->pr_keys[i], M_CTL);
4484         free(lun->write_buffer, M_CTL);
4485         if (lun->flags & CTL_LUN_MALLOCED)
4486                 free(lun, M_CTL);
4487
4488         STAILQ_FOREACH(nlun, &softc->lun_list, links) {
4489                 mtx_lock(&nlun->lun_lock);
4490                 ctl_est_ua_all(nlun, -1, CTL_UA_LUN_CHANGE);
4491                 mtx_unlock(&nlun->lun_lock);
4492         }
4493
4494         return (0);
4495 }
4496
4497 static void
4498 ctl_create_lun(struct ctl_be_lun *be_lun)
4499 {
4500         struct ctl_softc *softc;
4501
4502         softc = control_softc;
4503
4504         /*
4505          * ctl_alloc_lun() should handle all potential failure cases.
4506          */
4507         ctl_alloc_lun(softc, NULL, be_lun);
4508 }
4509
4510 int
4511 ctl_add_lun(struct ctl_be_lun *be_lun)
4512 {
4513         struct ctl_softc *softc = control_softc;
4514
4515         mtx_lock(&softc->ctl_lock);
4516         STAILQ_INSERT_TAIL(&softc->pending_lun_queue, be_lun, links);
4517         mtx_unlock(&softc->ctl_lock);
4518         wakeup(&softc->pending_lun_queue);
4519
4520         return (0);
4521 }
4522
4523 int
4524 ctl_enable_lun(struct ctl_be_lun *be_lun)
4525 {
4526         struct ctl_softc *softc;
4527         struct ctl_port *port, *nport;
4528         struct ctl_lun *lun;
4529         int retval;
4530
4531         lun = (struct ctl_lun *)be_lun->ctl_lun;
4532         softc = lun->ctl_softc;
4533
4534         mtx_lock(&softc->ctl_lock);
4535         mtx_lock(&lun->lun_lock);
4536         if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4537                 /*
4538                  * eh?  Why did we get called if the LUN is already
4539                  * enabled?
4540                  */
4541                 mtx_unlock(&lun->lun_lock);
4542                 mtx_unlock(&softc->ctl_lock);
4543                 return (0);
4544         }
4545         lun->flags &= ~CTL_LUN_DISABLED;
4546         mtx_unlock(&lun->lun_lock);
4547
4548         for (port = STAILQ_FIRST(&softc->port_list); port != NULL; port = nport) {
4549                 nport = STAILQ_NEXT(port, links);
4550                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0 ||
4551                     port->lun_map != NULL || port->lun_enable == NULL)
4552                         continue;
4553
4554                 /*
4555                  * Drop the lock while we call the FETD's enable routine.
4556                  * This can lead to a callback into CTL (at least in the
4557                  * case of the internal initiator frontend.
4558                  */
4559                 mtx_unlock(&softc->ctl_lock);
4560                 retval = port->lun_enable(port->targ_lun_arg, lun->lun);
4561                 mtx_lock(&softc->ctl_lock);
4562                 if (retval != 0) {
4563                         printf("%s: FETD %s port %d returned error "
4564                                "%d for lun_enable on lun %jd\n",
4565                                __func__, port->port_name, port->targ_port,
4566                                retval, (intmax_t)lun->lun);
4567                 }
4568         }
4569
4570         mtx_unlock(&softc->ctl_lock);
4571         ctl_isc_announce_lun(lun);
4572
4573         return (0);
4574 }
4575
4576 int
4577 ctl_disable_lun(struct ctl_be_lun *be_lun)
4578 {
4579         struct ctl_softc *softc;
4580         struct ctl_port *port;
4581         struct ctl_lun *lun;
4582         int retval;
4583
4584         lun = (struct ctl_lun *)be_lun->ctl_lun;
4585         softc = lun->ctl_softc;
4586
4587         mtx_lock(&softc->ctl_lock);
4588         mtx_lock(&lun->lun_lock);
4589         if (lun->flags & CTL_LUN_DISABLED) {
4590                 mtx_unlock(&lun->lun_lock);
4591                 mtx_unlock(&softc->ctl_lock);
4592                 return (0);
4593         }
4594         lun->flags |= CTL_LUN_DISABLED;
4595         mtx_unlock(&lun->lun_lock);
4596
4597         STAILQ_FOREACH(port, &softc->port_list, links) {
4598                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0 ||
4599                     port->lun_map != NULL || port->lun_disable == NULL)
4600                         continue;
4601
4602                 /*
4603                  * Drop the lock before we call the frontend's disable
4604                  * routine, to avoid lock order reversals.
4605                  *
4606                  * XXX KDM what happens if the frontend list changes while
4607                  * we're traversing it?  It's unlikely, but should be handled.
4608                  */
4609                 mtx_unlock(&softc->ctl_lock);
4610                 retval = port->lun_disable(port->targ_lun_arg, lun->lun);
4611                 mtx_lock(&softc->ctl_lock);
4612                 if (retval != 0) {
4613                         printf("%s: FETD %s port %d returned error "
4614                                "%d for lun_disable on lun %jd\n",
4615                                __func__, port->port_name, port->targ_port,
4616                                retval, (intmax_t)lun->lun);
4617                 }
4618         }
4619
4620         mtx_unlock(&softc->ctl_lock);
4621         ctl_isc_announce_lun(lun);
4622
4623         return (0);
4624 }
4625
4626 int
4627 ctl_start_lun(struct ctl_be_lun *be_lun)
4628 {
4629         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4630
4631         mtx_lock(&lun->lun_lock);
4632         lun->flags &= ~CTL_LUN_STOPPED;
4633         mtx_unlock(&lun->lun_lock);
4634         return (0);
4635 }
4636
4637 int
4638 ctl_stop_lun(struct ctl_be_lun *be_lun)
4639 {
4640         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4641
4642         mtx_lock(&lun->lun_lock);
4643         lun->flags |= CTL_LUN_STOPPED;
4644         mtx_unlock(&lun->lun_lock);
4645         return (0);
4646 }
4647
4648 int
4649 ctl_lun_offline(struct ctl_be_lun *be_lun)
4650 {
4651         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4652
4653         mtx_lock(&lun->lun_lock);
4654         lun->flags |= CTL_LUN_OFFLINE;
4655         mtx_unlock(&lun->lun_lock);
4656         return (0);
4657 }
4658
4659 int
4660 ctl_lun_online(struct ctl_be_lun *be_lun)
4661 {
4662         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4663
4664         mtx_lock(&lun->lun_lock);
4665         lun->flags &= ~CTL_LUN_OFFLINE;
4666         mtx_unlock(&lun->lun_lock);
4667         return (0);
4668 }
4669
4670 int
4671 ctl_lun_primary(struct ctl_be_lun *be_lun)
4672 {
4673         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4674
4675         mtx_lock(&lun->lun_lock);
4676         lun->flags |= CTL_LUN_PRIMARY_SC;
4677         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
4678         mtx_unlock(&lun->lun_lock);
4679         ctl_isc_announce_lun(lun);
4680         return (0);
4681 }
4682
4683 int
4684 ctl_lun_secondary(struct ctl_be_lun *be_lun)
4685 {
4686         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4687
4688         mtx_lock(&lun->lun_lock);
4689         lun->flags &= ~CTL_LUN_PRIMARY_SC;
4690         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
4691         mtx_unlock(&lun->lun_lock);
4692         ctl_isc_announce_lun(lun);
4693         return (0);
4694 }
4695
4696 int
4697 ctl_invalidate_lun(struct ctl_be_lun *be_lun)
4698 {
4699         struct ctl_softc *softc;
4700         struct ctl_lun *lun;
4701
4702         lun = (struct ctl_lun *)be_lun->ctl_lun;
4703         softc = lun->ctl_softc;
4704
4705         mtx_lock(&lun->lun_lock);
4706
4707         /*
4708          * The LUN needs to be disabled before it can be marked invalid.
4709          */
4710         if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4711                 mtx_unlock(&lun->lun_lock);
4712                 return (-1);
4713         }
4714         /*
4715          * Mark the LUN invalid.
4716          */
4717         lun->flags |= CTL_LUN_INVALID;
4718
4719         /*
4720          * If there is nothing in the OOA queue, go ahead and free the LUN.
4721          * If we have something in the OOA queue, we'll free it when the
4722          * last I/O completes.
4723          */
4724         if (TAILQ_EMPTY(&lun->ooa_queue)) {
4725                 mtx_unlock(&lun->lun_lock);
4726                 mtx_lock(&softc->ctl_lock);
4727                 ctl_free_lun(lun);
4728                 mtx_unlock(&softc->ctl_lock);
4729         } else
4730                 mtx_unlock(&lun->lun_lock);
4731
4732         return (0);
4733 }
4734
4735 int
4736 ctl_lun_inoperable(struct ctl_be_lun *be_lun)
4737 {
4738         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4739
4740         mtx_lock(&lun->lun_lock);
4741         lun->flags |= CTL_LUN_INOPERABLE;
4742         mtx_unlock(&lun->lun_lock);
4743         return (0);
4744 }
4745
4746 int
4747 ctl_lun_operable(struct ctl_be_lun *be_lun)
4748 {
4749         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4750
4751         mtx_lock(&lun->lun_lock);
4752         lun->flags &= ~CTL_LUN_INOPERABLE;
4753         mtx_unlock(&lun->lun_lock);
4754         return (0);
4755 }
4756
4757 void
4758 ctl_lun_capacity_changed(struct ctl_be_lun *be_lun)
4759 {
4760         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4761         union ctl_ha_msg msg;
4762
4763         mtx_lock(&lun->lun_lock);
4764         ctl_est_ua_all(lun, -1, CTL_UA_CAPACITY_CHANGED);
4765         mtx_unlock(&lun->lun_lock);
4766         if (lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
4767                 /* Send msg to other side. */
4768                 bzero(&msg.ua, sizeof(msg.ua));
4769                 msg.hdr.msg_type = CTL_MSG_UA;
4770                 msg.hdr.nexus.initid = -1;
4771                 msg.hdr.nexus.targ_port = -1;
4772                 msg.hdr.nexus.targ_lun = lun->lun;
4773                 msg.hdr.nexus.targ_mapped_lun = lun->lun;
4774                 msg.ua.ua_all = 1;
4775                 msg.ua.ua_set = 1;
4776                 msg.ua.ua_type = CTL_UA_CAPACITY_CHANGED;
4777                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg.ua),
4778                     M_WAITOK);
4779         }
4780 }
4781
4782 /*
4783  * Backend "memory move is complete" callback for requests that never
4784  * make it down to say RAIDCore's configuration code.
4785  */
4786 int
4787 ctl_config_move_done(union ctl_io *io)
4788 {
4789         int retval;
4790
4791         CTL_DEBUG_PRINT(("ctl_config_move_done\n"));
4792         KASSERT(io->io_hdr.io_type == CTL_IO_SCSI,
4793             ("Config I/O type isn't CTL_IO_SCSI (%d)!", io->io_hdr.io_type));
4794
4795         if ((io->io_hdr.port_status != 0) &&
4796             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
4797              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
4798                 /*
4799                  * For hardware error sense keys, the sense key
4800                  * specific value is defined to be a retry count,
4801                  * but we use it to pass back an internal FETD
4802                  * error code.  XXX KDM  Hopefully the FETD is only
4803                  * using 16 bits for an error code, since that's
4804                  * all the space we have in the sks field.
4805                  */
4806                 ctl_set_internal_failure(&io->scsiio,
4807                                          /*sks_valid*/ 1,
4808                                          /*retry_count*/
4809                                          io->io_hdr.port_status);
4810         }
4811
4812         if (ctl_debug & CTL_DEBUG_CDB_DATA)
4813                 ctl_data_print(io);
4814         if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN) ||
4815             ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
4816              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS) ||
4817             ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) {
4818                 /*
4819                  * XXX KDM just assuming a single pointer here, and not a
4820                  * S/G list.  If we start using S/G lists for config data,
4821                  * we'll need to know how to clean them up here as well.
4822                  */
4823                 if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
4824                         free(io->scsiio.kern_data_ptr, M_CTL);
4825                 ctl_done(io);
4826                 retval = CTL_RETVAL_COMPLETE;
4827         } else {
4828                 /*
4829                  * XXX KDM now we need to continue data movement.  Some
4830                  * options:
4831                  * - call ctl_scsiio() again?  We don't do this for data
4832                  *   writes, because for those at least we know ahead of
4833                  *   time where the write will go and how long it is.  For
4834                  *   config writes, though, that information is largely
4835                  *   contained within the write itself, thus we need to
4836                  *   parse out the data again.
4837                  *
4838                  * - Call some other function once the data is in?
4839                  */
4840
4841                 /*
4842                  * XXX KDM call ctl_scsiio() again for now, and check flag
4843                  * bits to see whether we're allocated or not.
4844                  */
4845                 retval = ctl_scsiio(&io->scsiio);
4846         }
4847         return (retval);
4848 }
4849
4850 /*
4851  * This gets called by a backend driver when it is done with a
4852  * data_submit method.
4853  */
4854 void
4855 ctl_data_submit_done(union ctl_io *io)
4856 {
4857         /*
4858          * If the IO_CONT flag is set, we need to call the supplied
4859          * function to continue processing the I/O, instead of completing
4860          * the I/O just yet.
4861          *
4862          * If there is an error, though, we don't want to keep processing.
4863          * Instead, just send status back to the initiator.
4864          */
4865         if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
4866             (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
4867             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
4868              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
4869                 io->scsiio.io_cont(io);
4870                 return;
4871         }
4872         ctl_done(io);
4873 }
4874
4875 /*
4876  * This gets called by a backend driver when it is done with a
4877  * configuration write.
4878  */
4879 void
4880 ctl_config_write_done(union ctl_io *io)
4881 {
4882         uint8_t *buf;
4883
4884         /*
4885          * If the IO_CONT flag is set, we need to call the supplied
4886          * function to continue processing the I/O, instead of completing
4887          * the I/O just yet.
4888          *
4889          * If there is an error, though, we don't want to keep processing.
4890          * Instead, just send status back to the initiator.
4891          */
4892         if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
4893             (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
4894             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
4895              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
4896                 io->scsiio.io_cont(io);
4897                 return;
4898         }
4899         /*
4900          * Since a configuration write can be done for commands that actually
4901          * have data allocated, like write buffer, and commands that have
4902          * no data, like start/stop unit, we need to check here.
4903          */
4904         if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
4905                 buf = io->scsiio.kern_data_ptr;
4906         else
4907                 buf = NULL;
4908         ctl_done(io);
4909         if (buf)
4910                 free(buf, M_CTL);
4911 }
4912
4913 void
4914 ctl_config_read_done(union ctl_io *io)
4915 {
4916         uint8_t *buf;
4917
4918         /*
4919          * If there is some error -- we are done, skip data transfer.
4920          */
4921         if ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0 ||
4922             ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
4923              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) {
4924                 if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
4925                         buf = io->scsiio.kern_data_ptr;
4926                 else
4927                         buf = NULL;
4928                 ctl_done(io);
4929                 if (buf)
4930                         free(buf, M_CTL);
4931                 return;
4932         }
4933
4934         /*
4935          * If the IO_CONT flag is set, we need to call the supplied
4936          * function to continue processing the I/O, instead of completing
4937          * the I/O just yet.
4938          */
4939         if (io->io_hdr.flags & CTL_FLAG_IO_CONT) {
4940                 io->scsiio.io_cont(io);
4941                 return;
4942         }
4943
4944         ctl_datamove(io);
4945 }
4946
4947 /*
4948  * SCSI release command.
4949  */
4950 int
4951 ctl_scsi_release(struct ctl_scsiio *ctsio)
4952 {
4953         int length, longid, thirdparty_id, resv_id;
4954         struct ctl_lun *lun;
4955         uint32_t residx;
4956
4957         length = 0;
4958         resv_id = 0;
4959
4960         CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
4961
4962         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
4963         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
4964
4965         switch (ctsio->cdb[0]) {
4966         case RELEASE_10: {
4967                 struct scsi_release_10 *cdb;
4968
4969                 cdb = (struct scsi_release_10 *)ctsio->cdb;
4970
4971                 if (cdb->byte2 & SR10_LONGID)
4972                         longid = 1;
4973                 else
4974                         thirdparty_id = cdb->thirdparty_id;
4975
4976                 resv_id = cdb->resv_id;
4977                 length = scsi_2btoul(cdb->length);
4978                 break;
4979         }
4980         }
4981
4982
4983         /*
4984          * XXX KDM right now, we only support LUN reservation.  We don't
4985          * support 3rd party reservations, or extent reservations, which
4986          * might actually need the parameter list.  If we've gotten this
4987          * far, we've got a LUN reservation.  Anything else got kicked out
4988          * above.  So, according to SPC, ignore the length.
4989          */
4990         length = 0;
4991
4992         if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
4993          && (length > 0)) {
4994                 ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
4995                 ctsio->kern_data_len = length;
4996                 ctsio->kern_total_len = length;
4997                 ctsio->kern_data_resid = 0;
4998                 ctsio->kern_rel_offset = 0;
4999                 ctsio->kern_sg_entries = 0;
5000                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5001                 ctsio->be_move_done = ctl_config_move_done;
5002                 ctl_datamove((union ctl_io *)ctsio);
5003
5004                 return (CTL_RETVAL_COMPLETE);
5005         }
5006
5007         if (length > 0)
5008                 thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5009
5010         mtx_lock(&lun->lun_lock);
5011
5012         /*
5013          * According to SPC, it is not an error for an intiator to attempt
5014          * to release a reservation on a LUN that isn't reserved, or that
5015          * is reserved by another initiator.  The reservation can only be
5016          * released, though, by the initiator who made it or by one of
5017          * several reset type events.
5018          */
5019         if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == residx))
5020                         lun->flags &= ~CTL_LUN_RESERVED;
5021
5022         mtx_unlock(&lun->lun_lock);
5023
5024         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5025                 free(ctsio->kern_data_ptr, M_CTL);
5026                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5027         }
5028
5029         ctl_set_success(ctsio);
5030         ctl_done((union ctl_io *)ctsio);
5031         return (CTL_RETVAL_COMPLETE);
5032 }
5033
5034 int
5035 ctl_scsi_reserve(struct ctl_scsiio *ctsio)
5036 {
5037         int extent, thirdparty, longid;
5038         int resv_id, length;
5039         uint64_t thirdparty_id;
5040         struct ctl_lun *lun;
5041         uint32_t residx;
5042
5043         extent = 0;
5044         thirdparty = 0;
5045         longid = 0;
5046         resv_id = 0;
5047         length = 0;
5048         thirdparty_id = 0;
5049
5050         CTL_DEBUG_PRINT(("ctl_reserve\n"));
5051
5052         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5053         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5054
5055         switch (ctsio->cdb[0]) {
5056         case RESERVE_10: {
5057                 struct scsi_reserve_10 *cdb;
5058
5059                 cdb = (struct scsi_reserve_10 *)ctsio->cdb;
5060
5061                 if (cdb->byte2 & SR10_LONGID)
5062                         longid = 1;
5063                 else
5064                         thirdparty_id = cdb->thirdparty_id;
5065
5066                 resv_id = cdb->resv_id;
5067                 length = scsi_2btoul(cdb->length);
5068                 break;
5069         }
5070         }
5071
5072         /*
5073          * XXX KDM right now, we only support LUN reservation.  We don't
5074          * support 3rd party reservations, or extent reservations, which
5075          * might actually need the parameter list.  If we've gotten this
5076          * far, we've got a LUN reservation.  Anything else got kicked out
5077          * above.  So, according to SPC, ignore the length.
5078          */
5079         length = 0;
5080
5081         if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5082          && (length > 0)) {
5083                 ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5084                 ctsio->kern_data_len = length;
5085                 ctsio->kern_total_len = length;
5086                 ctsio->kern_data_resid = 0;
5087                 ctsio->kern_rel_offset = 0;
5088                 ctsio->kern_sg_entries = 0;
5089                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5090                 ctsio->be_move_done = ctl_config_move_done;
5091                 ctl_datamove((union ctl_io *)ctsio);
5092
5093                 return (CTL_RETVAL_COMPLETE);
5094         }
5095
5096         if (length > 0)
5097                 thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
5098
5099         mtx_lock(&lun->lun_lock);
5100         if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx != residx)) {
5101                 ctl_set_reservation_conflict(ctsio);
5102                 goto bailout;
5103         }
5104
5105         lun->flags |= CTL_LUN_RESERVED;
5106         lun->res_idx = residx;
5107
5108         ctl_set_success(ctsio);
5109
5110 bailout:
5111         mtx_unlock(&lun->lun_lock);
5112
5113         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5114                 free(ctsio->kern_data_ptr, M_CTL);
5115                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5116         }
5117
5118         ctl_done((union ctl_io *)ctsio);
5119         return (CTL_RETVAL_COMPLETE);
5120 }
5121
5122 int
5123 ctl_start_stop(struct ctl_scsiio *ctsio)
5124 {
5125         struct scsi_start_stop_unit *cdb;
5126         struct ctl_lun *lun;
5127         int retval;
5128
5129         CTL_DEBUG_PRINT(("ctl_start_stop\n"));
5130
5131         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5132         retval = 0;
5133
5134         cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
5135
5136         /*
5137          * XXX KDM
5138          * We don't support the immediate bit on a stop unit.  In order to
5139          * do that, we would need to code up a way to know that a stop is
5140          * pending, and hold off any new commands until it completes, one
5141          * way or another.  Then we could accept or reject those commands
5142          * depending on its status.  We would almost need to do the reverse
5143          * of what we do below for an immediate start -- return the copy of
5144          * the ctl_io to the FETD with status to send to the host (and to
5145          * free the copy!) and then free the original I/O once the stop
5146          * actually completes.  That way, the OOA queue mechanism can work
5147          * to block commands that shouldn't proceed.  Another alternative
5148          * would be to put the copy in the queue in place of the original,
5149          * and return the original back to the caller.  That could be
5150          * slightly safer..
5151          */
5152         if ((cdb->byte2 & SSS_IMMED)
5153          && ((cdb->how & SSS_START) == 0)) {
5154                 ctl_set_invalid_field(ctsio,
5155                                       /*sks_valid*/ 1,
5156                                       /*command*/ 1,
5157                                       /*field*/ 1,
5158                                       /*bit_valid*/ 1,
5159                                       /*bit*/ 0);
5160                 ctl_done((union ctl_io *)ctsio);
5161                 return (CTL_RETVAL_COMPLETE);
5162         }
5163
5164         if ((lun->flags & CTL_LUN_PR_RESERVED)
5165          && ((cdb->how & SSS_START)==0)) {
5166                 uint32_t residx;
5167
5168                 residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5169                 if (ctl_get_prkey(lun, residx) == 0
5170                  || (lun->pr_res_idx!=residx && lun->res_type < 4)) {
5171
5172                         ctl_set_reservation_conflict(ctsio);
5173                         ctl_done((union ctl_io *)ctsio);
5174                         return (CTL_RETVAL_COMPLETE);
5175                 }
5176         }
5177
5178         /*
5179          * If there is no backend on this device, we can't start or stop
5180          * it.  In theory we shouldn't get any start/stop commands in the
5181          * first place at this level if the LUN doesn't have a backend.
5182          * That should get stopped by the command decode code.
5183          */
5184         if (lun->backend == NULL) {
5185                 ctl_set_invalid_opcode(ctsio);
5186                 ctl_done((union ctl_io *)ctsio);
5187                 return (CTL_RETVAL_COMPLETE);
5188         }
5189
5190         /*
5191          * XXX KDM Copan-specific offline behavior.
5192          * Figure out a reasonable way to port this?
5193          */
5194 #ifdef NEEDTOPORT
5195         mtx_lock(&lun->lun_lock);
5196
5197         if (((cdb->byte2 & SSS_ONOFFLINE) == 0)
5198          && (lun->flags & CTL_LUN_OFFLINE)) {
5199                 /*
5200                  * If the LUN is offline, and the on/offline bit isn't set,
5201                  * reject the start or stop.  Otherwise, let it through.
5202                  */
5203                 mtx_unlock(&lun->lun_lock);
5204                 ctl_set_lun_not_ready(ctsio);
5205                 ctl_done((union ctl_io *)ctsio);
5206         } else {
5207                 mtx_unlock(&lun->lun_lock);
5208 #endif /* NEEDTOPORT */
5209                 /*
5210                  * This could be a start or a stop when we're online,
5211                  * or a stop/offline or start/online.  A start or stop when
5212                  * we're offline is covered in the case above.
5213                  */
5214                 /*
5215                  * In the non-immediate case, we send the request to
5216                  * the backend and return status to the user when
5217                  * it is done.
5218                  *
5219                  * In the immediate case, we allocate a new ctl_io
5220                  * to hold a copy of the request, and send that to
5221                  * the backend.  We then set good status on the
5222                  * user's request and return it immediately.
5223                  */
5224                 if (cdb->byte2 & SSS_IMMED) {
5225                         union ctl_io *new_io;
5226
5227                         new_io = ctl_alloc_io(ctsio->io_hdr.pool);
5228                         ctl_copy_io((union ctl_io *)ctsio, new_io);
5229                         retval = lun->backend->config_write(new_io);
5230                         ctl_set_success(ctsio);
5231                         ctl_done((union ctl_io *)ctsio);
5232                 } else {
5233                         retval = lun->backend->config_write(
5234                                 (union ctl_io *)ctsio);
5235                 }
5236 #ifdef NEEDTOPORT
5237         }
5238 #endif
5239         return (retval);
5240 }
5241
5242 /*
5243  * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
5244  * we don't really do anything with the LBA and length fields if the user
5245  * passes them in.  Instead we'll just flush out the cache for the entire
5246  * LUN.
5247  */
5248 int
5249 ctl_sync_cache(struct ctl_scsiio *ctsio)
5250 {
5251         struct ctl_lun *lun;
5252         struct ctl_softc *softc;
5253         struct ctl_lba_len_flags *lbalen;
5254         uint64_t starting_lba;
5255         uint32_t block_count;
5256         int retval;
5257         uint8_t byte2;
5258
5259         CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
5260
5261         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5262         softc = lun->ctl_softc;
5263         retval = 0;
5264
5265         switch (ctsio->cdb[0]) {
5266         case SYNCHRONIZE_CACHE: {
5267                 struct scsi_sync_cache *cdb;
5268                 cdb = (struct scsi_sync_cache *)ctsio->cdb;
5269
5270                 starting_lba = scsi_4btoul(cdb->begin_lba);
5271                 block_count = scsi_2btoul(cdb->lb_count);
5272                 byte2 = cdb->byte2;
5273                 break;
5274         }
5275         case SYNCHRONIZE_CACHE_16: {
5276                 struct scsi_sync_cache_16 *cdb;
5277                 cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5278
5279                 starting_lba = scsi_8btou64(cdb->begin_lba);
5280                 block_count = scsi_4btoul(cdb->lb_count);
5281                 byte2 = cdb->byte2;
5282                 break;
5283         }
5284         default:
5285                 ctl_set_invalid_opcode(ctsio);
5286                 ctl_done((union ctl_io *)ctsio);
5287                 goto bailout;
5288                 break; /* NOTREACHED */
5289         }
5290
5291         /*
5292          * We check the LBA and length, but don't do anything with them.
5293          * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5294          * get flushed.  This check will just help satisfy anyone who wants
5295          * to see an error for an out of range LBA.
5296          */
5297         if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5298                 ctl_set_lba_out_of_range(ctsio);
5299                 ctl_done((union ctl_io *)ctsio);
5300                 goto bailout;
5301         }
5302
5303         /*
5304          * If this LUN has no backend, we can't flush the cache anyway.
5305          */
5306         if (lun->backend == NULL) {
5307                 ctl_set_invalid_opcode(ctsio);
5308                 ctl_done((union ctl_io *)ctsio);
5309                 goto bailout;
5310         }
5311
5312         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5313         lbalen->lba = starting_lba;
5314         lbalen->len = block_count;
5315         lbalen->flags = byte2;
5316
5317         /*
5318          * Check to see whether we're configured to send the SYNCHRONIZE
5319          * CACHE command directly to the back end.
5320          */
5321         mtx_lock(&lun->lun_lock);
5322         if ((softc->flags & CTL_FLAG_REAL_SYNC)
5323          && (++(lun->sync_count) >= lun->sync_interval)) {
5324                 lun->sync_count = 0;
5325                 mtx_unlock(&lun->lun_lock);
5326                 retval = lun->backend->config_write((union ctl_io *)ctsio);
5327         } else {
5328                 mtx_unlock(&lun->lun_lock);
5329                 ctl_set_success(ctsio);
5330                 ctl_done((union ctl_io *)ctsio);
5331         }
5332
5333 bailout:
5334
5335         return (retval);
5336 }
5337
5338 int
5339 ctl_format(struct ctl_scsiio *ctsio)
5340 {
5341         struct scsi_format *cdb;
5342         struct ctl_lun *lun;
5343         int length, defect_list_len;
5344
5345         CTL_DEBUG_PRINT(("ctl_format\n"));
5346
5347         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5348
5349         cdb = (struct scsi_format *)ctsio->cdb;
5350
5351         length = 0;
5352         if (cdb->byte2 & SF_FMTDATA) {
5353                 if (cdb->byte2 & SF_LONGLIST)
5354                         length = sizeof(struct scsi_format_header_long);
5355                 else
5356                         length = sizeof(struct scsi_format_header_short);
5357         }
5358
5359         if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5360          && (length > 0)) {
5361                 ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5362                 ctsio->kern_data_len = length;
5363                 ctsio->kern_total_len = length;
5364                 ctsio->kern_data_resid = 0;
5365                 ctsio->kern_rel_offset = 0;
5366                 ctsio->kern_sg_entries = 0;
5367                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5368                 ctsio->be_move_done = ctl_config_move_done;
5369                 ctl_datamove((union ctl_io *)ctsio);
5370
5371                 return (CTL_RETVAL_COMPLETE);
5372         }
5373
5374         defect_list_len = 0;
5375
5376         if (cdb->byte2 & SF_FMTDATA) {
5377                 if (cdb->byte2 & SF_LONGLIST) {
5378                         struct scsi_format_header_long *header;
5379
5380                         header = (struct scsi_format_header_long *)
5381                                 ctsio->kern_data_ptr;
5382
5383                         defect_list_len = scsi_4btoul(header->defect_list_len);
5384                         if (defect_list_len != 0) {
5385                                 ctl_set_invalid_field(ctsio,
5386                                                       /*sks_valid*/ 1,
5387                                                       /*command*/ 0,
5388                                                       /*field*/ 2,
5389                                                       /*bit_valid*/ 0,
5390                                                       /*bit*/ 0);
5391                                 goto bailout;
5392                         }
5393                 } else {
5394                         struct scsi_format_header_short *header;
5395
5396                         header = (struct scsi_format_header_short *)
5397                                 ctsio->kern_data_ptr;
5398
5399                         defect_list_len = scsi_2btoul(header->defect_list_len);
5400                         if (defect_list_len != 0) {
5401                                 ctl_set_invalid_field(ctsio,
5402                                                       /*sks_valid*/ 1,
5403                                                       /*command*/ 0,
5404                                                       /*field*/ 2,
5405                                                       /*bit_valid*/ 0,
5406                                                       /*bit*/ 0);
5407                                 goto bailout;
5408                         }
5409                 }
5410         }
5411
5412         /*
5413          * The format command will clear out the "Medium format corrupted"
5414          * status if set by the configuration code.  That status is really
5415          * just a way to notify the host that we have lost the media, and
5416          * get them to issue a command that will basically make them think
5417          * they're blowing away the media.
5418          */
5419         mtx_lock(&lun->lun_lock);
5420         lun->flags &= ~CTL_LUN_INOPERABLE;
5421         mtx_unlock(&lun->lun_lock);
5422
5423         ctl_set_success(ctsio);
5424 bailout:
5425
5426         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5427                 free(ctsio->kern_data_ptr, M_CTL);
5428                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5429         }
5430
5431         ctl_done((union ctl_io *)ctsio);
5432         return (CTL_RETVAL_COMPLETE);
5433 }
5434
5435 int
5436 ctl_read_buffer(struct ctl_scsiio *ctsio)
5437 {
5438         struct scsi_read_buffer *cdb;
5439         struct ctl_lun *lun;
5440         int buffer_offset, len;
5441         static uint8_t descr[4];
5442         static uint8_t echo_descr[4] = { 0 };
5443
5444         CTL_DEBUG_PRINT(("ctl_read_buffer\n"));
5445
5446         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5447         cdb = (struct scsi_read_buffer *)ctsio->cdb;
5448
5449         if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA &&
5450             (cdb->byte2 & RWB_MODE) != RWB_MODE_ECHO_DESCR &&
5451             (cdb->byte2 & RWB_MODE) != RWB_MODE_DESCR) {
5452                 ctl_set_invalid_field(ctsio,
5453                                       /*sks_valid*/ 1,
5454                                       /*command*/ 1,
5455                                       /*field*/ 1,
5456                                       /*bit_valid*/ 1,
5457                                       /*bit*/ 4);
5458                 ctl_done((union ctl_io *)ctsio);
5459                 return (CTL_RETVAL_COMPLETE);
5460         }
5461
5462         len = scsi_3btoul(cdb->length);
5463         buffer_offset = scsi_3btoul(cdb->offset);
5464
5465         if (buffer_offset + len > CTL_WRITE_BUFFER_SIZE) {
5466                 ctl_set_invalid_field(ctsio,
5467                                       /*sks_valid*/ 1,
5468                                       /*command*/ 1,
5469                                       /*field*/ 6,
5470                                       /*bit_valid*/ 0,
5471                                       /*bit*/ 0);
5472                 ctl_done((union ctl_io *)ctsio);
5473                 return (CTL_RETVAL_COMPLETE);
5474         }
5475
5476         if ((cdb->byte2 & RWB_MODE) == RWB_MODE_DESCR) {
5477                 descr[0] = 0;
5478                 scsi_ulto3b(CTL_WRITE_BUFFER_SIZE, &descr[1]);
5479                 ctsio->kern_data_ptr = descr;
5480                 len = min(len, sizeof(descr));
5481         } else if ((cdb->byte2 & RWB_MODE) == RWB_MODE_ECHO_DESCR) {
5482                 ctsio->kern_data_ptr = echo_descr;
5483                 len = min(len, sizeof(echo_descr));
5484         } else {
5485                 if (lun->write_buffer == NULL) {
5486                         lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE,
5487                             M_CTL, M_WAITOK);
5488                 }
5489                 ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5490         }
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         ctl_set_success(ctsio);
5497         ctsio->be_move_done = ctl_config_move_done;
5498         ctl_datamove((union ctl_io *)ctsio);
5499         return (CTL_RETVAL_COMPLETE);
5500 }
5501
5502 int
5503 ctl_write_buffer(struct ctl_scsiio *ctsio)
5504 {
5505         struct scsi_write_buffer *cdb;
5506         struct ctl_lun *lun;
5507         int buffer_offset, len;
5508
5509         CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5510
5511         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5512         cdb = (struct scsi_write_buffer *)ctsio->cdb;
5513
5514         if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA) {
5515                 ctl_set_invalid_field(ctsio,
5516                                       /*sks_valid*/ 1,
5517                                       /*command*/ 1,
5518                                       /*field*/ 1,
5519                                       /*bit_valid*/ 1,
5520                                       /*bit*/ 4);
5521                 ctl_done((union ctl_io *)ctsio);
5522                 return (CTL_RETVAL_COMPLETE);
5523         }
5524
5525         len = scsi_3btoul(cdb->length);
5526         buffer_offset = scsi_3btoul(cdb->offset);
5527
5528         if (buffer_offset + len > CTL_WRITE_BUFFER_SIZE) {
5529                 ctl_set_invalid_field(ctsio,
5530                                       /*sks_valid*/ 1,
5531                                       /*command*/ 1,
5532                                       /*field*/ 6,
5533                                       /*bit_valid*/ 0,
5534                                       /*bit*/ 0);
5535                 ctl_done((union ctl_io *)ctsio);
5536                 return (CTL_RETVAL_COMPLETE);
5537         }
5538
5539         /*
5540          * If we've got a kernel request that hasn't been malloced yet,
5541          * malloc it and tell the caller the data buffer is here.
5542          */
5543         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5544                 if (lun->write_buffer == NULL) {
5545                         lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE,
5546                             M_CTL, M_WAITOK);
5547                 }
5548                 ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5549                 ctsio->kern_data_len = len;
5550                 ctsio->kern_total_len = len;
5551                 ctsio->kern_data_resid = 0;
5552                 ctsio->kern_rel_offset = 0;
5553                 ctsio->kern_sg_entries = 0;
5554                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5555                 ctsio->be_move_done = ctl_config_move_done;
5556                 ctl_datamove((union ctl_io *)ctsio);
5557
5558                 return (CTL_RETVAL_COMPLETE);
5559         }
5560
5561         ctl_set_success(ctsio);
5562         ctl_done((union ctl_io *)ctsio);
5563         return (CTL_RETVAL_COMPLETE);
5564 }
5565
5566 int
5567 ctl_write_same(struct ctl_scsiio *ctsio)
5568 {
5569         struct ctl_lun *lun;
5570         struct ctl_lba_len_flags *lbalen;
5571         uint64_t lba;
5572         uint32_t num_blocks;
5573         int len, retval;
5574         uint8_t byte2;
5575
5576         retval = CTL_RETVAL_COMPLETE;
5577
5578         CTL_DEBUG_PRINT(("ctl_write_same\n"));
5579
5580         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5581
5582         switch (ctsio->cdb[0]) {
5583         case WRITE_SAME_10: {
5584                 struct scsi_write_same_10 *cdb;
5585
5586                 cdb = (struct scsi_write_same_10 *)ctsio->cdb;
5587
5588                 lba = scsi_4btoul(cdb->addr);
5589                 num_blocks = scsi_2btoul(cdb->length);
5590                 byte2 = cdb->byte2;
5591                 break;
5592         }
5593         case WRITE_SAME_16: {
5594                 struct scsi_write_same_16 *cdb;
5595
5596                 cdb = (struct scsi_write_same_16 *)ctsio->cdb;
5597
5598                 lba = scsi_8btou64(cdb->addr);
5599                 num_blocks = scsi_4btoul(cdb->length);
5600                 byte2 = cdb->byte2;
5601                 break;
5602         }
5603         default:
5604                 /*
5605                  * We got a command we don't support.  This shouldn't
5606                  * happen, commands should be filtered out above us.
5607                  */
5608                 ctl_set_invalid_opcode(ctsio);
5609                 ctl_done((union ctl_io *)ctsio);
5610
5611                 return (CTL_RETVAL_COMPLETE);
5612                 break; /* NOTREACHED */
5613         }
5614
5615         /* NDOB and ANCHOR flags can be used only together with UNMAP */
5616         if ((byte2 & SWS_UNMAP) == 0 &&
5617             (byte2 & (SWS_NDOB | SWS_ANCHOR)) != 0) {
5618                 ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
5619                     /*command*/ 1, /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 0);
5620                 ctl_done((union ctl_io *)ctsio);
5621                 return (CTL_RETVAL_COMPLETE);
5622         }
5623
5624         /*
5625          * The first check is to make sure we're in bounds, the second
5626          * check is to catch wrap-around problems.  If the lba + num blocks
5627          * is less than the lba, then we've wrapped around and the block
5628          * range is invalid anyway.
5629          */
5630         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5631          || ((lba + num_blocks) < lba)) {
5632                 ctl_set_lba_out_of_range(ctsio);
5633                 ctl_done((union ctl_io *)ctsio);
5634                 return (CTL_RETVAL_COMPLETE);
5635         }
5636
5637         /* Zero number of blocks means "to the last logical block" */
5638         if (num_blocks == 0) {
5639                 if ((lun->be_lun->maxlba + 1) - lba > UINT32_MAX) {
5640                         ctl_set_invalid_field(ctsio,
5641                                               /*sks_valid*/ 0,
5642                                               /*command*/ 1,
5643                                               /*field*/ 0,
5644                                               /*bit_valid*/ 0,
5645                                               /*bit*/ 0);
5646                         ctl_done((union ctl_io *)ctsio);
5647                         return (CTL_RETVAL_COMPLETE);
5648                 }
5649                 num_blocks = (lun->be_lun->maxlba + 1) - lba;
5650         }
5651
5652         len = lun->be_lun->blocksize;
5653
5654         /*
5655          * If we've got a kernel request that hasn't been malloced yet,
5656          * malloc it and tell the caller the data buffer is here.
5657          */
5658         if ((byte2 & SWS_NDOB) == 0 &&
5659             (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5660                 ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
5661                 ctsio->kern_data_len = len;
5662                 ctsio->kern_total_len = len;
5663                 ctsio->kern_data_resid = 0;
5664                 ctsio->kern_rel_offset = 0;
5665                 ctsio->kern_sg_entries = 0;
5666                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5667                 ctsio->be_move_done = ctl_config_move_done;
5668                 ctl_datamove((union ctl_io *)ctsio);
5669
5670                 return (CTL_RETVAL_COMPLETE);
5671         }
5672
5673         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5674         lbalen->lba = lba;
5675         lbalen->len = num_blocks;
5676         lbalen->flags = byte2;
5677         retval = lun->backend->config_write((union ctl_io *)ctsio);
5678
5679         return (retval);
5680 }
5681
5682 int
5683 ctl_unmap(struct ctl_scsiio *ctsio)
5684 {
5685         struct ctl_lun *lun;
5686         struct scsi_unmap *cdb;
5687         struct ctl_ptr_len_flags *ptrlen;
5688         struct scsi_unmap_header *hdr;
5689         struct scsi_unmap_desc *buf, *end, *endnz, *range;
5690         uint64_t lba;
5691         uint32_t num_blocks;
5692         int len, retval;
5693         uint8_t byte2;
5694
5695         retval = CTL_RETVAL_COMPLETE;
5696
5697         CTL_DEBUG_PRINT(("ctl_unmap\n"));
5698
5699         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5700         cdb = (struct scsi_unmap *)ctsio->cdb;
5701
5702         len = scsi_2btoul(cdb->length);
5703         byte2 = cdb->byte2;
5704
5705         /*
5706          * If we've got a kernel request that hasn't been malloced yet,
5707          * malloc it and tell the caller the data buffer is here.
5708          */
5709         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5710                 ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
5711                 ctsio->kern_data_len = len;
5712                 ctsio->kern_total_len = len;
5713                 ctsio->kern_data_resid = 0;
5714                 ctsio->kern_rel_offset = 0;
5715                 ctsio->kern_sg_entries = 0;
5716                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5717                 ctsio->be_move_done = ctl_config_move_done;
5718                 ctl_datamove((union ctl_io *)ctsio);
5719
5720                 return (CTL_RETVAL_COMPLETE);
5721         }
5722
5723         len = ctsio->kern_total_len - ctsio->kern_data_resid;
5724         hdr = (struct scsi_unmap_header *)ctsio->kern_data_ptr;
5725         if (len < sizeof (*hdr) ||
5726             len < (scsi_2btoul(hdr->length) + sizeof(hdr->length)) ||
5727             len < (scsi_2btoul(hdr->desc_length) + sizeof (*hdr)) ||
5728             scsi_2btoul(hdr->desc_length) % sizeof(*buf) != 0) {
5729                 ctl_set_invalid_field(ctsio,
5730                                       /*sks_valid*/ 0,
5731                                       /*command*/ 0,
5732                                       /*field*/ 0,
5733                                       /*bit_valid*/ 0,
5734                                       /*bit*/ 0);
5735                 goto done;
5736         }
5737         len = scsi_2btoul(hdr->desc_length);
5738         buf = (struct scsi_unmap_desc *)(hdr + 1);
5739         end = buf + len / sizeof(*buf);
5740
5741         endnz = buf;
5742         for (range = buf; range < end; range++) {
5743                 lba = scsi_8btou64(range->lba);
5744                 num_blocks = scsi_4btoul(range->length);
5745                 if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5746                  || ((lba + num_blocks) < lba)) {
5747                         ctl_set_lba_out_of_range(ctsio);
5748                         ctl_done((union ctl_io *)ctsio);
5749                         return (CTL_RETVAL_COMPLETE);
5750                 }
5751                 if (num_blocks != 0)
5752                         endnz = range + 1;
5753         }
5754
5755         /*
5756          * Block backend can not handle zero last range.
5757          * Filter it out and return if there is nothing left.
5758          */
5759         len = (uint8_t *)endnz - (uint8_t *)buf;
5760         if (len == 0) {
5761                 ctl_set_success(ctsio);
5762                 goto done;
5763         }
5764
5765         mtx_lock(&lun->lun_lock);
5766         ptrlen = (struct ctl_ptr_len_flags *)
5767             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5768         ptrlen->ptr = (void *)buf;
5769         ptrlen->len = len;
5770         ptrlen->flags = byte2;
5771         ctl_check_blocked(lun);
5772         mtx_unlock(&lun->lun_lock);
5773
5774         retval = lun->backend->config_write((union ctl_io *)ctsio);
5775         return (retval);
5776
5777 done:
5778         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5779                 free(ctsio->kern_data_ptr, M_CTL);
5780                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5781         }
5782         ctl_done((union ctl_io *)ctsio);
5783         return (CTL_RETVAL_COMPLETE);
5784 }
5785
5786 /*
5787  * Note that this function currently doesn't actually do anything inside
5788  * CTL to enforce things if the DQue bit is turned on.
5789  *
5790  * Also note that this function can't be used in the default case, because
5791  * the DQue bit isn't set in the changeable mask for the control mode page
5792  * anyway.  This is just here as an example for how to implement a page
5793  * handler, and a placeholder in case we want to allow the user to turn
5794  * tagged queueing on and off.
5795  *
5796  * The D_SENSE bit handling is functional, however, and will turn
5797  * descriptor sense on and off for a given LUN.
5798  */
5799 int
5800 ctl_control_page_handler(struct ctl_scsiio *ctsio,
5801                          struct ctl_page_index *page_index, uint8_t *page_ptr)
5802 {
5803         struct scsi_control_page *current_cp, *saved_cp, *user_cp;
5804         struct ctl_lun *lun;
5805         int set_ua;
5806         uint32_t initidx;
5807
5808         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5809         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5810         set_ua = 0;
5811
5812         user_cp = (struct scsi_control_page *)page_ptr;
5813         current_cp = (struct scsi_control_page *)
5814                 (page_index->page_data + (page_index->page_len *
5815                 CTL_PAGE_CURRENT));
5816         saved_cp = (struct scsi_control_page *)
5817                 (page_index->page_data + (page_index->page_len *
5818                 CTL_PAGE_SAVED));
5819
5820         mtx_lock(&lun->lun_lock);
5821         if (((current_cp->rlec & SCP_DSENSE) == 0)
5822          && ((user_cp->rlec & SCP_DSENSE) != 0)) {
5823                 /*
5824                  * Descriptor sense is currently turned off and the user
5825                  * wants to turn it on.
5826                  */
5827                 current_cp->rlec |= SCP_DSENSE;
5828                 saved_cp->rlec |= SCP_DSENSE;
5829                 lun->flags |= CTL_LUN_SENSE_DESC;
5830                 set_ua = 1;
5831         } else if (((current_cp->rlec & SCP_DSENSE) != 0)
5832                 && ((user_cp->rlec & SCP_DSENSE) == 0)) {
5833                 /*
5834                  * Descriptor sense is currently turned on, and the user
5835                  * wants to turn it off.
5836                  */
5837                 current_cp->rlec &= ~SCP_DSENSE;
5838                 saved_cp->rlec &= ~SCP_DSENSE;
5839                 lun->flags &= ~CTL_LUN_SENSE_DESC;
5840                 set_ua = 1;
5841         }
5842         if ((current_cp->queue_flags & SCP_QUEUE_ALG_MASK) !=
5843             (user_cp->queue_flags & SCP_QUEUE_ALG_MASK)) {
5844                 current_cp->queue_flags &= ~SCP_QUEUE_ALG_MASK;
5845                 current_cp->queue_flags |= user_cp->queue_flags & SCP_QUEUE_ALG_MASK;
5846                 saved_cp->queue_flags &= ~SCP_QUEUE_ALG_MASK;
5847                 saved_cp->queue_flags |= user_cp->queue_flags & SCP_QUEUE_ALG_MASK;
5848                 set_ua = 1;
5849         }
5850         if ((current_cp->eca_and_aen & SCP_SWP) !=
5851             (user_cp->eca_and_aen & SCP_SWP)) {
5852                 current_cp->eca_and_aen &= ~SCP_SWP;
5853                 current_cp->eca_and_aen |= user_cp->eca_and_aen & SCP_SWP;
5854                 saved_cp->eca_and_aen &= ~SCP_SWP;
5855                 saved_cp->eca_and_aen |= user_cp->eca_and_aen & SCP_SWP;
5856                 set_ua = 1;
5857         }
5858         if (set_ua != 0)
5859                 ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
5860         mtx_unlock(&lun->lun_lock);
5861
5862         return (0);
5863 }
5864
5865 int
5866 ctl_caching_sp_handler(struct ctl_scsiio *ctsio,
5867                      struct ctl_page_index *page_index, uint8_t *page_ptr)
5868 {
5869         struct scsi_caching_page *current_cp, *saved_cp, *user_cp;
5870         struct ctl_lun *lun;
5871         int set_ua;
5872         uint32_t initidx;
5873
5874         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5875         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5876         set_ua = 0;
5877
5878         user_cp = (struct scsi_caching_page *)page_ptr;
5879         current_cp = (struct scsi_caching_page *)
5880                 (page_index->page_data + (page_index->page_len *
5881                 CTL_PAGE_CURRENT));
5882         saved_cp = (struct scsi_caching_page *)
5883                 (page_index->page_data + (page_index->page_len *
5884                 CTL_PAGE_SAVED));
5885
5886         mtx_lock(&lun->lun_lock);
5887         if ((current_cp->flags1 & (SCP_WCE | SCP_RCD)) !=
5888             (user_cp->flags1 & (SCP_WCE | SCP_RCD))) {
5889                 current_cp->flags1 &= ~(SCP_WCE | SCP_RCD);
5890                 current_cp->flags1 |= user_cp->flags1 & (SCP_WCE | SCP_RCD);
5891                 saved_cp->flags1 &= ~(SCP_WCE | SCP_RCD);
5892                 saved_cp->flags1 |= user_cp->flags1 & (SCP_WCE | SCP_RCD);
5893                 set_ua = 1;
5894         }
5895         if (set_ua != 0)
5896                 ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
5897         mtx_unlock(&lun->lun_lock);
5898
5899         return (0);
5900 }
5901
5902 int
5903 ctl_debugconf_sp_select_handler(struct ctl_scsiio *ctsio,
5904                                 struct ctl_page_index *page_index,
5905                                 uint8_t *page_ptr)
5906 {
5907         uint8_t *c;
5908         int i;
5909
5910         c = ((struct copan_debugconf_subpage *)page_ptr)->ctl_time_io_secs;
5911         ctl_time_io_secs =
5912                 (c[0] << 8) |
5913                 (c[1] << 0) |
5914                 0;
5915         CTL_DEBUG_PRINT(("set ctl_time_io_secs to %d\n", ctl_time_io_secs));
5916         printf("set ctl_time_io_secs to %d\n", ctl_time_io_secs);
5917         printf("page data:");
5918         for (i=0; i<8; i++)
5919                 printf(" %.2x",page_ptr[i]);
5920         printf("\n");
5921         return (0);
5922 }
5923
5924 int
5925 ctl_debugconf_sp_sense_handler(struct ctl_scsiio *ctsio,
5926                                struct ctl_page_index *page_index,
5927                                int pc)
5928 {
5929         struct copan_debugconf_subpage *page;
5930
5931         page = (struct copan_debugconf_subpage *)page_index->page_data +
5932                 (page_index->page_len * pc);
5933
5934         switch (pc) {
5935         case SMS_PAGE_CTRL_CHANGEABLE >> 6:
5936         case SMS_PAGE_CTRL_DEFAULT >> 6:
5937         case SMS_PAGE_CTRL_SAVED >> 6:
5938                 /*
5939                  * We don't update the changable or default bits for this page.
5940                  */
5941                 break;
5942         case SMS_PAGE_CTRL_CURRENT >> 6:
5943                 page->ctl_time_io_secs[0] = ctl_time_io_secs >> 8;
5944                 page->ctl_time_io_secs[1] = ctl_time_io_secs >> 0;
5945                 break;
5946         default:
5947 #ifdef NEEDTOPORT
5948                 EPRINT(0, "Invalid PC %d!!", pc);
5949 #endif /* NEEDTOPORT */
5950                 break;
5951         }
5952         return (0);
5953 }
5954
5955
5956 static int
5957 ctl_do_mode_select(union ctl_io *io)
5958 {
5959         struct scsi_mode_page_header *page_header;
5960         struct ctl_page_index *page_index;
5961         struct ctl_scsiio *ctsio;
5962         int control_dev, page_len;
5963         int page_len_offset, page_len_size;
5964         union ctl_modepage_info *modepage_info;
5965         struct ctl_lun *lun;
5966         int *len_left, *len_used;
5967         int retval, i;
5968
5969         ctsio = &io->scsiio;
5970         page_index = NULL;
5971         page_len = 0;
5972         retval = CTL_RETVAL_COMPLETE;
5973
5974         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5975
5976         if (lun->be_lun->lun_type != T_DIRECT)
5977                 control_dev = 1;
5978         else
5979                 control_dev = 0;
5980
5981         modepage_info = (union ctl_modepage_info *)
5982                 ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
5983         len_left = &modepage_info->header.len_left;
5984         len_used = &modepage_info->header.len_used;
5985
5986 do_next_page:
5987
5988         page_header = (struct scsi_mode_page_header *)
5989                 (ctsio->kern_data_ptr + *len_used);
5990
5991         if (*len_left == 0) {
5992                 free(ctsio->kern_data_ptr, M_CTL);
5993                 ctl_set_success(ctsio);
5994                 ctl_done((union ctl_io *)ctsio);
5995                 return (CTL_RETVAL_COMPLETE);
5996         } else if (*len_left < sizeof(struct scsi_mode_page_header)) {
5997
5998                 free(ctsio->kern_data_ptr, M_CTL);
5999                 ctl_set_param_len_error(ctsio);
6000                 ctl_done((union ctl_io *)ctsio);
6001                 return (CTL_RETVAL_COMPLETE);
6002
6003         } else if ((page_header->page_code & SMPH_SPF)
6004                 && (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
6005
6006                 free(ctsio->kern_data_ptr, M_CTL);
6007                 ctl_set_param_len_error(ctsio);
6008                 ctl_done((union ctl_io *)ctsio);
6009                 return (CTL_RETVAL_COMPLETE);
6010         }
6011
6012
6013         /*
6014          * XXX KDM should we do something with the block descriptor?
6015          */
6016         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6017
6018                 if ((control_dev != 0)
6019                  && (lun->mode_pages.index[i].page_flags &
6020                      CTL_PAGE_FLAG_DISK_ONLY))
6021                         continue;
6022
6023                 if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
6024                     (page_header->page_code & SMPH_PC_MASK))
6025                         continue;
6026
6027                 /*
6028                  * If neither page has a subpage code, then we've got a
6029                  * match.
6030                  */
6031                 if (((lun->mode_pages.index[i].page_code & SMPH_SPF) == 0)
6032                  && ((page_header->page_code & SMPH_SPF) == 0)) {
6033                         page_index = &lun->mode_pages.index[i];
6034                         page_len = page_header->page_length;
6035                         break;
6036                 }
6037
6038                 /*
6039                  * If both pages have subpages, then the subpage numbers
6040                  * have to match.
6041                  */
6042                 if ((lun->mode_pages.index[i].page_code & SMPH_SPF)
6043                   && (page_header->page_code & SMPH_SPF)) {
6044                         struct scsi_mode_page_header_sp *sph;
6045
6046                         sph = (struct scsi_mode_page_header_sp *)page_header;
6047
6048                         if (lun->mode_pages.index[i].subpage ==
6049                             sph->subpage) {
6050                                 page_index = &lun->mode_pages.index[i];
6051                                 page_len = scsi_2btoul(sph->page_length);
6052                                 break;
6053                         }
6054                 }
6055         }
6056
6057         /*
6058          * If we couldn't find the page, or if we don't have a mode select
6059          * handler for it, send back an error to the user.
6060          */
6061         if ((page_index == NULL)
6062          || (page_index->select_handler == NULL)) {
6063                 ctl_set_invalid_field(ctsio,
6064                                       /*sks_valid*/ 1,
6065                                       /*command*/ 0,
6066                                       /*field*/ *len_used,
6067                                       /*bit_valid*/ 0,
6068                                       /*bit*/ 0);
6069                 free(ctsio->kern_data_ptr, M_CTL);
6070                 ctl_done((union ctl_io *)ctsio);
6071                 return (CTL_RETVAL_COMPLETE);
6072         }
6073
6074         if (page_index->page_code & SMPH_SPF) {
6075                 page_len_offset = 2;
6076                 page_len_size = 2;
6077         } else {
6078                 page_len_size = 1;
6079                 page_len_offset = 1;
6080         }
6081
6082         /*
6083          * If the length the initiator gives us isn't the one we specify in
6084          * the mode page header, or if they didn't specify enough data in
6085          * the CDB to avoid truncating this page, kick out the request.
6086          */
6087         if ((page_len != (page_index->page_len - page_len_offset -
6088                           page_len_size))
6089          || (*len_left < page_index->page_len)) {
6090
6091
6092                 ctl_set_invalid_field(ctsio,
6093                                       /*sks_valid*/ 1,
6094                                       /*command*/ 0,
6095                                       /*field*/ *len_used + page_len_offset,
6096                                       /*bit_valid*/ 0,
6097                                       /*bit*/ 0);
6098                 free(ctsio->kern_data_ptr, M_CTL);
6099                 ctl_done((union ctl_io *)ctsio);
6100                 return (CTL_RETVAL_COMPLETE);
6101         }
6102
6103         /*
6104          * Run through the mode page, checking to make sure that the bits
6105          * the user changed are actually legal for him to change.
6106          */
6107         for (i = 0; i < page_index->page_len; i++) {
6108                 uint8_t *user_byte, *change_mask, *current_byte;
6109                 int bad_bit;
6110                 int j;
6111
6112                 user_byte = (uint8_t *)page_header + i;
6113                 change_mask = page_index->page_data +
6114                               (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
6115                 current_byte = page_index->page_data +
6116                                (page_index->page_len * CTL_PAGE_CURRENT) + i;
6117
6118                 /*
6119                  * Check to see whether the user set any bits in this byte
6120                  * that he is not allowed to set.
6121                  */
6122                 if ((*user_byte & ~(*change_mask)) ==
6123                     (*current_byte & ~(*change_mask)))
6124                         continue;
6125
6126                 /*
6127                  * Go through bit by bit to determine which one is illegal.
6128                  */
6129                 bad_bit = 0;
6130                 for (j = 7; j >= 0; j--) {
6131                         if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
6132                             (((1 << i) & ~(*change_mask)) & *current_byte)) {
6133                                 bad_bit = i;
6134                                 break;
6135                         }
6136                 }
6137                 ctl_set_invalid_field(ctsio,
6138                                       /*sks_valid*/ 1,
6139                                       /*command*/ 0,
6140                                       /*field*/ *len_used + i,
6141                                       /*bit_valid*/ 1,
6142                                       /*bit*/ bad_bit);
6143                 free(ctsio->kern_data_ptr, M_CTL);
6144                 ctl_done((union ctl_io *)ctsio);
6145                 return (CTL_RETVAL_COMPLETE);
6146         }
6147
6148         /*
6149          * Decrement these before we call the page handler, since we may
6150          * end up getting called back one way or another before the handler
6151          * returns to this context.
6152          */
6153         *len_left -= page_index->page_len;
6154         *len_used += page_index->page_len;
6155
6156         retval = page_index->select_handler(ctsio, page_index,
6157                                             (uint8_t *)page_header);
6158
6159         /*
6160          * If the page handler returns CTL_RETVAL_QUEUED, then we need to
6161          * wait until this queued command completes to finish processing
6162          * the mode page.  If it returns anything other than
6163          * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
6164          * already set the sense information, freed the data pointer, and
6165          * completed the io for us.
6166          */
6167         if (retval != CTL_RETVAL_COMPLETE)
6168                 goto bailout_no_done;
6169
6170         /*
6171          * If the initiator sent us more than one page, parse the next one.
6172          */
6173         if (*len_left > 0)
6174                 goto do_next_page;
6175
6176         ctl_set_success(ctsio);
6177         free(ctsio->kern_data_ptr, M_CTL);
6178         ctl_done((union ctl_io *)ctsio);
6179
6180 bailout_no_done:
6181
6182         return (CTL_RETVAL_COMPLETE);
6183
6184 }
6185
6186 int
6187 ctl_mode_select(struct ctl_scsiio *ctsio)
6188 {
6189         int param_len, pf, sp;
6190         int header_size, bd_len;
6191         int len_left, len_used;
6192         struct ctl_page_index *page_index;
6193         struct ctl_lun *lun;
6194         int control_dev, page_len;
6195         union ctl_modepage_info *modepage_info;
6196         int retval;
6197
6198         pf = 0;
6199         sp = 0;
6200         page_len = 0;
6201         len_used = 0;
6202         len_left = 0;
6203         retval = 0;
6204         bd_len = 0;
6205         page_index = NULL;
6206
6207         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6208
6209         if (lun->be_lun->lun_type != T_DIRECT)
6210                 control_dev = 1;
6211         else
6212                 control_dev = 0;
6213
6214         switch (ctsio->cdb[0]) {
6215         case MODE_SELECT_6: {
6216                 struct scsi_mode_select_6 *cdb;
6217
6218                 cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
6219
6220                 pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6221                 sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6222
6223                 param_len = cdb->length;
6224                 header_size = sizeof(struct scsi_mode_header_6);
6225                 break;
6226         }
6227         case MODE_SELECT_10: {
6228                 struct scsi_mode_select_10 *cdb;
6229
6230                 cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
6231
6232                 pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6233                 sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6234
6235                 param_len = scsi_2btoul(cdb->length);
6236                 header_size = sizeof(struct scsi_mode_header_10);
6237                 break;
6238         }
6239         default:
6240                 ctl_set_invalid_opcode(ctsio);
6241                 ctl_done((union ctl_io *)ctsio);
6242                 return (CTL_RETVAL_COMPLETE);
6243                 break; /* NOTREACHED */
6244         }
6245
6246         /*
6247          * From SPC-3:
6248          * "A parameter list length of zero indicates that the Data-Out Buffer
6249          * shall be empty. This condition shall not be considered as an error."
6250          */
6251         if (param_len == 0) {
6252                 ctl_set_success(ctsio);
6253                 ctl_done((union ctl_io *)ctsio);
6254                 return (CTL_RETVAL_COMPLETE);
6255         }
6256
6257         /*
6258          * Since we'll hit this the first time through, prior to
6259          * allocation, we don't need to free a data buffer here.
6260          */
6261         if (param_len < header_size) {
6262                 ctl_set_param_len_error(ctsio);
6263                 ctl_done((union ctl_io *)ctsio);
6264                 return (CTL_RETVAL_COMPLETE);
6265         }
6266
6267         /*
6268          * Allocate the data buffer and grab the user's data.  In theory,
6269          * we shouldn't have to sanity check the parameter list length here
6270          * because the maximum size is 64K.  We should be able to malloc
6271          * that much without too many problems.
6272          */
6273         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6274                 ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6275                 ctsio->kern_data_len = param_len;
6276                 ctsio->kern_total_len = param_len;
6277                 ctsio->kern_data_resid = 0;
6278                 ctsio->kern_rel_offset = 0;
6279                 ctsio->kern_sg_entries = 0;
6280                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6281                 ctsio->be_move_done = ctl_config_move_done;
6282                 ctl_datamove((union ctl_io *)ctsio);
6283
6284                 return (CTL_RETVAL_COMPLETE);
6285         }
6286
6287         switch (ctsio->cdb[0]) {
6288         case MODE_SELECT_6: {
6289                 struct scsi_mode_header_6 *mh6;
6290
6291                 mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6292                 bd_len = mh6->blk_desc_len;
6293                 break;
6294         }
6295         case MODE_SELECT_10: {
6296                 struct scsi_mode_header_10 *mh10;
6297
6298                 mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6299                 bd_len = scsi_2btoul(mh10->blk_desc_len);
6300                 break;
6301         }
6302         default:
6303                 panic("Invalid CDB type %#x", ctsio->cdb[0]);
6304                 break;
6305         }
6306
6307         if (param_len < (header_size + bd_len)) {
6308                 free(ctsio->kern_data_ptr, M_CTL);
6309                 ctl_set_param_len_error(ctsio);
6310                 ctl_done((union ctl_io *)ctsio);
6311                 return (CTL_RETVAL_COMPLETE);
6312         }
6313
6314         /*
6315          * Set the IO_CONT flag, so that if this I/O gets passed to
6316          * ctl_config_write_done(), it'll get passed back to
6317          * ctl_do_mode_select() for further processing, or completion if
6318          * we're all done.
6319          */
6320         ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6321         ctsio->io_cont = ctl_do_mode_select;
6322
6323         modepage_info = (union ctl_modepage_info *)
6324                 ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6325
6326         memset(modepage_info, 0, sizeof(*modepage_info));
6327
6328         len_left = param_len - header_size - bd_len;
6329         len_used = header_size + bd_len;
6330
6331         modepage_info->header.len_left = len_left;
6332         modepage_info->header.len_used = len_used;
6333
6334         return (ctl_do_mode_select((union ctl_io *)ctsio));
6335 }
6336
6337 int
6338 ctl_mode_sense(struct ctl_scsiio *ctsio)
6339 {
6340         struct ctl_lun *lun;
6341         int pc, page_code, dbd, llba, subpage;
6342         int alloc_len, page_len, header_len, total_len;
6343         struct scsi_mode_block_descr *block_desc;
6344         struct ctl_page_index *page_index;
6345         int control_dev;
6346
6347         dbd = 0;
6348         llba = 0;
6349         block_desc = NULL;
6350         page_index = NULL;
6351
6352         CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6353
6354         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6355
6356         if (lun->be_lun->lun_type != T_DIRECT)
6357                 control_dev = 1;
6358         else
6359                 control_dev = 0;
6360
6361         switch (ctsio->cdb[0]) {
6362         case MODE_SENSE_6: {
6363                 struct scsi_mode_sense_6 *cdb;
6364
6365                 cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6366
6367                 header_len = sizeof(struct scsi_mode_hdr_6);
6368                 if (cdb->byte2 & SMS_DBD)
6369                         dbd = 1;
6370                 else
6371                         header_len += sizeof(struct scsi_mode_block_descr);
6372
6373                 pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6374                 page_code = cdb->page & SMS_PAGE_CODE;
6375                 subpage = cdb->subpage;
6376                 alloc_len = cdb->length;
6377                 break;
6378         }
6379         case MODE_SENSE_10: {
6380                 struct scsi_mode_sense_10 *cdb;
6381
6382                 cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6383
6384                 header_len = sizeof(struct scsi_mode_hdr_10);
6385
6386                 if (cdb->byte2 & SMS_DBD)
6387                         dbd = 1;
6388                 else
6389                         header_len += sizeof(struct scsi_mode_block_descr);
6390                 if (cdb->byte2 & SMS10_LLBAA)
6391                         llba = 1;
6392                 pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6393                 page_code = cdb->page & SMS_PAGE_CODE;
6394                 subpage = cdb->subpage;
6395                 alloc_len = scsi_2btoul(cdb->length);
6396                 break;
6397         }
6398         default:
6399                 ctl_set_invalid_opcode(ctsio);
6400                 ctl_done((union ctl_io *)ctsio);
6401                 return (CTL_RETVAL_COMPLETE);
6402                 break; /* NOTREACHED */
6403         }
6404
6405         /*
6406          * We have to make a first pass through to calculate the size of
6407          * the pages that match the user's query.  Then we allocate enough
6408          * memory to hold it, and actually copy the data into the buffer.
6409          */
6410         switch (page_code) {
6411         case SMS_ALL_PAGES_PAGE: {
6412                 int i;
6413
6414                 page_len = 0;
6415
6416                 /*
6417                  * At the moment, values other than 0 and 0xff here are
6418                  * reserved according to SPC-3.
6419                  */
6420                 if ((subpage != SMS_SUBPAGE_PAGE_0)
6421                  && (subpage != SMS_SUBPAGE_ALL)) {
6422                         ctl_set_invalid_field(ctsio,
6423                                               /*sks_valid*/ 1,
6424                                               /*command*/ 1,
6425                                               /*field*/ 3,
6426                                               /*bit_valid*/ 0,
6427                                               /*bit*/ 0);
6428                         ctl_done((union ctl_io *)ctsio);
6429                         return (CTL_RETVAL_COMPLETE);
6430                 }
6431
6432                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6433                         if ((control_dev != 0)
6434                          && (lun->mode_pages.index[i].page_flags &
6435                              CTL_PAGE_FLAG_DISK_ONLY))
6436                                 continue;
6437
6438                         /*
6439                          * We don't use this subpage if the user didn't
6440                          * request all subpages.
6441                          */
6442                         if ((lun->mode_pages.index[i].subpage != 0)
6443                          && (subpage == SMS_SUBPAGE_PAGE_0))
6444                                 continue;
6445
6446 #if 0
6447                         printf("found page %#x len %d\n",
6448                                lun->mode_pages.index[i].page_code &
6449                                SMPH_PC_MASK,
6450                                lun->mode_pages.index[i].page_len);
6451 #endif
6452                         page_len += lun->mode_pages.index[i].page_len;
6453                 }
6454                 break;
6455         }
6456         default: {
6457                 int i;
6458
6459                 page_len = 0;
6460
6461                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6462                         /* Look for the right page code */
6463                         if ((lun->mode_pages.index[i].page_code &
6464                              SMPH_PC_MASK) != page_code)
6465                                 continue;
6466
6467                         /* Look for the right subpage or the subpage wildcard*/
6468                         if ((lun->mode_pages.index[i].subpage != subpage)
6469                          && (subpage != SMS_SUBPAGE_ALL))
6470                                 continue;
6471
6472                         /* Make sure the page is supported for this dev type */
6473                         if ((control_dev != 0)
6474                          && (lun->mode_pages.index[i].page_flags &
6475                              CTL_PAGE_FLAG_DISK_ONLY))
6476                                 continue;
6477
6478 #if 0
6479                         printf("found page %#x len %d\n",
6480                                lun->mode_pages.index[i].page_code &
6481                                SMPH_PC_MASK,
6482                                lun->mode_pages.index[i].page_len);
6483 #endif
6484
6485                         page_len += lun->mode_pages.index[i].page_len;
6486                 }
6487
6488                 if (page_len == 0) {
6489                         ctl_set_invalid_field(ctsio,
6490                                               /*sks_valid*/ 1,
6491                                               /*command*/ 1,
6492                                               /*field*/ 2,
6493                                               /*bit_valid*/ 1,
6494                                               /*bit*/ 5);
6495                         ctl_done((union ctl_io *)ctsio);
6496                         return (CTL_RETVAL_COMPLETE);
6497                 }
6498                 break;
6499         }
6500         }
6501
6502         total_len = header_len + page_len;
6503 #if 0
6504         printf("header_len = %d, page_len = %d, total_len = %d\n",
6505                header_len, page_len, total_len);
6506 #endif
6507
6508         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6509         ctsio->kern_sg_entries = 0;
6510         ctsio->kern_data_resid = 0;
6511         ctsio->kern_rel_offset = 0;
6512         if (total_len < alloc_len) {
6513                 ctsio->residual = alloc_len - total_len;
6514                 ctsio->kern_data_len = total_len;
6515                 ctsio->kern_total_len = total_len;
6516         } else {
6517                 ctsio->residual = 0;
6518                 ctsio->kern_data_len = alloc_len;
6519                 ctsio->kern_total_len = alloc_len;
6520         }
6521
6522         switch (ctsio->cdb[0]) {
6523         case MODE_SENSE_6: {
6524                 struct scsi_mode_hdr_6 *header;
6525
6526                 header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
6527
6528                 header->datalen = MIN(total_len - 1, 254);
6529                 if (control_dev == 0) {
6530                         header->dev_specific = 0x10; /* DPOFUA */
6531                         if ((lun->be_lun->flags & CTL_LUN_FLAG_READONLY) ||
6532                             (lun->mode_pages.control_page[CTL_PAGE_CURRENT]
6533                             .eca_and_aen & SCP_SWP) != 0)
6534                                     header->dev_specific |= 0x80; /* WP */
6535                 }
6536                 if (dbd)
6537                         header->block_descr_len = 0;
6538                 else
6539                         header->block_descr_len =
6540                                 sizeof(struct scsi_mode_block_descr);
6541                 block_desc = (struct scsi_mode_block_descr *)&header[1];
6542                 break;
6543         }
6544         case MODE_SENSE_10: {
6545                 struct scsi_mode_hdr_10 *header;
6546                 int datalen;
6547
6548                 header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
6549
6550                 datalen = MIN(total_len - 2, 65533);
6551                 scsi_ulto2b(datalen, header->datalen);
6552                 if (control_dev == 0) {
6553                         header->dev_specific = 0x10; /* DPOFUA */
6554                         if ((lun->be_lun->flags & CTL_LUN_FLAG_READONLY) ||
6555                             (lun->mode_pages.control_page[CTL_PAGE_CURRENT]
6556                             .eca_and_aen & SCP_SWP) != 0)
6557                                     header->dev_specific |= 0x80; /* WP */
6558                 }
6559                 if (dbd)
6560                         scsi_ulto2b(0, header->block_descr_len);
6561                 else
6562                         scsi_ulto2b(sizeof(struct scsi_mode_block_descr),
6563                                     header->block_descr_len);
6564                 block_desc = (struct scsi_mode_block_descr *)&header[1];
6565                 break;
6566         }
6567         default:
6568                 panic("invalid CDB type %#x", ctsio->cdb[0]);
6569                 break; /* NOTREACHED */
6570         }
6571
6572         /*
6573          * If we've got a disk, use its blocksize in the block
6574          * descriptor.  Otherwise, just set it to 0.
6575          */
6576         if (dbd == 0) {
6577                 if (control_dev == 0)
6578                         scsi_ulto3b(lun->be_lun->blocksize,
6579                                     block_desc->block_len);
6580                 else
6581                         scsi_ulto3b(0, block_desc->block_len);
6582         }
6583
6584         switch (page_code) {
6585         case SMS_ALL_PAGES_PAGE: {
6586                 int i, data_used;
6587
6588                 data_used = header_len;
6589                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6590                         struct ctl_page_index *page_index;
6591
6592                         page_index = &lun->mode_pages.index[i];
6593
6594                         if ((control_dev != 0)
6595                          && (page_index->page_flags &
6596                             CTL_PAGE_FLAG_DISK_ONLY))
6597                                 continue;
6598
6599                         /*
6600                          * We don't use this subpage if the user didn't
6601                          * request all subpages.  We already checked (above)
6602                          * to make sure the user only specified a subpage
6603                          * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case.
6604                          */
6605                         if ((page_index->subpage != 0)
6606                          && (subpage == SMS_SUBPAGE_PAGE_0))
6607                                 continue;
6608
6609                         /*
6610                          * Call the handler, if it exists, to update the
6611                          * page to the latest values.
6612                          */
6613                         if (page_index->sense_handler != NULL)
6614                                 page_index->sense_handler(ctsio, page_index,pc);
6615
6616                         memcpy(ctsio->kern_data_ptr + data_used,
6617                                page_index->page_data +
6618                                (page_index->page_len * pc),
6619                                page_index->page_len);
6620                         data_used += page_index->page_len;
6621                 }
6622                 break;
6623         }
6624         default: {
6625                 int i, data_used;
6626
6627                 data_used = header_len;
6628
6629                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6630                         struct ctl_page_index *page_index;
6631
6632                         page_index = &lun->mode_pages.index[i];
6633
6634                         /* Look for the right page code */
6635                         if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6636                                 continue;
6637
6638                         /* Look for the right subpage or the subpage wildcard*/
6639                         if ((page_index->subpage != subpage)
6640                          && (subpage != SMS_SUBPAGE_ALL))
6641                                 continue;
6642
6643                         /* Make sure the page is supported for this dev type */
6644                         if ((control_dev != 0)
6645                          && (page_index->page_flags &
6646                              CTL_PAGE_FLAG_DISK_ONLY))
6647                                 continue;
6648
6649                         /*
6650                          * Call the handler, if it exists, to update the
6651                          * page to the latest values.
6652                          */
6653                         if (page_index->sense_handler != NULL)
6654                                 page_index->sense_handler(ctsio, page_index,pc);
6655
6656                         memcpy(ctsio->kern_data_ptr + data_used,
6657                                page_index->page_data +
6658                                (page_index->page_len * pc),
6659                                page_index->page_len);
6660                         data_used += page_index->page_len;
6661                 }
6662                 break;
6663         }
6664         }
6665
6666         ctl_set_success(ctsio);
6667         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6668         ctsio->be_move_done = ctl_config_move_done;
6669         ctl_datamove((union ctl_io *)ctsio);
6670         return (CTL_RETVAL_COMPLETE);
6671 }
6672
6673 int
6674 ctl_lbp_log_sense_handler(struct ctl_scsiio *ctsio,
6675                                struct ctl_page_index *page_index,
6676                                int pc)
6677 {
6678         struct ctl_lun *lun;
6679         struct scsi_log_param_header *phdr;
6680         uint8_t *data;
6681         uint64_t val;
6682
6683         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6684         data = page_index->page_data;
6685
6686         if (lun->backend->lun_attr != NULL &&
6687             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "blocksavail"))
6688              != UINT64_MAX) {
6689                 phdr = (struct scsi_log_param_header *)data;
6690                 scsi_ulto2b(0x0001, phdr->param_code);
6691                 phdr->param_control = SLP_LBIN | SLP_LP;
6692                 phdr->param_len = 8;
6693                 data = (uint8_t *)(phdr + 1);
6694                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6695                 data[4] = 0x02; /* per-pool */
6696                 data += phdr->param_len;
6697         }
6698
6699         if (lun->backend->lun_attr != NULL &&
6700             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "blocksused"))
6701              != UINT64_MAX) {
6702                 phdr = (struct scsi_log_param_header *)data;
6703                 scsi_ulto2b(0x0002, phdr->param_code);
6704                 phdr->param_control = SLP_LBIN | SLP_LP;
6705                 phdr->param_len = 8;
6706                 data = (uint8_t *)(phdr + 1);
6707                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6708                 data[4] = 0x01; /* per-LUN */
6709                 data += phdr->param_len;
6710         }
6711
6712         if (lun->backend->lun_attr != NULL &&
6713             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "poolblocksavail"))
6714              != UINT64_MAX) {
6715                 phdr = (struct scsi_log_param_header *)data;
6716                 scsi_ulto2b(0x00f1, phdr->param_code);
6717                 phdr->param_control = SLP_LBIN | SLP_LP;
6718                 phdr->param_len = 8;
6719                 data = (uint8_t *)(phdr + 1);
6720                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6721                 data[4] = 0x02; /* per-pool */
6722                 data += phdr->param_len;
6723         }
6724
6725         if (lun->backend->lun_attr != NULL &&
6726             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "poolblocksused"))
6727              != UINT64_MAX) {
6728                 phdr = (struct scsi_log_param_header *)data;
6729                 scsi_ulto2b(0x00f2, phdr->param_code);
6730                 phdr->param_control = SLP_LBIN | SLP_LP;
6731                 phdr->param_len = 8;
6732                 data = (uint8_t *)(phdr + 1);
6733                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6734                 data[4] = 0x02; /* per-pool */
6735                 data += phdr->param_len;
6736         }
6737
6738         page_index->page_len = data - page_index->page_data;
6739         return (0);
6740 }
6741
6742 int
6743 ctl_sap_log_sense_handler(struct ctl_scsiio *ctsio,
6744                                struct ctl_page_index *page_index,
6745                                int pc)
6746 {
6747         struct ctl_lun *lun;
6748         struct stat_page *data;
6749         uint64_t rn, wn, rb, wb;
6750         struct bintime rt, wt;
6751         int i;
6752
6753         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6754         data = (struct stat_page *)page_index->page_data;
6755
6756         scsi_ulto2b(SLP_SAP, data->sap.hdr.param_code);
6757         data->sap.hdr.param_control = SLP_LBIN;
6758         data->sap.hdr.param_len = sizeof(struct scsi_log_stat_and_perf) -
6759             sizeof(struct scsi_log_param_header);
6760         rn = wn = rb = wb = 0;
6761         bintime_clear(&rt);
6762         bintime_clear(&wt);
6763         for (i = 0; i < CTL_MAX_PORTS; i++) {
6764                 rn += lun->stats.ports[i].operations[CTL_STATS_READ];
6765                 wn += lun->stats.ports[i].operations[CTL_STATS_WRITE];
6766                 rb += lun->stats.ports[i].bytes[CTL_STATS_READ];
6767                 wb += lun->stats.ports[i].bytes[CTL_STATS_WRITE];
6768                 bintime_add(&rt, &lun->stats.ports[i].time[CTL_STATS_READ]);
6769                 bintime_add(&wt, &lun->stats.ports[i].time[CTL_STATS_WRITE]);
6770         }
6771         scsi_u64to8b(rn, data->sap.read_num);
6772         scsi_u64to8b(wn, data->sap.write_num);
6773         if (lun->stats.blocksize > 0) {
6774                 scsi_u64to8b(wb / lun->stats.blocksize,
6775                     data->sap.recvieved_lba);
6776                 scsi_u64to8b(rb / lun->stats.blocksize,
6777                     data->sap.transmitted_lba);
6778         }
6779         scsi_u64to8b((uint64_t)rt.sec * 1000 + rt.frac / (UINT64_MAX / 1000),
6780             data->sap.read_int);
6781         scsi_u64to8b((uint64_t)wt.sec * 1000 + wt.frac / (UINT64_MAX / 1000),
6782             data->sap.write_int);
6783         scsi_u64to8b(0, data->sap.weighted_num);
6784         scsi_u64to8b(0, data->sap.weighted_int);
6785         scsi_ulto2b(SLP_IT, data->it.hdr.param_code);
6786         data->it.hdr.param_control = SLP_LBIN;
6787         data->it.hdr.param_len = sizeof(struct scsi_log_idle_time) -
6788             sizeof(struct scsi_log_param_header);
6789 #ifdef CTL_TIME_IO
6790         scsi_u64to8b(lun->idle_time / SBT_1MS, data->it.idle_int);
6791 #endif
6792         scsi_ulto2b(SLP_TI, data->ti.hdr.param_code);
6793         data->it.hdr.param_control = SLP_LBIN;
6794         data->ti.hdr.param_len = sizeof(struct scsi_log_time_interval) -
6795             sizeof(struct scsi_log_param_header);
6796         scsi_ulto4b(3, data->ti.exponent);
6797         scsi_ulto4b(1, data->ti.integer);
6798
6799         page_index->page_len = sizeof(*data);
6800         return (0);
6801 }
6802
6803 int
6804 ctl_log_sense(struct ctl_scsiio *ctsio)
6805 {
6806         struct ctl_lun *lun;
6807         int i, pc, page_code, subpage;
6808         int alloc_len, total_len;
6809         struct ctl_page_index *page_index;
6810         struct scsi_log_sense *cdb;
6811         struct scsi_log_header *header;
6812
6813         CTL_DEBUG_PRINT(("ctl_log_sense\n"));
6814
6815         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6816         cdb = (struct scsi_log_sense *)ctsio->cdb;
6817         pc = (cdb->page & SLS_PAGE_CTRL_MASK) >> 6;
6818         page_code = cdb->page & SLS_PAGE_CODE;
6819         subpage = cdb->subpage;
6820         alloc_len = scsi_2btoul(cdb->length);
6821
6822         page_index = NULL;
6823         for (i = 0; i < CTL_NUM_LOG_PAGES; i++) {
6824                 page_index = &lun->log_pages.index[i];
6825
6826                 /* Look for the right page code */
6827                 if ((page_index->page_code & SL_PAGE_CODE) != page_code)
6828                         continue;
6829
6830                 /* Look for the right subpage or the subpage wildcard*/
6831                 if (page_index->subpage != subpage)
6832                         continue;
6833
6834                 break;
6835         }
6836         if (i >= CTL_NUM_LOG_PAGES) {
6837                 ctl_set_invalid_field(ctsio,
6838                                       /*sks_valid*/ 1,
6839                                       /*command*/ 1,
6840                                       /*field*/ 2,
6841                                       /*bit_valid*/ 0,
6842                                       /*bit*/ 0);
6843                 ctl_done((union ctl_io *)ctsio);
6844                 return (CTL_RETVAL_COMPLETE);
6845         }
6846
6847         total_len = sizeof(struct scsi_log_header) + page_index->page_len;
6848
6849         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6850         ctsio->kern_sg_entries = 0;
6851         ctsio->kern_data_resid = 0;
6852         ctsio->kern_rel_offset = 0;
6853         if (total_len < alloc_len) {
6854                 ctsio->residual = alloc_len - total_len;
6855                 ctsio->kern_data_len = total_len;
6856                 ctsio->kern_total_len = total_len;
6857         } else {
6858                 ctsio->residual = 0;
6859                 ctsio->kern_data_len = alloc_len;
6860                 ctsio->kern_total_len = alloc_len;
6861         }
6862
6863         header = (struct scsi_log_header *)ctsio->kern_data_ptr;
6864         header->page = page_index->page_code;
6865         if (page_index->subpage) {
6866                 header->page |= SL_SPF;
6867                 header->subpage = page_index->subpage;
6868         }
6869         scsi_ulto2b(page_index->page_len, header->datalen);
6870
6871         /*
6872          * Call the handler, if it exists, to update the
6873          * page to the latest values.
6874          */
6875         if (page_index->sense_handler != NULL)
6876                 page_index->sense_handler(ctsio, page_index, pc);
6877
6878         memcpy(header + 1, page_index->page_data, page_index->page_len);
6879
6880         ctl_set_success(ctsio);
6881         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6882         ctsio->be_move_done = ctl_config_move_done;
6883         ctl_datamove((union ctl_io *)ctsio);
6884         return (CTL_RETVAL_COMPLETE);
6885 }
6886
6887 int
6888 ctl_read_capacity(struct ctl_scsiio *ctsio)
6889 {
6890         struct scsi_read_capacity *cdb;
6891         struct scsi_read_capacity_data *data;
6892         struct ctl_lun *lun;
6893         uint32_t lba;
6894
6895         CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
6896
6897         cdb = (struct scsi_read_capacity *)ctsio->cdb;
6898
6899         lba = scsi_4btoul(cdb->addr);
6900         if (((cdb->pmi & SRC_PMI) == 0)
6901          && (lba != 0)) {
6902                 ctl_set_invalid_field(/*ctsio*/ ctsio,
6903                                       /*sks_valid*/ 1,
6904                                       /*command*/ 1,
6905                                       /*field*/ 2,
6906                                       /*bit_valid*/ 0,
6907                                       /*bit*/ 0);
6908                 ctl_done((union ctl_io *)ctsio);
6909                 return (CTL_RETVAL_COMPLETE);
6910         }
6911
6912         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6913
6914         ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
6915         data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
6916         ctsio->residual = 0;
6917         ctsio->kern_data_len = sizeof(*data);
6918         ctsio->kern_total_len = sizeof(*data);
6919         ctsio->kern_data_resid = 0;
6920         ctsio->kern_rel_offset = 0;
6921         ctsio->kern_sg_entries = 0;
6922
6923         /*
6924          * If the maximum LBA is greater than 0xfffffffe, the user must
6925          * issue a SERVICE ACTION IN (16) command, with the read capacity
6926          * serivce action set.
6927          */
6928         if (lun->be_lun->maxlba > 0xfffffffe)
6929                 scsi_ulto4b(0xffffffff, data->addr);
6930         else
6931                 scsi_ulto4b(lun->be_lun->maxlba, data->addr);
6932
6933         /*
6934          * XXX KDM this may not be 512 bytes...
6935          */
6936         scsi_ulto4b(lun->be_lun->blocksize, data->length);
6937
6938         ctl_set_success(ctsio);
6939         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6940         ctsio->be_move_done = ctl_config_move_done;
6941         ctl_datamove((union ctl_io *)ctsio);
6942         return (CTL_RETVAL_COMPLETE);
6943 }
6944
6945 int
6946 ctl_read_capacity_16(struct ctl_scsiio *ctsio)
6947 {
6948         struct scsi_read_capacity_16 *cdb;
6949         struct scsi_read_capacity_data_long *data;
6950         struct ctl_lun *lun;
6951         uint64_t lba;
6952         uint32_t alloc_len;
6953
6954         CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
6955
6956         cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
6957
6958         alloc_len = scsi_4btoul(cdb->alloc_len);
6959         lba = scsi_8btou64(cdb->addr);
6960
6961         if ((cdb->reladr & SRC16_PMI)
6962          && (lba != 0)) {
6963                 ctl_set_invalid_field(/*ctsio*/ ctsio,
6964                                       /*sks_valid*/ 1,
6965                                       /*command*/ 1,
6966                                       /*field*/ 2,
6967                                       /*bit_valid*/ 0,
6968                                       /*bit*/ 0);
6969                 ctl_done((union ctl_io *)ctsio);
6970                 return (CTL_RETVAL_COMPLETE);
6971         }
6972
6973         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6974
6975         ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
6976         data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
6977
6978         if (sizeof(*data) < alloc_len) {
6979                 ctsio->residual = alloc_len - sizeof(*data);
6980                 ctsio->kern_data_len = sizeof(*data);
6981                 ctsio->kern_total_len = sizeof(*data);
6982         } else {
6983                 ctsio->residual = 0;
6984                 ctsio->kern_data_len = alloc_len;
6985                 ctsio->kern_total_len = alloc_len;
6986         }
6987         ctsio->kern_data_resid = 0;
6988         ctsio->kern_rel_offset = 0;
6989         ctsio->kern_sg_entries = 0;
6990
6991         scsi_u64to8b(lun->be_lun->maxlba, data->addr);
6992         /* XXX KDM this may not be 512 bytes... */
6993         scsi_ulto4b(lun->be_lun->blocksize, data->length);
6994         data->prot_lbppbe = lun->be_lun->pblockexp & SRC16_LBPPBE;
6995         scsi_ulto2b(lun->be_lun->pblockoff & SRC16_LALBA_A, data->lalba_lbp);
6996         if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
6997                 data->lalba_lbp[0] |= SRC16_LBPME | SRC16_LBPRZ;
6998
6999         ctl_set_success(ctsio);
7000         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7001         ctsio->be_move_done = ctl_config_move_done;
7002         ctl_datamove((union ctl_io *)ctsio);
7003         return (CTL_RETVAL_COMPLETE);
7004 }
7005
7006 int
7007 ctl_get_lba_status(struct ctl_scsiio *ctsio)
7008 {
7009         struct scsi_get_lba_status *cdb;
7010         struct scsi_get_lba_status_data *data;
7011         struct ctl_lun *lun;
7012         struct ctl_lba_len_flags *lbalen;
7013         uint64_t lba;
7014         uint32_t alloc_len, total_len;
7015         int retval;
7016
7017         CTL_DEBUG_PRINT(("ctl_get_lba_status\n"));
7018
7019         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7020         cdb = (struct scsi_get_lba_status *)ctsio->cdb;
7021         lba = scsi_8btou64(cdb->addr);
7022         alloc_len = scsi_4btoul(cdb->alloc_len);
7023
7024         if (lba > lun->be_lun->maxlba) {
7025                 ctl_set_lba_out_of_range(ctsio);
7026                 ctl_done((union ctl_io *)ctsio);
7027                 return (CTL_RETVAL_COMPLETE);
7028         }
7029
7030         total_len = sizeof(*data) + sizeof(data->descr[0]);
7031         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7032         data = (struct scsi_get_lba_status_data *)ctsio->kern_data_ptr;
7033
7034         if (total_len < alloc_len) {
7035                 ctsio->residual = alloc_len - total_len;
7036                 ctsio->kern_data_len = total_len;
7037                 ctsio->kern_total_len = total_len;
7038         } else {
7039                 ctsio->residual = 0;
7040                 ctsio->kern_data_len = alloc_len;
7041                 ctsio->kern_total_len = alloc_len;
7042         }
7043         ctsio->kern_data_resid = 0;
7044         ctsio->kern_rel_offset = 0;
7045         ctsio->kern_sg_entries = 0;
7046
7047         /* Fill dummy data in case backend can't tell anything. */
7048         scsi_ulto4b(4 + sizeof(data->descr[0]), data->length);
7049         scsi_u64to8b(lba, data->descr[0].addr);
7050         scsi_ulto4b(MIN(UINT32_MAX, lun->be_lun->maxlba + 1 - lba),
7051             data->descr[0].length);
7052         data->descr[0].status = 0; /* Mapped or unknown. */
7053
7054         ctl_set_success(ctsio);
7055         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7056         ctsio->be_move_done = ctl_config_move_done;
7057
7058         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
7059         lbalen->lba = lba;
7060         lbalen->len = total_len;
7061         lbalen->flags = 0;
7062         retval = lun->backend->config_read((union ctl_io *)ctsio);
7063         return (CTL_RETVAL_COMPLETE);
7064 }
7065
7066 int
7067 ctl_read_defect(struct ctl_scsiio *ctsio)
7068 {
7069         struct scsi_read_defect_data_10 *ccb10;
7070         struct scsi_read_defect_data_12 *ccb12;
7071         struct scsi_read_defect_data_hdr_10 *data10;
7072         struct scsi_read_defect_data_hdr_12 *data12;
7073         uint32_t alloc_len, data_len;
7074         uint8_t format;
7075
7076         CTL_DEBUG_PRINT(("ctl_read_defect\n"));
7077
7078         if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
7079                 ccb10 = (struct scsi_read_defect_data_10 *)&ctsio->cdb;
7080                 format = ccb10->format;
7081                 alloc_len = scsi_2btoul(ccb10->alloc_length);
7082                 data_len = sizeof(*data10);
7083         } else {
7084                 ccb12 = (struct scsi_read_defect_data_12 *)&ctsio->cdb;
7085                 format = ccb12->format;
7086                 alloc_len = scsi_4btoul(ccb12->alloc_length);
7087                 data_len = sizeof(*data12);
7088         }
7089         if (alloc_len == 0) {
7090                 ctl_set_success(ctsio);
7091                 ctl_done((union ctl_io *)ctsio);
7092                 return (CTL_RETVAL_COMPLETE);
7093         }
7094
7095         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
7096         if (data_len < alloc_len) {
7097                 ctsio->residual = alloc_len - data_len;
7098                 ctsio->kern_data_len = data_len;
7099                 ctsio->kern_total_len = data_len;
7100         } else {
7101                 ctsio->residual = 0;
7102                 ctsio->kern_data_len = alloc_len;
7103                 ctsio->kern_total_len = alloc_len;
7104         }
7105         ctsio->kern_data_resid = 0;
7106         ctsio->kern_rel_offset = 0;
7107         ctsio->kern_sg_entries = 0;
7108
7109         if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
7110                 data10 = (struct scsi_read_defect_data_hdr_10 *)
7111                     ctsio->kern_data_ptr;
7112                 data10->format = format;
7113                 scsi_ulto2b(0, data10->length);
7114         } else {
7115                 data12 = (struct scsi_read_defect_data_hdr_12 *)
7116                     ctsio->kern_data_ptr;
7117                 data12->format = format;
7118                 scsi_ulto2b(0, data12->generation);
7119                 scsi_ulto4b(0, data12->length);
7120         }
7121
7122         ctl_set_success(ctsio);
7123         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7124         ctsio->be_move_done = ctl_config_move_done;
7125         ctl_datamove((union ctl_io *)ctsio);
7126         return (CTL_RETVAL_COMPLETE);
7127 }
7128
7129 int
7130 ctl_report_tagret_port_groups(struct ctl_scsiio *ctsio)
7131 {
7132         struct scsi_maintenance_in *cdb;
7133         int retval;
7134         int alloc_len, ext, total_len = 0, g, pc, pg, gs, os;
7135         int num_target_port_groups, num_target_ports;
7136         struct ctl_lun *lun;
7137         struct ctl_softc *softc;
7138         struct ctl_port *port;
7139         struct scsi_target_group_data *rtg_ptr;
7140         struct scsi_target_group_data_extended *rtg_ext_ptr;
7141         struct scsi_target_port_group_descriptor *tpg_desc;
7142
7143         CTL_DEBUG_PRINT(("ctl_report_tagret_port_groups\n"));
7144
7145         cdb = (struct scsi_maintenance_in *)ctsio->cdb;
7146         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7147         softc = lun->ctl_softc;
7148
7149         retval = CTL_RETVAL_COMPLETE;
7150
7151         switch (cdb->byte2 & STG_PDF_MASK) {
7152         case STG_PDF_LENGTH:
7153                 ext = 0;
7154                 break;
7155         case STG_PDF_EXTENDED:
7156                 ext = 1;
7157                 break;
7158         default:
7159                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7160                                       /*sks_valid*/ 1,
7161                                       /*command*/ 1,
7162                                       /*field*/ 2,
7163                                       /*bit_valid*/ 1,
7164                                       /*bit*/ 5);
7165                 ctl_done((union ctl_io *)ctsio);
7166                 return(retval);
7167         }
7168
7169         if (softc->is_single)
7170                 num_target_port_groups = 1;
7171         else
7172                 num_target_port_groups = NUM_TARGET_PORT_GROUPS;
7173         num_target_ports = 0;
7174         mtx_lock(&softc->ctl_lock);
7175         STAILQ_FOREACH(port, &softc->port_list, links) {
7176                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7177                         continue;
7178                 if (ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
7179                         continue;
7180                 num_target_ports++;
7181         }
7182         mtx_unlock(&softc->ctl_lock);
7183
7184         if (ext)
7185                 total_len = sizeof(struct scsi_target_group_data_extended);
7186         else
7187                 total_len = sizeof(struct scsi_target_group_data);
7188         total_len += sizeof(struct scsi_target_port_group_descriptor) *
7189                 num_target_port_groups +
7190             sizeof(struct scsi_target_port_descriptor) * num_target_ports;
7191
7192         alloc_len = scsi_4btoul(cdb->length);
7193
7194         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7195
7196         ctsio->kern_sg_entries = 0;
7197
7198         if (total_len < alloc_len) {
7199                 ctsio->residual = alloc_len - total_len;
7200                 ctsio->kern_data_len = total_len;
7201                 ctsio->kern_total_len = total_len;
7202         } else {
7203                 ctsio->residual = 0;
7204                 ctsio->kern_data_len = alloc_len;
7205                 ctsio->kern_total_len = alloc_len;
7206         }
7207         ctsio->kern_data_resid = 0;
7208         ctsio->kern_rel_offset = 0;
7209
7210         if (ext) {
7211                 rtg_ext_ptr = (struct scsi_target_group_data_extended *)
7212                     ctsio->kern_data_ptr;
7213                 scsi_ulto4b(total_len - 4, rtg_ext_ptr->length);
7214                 rtg_ext_ptr->format_type = 0x10;
7215                 rtg_ext_ptr->implicit_transition_time = 0;
7216                 tpg_desc = &rtg_ext_ptr->groups[0];
7217         } else {
7218                 rtg_ptr = (struct scsi_target_group_data *)
7219                     ctsio->kern_data_ptr;
7220                 scsi_ulto4b(total_len - 4, rtg_ptr->length);
7221                 tpg_desc = &rtg_ptr->groups[0];
7222         }
7223
7224         mtx_lock(&softc->ctl_lock);
7225         pg = softc->port_min / softc->port_cnt;
7226         if (softc->ha_link == CTL_HA_LINK_OFFLINE)
7227                 gs = TPG_ASYMMETRIC_ACCESS_UNAVAILABLE;
7228         else if (softc->ha_link == CTL_HA_LINK_UNKNOWN)
7229                 gs = TPG_ASYMMETRIC_ACCESS_TRANSITIONING;
7230         else if (softc->ha_mode == CTL_HA_MODE_ACT_STBY)
7231                 gs = TPG_ASYMMETRIC_ACCESS_STANDBY;
7232         else
7233                 gs = TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7234         if (lun->flags & CTL_LUN_PRIMARY_SC) {
7235                 os = gs;
7236                 gs = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7237         } else
7238                 os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7239         for (g = 0; g < num_target_port_groups; g++) {
7240                 tpg_desc->pref_state = (g == pg) ? gs : os;
7241                 tpg_desc->support = TPG_AO_SUP | TPG_AN_SUP | TPG_S_SUP |
7242                     TPG_U_SUP | TPG_T_SUP;
7243                 scsi_ulto2b(g + 1, tpg_desc->target_port_group);
7244                 tpg_desc->status = TPG_IMPLICIT;
7245                 pc = 0;
7246                 STAILQ_FOREACH(port, &softc->port_list, links) {
7247                         if (port->targ_port < g * softc->port_cnt ||
7248                             port->targ_port >= (g + 1) * softc->port_cnt)
7249                                 continue;
7250                         if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7251                                 continue;
7252                         if (ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
7253                                 continue;
7254                         scsi_ulto2b(port->targ_port, tpg_desc->descriptors[pc].
7255                             relative_target_port_identifier);
7256                         pc++;
7257                 }
7258                 tpg_desc->target_port_count = pc;
7259                 tpg_desc = (struct scsi_target_port_group_descriptor *)
7260                     &tpg_desc->descriptors[pc];
7261         }
7262         mtx_unlock(&softc->ctl_lock);
7263
7264         ctl_set_success(ctsio);
7265         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7266         ctsio->be_move_done = ctl_config_move_done;
7267         ctl_datamove((union ctl_io *)ctsio);
7268         return(retval);
7269 }
7270
7271 int
7272 ctl_report_supported_opcodes(struct ctl_scsiio *ctsio)
7273 {
7274         struct ctl_lun *lun;
7275         struct scsi_report_supported_opcodes *cdb;
7276         const struct ctl_cmd_entry *entry, *sentry;
7277         struct scsi_report_supported_opcodes_all *all;
7278         struct scsi_report_supported_opcodes_descr *descr;
7279         struct scsi_report_supported_opcodes_one *one;
7280         int retval;
7281         int alloc_len, total_len;
7282         int opcode, service_action, i, j, num;
7283
7284         CTL_DEBUG_PRINT(("ctl_report_supported_opcodes\n"));
7285
7286         cdb = (struct scsi_report_supported_opcodes *)ctsio->cdb;
7287         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7288
7289         retval = CTL_RETVAL_COMPLETE;
7290
7291         opcode = cdb->requested_opcode;
7292         service_action = scsi_2btoul(cdb->requested_service_action);
7293         switch (cdb->options & RSO_OPTIONS_MASK) {
7294         case RSO_OPTIONS_ALL:
7295                 num = 0;
7296                 for (i = 0; i < 256; i++) {
7297                         entry = &ctl_cmd_table[i];
7298                         if (entry->flags & CTL_CMD_FLAG_SA5) {
7299                                 for (j = 0; j < 32; j++) {
7300                                         sentry = &((const struct ctl_cmd_entry *)
7301                                             entry->execute)[j];
7302                                         if (ctl_cmd_applicable(
7303                                             lun->be_lun->lun_type, sentry))
7304                                                 num++;
7305                                 }
7306                         } else {
7307                                 if (ctl_cmd_applicable(lun->be_lun->lun_type,
7308                                     entry))
7309                                         num++;
7310                         }
7311                 }
7312                 total_len = sizeof(struct scsi_report_supported_opcodes_all) +
7313                     num * sizeof(struct scsi_report_supported_opcodes_descr);
7314                 break;
7315         case RSO_OPTIONS_OC:
7316                 if (ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) {
7317                         ctl_set_invalid_field(/*ctsio*/ ctsio,
7318                                               /*sks_valid*/ 1,
7319                                               /*command*/ 1,
7320                                               /*field*/ 2,
7321                                               /*bit_valid*/ 1,
7322                                               /*bit*/ 2);
7323                         ctl_done((union ctl_io *)ctsio);
7324                         return (CTL_RETVAL_COMPLETE);
7325                 }
7326                 total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7327                 break;
7328         case RSO_OPTIONS_OC_SA:
7329                 if ((ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) == 0 ||
7330                     service_action >= 32) {
7331                         ctl_set_invalid_field(/*ctsio*/ ctsio,
7332                                               /*sks_valid*/ 1,
7333                                               /*command*/ 1,
7334                                               /*field*/ 2,
7335                                               /*bit_valid*/ 1,
7336                                               /*bit*/ 2);
7337                         ctl_done((union ctl_io *)ctsio);
7338                         return (CTL_RETVAL_COMPLETE);
7339                 }
7340                 total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7341                 break;
7342         default:
7343                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7344                                       /*sks_valid*/ 1,
7345                                       /*command*/ 1,
7346                                       /*field*/ 2,
7347                                       /*bit_valid*/ 1,
7348                                       /*bit*/ 2);
7349                 ctl_done((union ctl_io *)ctsio);
7350                 return (CTL_RETVAL_COMPLETE);
7351         }
7352
7353         alloc_len = scsi_4btoul(cdb->length);
7354
7355         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7356
7357         ctsio->kern_sg_entries = 0;
7358
7359         if (total_len < alloc_len) {
7360                 ctsio->residual = alloc_len - total_len;
7361                 ctsio->kern_data_len = total_len;
7362                 ctsio->kern_total_len = total_len;
7363         } else {
7364                 ctsio->residual = 0;
7365                 ctsio->kern_data_len = alloc_len;
7366                 ctsio->kern_total_len = alloc_len;
7367         }
7368         ctsio->kern_data_resid = 0;
7369         ctsio->kern_rel_offset = 0;
7370
7371         switch (cdb->options & RSO_OPTIONS_MASK) {
7372         case RSO_OPTIONS_ALL:
7373                 all = (struct scsi_report_supported_opcodes_all *)
7374                     ctsio->kern_data_ptr;
7375                 num = 0;
7376                 for (i = 0; i < 256; i++) {
7377                         entry = &ctl_cmd_table[i];
7378                         if (entry->flags & CTL_CMD_FLAG_SA5) {
7379                                 for (j = 0; j < 32; j++) {
7380                                         sentry = &((const struct ctl_cmd_entry *)
7381                                             entry->execute)[j];
7382                                         if (!ctl_cmd_applicable(
7383                                             lun->be_lun->lun_type, sentry))
7384                                                 continue;
7385                                         descr = &all->descr[num++];
7386                                         descr->opcode = i;
7387                                         scsi_ulto2b(j, descr->service_action);
7388                                         descr->flags = RSO_SERVACTV;
7389                                         scsi_ulto2b(sentry->length,
7390                                             descr->cdb_length);
7391                                 }
7392                         } else {
7393                                 if (!ctl_cmd_applicable(lun->be_lun->lun_type,
7394                                     entry))
7395                                         continue;
7396                                 descr = &all->descr[num++];
7397                                 descr->opcode = i;
7398                                 scsi_ulto2b(0, descr->service_action);
7399                                 descr->flags = 0;
7400                                 scsi_ulto2b(entry->length, descr->cdb_length);
7401                         }
7402                 }
7403                 scsi_ulto4b(
7404                     num * sizeof(struct scsi_report_supported_opcodes_descr),
7405                     all->length);
7406                 break;
7407         case RSO_OPTIONS_OC:
7408                 one = (struct scsi_report_supported_opcodes_one *)
7409                     ctsio->kern_data_ptr;
7410                 entry = &ctl_cmd_table[opcode];
7411                 goto fill_one;
7412         case RSO_OPTIONS_OC_SA:
7413                 one = (struct scsi_report_supported_opcodes_one *)
7414                     ctsio->kern_data_ptr;
7415                 entry = &ctl_cmd_table[opcode];
7416                 entry = &((const struct ctl_cmd_entry *)
7417                     entry->execute)[service_action];
7418 fill_one:
7419                 if (ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
7420                         one->support = 3;
7421                         scsi_ulto2b(entry->length, one->cdb_length);
7422                         one->cdb_usage[0] = opcode;
7423                         memcpy(&one->cdb_usage[1], entry->usage,
7424                             entry->length - 1);
7425                 } else
7426                         one->support = 1;
7427                 break;
7428         }
7429
7430         ctl_set_success(ctsio);
7431         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7432         ctsio->be_move_done = ctl_config_move_done;
7433         ctl_datamove((union ctl_io *)ctsio);
7434         return(retval);
7435 }
7436
7437 int
7438 ctl_report_supported_tmf(struct ctl_scsiio *ctsio)
7439 {
7440         struct scsi_report_supported_tmf *cdb;
7441         struct scsi_report_supported_tmf_data *data;
7442         int retval;
7443         int alloc_len, total_len;
7444
7445         CTL_DEBUG_PRINT(("ctl_report_supported_tmf\n"));
7446
7447         cdb = (struct scsi_report_supported_tmf *)ctsio->cdb;
7448
7449         retval = CTL_RETVAL_COMPLETE;
7450
7451         total_len = sizeof(struct scsi_report_supported_tmf_data);
7452         alloc_len = scsi_4btoul(cdb->length);
7453
7454         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7455
7456         ctsio->kern_sg_entries = 0;
7457
7458         if (total_len < alloc_len) {
7459                 ctsio->residual = alloc_len - total_len;
7460                 ctsio->kern_data_len = total_len;
7461                 ctsio->kern_total_len = total_len;
7462         } else {
7463                 ctsio->residual = 0;
7464                 ctsio->kern_data_len = alloc_len;
7465                 ctsio->kern_total_len = alloc_len;
7466         }
7467         ctsio->kern_data_resid = 0;
7468         ctsio->kern_rel_offset = 0;
7469
7470         data = (struct scsi_report_supported_tmf_data *)ctsio->kern_data_ptr;
7471         data->byte1 |= RST_ATS | RST_ATSS | RST_CTSS | RST_LURS | RST_QTS |
7472             RST_TRS;
7473         data->byte2 |= RST_QAES | RST_QTSS | RST_ITNRS;
7474
7475         ctl_set_success(ctsio);
7476         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7477         ctsio->be_move_done = ctl_config_move_done;
7478         ctl_datamove((union ctl_io *)ctsio);
7479         return (retval);
7480 }
7481
7482 int
7483 ctl_report_timestamp(struct ctl_scsiio *ctsio)
7484 {
7485         struct scsi_report_timestamp *cdb;
7486         struct scsi_report_timestamp_data *data;
7487         struct timeval tv;
7488         int64_t timestamp;
7489         int retval;
7490         int alloc_len, total_len;
7491
7492         CTL_DEBUG_PRINT(("ctl_report_timestamp\n"));
7493
7494         cdb = (struct scsi_report_timestamp *)ctsio->cdb;
7495
7496         retval = CTL_RETVAL_COMPLETE;
7497
7498         total_len = sizeof(struct scsi_report_timestamp_data);
7499         alloc_len = scsi_4btoul(cdb->length);
7500
7501         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7502
7503         ctsio->kern_sg_entries = 0;
7504
7505         if (total_len < alloc_len) {
7506                 ctsio->residual = alloc_len - total_len;
7507                 ctsio->kern_data_len = total_len;
7508                 ctsio->kern_total_len = total_len;
7509         } else {
7510                 ctsio->residual = 0;
7511                 ctsio->kern_data_len = alloc_len;
7512                 ctsio->kern_total_len = alloc_len;
7513         }
7514         ctsio->kern_data_resid = 0;
7515         ctsio->kern_rel_offset = 0;
7516
7517         data = (struct scsi_report_timestamp_data *)ctsio->kern_data_ptr;
7518         scsi_ulto2b(sizeof(*data) - 2, data->length);
7519         data->origin = RTS_ORIG_OUTSIDE;
7520         getmicrotime(&tv);
7521         timestamp = (int64_t)tv.tv_sec * 1000 + tv.tv_usec / 1000;
7522         scsi_ulto4b(timestamp >> 16, data->timestamp);
7523         scsi_ulto2b(timestamp & 0xffff, &data->timestamp[4]);
7524
7525         ctl_set_success(ctsio);
7526         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7527         ctsio->be_move_done = ctl_config_move_done;
7528         ctl_datamove((union ctl_io *)ctsio);
7529         return (retval);
7530 }
7531
7532 int
7533 ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7534 {
7535         struct scsi_per_res_in *cdb;
7536         int alloc_len, total_len = 0;
7537         /* struct scsi_per_res_in_rsrv in_data; */
7538         struct ctl_lun *lun;
7539         struct ctl_softc *softc;
7540         uint64_t key;
7541
7542         CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7543
7544         cdb = (struct scsi_per_res_in *)ctsio->cdb;
7545
7546         alloc_len = scsi_2btoul(cdb->length);
7547
7548         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7549         softc = lun->ctl_softc;
7550
7551 retry:
7552         mtx_lock(&lun->lun_lock);
7553         switch (cdb->action) {
7554         case SPRI_RK: /* read keys */
7555                 total_len = sizeof(struct scsi_per_res_in_keys) +
7556                         lun->pr_key_count *
7557                         sizeof(struct scsi_per_res_key);
7558                 break;
7559         case SPRI_RR: /* read reservation */
7560                 if (lun->flags & CTL_LUN_PR_RESERVED)
7561                         total_len = sizeof(struct scsi_per_res_in_rsrv);
7562                 else
7563                         total_len = sizeof(struct scsi_per_res_in_header);
7564                 break;
7565         case SPRI_RC: /* report capabilities */
7566                 total_len = sizeof(struct scsi_per_res_cap);
7567                 break;
7568         case SPRI_RS: /* read full status */
7569                 total_len = sizeof(struct scsi_per_res_in_header) +
7570                     (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7571                     lun->pr_key_count;
7572                 break;
7573         default:
7574                 panic("Invalid PR type %x", cdb->action);
7575         }
7576         mtx_unlock(&lun->lun_lock);
7577
7578         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7579
7580         if (total_len < alloc_len) {
7581                 ctsio->residual = alloc_len - total_len;
7582                 ctsio->kern_data_len = total_len;
7583                 ctsio->kern_total_len = total_len;
7584         } else {
7585                 ctsio->residual = 0;
7586                 ctsio->kern_data_len = alloc_len;
7587                 ctsio->kern_total_len = alloc_len;
7588         }
7589
7590         ctsio->kern_data_resid = 0;
7591         ctsio->kern_rel_offset = 0;
7592         ctsio->kern_sg_entries = 0;
7593
7594         mtx_lock(&lun->lun_lock);
7595         switch (cdb->action) {
7596         case SPRI_RK: { // read keys
7597         struct scsi_per_res_in_keys *res_keys;
7598                 int i, key_count;
7599
7600                 res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7601
7602                 /*
7603                  * We had to drop the lock to allocate our buffer, which
7604                  * leaves time for someone to come in with another
7605                  * persistent reservation.  (That is unlikely, though,
7606                  * since this should be the only persistent reservation
7607                  * command active right now.)
7608                  */
7609                 if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7610                     (lun->pr_key_count *
7611                      sizeof(struct scsi_per_res_key)))){
7612                         mtx_unlock(&lun->lun_lock);
7613                         free(ctsio->kern_data_ptr, M_CTL);
7614                         printf("%s: reservation length changed, retrying\n",
7615                                __func__);
7616                         goto retry;
7617                 }
7618
7619                 scsi_ulto4b(lun->PRGeneration, res_keys->header.generation);
7620
7621                 scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7622                              lun->pr_key_count, res_keys->header.length);
7623
7624                 for (i = 0, key_count = 0; i < CTL_MAX_INITIATORS; i++) {
7625                         if ((key = ctl_get_prkey(lun, i)) == 0)
7626                                 continue;
7627
7628                         /*
7629                          * We used lun->pr_key_count to calculate the
7630                          * size to allocate.  If it turns out the number of
7631                          * initiators with the registered flag set is
7632                          * larger than that (i.e. they haven't been kept in
7633                          * sync), we've got a problem.
7634                          */
7635                         if (key_count >= lun->pr_key_count) {
7636 #ifdef NEEDTOPORT
7637                                 csevent_log(CSC_CTL | CSC_SHELF_SW |
7638                                             CTL_PR_ERROR,
7639                                             csevent_LogType_Fault,
7640                                             csevent_AlertLevel_Yellow,
7641                                             csevent_FRU_ShelfController,
7642                                             csevent_FRU_Firmware,
7643                                         csevent_FRU_Unknown,
7644                                             "registered keys %d >= key "
7645                                             "count %d", key_count,
7646                                             lun->pr_key_count);
7647 #endif
7648                                 key_count++;
7649                                 continue;
7650                         }
7651                         scsi_u64to8b(key, res_keys->keys[key_count].key);
7652                         key_count++;
7653                 }
7654                 break;
7655         }
7656         case SPRI_RR: { // read reservation
7657                 struct scsi_per_res_in_rsrv *res;
7658                 int tmp_len, header_only;
7659
7660                 res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7661
7662                 scsi_ulto4b(lun->PRGeneration, res->header.generation);
7663
7664                 if (lun->flags & CTL_LUN_PR_RESERVED)
7665                 {
7666                         tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7667                         scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7668                                     res->header.length);
7669                         header_only = 0;
7670                 } else {
7671                         tmp_len = sizeof(struct scsi_per_res_in_header);
7672                         scsi_ulto4b(0, res->header.length);
7673                         header_only = 1;
7674                 }
7675
7676                 /*
7677                  * We had to drop the lock to allocate our buffer, which
7678                  * leaves time for someone to come in with another
7679                  * persistent reservation.  (That is unlikely, though,
7680                  * since this should be the only persistent reservation
7681                  * command active right now.)
7682                  */
7683                 if (tmp_len != total_len) {
7684                         mtx_unlock(&lun->lun_lock);
7685                         free(ctsio->kern_data_ptr, M_CTL);
7686                         printf("%s: reservation status changed, retrying\n",
7687                                __func__);
7688                         goto retry;
7689                 }
7690
7691                 /*
7692                  * No reservation held, so we're done.
7693                  */
7694                 if (header_only != 0)
7695                         break;
7696
7697                 /*
7698                  * If the registration is an All Registrants type, the key
7699                  * is 0, since it doesn't really matter.
7700                  */
7701                 if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7702                         scsi_u64to8b(ctl_get_prkey(lun, lun->pr_res_idx),
7703                             res->data.reservation);
7704                 }
7705                 res->data.scopetype = lun->res_type;
7706                 break;
7707         }
7708         case SPRI_RC:     //report capabilities
7709         {
7710                 struct scsi_per_res_cap *res_cap;
7711                 uint16_t type_mask;
7712
7713                 res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7714                 scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7715                 res_cap->flags2 |= SPRI_TMV | SPRI_ALLOW_5;
7716                 type_mask = SPRI_TM_WR_EX_AR |
7717                             SPRI_TM_EX_AC_RO |
7718                             SPRI_TM_WR_EX_RO |
7719                             SPRI_TM_EX_AC |
7720                             SPRI_TM_WR_EX |
7721                             SPRI_TM_EX_AC_AR;
7722                 scsi_ulto2b(type_mask, res_cap->type_mask);
7723                 break;
7724         }
7725         case SPRI_RS: { // read full status
7726                 struct scsi_per_res_in_full *res_status;
7727                 struct scsi_per_res_in_full_desc *res_desc;
7728                 struct ctl_port *port;
7729                 int i, len;
7730
7731                 res_status = (struct scsi_per_res_in_full*)ctsio->kern_data_ptr;
7732
7733                 /*
7734                  * We had to drop the lock to allocate our buffer, which
7735                  * leaves time for someone to come in with another
7736                  * persistent reservation.  (That is unlikely, though,
7737                  * since this should be the only persistent reservation
7738                  * command active right now.)
7739                  */
7740                 if (total_len < (sizeof(struct scsi_per_res_in_header) +
7741                     (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7742                      lun->pr_key_count)){
7743                         mtx_unlock(&lun->lun_lock);
7744                         free(ctsio->kern_data_ptr, M_CTL);
7745                         printf("%s: reservation length changed, retrying\n",
7746                                __func__);
7747                         goto retry;
7748                 }
7749
7750                 scsi_ulto4b(lun->PRGeneration, res_status->header.generation);
7751
7752                 res_desc = &res_status->desc[0];
7753                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
7754                         if ((key = ctl_get_prkey(lun, i)) == 0)
7755                                 continue;
7756
7757                         scsi_u64to8b(key, res_desc->res_key.key);
7758                         if ((lun->flags & CTL_LUN_PR_RESERVED) &&
7759                             (lun->pr_res_idx == i ||
7760                              lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS)) {
7761                                 res_desc->flags = SPRI_FULL_R_HOLDER;
7762                                 res_desc->scopetype = lun->res_type;
7763                         }
7764                         scsi_ulto2b(i / CTL_MAX_INIT_PER_PORT,
7765                             res_desc->rel_trgt_port_id);
7766                         len = 0;
7767                         port = softc->ctl_ports[i / CTL_MAX_INIT_PER_PORT];
7768                         if (port != NULL)
7769                                 len = ctl_create_iid(port,
7770                                     i % CTL_MAX_INIT_PER_PORT,
7771                                     res_desc->transport_id);
7772                         scsi_ulto4b(len, res_desc->additional_length);
7773                         res_desc = (struct scsi_per_res_in_full_desc *)
7774                             &res_desc->transport_id[len];
7775                 }
7776                 scsi_ulto4b((uint8_t *)res_desc - (uint8_t *)&res_status->desc[0],
7777                     res_status->header.length);
7778                 break;
7779         }
7780         default:
7781                 /*
7782                  * This is a bug, because we just checked for this above,
7783                  * and should have returned an error.
7784                  */
7785                 panic("Invalid PR type %x", cdb->action);
7786                 break; /* NOTREACHED */
7787         }
7788         mtx_unlock(&lun->lun_lock);
7789
7790         ctl_set_success(ctsio);
7791         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7792         ctsio->be_move_done = ctl_config_move_done;
7793         ctl_datamove((union ctl_io *)ctsio);
7794         return (CTL_RETVAL_COMPLETE);
7795 }
7796
7797 /*
7798  * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
7799  * it should return.
7800  */
7801 static int
7802 ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
7803                 uint64_t sa_res_key, uint8_t type, uint32_t residx,
7804                 struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
7805                 struct scsi_per_res_out_parms* param)
7806 {
7807         union ctl_ha_msg persis_io;
7808         int i;
7809
7810         mtx_lock(&lun->lun_lock);
7811         if (sa_res_key == 0) {
7812                 if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
7813                         /* validate scope and type */
7814                         if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7815                              SPR_LU_SCOPE) {
7816                                 mtx_unlock(&lun->lun_lock);
7817                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7818                                                       /*sks_valid*/ 1,
7819                                                       /*command*/ 1,
7820                                                       /*field*/ 2,
7821                                                       /*bit_valid*/ 1,
7822                                                       /*bit*/ 4);
7823                                 ctl_done((union ctl_io *)ctsio);
7824                                 return (1);
7825                         }
7826
7827                         if (type>8 || type==2 || type==4 || type==0) {
7828                                 mtx_unlock(&lun->lun_lock);
7829                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7830                                                       /*sks_valid*/ 1,
7831                                                       /*command*/ 1,
7832                                                       /*field*/ 2,
7833                                                       /*bit_valid*/ 1,
7834                                                       /*bit*/ 0);
7835                                 ctl_done((union ctl_io *)ctsio);
7836                                 return (1);
7837                         }
7838
7839                         /*
7840                          * Unregister everybody else and build UA for
7841                          * them
7842                          */
7843                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
7844                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
7845                                         continue;
7846
7847                                 ctl_clr_prkey(lun, i);
7848                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7849                         }
7850                         lun->pr_key_count = 1;
7851                         lun->res_type = type;
7852                         if (lun->res_type != SPR_TYPE_WR_EX_AR
7853                          && lun->res_type != SPR_TYPE_EX_AC_AR)
7854                                 lun->pr_res_idx = residx;
7855                         lun->PRGeneration++;
7856                         mtx_unlock(&lun->lun_lock);
7857
7858                         /* send msg to other side */
7859                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7860                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7861                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7862                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
7863                         persis_io.pr.pr_info.res_type = type;
7864                         memcpy(persis_io.pr.pr_info.sa_res_key,
7865                                param->serv_act_res_key,
7866                                sizeof(param->serv_act_res_key));
7867                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
7868                             sizeof(persis_io.pr), M_WAITOK);
7869                 } else {
7870                         /* not all registrants */
7871                         mtx_unlock(&lun->lun_lock);
7872                         free(ctsio->kern_data_ptr, M_CTL);
7873                         ctl_set_invalid_field(ctsio,
7874                                               /*sks_valid*/ 1,
7875                                               /*command*/ 0,
7876                                               /*field*/ 8,
7877                                               /*bit_valid*/ 0,
7878                                               /*bit*/ 0);
7879                         ctl_done((union ctl_io *)ctsio);
7880                         return (1);
7881                 }
7882         } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
7883                 || !(lun->flags & CTL_LUN_PR_RESERVED)) {
7884                 int found = 0;
7885
7886                 if (res_key == sa_res_key) {
7887                         /* special case */
7888                         /*
7889                          * The spec implies this is not good but doesn't
7890                          * say what to do. There are two choices either
7891                          * generate a res conflict or check condition
7892                          * with illegal field in parameter data. Since
7893                          * that is what is done when the sa_res_key is
7894                          * zero I'll take that approach since this has
7895                          * to do with the sa_res_key.
7896                          */
7897                         mtx_unlock(&lun->lun_lock);
7898                         free(ctsio->kern_data_ptr, M_CTL);
7899                         ctl_set_invalid_field(ctsio,
7900                                               /*sks_valid*/ 1,
7901                                               /*command*/ 0,
7902                                               /*field*/ 8,
7903                                               /*bit_valid*/ 0,
7904                                               /*bit*/ 0);
7905                         ctl_done((union ctl_io *)ctsio);
7906                         return (1);
7907                 }
7908
7909                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
7910                         if (ctl_get_prkey(lun, i) != sa_res_key)
7911                                 continue;
7912
7913                         found = 1;
7914                         ctl_clr_prkey(lun, i);
7915                         lun->pr_key_count--;
7916                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7917                 }
7918                 if (!found) {
7919                         mtx_unlock(&lun->lun_lock);
7920                         free(ctsio->kern_data_ptr, M_CTL);
7921                         ctl_set_reservation_conflict(ctsio);
7922                         ctl_done((union ctl_io *)ctsio);
7923                         return (CTL_RETVAL_COMPLETE);
7924                 }
7925                 lun->PRGeneration++;
7926                 mtx_unlock(&lun->lun_lock);
7927
7928                 /* send msg to other side */
7929                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7930                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7931                 persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7932                 persis_io.pr.pr_info.residx = lun->pr_res_idx;
7933                 persis_io.pr.pr_info.res_type = type;
7934                 memcpy(persis_io.pr.pr_info.sa_res_key,
7935                        param->serv_act_res_key,
7936                        sizeof(param->serv_act_res_key));
7937                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
7938                     sizeof(persis_io.pr), M_WAITOK);
7939         } else {
7940                 /* Reserved but not all registrants */
7941                 /* sa_res_key is res holder */
7942                 if (sa_res_key == ctl_get_prkey(lun, lun->pr_res_idx)) {
7943                         /* validate scope and type */
7944                         if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7945                              SPR_LU_SCOPE) {
7946                                 mtx_unlock(&lun->lun_lock);
7947                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7948                                                       /*sks_valid*/ 1,
7949                                                       /*command*/ 1,
7950                                                       /*field*/ 2,
7951                                                       /*bit_valid*/ 1,
7952                                                       /*bit*/ 4);
7953                                 ctl_done((union ctl_io *)ctsio);
7954                                 return (1);
7955                         }
7956
7957                         if (type>8 || type==2 || type==4 || type==0) {
7958                                 mtx_unlock(&lun->lun_lock);
7959                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7960                                                       /*sks_valid*/ 1,
7961                                                       /*command*/ 1,
7962                                                       /*field*/ 2,
7963                                                       /*bit_valid*/ 1,
7964                                                       /*bit*/ 0);
7965                                 ctl_done((union ctl_io *)ctsio);
7966                                 return (1);
7967                         }
7968
7969                         /*
7970                          * Do the following:
7971                          * if sa_res_key != res_key remove all
7972                          * registrants w/sa_res_key and generate UA
7973                          * for these registrants(Registrations
7974                          * Preempted) if it wasn't an exclusive
7975                          * reservation generate UA(Reservations
7976                          * Preempted) for all other registered nexuses
7977                          * if the type has changed. Establish the new
7978                          * reservation and holder. If res_key and
7979                          * sa_res_key are the same do the above
7980                          * except don't unregister the res holder.
7981                          */
7982
7983                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
7984                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
7985                                         continue;
7986
7987                                 if (sa_res_key == ctl_get_prkey(lun, i)) {
7988                                         ctl_clr_prkey(lun, i);
7989                                         lun->pr_key_count--;
7990                                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7991                                 } else if (type != lun->res_type
7992                                         && (lun->res_type == SPR_TYPE_WR_EX_RO
7993                                          || lun->res_type ==SPR_TYPE_EX_AC_RO)){
7994                                         ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
7995                                 }
7996                         }
7997                         lun->res_type = type;
7998                         if (lun->res_type != SPR_TYPE_WR_EX_AR
7999                          && lun->res_type != SPR_TYPE_EX_AC_AR)
8000                                 lun->pr_res_idx = residx;
8001                         else
8002                                 lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8003                         lun->PRGeneration++;
8004                         mtx_unlock(&lun->lun_lock);
8005
8006                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8007                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8008                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8009                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8010                         persis_io.pr.pr_info.res_type = type;
8011                         memcpy(persis_io.pr.pr_info.sa_res_key,
8012                                param->serv_act_res_key,
8013                                sizeof(param->serv_act_res_key));
8014                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8015                             sizeof(persis_io.pr), M_WAITOK);
8016                 } else {
8017                         /*
8018                          * sa_res_key is not the res holder just
8019                          * remove registrants
8020                          */
8021                         int found=0;
8022
8023                         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8024                                 if (sa_res_key != ctl_get_prkey(lun, i))
8025                                         continue;
8026
8027                                 found = 1;
8028                                 ctl_clr_prkey(lun, i);
8029                                 lun->pr_key_count--;
8030                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8031                         }
8032
8033                         if (!found) {
8034                                 mtx_unlock(&lun->lun_lock);
8035                                 free(ctsio->kern_data_ptr, M_CTL);
8036                                 ctl_set_reservation_conflict(ctsio);
8037                                 ctl_done((union ctl_io *)ctsio);
8038                                 return (1);
8039                         }
8040                         lun->PRGeneration++;
8041                         mtx_unlock(&lun->lun_lock);
8042
8043                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8044                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8045                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8046                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8047                         persis_io.pr.pr_info.res_type = type;
8048                         memcpy(persis_io.pr.pr_info.sa_res_key,
8049                                param->serv_act_res_key,
8050                                sizeof(param->serv_act_res_key));
8051                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8052                             sizeof(persis_io.pr), M_WAITOK);
8053                 }
8054         }
8055         return (0);
8056 }
8057
8058 static void
8059 ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
8060 {
8061         uint64_t sa_res_key;
8062         int i;
8063
8064         sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
8065
8066         if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8067          || lun->pr_res_idx == CTL_PR_NO_RESERVATION
8068          || sa_res_key != ctl_get_prkey(lun, lun->pr_res_idx)) {
8069                 if (sa_res_key == 0) {
8070                         /*
8071                          * Unregister everybody else and build UA for
8072                          * them
8073                          */
8074                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
8075                                 if (i == msg->pr.pr_info.residx ||
8076                                     ctl_get_prkey(lun, i) == 0)
8077                                         continue;
8078
8079                                 ctl_clr_prkey(lun, i);
8080                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8081                         }
8082
8083                         lun->pr_key_count = 1;
8084                         lun->res_type = msg->pr.pr_info.res_type;
8085                         if (lun->res_type != SPR_TYPE_WR_EX_AR
8086                          && lun->res_type != SPR_TYPE_EX_AC_AR)
8087                                 lun->pr_res_idx = msg->pr.pr_info.residx;
8088                 } else {
8089                         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8090                                 if (sa_res_key == ctl_get_prkey(lun, i))
8091                                         continue;
8092
8093                                 ctl_clr_prkey(lun, i);
8094                                 lun->pr_key_count--;
8095                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8096                         }
8097                 }
8098         } else {
8099                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8100                         if (i == msg->pr.pr_info.residx ||
8101                             ctl_get_prkey(lun, i) == 0)
8102                                 continue;
8103
8104                         if (sa_res_key == ctl_get_prkey(lun, i)) {
8105                                 ctl_clr_prkey(lun, i);
8106                                 lun->pr_key_count--;
8107                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8108                         } else if (msg->pr.pr_info.res_type != lun->res_type
8109                                 && (lun->res_type == SPR_TYPE_WR_EX_RO
8110                                  || lun->res_type == SPR_TYPE_EX_AC_RO)) {
8111                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8112                         }
8113                 }
8114                 lun->res_type = msg->pr.pr_info.res_type;
8115                 if (lun->res_type != SPR_TYPE_WR_EX_AR
8116                  && lun->res_type != SPR_TYPE_EX_AC_AR)
8117                         lun->pr_res_idx = msg->pr.pr_info.residx;
8118                 else
8119                         lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8120         }
8121         lun->PRGeneration++;
8122
8123 }
8124
8125
8126 int
8127 ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
8128 {
8129         int retval;
8130         u_int32_t param_len;
8131         struct scsi_per_res_out *cdb;
8132         struct ctl_lun *lun;
8133         struct scsi_per_res_out_parms* param;
8134         struct ctl_softc *softc;
8135         uint32_t residx;
8136         uint64_t res_key, sa_res_key, key;
8137         uint8_t type;
8138         union ctl_ha_msg persis_io;
8139         int    i;
8140
8141         CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n"));
8142
8143         retval = CTL_RETVAL_COMPLETE;
8144
8145         cdb = (struct scsi_per_res_out *)ctsio->cdb;
8146         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8147         softc = lun->ctl_softc;
8148
8149         /*
8150          * We only support whole-LUN scope.  The scope & type are ignored for
8151          * register, register and ignore existing key and clear.
8152          * We sometimes ignore scope and type on preempts too!!
8153          * Verify reservation type here as well.
8154          */
8155         type = cdb->scope_type & SPR_TYPE_MASK;
8156         if ((cdb->action == SPRO_RESERVE)
8157          || (cdb->action == SPRO_RELEASE)) {
8158                 if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
8159                         ctl_set_invalid_field(/*ctsio*/ ctsio,
8160                                               /*sks_valid*/ 1,
8161                                               /*command*/ 1,
8162                                               /*field*/ 2,
8163                                               /*bit_valid*/ 1,
8164                                               /*bit*/ 4);
8165                         ctl_done((union ctl_io *)ctsio);
8166                         return (CTL_RETVAL_COMPLETE);
8167                 }
8168
8169                 if (type>8 || type==2 || type==4 || type==0) {
8170                         ctl_set_invalid_field(/*ctsio*/ ctsio,
8171                                               /*sks_valid*/ 1,
8172                                               /*command*/ 1,
8173                                               /*field*/ 2,
8174                                               /*bit_valid*/ 1,
8175                                               /*bit*/ 0);
8176                         ctl_done((union ctl_io *)ctsio);
8177                         return (CTL_RETVAL_COMPLETE);
8178                 }
8179         }
8180
8181         param_len = scsi_4btoul(cdb->length);
8182
8183         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
8184                 ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
8185                 ctsio->kern_data_len = param_len;
8186                 ctsio->kern_total_len = param_len;
8187                 ctsio->kern_data_resid = 0;
8188                 ctsio->kern_rel_offset = 0;
8189                 ctsio->kern_sg_entries = 0;
8190                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
8191                 ctsio->be_move_done = ctl_config_move_done;
8192                 ctl_datamove((union ctl_io *)ctsio);
8193
8194                 return (CTL_RETVAL_COMPLETE);
8195         }
8196
8197         param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
8198
8199         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
8200         res_key = scsi_8btou64(param->res_key.key);
8201         sa_res_key = scsi_8btou64(param->serv_act_res_key);
8202
8203         /*
8204          * Validate the reservation key here except for SPRO_REG_IGNO
8205          * This must be done for all other service actions
8206          */
8207         if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
8208                 mtx_lock(&lun->lun_lock);
8209                 if ((key = ctl_get_prkey(lun, residx)) != 0) {
8210                         if (res_key != key) {
8211                                 /*
8212                                  * The current key passed in doesn't match
8213                                  * the one the initiator previously
8214                                  * registered.
8215                                  */
8216                                 mtx_unlock(&lun->lun_lock);
8217                                 free(ctsio->kern_data_ptr, M_CTL);
8218                                 ctl_set_reservation_conflict(ctsio);
8219                                 ctl_done((union ctl_io *)ctsio);
8220                                 return (CTL_RETVAL_COMPLETE);
8221                         }
8222                 } else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
8223                         /*
8224                          * We are not registered
8225                          */
8226                         mtx_unlock(&lun->lun_lock);
8227                         free(ctsio->kern_data_ptr, M_CTL);
8228                         ctl_set_reservation_conflict(ctsio);
8229                         ctl_done((union ctl_io *)ctsio);
8230                         return (CTL_RETVAL_COMPLETE);
8231                 } else if (res_key != 0) {
8232                         /*
8233                          * We are not registered and trying to register but
8234                          * the register key isn't zero.
8235                          */
8236                         mtx_unlock(&lun->lun_lock);
8237                         free(ctsio->kern_data_ptr, M_CTL);
8238                         ctl_set_reservation_conflict(ctsio);
8239                         ctl_done((union ctl_io *)ctsio);
8240                         return (CTL_RETVAL_COMPLETE);
8241                 }
8242                 mtx_unlock(&lun->lun_lock);
8243         }
8244
8245         switch (cdb->action & SPRO_ACTION_MASK) {
8246         case SPRO_REGISTER:
8247         case SPRO_REG_IGNO: {
8248
8249 #if 0
8250                 printf("Registration received\n");
8251 #endif
8252
8253                 /*
8254                  * We don't support any of these options, as we report in
8255                  * the read capabilities request (see
8256                  * ctl_persistent_reserve_in(), above).
8257                  */
8258                 if ((param->flags & SPR_SPEC_I_PT)
8259                  || (param->flags & SPR_ALL_TG_PT)
8260                  || (param->flags & SPR_APTPL)) {
8261                         int bit_ptr;
8262
8263                         if (param->flags & SPR_APTPL)
8264                                 bit_ptr = 0;
8265                         else if (param->flags & SPR_ALL_TG_PT)
8266                                 bit_ptr = 2;
8267                         else /* SPR_SPEC_I_PT */
8268                                 bit_ptr = 3;
8269
8270                         free(ctsio->kern_data_ptr, M_CTL);
8271                         ctl_set_invalid_field(ctsio,
8272                                               /*sks_valid*/ 1,
8273                                               /*command*/ 0,
8274                                               /*field*/ 20,
8275                                               /*bit_valid*/ 1,
8276                                               /*bit*/ bit_ptr);
8277                         ctl_done((union ctl_io *)ctsio);
8278                         return (CTL_RETVAL_COMPLETE);
8279                 }
8280
8281                 mtx_lock(&lun->lun_lock);
8282
8283                 /*
8284                  * The initiator wants to clear the
8285                  * key/unregister.
8286                  */
8287                 if (sa_res_key == 0) {
8288                         if ((res_key == 0
8289                           && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
8290                          || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
8291                           && ctl_get_prkey(lun, residx) == 0)) {
8292                                 mtx_unlock(&lun->lun_lock);
8293                                 goto done;
8294                         }
8295
8296                         ctl_clr_prkey(lun, residx);
8297                         lun->pr_key_count--;
8298
8299                         if (residx == lun->pr_res_idx) {
8300                                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8301                                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8302
8303                                 if ((lun->res_type == SPR_TYPE_WR_EX_RO
8304                                   || lun->res_type == SPR_TYPE_EX_AC_RO)
8305                                  && lun->pr_key_count) {
8306                                         /*
8307                                          * If the reservation is a registrants
8308                                          * only type we need to generate a UA
8309                                          * for other registered inits.  The
8310                                          * sense code should be RESERVATIONS
8311                                          * RELEASED
8312                                          */
8313
8314                                         for (i = softc->init_min; i < softc->init_max; i++){
8315                                                 if (ctl_get_prkey(lun, i) == 0)
8316                                                         continue;
8317                                                 ctl_est_ua(lun, i,
8318                                                     CTL_UA_RES_RELEASE);
8319                                         }
8320                                 }
8321                                 lun->res_type = 0;
8322                         } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8323                                 if (lun->pr_key_count==0) {
8324                                         lun->flags &= ~CTL_LUN_PR_RESERVED;
8325                                         lun->res_type = 0;
8326                                         lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8327                                 }
8328                         }
8329                         lun->PRGeneration++;
8330                         mtx_unlock(&lun->lun_lock);
8331
8332                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8333                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8334                         persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
8335                         persis_io.pr.pr_info.residx = residx;
8336                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8337                             sizeof(persis_io.pr), M_WAITOK);
8338                 } else /* sa_res_key != 0 */ {
8339
8340                         /*
8341                          * If we aren't registered currently then increment
8342                          * the key count and set the registered flag.
8343                          */
8344                         ctl_alloc_prkey(lun, residx);
8345                         if (ctl_get_prkey(lun, residx) == 0)
8346                                 lun->pr_key_count++;
8347                         ctl_set_prkey(lun, residx, sa_res_key);
8348                         lun->PRGeneration++;
8349                         mtx_unlock(&lun->lun_lock);
8350
8351                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8352                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8353                         persis_io.pr.pr_info.action = CTL_PR_REG_KEY;
8354                         persis_io.pr.pr_info.residx = residx;
8355                         memcpy(persis_io.pr.pr_info.sa_res_key,
8356                                param->serv_act_res_key,
8357                                sizeof(param->serv_act_res_key));
8358                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8359                             sizeof(persis_io.pr), M_WAITOK);
8360                 }
8361
8362                 break;
8363         }
8364         case SPRO_RESERVE:
8365 #if 0
8366                 printf("Reserve executed type %d\n", type);
8367 #endif
8368                 mtx_lock(&lun->lun_lock);
8369                 if (lun->flags & CTL_LUN_PR_RESERVED) {
8370                         /*
8371                          * if this isn't the reservation holder and it's
8372                          * not a "all registrants" type or if the type is
8373                          * different then we have a conflict
8374                          */
8375                         if ((lun->pr_res_idx != residx
8376                           && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8377                          || lun->res_type != type) {
8378                                 mtx_unlock(&lun->lun_lock);
8379                                 free(ctsio->kern_data_ptr, M_CTL);
8380                                 ctl_set_reservation_conflict(ctsio);
8381                                 ctl_done((union ctl_io *)ctsio);
8382                                 return (CTL_RETVAL_COMPLETE);
8383                         }
8384                         mtx_unlock(&lun->lun_lock);
8385                 } else /* create a reservation */ {
8386                         /*
8387                          * If it's not an "all registrants" type record
8388                          * reservation holder
8389                          */
8390                         if (type != SPR_TYPE_WR_EX_AR
8391                          && type != SPR_TYPE_EX_AC_AR)
8392                                 lun->pr_res_idx = residx; /* Res holder */
8393                         else
8394                                 lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8395
8396                         lun->flags |= CTL_LUN_PR_RESERVED;
8397                         lun->res_type = type;
8398
8399                         mtx_unlock(&lun->lun_lock);
8400
8401                         /* send msg to other side */
8402                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8403                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8404                         persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8405                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8406                         persis_io.pr.pr_info.res_type = type;
8407                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8408                             sizeof(persis_io.pr), M_WAITOK);
8409                 }
8410                 break;
8411
8412         case SPRO_RELEASE:
8413                 mtx_lock(&lun->lun_lock);
8414                 if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8415                         /* No reservation exists return good status */
8416                         mtx_unlock(&lun->lun_lock);
8417                         goto done;
8418                 }
8419                 /*
8420                  * Is this nexus a reservation holder?
8421                  */
8422                 if (lun->pr_res_idx != residx
8423                  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8424                         /*
8425                          * not a res holder return good status but
8426                          * do nothing
8427                          */
8428                         mtx_unlock(&lun->lun_lock);
8429                         goto done;
8430                 }
8431
8432                 if (lun->res_type != type) {
8433                         mtx_unlock(&lun->lun_lock);
8434                         free(ctsio->kern_data_ptr, M_CTL);
8435                         ctl_set_illegal_pr_release(ctsio);
8436                         ctl_done((union ctl_io *)ctsio);
8437                         return (CTL_RETVAL_COMPLETE);
8438                 }
8439
8440                 /* okay to release */
8441                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8442                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8443                 lun->res_type = 0;
8444
8445                 /*
8446                  * if this isn't an exclusive access
8447                  * res generate UA for all other
8448                  * registrants.
8449                  */
8450                 if (type != SPR_TYPE_EX_AC
8451                  && type != SPR_TYPE_WR_EX) {
8452                         for (i = softc->init_min; i < softc->init_max; i++) {
8453                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
8454                                         continue;
8455                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8456                         }
8457                 }
8458                 mtx_unlock(&lun->lun_lock);
8459
8460                 /* Send msg to other side */
8461                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8462                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8463                 persis_io.pr.pr_info.action = CTL_PR_RELEASE;
8464                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8465                      sizeof(persis_io.pr), M_WAITOK);
8466                 break;
8467
8468         case SPRO_CLEAR:
8469                 /* send msg to other side */
8470
8471                 mtx_lock(&lun->lun_lock);
8472                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8473                 lun->res_type = 0;
8474                 lun->pr_key_count = 0;
8475                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8476
8477                 ctl_clr_prkey(lun, residx);
8478                 for (i = 0; i < CTL_MAX_INITIATORS; i++)
8479                         if (ctl_get_prkey(lun, i) != 0) {
8480                                 ctl_clr_prkey(lun, i);
8481                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8482                         }
8483                 lun->PRGeneration++;
8484                 mtx_unlock(&lun->lun_lock);
8485
8486                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8487                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8488                 persis_io.pr.pr_info.action = CTL_PR_CLEAR;
8489                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8490                      sizeof(persis_io.pr), M_WAITOK);
8491                 break;
8492
8493         case SPRO_PREEMPT:
8494         case SPRO_PRE_ABO: {
8495                 int nretval;
8496
8497                 nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8498                                           residx, ctsio, cdb, param);
8499                 if (nretval != 0)
8500                         return (CTL_RETVAL_COMPLETE);
8501                 break;
8502         }
8503         default:
8504                 panic("Invalid PR type %x", cdb->action);
8505         }
8506
8507 done:
8508         free(ctsio->kern_data_ptr, M_CTL);
8509         ctl_set_success(ctsio);
8510         ctl_done((union ctl_io *)ctsio);
8511
8512         return (retval);
8513 }
8514
8515 /*
8516  * This routine is for handling a message from the other SC pertaining to
8517  * persistent reserve out. All the error checking will have been done
8518  * so only perorming the action need be done here to keep the two
8519  * in sync.
8520  */
8521 static void
8522 ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg)
8523 {
8524         struct ctl_lun *lun;
8525         struct ctl_softc *softc;
8526         int i;
8527         uint32_t residx, targ_lun;
8528
8529         softc = control_softc;
8530         targ_lun = msg->hdr.nexus.targ_mapped_lun;
8531         mtx_lock(&softc->ctl_lock);
8532         if ((targ_lun >= CTL_MAX_LUNS) ||
8533             ((lun = softc->ctl_luns[targ_lun]) == NULL)) {
8534                 mtx_unlock(&softc->ctl_lock);
8535                 return;
8536         }
8537         mtx_lock(&lun->lun_lock);
8538         mtx_unlock(&softc->ctl_lock);
8539         if (lun->flags & CTL_LUN_DISABLED) {
8540                 mtx_unlock(&lun->lun_lock);
8541                 return;
8542         }
8543         residx = ctl_get_initindex(&msg->hdr.nexus);
8544         switch(msg->pr.pr_info.action) {
8545         case CTL_PR_REG_KEY:
8546                 ctl_alloc_prkey(lun, msg->pr.pr_info.residx);
8547                 if (ctl_get_prkey(lun, msg->pr.pr_info.residx) == 0)
8548                         lun->pr_key_count++;
8549                 ctl_set_prkey(lun, msg->pr.pr_info.residx,
8550                     scsi_8btou64(msg->pr.pr_info.sa_res_key));
8551                 lun->PRGeneration++;
8552                 break;
8553
8554         case CTL_PR_UNREG_KEY:
8555                 ctl_clr_prkey(lun, msg->pr.pr_info.residx);
8556                 lun->pr_key_count--;
8557
8558                 /* XXX Need to see if the reservation has been released */
8559                 /* if so do we need to generate UA? */
8560                 if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8561                         lun->flags &= ~CTL_LUN_PR_RESERVED;
8562                         lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8563
8564                         if ((lun->res_type == SPR_TYPE_WR_EX_RO
8565                           || lun->res_type == SPR_TYPE_EX_AC_RO)
8566                          && lun->pr_key_count) {
8567                                 /*
8568                                  * If the reservation is a registrants
8569                                  * only type we need to generate a UA
8570                                  * for other registered inits.  The
8571                                  * sense code should be RESERVATIONS
8572                                  * RELEASED
8573                                  */
8574
8575                                 for (i = softc->init_min; i < softc->init_max; i++) {
8576                                         if (ctl_get_prkey(lun, i) == 0)
8577                                                 continue;
8578
8579                                         ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8580                                 }
8581                         }
8582                         lun->res_type = 0;
8583                 } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8584                         if (lun->pr_key_count==0) {
8585                                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8586                                 lun->res_type = 0;
8587                                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8588                         }
8589                 }
8590                 lun->PRGeneration++;
8591                 break;
8592
8593         case CTL_PR_RESERVE:
8594                 lun->flags |= CTL_LUN_PR_RESERVED;
8595                 lun->res_type = msg->pr.pr_info.res_type;
8596                 lun->pr_res_idx = msg->pr.pr_info.residx;
8597
8598                 break;
8599
8600         case CTL_PR_RELEASE:
8601                 /*
8602                  * if this isn't an exclusive access res generate UA for all
8603                  * other registrants.
8604                  */
8605                 if (lun->res_type != SPR_TYPE_EX_AC
8606                  && lun->res_type != SPR_TYPE_WR_EX) {
8607                         for (i = softc->init_min; i < softc->init_max; i++)
8608                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
8609                                         continue;
8610                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8611                 }
8612
8613                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8614                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8615                 lun->res_type = 0;
8616                 break;
8617
8618         case CTL_PR_PREEMPT:
8619                 ctl_pro_preempt_other(lun, msg);
8620                 break;
8621         case CTL_PR_CLEAR:
8622                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8623                 lun->res_type = 0;
8624                 lun->pr_key_count = 0;
8625                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8626
8627                 for (i=0; i < CTL_MAX_INITIATORS; i++) {
8628                         if (ctl_get_prkey(lun, i) == 0)
8629                                 continue;
8630                         ctl_clr_prkey(lun, i);
8631                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8632                 }
8633                 lun->PRGeneration++;
8634                 break;
8635         }
8636
8637         mtx_unlock(&lun->lun_lock);
8638 }
8639
8640 int
8641 ctl_read_write(struct ctl_scsiio *ctsio)
8642 {
8643         struct ctl_lun *lun;
8644         struct ctl_lba_len_flags *lbalen;
8645         uint64_t lba;
8646         uint32_t num_blocks;
8647         int flags, retval;
8648         int isread;
8649
8650         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8651
8652         CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
8653
8654         flags = 0;
8655         retval = CTL_RETVAL_COMPLETE;
8656
8657         isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
8658               || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
8659         switch (ctsio->cdb[0]) {
8660         case READ_6:
8661         case WRITE_6: {
8662                 struct scsi_rw_6 *cdb;
8663
8664                 cdb = (struct scsi_rw_6 *)ctsio->cdb;
8665
8666                 lba = scsi_3btoul(cdb->addr);
8667                 /* only 5 bits are valid in the most significant address byte */
8668                 lba &= 0x1fffff;
8669                 num_blocks = cdb->length;
8670                 /*
8671                  * This is correct according to SBC-2.
8672                  */
8673                 if (num_blocks == 0)
8674                         num_blocks = 256;
8675                 break;
8676         }
8677         case READ_10:
8678         case WRITE_10: {
8679                 struct scsi_rw_10 *cdb;
8680
8681                 cdb = (struct scsi_rw_10 *)ctsio->cdb;
8682                 if (cdb->byte2 & SRW10_FUA)
8683                         flags |= CTL_LLF_FUA;
8684                 if (cdb->byte2 & SRW10_DPO)
8685                         flags |= CTL_LLF_DPO;
8686                 lba = scsi_4btoul(cdb->addr);
8687                 num_blocks = scsi_2btoul(cdb->length);
8688                 break;
8689         }
8690         case WRITE_VERIFY_10: {
8691                 struct scsi_write_verify_10 *cdb;
8692
8693                 cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
8694                 flags |= CTL_LLF_FUA;
8695                 if (cdb->byte2 & SWV_DPO)
8696                         flags |= CTL_LLF_DPO;
8697                 lba = scsi_4btoul(cdb->addr);
8698                 num_blocks = scsi_2btoul(cdb->length);
8699                 break;
8700         }
8701         case READ_12:
8702         case WRITE_12: {
8703                 struct scsi_rw_12 *cdb;
8704
8705                 cdb = (struct scsi_rw_12 *)ctsio->cdb;
8706                 if (cdb->byte2 & SRW12_FUA)
8707                         flags |= CTL_LLF_FUA;
8708                 if (cdb->byte2 & SRW12_DPO)
8709                         flags |= CTL_LLF_DPO;
8710                 lba = scsi_4btoul(cdb->addr);
8711                 num_blocks = scsi_4btoul(cdb->length);
8712                 break;
8713         }
8714         case WRITE_VERIFY_12: {
8715                 struct scsi_write_verify_12 *cdb;
8716
8717                 cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
8718                 flags |= CTL_LLF_FUA;
8719                 if (cdb->byte2 & SWV_DPO)
8720                         flags |= CTL_LLF_DPO;
8721                 lba = scsi_4btoul(cdb->addr);
8722                 num_blocks = scsi_4btoul(cdb->length);
8723                 break;
8724         }
8725         case READ_16:
8726         case WRITE_16: {
8727                 struct scsi_rw_16 *cdb;
8728
8729                 cdb = (struct scsi_rw_16 *)ctsio->cdb;
8730                 if (cdb->byte2 & SRW12_FUA)
8731                         flags |= CTL_LLF_FUA;
8732                 if (cdb->byte2 & SRW12_DPO)
8733                         flags |= CTL_LLF_DPO;
8734                 lba = scsi_8btou64(cdb->addr);
8735                 num_blocks = scsi_4btoul(cdb->length);
8736                 break;
8737         }
8738         case WRITE_ATOMIC_16: {
8739                 struct scsi_rw_16 *cdb;
8740
8741                 if (lun->be_lun->atomicblock == 0) {
8742                         ctl_set_invalid_opcode(ctsio);
8743                         ctl_done((union ctl_io *)ctsio);
8744                         return (CTL_RETVAL_COMPLETE);
8745                 }
8746
8747                 cdb = (struct scsi_rw_16 *)ctsio->cdb;
8748                 if (cdb->byte2 & SRW12_FUA)
8749                         flags |= CTL_LLF_FUA;
8750                 if (cdb->byte2 & SRW12_DPO)
8751                         flags |= CTL_LLF_DPO;
8752                 lba = scsi_8btou64(cdb->addr);
8753                 num_blocks = scsi_4btoul(cdb->length);
8754                 if (num_blocks > lun->be_lun->atomicblock) {
8755                         ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
8756                             /*command*/ 1, /*field*/ 12, /*bit_valid*/ 0,
8757                             /*bit*/ 0);
8758                         ctl_done((union ctl_io *)ctsio);
8759                         return (CTL_RETVAL_COMPLETE);
8760                 }
8761                 break;
8762         }
8763         case WRITE_VERIFY_16: {
8764                 struct scsi_write_verify_16 *cdb;
8765
8766                 cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
8767                 flags |= CTL_LLF_FUA;
8768                 if (cdb->byte2 & SWV_DPO)
8769                         flags |= CTL_LLF_DPO;
8770                 lba = scsi_8btou64(cdb->addr);
8771                 num_blocks = scsi_4btoul(cdb->length);
8772                 break;
8773         }
8774         default:
8775                 /*
8776                  * We got a command we don't support.  This shouldn't
8777                  * happen, commands should be filtered out above us.
8778                  */
8779                 ctl_set_invalid_opcode(ctsio);
8780                 ctl_done((union ctl_io *)ctsio);
8781
8782                 return (CTL_RETVAL_COMPLETE);
8783                 break; /* NOTREACHED */
8784         }
8785
8786         /*
8787          * The first check is to make sure we're in bounds, the second
8788          * check is to catch wrap-around problems.  If the lba + num blocks
8789          * is less than the lba, then we've wrapped around and the block
8790          * range is invalid anyway.
8791          */
8792         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8793          || ((lba + num_blocks) < lba)) {
8794                 ctl_set_lba_out_of_range(ctsio);
8795                 ctl_done((union ctl_io *)ctsio);
8796                 return (CTL_RETVAL_COMPLETE);
8797         }
8798
8799         /*
8800          * According to SBC-3, a transfer length of 0 is not an error.
8801          * Note that this cannot happen with WRITE(6) or READ(6), since 0
8802          * translates to 256 blocks for those commands.
8803          */
8804         if (num_blocks == 0) {
8805                 ctl_set_success(ctsio);
8806                 ctl_done((union ctl_io *)ctsio);
8807                 return (CTL_RETVAL_COMPLETE);
8808         }
8809
8810         /* Set FUA and/or DPO if caches are disabled. */
8811         if (isread) {
8812                 if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
8813                     SCP_RCD) != 0)
8814                         flags |= CTL_LLF_FUA | CTL_LLF_DPO;
8815         } else {
8816                 if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
8817                     SCP_WCE) == 0)
8818                         flags |= CTL_LLF_FUA;
8819         }
8820
8821         lbalen = (struct ctl_lba_len_flags *)
8822             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8823         lbalen->lba = lba;
8824         lbalen->len = num_blocks;
8825         lbalen->flags = (isread ? CTL_LLF_READ : CTL_LLF_WRITE) | flags;
8826
8827         ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
8828         ctsio->kern_rel_offset = 0;
8829
8830         CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
8831
8832         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8833
8834         return (retval);
8835 }
8836
8837 static int
8838 ctl_cnw_cont(union ctl_io *io)
8839 {
8840         struct ctl_scsiio *ctsio;
8841         struct ctl_lun *lun;
8842         struct ctl_lba_len_flags *lbalen;
8843         int retval;
8844
8845         ctsio = &io->scsiio;
8846         ctsio->io_hdr.status = CTL_STATUS_NONE;
8847         ctsio->io_hdr.flags &= ~CTL_FLAG_IO_CONT;
8848         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8849         lbalen = (struct ctl_lba_len_flags *)
8850             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8851         lbalen->flags &= ~CTL_LLF_COMPARE;
8852         lbalen->flags |= CTL_LLF_WRITE;
8853
8854         CTL_DEBUG_PRINT(("ctl_cnw_cont: calling data_submit()\n"));
8855         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8856         return (retval);
8857 }
8858
8859 int
8860 ctl_cnw(struct ctl_scsiio *ctsio)
8861 {
8862         struct ctl_lun *lun;
8863         struct ctl_lba_len_flags *lbalen;
8864         uint64_t lba;
8865         uint32_t num_blocks;
8866         int flags, retval;
8867
8868         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8869
8870         CTL_DEBUG_PRINT(("ctl_cnw: command: %#x\n", ctsio->cdb[0]));
8871
8872         flags = 0;
8873         retval = CTL_RETVAL_COMPLETE;
8874
8875         switch (ctsio->cdb[0]) {
8876         case COMPARE_AND_WRITE: {
8877                 struct scsi_compare_and_write *cdb;
8878
8879                 cdb = (struct scsi_compare_and_write *)ctsio->cdb;
8880                 if (cdb->byte2 & SRW10_FUA)
8881                         flags |= CTL_LLF_FUA;
8882                 if (cdb->byte2 & SRW10_DPO)
8883                         flags |= CTL_LLF_DPO;
8884                 lba = scsi_8btou64(cdb->addr);
8885                 num_blocks = cdb->length;
8886                 break;
8887         }
8888         default:
8889                 /*
8890                  * We got a command we don't support.  This shouldn't
8891                  * happen, commands should be filtered out above us.
8892                  */
8893                 ctl_set_invalid_opcode(ctsio);
8894                 ctl_done((union ctl_io *)ctsio);
8895
8896                 return (CTL_RETVAL_COMPLETE);
8897                 break; /* NOTREACHED */
8898         }
8899
8900         /*
8901          * The first check is to make sure we're in bounds, the second
8902          * check is to catch wrap-around problems.  If the lba + num blocks
8903          * is less than the lba, then we've wrapped around and the block
8904          * range is invalid anyway.
8905          */
8906         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8907          || ((lba + num_blocks) < lba)) {
8908                 ctl_set_lba_out_of_range(ctsio);
8909                 ctl_done((union ctl_io *)ctsio);
8910                 return (CTL_RETVAL_COMPLETE);
8911         }
8912
8913         /*
8914          * According to SBC-3, a transfer length of 0 is not an error.
8915          */
8916         if (num_blocks == 0) {
8917                 ctl_set_success(ctsio);
8918                 ctl_done((union ctl_io *)ctsio);
8919                 return (CTL_RETVAL_COMPLETE);
8920         }
8921
8922         /* Set FUA if write cache is disabled. */
8923         if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
8924             SCP_WCE) == 0)
8925                 flags |= CTL_LLF_FUA;
8926
8927         ctsio->kern_total_len = 2 * num_blocks * lun->be_lun->blocksize;
8928         ctsio->kern_rel_offset = 0;
8929
8930         /*
8931          * Set the IO_CONT flag, so that if this I/O gets passed to
8932          * ctl_data_submit_done(), it'll get passed back to
8933          * ctl_ctl_cnw_cont() for further processing.
8934          */
8935         ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
8936         ctsio->io_cont = ctl_cnw_cont;
8937
8938         lbalen = (struct ctl_lba_len_flags *)
8939             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8940         lbalen->lba = lba;
8941         lbalen->len = num_blocks;
8942         lbalen->flags = CTL_LLF_COMPARE | flags;
8943
8944         CTL_DEBUG_PRINT(("ctl_cnw: calling data_submit()\n"));
8945         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8946         return (retval);
8947 }
8948
8949 int
8950 ctl_verify(struct ctl_scsiio *ctsio)
8951 {
8952         struct ctl_lun *lun;
8953         struct ctl_lba_len_flags *lbalen;
8954         uint64_t lba;
8955         uint32_t num_blocks;
8956         int bytchk, flags;
8957         int retval;
8958
8959         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8960
8961         CTL_DEBUG_PRINT(("ctl_verify: command: %#x\n", ctsio->cdb[0]));
8962
8963         bytchk = 0;
8964         flags = CTL_LLF_FUA;
8965         retval = CTL_RETVAL_COMPLETE;
8966
8967         switch (ctsio->cdb[0]) {
8968         case VERIFY_10: {
8969                 struct scsi_verify_10 *cdb;
8970
8971                 cdb = (struct scsi_verify_10 *)ctsio->cdb;
8972                 if (cdb->byte2 & SVFY_BYTCHK)
8973                         bytchk = 1;
8974                 if (cdb->byte2 & SVFY_DPO)
8975                         flags |= CTL_LLF_DPO;
8976                 lba = scsi_4btoul(cdb->addr);
8977                 num_blocks = scsi_2btoul(cdb->length);
8978                 break;
8979         }
8980         case VERIFY_12: {
8981                 struct scsi_verify_12 *cdb;
8982
8983                 cdb = (struct scsi_verify_12 *)ctsio->cdb;
8984                 if (cdb->byte2 & SVFY_BYTCHK)
8985                         bytchk = 1;
8986                 if (cdb->byte2 & SVFY_DPO)
8987                         flags |= CTL_LLF_DPO;
8988                 lba = scsi_4btoul(cdb->addr);
8989                 num_blocks = scsi_4btoul(cdb->length);
8990                 break;
8991         }
8992         case VERIFY_16: {
8993                 struct scsi_rw_16 *cdb;
8994
8995                 cdb = (struct scsi_rw_16 *)ctsio->cdb;
8996                 if (cdb->byte2 & SVFY_BYTCHK)
8997                         bytchk = 1;
8998                 if (cdb->byte2 & SVFY_DPO)
8999                         flags |= CTL_LLF_DPO;
9000                 lba = scsi_8btou64(cdb->addr);
9001                 num_blocks = scsi_4btoul(cdb->length);
9002                 break;
9003         }
9004         default:
9005                 /*
9006                  * We got a command we don't support.  This shouldn't
9007                  * happen, commands should be filtered out above us.
9008                  */
9009                 ctl_set_invalid_opcode(ctsio);
9010                 ctl_done((union ctl_io *)ctsio);
9011                 return (CTL_RETVAL_COMPLETE);
9012         }
9013
9014         /*
9015          * The first check is to make sure we're in bounds, the second
9016          * check is to catch wrap-around problems.  If the lba + num blocks
9017          * is less than the lba, then we've wrapped around and the block
9018          * range is invalid anyway.
9019          */
9020         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9021          || ((lba + num_blocks) < lba)) {
9022                 ctl_set_lba_out_of_range(ctsio);
9023                 ctl_done((union ctl_io *)ctsio);
9024                 return (CTL_RETVAL_COMPLETE);
9025         }
9026
9027         /*
9028          * According to SBC-3, a transfer length of 0 is not an error.
9029          */
9030         if (num_blocks == 0) {
9031                 ctl_set_success(ctsio);
9032                 ctl_done((union ctl_io *)ctsio);
9033                 return (CTL_RETVAL_COMPLETE);
9034         }
9035
9036         lbalen = (struct ctl_lba_len_flags *)
9037             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9038         lbalen->lba = lba;
9039         lbalen->len = num_blocks;
9040         if (bytchk) {
9041                 lbalen->flags = CTL_LLF_COMPARE | flags;
9042                 ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9043         } else {
9044                 lbalen->flags = CTL_LLF_VERIFY | flags;
9045                 ctsio->kern_total_len = 0;
9046         }
9047         ctsio->kern_rel_offset = 0;
9048
9049         CTL_DEBUG_PRINT(("ctl_verify: calling data_submit()\n"));
9050         retval = lun->backend->data_submit((union ctl_io *)ctsio);
9051         return (retval);
9052 }
9053
9054 int
9055 ctl_report_luns(struct ctl_scsiio *ctsio)
9056 {
9057         struct ctl_softc *softc = control_softc;
9058         struct scsi_report_luns *cdb;
9059         struct scsi_report_luns_data *lun_data;
9060         struct ctl_lun *lun, *request_lun;
9061         struct ctl_port *port;
9062         int num_luns, retval;
9063         uint32_t alloc_len, lun_datalen;
9064         int num_filled, well_known;
9065         uint32_t initidx, targ_lun_id, lun_id;
9066
9067         retval = CTL_RETVAL_COMPLETE;
9068         well_known = 0;
9069
9070         cdb = (struct scsi_report_luns *)ctsio->cdb;
9071         port = ctl_io_port(&ctsio->io_hdr);
9072
9073         CTL_DEBUG_PRINT(("ctl_report_luns\n"));
9074
9075         mtx_lock(&softc->ctl_lock);
9076         num_luns = 0;
9077         for (targ_lun_id = 0; targ_lun_id < CTL_MAX_LUNS; targ_lun_id++) {
9078                 if (ctl_lun_map_from_port(port, targ_lun_id) < CTL_MAX_LUNS)
9079                         num_luns++;
9080         }
9081         mtx_unlock(&softc->ctl_lock);
9082
9083         switch (cdb->select_report) {
9084         case RPL_REPORT_DEFAULT:
9085         case RPL_REPORT_ALL:
9086                 break;
9087         case RPL_REPORT_WELLKNOWN:
9088                 well_known = 1;
9089                 num_luns = 0;
9090                 break;
9091         default:
9092                 ctl_set_invalid_field(ctsio,
9093                                       /*sks_valid*/ 1,
9094                                       /*command*/ 1,
9095                                       /*field*/ 2,
9096                                       /*bit_valid*/ 0,
9097                                       /*bit*/ 0);
9098                 ctl_done((union ctl_io *)ctsio);
9099                 return (retval);
9100                 break; /* NOTREACHED */
9101         }
9102
9103         alloc_len = scsi_4btoul(cdb->length);
9104         /*
9105          * The initiator has to allocate at least 16 bytes for this request,
9106          * so he can at least get the header and the first LUN.  Otherwise
9107          * we reject the request (per SPC-3 rev 14, section 6.21).
9108          */
9109         if (alloc_len < (sizeof(struct scsi_report_luns_data) +
9110             sizeof(struct scsi_report_luns_lundata))) {
9111                 ctl_set_invalid_field(ctsio,
9112                                       /*sks_valid*/ 1,
9113                                       /*command*/ 1,
9114                                       /*field*/ 6,
9115                                       /*bit_valid*/ 0,
9116                                       /*bit*/ 0);
9117                 ctl_done((union ctl_io *)ctsio);
9118                 return (retval);
9119         }
9120
9121         request_lun = (struct ctl_lun *)
9122                 ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9123
9124         lun_datalen = sizeof(*lun_data) +
9125                 (num_luns * sizeof(struct scsi_report_luns_lundata));
9126
9127         ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO);
9128         lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
9129         ctsio->kern_sg_entries = 0;
9130
9131         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9132
9133         mtx_lock(&softc->ctl_lock);
9134         for (targ_lun_id = 0, num_filled = 0; targ_lun_id < CTL_MAX_LUNS && num_filled < num_luns; targ_lun_id++) {
9135                 lun_id = ctl_lun_map_from_port(port, targ_lun_id);
9136                 if (lun_id >= CTL_MAX_LUNS)
9137                         continue;
9138                 lun = softc->ctl_luns[lun_id];
9139                 if (lun == NULL)
9140                         continue;
9141
9142                 if (targ_lun_id <= 0xff) {
9143                         /*
9144                          * Peripheral addressing method, bus number 0.
9145                          */
9146                         lun_data->luns[num_filled].lundata[0] =
9147                                 RPL_LUNDATA_ATYP_PERIPH;
9148                         lun_data->luns[num_filled].lundata[1] = targ_lun_id;
9149                         num_filled++;
9150                 } else if (targ_lun_id <= 0x3fff) {
9151                         /*
9152                          * Flat addressing method.
9153                          */
9154                         lun_data->luns[num_filled].lundata[0] =
9155                                 RPL_LUNDATA_ATYP_FLAT | (targ_lun_id >> 8);
9156                         lun_data->luns[num_filled].lundata[1] =
9157                                 (targ_lun_id & 0xff);
9158                         num_filled++;
9159                 } else if (targ_lun_id <= 0xffffff) {
9160                         /*
9161                          * Extended flat addressing method.
9162                          */
9163                         lun_data->luns[num_filled].lundata[0] =
9164                             RPL_LUNDATA_ATYP_EXTLUN | 0x12;
9165                         scsi_ulto3b(targ_lun_id,
9166                             &lun_data->luns[num_filled].lundata[1]);
9167                         num_filled++;
9168                 } else {
9169                         printf("ctl_report_luns: bogus LUN number %jd, "
9170                                "skipping\n", (intmax_t)targ_lun_id);
9171                 }
9172                 /*
9173                  * According to SPC-3, rev 14 section 6.21:
9174                  *
9175                  * "The execution of a REPORT LUNS command to any valid and
9176                  * installed logical unit shall clear the REPORTED LUNS DATA
9177                  * HAS CHANGED unit attention condition for all logical
9178                  * units of that target with respect to the requesting
9179                  * initiator. A valid and installed logical unit is one
9180                  * having a PERIPHERAL QUALIFIER of 000b in the standard
9181                  * INQUIRY data (see 6.4.2)."
9182                  *
9183                  * If request_lun is NULL, the LUN this report luns command
9184                  * was issued to is either disabled or doesn't exist. In that
9185                  * case, we shouldn't clear any pending lun change unit
9186                  * attention.
9187                  */
9188                 if (request_lun != NULL) {
9189                         mtx_lock(&lun->lun_lock);
9190                         ctl_clr_ua(lun, initidx, CTL_UA_LUN_CHANGE);
9191                         mtx_unlock(&lun->lun_lock);
9192                 }
9193         }
9194         mtx_unlock(&softc->ctl_lock);
9195
9196         /*
9197          * It's quite possible that we've returned fewer LUNs than we allocated
9198          * space for.  Trim it.
9199          */
9200         lun_datalen = sizeof(*lun_data) +
9201                 (num_filled * sizeof(struct scsi_report_luns_lundata));
9202
9203         if (lun_datalen < alloc_len) {
9204                 ctsio->residual = alloc_len - lun_datalen;
9205                 ctsio->kern_data_len = lun_datalen;
9206                 ctsio->kern_total_len = lun_datalen;
9207         } else {
9208                 ctsio->residual = 0;
9209                 ctsio->kern_data_len = alloc_len;
9210                 ctsio->kern_total_len = alloc_len;
9211         }
9212         ctsio->kern_data_resid = 0;
9213         ctsio->kern_rel_offset = 0;
9214         ctsio->kern_sg_entries = 0;
9215
9216         /*
9217          * We set this to the actual data length, regardless of how much
9218          * space we actually have to return results.  If the user looks at
9219          * this value, he'll know whether or not he allocated enough space
9220          * and reissue the command if necessary.  We don't support well
9221          * known logical units, so if the user asks for that, return none.
9222          */
9223         scsi_ulto4b(lun_datalen - 8, lun_data->length);
9224
9225         /*
9226          * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
9227          * this request.
9228          */
9229         ctl_set_success(ctsio);
9230         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9231         ctsio->be_move_done = ctl_config_move_done;
9232         ctl_datamove((union ctl_io *)ctsio);
9233         return (retval);
9234 }
9235
9236 int
9237 ctl_request_sense(struct ctl_scsiio *ctsio)
9238 {
9239         struct scsi_request_sense *cdb;
9240         struct scsi_sense_data *sense_ptr;
9241         struct ctl_softc *ctl_softc;
9242         struct ctl_lun *lun;
9243         uint32_t initidx;
9244         int have_error;
9245         scsi_sense_data_type sense_format;
9246         ctl_ua_type ua_type;
9247
9248         cdb = (struct scsi_request_sense *)ctsio->cdb;
9249
9250         ctl_softc = control_softc;
9251         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9252
9253         CTL_DEBUG_PRINT(("ctl_request_sense\n"));
9254
9255         /*
9256          * Determine which sense format the user wants.
9257          */
9258         if (cdb->byte2 & SRS_DESC)
9259                 sense_format = SSD_TYPE_DESC;
9260         else
9261                 sense_format = SSD_TYPE_FIXED;
9262
9263         ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
9264         sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
9265         ctsio->kern_sg_entries = 0;
9266
9267         /*
9268          * struct scsi_sense_data, which is currently set to 256 bytes, is
9269          * larger than the largest allowed value for the length field in the
9270          * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
9271          */
9272         ctsio->residual = 0;
9273         ctsio->kern_data_len = cdb->length;
9274         ctsio->kern_total_len = cdb->length;
9275
9276         ctsio->kern_data_resid = 0;
9277         ctsio->kern_rel_offset = 0;
9278         ctsio->kern_sg_entries = 0;
9279
9280         /*
9281          * If we don't have a LUN, we don't have any pending sense.
9282          */
9283         if (lun == NULL)
9284                 goto no_sense;
9285
9286         have_error = 0;
9287         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9288         /*
9289          * Check for pending sense, and then for pending unit attentions.
9290          * Pending sense gets returned first, then pending unit attentions.
9291          */
9292         mtx_lock(&lun->lun_lock);
9293 #ifdef CTL_WITH_CA
9294         if (ctl_is_set(lun->have_ca, initidx)) {
9295                 scsi_sense_data_type stored_format;
9296
9297                 /*
9298                  * Check to see which sense format was used for the stored
9299                  * sense data.
9300                  */
9301                 stored_format = scsi_sense_type(&lun->pending_sense[initidx]);
9302
9303                 /*
9304                  * If the user requested a different sense format than the
9305                  * one we stored, then we need to convert it to the other
9306                  * format.  If we're going from descriptor to fixed format
9307                  * sense data, we may lose things in translation, depending
9308                  * on what options were used.
9309                  *
9310                  * If the stored format is SSD_TYPE_NONE (i.e. invalid),
9311                  * for some reason we'll just copy it out as-is.
9312                  */
9313                 if ((stored_format == SSD_TYPE_FIXED)
9314                  && (sense_format == SSD_TYPE_DESC))
9315                         ctl_sense_to_desc((struct scsi_sense_data_fixed *)
9316                             &lun->pending_sense[initidx],
9317                             (struct scsi_sense_data_desc *)sense_ptr);
9318                 else if ((stored_format == SSD_TYPE_DESC)
9319                       && (sense_format == SSD_TYPE_FIXED))
9320                         ctl_sense_to_fixed((struct scsi_sense_data_desc *)
9321                             &lun->pending_sense[initidx],
9322                             (struct scsi_sense_data_fixed *)sense_ptr);
9323                 else
9324                         memcpy(sense_ptr, &lun->pending_sense[initidx],
9325                                MIN(sizeof(*sense_ptr),
9326                                sizeof(lun->pending_sense[initidx])));
9327
9328                 ctl_clear_mask(lun->have_ca, initidx);
9329                 have_error = 1;
9330         } else
9331 #endif
9332         {
9333                 ua_type = ctl_build_ua(lun, initidx, sense_ptr, sense_format);
9334                 if (ua_type != CTL_UA_NONE)
9335                         have_error = 1;
9336                 if (ua_type == CTL_UA_LUN_CHANGE) {
9337                         mtx_unlock(&lun->lun_lock);
9338                         mtx_lock(&ctl_softc->ctl_lock);
9339                         ctl_clr_ua_allluns(ctl_softc, initidx, ua_type);
9340                         mtx_unlock(&ctl_softc->ctl_lock);
9341                         mtx_lock(&lun->lun_lock);
9342                 }
9343
9344         }
9345         mtx_unlock(&lun->lun_lock);
9346
9347         /*
9348          * We already have a pending error, return it.
9349          */
9350         if (have_error != 0) {
9351                 /*
9352                  * We report the SCSI status as OK, since the status of the
9353                  * request sense command itself is OK.
9354                  * We report 0 for the sense length, because we aren't doing
9355                  * autosense in this case.  We're reporting sense as
9356                  * parameter data.
9357                  */
9358                 ctl_set_success(ctsio);
9359                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9360                 ctsio->be_move_done = ctl_config_move_done;
9361                 ctl_datamove((union ctl_io *)ctsio);
9362                 return (CTL_RETVAL_COMPLETE);
9363         }
9364
9365 no_sense:
9366
9367         /*
9368          * No sense information to report, so we report that everything is
9369          * okay.
9370          */
9371         ctl_set_sense_data(sense_ptr,
9372                            lun,
9373                            sense_format,
9374                            /*current_error*/ 1,
9375                            /*sense_key*/ SSD_KEY_NO_SENSE,
9376                            /*asc*/ 0x00,
9377                            /*ascq*/ 0x00,
9378                            SSD_ELEM_NONE);
9379
9380         /*
9381          * We report 0 for the sense length, because we aren't doing
9382          * autosense in this case.  We're reporting sense as parameter data.
9383          */
9384         ctl_set_success(ctsio);
9385         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9386         ctsio->be_move_done = ctl_config_move_done;
9387         ctl_datamove((union ctl_io *)ctsio);
9388         return (CTL_RETVAL_COMPLETE);
9389 }
9390
9391 int
9392 ctl_tur(struct ctl_scsiio *ctsio)
9393 {
9394
9395         CTL_DEBUG_PRINT(("ctl_tur\n"));
9396
9397         ctl_set_success(ctsio);
9398         ctl_done((union ctl_io *)ctsio);
9399
9400         return (CTL_RETVAL_COMPLETE);
9401 }
9402
9403 /*
9404  * SCSI VPD page 0x00, the Supported VPD Pages page.
9405  */
9406 static int
9407 ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
9408 {
9409         struct scsi_vpd_supported_pages *pages;
9410         int sup_page_size;
9411         struct ctl_lun *lun;
9412         int p;
9413
9414         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9415
9416         sup_page_size = sizeof(struct scsi_vpd_supported_pages) *
9417             SCSI_EVPD_NUM_SUPPORTED_PAGES;
9418         ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO);
9419         pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
9420         ctsio->kern_sg_entries = 0;
9421
9422         if (sup_page_size < alloc_len) {
9423                 ctsio->residual = alloc_len - sup_page_size;
9424                 ctsio->kern_data_len = sup_page_size;
9425                 ctsio->kern_total_len = sup_page_size;
9426         } else {
9427                 ctsio->residual = 0;
9428                 ctsio->kern_data_len = alloc_len;
9429                 ctsio->kern_total_len = alloc_len;
9430         }
9431         ctsio->kern_data_resid = 0;
9432         ctsio->kern_rel_offset = 0;
9433         ctsio->kern_sg_entries = 0;
9434
9435         /*
9436          * The control device is always connected.  The disk device, on the
9437          * other hand, may not be online all the time.  Need to change this
9438          * to figure out whether the disk device is actually online or not.
9439          */
9440         if (lun != NULL)
9441                 pages->device = (SID_QUAL_LU_CONNECTED << 5) |
9442                                 lun->be_lun->lun_type;
9443         else
9444                 pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9445
9446         p = 0;
9447         /* Supported VPD pages */
9448         pages->page_list[p++] = SVPD_SUPPORTED_PAGES;
9449         /* Serial Number */
9450         pages->page_list[p++] = SVPD_UNIT_SERIAL_NUMBER;
9451         /* Device Identification */
9452         pages->page_list[p++] = SVPD_DEVICE_ID;
9453         /* Extended INQUIRY Data */
9454         pages->page_list[p++] = SVPD_EXTENDED_INQUIRY_DATA;
9455         /* Mode Page Policy */
9456         pages->page_list[p++] = SVPD_MODE_PAGE_POLICY;
9457         /* SCSI Ports */
9458         pages->page_list[p++] = SVPD_SCSI_PORTS;
9459         /* Third-party Copy */
9460         pages->page_list[p++] = SVPD_SCSI_TPC;
9461         if (lun != NULL && lun->be_lun->lun_type == T_DIRECT) {
9462                 /* Block limits */
9463                 pages->page_list[p++] = SVPD_BLOCK_LIMITS;
9464                 /* Block Device Characteristics */
9465                 pages->page_list[p++] = SVPD_BDC;
9466                 /* Logical Block Provisioning */
9467                 pages->page_list[p++] = SVPD_LBP;
9468         }
9469         pages->length = p;
9470
9471         ctl_set_success(ctsio);
9472         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9473         ctsio->be_move_done = ctl_config_move_done;
9474         ctl_datamove((union ctl_io *)ctsio);
9475         return (CTL_RETVAL_COMPLETE);
9476 }
9477
9478 /*
9479  * SCSI VPD page 0x80, the Unit Serial Number page.
9480  */
9481 static int
9482 ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
9483 {
9484         struct scsi_vpd_unit_serial_number *sn_ptr;
9485         struct ctl_lun *lun;
9486         int data_len;
9487
9488         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9489
9490         data_len = 4 + CTL_SN_LEN;
9491         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9492         sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
9493         if (data_len < alloc_len) {
9494                 ctsio->residual = alloc_len - data_len;
9495                 ctsio->kern_data_len = data_len;
9496                 ctsio->kern_total_len = data_len;
9497         } else {
9498                 ctsio->residual = 0;
9499                 ctsio->kern_data_len = alloc_len;
9500                 ctsio->kern_total_len = alloc_len;
9501         }
9502         ctsio->kern_data_resid = 0;
9503         ctsio->kern_rel_offset = 0;
9504         ctsio->kern_sg_entries = 0;
9505
9506         /*
9507          * The control device is always connected.  The disk device, on the
9508          * other hand, may not be online all the time.  Need to change this
9509          * to figure out whether the disk device is actually online or not.
9510          */
9511         if (lun != NULL)
9512                 sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9513                                   lun->be_lun->lun_type;
9514         else
9515                 sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9516
9517         sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9518         sn_ptr->length = CTL_SN_LEN;
9519         /*
9520          * If we don't have a LUN, we just leave the serial number as
9521          * all spaces.
9522          */
9523         if (lun != NULL) {
9524                 strncpy((char *)sn_ptr->serial_num,
9525                         (char *)lun->be_lun->serial_num, CTL_SN_LEN);
9526         } else
9527                 memset(sn_ptr->serial_num, 0x20, CTL_SN_LEN);
9528
9529         ctl_set_success(ctsio);
9530         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9531         ctsio->be_move_done = ctl_config_move_done;
9532         ctl_datamove((union ctl_io *)ctsio);
9533         return (CTL_RETVAL_COMPLETE);
9534 }
9535
9536
9537 /*
9538  * SCSI VPD page 0x86, the Extended INQUIRY Data page.
9539  */
9540 static int
9541 ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len)
9542 {
9543         struct scsi_vpd_extended_inquiry_data *eid_ptr;
9544         struct ctl_lun *lun;
9545         int data_len;
9546
9547         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9548
9549         data_len = sizeof(struct scsi_vpd_extended_inquiry_data);
9550         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9551         eid_ptr = (struct scsi_vpd_extended_inquiry_data *)ctsio->kern_data_ptr;
9552         ctsio->kern_sg_entries = 0;
9553
9554         if (data_len < alloc_len) {
9555                 ctsio->residual = alloc_len - data_len;
9556                 ctsio->kern_data_len = data_len;
9557                 ctsio->kern_total_len = data_len;
9558         } else {
9559                 ctsio->residual = 0;
9560                 ctsio->kern_data_len = alloc_len;
9561                 ctsio->kern_total_len = alloc_len;
9562         }
9563         ctsio->kern_data_resid = 0;
9564         ctsio->kern_rel_offset = 0;
9565         ctsio->kern_sg_entries = 0;
9566
9567         /*
9568          * The control device is always connected.  The disk device, on the
9569          * other hand, may not be online all the time.
9570          */
9571         if (lun != NULL)
9572                 eid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9573                                      lun->be_lun->lun_type;
9574         else
9575                 eid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9576         eid_ptr->page_code = SVPD_EXTENDED_INQUIRY_DATA;
9577         scsi_ulto2b(data_len - 4, eid_ptr->page_length);
9578         /*
9579          * We support head of queue, ordered and simple tags.
9580          */
9581         eid_ptr->flags2 = SVPD_EID_HEADSUP | SVPD_EID_ORDSUP | SVPD_EID_SIMPSUP;
9582         /*
9583          * Volatile cache supported.
9584          */
9585         eid_ptr->flags3 = SVPD_EID_V_SUP;
9586
9587         /*
9588          * This means that we clear the REPORTED LUNS DATA HAS CHANGED unit
9589          * attention for a particular IT nexus on all LUNs once we report
9590          * it to that nexus once.  This bit is required as of SPC-4.
9591          */
9592         eid_ptr->flags4 = SVPD_EID_LUICLT;
9593
9594         /*
9595          * XXX KDM in order to correctly answer this, we would need
9596          * information from the SIM to determine how much sense data it
9597          * can send.  So this would really be a path inquiry field, most
9598          * likely.  This can be set to a maximum of 252 according to SPC-4,
9599          * but the hardware may or may not be able to support that much.
9600          * 0 just means that the maximum sense data length is not reported.
9601          */
9602         eid_ptr->max_sense_length = 0;
9603
9604         ctl_set_success(ctsio);
9605         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9606         ctsio->be_move_done = ctl_config_move_done;
9607         ctl_datamove((union ctl_io *)ctsio);
9608         return (CTL_RETVAL_COMPLETE);
9609 }
9610
9611 static int
9612 ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len)
9613 {
9614         struct scsi_vpd_mode_page_policy *mpp_ptr;
9615         struct ctl_lun *lun;
9616         int data_len;
9617
9618         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9619
9620         data_len = sizeof(struct scsi_vpd_mode_page_policy) +
9621             sizeof(struct scsi_vpd_mode_page_policy_descr);
9622
9623         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9624         mpp_ptr = (struct scsi_vpd_mode_page_policy *)ctsio->kern_data_ptr;
9625         ctsio->kern_sg_entries = 0;
9626
9627         if (data_len < alloc_len) {
9628                 ctsio->residual = alloc_len - data_len;
9629                 ctsio->kern_data_len = data_len;
9630                 ctsio->kern_total_len = data_len;
9631         } else {
9632                 ctsio->residual = 0;
9633                 ctsio->kern_data_len = alloc_len;
9634                 ctsio->kern_total_len = alloc_len;
9635         }
9636         ctsio->kern_data_resid = 0;
9637         ctsio->kern_rel_offset = 0;
9638         ctsio->kern_sg_entries = 0;
9639
9640         /*
9641          * The control device is always connected.  The disk device, on the
9642          * other hand, may not be online all the time.
9643          */
9644         if (lun != NULL)
9645                 mpp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9646                                      lun->be_lun->lun_type;
9647         else
9648                 mpp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9649         mpp_ptr->page_code = SVPD_MODE_PAGE_POLICY;
9650         scsi_ulto2b(data_len - 4, mpp_ptr->page_length);
9651         mpp_ptr->descr[0].page_code = 0x3f;
9652         mpp_ptr->descr[0].subpage_code = 0xff;
9653         mpp_ptr->descr[0].policy = SVPD_MPP_SHARED;
9654
9655         ctl_set_success(ctsio);
9656         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9657         ctsio->be_move_done = ctl_config_move_done;
9658         ctl_datamove((union ctl_io *)ctsio);
9659         return (CTL_RETVAL_COMPLETE);
9660 }
9661
9662 /*
9663  * SCSI VPD page 0x83, the Device Identification page.
9664  */
9665 static int
9666 ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
9667 {
9668         struct scsi_vpd_device_id *devid_ptr;
9669         struct scsi_vpd_id_descriptor *desc;
9670         struct ctl_softc *softc;
9671         struct ctl_lun *lun;
9672         struct ctl_port *port;
9673         int data_len;
9674         uint8_t proto;
9675
9676         softc = control_softc;
9677
9678         port = ctl_io_port(&ctsio->io_hdr);
9679         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9680
9681         data_len = sizeof(struct scsi_vpd_device_id) +
9682             sizeof(struct scsi_vpd_id_descriptor) +
9683                 sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
9684             sizeof(struct scsi_vpd_id_descriptor) +
9685                 sizeof(struct scsi_vpd_id_trgt_port_grp_id);
9686         if (lun && lun->lun_devid)
9687                 data_len += lun->lun_devid->len;
9688         if (port && port->port_devid)
9689                 data_len += port->port_devid->len;
9690         if (port && port->target_devid)
9691                 data_len += port->target_devid->len;
9692
9693         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9694         devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
9695         ctsio->kern_sg_entries = 0;
9696
9697         if (data_len < alloc_len) {
9698                 ctsio->residual = alloc_len - data_len;
9699                 ctsio->kern_data_len = data_len;
9700                 ctsio->kern_total_len = data_len;
9701         } else {
9702                 ctsio->residual = 0;
9703                 ctsio->kern_data_len = alloc_len;
9704                 ctsio->kern_total_len = alloc_len;
9705         }
9706         ctsio->kern_data_resid = 0;
9707         ctsio->kern_rel_offset = 0;
9708         ctsio->kern_sg_entries = 0;
9709
9710         /*
9711          * The control device is always connected.  The disk device, on the
9712          * other hand, may not be online all the time.
9713          */
9714         if (lun != NULL)
9715                 devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9716                                      lun->be_lun->lun_type;
9717         else
9718                 devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9719         devid_ptr->page_code = SVPD_DEVICE_ID;
9720         scsi_ulto2b(data_len - 4, devid_ptr->length);
9721
9722         if (port && port->port_type == CTL_PORT_FC)
9723                 proto = SCSI_PROTO_FC << 4;
9724         else if (port && port->port_type == CTL_PORT_ISCSI)
9725                 proto = SCSI_PROTO_ISCSI << 4;
9726         else
9727                 proto = SCSI_PROTO_SPI << 4;
9728         desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
9729
9730         /*
9731          * We're using a LUN association here.  i.e., this device ID is a
9732          * per-LUN identifier.
9733          */
9734         if (lun && lun->lun_devid) {
9735                 memcpy(desc, lun->lun_devid->data, lun->lun_devid->len);
9736                 desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9737                     lun->lun_devid->len);
9738         }
9739
9740         /*
9741          * This is for the WWPN which is a port association.
9742          */
9743         if (port && port->port_devid) {
9744                 memcpy(desc, port->port_devid->data, port->port_devid->len);
9745                 desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9746                     port->port_devid->len);
9747         }
9748
9749         /*
9750          * This is for the Relative Target Port(type 4h) identifier
9751          */
9752         desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9753         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9754             SVPD_ID_TYPE_RELTARG;
9755         desc->length = 4;
9756         scsi_ulto2b(ctsio->io_hdr.nexus.targ_port, &desc->identifier[2]);
9757         desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9758             sizeof(struct scsi_vpd_id_rel_trgt_port_id));
9759
9760         /*
9761          * This is for the Target Port Group(type 5h) identifier
9762          */
9763         desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9764         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9765             SVPD_ID_TYPE_TPORTGRP;
9766         desc->length = 4;
9767         scsi_ulto2b(ctsio->io_hdr.nexus.targ_port / softc->port_cnt + 1,
9768             &desc->identifier[2]);
9769         desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9770             sizeof(struct scsi_vpd_id_trgt_port_grp_id));
9771
9772         /*
9773          * This is for the Target identifier
9774          */
9775         if (port && port->target_devid) {
9776                 memcpy(desc, port->target_devid->data, port->target_devid->len);
9777         }
9778
9779         ctl_set_success(ctsio);
9780         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9781         ctsio->be_move_done = ctl_config_move_done;
9782         ctl_datamove((union ctl_io *)ctsio);
9783         return (CTL_RETVAL_COMPLETE);
9784 }
9785
9786 static int
9787 ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio, int alloc_len)
9788 {
9789         struct ctl_softc *softc = control_softc;
9790         struct scsi_vpd_scsi_ports *sp;
9791         struct scsi_vpd_port_designation *pd;
9792         struct scsi_vpd_port_designation_cont *pdc;
9793         struct ctl_lun *lun;
9794         struct ctl_port *port;
9795         int data_len, num_target_ports, iid_len, id_len;
9796
9797         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9798
9799         num_target_ports = 0;
9800         iid_len = 0;
9801         id_len = 0;
9802         mtx_lock(&softc->ctl_lock);
9803         STAILQ_FOREACH(port, &softc->port_list, links) {
9804                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
9805                         continue;
9806                 if (lun != NULL &&
9807                     ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
9808                         continue;
9809                 num_target_ports++;
9810                 if (port->init_devid)
9811                         iid_len += port->init_devid->len;
9812                 if (port->port_devid)
9813                         id_len += port->port_devid->len;
9814         }
9815         mtx_unlock(&softc->ctl_lock);
9816
9817         data_len = sizeof(struct scsi_vpd_scsi_ports) +
9818             num_target_ports * (sizeof(struct scsi_vpd_port_designation) +
9819              sizeof(struct scsi_vpd_port_designation_cont)) + iid_len + id_len;
9820         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9821         sp = (struct scsi_vpd_scsi_ports *)ctsio->kern_data_ptr;
9822         ctsio->kern_sg_entries = 0;
9823
9824         if (data_len < alloc_len) {
9825                 ctsio->residual = alloc_len - data_len;
9826                 ctsio->kern_data_len = data_len;
9827                 ctsio->kern_total_len = data_len;
9828         } else {
9829                 ctsio->residual = 0;
9830                 ctsio->kern_data_len = alloc_len;
9831                 ctsio->kern_total_len = alloc_len;
9832         }
9833         ctsio->kern_data_resid = 0;
9834         ctsio->kern_rel_offset = 0;
9835         ctsio->kern_sg_entries = 0;
9836
9837         /*
9838          * The control device is always connected.  The disk device, on the
9839          * other hand, may not be online all the time.  Need to change this
9840          * to figure out whether the disk device is actually online or not.
9841          */
9842         if (lun != NULL)
9843                 sp->device = (SID_QUAL_LU_CONNECTED << 5) |
9844                                   lun->be_lun->lun_type;
9845         else
9846                 sp->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9847
9848         sp->page_code = SVPD_SCSI_PORTS;
9849         scsi_ulto2b(data_len - sizeof(struct scsi_vpd_scsi_ports),
9850             sp->page_length);
9851         pd = &sp->design[0];
9852
9853         mtx_lock(&softc->ctl_lock);
9854         STAILQ_FOREACH(port, &softc->port_list, links) {
9855                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
9856                         continue;
9857                 if (lun != NULL &&
9858                     ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
9859                         continue;
9860                 scsi_ulto2b(port->targ_port, pd->relative_port_id);
9861                 if (port->init_devid) {
9862                         iid_len = port->init_devid->len;
9863                         memcpy(pd->initiator_transportid,
9864                             port->init_devid->data, port->init_devid->len);
9865                 } else
9866                         iid_len = 0;
9867                 scsi_ulto2b(iid_len, pd->initiator_transportid_length);
9868                 pdc = (struct scsi_vpd_port_designation_cont *)
9869                     (&pd->initiator_transportid[iid_len]);
9870                 if (port->port_devid) {
9871                         id_len = port->port_devid->len;
9872                         memcpy(pdc->target_port_descriptors,
9873                             port->port_devid->data, port->port_devid->len);
9874                 } else
9875                         id_len = 0;
9876                 scsi_ulto2b(id_len, pdc->target_port_descriptors_length);
9877                 pd = (struct scsi_vpd_port_designation *)
9878                     ((uint8_t *)pdc->target_port_descriptors + id_len);
9879         }
9880         mtx_unlock(&softc->ctl_lock);
9881
9882         ctl_set_success(ctsio);
9883         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9884         ctsio->be_move_done = ctl_config_move_done;
9885         ctl_datamove((union ctl_io *)ctsio);
9886         return (CTL_RETVAL_COMPLETE);
9887 }
9888
9889 static int
9890 ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio, int alloc_len)
9891 {
9892         struct scsi_vpd_block_limits *bl_ptr;
9893         struct ctl_lun *lun;
9894         int bs;
9895
9896         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9897
9898         ctsio->kern_data_ptr = malloc(sizeof(*bl_ptr), M_CTL, M_WAITOK | M_ZERO);
9899         bl_ptr = (struct scsi_vpd_block_limits *)ctsio->kern_data_ptr;
9900         ctsio->kern_sg_entries = 0;
9901
9902         if (sizeof(*bl_ptr) < alloc_len) {
9903                 ctsio->residual = alloc_len - sizeof(*bl_ptr);
9904                 ctsio->kern_data_len = sizeof(*bl_ptr);
9905                 ctsio->kern_total_len = sizeof(*bl_ptr);
9906         } else {
9907                 ctsio->residual = 0;
9908                 ctsio->kern_data_len = alloc_len;
9909                 ctsio->kern_total_len = alloc_len;
9910         }
9911         ctsio->kern_data_resid = 0;
9912         ctsio->kern_rel_offset = 0;
9913         ctsio->kern_sg_entries = 0;
9914
9915         /*
9916          * The control device is always connected.  The disk device, on the
9917          * other hand, may not be online all the time.  Need to change this
9918          * to figure out whether the disk device is actually online or not.
9919          */
9920         if (lun != NULL)
9921                 bl_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9922                                   lun->be_lun->lun_type;
9923         else
9924                 bl_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9925
9926         bl_ptr->page_code = SVPD_BLOCK_LIMITS;
9927         scsi_ulto2b(sizeof(*bl_ptr) - 4, bl_ptr->page_length);
9928         bl_ptr->max_cmp_write_len = 0xff;
9929         scsi_ulto4b(0xffffffff, bl_ptr->max_txfer_len);
9930         if (lun != NULL) {
9931                 bs = lun->be_lun->blocksize;
9932                 scsi_ulto4b(lun->be_lun->opttxferlen, bl_ptr->opt_txfer_len);
9933                 if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
9934                         scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_lba_cnt);
9935                         scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_blk_cnt);
9936                         if (lun->be_lun->ublockexp != 0) {
9937                                 scsi_ulto4b((1 << lun->be_lun->ublockexp),
9938                                     bl_ptr->opt_unmap_grain);
9939                                 scsi_ulto4b(0x80000000 | lun->be_lun->ublockoff,
9940                                     bl_ptr->unmap_grain_align);
9941                         }
9942                 }
9943                 scsi_ulto4b(lun->be_lun->atomicblock,
9944                     bl_ptr->max_atomic_transfer_length);
9945                 scsi_ulto4b(0, bl_ptr->atomic_alignment);
9946                 scsi_ulto4b(0, bl_ptr->atomic_transfer_length_granularity);
9947         }
9948         scsi_u64to8b(UINT64_MAX, bl_ptr->max_write_same_length);
9949
9950         ctl_set_success(ctsio);
9951         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9952         ctsio->be_move_done = ctl_config_move_done;
9953         ctl_datamove((union ctl_io *)ctsio);
9954         return (CTL_RETVAL_COMPLETE);
9955 }
9956
9957 static int
9958 ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len)
9959 {
9960         struct scsi_vpd_block_device_characteristics *bdc_ptr;
9961         struct ctl_lun *lun;
9962         const char *value;
9963         u_int i;
9964
9965         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9966
9967         ctsio->kern_data_ptr = malloc(sizeof(*bdc_ptr), M_CTL, M_WAITOK | M_ZERO);
9968         bdc_ptr = (struct scsi_vpd_block_device_characteristics *)ctsio->kern_data_ptr;
9969         ctsio->kern_sg_entries = 0;
9970
9971         if (sizeof(*bdc_ptr) < alloc_len) {
9972                 ctsio->residual = alloc_len - sizeof(*bdc_ptr);
9973                 ctsio->kern_data_len = sizeof(*bdc_ptr);
9974                 ctsio->kern_total_len = sizeof(*bdc_ptr);
9975         } else {
9976                 ctsio->residual = 0;
9977                 ctsio->kern_data_len = alloc_len;
9978                 ctsio->kern_total_len = alloc_len;
9979         }
9980         ctsio->kern_data_resid = 0;
9981         ctsio->kern_rel_offset = 0;
9982         ctsio->kern_sg_entries = 0;
9983
9984         /*
9985          * The control device is always connected.  The disk device, on the
9986          * other hand, may not be online all the time.  Need to change this
9987          * to figure out whether the disk device is actually online or not.
9988          */
9989         if (lun != NULL)
9990                 bdc_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9991                                   lun->be_lun->lun_type;
9992         else
9993                 bdc_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9994         bdc_ptr->page_code = SVPD_BDC;
9995         scsi_ulto2b(sizeof(*bdc_ptr) - 4, bdc_ptr->page_length);
9996         if (lun != NULL &&
9997             (value = ctl_get_opt(&lun->be_lun->options, "rpm")) != NULL)
9998                 i = strtol(value, NULL, 0);
9999         else
10000                 i = CTL_DEFAULT_ROTATION_RATE;
10001         scsi_ulto2b(i, bdc_ptr->medium_rotation_rate);
10002         if (lun != NULL &&
10003             (value = ctl_get_opt(&lun->be_lun->options, "formfactor")) != NULL)
10004                 i = strtol(value, NULL, 0);
10005         else
10006                 i = 0;
10007         bdc_ptr->wab_wac_ff = (i & 0x0f);
10008         bdc_ptr->flags = SVPD_FUAB | SVPD_VBULS;
10009
10010         ctl_set_success(ctsio);
10011         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10012         ctsio->be_move_done = ctl_config_move_done;
10013         ctl_datamove((union ctl_io *)ctsio);
10014         return (CTL_RETVAL_COMPLETE);
10015 }
10016
10017 static int
10018 ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len)
10019 {
10020         struct scsi_vpd_logical_block_prov *lbp_ptr;
10021         struct ctl_lun *lun;
10022
10023         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10024
10025         ctsio->kern_data_ptr = malloc(sizeof(*lbp_ptr), M_CTL, M_WAITOK | M_ZERO);
10026         lbp_ptr = (struct scsi_vpd_logical_block_prov *)ctsio->kern_data_ptr;
10027         ctsio->kern_sg_entries = 0;
10028
10029         if (sizeof(*lbp_ptr) < alloc_len) {
10030                 ctsio->residual = alloc_len - sizeof(*lbp_ptr);
10031                 ctsio->kern_data_len = sizeof(*lbp_ptr);
10032                 ctsio->kern_total_len = sizeof(*lbp_ptr);
10033         } else {
10034                 ctsio->residual = 0;
10035                 ctsio->kern_data_len = alloc_len;
10036                 ctsio->kern_total_len = alloc_len;
10037         }
10038         ctsio->kern_data_resid = 0;
10039         ctsio->kern_rel_offset = 0;
10040         ctsio->kern_sg_entries = 0;
10041
10042         /*
10043          * The control device is always connected.  The disk device, on the
10044          * other hand, may not be online all the time.  Need to change this
10045          * to figure out whether the disk device is actually online or not.
10046          */
10047         if (lun != NULL)
10048                 lbp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10049                                   lun->be_lun->lun_type;
10050         else
10051                 lbp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10052
10053         lbp_ptr->page_code = SVPD_LBP;
10054         scsi_ulto2b(sizeof(*lbp_ptr) - 4, lbp_ptr->page_length);
10055         lbp_ptr->threshold_exponent = CTL_LBP_EXPONENT;
10056         if (lun != NULL && lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
10057                 lbp_ptr->flags = SVPD_LBP_UNMAP | SVPD_LBP_WS16 |
10058                     SVPD_LBP_WS10 | SVPD_LBP_RZ | SVPD_LBP_ANC_SUP;
10059                 lbp_ptr->prov_type = SVPD_LBP_THIN;
10060         }
10061
10062         ctl_set_success(ctsio);
10063         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10064         ctsio->be_move_done = ctl_config_move_done;
10065         ctl_datamove((union ctl_io *)ctsio);
10066         return (CTL_RETVAL_COMPLETE);
10067 }
10068
10069 /*
10070  * INQUIRY with the EVPD bit set.
10071  */
10072 static int
10073 ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
10074 {
10075         struct ctl_lun *lun;
10076         struct scsi_inquiry *cdb;
10077         int alloc_len, retval;
10078
10079         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10080         cdb = (struct scsi_inquiry *)ctsio->cdb;
10081         alloc_len = scsi_2btoul(cdb->length);
10082
10083         switch (cdb->page_code) {
10084         case SVPD_SUPPORTED_PAGES:
10085                 retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
10086                 break;
10087         case SVPD_UNIT_SERIAL_NUMBER:
10088                 retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
10089                 break;
10090         case SVPD_DEVICE_ID:
10091                 retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
10092                 break;
10093         case SVPD_EXTENDED_INQUIRY_DATA:
10094                 retval = ctl_inquiry_evpd_eid(ctsio, alloc_len);
10095                 break;
10096         case SVPD_MODE_PAGE_POLICY:
10097                 retval = ctl_inquiry_evpd_mpp(ctsio, alloc_len);
10098                 break;
10099         case SVPD_SCSI_PORTS:
10100                 retval = ctl_inquiry_evpd_scsi_ports(ctsio, alloc_len);
10101                 break;
10102         case SVPD_SCSI_TPC:
10103                 retval = ctl_inquiry_evpd_tpc(ctsio, alloc_len);
10104                 break;
10105         case SVPD_BLOCK_LIMITS:
10106                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
10107                         goto err;
10108                 retval = ctl_inquiry_evpd_block_limits(ctsio, alloc_len);
10109                 break;
10110         case SVPD_BDC:
10111                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
10112                         goto err;
10113                 retval = ctl_inquiry_evpd_bdc(ctsio, alloc_len);
10114                 break;
10115         case SVPD_LBP:
10116                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
10117                         goto err;
10118                 retval = ctl_inquiry_evpd_lbp(ctsio, alloc_len);
10119                 break;
10120         default:
10121 err:
10122                 ctl_set_invalid_field(ctsio,
10123                                       /*sks_valid*/ 1,
10124                                       /*command*/ 1,
10125                                       /*field*/ 2,
10126                                       /*bit_valid*/ 0,
10127                                       /*bit*/ 0);
10128                 ctl_done((union ctl_io *)ctsio);
10129                 retval = CTL_RETVAL_COMPLETE;
10130                 break;
10131         }
10132
10133         return (retval);
10134 }
10135
10136 /*
10137  * Standard INQUIRY data.
10138  */
10139 static int
10140 ctl_inquiry_std(struct ctl_scsiio *ctsio)
10141 {
10142         struct scsi_inquiry_data *inq_ptr;
10143         struct scsi_inquiry *cdb;
10144         struct ctl_softc *softc;
10145         struct ctl_port *port;
10146         struct ctl_lun *lun;
10147         char *val;
10148         uint32_t alloc_len, data_len;
10149         ctl_port_type port_type;
10150
10151         softc = control_softc;
10152
10153         /*
10154          * Figure out whether we're talking to a Fibre Channel port or not.
10155          * We treat the ioctl front end, and any SCSI adapters, as packetized
10156          * SCSI front ends.
10157          */
10158         port = ctl_io_port(&ctsio->io_hdr);
10159         if (port != NULL)
10160                 port_type = port->port_type;
10161         else
10162                 port_type = CTL_PORT_SCSI;
10163         if (port_type == CTL_PORT_IOCTL || port_type == CTL_PORT_INTERNAL)
10164                 port_type = CTL_PORT_SCSI;
10165
10166         lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10167         cdb = (struct scsi_inquiry *)ctsio->cdb;
10168         alloc_len = scsi_2btoul(cdb->length);
10169
10170         /*
10171          * We malloc the full inquiry data size here and fill it
10172          * in.  If the user only asks for less, we'll give him
10173          * that much.
10174          */
10175         data_len = offsetof(struct scsi_inquiry_data, vendor_specific1);
10176         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10177         inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
10178         ctsio->kern_sg_entries = 0;
10179         ctsio->kern_data_resid = 0;
10180         ctsio->kern_rel_offset = 0;
10181
10182         if (data_len < alloc_len) {
10183                 ctsio->residual = alloc_len - data_len;
10184                 ctsio->kern_data_len = data_len;
10185                 ctsio->kern_total_len = data_len;
10186         } else {
10187                 ctsio->residual = 0;
10188                 ctsio->kern_data_len = alloc_len;
10189                 ctsio->kern_total_len = alloc_len;
10190         }
10191
10192         if (lun != NULL) {
10193                 if ((lun->flags & CTL_LUN_PRIMARY_SC) ||
10194                     softc->ha_link >= CTL_HA_LINK_UNKNOWN) {
10195                         inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10196                             lun->be_lun->lun_type;
10197                 } else {
10198                         inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) |
10199                             lun->be_lun->lun_type;
10200                 }
10201         } else
10202                 inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE;
10203
10204         /* RMB in byte 2 is 0 */
10205         inq_ptr->version = SCSI_REV_SPC4;
10206
10207         /*
10208          * According to SAM-3, even if a device only supports a single
10209          * level of LUN addressing, it should still set the HISUP bit:
10210          *
10211          * 4.9.1 Logical unit numbers overview
10212          *
10213          * All logical unit number formats described in this standard are
10214          * hierarchical in structure even when only a single level in that
10215          * hierarchy is used. The HISUP bit shall be set to one in the
10216          * standard INQUIRY data (see SPC-2) when any logical unit number
10217          * format described in this standard is used.  Non-hierarchical
10218          * formats are outside the scope of this standard.
10219          *
10220          * Therefore we set the HiSup bit here.
10221          *
10222          * The reponse format is 2, per SPC-3.
10223          */
10224         inq_ptr->response_format = SID_HiSup | 2;
10225
10226         inq_ptr->additional_length = data_len -
10227             (offsetof(struct scsi_inquiry_data, additional_length) + 1);
10228         CTL_DEBUG_PRINT(("additional_length = %d\n",
10229                          inq_ptr->additional_length));
10230
10231         inq_ptr->spc3_flags = SPC3_SID_3PC | SPC3_SID_TPGS_IMPLICIT;
10232         /* 16 bit addressing */
10233         if (port_type == CTL_PORT_SCSI)
10234                 inq_ptr->spc2_flags = SPC2_SID_ADDR16;
10235         /* XXX set the SID_MultiP bit here if we're actually going to
10236            respond on multiple ports */
10237         inq_ptr->spc2_flags |= SPC2_SID_MultiP;
10238
10239         /* 16 bit data bus, synchronous transfers */
10240         if (port_type == CTL_PORT_SCSI)
10241                 inq_ptr->flags = SID_WBus16 | SID_Sync;
10242         /*
10243          * XXX KDM do we want to support tagged queueing on the control
10244          * device at all?
10245          */
10246         if ((lun == NULL)
10247          || (lun->be_lun->lun_type != T_PROCESSOR))
10248                 inq_ptr->flags |= SID_CmdQue;
10249         /*
10250          * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
10251          * We have 8 bytes for the vendor name, and 16 bytes for the device
10252          * name and 4 bytes for the revision.
10253          */
10254         if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10255             "vendor")) == NULL) {
10256                 strncpy(inq_ptr->vendor, CTL_VENDOR, sizeof(inq_ptr->vendor));
10257         } else {
10258                 memset(inq_ptr->vendor, ' ', sizeof(inq_ptr->vendor));
10259                 strncpy(inq_ptr->vendor, val,
10260                     min(sizeof(inq_ptr->vendor), strlen(val)));
10261         }
10262         if (lun == NULL) {
10263                 strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10264                     sizeof(inq_ptr->product));
10265         } else if ((val = ctl_get_opt(&lun->be_lun->options, "product")) == NULL) {
10266                 switch (lun->be_lun->lun_type) {
10267                 case T_DIRECT:
10268                         strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10269                             sizeof(inq_ptr->product));
10270                         break;
10271                 case T_PROCESSOR:
10272                         strncpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT,
10273                             sizeof(inq_ptr->product));
10274                         break;
10275                 default:
10276                         strncpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT,
10277                             sizeof(inq_ptr->product));
10278                         break;
10279                 }
10280         } else {
10281                 memset(inq_ptr->product, ' ', sizeof(inq_ptr->product));
10282                 strncpy(inq_ptr->product, val,
10283                     min(sizeof(inq_ptr->product), strlen(val)));
10284         }
10285
10286         /*
10287          * XXX make this a macro somewhere so it automatically gets
10288          * incremented when we make changes.
10289          */
10290         if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10291             "revision")) == NULL) {
10292                 strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
10293         } else {
10294                 memset(inq_ptr->revision, ' ', sizeof(inq_ptr->revision));
10295                 strncpy(inq_ptr->revision, val,
10296                     min(sizeof(inq_ptr->revision), strlen(val)));
10297         }
10298
10299         /*
10300          * For parallel SCSI, we support double transition and single
10301          * transition clocking.  We also support QAS (Quick Arbitration
10302          * and Selection) and Information Unit transfers on both the
10303          * control and array devices.
10304          */
10305         if (port_type == CTL_PORT_SCSI)
10306                 inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
10307                                     SID_SPI_IUS;
10308
10309         /* SAM-5 (no version claimed) */
10310         scsi_ulto2b(0x00A0, inq_ptr->version1);
10311         /* SPC-4 (no version claimed) */
10312         scsi_ulto2b(0x0460, inq_ptr->version2);
10313         if (port_type == CTL_PORT_FC) {
10314                 /* FCP-2 ANSI INCITS.350:2003 */
10315                 scsi_ulto2b(0x0917, inq_ptr->version3);
10316         } else if (port_type == CTL_PORT_SCSI) {
10317                 /* SPI-4 ANSI INCITS.362:200x */
10318                 scsi_ulto2b(0x0B56, inq_ptr->version3);
10319         } else if (port_type == CTL_PORT_ISCSI) {
10320                 /* iSCSI (no version claimed) */
10321                 scsi_ulto2b(0x0960, inq_ptr->version3);
10322         } else if (port_type == CTL_PORT_SAS) {
10323                 /* SAS (no version claimed) */
10324                 scsi_ulto2b(0x0BE0, inq_ptr->version3);
10325         }
10326
10327         if (lun == NULL) {
10328                 /* SBC-4 (no version claimed) */
10329                 scsi_ulto2b(0x0600, inq_ptr->version4);
10330         } else {
10331                 switch (lun->be_lun->lun_type) {
10332                 case T_DIRECT:
10333                         /* SBC-4 (no version claimed) */
10334                         scsi_ulto2b(0x0600, inq_ptr->version4);
10335                         break;
10336                 case T_PROCESSOR:
10337                 default:
10338                         break;
10339                 }
10340         }
10341
10342         ctl_set_success(ctsio);
10343         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10344         ctsio->be_move_done = ctl_config_move_done;
10345         ctl_datamove((union ctl_io *)ctsio);
10346         return (CTL_RETVAL_COMPLETE);
10347 }
10348
10349 int
10350 ctl_inquiry(struct ctl_scsiio *ctsio)
10351 {
10352         struct scsi_inquiry *cdb;
10353         int retval;
10354
10355         CTL_DEBUG_PRINT(("ctl_inquiry\n"));
10356
10357         cdb = (struct scsi_inquiry *)ctsio->cdb;
10358         if (cdb->byte2 & SI_EVPD)
10359                 retval = ctl_inquiry_evpd(ctsio);
10360         else if (cdb->page_code == 0)
10361                 retval = ctl_inquiry_std(ctsio);
10362         else {
10363                 ctl_set_invalid_field(ctsio,
10364                                       /*sks_valid*/ 1,
10365                                       /*command*/ 1,
10366                                       /*field*/ 2,
10367                                       /*bit_valid*/ 0,
10368                                       /*bit*/ 0);
10369                 ctl_done((union ctl_io *)ctsio);
10370                 return (CTL_RETVAL_COMPLETE);
10371         }
10372
10373         return (retval);
10374 }
10375
10376 /*
10377  * For known CDB types, parse the LBA and length.
10378  */
10379 static int
10380 ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len)
10381 {
10382         if (io->io_hdr.io_type != CTL_IO_SCSI)
10383                 return (1);
10384
10385         switch (io->scsiio.cdb[0]) {
10386         case COMPARE_AND_WRITE: {
10387                 struct scsi_compare_and_write *cdb;
10388
10389                 cdb = (struct scsi_compare_and_write *)io->scsiio.cdb;
10390
10391                 *lba = scsi_8btou64(cdb->addr);
10392                 *len = cdb->length;
10393                 break;
10394         }
10395         case READ_6:
10396         case WRITE_6: {
10397                 struct scsi_rw_6 *cdb;
10398
10399                 cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
10400
10401                 *lba = scsi_3btoul(cdb->addr);
10402                 /* only 5 bits are valid in the most significant address byte */
10403                 *lba &= 0x1fffff;
10404                 *len = cdb->length;
10405                 break;
10406         }
10407         case READ_10:
10408         case WRITE_10: {
10409                 struct scsi_rw_10 *cdb;
10410
10411                 cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
10412
10413                 *lba = scsi_4btoul(cdb->addr);
10414                 *len = scsi_2btoul(cdb->length);
10415                 break;
10416         }
10417         case WRITE_VERIFY_10: {
10418                 struct scsi_write_verify_10 *cdb;
10419
10420                 cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
10421
10422                 *lba = scsi_4btoul(cdb->addr);
10423                 *len = scsi_2btoul(cdb->length);
10424                 break;
10425         }
10426         case READ_12:
10427         case WRITE_12: {
10428                 struct scsi_rw_12 *cdb;
10429
10430                 cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
10431
10432                 *lba = scsi_4btoul(cdb->addr);
10433                 *len = scsi_4btoul(cdb->length);
10434                 break;
10435         }
10436         case WRITE_VERIFY_12: {
10437                 struct scsi_write_verify_12 *cdb;
10438
10439                 cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
10440
10441                 *lba = scsi_4btoul(cdb->addr);
10442                 *len = scsi_4btoul(cdb->length);
10443                 break;
10444         }
10445         case READ_16:
10446         case WRITE_16:
10447         case WRITE_ATOMIC_16: {
10448                 struct scsi_rw_16 *cdb;
10449
10450                 cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
10451
10452                 *lba = scsi_8btou64(cdb->addr);
10453                 *len = scsi_4btoul(cdb->length);
10454                 break;
10455         }
10456         case WRITE_VERIFY_16: {
10457                 struct scsi_write_verify_16 *cdb;
10458
10459                 cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
10460
10461                 *lba = scsi_8btou64(cdb->addr);
10462                 *len = scsi_4btoul(cdb->length);
10463                 break;
10464         }
10465         case WRITE_SAME_10: {
10466                 struct scsi_write_same_10 *cdb;
10467
10468                 cdb = (struct scsi_write_same_10 *)io->scsiio.cdb;
10469
10470                 *lba = scsi_4btoul(cdb->addr);
10471                 *len = scsi_2btoul(cdb->length);
10472                 break;
10473         }
10474         case WRITE_SAME_16: {
10475                 struct scsi_write_same_16 *cdb;
10476
10477                 cdb = (struct scsi_write_same_16 *)io->scsiio.cdb;
10478
10479                 *lba = scsi_8btou64(cdb->addr);
10480                 *len = scsi_4btoul(cdb->length);
10481                 break;
10482         }
10483         case VERIFY_10: {
10484                 struct scsi_verify_10 *cdb;
10485
10486                 cdb = (struct scsi_verify_10 *)io->scsiio.cdb;
10487
10488                 *lba = scsi_4btoul(cdb->addr);
10489                 *len = scsi_2btoul(cdb->length);
10490                 break;
10491         }
10492         case VERIFY_12: {
10493                 struct scsi_verify_12 *cdb;
10494
10495                 cdb = (struct scsi_verify_12 *)io->scsiio.cdb;
10496
10497                 *lba = scsi_4btoul(cdb->addr);
10498                 *len = scsi_4btoul(cdb->length);
10499                 break;
10500         }
10501         case VERIFY_16: {
10502                 struct scsi_verify_16 *cdb;
10503
10504                 cdb = (struct scsi_verify_16 *)io->scsiio.cdb;
10505
10506                 *lba = scsi_8btou64(cdb->addr);
10507                 *len = scsi_4btoul(cdb->length);
10508                 break;
10509         }
10510         case UNMAP: {
10511                 *lba = 0;
10512                 *len = UINT64_MAX;
10513                 break;
10514         }
10515         case SERVICE_ACTION_IN: {       /* GET LBA STATUS */
10516                 struct scsi_get_lba_status *cdb;
10517
10518                 cdb = (struct scsi_get_lba_status *)io->scsiio.cdb;
10519                 *lba = scsi_8btou64(cdb->addr);
10520                 *len = UINT32_MAX;
10521                 break;
10522         }
10523         default:
10524                 return (1);
10525                 break; /* NOTREACHED */
10526         }
10527
10528         return (0);
10529 }
10530
10531 static ctl_action
10532 ctl_extent_check_lba(uint64_t lba1, uint64_t len1, uint64_t lba2, uint64_t len2,
10533     bool seq)
10534 {
10535         uint64_t endlba1, endlba2;
10536
10537         endlba1 = lba1 + len1 - (seq ? 0 : 1);
10538         endlba2 = lba2 + len2 - 1;
10539
10540         if ((endlba1 < lba2) || (endlba2 < lba1))
10541                 return (CTL_ACTION_PASS);
10542         else
10543                 return (CTL_ACTION_BLOCK);
10544 }
10545
10546 static int
10547 ctl_extent_check_unmap(union ctl_io *io, uint64_t lba2, uint64_t len2)
10548 {
10549         struct ctl_ptr_len_flags *ptrlen;
10550         struct scsi_unmap_desc *buf, *end, *range;
10551         uint64_t lba;
10552         uint32_t len;
10553
10554         /* If not UNMAP -- go other way. */
10555         if (io->io_hdr.io_type != CTL_IO_SCSI ||
10556             io->scsiio.cdb[0] != UNMAP)
10557                 return (CTL_ACTION_ERROR);
10558
10559         /* If UNMAP without data -- block and wait for data. */
10560         ptrlen = (struct ctl_ptr_len_flags *)
10561             &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
10562         if ((io->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0 ||
10563             ptrlen->ptr == NULL)
10564                 return (CTL_ACTION_BLOCK);
10565
10566         /* UNMAP with data -- check for collision. */
10567         buf = (struct scsi_unmap_desc *)ptrlen->ptr;
10568         end = buf + ptrlen->len / sizeof(*buf);
10569         for (range = buf; range < end; range++) {
10570                 lba = scsi_8btou64(range->lba);
10571                 len = scsi_4btoul(range->length);
10572                 if ((lba < lba2 + len2) && (lba + len > lba2))
10573                         return (CTL_ACTION_BLOCK);
10574         }
10575         return (CTL_ACTION_PASS);
10576 }
10577
10578 static ctl_action
10579 ctl_extent_check(union ctl_io *io1, union ctl_io *io2, bool seq)
10580 {
10581         uint64_t lba1, lba2;
10582         uint64_t len1, len2;
10583         int retval;
10584
10585         if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
10586                 return (CTL_ACTION_ERROR);
10587
10588         retval = ctl_extent_check_unmap(io1, lba2, len2);
10589         if (retval != CTL_ACTION_ERROR)
10590                 return (retval);
10591
10592         if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
10593                 return (CTL_ACTION_ERROR);
10594
10595         if (io1->io_hdr.flags & CTL_FLAG_SERSEQ_DONE)
10596                 seq = FALSE;
10597         return (ctl_extent_check_lba(lba1, len1, lba2, len2, seq));
10598 }
10599
10600 static ctl_action
10601 ctl_extent_check_seq(union ctl_io *io1, union ctl_io *io2)
10602 {
10603         uint64_t lba1, lba2;
10604         uint64_t len1, len2;
10605
10606         if (io1->io_hdr.flags & CTL_FLAG_SERSEQ_DONE)
10607                 return (CTL_ACTION_PASS);
10608         if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
10609                 return (CTL_ACTION_ERROR);
10610         if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
10611                 return (CTL_ACTION_ERROR);
10612
10613         if (lba1 + len1 == lba2)
10614                 return (CTL_ACTION_BLOCK);
10615         return (CTL_ACTION_PASS);
10616 }
10617
10618 static ctl_action
10619 ctl_check_for_blockage(struct ctl_lun *lun, union ctl_io *pending_io,
10620     union ctl_io *ooa_io)
10621 {
10622         const struct ctl_cmd_entry *pending_entry, *ooa_entry;
10623         ctl_serialize_action *serialize_row;
10624
10625         /*
10626          * The initiator attempted multiple untagged commands at the same
10627          * time.  Can't do that.
10628          */
10629         if ((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10630          && (ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10631          && ((pending_io->io_hdr.nexus.targ_port ==
10632               ooa_io->io_hdr.nexus.targ_port)
10633           && (pending_io->io_hdr.nexus.initid ==
10634               ooa_io->io_hdr.nexus.initid))
10635          && ((ooa_io->io_hdr.flags & (CTL_FLAG_ABORT |
10636               CTL_FLAG_STATUS_SENT)) == 0))
10637                 return (CTL_ACTION_OVERLAP);
10638
10639         /*
10640          * The initiator attempted to send multiple tagged commands with
10641          * the same ID.  (It's fine if different initiators have the same
10642          * tag ID.)
10643          *
10644          * Even if all of those conditions are true, we don't kill the I/O
10645          * if the command ahead of us has been aborted.  We won't end up
10646          * sending it to the FETD, and it's perfectly legal to resend a
10647          * command with the same tag number as long as the previous
10648          * instance of this tag number has been aborted somehow.
10649          */
10650         if ((pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10651          && (ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10652          && (pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
10653          && ((pending_io->io_hdr.nexus.targ_port ==
10654               ooa_io->io_hdr.nexus.targ_port)
10655           && (pending_io->io_hdr.nexus.initid ==
10656               ooa_io->io_hdr.nexus.initid))
10657          && ((ooa_io->io_hdr.flags & (CTL_FLAG_ABORT |
10658               CTL_FLAG_STATUS_SENT)) == 0))
10659                 return (CTL_ACTION_OVERLAP_TAG);
10660
10661         /*
10662          * If we get a head of queue tag, SAM-3 says that we should
10663          * immediately execute it.
10664          *
10665          * What happens if this command would normally block for some other
10666          * reason?  e.g. a request sense with a head of queue tag
10667          * immediately after a write.  Normally that would block, but this
10668          * will result in its getting executed immediately...
10669          *
10670          * We currently return "pass" instead of "skip", so we'll end up
10671          * going through the rest of the queue to check for overlapped tags.
10672          *
10673          * XXX KDM check for other types of blockage first??
10674          */
10675         if (pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10676                 return (CTL_ACTION_PASS);
10677
10678         /*
10679          * Ordered tags have to block until all items ahead of them
10680          * have completed.  If we get called with an ordered tag, we always
10681          * block, if something else is ahead of us in the queue.
10682          */
10683         if (pending_io->scsiio.tag_type == CTL_TAG_ORDERED)
10684                 return (CTL_ACTION_BLOCK);
10685
10686         /*
10687          * Simple tags get blocked until all head of queue and ordered tags
10688          * ahead of them have completed.  I'm lumping untagged commands in
10689          * with simple tags here.  XXX KDM is that the right thing to do?
10690          */
10691         if (((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10692           || (pending_io->scsiio.tag_type == CTL_TAG_SIMPLE))
10693          && ((ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10694           || (ooa_io->scsiio.tag_type == CTL_TAG_ORDERED)))
10695                 return (CTL_ACTION_BLOCK);
10696
10697         pending_entry = ctl_get_cmd_entry(&pending_io->scsiio, NULL);
10698         ooa_entry = ctl_get_cmd_entry(&ooa_io->scsiio, NULL);
10699
10700         serialize_row = ctl_serialize_table[ooa_entry->seridx];
10701
10702         switch (serialize_row[pending_entry->seridx]) {
10703         case CTL_SER_BLOCK:
10704                 return (CTL_ACTION_BLOCK);
10705         case CTL_SER_EXTENT:
10706                 return (ctl_extent_check(ooa_io, pending_io,
10707                     (lun->be_lun && lun->be_lun->serseq == CTL_LUN_SERSEQ_ON)));
10708         case CTL_SER_EXTENTOPT:
10709                 if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT].queue_flags
10710                     & SCP_QUEUE_ALG_MASK) != SCP_QUEUE_ALG_UNRESTRICTED)
10711                         return (ctl_extent_check(ooa_io, pending_io,
10712                             (lun->be_lun &&
10713                              lun->be_lun->serseq == CTL_LUN_SERSEQ_ON)));
10714                 return (CTL_ACTION_PASS);
10715         case CTL_SER_EXTENTSEQ:
10716                 if (lun->be_lun && lun->be_lun->serseq != CTL_LUN_SERSEQ_OFF)
10717                         return (ctl_extent_check_seq(ooa_io, pending_io));
10718                 return (CTL_ACTION_PASS);
10719         case CTL_SER_PASS:
10720                 return (CTL_ACTION_PASS);
10721         case CTL_SER_BLOCKOPT:
10722                 if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT].queue_flags
10723                     & SCP_QUEUE_ALG_MASK) != SCP_QUEUE_ALG_UNRESTRICTED)
10724                         return (CTL_ACTION_BLOCK);
10725                 return (CTL_ACTION_PASS);
10726         case CTL_SER_SKIP:
10727                 return (CTL_ACTION_SKIP);
10728         default:
10729                 panic("invalid serialization value %d",
10730                       serialize_row[pending_entry->seridx]);
10731         }
10732
10733         return (CTL_ACTION_ERROR);
10734 }
10735
10736 /*
10737  * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
10738  * Assumptions:
10739  * - pending_io is generally either incoming, or on the blocked queue
10740  * - starting I/O is the I/O we want to start the check with.
10741  */
10742 static ctl_action
10743 ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
10744               union ctl_io *starting_io)
10745 {
10746         union ctl_io *ooa_io;
10747         ctl_action action;
10748
10749         mtx_assert(&lun->lun_lock, MA_OWNED);
10750
10751         /*
10752          * Run back along the OOA queue, starting with the current
10753          * blocked I/O and going through every I/O before it on the
10754          * queue.  If starting_io is NULL, we'll just end up returning
10755          * CTL_ACTION_PASS.
10756          */
10757         for (ooa_io = starting_io; ooa_io != NULL;
10758              ooa_io = (union ctl_io *)TAILQ_PREV(&ooa_io->io_hdr, ctl_ooaq,
10759              ooa_links)){
10760
10761                 /*
10762                  * This routine just checks to see whether
10763                  * cur_blocked is blocked by ooa_io, which is ahead
10764                  * of it in the queue.  It doesn't queue/dequeue
10765                  * cur_blocked.
10766                  */
10767                 action = ctl_check_for_blockage(lun, pending_io, ooa_io);
10768                 switch (action) {
10769                 case CTL_ACTION_BLOCK:
10770                 case CTL_ACTION_OVERLAP:
10771                 case CTL_ACTION_OVERLAP_TAG:
10772                 case CTL_ACTION_SKIP:
10773                 case CTL_ACTION_ERROR:
10774                         return (action);
10775                         break; /* NOTREACHED */
10776                 case CTL_ACTION_PASS:
10777                         break;
10778                 default:
10779                         panic("invalid action %d", action);
10780                         break;  /* NOTREACHED */
10781                 }
10782         }
10783
10784         return (CTL_ACTION_PASS);
10785 }
10786
10787 /*
10788  * Assumptions:
10789  * - An I/O has just completed, and has been removed from the per-LUN OOA
10790  *   queue, so some items on the blocked queue may now be unblocked.
10791  */
10792 static int
10793 ctl_check_blocked(struct ctl_lun *lun)
10794 {
10795         struct ctl_softc *softc = lun->ctl_softc;
10796         union ctl_io *cur_blocked, *next_blocked;
10797
10798         mtx_assert(&lun->lun_lock, MA_OWNED);
10799
10800         /*
10801          * Run forward from the head of the blocked queue, checking each
10802          * entry against the I/Os prior to it on the OOA queue to see if
10803          * there is still any blockage.
10804          *
10805          * We cannot use the TAILQ_FOREACH() macro, because it can't deal
10806          * with our removing a variable on it while it is traversing the
10807          * list.
10808          */
10809         for (cur_blocked = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue);
10810              cur_blocked != NULL; cur_blocked = next_blocked) {
10811                 union ctl_io *prev_ooa;
10812                 ctl_action action;
10813
10814                 next_blocked = (union ctl_io *)TAILQ_NEXT(&cur_blocked->io_hdr,
10815                                                           blocked_links);
10816
10817                 prev_ooa = (union ctl_io *)TAILQ_PREV(&cur_blocked->io_hdr,
10818                                                       ctl_ooaq, ooa_links);
10819
10820                 /*
10821                  * If cur_blocked happens to be the first item in the OOA
10822                  * queue now, prev_ooa will be NULL, and the action
10823                  * returned will just be CTL_ACTION_PASS.
10824                  */
10825                 action = ctl_check_ooa(lun, cur_blocked, prev_ooa);
10826
10827                 switch (action) {
10828                 case CTL_ACTION_BLOCK:
10829                         /* Nothing to do here, still blocked */
10830                         break;
10831                 case CTL_ACTION_OVERLAP:
10832                 case CTL_ACTION_OVERLAP_TAG:
10833                         /*
10834                          * This shouldn't happen!  In theory we've already
10835                          * checked this command for overlap...
10836                          */
10837                         break;
10838                 case CTL_ACTION_PASS:
10839                 case CTL_ACTION_SKIP: {
10840                         const struct ctl_cmd_entry *entry;
10841
10842                         /*
10843                          * The skip case shouldn't happen, this transaction
10844                          * should have never made it onto the blocked queue.
10845                          */
10846                         /*
10847                          * This I/O is no longer blocked, we can remove it
10848                          * from the blocked queue.  Since this is a TAILQ
10849                          * (doubly linked list), we can do O(1) removals
10850                          * from any place on the list.
10851                          */
10852                         TAILQ_REMOVE(&lun->blocked_queue, &cur_blocked->io_hdr,
10853                                      blocked_links);
10854                         cur_blocked->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
10855
10856                         if ((softc->ha_mode != CTL_HA_MODE_XFER) &&
10857                             (cur_blocked->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)){
10858                                 /*
10859                                  * Need to send IO back to original side to
10860                                  * run
10861                                  */
10862                                 union ctl_ha_msg msg_info;
10863
10864                                 cur_blocked->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
10865                                 msg_info.hdr.original_sc =
10866                                         cur_blocked->io_hdr.original_sc;
10867                                 msg_info.hdr.serializing_sc = cur_blocked;
10868                                 msg_info.hdr.msg_type = CTL_MSG_R2R;
10869                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
10870                                     sizeof(msg_info.hdr), M_NOWAIT);
10871                                 break;
10872                         }
10873                         entry = ctl_get_cmd_entry(&cur_blocked->scsiio, NULL);
10874
10875                         /*
10876                          * Check this I/O for LUN state changes that may
10877                          * have happened while this command was blocked.
10878                          * The LUN state may have been changed by a command
10879                          * ahead of us in the queue, so we need to re-check
10880                          * for any states that can be caused by SCSI
10881                          * commands.
10882                          */
10883                         if (ctl_scsiio_lun_check(lun, entry,
10884                                                  &cur_blocked->scsiio) == 0) {
10885                                 cur_blocked->io_hdr.flags |=
10886                                                       CTL_FLAG_IS_WAS_ON_RTR;
10887                                 ctl_enqueue_rtr(cur_blocked);
10888                         } else
10889                                 ctl_done(cur_blocked);
10890                         break;
10891                 }
10892                 default:
10893                         /*
10894                          * This probably shouldn't happen -- we shouldn't
10895                          * get CTL_ACTION_ERROR, or anything else.
10896                          */
10897                         break;
10898                 }
10899         }
10900
10901         return (CTL_RETVAL_COMPLETE);
10902 }
10903
10904 /*
10905  * This routine (with one exception) checks LUN flags that can be set by
10906  * commands ahead of us in the OOA queue.  These flags have to be checked
10907  * when a command initially comes in, and when we pull a command off the
10908  * blocked queue and are preparing to execute it.  The reason we have to
10909  * check these flags for commands on the blocked queue is that the LUN
10910  * state may have been changed by a command ahead of us while we're on the
10911  * blocked queue.
10912  *
10913  * Ordering is somewhat important with these checks, so please pay
10914  * careful attention to the placement of any new checks.
10915  */
10916 static int
10917 ctl_scsiio_lun_check(struct ctl_lun *lun,
10918     const struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
10919 {
10920         struct ctl_softc *softc = lun->ctl_softc;
10921         int retval;
10922         uint32_t residx;
10923
10924         retval = 0;
10925
10926         mtx_assert(&lun->lun_lock, MA_OWNED);
10927
10928         /*
10929          * If this shelf is a secondary shelf controller, we may have to
10930          * reject some commands disallowed by HA mode and link state.
10931          */
10932         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
10933                 if (softc->ha_link == CTL_HA_LINK_OFFLINE &&
10934                     (entry->flags & CTL_CMD_FLAG_OK_ON_UNAVAIL) == 0) {
10935                         ctl_set_lun_unavail(ctsio);
10936                         retval = 1;
10937                         goto bailout;
10938                 }
10939                 if ((lun->flags & CTL_LUN_PEER_SC_PRIMARY) == 0 &&
10940                     (entry->flags & CTL_CMD_FLAG_OK_ON_UNAVAIL) == 0) {
10941                         ctl_set_lun_transit(ctsio);
10942                         retval = 1;
10943                         goto bailout;
10944                 }
10945                 if (softc->ha_mode == CTL_HA_MODE_ACT_STBY &&
10946                     (entry->flags & CTL_CMD_FLAG_OK_ON_STANDBY) == 0) {
10947                         ctl_set_lun_standby(ctsio);
10948                         retval = 1;
10949                         goto bailout;
10950                 }
10951
10952                 /* The rest of checks are only done on executing side */
10953                 if (softc->ha_mode == CTL_HA_MODE_XFER)
10954                         goto bailout;
10955         }
10956
10957         if (entry->pattern & CTL_LUN_PAT_WRITE) {
10958                 if (lun->be_lun &&
10959                     lun->be_lun->flags & CTL_LUN_FLAG_READONLY) {
10960                         ctl_set_hw_write_protected(ctsio);
10961                         retval = 1;
10962                         goto bailout;
10963                 }
10964                 if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT]
10965                     .eca_and_aen & SCP_SWP) != 0) {
10966                         ctl_set_sense(ctsio, /*current_error*/ 1,
10967                             /*sense_key*/ SSD_KEY_DATA_PROTECT,
10968                             /*asc*/ 0x27, /*ascq*/ 0x02, SSD_ELEM_NONE);
10969                         retval = 1;
10970                         goto bailout;
10971                 }
10972         }
10973
10974         /*
10975          * Check for a reservation conflict.  If this command isn't allowed
10976          * even on reserved LUNs, and if this initiator isn't the one who
10977          * reserved us, reject the command with a reservation conflict.
10978          */
10979         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
10980         if ((lun->flags & CTL_LUN_RESERVED)
10981          && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
10982                 if (lun->res_idx != residx) {
10983                         ctl_set_reservation_conflict(ctsio);
10984                         retval = 1;
10985                         goto bailout;
10986                 }
10987         }
10988
10989         if ((lun->flags & CTL_LUN_PR_RESERVED) == 0 ||
10990             (entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV)) {
10991                 /* No reservation or command is allowed. */;
10992         } else if ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_WRESV) &&
10993             (lun->res_type == SPR_TYPE_WR_EX ||
10994              lun->res_type == SPR_TYPE_WR_EX_RO ||
10995              lun->res_type == SPR_TYPE_WR_EX_AR)) {
10996                 /* The command is allowed for Write Exclusive resv. */;
10997         } else {
10998                 /*
10999                  * if we aren't registered or it's a res holder type
11000                  * reservation and this isn't the res holder then set a
11001                  * conflict.
11002                  */
11003                 if (ctl_get_prkey(lun, residx) == 0
11004                  || (residx != lun->pr_res_idx && lun->res_type < 4)) {
11005                         ctl_set_reservation_conflict(ctsio);
11006                         retval = 1;
11007                         goto bailout;
11008                 }
11009         }
11010
11011         if ((lun->flags & CTL_LUN_OFFLINE)
11012          && ((entry->flags & CTL_CMD_FLAG_OK_ON_STANDBY) == 0)) {
11013                 ctl_set_lun_not_ready(ctsio);
11014                 retval = 1;
11015                 goto bailout;
11016         }
11017
11018         if ((lun->flags & CTL_LUN_STOPPED)
11019          && ((entry->flags & CTL_CMD_FLAG_OK_ON_STOPPED) == 0)) {
11020                 /* "Logical unit not ready, initializing cmd. required" */
11021                 ctl_set_lun_stopped(ctsio);
11022                 retval = 1;
11023                 goto bailout;
11024         }
11025
11026         if ((lun->flags & CTL_LUN_INOPERABLE)
11027          && ((entry->flags & CTL_CMD_FLAG_OK_ON_INOPERABLE) == 0)) {
11028                 /* "Medium format corrupted" */
11029                 ctl_set_medium_format_corrupted(ctsio);
11030                 retval = 1;
11031                 goto bailout;
11032         }
11033
11034 bailout:
11035         return (retval);
11036 }
11037
11038 static void
11039 ctl_failover_io(union ctl_io *io, int have_lock)
11040 {
11041         ctl_set_busy(&io->scsiio);
11042         ctl_done(io);
11043 }
11044
11045 static void
11046 ctl_failover_lun(struct ctl_lun *lun)
11047 {
11048         struct ctl_softc *softc = lun->ctl_softc;
11049         struct ctl_io_hdr *io, *next_io;
11050
11051         CTL_DEBUG_PRINT(("FAILOVER for lun %ju\n", lun->lun));
11052         if (softc->ha_mode == CTL_HA_MODE_XFER) {
11053                 TAILQ_FOREACH_SAFE(io, &lun->ooa_queue, ooa_links, next_io) {
11054                         /* We are master */
11055                         if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
11056                                 if (io->flags & CTL_FLAG_IO_ACTIVE) {
11057                                         io->flags |= CTL_FLAG_ABORT;
11058                                         io->flags |= CTL_FLAG_FAILOVER;
11059                                 } else { /* This can be only due to DATAMOVE */
11060                                         io->msg_type = CTL_MSG_DATAMOVE_DONE;
11061                                         io->flags &= ~CTL_FLAG_DMA_INPROG;
11062                                         io->flags |= CTL_FLAG_IO_ACTIVE;
11063                                         io->port_status = 31340;
11064                                         ctl_enqueue_isc((union ctl_io *)io);
11065                                 }
11066                         }
11067                         /* We are slave */
11068                         if (io->flags & CTL_FLAG_SENT_2OTHER_SC) {
11069                                 io->flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11070                                 if (io->flags & CTL_FLAG_IO_ACTIVE) {
11071                                         io->flags |= CTL_FLAG_FAILOVER;
11072                                 } else {
11073                                         ctl_set_busy(&((union ctl_io *)io)->
11074                                             scsiio);
11075                                         ctl_done((union ctl_io *)io);
11076                                 }
11077                         }
11078                 }
11079         } else { /* SERIALIZE modes */
11080                 TAILQ_FOREACH_SAFE(io, &lun->blocked_queue, blocked_links,
11081                     next_io) {
11082                         /* We are master */
11083                         if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
11084                                 TAILQ_REMOVE(&lun->blocked_queue, io,
11085                                     blocked_links);
11086                                 io->flags &= ~CTL_FLAG_BLOCKED;
11087                                 TAILQ_REMOVE(&lun->ooa_queue, io, ooa_links);
11088                                 ctl_free_io((union ctl_io *)io);
11089                         }
11090                 }
11091                 TAILQ_FOREACH_SAFE(io, &lun->ooa_queue, ooa_links, next_io) {
11092                         /* We are master */
11093                         if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
11094                                 TAILQ_REMOVE(&lun->ooa_queue, io, ooa_links);
11095                                 ctl_free_io((union ctl_io *)io);
11096                         }
11097                         /* We are slave */
11098                         if (io->flags & CTL_FLAG_SENT_2OTHER_SC) {
11099                                 io->flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11100                                 if (!(io->flags & CTL_FLAG_IO_ACTIVE)) {
11101                                         ctl_set_busy(&((union ctl_io *)io)->
11102                                             scsiio);
11103                                         ctl_done((union ctl_io *)io);
11104                                 }
11105                         }
11106                 }
11107                 ctl_check_blocked(lun);
11108         }
11109 }
11110
11111 static int
11112 ctl_scsiio_precheck(struct ctl_softc *softc, struct ctl_scsiio *ctsio)
11113 {
11114         struct ctl_lun *lun;
11115         const struct ctl_cmd_entry *entry;
11116         uint32_t initidx, targ_lun;
11117         int retval;
11118
11119         retval = 0;
11120
11121         lun = NULL;
11122
11123         targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
11124         if ((targ_lun < CTL_MAX_LUNS)
11125          && ((lun = softc->ctl_luns[targ_lun]) != NULL)) {
11126                 /*
11127                  * If the LUN is invalid, pretend that it doesn't exist.
11128                  * It will go away as soon as all pending I/O has been
11129                  * completed.
11130                  */
11131                 mtx_lock(&lun->lun_lock);
11132                 if (lun->flags & CTL_LUN_DISABLED) {
11133                         mtx_unlock(&lun->lun_lock);
11134                         lun = NULL;
11135                         ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
11136                         ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
11137                 } else {
11138                         ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = lun;
11139                         ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr =
11140                                 lun->be_lun;
11141
11142                         /*
11143                          * Every I/O goes into the OOA queue for a
11144                          * particular LUN, and stays there until completion.
11145                          */
11146 #ifdef CTL_TIME_IO
11147                         if (TAILQ_EMPTY(&lun->ooa_queue)) {
11148                                 lun->idle_time += getsbinuptime() -
11149                                     lun->last_busy;
11150                         }
11151 #endif
11152                         TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr,
11153                             ooa_links);
11154                 }
11155         } else {
11156                 ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
11157                 ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
11158         }
11159
11160         /* Get command entry and return error if it is unsuppotyed. */
11161         entry = ctl_validate_command(ctsio);
11162         if (entry == NULL) {
11163                 if (lun)
11164                         mtx_unlock(&lun->lun_lock);
11165                 return (retval);
11166         }
11167
11168         ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
11169         ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
11170
11171         /*
11172          * Check to see whether we can send this command to LUNs that don't
11173          * exist.  This should pretty much only be the case for inquiry
11174          * and request sense.  Further checks, below, really require having
11175          * a LUN, so we can't really check the command anymore.  Just put
11176          * it on the rtr queue.
11177          */
11178         if (lun == NULL) {
11179                 if (entry->flags & CTL_CMD_FLAG_OK_ON_NO_LUN) {
11180                         ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11181                         ctl_enqueue_rtr((union ctl_io *)ctsio);
11182                         return (retval);
11183                 }
11184
11185                 ctl_set_unsupported_lun(ctsio);
11186                 ctl_done((union ctl_io *)ctsio);
11187                 CTL_DEBUG_PRINT(("ctl_scsiio_precheck: bailing out due to invalid LUN\n"));
11188                 return (retval);
11189         } else {
11190                 /*
11191                  * Make sure we support this particular command on this LUN.
11192                  * e.g., we don't support writes to the control LUN.
11193                  */
11194                 if (!ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
11195                         mtx_unlock(&lun->lun_lock);
11196                         ctl_set_invalid_opcode(ctsio);
11197                         ctl_done((union ctl_io *)ctsio);
11198                         return (retval);
11199                 }
11200         }
11201
11202         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11203
11204 #ifdef CTL_WITH_CA
11205         /*
11206          * If we've got a request sense, it'll clear the contingent
11207          * allegiance condition.  Otherwise, if we have a CA condition for
11208          * this initiator, clear it, because it sent down a command other
11209          * than request sense.
11210          */
11211         if ((ctsio->cdb[0] != REQUEST_SENSE)
11212          && (ctl_is_set(lun->have_ca, initidx)))
11213                 ctl_clear_mask(lun->have_ca, initidx);
11214 #endif
11215
11216         /*
11217          * If the command has this flag set, it handles its own unit
11218          * attention reporting, we shouldn't do anything.  Otherwise we
11219          * check for any pending unit attentions, and send them back to the
11220          * initiator.  We only do this when a command initially comes in,
11221          * not when we pull it off the blocked queue.
11222          *
11223          * According to SAM-3, section 5.3.2, the order that things get
11224          * presented back to the host is basically unit attentions caused
11225          * by some sort of reset event, busy status, reservation conflicts
11226          * or task set full, and finally any other status.
11227          *
11228          * One issue here is that some of the unit attentions we report
11229          * don't fall into the "reset" category (e.g. "reported luns data
11230          * has changed").  So reporting it here, before the reservation
11231          * check, may be technically wrong.  I guess the only thing to do
11232          * would be to check for and report the reset events here, and then
11233          * check for the other unit attention types after we check for a
11234          * reservation conflict.
11235          *
11236          * XXX KDM need to fix this
11237          */
11238         if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
11239                 ctl_ua_type ua_type;
11240
11241                 ua_type = ctl_build_ua(lun, initidx, &ctsio->sense_data,
11242                     SSD_TYPE_NONE);
11243                 if (ua_type != CTL_UA_NONE) {
11244                         mtx_unlock(&lun->lun_lock);
11245                         ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
11246                         ctsio->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
11247                         ctsio->sense_len = SSD_FULL_SIZE;
11248                         ctl_done((union ctl_io *)ctsio);
11249                         return (retval);
11250                 }
11251         }
11252
11253
11254         if (ctl_scsiio_lun_check(lun, entry, ctsio) != 0) {
11255                 mtx_unlock(&lun->lun_lock);
11256                 ctl_done((union ctl_io *)ctsio);
11257                 return (retval);
11258         }
11259
11260         /*
11261          * XXX CHD this is where we want to send IO to other side if
11262          * this LUN is secondary on this SC. We will need to make a copy
11263          * of the IO and flag the IO on this side as SENT_2OTHER and the flag
11264          * the copy we send as FROM_OTHER.
11265          * We also need to stuff the address of the original IO so we can
11266          * find it easily. Something similar will need be done on the other
11267          * side so when we are done we can find the copy.
11268          */
11269         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
11270             (lun->flags & CTL_LUN_PEER_SC_PRIMARY) != 0 &&
11271             (entry->flags & CTL_CMD_FLAG_RUN_HERE) == 0) {
11272                 union ctl_ha_msg msg_info;
11273                 int isc_retval;
11274
11275                 ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11276                 ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11277                 mtx_unlock(&lun->lun_lock);
11278
11279                 msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
11280                 msg_info.hdr.original_sc = (union ctl_io *)ctsio;
11281                 msg_info.hdr.serializing_sc = NULL;
11282                 msg_info.hdr.nexus = ctsio->io_hdr.nexus;
11283                 msg_info.scsi.tag_num = ctsio->tag_num;
11284                 msg_info.scsi.tag_type = ctsio->tag_type;
11285                 msg_info.scsi.cdb_len = ctsio->cdb_len;
11286                 memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
11287
11288                 if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11289                     sizeof(msg_info.scsi) - sizeof(msg_info.scsi.sense_data),
11290                     M_WAITOK)) > CTL_HA_STATUS_SUCCESS) {
11291                         ctl_set_busy(ctsio);
11292                         ctl_done((union ctl_io *)ctsio);
11293                         return (retval);
11294                 }
11295                 return (retval);
11296         }
11297
11298         switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
11299                               (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr,
11300                               ctl_ooaq, ooa_links))) {
11301         case CTL_ACTION_BLOCK:
11302                 ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
11303                 TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
11304                                   blocked_links);
11305                 mtx_unlock(&lun->lun_lock);
11306                 return (retval);
11307         case CTL_ACTION_PASS:
11308         case CTL_ACTION_SKIP:
11309                 ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11310                 mtx_unlock(&lun->lun_lock);
11311                 ctl_enqueue_rtr((union ctl_io *)ctsio);
11312                 break;
11313         case CTL_ACTION_OVERLAP:
11314                 mtx_unlock(&lun->lun_lock);
11315                 ctl_set_overlapped_cmd(ctsio);
11316                 ctl_done((union ctl_io *)ctsio);
11317                 break;
11318         case CTL_ACTION_OVERLAP_TAG:
11319                 mtx_unlock(&lun->lun_lock);
11320                 ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
11321                 ctl_done((union ctl_io *)ctsio);
11322                 break;
11323         case CTL_ACTION_ERROR:
11324         default:
11325                 mtx_unlock(&lun->lun_lock);
11326                 ctl_set_internal_failure(ctsio,
11327                                          /*sks_valid*/ 0,
11328                                          /*retry_count*/ 0);
11329                 ctl_done((union ctl_io *)ctsio);
11330                 break;
11331         }
11332         return (retval);
11333 }
11334
11335 const struct ctl_cmd_entry *
11336 ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa)
11337 {
11338         const struct ctl_cmd_entry *entry;
11339         int service_action;
11340
11341         entry = &ctl_cmd_table[ctsio->cdb[0]];
11342         if (sa)
11343                 *sa = ((entry->flags & CTL_CMD_FLAG_SA5) != 0);
11344         if (entry->flags & CTL_CMD_FLAG_SA5) {
11345                 service_action = ctsio->cdb[1] & SERVICE_ACTION_MASK;
11346                 entry = &((const struct ctl_cmd_entry *)
11347                     entry->execute)[service_action];
11348         }
11349         return (entry);
11350 }
11351
11352 const struct ctl_cmd_entry *
11353 ctl_validate_command(struct ctl_scsiio *ctsio)
11354 {
11355         const struct ctl_cmd_entry *entry;
11356         int i, sa;
11357         uint8_t diff;
11358
11359         entry = ctl_get_cmd_entry(ctsio, &sa);
11360         if (entry->execute == NULL) {
11361                 if (sa)
11362                         ctl_set_invalid_field(ctsio,
11363                                               /*sks_valid*/ 1,
11364                                               /*command*/ 1,
11365                                               /*field*/ 1,
11366                                               /*bit_valid*/ 1,
11367                                               /*bit*/ 4);
11368                 else
11369                         ctl_set_invalid_opcode(ctsio);
11370                 ctl_done((union ctl_io *)ctsio);
11371                 return (NULL);
11372         }
11373         KASSERT(entry->length > 0,
11374             ("Not defined length for command 0x%02x/0x%02x",
11375              ctsio->cdb[0], ctsio->cdb[1]));
11376         for (i = 1; i < entry->length; i++) {
11377                 diff = ctsio->cdb[i] & ~entry->usage[i - 1];
11378                 if (diff == 0)
11379                         continue;
11380                 ctl_set_invalid_field(ctsio,
11381                                       /*sks_valid*/ 1,
11382                                       /*command*/ 1,
11383                                       /*field*/ i,
11384                                       /*bit_valid*/ 1,
11385                                       /*bit*/ fls(diff) - 1);
11386                 ctl_done((union ctl_io *)ctsio);
11387                 return (NULL);
11388         }
11389         return (entry);
11390 }
11391
11392 static int
11393 ctl_cmd_applicable(uint8_t lun_type, const struct ctl_cmd_entry *entry)
11394 {
11395
11396         switch (lun_type) {
11397         case T_PROCESSOR:
11398                 if ((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0)
11399                         return (0);
11400                 break;
11401         case T_DIRECT:
11402                 if ((entry->flags & CTL_CMD_FLAG_OK_ON_SLUN) == 0)
11403                         return (0);
11404                 break;
11405         default:
11406                 return (0);
11407         }
11408         return (1);
11409 }
11410
11411 static int
11412 ctl_scsiio(struct ctl_scsiio *ctsio)
11413 {
11414         int retval;
11415         const struct ctl_cmd_entry *entry;
11416
11417         retval = CTL_RETVAL_COMPLETE;
11418
11419         CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
11420
11421         entry = ctl_get_cmd_entry(ctsio, NULL);
11422
11423         /*
11424          * If this I/O has been aborted, just send it straight to
11425          * ctl_done() without executing it.
11426          */
11427         if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
11428                 ctl_done((union ctl_io *)ctsio);
11429                 goto bailout;
11430         }
11431
11432         /*
11433          * All the checks should have been handled by ctl_scsiio_precheck().
11434          * We should be clear now to just execute the I/O.
11435          */
11436         retval = entry->execute(ctsio);
11437
11438 bailout:
11439         return (retval);
11440 }
11441
11442 /*
11443  * Since we only implement one target right now, a bus reset simply resets
11444  * our single target.
11445  */
11446 static int
11447 ctl_bus_reset(struct ctl_softc *softc, union ctl_io *io)
11448 {
11449         return(ctl_target_reset(softc, io, CTL_UA_BUS_RESET));
11450 }
11451
11452 static int
11453 ctl_target_reset(struct ctl_softc *softc, union ctl_io *io,
11454                  ctl_ua_type ua_type)
11455 {
11456         struct ctl_port *port;
11457         struct ctl_lun *lun;
11458         int retval;
11459
11460         if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11461                 union ctl_ha_msg msg_info;
11462
11463                 msg_info.hdr.nexus = io->io_hdr.nexus;
11464                 if (ua_type==CTL_UA_TARG_RESET)
11465                         msg_info.task.task_action = CTL_TASK_TARGET_RESET;
11466                 else
11467                         msg_info.task.task_action = CTL_TASK_BUS_RESET;
11468                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11469                 msg_info.hdr.original_sc = NULL;
11470                 msg_info.hdr.serializing_sc = NULL;
11471                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11472                     sizeof(msg_info.task), M_WAITOK);
11473         }
11474         retval = 0;
11475
11476         mtx_lock(&softc->ctl_lock);
11477         port = softc->ctl_ports[io->io_hdr.nexus.targ_port];
11478         STAILQ_FOREACH(lun, &softc->lun_list, links) {
11479                 if (port != NULL &&
11480                     ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
11481                         continue;
11482                 retval += ctl_do_lun_reset(lun, io, ua_type);
11483         }
11484         mtx_unlock(&softc->ctl_lock);
11485         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11486         return (retval);
11487 }
11488
11489 /*
11490  * The LUN should always be set.  The I/O is optional, and is used to
11491  * distinguish between I/Os sent by this initiator, and by other
11492  * initiators.  We set unit attention for initiators other than this one.
11493  * SAM-3 is vague on this point.  It does say that a unit attention should
11494  * be established for other initiators when a LUN is reset (see section
11495  * 5.7.3), but it doesn't specifically say that the unit attention should
11496  * be established for this particular initiator when a LUN is reset.  Here
11497  * is the relevant text, from SAM-3 rev 8:
11498  *
11499  * 5.7.2 When a SCSI initiator port aborts its own tasks
11500  *
11501  * When a SCSI initiator port causes its own task(s) to be aborted, no
11502  * notification that the task(s) have been aborted shall be returned to
11503  * the SCSI initiator port other than the completion response for the
11504  * command or task management function action that caused the task(s) to
11505  * be aborted and notification(s) associated with related effects of the
11506  * action (e.g., a reset unit attention condition).
11507  *
11508  * XXX KDM for now, we're setting unit attention for all initiators.
11509  */
11510 static int
11511 ctl_do_lun_reset(struct ctl_lun *lun, union ctl_io *io, ctl_ua_type ua_type)
11512 {
11513         union ctl_io *xio;
11514 #if 0
11515         uint32_t initidx;
11516 #endif
11517 #ifdef CTL_WITH_CA
11518         int i;
11519 #endif
11520
11521         mtx_lock(&lun->lun_lock);
11522         /*
11523          * Run through the OOA queue and abort each I/O.
11524          */
11525         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11526              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11527                 xio->io_hdr.flags |= CTL_FLAG_ABORT | CTL_FLAG_ABORT_STATUS;
11528         }
11529
11530         /*
11531          * This version sets unit attention for every
11532          */
11533 #if 0
11534         initidx = ctl_get_initindex(&io->io_hdr.nexus);
11535         ctl_est_ua_all(lun, initidx, ua_type);
11536 #else
11537         ctl_est_ua_all(lun, -1, ua_type);
11538 #endif
11539
11540         /*
11541          * A reset (any kind, really) clears reservations established with
11542          * RESERVE/RELEASE.  It does not clear reservations established
11543          * with PERSISTENT RESERVE OUT, but we don't support that at the
11544          * moment anyway.  See SPC-2, section 5.6.  SPC-3 doesn't address
11545          * reservations made with the RESERVE/RELEASE commands, because
11546          * those commands are obsolete in SPC-3.
11547          */
11548         lun->flags &= ~CTL_LUN_RESERVED;
11549
11550 #ifdef CTL_WITH_CA
11551         for (i = 0; i < CTL_MAX_INITIATORS; i++)
11552                 ctl_clear_mask(lun->have_ca, i);
11553 #endif
11554         mtx_unlock(&lun->lun_lock);
11555
11556         return (0);
11557 }
11558
11559 static int
11560 ctl_lun_reset(struct ctl_softc *softc, union ctl_io *io)
11561 {
11562         struct ctl_lun *lun;
11563         uint32_t targ_lun;
11564         int retval;
11565
11566         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11567         mtx_lock(&softc->ctl_lock);
11568         if ((targ_lun >= CTL_MAX_LUNS) ||
11569             (lun = softc->ctl_luns[targ_lun]) == NULL) {
11570                 mtx_unlock(&softc->ctl_lock);
11571                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
11572                 return (1);
11573         }
11574         retval = ctl_do_lun_reset(lun, io, CTL_UA_LUN_RESET);
11575         mtx_unlock(&softc->ctl_lock);
11576         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11577
11578         if ((io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) == 0) {
11579                 union ctl_ha_msg msg_info;
11580
11581                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11582                 msg_info.hdr.nexus = io->io_hdr.nexus;
11583                 msg_info.task.task_action = CTL_TASK_LUN_RESET;
11584                 msg_info.hdr.original_sc = NULL;
11585                 msg_info.hdr.serializing_sc = NULL;
11586                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11587                     sizeof(msg_info.task), M_WAITOK);
11588         }
11589         return (retval);
11590 }
11591
11592 static void
11593 ctl_abort_tasks_lun(struct ctl_lun *lun, uint32_t targ_port, uint32_t init_id,
11594     int other_sc)
11595 {
11596         union ctl_io *xio;
11597
11598         mtx_assert(&lun->lun_lock, MA_OWNED);
11599
11600         /*
11601          * Run through the OOA queue and attempt to find the given I/O.
11602          * The target port, initiator ID, tag type and tag number have to
11603          * match the values that we got from the initiator.  If we have an
11604          * untagged command to abort, simply abort the first untagged command
11605          * we come to.  We only allow one untagged command at a time of course.
11606          */
11607         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11608              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11609
11610                 if ((targ_port == UINT32_MAX ||
11611                      targ_port == xio->io_hdr.nexus.targ_port) &&
11612                     (init_id == UINT32_MAX ||
11613                      init_id == xio->io_hdr.nexus.initid)) {
11614                         if (targ_port != xio->io_hdr.nexus.targ_port ||
11615                             init_id != xio->io_hdr.nexus.initid)
11616                                 xio->io_hdr.flags |= CTL_FLAG_ABORT_STATUS;
11617                         xio->io_hdr.flags |= CTL_FLAG_ABORT;
11618                         if (!other_sc && !(lun->flags & CTL_LUN_PRIMARY_SC)) {
11619                                 union ctl_ha_msg msg_info;
11620
11621                                 msg_info.hdr.nexus = xio->io_hdr.nexus;
11622                                 msg_info.task.task_action = CTL_TASK_ABORT_TASK;
11623                                 msg_info.task.tag_num = xio->scsiio.tag_num;
11624                                 msg_info.task.tag_type = xio->scsiio.tag_type;
11625                                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11626                                 msg_info.hdr.original_sc = NULL;
11627                                 msg_info.hdr.serializing_sc = NULL;
11628                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11629                                     sizeof(msg_info.task), M_NOWAIT);
11630                         }
11631                 }
11632         }
11633 }
11634
11635 static int
11636 ctl_abort_task_set(union ctl_io *io)
11637 {
11638         struct ctl_softc *softc = control_softc;
11639         struct ctl_lun *lun;
11640         uint32_t targ_lun;
11641
11642         /*
11643          * Look up the LUN.
11644          */
11645         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11646         mtx_lock(&softc->ctl_lock);
11647         if ((targ_lun >= CTL_MAX_LUNS) ||
11648             (lun = softc->ctl_luns[targ_lun]) == NULL) {
11649                 mtx_unlock(&softc->ctl_lock);
11650                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
11651                 return (1);
11652         }
11653
11654         mtx_lock(&lun->lun_lock);
11655         mtx_unlock(&softc->ctl_lock);
11656         if (io->taskio.task_action == CTL_TASK_ABORT_TASK_SET) {
11657                 ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
11658                     io->io_hdr.nexus.initid,
11659                     (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
11660         } else { /* CTL_TASK_CLEAR_TASK_SET */
11661                 ctl_abort_tasks_lun(lun, UINT32_MAX, UINT32_MAX,
11662                     (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
11663         }
11664         mtx_unlock(&lun->lun_lock);
11665         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11666         return (0);
11667 }
11668
11669 static int
11670 ctl_i_t_nexus_reset(union ctl_io *io)
11671 {
11672         struct ctl_softc *softc = control_softc;
11673         struct ctl_lun *lun;
11674         uint32_t initidx;
11675
11676         if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11677                 union ctl_ha_msg msg_info;
11678
11679                 msg_info.hdr.nexus = io->io_hdr.nexus;
11680                 msg_info.task.task_action = CTL_TASK_I_T_NEXUS_RESET;
11681                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11682                 msg_info.hdr.original_sc = NULL;
11683                 msg_info.hdr.serializing_sc = NULL;
11684                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11685                     sizeof(msg_info.task), M_WAITOK);
11686         }
11687
11688         initidx = ctl_get_initindex(&io->io_hdr.nexus);
11689         mtx_lock(&softc->ctl_lock);
11690         STAILQ_FOREACH(lun, &softc->lun_list, links) {
11691                 mtx_lock(&lun->lun_lock);
11692                 ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
11693                     io->io_hdr.nexus.initid, 1);
11694 #ifdef CTL_WITH_CA
11695                 ctl_clear_mask(lun->have_ca, initidx);
11696 #endif
11697                 if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == initidx))
11698                         lun->flags &= ~CTL_LUN_RESERVED;
11699                 ctl_est_ua(lun, initidx, CTL_UA_I_T_NEXUS_LOSS);
11700                 mtx_unlock(&lun->lun_lock);
11701         }
11702         mtx_unlock(&softc->ctl_lock);
11703         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11704         return (0);
11705 }
11706
11707 static int
11708 ctl_abort_task(union ctl_io *io)
11709 {
11710         union ctl_io *xio;
11711         struct ctl_lun *lun;
11712         struct ctl_softc *softc;
11713 #if 0
11714         struct sbuf sb;
11715         char printbuf[128];
11716 #endif
11717         int found;
11718         uint32_t targ_lun;
11719
11720         softc = control_softc;
11721         found = 0;
11722
11723         /*
11724          * Look up the LUN.
11725          */
11726         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11727         mtx_lock(&softc->ctl_lock);
11728         if ((targ_lun >= CTL_MAX_LUNS) ||
11729             (lun = softc->ctl_luns[targ_lun]) == NULL) {
11730                 mtx_unlock(&softc->ctl_lock);
11731                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
11732                 return (1);
11733         }
11734
11735 #if 0
11736         printf("ctl_abort_task: called for lun %lld, tag %d type %d\n",
11737                lun->lun, io->taskio.tag_num, io->taskio.tag_type);
11738 #endif
11739
11740         mtx_lock(&lun->lun_lock);
11741         mtx_unlock(&softc->ctl_lock);
11742         /*
11743          * Run through the OOA queue and attempt to find the given I/O.
11744          * The target port, initiator ID, tag type and tag number have to
11745          * match the values that we got from the initiator.  If we have an
11746          * untagged command to abort, simply abort the first untagged command
11747          * we come to.  We only allow one untagged command at a time of course.
11748          */
11749         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11750              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11751 #if 0
11752                 sbuf_new(&sb, printbuf, sizeof(printbuf), SBUF_FIXEDLEN);
11753
11754                 sbuf_printf(&sb, "LUN %lld tag %d type %d%s%s%s%s: ",
11755                             lun->lun, xio->scsiio.tag_num,
11756                             xio->scsiio.tag_type,
11757                             (xio->io_hdr.blocked_links.tqe_prev
11758                             == NULL) ? "" : " BLOCKED",
11759                             (xio->io_hdr.flags &
11760                             CTL_FLAG_DMA_INPROG) ? " DMA" : "",
11761                             (xio->io_hdr.flags &
11762                             CTL_FLAG_ABORT) ? " ABORT" : "",
11763                             (xio->io_hdr.flags &
11764                             CTL_FLAG_IS_WAS_ON_RTR ? " RTR" : ""));
11765                 ctl_scsi_command_string(&xio->scsiio, NULL, &sb);
11766                 sbuf_finish(&sb);
11767                 printf("%s\n", sbuf_data(&sb));
11768 #endif
11769
11770                 if ((xio->io_hdr.nexus.targ_port != io->io_hdr.nexus.targ_port)
11771                  || (xio->io_hdr.nexus.initid != io->io_hdr.nexus.initid)
11772                  || (xio->io_hdr.flags & CTL_FLAG_ABORT))
11773                         continue;
11774
11775                 /*
11776                  * If the abort says that the task is untagged, the
11777                  * task in the queue must be untagged.  Otherwise,
11778                  * we just check to see whether the tag numbers
11779                  * match.  This is because the QLogic firmware
11780                  * doesn't pass back the tag type in an abort
11781                  * request.
11782                  */
11783 #if 0
11784                 if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
11785                   && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
11786                  || (xio->scsiio.tag_num == io->taskio.tag_num))
11787 #endif
11788                 /*
11789                  * XXX KDM we've got problems with FC, because it
11790                  * doesn't send down a tag type with aborts.  So we
11791                  * can only really go by the tag number...
11792                  * This may cause problems with parallel SCSI.
11793                  * Need to figure that out!!
11794                  */
11795                 if (xio->scsiio.tag_num == io->taskio.tag_num) {
11796                         xio->io_hdr.flags |= CTL_FLAG_ABORT;
11797                         found = 1;
11798                         if ((io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) == 0 &&
11799                             !(lun->flags & CTL_LUN_PRIMARY_SC)) {
11800                                 union ctl_ha_msg msg_info;
11801
11802                                 msg_info.hdr.nexus = io->io_hdr.nexus;
11803                                 msg_info.task.task_action = CTL_TASK_ABORT_TASK;
11804                                 msg_info.task.tag_num = io->taskio.tag_num;
11805                                 msg_info.task.tag_type = io->taskio.tag_type;
11806                                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11807                                 msg_info.hdr.original_sc = NULL;
11808                                 msg_info.hdr.serializing_sc = NULL;
11809 #if 0
11810                                 printf("Sent Abort to other side\n");
11811 #endif
11812                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11813                                     sizeof(msg_info.task), M_NOWAIT);
11814                         }
11815 #if 0
11816                         printf("ctl_abort_task: found I/O to abort\n");
11817 #endif
11818                 }
11819         }
11820         mtx_unlock(&lun->lun_lock);
11821
11822         if (found == 0) {
11823                 /*
11824                  * This isn't really an error.  It's entirely possible for
11825                  * the abort and command completion to cross on the wire.
11826                  * This is more of an informative/diagnostic error.
11827                  */
11828 #if 0
11829                 printf("ctl_abort_task: ABORT sent for nonexistent I/O: "
11830                        "%u:%u:%u tag %d type %d\n",
11831                        io->io_hdr.nexus.initid,
11832                        io->io_hdr.nexus.targ_port,
11833                        io->io_hdr.nexus.targ_lun, io->taskio.tag_num,
11834                        io->taskio.tag_type);
11835 #endif
11836         }
11837         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11838         return (0);
11839 }
11840
11841 static int
11842 ctl_query_task(union ctl_io *io, int task_set)
11843 {
11844         union ctl_io *xio;
11845         struct ctl_lun *lun;
11846         struct ctl_softc *softc;
11847         int found = 0;
11848         uint32_t targ_lun;
11849
11850         softc = control_softc;
11851         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11852         mtx_lock(&softc->ctl_lock);
11853         if ((targ_lun >= CTL_MAX_LUNS) ||
11854             (lun = softc->ctl_luns[targ_lun]) == NULL) {
11855                 mtx_unlock(&softc->ctl_lock);
11856                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
11857                 return (1);
11858         }
11859         mtx_lock(&lun->lun_lock);
11860         mtx_unlock(&softc->ctl_lock);
11861         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11862              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11863
11864                 if ((xio->io_hdr.nexus.targ_port != io->io_hdr.nexus.targ_port)
11865                  || (xio->io_hdr.nexus.initid != io->io_hdr.nexus.initid)
11866                  || (xio->io_hdr.flags & CTL_FLAG_ABORT))
11867                         continue;
11868
11869                 if (task_set || xio->scsiio.tag_num == io->taskio.tag_num) {
11870                         found = 1;
11871                         break;
11872                 }
11873         }
11874         mtx_unlock(&lun->lun_lock);
11875         if (found)
11876                 io->taskio.task_status = CTL_TASK_FUNCTION_SUCCEEDED;
11877         else
11878                 io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11879         return (0);
11880 }
11881
11882 static int
11883 ctl_query_async_event(union ctl_io *io)
11884 {
11885         struct ctl_lun *lun;
11886         struct ctl_softc *softc;
11887         ctl_ua_type ua;
11888         uint32_t targ_lun, initidx;
11889
11890         softc = control_softc;
11891         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11892         mtx_lock(&softc->ctl_lock);
11893         if ((targ_lun >= CTL_MAX_LUNS) ||
11894             (lun = softc->ctl_luns[targ_lun]) == NULL) {
11895                 mtx_unlock(&softc->ctl_lock);
11896                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
11897                 return (1);
11898         }
11899         mtx_lock(&lun->lun_lock);
11900         mtx_unlock(&softc->ctl_lock);
11901         initidx = ctl_get_initindex(&io->io_hdr.nexus);
11902         ua = ctl_build_qae(lun, initidx, io->taskio.task_resp);
11903         mtx_unlock(&lun->lun_lock);
11904         if (ua != CTL_UA_NONE)
11905                 io->taskio.task_status = CTL_TASK_FUNCTION_SUCCEEDED;
11906         else
11907                 io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11908         return (0);
11909 }
11910
11911 static void
11912 ctl_run_task(union ctl_io *io)
11913 {
11914         struct ctl_softc *softc = control_softc;
11915         int retval = 1;
11916
11917         CTL_DEBUG_PRINT(("ctl_run_task\n"));
11918         KASSERT(io->io_hdr.io_type == CTL_IO_TASK,
11919             ("ctl_run_task: Unextected io_type %d\n", io->io_hdr.io_type));
11920         io->taskio.task_status = CTL_TASK_FUNCTION_NOT_SUPPORTED;
11921         bzero(io->taskio.task_resp, sizeof(io->taskio.task_resp));
11922         switch (io->taskio.task_action) {
11923         case CTL_TASK_ABORT_TASK:
11924                 retval = ctl_abort_task(io);
11925                 break;
11926         case CTL_TASK_ABORT_TASK_SET:
11927         case CTL_TASK_CLEAR_TASK_SET:
11928                 retval = ctl_abort_task_set(io);
11929                 break;
11930         case CTL_TASK_CLEAR_ACA:
11931                 break;
11932         case CTL_TASK_I_T_NEXUS_RESET:
11933                 retval = ctl_i_t_nexus_reset(io);
11934                 break;
11935         case CTL_TASK_LUN_RESET:
11936                 retval = ctl_lun_reset(softc, io);
11937                 break;
11938         case CTL_TASK_TARGET_RESET:
11939                 retval = ctl_target_reset(softc, io, CTL_UA_TARG_RESET);
11940                 break;
11941         case CTL_TASK_BUS_RESET:
11942                 retval = ctl_bus_reset(softc, io);
11943                 break;
11944         case CTL_TASK_PORT_LOGIN:
11945                 break;
11946         case CTL_TASK_PORT_LOGOUT:
11947                 break;
11948         case CTL_TASK_QUERY_TASK:
11949                 retval = ctl_query_task(io, 0);
11950                 break;
11951         case CTL_TASK_QUERY_TASK_SET:
11952                 retval = ctl_query_task(io, 1);
11953                 break;
11954         case CTL_TASK_QUERY_ASYNC_EVENT:
11955                 retval = ctl_query_async_event(io);
11956                 break;
11957         default:
11958                 printf("%s: got unknown task management event %d\n",
11959                        __func__, io->taskio.task_action);
11960                 break;
11961         }
11962         if (retval == 0)
11963                 io->io_hdr.status = CTL_SUCCESS;
11964         else
11965                 io->io_hdr.status = CTL_ERROR;
11966         ctl_done(io);
11967 }
11968
11969 /*
11970  * For HA operation.  Handle commands that come in from the other
11971  * controller.
11972  */
11973 static void
11974 ctl_handle_isc(union ctl_io *io)
11975 {
11976         int free_io;
11977         struct ctl_lun *lun;
11978         struct ctl_softc *softc;
11979         uint32_t targ_lun;
11980
11981         softc = control_softc;
11982
11983         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11984         lun = softc->ctl_luns[targ_lun];
11985
11986         switch (io->io_hdr.msg_type) {
11987         case CTL_MSG_SERIALIZE:
11988                 free_io = ctl_serialize_other_sc_cmd(&io->scsiio);
11989                 break;
11990         case CTL_MSG_R2R: {
11991                 const struct ctl_cmd_entry *entry;
11992
11993                 /*
11994                  * This is only used in SER_ONLY mode.
11995                  */
11996                 free_io = 0;
11997                 entry = ctl_get_cmd_entry(&io->scsiio, NULL);
11998                 mtx_lock(&lun->lun_lock);
11999                 if (ctl_scsiio_lun_check(lun,
12000                     entry, (struct ctl_scsiio *)io) != 0) {
12001                         mtx_unlock(&lun->lun_lock);
12002                         ctl_done(io);
12003                         break;
12004                 }
12005                 io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
12006                 mtx_unlock(&lun->lun_lock);
12007                 ctl_enqueue_rtr(io);
12008                 break;
12009         }
12010         case CTL_MSG_FINISH_IO:
12011                 if (softc->ha_mode == CTL_HA_MODE_XFER) {
12012                         free_io = 0;
12013                         ctl_done(io);
12014                 } else {
12015                         free_io = 1;
12016                         mtx_lock(&lun->lun_lock);
12017                         TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr,
12018                                      ooa_links);
12019                         ctl_check_blocked(lun);
12020                         mtx_unlock(&lun->lun_lock);
12021                 }
12022                 break;
12023         case CTL_MSG_PERS_ACTION:
12024                 ctl_hndl_per_res_out_on_other_sc(
12025                         (union ctl_ha_msg *)&io->presio.pr_msg);
12026                 free_io = 1;
12027                 break;
12028         case CTL_MSG_BAD_JUJU:
12029                 free_io = 0;
12030                 ctl_done(io);
12031                 break;
12032         case CTL_MSG_DATAMOVE:
12033                 /* Only used in XFER mode */
12034                 free_io = 0;
12035                 ctl_datamove_remote(io);
12036                 break;
12037         case CTL_MSG_DATAMOVE_DONE:
12038                 /* Only used in XFER mode */
12039                 free_io = 0;
12040                 io->scsiio.be_move_done(io);
12041                 break;
12042         case CTL_MSG_FAILOVER:
12043                 mtx_lock(&lun->lun_lock);
12044                 ctl_failover_lun(lun);
12045                 mtx_unlock(&lun->lun_lock);
12046                 free_io = 1;
12047                 break;
12048         default:
12049                 free_io = 1;
12050                 printf("%s: Invalid message type %d\n",
12051                        __func__, io->io_hdr.msg_type);
12052                 break;
12053         }
12054         if (free_io)
12055                 ctl_free_io(io);
12056
12057 }
12058
12059
12060 /*
12061  * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
12062  * there is no match.
12063  */
12064 static ctl_lun_error_pattern
12065 ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
12066 {
12067         const struct ctl_cmd_entry *entry;
12068         ctl_lun_error_pattern filtered_pattern, pattern;
12069
12070         pattern = desc->error_pattern;
12071
12072         /*
12073          * XXX KDM we need more data passed into this function to match a
12074          * custom pattern, and we actually need to implement custom pattern
12075          * matching.
12076          */
12077         if (pattern & CTL_LUN_PAT_CMD)
12078                 return (CTL_LUN_PAT_CMD);
12079
12080         if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
12081                 return (CTL_LUN_PAT_ANY);
12082
12083         entry = ctl_get_cmd_entry(ctsio, NULL);
12084
12085         filtered_pattern = entry->pattern & pattern;
12086
12087         /*
12088          * If the user requested specific flags in the pattern (e.g.
12089          * CTL_LUN_PAT_RANGE), make sure the command supports all of those
12090          * flags.
12091          *
12092          * If the user did not specify any flags, it doesn't matter whether
12093          * or not the command supports the flags.
12094          */
12095         if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
12096              (pattern & ~CTL_LUN_PAT_MASK))
12097                 return (CTL_LUN_PAT_NONE);
12098
12099         /*
12100          * If the user asked for a range check, see if the requested LBA
12101          * range overlaps with this command's LBA range.
12102          */
12103         if (filtered_pattern & CTL_LUN_PAT_RANGE) {
12104                 uint64_t lba1;
12105                 uint64_t len1;
12106                 ctl_action action;
12107                 int retval;
12108
12109                 retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
12110                 if (retval != 0)
12111                         return (CTL_LUN_PAT_NONE);
12112
12113                 action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
12114                                               desc->lba_range.len, FALSE);
12115                 /*
12116                  * A "pass" means that the LBA ranges don't overlap, so
12117                  * this doesn't match the user's range criteria.
12118                  */
12119                 if (action == CTL_ACTION_PASS)
12120                         return (CTL_LUN_PAT_NONE);
12121         }
12122
12123         return (filtered_pattern);
12124 }
12125
12126 static void
12127 ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
12128 {
12129         struct ctl_error_desc *desc, *desc2;
12130
12131         mtx_assert(&lun->lun_lock, MA_OWNED);
12132
12133         STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
12134                 ctl_lun_error_pattern pattern;
12135                 /*
12136                  * Check to see whether this particular command matches
12137                  * the pattern in the descriptor.
12138                  */
12139                 pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
12140                 if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
12141                         continue;
12142
12143                 switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
12144                 case CTL_LUN_INJ_ABORTED:
12145                         ctl_set_aborted(&io->scsiio);
12146                         break;
12147                 case CTL_LUN_INJ_MEDIUM_ERR:
12148                         ctl_set_medium_error(&io->scsiio,
12149                             (io->io_hdr.flags & CTL_FLAG_DATA_MASK) !=
12150                              CTL_FLAG_DATA_OUT);
12151                         break;
12152                 case CTL_LUN_INJ_UA:
12153                         /* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
12154                          * OCCURRED */
12155                         ctl_set_ua(&io->scsiio, 0x29, 0x00);
12156                         break;
12157                 case CTL_LUN_INJ_CUSTOM:
12158                         /*
12159                          * We're assuming the user knows what he is doing.
12160                          * Just copy the sense information without doing
12161                          * checks.
12162                          */
12163                         bcopy(&desc->custom_sense, &io->scsiio.sense_data,
12164                               MIN(sizeof(desc->custom_sense),
12165                                   sizeof(io->scsiio.sense_data)));
12166                         io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
12167                         io->scsiio.sense_len = SSD_FULL_SIZE;
12168                         io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
12169                         break;
12170                 case CTL_LUN_INJ_NONE:
12171                 default:
12172                         /*
12173                          * If this is an error injection type we don't know
12174                          * about, clear the continuous flag (if it is set)
12175                          * so it will get deleted below.
12176                          */
12177                         desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
12178                         break;
12179                 }
12180                 /*
12181                  * By default, each error injection action is a one-shot
12182                  */
12183                 if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
12184                         continue;
12185
12186                 STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
12187
12188                 free(desc, M_CTL);
12189         }
12190 }
12191
12192 #ifdef CTL_IO_DELAY
12193 static void
12194 ctl_datamove_timer_wakeup(void *arg)
12195 {
12196         union ctl_io *io;
12197
12198         io = (union ctl_io *)arg;
12199
12200         ctl_datamove(io);
12201 }
12202 #endif /* CTL_IO_DELAY */
12203
12204 void
12205 ctl_datamove(union ctl_io *io)
12206 {
12207         struct ctl_lun *lun;
12208         void (*fe_datamove)(union ctl_io *io);
12209
12210         mtx_assert(&control_softc->ctl_lock, MA_NOTOWNED);
12211
12212         CTL_DEBUG_PRINT(("ctl_datamove\n"));
12213
12214         lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12215 #ifdef CTL_TIME_IO
12216         if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12217                 char str[256];
12218                 char path_str[64];
12219                 struct sbuf sb;
12220
12221                 ctl_scsi_path_string(io, path_str, sizeof(path_str));
12222                 sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12223
12224                 sbuf_cat(&sb, path_str);
12225                 switch (io->io_hdr.io_type) {
12226                 case CTL_IO_SCSI:
12227                         ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12228                         sbuf_printf(&sb, "\n");
12229                         sbuf_cat(&sb, path_str);
12230                         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12231                                     io->scsiio.tag_num, io->scsiio.tag_type);
12232                         break;
12233                 case CTL_IO_TASK:
12234                         sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12235                                     "Tag Type: %d\n", io->taskio.task_action,
12236                                     io->taskio.tag_num, io->taskio.tag_type);
12237                         break;
12238                 default:
12239                         printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12240                         panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12241                         break;
12242                 }
12243                 sbuf_cat(&sb, path_str);
12244                 sbuf_printf(&sb, "ctl_datamove: %jd seconds\n",
12245                             (intmax_t)time_uptime - io->io_hdr.start_time);
12246                 sbuf_finish(&sb);
12247                 printf("%s", sbuf_data(&sb));
12248         }
12249 #endif /* CTL_TIME_IO */
12250
12251 #ifdef CTL_IO_DELAY
12252         if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
12253                 io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
12254         } else {
12255                 if ((lun != NULL)
12256                  && (lun->delay_info.datamove_delay > 0)) {
12257
12258                         callout_init(&io->io_hdr.delay_callout, /*mpsafe*/ 1);
12259                         io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
12260                         callout_reset(&io->io_hdr.delay_callout,
12261                                       lun->delay_info.datamove_delay * hz,
12262                                       ctl_datamove_timer_wakeup, io);
12263                         if (lun->delay_info.datamove_type ==
12264                             CTL_DELAY_TYPE_ONESHOT)
12265                                 lun->delay_info.datamove_delay = 0;
12266                         return;
12267                 }
12268         }
12269 #endif
12270
12271         /*
12272          * This command has been aborted.  Set the port status, so we fail
12273          * the data move.
12274          */
12275         if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12276                 printf("ctl_datamove: tag 0x%04x on (%u:%u:%u) aborted\n",
12277                        io->scsiio.tag_num, io->io_hdr.nexus.initid,
12278                        io->io_hdr.nexus.targ_port,
12279                        io->io_hdr.nexus.targ_lun);
12280                 io->io_hdr.port_status = 31337;
12281                 /*
12282                  * Note that the backend, in this case, will get the
12283                  * callback in its context.  In other cases it may get
12284                  * called in the frontend's interrupt thread context.
12285                  */
12286                 io->scsiio.be_move_done(io);
12287                 return;
12288         }
12289
12290         /* Don't confuse frontend with zero length data move. */
12291         if (io->scsiio.kern_data_len == 0) {
12292                 io->scsiio.be_move_done(io);
12293                 return;
12294         }
12295
12296         /*
12297          * If we're in XFER mode and this I/O is from the other shelf
12298          * controller, we need to send the DMA to the other side to
12299          * actually transfer the data to/from the host.  In serialize only
12300          * mode the transfer happens below CTL and ctl_datamove() is only
12301          * called on the machine that originally received the I/O.
12302          */
12303         if ((control_softc->ha_mode == CTL_HA_MODE_XFER)
12304          && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
12305                 union ctl_ha_msg msg;
12306                 uint32_t sg_entries_sent;
12307                 int do_sg_copy;
12308                 int i;
12309
12310                 memset(&msg, 0, sizeof(msg));
12311                 msg.hdr.msg_type = CTL_MSG_DATAMOVE;
12312                 msg.hdr.original_sc = io->io_hdr.original_sc;
12313                 msg.hdr.serializing_sc = io;
12314                 msg.hdr.nexus = io->io_hdr.nexus;
12315                 msg.hdr.status = io->io_hdr.status;
12316                 msg.dt.flags = io->io_hdr.flags;
12317                 /*
12318                  * We convert everything into a S/G list here.  We can't
12319                  * pass by reference, only by value between controllers.
12320                  * So we can't pass a pointer to the S/G list, only as many
12321                  * S/G entries as we can fit in here.  If it's possible for
12322                  * us to get more than CTL_HA_MAX_SG_ENTRIES S/G entries,
12323                  * then we need to break this up into multiple transfers.
12324                  */
12325                 if (io->scsiio.kern_sg_entries == 0) {
12326                         msg.dt.kern_sg_entries = 1;
12327 #if 0
12328                         /*
12329                          * Convert to a physical address if this is a
12330                          * virtual address.
12331                          */
12332                         if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
12333                                 msg.dt.sg_list[0].addr =
12334                                         io->scsiio.kern_data_ptr;
12335                         } else {
12336                                 /*
12337                                  * XXX KDM use busdma here!
12338                                  */
12339                                 msg.dt.sg_list[0].addr = (void *)
12340                                         vtophys(io->scsiio.kern_data_ptr);
12341                         }
12342 #else
12343                         KASSERT((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0,
12344                             ("HA does not support BUS_ADDR"));
12345                         msg.dt.sg_list[0].addr = io->scsiio.kern_data_ptr;
12346 #endif
12347
12348                         msg.dt.sg_list[0].len = io->scsiio.kern_data_len;
12349                         do_sg_copy = 0;
12350                 } else {
12351                         msg.dt.kern_sg_entries = io->scsiio.kern_sg_entries;
12352                         do_sg_copy = 1;
12353                 }
12354
12355                 msg.dt.kern_data_len = io->scsiio.kern_data_len;
12356                 msg.dt.kern_total_len = io->scsiio.kern_total_len;
12357                 msg.dt.kern_data_resid = io->scsiio.kern_data_resid;
12358                 msg.dt.kern_rel_offset = io->scsiio.kern_rel_offset;
12359                 msg.dt.sg_sequence = 0;
12360
12361                 /*
12362                  * Loop until we've sent all of the S/G entries.  On the
12363                  * other end, we'll recompose these S/G entries into one
12364                  * contiguous list before passing it to the
12365                  */
12366                 for (sg_entries_sent = 0; sg_entries_sent <
12367                      msg.dt.kern_sg_entries; msg.dt.sg_sequence++) {
12368                         msg.dt.cur_sg_entries = MIN((sizeof(msg.dt.sg_list)/
12369                                 sizeof(msg.dt.sg_list[0])),
12370                                 msg.dt.kern_sg_entries - sg_entries_sent);
12371
12372                         if (do_sg_copy != 0) {
12373                                 struct ctl_sg_entry *sgl;
12374                                 int j;
12375
12376                                 sgl = (struct ctl_sg_entry *)
12377                                         io->scsiio.kern_data_ptr;
12378                                 /*
12379                                  * If this is in cached memory, flush the cache
12380                                  * before we send the DMA request to the other
12381                                  * controller.  We want to do this in either
12382                                  * the * read or the write case.  The read
12383                                  * case is straightforward.  In the write
12384                                  * case, we want to make sure nothing is
12385                                  * in the local cache that could overwrite
12386                                  * the DMAed data.
12387                                  */
12388
12389                                 for (i = sg_entries_sent, j = 0;
12390                                      i < msg.dt.cur_sg_entries; i++, j++) {
12391 #if 0
12392                                         if ((io->io_hdr.flags &
12393                                              CTL_FLAG_BUS_ADDR) == 0) {
12394                                                 /*
12395                                                  * XXX KDM use busdma.
12396                                                  */
12397                                                 msg.dt.sg_list[j].addr =(void *)
12398                                                        vtophys(sgl[i].addr);
12399                                         } else {
12400                                                 msg.dt.sg_list[j].addr =
12401                                                         sgl[i].addr;
12402                                         }
12403 #else
12404                                         KASSERT((io->io_hdr.flags &
12405                                             CTL_FLAG_BUS_ADDR) == 0,
12406                                             ("HA does not support BUS_ADDR"));
12407                                         msg.dt.sg_list[j].addr = sgl[i].addr;
12408 #endif
12409                                         msg.dt.sg_list[j].len = sgl[i].len;
12410                                 }
12411                         }
12412
12413                         sg_entries_sent += msg.dt.cur_sg_entries;
12414                         if (sg_entries_sent >= msg.dt.kern_sg_entries)
12415                                 msg.dt.sg_last = 1;
12416                         else
12417                                 msg.dt.sg_last = 0;
12418
12419                         if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12420                             sizeof(msg.dt) - sizeof(msg.dt.sg_list) +
12421                             sizeof(struct ctl_sg_entry)*msg.dt.cur_sg_entries,
12422                             M_WAITOK) > CTL_HA_STATUS_SUCCESS) {
12423                                 io->io_hdr.port_status = 31341;
12424                                 io->scsiio.be_move_done(io);
12425                                 return;
12426                         }
12427
12428                         msg.dt.sent_sg_entries = sg_entries_sent;
12429                 }
12430
12431                 /*
12432                  * Officially handover the request from us to peer.
12433                  * If failover has just happened, then we must return error.
12434                  * If failover happen just after, then it is not our problem.
12435                  */
12436                 if (lun)
12437                         mtx_lock(&lun->lun_lock);
12438                 if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12439                         if (lun)
12440                                 mtx_unlock(&lun->lun_lock);
12441                         io->io_hdr.port_status = 31342;
12442                         io->scsiio.be_move_done(io);
12443                         return;
12444                 }
12445                 io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12446                 io->io_hdr.flags |= CTL_FLAG_DMA_INPROG;
12447                 if (lun)
12448                         mtx_unlock(&lun->lun_lock);
12449         } else {
12450
12451                 /*
12452                  * Lookup the fe_datamove() function for this particular
12453                  * front end.
12454                  */
12455                 fe_datamove = ctl_io_port(&io->io_hdr)->fe_datamove;
12456
12457                 fe_datamove(io);
12458         }
12459 }
12460
12461 static void
12462 ctl_send_datamove_done(union ctl_io *io, int have_lock)
12463 {
12464         union ctl_ha_msg msg;
12465
12466         memset(&msg, 0, sizeof(msg));
12467
12468         msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
12469         msg.hdr.original_sc = io;
12470         msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
12471         msg.hdr.nexus = io->io_hdr.nexus;
12472         msg.hdr.status = io->io_hdr.status;
12473         msg.scsi.tag_num = io->scsiio.tag_num;
12474         msg.scsi.tag_type = io->scsiio.tag_type;
12475         msg.scsi.scsi_status = io->scsiio.scsi_status;
12476         memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
12477                io->scsiio.sense_len);
12478         msg.scsi.sense_len = io->scsiio.sense_len;
12479         msg.scsi.sense_residual = io->scsiio.sense_residual;
12480         msg.scsi.fetd_status = io->io_hdr.port_status;
12481         msg.scsi.residual = io->scsiio.residual;
12482         io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12483
12484         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12485                 ctl_failover_io(io, /*have_lock*/ have_lock);
12486                 return;
12487         }
12488
12489         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12490             sizeof(msg.scsi) - sizeof(msg.scsi.sense_data) +
12491             msg.scsi.sense_len, M_WAITOK);
12492 }
12493
12494 /*
12495  * The DMA to the remote side is done, now we need to tell the other side
12496  * we're done so it can continue with its data movement.
12497  */
12498 static void
12499 ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
12500 {
12501         union ctl_io *io;
12502         int i;
12503
12504         io = rq->context;
12505
12506         if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12507                 printf("%s: ISC DMA write failed with error %d", __func__,
12508                        rq->ret);
12509                 ctl_set_internal_failure(&io->scsiio,
12510                                          /*sks_valid*/ 1,
12511                                          /*retry_count*/ rq->ret);
12512         }
12513
12514         ctl_dt_req_free(rq);
12515
12516         for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12517                 free(io->io_hdr.local_sglist[i].addr, M_CTL);
12518         free(io->io_hdr.remote_sglist, M_CTL);
12519         io->io_hdr.remote_sglist = NULL;
12520         io->io_hdr.local_sglist = NULL;
12521
12522         /*
12523          * The data is in local and remote memory, so now we need to send
12524          * status (good or back) back to the other side.
12525          */
12526         ctl_send_datamove_done(io, /*have_lock*/ 0);
12527 }
12528
12529 /*
12530  * We've moved the data from the host/controller into local memory.  Now we
12531  * need to push it over to the remote controller's memory.
12532  */
12533 static int
12534 ctl_datamove_remote_dm_write_cb(union ctl_io *io)
12535 {
12536         int retval;
12537
12538         retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
12539                                           ctl_datamove_remote_write_cb);
12540         return (retval);
12541 }
12542
12543 static void
12544 ctl_datamove_remote_write(union ctl_io *io)
12545 {
12546         int retval;
12547         void (*fe_datamove)(union ctl_io *io);
12548
12549         /*
12550          * - Get the data from the host/HBA into local memory.
12551          * - DMA memory from the local controller to the remote controller.
12552          * - Send status back to the remote controller.
12553          */
12554
12555         retval = ctl_datamove_remote_sgl_setup(io);
12556         if (retval != 0)
12557                 return;
12558
12559         /* Switch the pointer over so the FETD knows what to do */
12560         io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12561
12562         /*
12563          * Use a custom move done callback, since we need to send completion
12564          * back to the other controller, not to the backend on this side.
12565          */
12566         io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
12567
12568         fe_datamove = ctl_io_port(&io->io_hdr)->fe_datamove;
12569         fe_datamove(io);
12570 }
12571
12572 static int
12573 ctl_datamove_remote_dm_read_cb(union ctl_io *io)
12574 {
12575 #if 0
12576         char str[256];
12577         char path_str[64];
12578         struct sbuf sb;
12579 #endif
12580         int i;
12581
12582         for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12583                 free(io->io_hdr.local_sglist[i].addr, M_CTL);
12584         free(io->io_hdr.remote_sglist, M_CTL);
12585         io->io_hdr.remote_sglist = NULL;
12586         io->io_hdr.local_sglist = NULL;
12587
12588 #if 0
12589         scsi_path_string(io, path_str, sizeof(path_str));
12590         sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12591         sbuf_cat(&sb, path_str);
12592         scsi_command_string(&io->scsiio, NULL, &sb);
12593         sbuf_printf(&sb, "\n");
12594         sbuf_cat(&sb, path_str);
12595         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12596                     io->scsiio.tag_num, io->scsiio.tag_type);
12597         sbuf_cat(&sb, path_str);
12598         sbuf_printf(&sb, "%s: flags %#x, status %#x\n", __func__,
12599                     io->io_hdr.flags, io->io_hdr.status);
12600         sbuf_finish(&sb);
12601         printk("%s", sbuf_data(&sb));
12602 #endif
12603
12604
12605         /*
12606          * The read is done, now we need to send status (good or bad) back
12607          * to the other side.
12608          */
12609         ctl_send_datamove_done(io, /*have_lock*/ 0);
12610
12611         return (0);
12612 }
12613
12614 static void
12615 ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
12616 {
12617         union ctl_io *io;
12618         void (*fe_datamove)(union ctl_io *io);
12619
12620         io = rq->context;
12621
12622         if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12623                 printf("%s: ISC DMA read failed with error %d\n", __func__,
12624                        rq->ret);
12625                 ctl_set_internal_failure(&io->scsiio,
12626                                          /*sks_valid*/ 1,
12627                                          /*retry_count*/ rq->ret);
12628         }
12629
12630         ctl_dt_req_free(rq);
12631
12632         /* Switch the pointer over so the FETD knows what to do */
12633         io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12634
12635         /*
12636          * Use a custom move done callback, since we need to send completion
12637          * back to the other controller, not to the backend on this side.
12638          */
12639         io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
12640
12641         /* XXX KDM add checks like the ones in ctl_datamove? */
12642
12643         fe_datamove = ctl_io_port(&io->io_hdr)->fe_datamove;
12644         fe_datamove(io);
12645 }
12646
12647 static int
12648 ctl_datamove_remote_sgl_setup(union ctl_io *io)
12649 {
12650         struct ctl_sg_entry *local_sglist;
12651         struct ctl_softc *softc;
12652         uint32_t len_to_go;
12653         int retval;
12654         int i;
12655
12656         retval = 0;
12657         softc = control_softc;
12658         local_sglist = io->io_hdr.local_sglist;
12659         len_to_go = io->scsiio.kern_data_len;
12660
12661         /*
12662          * The difficult thing here is that the size of the various
12663          * S/G segments may be different than the size from the
12664          * remote controller.  That'll make it harder when DMAing
12665          * the data back to the other side.
12666          */
12667         for (i = 0; len_to_go > 0; i++) {
12668                 local_sglist[i].len = MIN(len_to_go, CTL_HA_DATAMOVE_SEGMENT);
12669                 local_sglist[i].addr =
12670                     malloc(local_sglist[i].len, M_CTL, M_WAITOK);
12671
12672                 len_to_go -= local_sglist[i].len;
12673         }
12674         /*
12675          * Reset the number of S/G entries accordingly.  The original
12676          * number of S/G entries is available in rem_sg_entries.
12677          */
12678         io->scsiio.kern_sg_entries = i;
12679
12680 #if 0
12681         printf("%s: kern_sg_entries = %d\n", __func__,
12682                io->scsiio.kern_sg_entries);
12683         for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12684                 printf("%s: sg[%d] = %p, %lu\n", __func__, i,
12685                        local_sglist[i].addr, local_sglist[i].len);
12686 #endif
12687
12688         return (retval);
12689 }
12690
12691 static int
12692 ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
12693                          ctl_ha_dt_cb callback)
12694 {
12695         struct ctl_ha_dt_req *rq;
12696         struct ctl_sg_entry *remote_sglist, *local_sglist;
12697         uint32_t local_used, remote_used, total_used;
12698         int i, j, isc_ret;
12699
12700         rq = ctl_dt_req_alloc();
12701
12702         /*
12703          * If we failed to allocate the request, and if the DMA didn't fail
12704          * anyway, set busy status.  This is just a resource allocation
12705          * failure.
12706          */
12707         if ((rq == NULL)
12708          && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
12709              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS))
12710                 ctl_set_busy(&io->scsiio);
12711
12712         if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
12713             (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS) {
12714
12715                 if (rq != NULL)
12716                         ctl_dt_req_free(rq);
12717
12718                 /*
12719                  * The data move failed.  We need to return status back
12720                  * to the other controller.  No point in trying to DMA
12721                  * data to the remote controller.
12722                  */
12723
12724                 ctl_send_datamove_done(io, /*have_lock*/ 0);
12725
12726                 return (1);
12727         }
12728
12729         local_sglist = io->io_hdr.local_sglist;
12730         remote_sglist = io->io_hdr.remote_sglist;
12731         local_used = 0;
12732         remote_used = 0;
12733         total_used = 0;
12734
12735         /*
12736          * Pull/push the data over the wire from/to the other controller.
12737          * This takes into account the possibility that the local and
12738          * remote sglists may not be identical in terms of the size of
12739          * the elements and the number of elements.
12740          *
12741          * One fundamental assumption here is that the length allocated for
12742          * both the local and remote sglists is identical.  Otherwise, we've
12743          * essentially got a coding error of some sort.
12744          */
12745         isc_ret = CTL_HA_STATUS_SUCCESS;
12746         for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
12747                 uint32_t cur_len;
12748                 uint8_t *tmp_ptr;
12749
12750                 rq->command = command;
12751                 rq->context = io;
12752
12753                 /*
12754                  * Both pointers should be aligned.  But it is possible
12755                  * that the allocation length is not.  They should both
12756                  * also have enough slack left over at the end, though,
12757                  * to round up to the next 8 byte boundary.
12758                  */
12759                 cur_len = MIN(local_sglist[i].len - local_used,
12760                               remote_sglist[j].len - remote_used);
12761                 rq->size = cur_len;
12762
12763                 tmp_ptr = (uint8_t *)local_sglist[i].addr;
12764                 tmp_ptr += local_used;
12765
12766 #if 0
12767                 /* Use physical addresses when talking to ISC hardware */
12768                 if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
12769                         /* XXX KDM use busdma */
12770                         rq->local = vtophys(tmp_ptr);
12771                 } else
12772                         rq->local = tmp_ptr;
12773 #else
12774                 KASSERT((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0,
12775                     ("HA does not support BUS_ADDR"));
12776                 rq->local = tmp_ptr;
12777 #endif
12778
12779                 tmp_ptr = (uint8_t *)remote_sglist[j].addr;
12780                 tmp_ptr += remote_used;
12781                 rq->remote = tmp_ptr;
12782
12783                 rq->callback = NULL;
12784
12785                 local_used += cur_len;
12786                 if (local_used >= local_sglist[i].len) {
12787                         i++;
12788                         local_used = 0;
12789                 }
12790
12791                 remote_used += cur_len;
12792                 if (remote_used >= remote_sglist[j].len) {
12793                         j++;
12794                         remote_used = 0;
12795                 }
12796                 total_used += cur_len;
12797
12798                 if (total_used >= io->scsiio.kern_data_len)
12799                         rq->callback = callback;
12800
12801 #if 0
12802                 printf("%s: %s: local %p remote %p size %d\n", __func__,
12803                        (command == CTL_HA_DT_CMD_WRITE) ? "WRITE" : "READ",
12804                        rq->local, rq->remote, rq->size);
12805 #endif
12806
12807                 isc_ret = ctl_dt_single(rq);
12808                 if (isc_ret > CTL_HA_STATUS_SUCCESS)
12809                         break;
12810         }
12811         if (isc_ret != CTL_HA_STATUS_WAIT) {
12812                 rq->ret = isc_ret;
12813                 callback(rq);
12814         }
12815
12816         return (0);
12817 }
12818
12819 static void
12820 ctl_datamove_remote_read(union ctl_io *io)
12821 {
12822         int retval;
12823         int i;
12824
12825         /*
12826          * This will send an error to the other controller in the case of a
12827          * failure.
12828          */
12829         retval = ctl_datamove_remote_sgl_setup(io);
12830         if (retval != 0)
12831                 return;
12832
12833         retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
12834                                           ctl_datamove_remote_read_cb);
12835         if (retval != 0) {
12836                 /*
12837                  * Make sure we free memory if there was an error..  The
12838                  * ctl_datamove_remote_xfer() function will send the
12839                  * datamove done message, or call the callback with an
12840                  * error if there is a problem.
12841                  */
12842                 for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12843                         free(io->io_hdr.local_sglist[i].addr, M_CTL);
12844                 free(io->io_hdr.remote_sglist, M_CTL);
12845                 io->io_hdr.remote_sglist = NULL;
12846                 io->io_hdr.local_sglist = NULL;
12847         }
12848 }
12849
12850 /*
12851  * Process a datamove request from the other controller.  This is used for
12852  * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
12853  * first.  Once that is complete, the data gets DMAed into the remote
12854  * controller's memory.  For reads, we DMA from the remote controller's
12855  * memory into our memory first, and then move it out to the FETD.
12856  */
12857 static void
12858 ctl_datamove_remote(union ctl_io *io)
12859 {
12860
12861         mtx_assert(&control_softc->ctl_lock, MA_NOTOWNED);
12862
12863         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12864                 ctl_failover_io(io, /*have_lock*/ 0);
12865                 return;
12866         }
12867
12868         /*
12869          * Note that we look for an aborted I/O here, but don't do some of
12870          * the other checks that ctl_datamove() normally does.
12871          * We don't need to run the datamove delay code, since that should
12872          * have been done if need be on the other controller.
12873          */
12874         if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12875                 printf("%s: tag 0x%04x on (%u:%u:%u) aborted\n", __func__,
12876                        io->scsiio.tag_num, io->io_hdr.nexus.initid,
12877                        io->io_hdr.nexus.targ_port,
12878                        io->io_hdr.nexus.targ_lun);
12879                 io->io_hdr.port_status = 31338;
12880                 ctl_send_datamove_done(io, /*have_lock*/ 0);
12881                 return;
12882         }
12883
12884         if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT)
12885                 ctl_datamove_remote_write(io);
12886         else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN)
12887                 ctl_datamove_remote_read(io);
12888         else {
12889                 io->io_hdr.port_status = 31339;
12890                 ctl_send_datamove_done(io, /*have_lock*/ 0);
12891         }
12892 }
12893
12894 static int
12895 ctl_process_done(union ctl_io *io)
12896 {
12897         struct ctl_lun *lun;
12898         struct ctl_softc *softc = control_softc;
12899         void (*fe_done)(union ctl_io *io);
12900         union ctl_ha_msg msg;
12901         uint32_t targ_port = io->io_hdr.nexus.targ_port;
12902
12903         CTL_DEBUG_PRINT(("ctl_process_done\n"));
12904
12905         if ((io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) == 0)
12906                 fe_done = softc->ctl_ports[targ_port]->fe_done;
12907         else
12908                 fe_done = NULL;
12909
12910 #ifdef CTL_TIME_IO
12911         if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12912                 char str[256];
12913                 char path_str[64];
12914                 struct sbuf sb;
12915
12916                 ctl_scsi_path_string(io, path_str, sizeof(path_str));
12917                 sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12918
12919                 sbuf_cat(&sb, path_str);
12920                 switch (io->io_hdr.io_type) {
12921                 case CTL_IO_SCSI:
12922                         ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12923                         sbuf_printf(&sb, "\n");
12924                         sbuf_cat(&sb, path_str);
12925                         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12926                                     io->scsiio.tag_num, io->scsiio.tag_type);
12927                         break;
12928                 case CTL_IO_TASK:
12929                         sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12930                                     "Tag Type: %d\n", io->taskio.task_action,
12931                                     io->taskio.tag_num, io->taskio.tag_type);
12932                         break;
12933                 default:
12934                         printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12935                         panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12936                         break;
12937                 }
12938                 sbuf_cat(&sb, path_str);
12939                 sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
12940                             (intmax_t)time_uptime - io->io_hdr.start_time);
12941                 sbuf_finish(&sb);
12942                 printf("%s", sbuf_data(&sb));
12943         }
12944 #endif /* CTL_TIME_IO */
12945
12946         switch (io->io_hdr.io_type) {
12947         case CTL_IO_SCSI:
12948                 break;
12949         case CTL_IO_TASK:
12950                 if (ctl_debug & CTL_DEBUG_INFO)
12951                         ctl_io_error_print(io, NULL);
12952                 if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
12953                         ctl_free_io(io);
12954                 else
12955                         fe_done(io);
12956                 return (CTL_RETVAL_COMPLETE);
12957         default:
12958                 panic("ctl_process_done: invalid io type %d\n",
12959                       io->io_hdr.io_type);
12960                 break; /* NOTREACHED */
12961         }
12962
12963         lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12964         if (lun == NULL) {
12965                 CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
12966                                  io->io_hdr.nexus.targ_mapped_lun));
12967                 goto bailout;
12968         }
12969
12970         mtx_lock(&lun->lun_lock);
12971
12972         /*
12973          * Check to see if we have any errors to inject here.  We only
12974          * inject errors for commands that don't already have errors set.
12975          */
12976         if ((STAILQ_FIRST(&lun->error_list) != NULL) &&
12977             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS) &&
12978             ((io->io_hdr.flags & CTL_FLAG_STATUS_SENT) == 0))
12979                 ctl_inject_error(lun, io);
12980
12981         /*
12982          * XXX KDM how do we treat commands that aren't completed
12983          * successfully?
12984          *
12985          * XXX KDM should we also track I/O latency?
12986          */
12987         if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS &&
12988             io->io_hdr.io_type == CTL_IO_SCSI) {
12989 #ifdef CTL_TIME_IO
12990                 struct bintime cur_bt;
12991 #endif
12992                 int type;
12993
12994                 if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
12995                     CTL_FLAG_DATA_IN)
12996                         type = CTL_STATS_READ;
12997                 else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
12998                     CTL_FLAG_DATA_OUT)
12999                         type = CTL_STATS_WRITE;
13000                 else
13001                         type = CTL_STATS_NO_IO;
13002
13003                 lun->stats.ports[targ_port].bytes[type] +=
13004                     io->scsiio.kern_total_len;
13005                 lun->stats.ports[targ_port].operations[type]++;
13006 #ifdef CTL_TIME_IO
13007                 bintime_add(&lun->stats.ports[targ_port].dma_time[type],
13008                    &io->io_hdr.dma_bt);
13009                 lun->stats.ports[targ_port].num_dmas[type] +=
13010                     io->io_hdr.num_dmas;
13011                 getbintime(&cur_bt);
13012                 bintime_sub(&cur_bt, &io->io_hdr.start_bt);
13013                 bintime_add(&lun->stats.ports[targ_port].time[type], &cur_bt);
13014 #endif
13015         }
13016
13017         /*
13018          * Remove this from the OOA queue.
13019          */
13020         TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
13021 #ifdef CTL_TIME_IO
13022         if (TAILQ_EMPTY(&lun->ooa_queue))
13023                 lun->last_busy = getsbinuptime();
13024 #endif
13025
13026         /*
13027          * Run through the blocked queue on this LUN and see if anything
13028          * has become unblocked, now that this transaction is done.
13029          */
13030         ctl_check_blocked(lun);
13031
13032         /*
13033          * If the LUN has been invalidated, free it if there is nothing
13034          * left on its OOA queue.
13035          */
13036         if ((lun->flags & CTL_LUN_INVALID)
13037          && TAILQ_EMPTY(&lun->ooa_queue)) {
13038                 mtx_unlock(&lun->lun_lock);
13039                 mtx_lock(&softc->ctl_lock);
13040                 ctl_free_lun(lun);
13041                 mtx_unlock(&softc->ctl_lock);
13042         } else
13043                 mtx_unlock(&lun->lun_lock);
13044
13045 bailout:
13046
13047         /*
13048          * If this command has been aborted, make sure we set the status
13049          * properly.  The FETD is responsible for freeing the I/O and doing
13050          * whatever it needs to do to clean up its state.
13051          */
13052         if (io->io_hdr.flags & CTL_FLAG_ABORT)
13053                 ctl_set_task_aborted(&io->scsiio);
13054
13055         /*
13056          * If enabled, print command error status.
13057          */
13058         if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS &&
13059             (ctl_debug & CTL_DEBUG_INFO) != 0)
13060                 ctl_io_error_print(io, NULL);
13061
13062         /*
13063          * Tell the FETD or the other shelf controller we're done with this
13064          * command.  Note that only SCSI commands get to this point.  Task
13065          * management commands are completed above.
13066          */
13067         if ((softc->ha_mode != CTL_HA_MODE_XFER) &&
13068             (io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)) {
13069                 memset(&msg, 0, sizeof(msg));
13070                 msg.hdr.msg_type = CTL_MSG_FINISH_IO;
13071                 msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
13072                 msg.hdr.nexus = io->io_hdr.nexus;
13073                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13074                     sizeof(msg.scsi) - sizeof(msg.scsi.sense_data),
13075                     M_WAITOK);
13076         }
13077         if ((softc->ha_mode == CTL_HA_MODE_XFER)
13078          && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
13079                 memset(&msg, 0, sizeof(msg));
13080                 msg.hdr.msg_type = CTL_MSG_FINISH_IO;
13081                 msg.hdr.original_sc = io->io_hdr.original_sc;
13082                 msg.hdr.nexus = io->io_hdr.nexus;
13083                 msg.hdr.status = io->io_hdr.status;
13084                 msg.scsi.scsi_status = io->scsiio.scsi_status;
13085                 msg.scsi.tag_num = io->scsiio.tag_num;
13086                 msg.scsi.tag_type = io->scsiio.tag_type;
13087                 msg.scsi.sense_len = io->scsiio.sense_len;
13088                 msg.scsi.sense_residual = io->scsiio.sense_residual;
13089                 msg.scsi.residual = io->scsiio.residual;
13090                 memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
13091                        io->scsiio.sense_len);
13092
13093                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13094                     sizeof(msg.scsi) - sizeof(msg.scsi.sense_data) +
13095                     msg.scsi.sense_len, M_WAITOK);
13096                 ctl_free_io(io);
13097         } else
13098                 fe_done(io);
13099
13100         return (CTL_RETVAL_COMPLETE);
13101 }
13102
13103 #ifdef CTL_WITH_CA
13104 /*
13105  * Front end should call this if it doesn't do autosense.  When the request
13106  * sense comes back in from the initiator, we'll dequeue this and send it.
13107  */
13108 int
13109 ctl_queue_sense(union ctl_io *io)
13110 {
13111         struct ctl_lun *lun;
13112         struct ctl_port *port;
13113         struct ctl_softc *softc;
13114         uint32_t initidx, targ_lun;
13115
13116         softc = control_softc;
13117
13118         CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
13119
13120         /*
13121          * LUN lookup will likely move to the ctl_work_thread() once we
13122          * have our new queueing infrastructure (that doesn't put things on
13123          * a per-LUN queue initially).  That is so that we can handle
13124          * things like an INQUIRY to a LUN that we don't have enabled.  We
13125          * can't deal with that right now.
13126          */
13127         mtx_lock(&softc->ctl_lock);
13128
13129         /*
13130          * If we don't have a LUN for this, just toss the sense
13131          * information.
13132          */
13133         port = ctl_io_port(&ctsio->io_hdr);
13134         targ_lun = ctl_lun_map_from_port(port, io->io_hdr.nexus.targ_lun);
13135         if ((targ_lun < CTL_MAX_LUNS)
13136          && (softc->ctl_luns[targ_lun] != NULL))
13137                 lun = softc->ctl_luns[targ_lun];
13138         else
13139                 goto bailout;
13140
13141         initidx = ctl_get_initindex(&io->io_hdr.nexus);
13142
13143         mtx_lock(&lun->lun_lock);
13144         /*
13145          * Already have CA set for this LUN...toss the sense information.
13146          */
13147         if (ctl_is_set(lun->have_ca, initidx)) {
13148                 mtx_unlock(&lun->lun_lock);
13149                 goto bailout;
13150         }
13151
13152         memcpy(&lun->pending_sense[initidx], &io->scsiio.sense_data,
13153                MIN(sizeof(lun->pending_sense[initidx]),
13154                sizeof(io->scsiio.sense_data)));
13155         ctl_set_mask(lun->have_ca, initidx);
13156         mtx_unlock(&lun->lun_lock);
13157
13158 bailout:
13159         mtx_unlock(&softc->ctl_lock);
13160
13161         ctl_free_io(io);
13162
13163         return (CTL_RETVAL_COMPLETE);
13164 }
13165 #endif
13166
13167 /*
13168  * Primary command inlet from frontend ports.  All SCSI and task I/O
13169  * requests must go through this function.
13170  */
13171 int
13172 ctl_queue(union ctl_io *io)
13173 {
13174         struct ctl_port *port;
13175
13176         CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
13177
13178 #ifdef CTL_TIME_IO
13179         io->io_hdr.start_time = time_uptime;
13180         getbintime(&io->io_hdr.start_bt);
13181 #endif /* CTL_TIME_IO */
13182
13183         /* Map FE-specific LUN ID into global one. */
13184         port = ctl_io_port(&io->io_hdr);
13185         io->io_hdr.nexus.targ_mapped_lun =
13186             ctl_lun_map_from_port(port, io->io_hdr.nexus.targ_lun);
13187
13188         switch (io->io_hdr.io_type) {
13189         case CTL_IO_SCSI:
13190         case CTL_IO_TASK:
13191                 if (ctl_debug & CTL_DEBUG_CDB)
13192                         ctl_io_print(io);
13193                 ctl_enqueue_incoming(io);
13194                 break;
13195         default:
13196                 printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
13197                 return (EINVAL);
13198         }
13199
13200         return (CTL_RETVAL_COMPLETE);
13201 }
13202
13203 #ifdef CTL_IO_DELAY
13204 static void
13205 ctl_done_timer_wakeup(void *arg)
13206 {
13207         union ctl_io *io;
13208
13209         io = (union ctl_io *)arg;
13210         ctl_done(io);
13211 }
13212 #endif /* CTL_IO_DELAY */
13213
13214 void
13215 ctl_serseq_done(union ctl_io *io)
13216 {
13217         struct ctl_lun *lun;
13218
13219         lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13220         if (lun->be_lun == NULL ||
13221             lun->be_lun->serseq == CTL_LUN_SERSEQ_OFF)
13222                 return;
13223         mtx_lock(&lun->lun_lock);
13224         io->io_hdr.flags |= CTL_FLAG_SERSEQ_DONE;
13225         ctl_check_blocked(lun);
13226         mtx_unlock(&lun->lun_lock);
13227 }
13228
13229 void
13230 ctl_done(union ctl_io *io)
13231 {
13232
13233         /*
13234          * Enable this to catch duplicate completion issues.
13235          */
13236 #if 0
13237         if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
13238                 printf("%s: type %d msg %d cdb %x iptl: "
13239                        "%u:%u:%u tag 0x%04x "
13240                        "flag %#x status %x\n",
13241                         __func__,
13242                         io->io_hdr.io_type,
13243                         io->io_hdr.msg_type,
13244                         io->scsiio.cdb[0],
13245                         io->io_hdr.nexus.initid,
13246                         io->io_hdr.nexus.targ_port,
13247                         io->io_hdr.nexus.targ_lun,
13248                         (io->io_hdr.io_type ==
13249                         CTL_IO_TASK) ?
13250                         io->taskio.tag_num :
13251                         io->scsiio.tag_num,
13252                         io->io_hdr.flags,
13253                         io->io_hdr.status);
13254         } else
13255                 io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
13256 #endif
13257
13258         /*
13259          * This is an internal copy of an I/O, and should not go through
13260          * the normal done processing logic.
13261          */
13262         if (io->io_hdr.flags & CTL_FLAG_INT_COPY)
13263                 return;
13264
13265 #ifdef CTL_IO_DELAY
13266         if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
13267                 struct ctl_lun *lun;
13268
13269                 lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13270
13271                 io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
13272         } else {
13273                 struct ctl_lun *lun;
13274
13275                 lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13276
13277                 if ((lun != NULL)
13278                  && (lun->delay_info.done_delay > 0)) {
13279
13280                         callout_init(&io->io_hdr.delay_callout, /*mpsafe*/ 1);
13281                         io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
13282                         callout_reset(&io->io_hdr.delay_callout,
13283                                       lun->delay_info.done_delay * hz,
13284                                       ctl_done_timer_wakeup, io);
13285                         if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
13286                                 lun->delay_info.done_delay = 0;
13287                         return;
13288                 }
13289         }
13290 #endif /* CTL_IO_DELAY */
13291
13292         ctl_enqueue_done(io);
13293 }
13294
13295 static void
13296 ctl_work_thread(void *arg)
13297 {
13298         struct ctl_thread *thr = (struct ctl_thread *)arg;
13299         struct ctl_softc *softc = thr->ctl_softc;
13300         union ctl_io *io;
13301         int retval;
13302
13303         CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
13304
13305         for (;;) {
13306                 retval = 0;
13307
13308                 /*
13309                  * We handle the queues in this order:
13310                  * - ISC
13311                  * - done queue (to free up resources, unblock other commands)
13312                  * - RtR queue
13313                  * - incoming queue
13314                  *
13315                  * If those queues are empty, we break out of the loop and
13316                  * go to sleep.
13317                  */
13318                 mtx_lock(&thr->queue_lock);
13319                 io = (union ctl_io *)STAILQ_FIRST(&thr->isc_queue);
13320                 if (io != NULL) {
13321                         STAILQ_REMOVE_HEAD(&thr->isc_queue, links);
13322                         mtx_unlock(&thr->queue_lock);
13323                         ctl_handle_isc(io);
13324                         continue;
13325                 }
13326                 io = (union ctl_io *)STAILQ_FIRST(&thr->done_queue);
13327                 if (io != NULL) {
13328                         STAILQ_REMOVE_HEAD(&thr->done_queue, links);
13329                         /* clear any blocked commands, call fe_done */
13330                         mtx_unlock(&thr->queue_lock);
13331                         retval = ctl_process_done(io);
13332                         continue;
13333                 }
13334                 io = (union ctl_io *)STAILQ_FIRST(&thr->incoming_queue);
13335                 if (io != NULL) {
13336                         STAILQ_REMOVE_HEAD(&thr->incoming_queue, links);
13337                         mtx_unlock(&thr->queue_lock);
13338                         if (io->io_hdr.io_type == CTL_IO_TASK)
13339                                 ctl_run_task(io);
13340                         else
13341                                 ctl_scsiio_precheck(softc, &io->scsiio);
13342                         continue;
13343                 }
13344                 io = (union ctl_io *)STAILQ_FIRST(&thr->rtr_queue);
13345                 if (io != NULL) {
13346                         STAILQ_REMOVE_HEAD(&thr->rtr_queue, links);
13347                         mtx_unlock(&thr->queue_lock);
13348                         retval = ctl_scsiio(&io->scsiio);
13349                         if (retval != CTL_RETVAL_COMPLETE)
13350                                 CTL_DEBUG_PRINT(("ctl_scsiio failed\n"));
13351                         continue;
13352                 }
13353
13354                 /* Sleep until we have something to do. */
13355                 mtx_sleep(thr, &thr->queue_lock, PDROP | PRIBIO, "-", 0);
13356         }
13357 }
13358
13359 static void
13360 ctl_lun_thread(void *arg)
13361 {
13362         struct ctl_softc *softc = (struct ctl_softc *)arg;
13363         struct ctl_be_lun *be_lun;
13364         int retval;
13365
13366         CTL_DEBUG_PRINT(("ctl_lun_thread starting\n"));
13367
13368         for (;;) {
13369                 retval = 0;
13370                 mtx_lock(&softc->ctl_lock);
13371                 be_lun = STAILQ_FIRST(&softc->pending_lun_queue);
13372                 if (be_lun != NULL) {
13373                         STAILQ_REMOVE_HEAD(&softc->pending_lun_queue, links);
13374                         mtx_unlock(&softc->ctl_lock);
13375                         ctl_create_lun(be_lun);
13376                         continue;
13377                 }
13378
13379                 /* Sleep until we have something to do. */
13380                 mtx_sleep(&softc->pending_lun_queue, &softc->ctl_lock,
13381                     PDROP | PRIBIO, "-", 0);
13382         }
13383 }
13384
13385 static void
13386 ctl_thresh_thread(void *arg)
13387 {
13388         struct ctl_softc *softc = (struct ctl_softc *)arg;
13389         struct ctl_lun *lun;
13390         struct ctl_be_lun *be_lun;
13391         struct scsi_da_rw_recovery_page *rwpage;
13392         struct ctl_logical_block_provisioning_page *page;
13393         const char *attr;
13394         union ctl_ha_msg msg;
13395         uint64_t thres, val;
13396         int i, e, set;
13397
13398         CTL_DEBUG_PRINT(("ctl_thresh_thread starting\n"));
13399
13400         for (;;) {
13401                 mtx_lock(&softc->ctl_lock);
13402                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
13403                         be_lun = lun->be_lun;
13404                         if ((lun->flags & CTL_LUN_DISABLED) ||
13405                             (lun->flags & CTL_LUN_OFFLINE) ||
13406                             lun->backend->lun_attr == NULL)
13407                                 continue;
13408                         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
13409                             softc->ha_mode == CTL_HA_MODE_XFER)
13410                                 continue;
13411                         rwpage = &lun->mode_pages.rw_er_page[CTL_PAGE_CURRENT];
13412                         if ((rwpage->byte8 & SMS_RWER_LBPERE) == 0)
13413                                 continue;
13414                         e = 0;
13415                         page = &lun->mode_pages.lbp_page[CTL_PAGE_CURRENT];
13416                         for (i = 0; i < CTL_NUM_LBP_THRESH; i++) {
13417                                 if ((page->descr[i].flags & SLBPPD_ENABLED) == 0)
13418                                         continue;
13419                                 thres = scsi_4btoul(page->descr[i].count);
13420                                 thres <<= CTL_LBP_EXPONENT;
13421                                 switch (page->descr[i].resource) {
13422                                 case 0x01:
13423                                         attr = "blocksavail";
13424                                         break;
13425                                 case 0x02:
13426                                         attr = "blocksused";
13427                                         break;
13428                                 case 0xf1:
13429                                         attr = "poolblocksavail";
13430                                         break;
13431                                 case 0xf2:
13432                                         attr = "poolblocksused";
13433                                         break;
13434                                 default:
13435                                         continue;
13436                                 }
13437                                 mtx_unlock(&softc->ctl_lock); // XXX
13438                                 val = lun->backend->lun_attr(
13439                                     lun->be_lun->be_lun, attr);
13440                                 mtx_lock(&softc->ctl_lock);
13441                                 if (val == UINT64_MAX)
13442                                         continue;
13443                                 if ((page->descr[i].flags & SLBPPD_ARMING_MASK)
13444                                     == SLBPPD_ARMING_INC)
13445                                         e = (val >= thres);
13446                                 else
13447                                         e = (val <= thres);
13448                                 if (e)
13449                                         break;
13450                         }
13451                         mtx_lock(&lun->lun_lock);
13452                         if (e) {
13453                                 scsi_u64to8b((uint8_t *)&page->descr[i] -
13454                                     (uint8_t *)page, lun->ua_tpt_info);
13455                                 if (lun->lasttpt == 0 ||
13456                                     time_uptime - lun->lasttpt >= CTL_LBP_UA_PERIOD) {
13457                                         lun->lasttpt = time_uptime;
13458                                         ctl_est_ua_all(lun, -1, CTL_UA_THIN_PROV_THRES);
13459                                         set = 1;
13460                                 } else
13461                                         set = 0;
13462                         } else {
13463                                 lun->lasttpt = 0;
13464                                 ctl_clr_ua_all(lun, -1, CTL_UA_THIN_PROV_THRES);
13465                                 set = -1;
13466                         }
13467                         mtx_unlock(&lun->lun_lock);
13468                         if (set != 0 &&
13469                             lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
13470                                 /* Send msg to other side. */
13471                                 bzero(&msg.ua, sizeof(msg.ua));
13472                                 msg.hdr.msg_type = CTL_MSG_UA;
13473                                 msg.hdr.nexus.initid = -1;
13474                                 msg.hdr.nexus.targ_port = -1;
13475                                 msg.hdr.nexus.targ_lun = lun->lun;
13476                                 msg.hdr.nexus.targ_mapped_lun = lun->lun;
13477                                 msg.ua.ua_all = 1;
13478                                 msg.ua.ua_set = (set > 0);
13479                                 msg.ua.ua_type = CTL_UA_THIN_PROV_THRES;
13480                                 memcpy(msg.ua.ua_info, lun->ua_tpt_info, 8);
13481                                 mtx_unlock(&softc->ctl_lock); // XXX
13482                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13483                                     sizeof(msg.ua), M_WAITOK);
13484                                 mtx_lock(&softc->ctl_lock);
13485                         }
13486                 }
13487                 mtx_unlock(&softc->ctl_lock);
13488                 pause("-", CTL_LBP_PERIOD * hz);
13489         }
13490 }
13491
13492 static void
13493 ctl_enqueue_incoming(union ctl_io *io)
13494 {
13495         struct ctl_softc *softc = control_softc;
13496         struct ctl_thread *thr;
13497         u_int idx;
13498
13499         idx = (io->io_hdr.nexus.targ_port * 127 +
13500                io->io_hdr.nexus.initid) % worker_threads;
13501         thr = &softc->threads[idx];
13502         mtx_lock(&thr->queue_lock);
13503         STAILQ_INSERT_TAIL(&thr->incoming_queue, &io->io_hdr, links);
13504         mtx_unlock(&thr->queue_lock);
13505         wakeup(thr);
13506 }
13507
13508 static void
13509 ctl_enqueue_rtr(union ctl_io *io)
13510 {
13511         struct ctl_softc *softc = control_softc;
13512         struct ctl_thread *thr;
13513
13514         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13515         mtx_lock(&thr->queue_lock);
13516         STAILQ_INSERT_TAIL(&thr->rtr_queue, &io->io_hdr, links);
13517         mtx_unlock(&thr->queue_lock);
13518         wakeup(thr);
13519 }
13520
13521 static void
13522 ctl_enqueue_done(union ctl_io *io)
13523 {
13524         struct ctl_softc *softc = control_softc;
13525         struct ctl_thread *thr;
13526
13527         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13528         mtx_lock(&thr->queue_lock);
13529         STAILQ_INSERT_TAIL(&thr->done_queue, &io->io_hdr, links);
13530         mtx_unlock(&thr->queue_lock);
13531         wakeup(thr);
13532 }
13533
13534 static void
13535 ctl_enqueue_isc(union ctl_io *io)
13536 {
13537         struct ctl_softc *softc = control_softc;
13538         struct ctl_thread *thr;
13539
13540         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13541         mtx_lock(&thr->queue_lock);
13542         STAILQ_INSERT_TAIL(&thr->isc_queue, &io->io_hdr, links);
13543         mtx_unlock(&thr->queue_lock);
13544         wakeup(thr);
13545 }
13546
13547 /*
13548  *  vim: ts=8
13549  */