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_lun_reset(struct ctl_lun *lun, union ctl_io *io,
 418                          ctl_ua_type ua_type);
 419 static int ctl_abort_task(union ctl_io *io);
 420 static int ctl_abort_task_set(union ctl_io *io);
 421 static int ctl_i_t_nexus_reset(union ctl_io *io);
 422 static void ctl_run_task(union ctl_io *io);
 423 #ifdef CTL_IO_DELAY
 424 static void ctl_datamove_timer_wakeup(void *arg);
 425 static void ctl_done_timer_wakeup(void *arg);
 426 #endif /* CTL_IO_DELAY */
 427
 428 static void ctl_send_datamove_done(union ctl_io *io, int have_lock);
 429 static void ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq);
 430 static int ctl_datamove_remote_dm_write_cb(union ctl_io *io);
 431 static void ctl_datamove_remote_write(union ctl_io *io);
 432 static int ctl_datamove_remote_dm_read_cb(union ctl_io *io);
 433 static void ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq);
 434 static int ctl_datamove_remote_sgl_setup(union ctl_io *io);
 435 static int ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
 436                                     ctl_ha_dt_cb callback);
 437 static void ctl_datamove_remote_read(union ctl_io *io);
 438 static void ctl_datamove_remote(union ctl_io *io);
 439 static int ctl_process_done(union ctl_io *io);
 440 static void ctl_lun_thread(void *arg);
 441 static void ctl_thresh_thread(void *arg);
 442 static void ctl_work_thread(void *arg);
 443 static void ctl_enqueue_incoming(union ctl_io *io);
 444 static void ctl_enqueue_rtr(union ctl_io *io);
 445 static void ctl_enqueue_done(union ctl_io *io);
 446 static void ctl_enqueue_isc(union ctl_io *io);
 447 static const struct ctl_cmd_entry *
 448     ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa);
 449 static const struct ctl_cmd_entry *
 450     ctl_validate_command(struct ctl_scsiio *ctsio);
 451 static int ctl_cmd_applicable(uint8_t lun_type,
 452     const struct ctl_cmd_entry *entry);
 453
 454 static uint64_t ctl_get_prkey(struct ctl_lun *lun, uint32_t residx);
 455 static void ctl_clr_prkey(struct ctl_lun *lun, uint32_t residx);
 456 static void ctl_alloc_prkey(struct ctl_lun *lun, uint32_t residx);
 457 static void ctl_set_prkey(struct ctl_lun *lun, uint32_t residx, uint64_t key);
 458
 459 /*
 460  * Load the serialization table.  This isn't very pretty, but is probably
 461  * the easiest way to do it.
 462  */
 463 #include "ctl_ser_table.c"
 464
 465 /*
 466  * We only need to define open, close and ioctl routines for this driver.
 467  */
 468 static struct cdevsw ctl_cdevsw = {
 469         .d_version =    D_VERSION,
 470         .d_flags =      0,
 471         .d_open =       ctl_open,
 472         .d_close =      ctl_close,
 473         .d_ioctl =      ctl_ioctl,
 474         .d_name =       "ctl",
 475 };
 476
 477
 478 MALLOC_DEFINE(M_CTL, "ctlmem", "Memory used for CTL");
 479
 480 static int ctl_module_event_handler(module_t, int /*modeventtype_t*/, void *);
 481
 482 static moduledata_t ctl_moduledata = {
 483         "ctl",
 484         ctl_module_event_handler,
 485         NULL
 486 };
 487
 488 DECLARE_MODULE(ctl, ctl_moduledata, SI_SUB_CONFIGURE, SI_ORDER_THIRD);
 489 MODULE_VERSION(ctl, 1);
 490
 491 static struct ctl_frontend ha_frontend =
 492 {
 493         .name = "ha",
 494 };
 495
 496 static void
 497 ctl_isc_handler_finish_xfer(struct ctl_softc *ctl_softc,
 498                             union ctl_ha_msg *msg_info)
 499 {
 500         struct ctl_scsiio *ctsio;
 501
 502         if (msg_info->hdr.original_sc == NULL) {
 503                 printf("%s: original_sc == NULL!\n", __func__);
 504                 /* XXX KDM now what? */
 505                 return;
 506         }
 507
 508         ctsio = &msg_info->hdr.original_sc->scsiio;
 509         ctsio->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
 510         ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
 511         ctsio->io_hdr.status = msg_info->hdr.status;
 512         ctsio->scsi_status = msg_info->scsi.scsi_status;
 513         ctsio->sense_len = msg_info->scsi.sense_len;
 514         ctsio->sense_residual = msg_info->scsi.sense_residual;
 515         ctsio->residual = msg_info->scsi.residual;
 516         memcpy(&ctsio->sense_data, &msg_info->scsi.sense_data,
 517                msg_info->scsi.sense_len);
 518         memcpy(&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
 519                &msg_info->scsi.lbalen, sizeof(msg_info->scsi.lbalen));
 520         ctl_enqueue_isc((union ctl_io *)ctsio);
 521 }
 522
 523 static void
 524 ctl_isc_handler_finish_ser_only(struct ctl_softc *ctl_softc,
 525                                 union ctl_ha_msg *msg_info)
 526 {
 527         struct ctl_scsiio *ctsio;
 528
 529         if (msg_info->hdr.serializing_sc == NULL) {
 530                 printf("%s: serializing_sc == NULL!\n", __func__);
 531                 /* XXX KDM now what? */
 532                 return;
 533         }
 534
 535         ctsio = &msg_info->hdr.serializing_sc->scsiio;
 536         ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
 537         ctl_enqueue_isc((union ctl_io *)ctsio);
 538 }
 539
 540 void
 541 ctl_isc_announce_lun(struct ctl_lun *lun)
 542 {
 543         struct ctl_softc *softc = lun->ctl_softc;
 544         union ctl_ha_msg *msg;
 545         struct ctl_ha_msg_lun_pr_key pr_key;
 546         int i, k;
 547
 548         if (softc->ha_link != CTL_HA_LINK_ONLINE)
 549                 return;
 550         mtx_lock(&lun->lun_lock);
 551         i = sizeof(msg->lun);
 552         if (lun->lun_devid)
 553                 i += lun->lun_devid->len;
 554         i += sizeof(pr_key) * lun->pr_key_count;
 555 alloc:
 556         mtx_unlock(&lun->lun_lock);
 557         msg = malloc(i, M_CTL, M_WAITOK);
 558         mtx_lock(&lun->lun_lock);
 559         k = sizeof(msg->lun);
 560         if (lun->lun_devid)
 561                 k += lun->lun_devid->len;
 562         k += sizeof(pr_key) * lun->pr_key_count;
 563         if (i < k) {
 564                 free(msg, M_CTL);
 565                 i = k;
 566                 goto alloc;
 567         }
 568         bzero(&msg->lun, sizeof(msg->lun));
 569         msg->hdr.msg_type = CTL_MSG_LUN_SYNC;
 570         msg->hdr.nexus.targ_lun = lun->lun;
 571         msg->hdr.nexus.targ_mapped_lun = lun->lun;
 572         msg->lun.flags = lun->flags;
 573         msg->lun.pr_generation = lun->PRGeneration;
 574         msg->lun.pr_res_idx = lun->pr_res_idx;
 575         msg->lun.pr_res_type = lun->res_type;
 576         msg->lun.pr_key_count = lun->pr_key_count;
 577         i = 0;
 578         if (lun->lun_devid) {
 579                 msg->lun.lun_devid_len = lun->lun_devid->len;
 580                 memcpy(&msg->lun.data[i], lun->lun_devid->data,
 581                     msg->lun.lun_devid_len);
 582                 i += msg->lun.lun_devid_len;
 583         }
 584         for (k = 0; k < CTL_MAX_INITIATORS; k++) {
 585                 if ((pr_key.pr_key = ctl_get_prkey(lun, k)) == 0)
 586                         continue;
 587                 pr_key.pr_iid = k;
 588                 memcpy(&msg->lun.data[i], &pr_key, sizeof(pr_key));
 589                 i += sizeof(pr_key);
 590         }
 591         mtx_unlock(&lun->lun_lock);
 592         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->port, sizeof(msg->port) + i,
 593             M_WAITOK);
 594         free(msg, M_CTL);
 595 }
 596
 597 void
 598 ctl_isc_announce_port(struct ctl_port *port)
 599 {
 600         struct ctl_softc *softc = control_softc;
 601         union ctl_ha_msg *msg;
 602         int i;
 603
 604         if (port->targ_port < softc->port_min ||
 605             port->targ_port >= softc->port_max ||
 606             softc->ha_link != CTL_HA_LINK_ONLINE)
 607                 return;
 608         i = sizeof(msg->port) + strlen(port->port_name) + 1;
 609         if (port->lun_map)
 610                 i += sizeof(uint32_t) * CTL_MAX_LUNS;
 611         if (port->port_devid)
 612                 i += port->port_devid->len;
 613         if (port->target_devid)
 614                 i += port->target_devid->len;
 615         msg = malloc(i, M_CTL, M_WAITOK);
 616         bzero(&msg->port, sizeof(msg->port));
 617         msg->hdr.msg_type = CTL_MSG_PORT_SYNC;
 618         msg->hdr.nexus.targ_port = port->targ_port;
 619         msg->port.port_type = port->port_type;
 620         msg->port.physical_port = port->physical_port;
 621         msg->port.virtual_port = port->virtual_port;
 622         msg->port.status = port->status;
 623         i = 0;
 624         msg->port.name_len = sprintf(&msg->port.data[i],
 625             "%d:%s", softc->ha_id, port->port_name) + 1;
 626         i += msg->port.name_len;
 627         if (port->lun_map) {
 628                 msg->port.lun_map_len = sizeof(uint32_t) * CTL_MAX_LUNS;
 629                 memcpy(&msg->port.data[i], port->lun_map,
 630                     msg->port.lun_map_len);
 631                 i += msg->port.lun_map_len;
 632         }
 633         if (port->port_devid) {
 634                 msg->port.port_devid_len = port->port_devid->len;
 635                 memcpy(&msg->port.data[i], port->port_devid->data,
 636                     msg->port.port_devid_len);
 637                 i += msg->port.port_devid_len;
 638         }
 639         if (port->target_devid) {
 640                 msg->port.target_devid_len = port->target_devid->len;
 641                 memcpy(&msg->port.data[i], port->target_devid->data,
 642                     msg->port.target_devid_len);
 643                 i += msg->port.target_devid_len;
 644         }
 645         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->port, sizeof(msg->port) + i,
 646             M_WAITOK);
 647         free(msg, M_CTL);
 648 }
 649
 650 static void
 651 ctl_isc_ha_link_up(struct ctl_softc *softc)
 652 {
 653         struct ctl_port *port;
 654         struct ctl_lun *lun;
 655
 656         STAILQ_FOREACH(port, &softc->port_list, links)
 657                 ctl_isc_announce_port(port);
 658         STAILQ_FOREACH(lun, &softc->lun_list, links)
 659                 ctl_isc_announce_lun(lun);
 660 }
 661
 662 static void
 663 ctl_isc_ha_link_down(struct ctl_softc *softc)
 664 {
 665         struct ctl_port *port;
 666         struct ctl_lun *lun;
 667         union ctl_io *io;
 668
 669         mtx_lock(&softc->ctl_lock);
 670         STAILQ_FOREACH(lun, &softc->lun_list, links) {
 671                 mtx_lock(&lun->lun_lock);
 672                 if (lun->flags & CTL_LUN_PEER_SC_PRIMARY) {
 673                         lun->flags &= ~CTL_LUN_PEER_SC_PRIMARY;
 674                         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
 675                 }
 676                 mtx_unlock(&lun->lun_lock);
 677
 678                 mtx_unlock(&softc->ctl_lock);
 679                 io = ctl_alloc_io(softc->othersc_pool);
 680                 mtx_lock(&softc->ctl_lock);
 681                 ctl_zero_io(io);
 682                 io->io_hdr.msg_type = CTL_MSG_FAILOVER;
 683                 io->io_hdr.nexus.targ_mapped_lun = lun->lun;
 684                 ctl_enqueue_isc(io);
 685         }
 686
 687         STAILQ_FOREACH(port, &softc->port_list, links) {
 688                 if (port->targ_port >= softc->port_min &&
 689                     port->targ_port < softc->port_max)
 690                         continue;
 691                 port->status &= ~CTL_PORT_STATUS_ONLINE;
 692         }
 693         mtx_unlock(&softc->ctl_lock);
 694 }
 695
 696 static void
 697 ctl_isc_ua(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
 698 {
 699         struct ctl_lun *lun;
 700         uint32_t iid = ctl_get_initindex(&msg->hdr.nexus);
 701
 702         mtx_lock(&softc->ctl_lock);
 703         if (msg->hdr.nexus.targ_lun < CTL_MAX_LUNS &&
 704             (lun = softc->ctl_luns[msg->hdr.nexus.targ_mapped_lun]) != NULL) {
 705                 mtx_lock(&lun->lun_lock);
 706                 mtx_unlock(&softc->ctl_lock);
 707                 if (msg->ua.ua_all) {
 708                         if (msg->ua.ua_set)
 709                                 ctl_est_ua_all(lun, iid, msg->ua.ua_type);
 710                         else
 711                                 ctl_clr_ua_all(lun, iid, msg->ua.ua_type);
 712                 } else {
 713                         if (msg->ua.ua_set)
 714                                 ctl_est_ua(lun, iid, msg->ua.ua_type);
 715                         else
 716                                 ctl_clr_ua(lun, iid, msg->ua.ua_type);
 717                 }
 718                 mtx_unlock(&lun->lun_lock);
 719         } else
 720                 mtx_unlock(&softc->ctl_lock);
 721 }
 722
 723 static void
 724 ctl_isc_lun_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
 725 {
 726         struct ctl_lun *lun;
 727         struct ctl_ha_msg_lun_pr_key pr_key;
 728         int i, k;
 729         ctl_lun_flags oflags;
 730         uint32_t targ_lun;
 731
 732         targ_lun = msg->hdr.nexus.targ_mapped_lun;
 733         mtx_lock(&softc->ctl_lock);
 734         if ((targ_lun >= CTL_MAX_LUNS) ||
 735             ((lun = softc->ctl_luns[targ_lun]) == NULL)) {
 736                 mtx_unlock(&softc->ctl_lock);
 737                 return;
 738         }
 739         mtx_lock(&lun->lun_lock);
 740         mtx_unlock(&softc->ctl_lock);
 741         if (lun->flags & CTL_LUN_DISABLED) {
 742                 mtx_unlock(&lun->lun_lock);
 743                 return;
 744         }
 745         i = (lun->lun_devid != NULL) ? lun->lun_devid->len : 0;
 746         if (msg->lun.lun_devid_len != i || (i > 0 &&
 747             memcmp(&msg->lun.data[0], lun->lun_devid->data, i) != 0)) {
 748                 mtx_unlock(&lun->lun_lock);
 749                 printf("%s: Received conflicting HA LUN %d\n",
 750                     __func__, msg->hdr.nexus.targ_lun);
 751                 return;
 752         } else {
 753                 /* Record whether peer is primary. */
 754                 oflags = lun->flags;
 755                 if ((msg->lun.flags & CTL_LUN_PRIMARY_SC) &&
 756                     (msg->lun.flags & CTL_LUN_DISABLED) == 0)
 757                         lun->flags |= CTL_LUN_PEER_SC_PRIMARY;
 758                 else
 759                         lun->flags &= ~CTL_LUN_PEER_SC_PRIMARY;
 760                 if (oflags != lun->flags)
 761                         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
 762
 763                 /* If peer is primary and we are not -- use data */
 764                 if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
 765                     (lun->flags & CTL_LUN_PEER_SC_PRIMARY)) {
 766                         lun->PRGeneration = msg->lun.pr_generation;
 767                         lun->pr_res_idx = msg->lun.pr_res_idx;
 768                         lun->res_type = msg->lun.pr_res_type;
 769                         lun->pr_key_count = msg->lun.pr_key_count;
 770                         for (k = 0; k < CTL_MAX_INITIATORS; k++)
 771                                 ctl_clr_prkey(lun, k);
 772                         for (k = 0; k < msg->lun.pr_key_count; k++) {
 773                                 memcpy(&pr_key, &msg->lun.data[i],
 774                                     sizeof(pr_key));
 775                                 ctl_alloc_prkey(lun, pr_key.pr_iid);
 776                                 ctl_set_prkey(lun, pr_key.pr_iid,
 777                                     pr_key.pr_key);
 778                                 i += sizeof(pr_key);
 779                         }
 780                 }
 781
 782                 mtx_unlock(&lun->lun_lock);
 783                 CTL_DEBUG_PRINT(("%s: Known LUN %d, peer is %s\n",
 784                     __func__, msg->hdr.nexus.targ_lun,
 785                     (msg->lun.flags & CTL_LUN_PRIMARY_SC) ?
 786                     "primary" : "secondary"));
 787
 788                 /* If we are primary but peer doesn't know -- notify */
 789                 if ((lun->flags & CTL_LUN_PRIMARY_SC) &&
 790                     (msg->lun.flags & CTL_LUN_PEER_SC_PRIMARY) == 0)
 791                         ctl_isc_announce_lun(lun);
 792         }
 793 }
 794
 795 static void
 796 ctl_isc_port_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
 797 {
 798         struct ctl_port *port;
 799         struct ctl_lun *lun;
 800         int i, new;
 801
 802         port = softc->ctl_ports[msg->hdr.nexus.targ_port];
 803         if (port == NULL) {
 804                 CTL_DEBUG_PRINT(("%s: New port %d\n", __func__,
 805                     msg->hdr.nexus.targ_port));
 806                 new = 1;
 807                 port = malloc(sizeof(*port), M_CTL, M_WAITOK | M_ZERO);
 808                 port->frontend = &ha_frontend;
 809                 port->targ_port = msg->hdr.nexus.targ_port;
 810         } else if (port->frontend == &ha_frontend) {
 811                 CTL_DEBUG_PRINT(("%s: Updated port %d\n", __func__,
 812                     msg->hdr.nexus.targ_port));
 813                 new = 0;
 814         } else {
 815                 printf("%s: Received conflicting HA port %d\n",
 816                     __func__, msg->hdr.nexus.targ_port);
 817                 return;
 818         }
 819         port->port_type = msg->port.port_type;
 820         port->physical_port = msg->port.physical_port;
 821         port->virtual_port = msg->port.virtual_port;
 822         port->status = msg->port.status;
 823         i = 0;
 824         free(port->port_name, M_CTL);
 825         port->port_name = strndup(&msg->port.data[i], msg->port.name_len,
 826             M_CTL);
 827         i += msg->port.name_len;
 828         if (msg->port.lun_map_len != 0) {
 829                 if (port->lun_map == NULL)
 830                         port->lun_map = malloc(sizeof(uint32_t) * CTL_MAX_LUNS,
 831                             M_CTL, M_WAITOK);
 832                 memcpy(port->lun_map, &msg->port.data[i],
 833                     sizeof(uint32_t) * CTL_MAX_LUNS);
 834                 i += msg->port.lun_map_len;
 835         } else {
 836                 free(port->lun_map, M_CTL);
 837                 port->lun_map = NULL;
 838         }
 839         if (msg->port.port_devid_len != 0) {
 840                 if (port->port_devid == NULL ||
 841                     port->port_devid->len != msg->port.port_devid_len) {
 842                         free(port->port_devid, M_CTL);
 843                         port->port_devid = malloc(sizeof(struct ctl_devid) +
 844                             msg->port.port_devid_len, M_CTL, M_WAITOK);
 845                 }
 846                 memcpy(port->port_devid->data, &msg->port.data[i],
 847                     msg->port.port_devid_len);
 848                 port->port_devid->len = msg->port.port_devid_len;
 849                 i += msg->port.port_devid_len;
 850         } else {
 851                 free(port->port_devid, M_CTL);
 852                 port->port_devid = NULL;
 853         }
 854         if (msg->port.target_devid_len != 0) {
 855                 if (port->target_devid == NULL ||
 856                     port->target_devid->len != msg->port.target_devid_len) {
 857                         free(port->target_devid, M_CTL);
 858                         port->target_devid = malloc(sizeof(struct ctl_devid) +
 859                             msg->port.target_devid_len, M_CTL, M_WAITOK);
 860                 }
 861                 memcpy(port->target_devid->data, &msg->port.data[i],
 862                     msg->port.target_devid_len);
 863                 port->target_devid->len = msg->port.target_devid_len;
 864                 i += msg->port.target_devid_len;
 865         } else {
 866                 free(port->port_devid, M_CTL);
 867                 port->port_devid = NULL;
 868         }
 869         if (new) {
 870                 if (ctl_port_register(port) != 0) {
 871                         printf("%s: ctl_port_register() failed with error\n",
 872                             __func__);
 873                 }
 874         }
 875         mtx_lock(&softc->ctl_lock);
 876         STAILQ_FOREACH(lun, &softc->lun_list, links) {
 877                 if (ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
 878                         continue;
 879                 mtx_lock(&lun->lun_lock);
 880                 ctl_est_ua_all(lun, -1, CTL_UA_INQ_CHANGE);
 881                 mtx_unlock(&lun->lun_lock);
 882         }
 883         mtx_unlock(&softc->ctl_lock);
 884 }
 885
 886 /*
 887  * ISC (Inter Shelf Communication) event handler.  Events from the HA
 888  * subsystem come in here.
 889  */
 890 static void
 891 ctl_isc_event_handler(ctl_ha_channel channel, ctl_ha_event event, int param)
 892 {
 893         struct ctl_softc *softc;
 894         union ctl_io *io;
 895         struct ctl_prio *presio;
 896         ctl_ha_status isc_status;
 897
 898         softc = control_softc;
 899         CTL_DEBUG_PRINT(("CTL: Isc Msg event %d\n", event));
 900         if (event == CTL_HA_EVT_MSG_RECV) {
 901                 union ctl_ha_msg *msg, msgbuf;
 902
 903                 if (param > sizeof(msgbuf))
 904                         msg = malloc(param, M_CTL, M_WAITOK);
 905                 else
 906                         msg = &msgbuf;
 907                 isc_status = ctl_ha_msg_recv(CTL_HA_CHAN_CTL, msg, param,
 908                     M_WAITOK);
 909                 if (isc_status != CTL_HA_STATUS_SUCCESS) {
 910                         printf("%s: Error receiving message: %d\n",
 911                             __func__, isc_status);
 912                         if (msg != &msgbuf)
 913                                 free(msg, M_CTL);
 914                         return;
 915                 }
 916
 917                 CTL_DEBUG_PRINT(("CTL: msg_type %d\n", msg->msg_type));
 918                 switch (msg->hdr.msg_type) {
 919                 case CTL_MSG_SERIALIZE:
 920                         io = ctl_alloc_io(softc->othersc_pool);
 921                         ctl_zero_io(io);
 922                         // populate ctsio from msg
 923                         io->io_hdr.io_type = CTL_IO_SCSI;
 924                         io->io_hdr.msg_type = CTL_MSG_SERIALIZE;
 925                         io->io_hdr.original_sc = msg->hdr.original_sc;
 926                         io->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC |
 927                                             CTL_FLAG_IO_ACTIVE;
 928                         /*
 929                          * If we're in serialization-only mode, we don't
 930                          * want to go through full done processing.  Thus
 931                          * the COPY flag.
 932                          *
 933                          * XXX KDM add another flag that is more specific.
 934                          */
 935                         if (softc->ha_mode != CTL_HA_MODE_XFER)
 936                                 io->io_hdr.flags |= CTL_FLAG_INT_COPY;
 937                         io->io_hdr.nexus = msg->hdr.nexus;
 938 #if 0
 939                         printf("port %u, iid %u, lun %u\n",
 940                                io->io_hdr.nexus.targ_port,
 941                                io->io_hdr.nexus.initid,
 942                                io->io_hdr.nexus.targ_lun);
 943 #endif
 944                         io->scsiio.tag_num = msg->scsi.tag_num;
 945                         io->scsiio.tag_type = msg->scsi.tag_type;
 946 #ifdef CTL_TIME_IO
 947                         io->io_hdr.start_time = time_uptime;
 948                         getbintime(&io->io_hdr.start_bt);
 949 #endif /* CTL_TIME_IO */
 950                         io->scsiio.cdb_len = msg->scsi.cdb_len;
 951                         memcpy(io->scsiio.cdb, msg->scsi.cdb,
 952                                CTL_MAX_CDBLEN);
 953                         if (softc->ha_mode == CTL_HA_MODE_XFER) {
 954                                 const struct ctl_cmd_entry *entry;
 955
 956                                 entry = ctl_get_cmd_entry(&io->scsiio, NULL);
 957                                 io->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
 958                                 io->io_hdr.flags |=
 959                                         entry->flags & CTL_FLAG_DATA_MASK;
 960                         }
 961                         ctl_enqueue_isc(io);
 962                         break;
 963
 964                 /* Performed on the Originating SC, XFER mode only */
 965                 case CTL_MSG_DATAMOVE: {
 966                         struct ctl_sg_entry *sgl;
 967                         int i, j;
 968
 969                         io = msg->hdr.original_sc;
 970                         if (io == NULL) {
 971                                 printf("%s: original_sc == NULL!\n", __func__);
 972                                 /* XXX KDM do something here */
 973                                 break;
 974                         }
 975                         io->io_hdr.msg_type = CTL_MSG_DATAMOVE;
 976                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
 977                         /*
 978                          * Keep track of this, we need to send it back over
 979                          * when the datamove is complete.
 980                          */
 981                         io->io_hdr.serializing_sc = msg->hdr.serializing_sc;
 982
 983                         if (msg->dt.sg_sequence == 0) {
 984                                 i = msg->dt.kern_sg_entries +
 985                                     io->scsiio.kern_data_len /
 986                                     CTL_HA_DATAMOVE_SEGMENT + 1;
 987                                 sgl = malloc(sizeof(*sgl) * i, M_CTL,
 988                                     M_WAITOK | M_ZERO);
 989                                 io->io_hdr.remote_sglist = sgl;
 990                                 io->io_hdr.local_sglist =
 991                                     &sgl[msg->dt.kern_sg_entries];
 992
 993                                 io->scsiio.kern_data_ptr = (uint8_t *)sgl;
 994
 995                                 io->scsiio.kern_sg_entries =
 996                                         msg->dt.kern_sg_entries;
 997                                 io->scsiio.rem_sg_entries =
 998                                         msg->dt.kern_sg_entries;
 999                                 io->scsiio.kern_data_len =
1000                                         msg->dt.kern_data_len;
1001                                 io->scsiio.kern_total_len =
1002                                         msg->dt.kern_total_len;
1003                                 io->scsiio.kern_data_resid =
1004                                         msg->dt.kern_data_resid;
1005                                 io->scsiio.kern_rel_offset =
1006                                         msg->dt.kern_rel_offset;
1007                                 io->io_hdr.flags &= ~CTL_FLAG_BUS_ADDR;
1008                                 io->io_hdr.flags |= msg->dt.flags &
1009                                     CTL_FLAG_BUS_ADDR;
1010                         } else
1011                                 sgl = (struct ctl_sg_entry *)
1012                                         io->scsiio.kern_data_ptr;
1013
1014                         for (i = msg->dt.sent_sg_entries, j = 0;
1015                              i < (msg->dt.sent_sg_entries +
1016                              msg->dt.cur_sg_entries); i++, j++) {
1017                                 sgl[i].addr = msg->dt.sg_list[j].addr;
1018                                 sgl[i].len = msg->dt.sg_list[j].len;
1019
1020 #if 0
1021                                 printf("%s: L: %p,%d -> %p,%d j=%d, i=%d\n",
1022                                        __func__,
1023                                        msg->dt.sg_list[j].addr,
1024                                        msg->dt.sg_list[j].len,
1025                                        sgl[i].addr, sgl[i].len, j, i);
1026 #endif
1027                         }
1028
1029                         /*
1030                          * If this is the last piece of the I/O, we've got
1031                          * the full S/G list.  Queue processing in the thread.
1032                          * Otherwise wait for the next piece.
1033                          */
1034                         if (msg->dt.sg_last != 0)
1035                                 ctl_enqueue_isc(io);
1036                         break;
1037                 }
1038                 /* Performed on the Serializing (primary) SC, XFER mode only */
1039                 case CTL_MSG_DATAMOVE_DONE: {
1040                         if (msg->hdr.serializing_sc == NULL) {
1041                                 printf("%s: serializing_sc == NULL!\n",
1042                                        __func__);
1043                                 /* XXX KDM now what? */
1044                                 break;
1045                         }
1046                         /*
1047                          * We grab the sense information here in case
1048                          * there was a failure, so we can return status
1049                          * back to the initiator.
1050                          */
1051                         io = msg->hdr.serializing_sc;
1052                         io->io_hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
1053                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1054                         io->io_hdr.port_status = msg->scsi.fetd_status;
1055                         io->scsiio.residual = msg->scsi.residual;
1056                         if (msg->hdr.status != CTL_STATUS_NONE) {
1057                                 io->io_hdr.status = msg->hdr.status;
1058                                 io->scsiio.scsi_status = msg->scsi.scsi_status;
1059                                 io->scsiio.sense_len = msg->scsi.sense_len;
1060                                 io->scsiio.sense_residual =msg->scsi.sense_residual;
1061                                 memcpy(&io->scsiio.sense_data,
1062                                     &msg->scsi.sense_data,
1063                                     msg->scsi.sense_len);
1064                         }
1065                         ctl_enqueue_isc(io);
1066                         break;
1067                 }
1068
1069                 /* Preformed on Originating SC, SER_ONLY mode */
1070                 case CTL_MSG_R2R:
1071                         io = msg->hdr.original_sc;
1072                         if (io == NULL) {
1073                                 printf("%s: original_sc == NULL!\n",
1074                                     __func__);
1075                                 break;
1076                         }
1077                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1078                         io->io_hdr.msg_type = CTL_MSG_R2R;
1079                         io->io_hdr.serializing_sc = msg->hdr.serializing_sc;
1080                         ctl_enqueue_isc(io);
1081                         break;
1082
1083                 /*
1084                  * Performed on Serializing(i.e. primary SC) SC in SER_ONLY
1085                  * mode.
1086                  * Performed on the Originating (i.e. secondary) SC in XFER
1087                  * mode
1088                  */
1089                 case CTL_MSG_FINISH_IO:
1090                         if (softc->ha_mode == CTL_HA_MODE_XFER)
1091                                 ctl_isc_handler_finish_xfer(softc, msg);
1092                         else
1093                                 ctl_isc_handler_finish_ser_only(softc, msg);
1094                         break;
1095
1096                 /* Preformed on Originating SC */
1097                 case CTL_MSG_BAD_JUJU:
1098                         io = msg->hdr.original_sc;
1099                         if (io == NULL) {
1100                                 printf("%s: Bad JUJU!, original_sc is NULL!\n",
1101                                        __func__);
1102                                 break;
1103                         }
1104                         ctl_copy_sense_data(msg, io);
1105                         /*
1106                          * IO should have already been cleaned up on other
1107                          * SC so clear this flag so we won't send a message
1108                          * back to finish the IO there.
1109                          */
1110                         io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
1111                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1112
1113                         /* io = msg->hdr.serializing_sc; */
1114                         io->io_hdr.msg_type = CTL_MSG_BAD_JUJU;
1115                         ctl_enqueue_isc(io);
1116                         break;
1117
1118                 /* Handle resets sent from the other side */
1119                 case CTL_MSG_MANAGE_TASKS: {
1120                         struct ctl_taskio *taskio;
1121                         taskio = (struct ctl_taskio *)ctl_alloc_io(
1122                             softc->othersc_pool);
1123                         ctl_zero_io((union ctl_io *)taskio);
1124                         taskio->io_hdr.io_type = CTL_IO_TASK;
1125                         taskio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
1126                         taskio->io_hdr.nexus = msg->hdr.nexus;
1127                         taskio->task_action = msg->task.task_action;
1128                         taskio->tag_num = msg->task.tag_num;
1129                         taskio->tag_type = msg->task.tag_type;
1130 #ifdef CTL_TIME_IO
1131                         taskio->io_hdr.start_time = time_uptime;
1132                         getbintime(&taskio->io_hdr.start_bt);
1133 #endif /* CTL_TIME_IO */
1134                         ctl_run_task((union ctl_io *)taskio);
1135                         break;
1136                 }
1137                 /* Persistent Reserve action which needs attention */
1138                 case CTL_MSG_PERS_ACTION:
1139                         presio = (struct ctl_prio *)ctl_alloc_io(
1140                             softc->othersc_pool);
1141                         ctl_zero_io((union ctl_io *)presio);
1142                         presio->io_hdr.msg_type = CTL_MSG_PERS_ACTION;
1143                         presio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
1144                         presio->io_hdr.nexus = msg->hdr.nexus;
1145                         presio->pr_msg = msg->pr;
1146                         ctl_enqueue_isc((union ctl_io *)presio);
1147                         break;
1148                 case CTL_MSG_UA:
1149                         ctl_isc_ua(softc, msg, param);
1150                         break;
1151                 case CTL_MSG_PORT_SYNC:
1152                         ctl_isc_port_sync(softc, msg, param);
1153                         break;
1154                 case CTL_MSG_LUN_SYNC:
1155                         ctl_isc_lun_sync(softc, msg, param);
1156                         break;
1157                 default:
1158                         printf("Received HA message of unknown type %d\n",
1159                             msg->hdr.msg_type);
1160                         break;
1161                 }
1162                 if (msg != &msgbuf)
1163                         free(msg, M_CTL);
1164         } else if (event == CTL_HA_EVT_LINK_CHANGE) {
1165                 printf("CTL: HA link status changed from %d to %d\n",
1166                     softc->ha_link, param);
1167                 if (param == softc->ha_link)
1168                         return;
1169                 if (softc->ha_link == CTL_HA_LINK_ONLINE) {
1170                         softc->ha_link = param;
1171                         ctl_isc_ha_link_down(softc);
1172                 } else {
1173                         softc->ha_link = param;
1174                         if (softc->ha_link == CTL_HA_LINK_ONLINE)
1175                                 ctl_isc_ha_link_up(softc);
1176                 }
1177                 return;
1178         } else {
1179                 printf("ctl_isc_event_handler: Unknown event %d\n", event);
1180                 return;
1181         }
1182 }
1183
1184 static void
1185 ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest)
1186 {
1187
1188         memcpy(&dest->scsiio.sense_data, &src->scsi.sense_data,
1189             src->scsi.sense_len);
1190         dest->scsiio.scsi_status = src->scsi.scsi_status;
1191         dest->scsiio.sense_len = src->scsi.sense_len;
1192         dest->io_hdr.status = src->hdr.status;
1193 }
1194
1195 static void
1196 ctl_copy_sense_data_back(union ctl_io *src, union ctl_ha_msg *dest)
1197 {
1198
1199         memcpy(&dest->scsi.sense_data, &src->scsiio.sense_data,
1200             src->scsiio.sense_len);
1201         dest->scsi.scsi_status = src->scsiio.scsi_status;
1202         dest->scsi.sense_len = src->scsiio.sense_len;
1203         dest->hdr.status = src->io_hdr.status;
1204 }
1205
1206 void
1207 ctl_est_ua(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua)
1208 {
1209         struct ctl_softc *softc = lun->ctl_softc;
1210         ctl_ua_type *pu;
1211
1212         if (initidx < softc->init_min || initidx >= softc->init_max)
1213                 return;
1214         mtx_assert(&lun->lun_lock, MA_OWNED);
1215         pu = lun->pending_ua[initidx / CTL_MAX_INIT_PER_PORT];
1216         if (pu == NULL)
1217                 return;
1218         pu[initidx % CTL_MAX_INIT_PER_PORT] |= ua;
1219 }
1220
1221 void
1222 ctl_est_ua_port(struct ctl_lun *lun, int port, uint32_t except, ctl_ua_type ua)
1223 {
1224         int i;
1225
1226         mtx_assert(&lun->lun_lock, MA_OWNED);
1227         if (lun->pending_ua[port] == NULL)
1228                 return;
1229         for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1230                 if (port * CTL_MAX_INIT_PER_PORT + i == except)
1231                         continue;
1232                 lun->pending_ua[port][i] |= ua;
1233         }
1234 }
1235
1236 void
1237 ctl_est_ua_all(struct ctl_lun *lun, uint32_t except, ctl_ua_type ua)
1238 {
1239         struct ctl_softc *softc = lun->ctl_softc;
1240         int i;
1241
1242         mtx_assert(&lun->lun_lock, MA_OWNED);
1243         for (i = softc->port_min; i < softc->port_max; i++)
1244                 ctl_est_ua_port(lun, i, except, ua);
1245 }
1246
1247 void
1248 ctl_clr_ua(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua)
1249 {
1250         struct ctl_softc *softc = lun->ctl_softc;
1251         ctl_ua_type *pu;
1252
1253         if (initidx < softc->init_min || initidx >= softc->init_max)
1254                 return;
1255         mtx_assert(&lun->lun_lock, MA_OWNED);
1256         pu = lun->pending_ua[initidx / CTL_MAX_INIT_PER_PORT];
1257         if (pu == NULL)
1258                 return;
1259         pu[initidx % CTL_MAX_INIT_PER_PORT] &= ~ua;
1260 }
1261
1262 void
1263 ctl_clr_ua_all(struct ctl_lun *lun, uint32_t except, ctl_ua_type ua)
1264 {
1265         struct ctl_softc *softc = lun->ctl_softc;
1266         int i, j;
1267
1268         mtx_assert(&lun->lun_lock, MA_OWNED);
1269         for (i = softc->port_min; i < softc->port_max; i++) {
1270                 if (lun->pending_ua[i] == NULL)
1271                         continue;
1272                 for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
1273                         if (i * CTL_MAX_INIT_PER_PORT + j == except)
1274                                 continue;
1275                         lun->pending_ua[i][j] &= ~ua;
1276                 }
1277         }
1278 }
1279
1280 void
1281 ctl_clr_ua_allluns(struct ctl_softc *ctl_softc, uint32_t initidx,
1282     ctl_ua_type ua_type)
1283 {
1284         struct ctl_lun *lun;
1285
1286         mtx_assert(&ctl_softc->ctl_lock, MA_OWNED);
1287         STAILQ_FOREACH(lun, &ctl_softc->lun_list, links) {
1288                 mtx_lock(&lun->lun_lock);
1289                 ctl_clr_ua(lun, initidx, ua_type);
1290                 mtx_unlock(&lun->lun_lock);
1291         }
1292 }
1293
1294 static int
1295 ctl_ha_role_sysctl(SYSCTL_HANDLER_ARGS)
1296 {
1297         struct ctl_softc *softc = (struct ctl_softc *)arg1;
1298         struct ctl_lun *lun;
1299         struct ctl_lun_req ireq;
1300         int error, value;
1301
1302         value = (softc->flags & CTL_FLAG_ACTIVE_SHELF) ? 0 : 1;
1303         error = sysctl_handle_int(oidp, &value, 0, req);
1304         if ((error != 0) || (req->newptr == NULL))
1305                 return (error);
1306
1307         mtx_lock(&softc->ctl_lock);
1308         if (value == 0)
1309                 softc->flags |= CTL_FLAG_ACTIVE_SHELF;
1310         else
1311                 softc->flags &= ~CTL_FLAG_ACTIVE_SHELF;
1312         STAILQ_FOREACH(lun, &softc->lun_list, links) {
1313                 mtx_unlock(&softc->ctl_lock);
1314                 bzero(&ireq, sizeof(ireq));
1315                 ireq.reqtype = CTL_LUNREQ_MODIFY;
1316                 ireq.reqdata.modify.lun_id = lun->lun;
1317                 lun->backend->ioctl(NULL, CTL_LUN_REQ, (caddr_t)&ireq, 0,
1318                     curthread);
1319                 if (ireq.status != CTL_LUN_OK) {
1320                         printf("%s: CTL_LUNREQ_MODIFY returned %d '%s'\n",
1321                             __func__, ireq.status, ireq.error_str);
1322                 }
1323                 mtx_lock(&softc->ctl_lock);
1324         }
1325         mtx_unlock(&softc->ctl_lock);
1326         return (0);
1327 }
1328
1329 static int
1330 ctl_init(void)
1331 {
1332         struct ctl_softc *softc;
1333         void *other_pool;
1334         int i, error, retval;
1335
1336         retval = 0;
1337         control_softc = malloc(sizeof(*control_softc), M_DEVBUF,
1338                                M_WAITOK | M_ZERO);
1339         softc = control_softc;
1340
1341         softc->dev = make_dev(&ctl_cdevsw, 0, UID_ROOT, GID_OPERATOR, 0600,
1342                               "cam/ctl");
1343
1344         softc->dev->si_drv1 = softc;
1345
1346         sysctl_ctx_init(&softc->sysctl_ctx);
1347         softc->sysctl_tree = SYSCTL_ADD_NODE(&softc->sysctl_ctx,
1348                 SYSCTL_STATIC_CHILDREN(_kern_cam), OID_AUTO, "ctl",
1349                 CTLFLAG_RD, 0, "CAM Target Layer");
1350
1351         if (softc->sysctl_tree == NULL) {
1352                 printf("%s: unable to allocate sysctl tree\n", __func__);
1353                 destroy_dev(softc->dev);
1354                 free(control_softc, M_DEVBUF);
1355                 control_softc = NULL;
1356                 return (ENOMEM);
1357         }
1358
1359         mtx_init(&softc->ctl_lock, "CTL mutex", NULL, MTX_DEF);
1360         softc->io_zone = uma_zcreate("CTL IO", sizeof(union ctl_io),
1361             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1362         softc->open_count = 0;
1363
1364         /*
1365          * Default to actually sending a SYNCHRONIZE CACHE command down to
1366          * the drive.
1367          */
1368         softc->flags = CTL_FLAG_REAL_SYNC;
1369
1370         SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1371             OID_AUTO, "ha_mode", CTLFLAG_RDTUN, (int *)&softc->ha_mode, 0,
1372             "HA mode (0 - act/stby, 1 - serialize only, 2 - xfer)");
1373
1374         /*
1375          * In Copan's HA scheme, the "master" and "slave" roles are
1376          * figured out through the slot the controller is in.  Although it
1377          * is an active/active system, someone has to be in charge.
1378          */
1379         SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1380             OID_AUTO, "ha_id", CTLFLAG_RDTUN, &softc->ha_id, 0,
1381             "HA head ID (0 - no HA)");
1382         if (softc->ha_id == 0 || softc->ha_id > NUM_TARGET_PORT_GROUPS) {
1383                 softc->flags |= CTL_FLAG_ACTIVE_SHELF;
1384                 softc->is_single = 1;
1385                 softc->port_cnt = CTL_MAX_PORTS;
1386                 softc->port_min = 0;
1387         } else {
1388                 softc->port_cnt = CTL_MAX_PORTS / NUM_TARGET_PORT_GROUPS;
1389                 softc->port_min = (softc->ha_id - 1) * softc->port_cnt;
1390         }
1391         softc->port_max = softc->port_min + softc->port_cnt;
1392         softc->init_min = softc->port_min * CTL_MAX_INIT_PER_PORT;
1393         softc->init_max = softc->port_max * CTL_MAX_INIT_PER_PORT;
1394
1395         SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1396             OID_AUTO, "ha_link", CTLFLAG_RD, (int *)&softc->ha_link, 0,
1397             "HA link state (0 - offline, 1 - unknown, 2 - online)");
1398
1399         STAILQ_INIT(&softc->lun_list);
1400         STAILQ_INIT(&softc->pending_lun_queue);
1401         STAILQ_INIT(&softc->fe_list);
1402         STAILQ_INIT(&softc->port_list);
1403         STAILQ_INIT(&softc->be_list);
1404         ctl_tpc_init(softc);
1405
1406         if (ctl_pool_create(softc, "othersc", CTL_POOL_ENTRIES_OTHER_SC,
1407                             &other_pool) != 0)
1408         {
1409                 printf("ctl: can't allocate %d entry other SC pool, "
1410                        "exiting\n", CTL_POOL_ENTRIES_OTHER_SC);
1411                 return (ENOMEM);
1412         }
1413         softc->othersc_pool = other_pool;
1414
1415         if (worker_threads <= 0)
1416                 worker_threads = max(1, mp_ncpus / 4);
1417         if (worker_threads > CTL_MAX_THREADS)
1418                 worker_threads = CTL_MAX_THREADS;
1419
1420         for (i = 0; i < worker_threads; i++) {
1421                 struct ctl_thread *thr = &softc->threads[i];
1422
1423                 mtx_init(&thr->queue_lock, "CTL queue mutex", NULL, MTX_DEF);
1424                 thr->ctl_softc = softc;
1425                 STAILQ_INIT(&thr->incoming_queue);
1426                 STAILQ_INIT(&thr->rtr_queue);
1427                 STAILQ_INIT(&thr->done_queue);
1428                 STAILQ_INIT(&thr->isc_queue);
1429
1430                 error = kproc_kthread_add(ctl_work_thread, thr,
1431                     &softc->ctl_proc, &thr->thread, 0, 0, "ctl", "work%d", i);
1432                 if (error != 0) {
1433                         printf("error creating CTL work thread!\n");
1434                         ctl_pool_free(other_pool);
1435                         return (error);
1436                 }
1437         }
1438         error = kproc_kthread_add(ctl_lun_thread, softc,
1439             &softc->ctl_proc, NULL, 0, 0, "ctl", "lun");
1440         if (error != 0) {
1441                 printf("error creating CTL lun thread!\n");
1442                 ctl_pool_free(other_pool);
1443                 return (error);
1444         }
1445         error = kproc_kthread_add(ctl_thresh_thread, softc,
1446             &softc->ctl_proc, NULL, 0, 0, "ctl", "thresh");
1447         if (error != 0) {
1448                 printf("error creating CTL threshold thread!\n");
1449                 ctl_pool_free(other_pool);
1450                 return (error);
1451         }
1452
1453         SYSCTL_ADD_PROC(&softc->sysctl_ctx,SYSCTL_CHILDREN(softc->sysctl_tree),
1454             OID_AUTO, "ha_role", CTLTYPE_INT | CTLFLAG_RWTUN,
1455             softc, 0, ctl_ha_role_sysctl, "I", "HA role for this head");
1456
1457         if (softc->is_single == 0) {
1458                 ctl_frontend_register(&ha_frontend);
1459                 if (ctl_ha_msg_init(softc) != CTL_HA_STATUS_SUCCESS) {
1460                         printf("ctl_init: ctl_ha_msg_init failed.\n");
1461                         softc->is_single = 1;
1462                 } else
1463                 if (ctl_ha_msg_register(CTL_HA_CHAN_CTL, ctl_isc_event_handler)
1464                     != CTL_HA_STATUS_SUCCESS) {
1465                         printf("ctl_init: ctl_ha_msg_register failed.\n");
1466                         softc->is_single = 1;
1467                 }
1468         }
1469         return (0);
1470 }
1471
1472 void
1473 ctl_shutdown(void)
1474 {
1475         struct ctl_softc *softc;
1476         struct ctl_lun *lun, *next_lun;
1477
1478         softc = (struct ctl_softc *)control_softc;
1479
1480         if (softc->is_single == 0) {
1481                 if (ctl_ha_msg_deregister(CTL_HA_CHAN_CTL)
1482                     != CTL_HA_STATUS_SUCCESS) {
1483                         printf("ctl_shutdown: ctl_ha_msg_deregister failed.\n");
1484                 }
1485                 if (ctl_ha_msg_shutdown(softc) != CTL_HA_STATUS_SUCCESS) {
1486                         printf("ctl_shutdown: ctl_ha_msg_shutdown failed.\n");
1487                 }
1488                 ctl_frontend_deregister(&ha_frontend);
1489         }
1490
1491         mtx_lock(&softc->ctl_lock);
1492
1493         /*
1494          * Free up each LUN.
1495          */
1496         for (lun = STAILQ_FIRST(&softc->lun_list); lun != NULL; lun = next_lun){
1497                 next_lun = STAILQ_NEXT(lun, links);
1498                 ctl_free_lun(lun);
1499         }
1500
1501         mtx_unlock(&softc->ctl_lock);
1502
1503 #if 0
1504         ctl_shutdown_thread(softc->work_thread);
1505         mtx_destroy(&softc->queue_lock);
1506 #endif
1507
1508         ctl_tpc_shutdown(softc);
1509         uma_zdestroy(softc->io_zone);
1510         mtx_destroy(&softc->ctl_lock);
1511
1512         destroy_dev(softc->dev);
1513
1514         sysctl_ctx_free(&softc->sysctl_ctx);
1515
1516         free(control_softc, M_DEVBUF);
1517         control_softc = NULL;
1518 }
1519
1520 static int
1521 ctl_module_event_handler(module_t mod, int what, void *arg)
1522 {
1523
1524         switch (what) {
1525         case MOD_LOAD:
1526                 return (ctl_init());
1527         case MOD_UNLOAD:
1528                 return (EBUSY);
1529         default:
1530                 return (EOPNOTSUPP);
1531         }
1532 }
1533
1534 /*
1535  * XXX KDM should we do some access checks here?  Bump a reference count to
1536  * prevent a CTL module from being unloaded while someone has it open?
1537  */
1538 static int
1539 ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td)
1540 {
1541         return (0);
1542 }
1543
1544 static int
1545 ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td)
1546 {
1547         return (0);
1548 }
1549
1550 /*
1551  * Remove an initiator by port number and initiator ID.
1552  * Returns 0 for success, -1 for failure.
1553  */
1554 int
1555 ctl_remove_initiator(struct ctl_port *port, int iid)
1556 {
1557         struct ctl_softc *softc = control_softc;
1558
1559         mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1560
1561         if (iid > CTL_MAX_INIT_PER_PORT) {
1562                 printf("%s: initiator ID %u > maximun %u!\n",
1563                        __func__, iid, CTL_MAX_INIT_PER_PORT);
1564                 return (-1);
1565         }
1566
1567         mtx_lock(&softc->ctl_lock);
1568         port->wwpn_iid[iid].in_use--;
1569         port->wwpn_iid[iid].last_use = time_uptime;
1570         mtx_unlock(&softc->ctl_lock);
1571
1572         return (0);
1573 }
1574
1575 /*
1576  * Add an initiator to the initiator map.
1577  * Returns iid for success, < 0 for failure.
1578  */
1579 int
1580 ctl_add_initiator(struct ctl_port *port, int iid, uint64_t wwpn, char *name)
1581 {
1582         struct ctl_softc *softc = control_softc;
1583         time_t best_time;
1584         int i, best;
1585
1586         mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
1587
1588         if (iid >= CTL_MAX_INIT_PER_PORT) {
1589                 printf("%s: WWPN %#jx initiator ID %u > maximum %u!\n",
1590                        __func__, wwpn, iid, CTL_MAX_INIT_PER_PORT);
1591                 free(name, M_CTL);
1592                 return (-1);
1593         }
1594
1595         mtx_lock(&softc->ctl_lock);
1596
1597         if (iid < 0 && (wwpn != 0 || name != NULL)) {
1598                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1599                         if (wwpn != 0 && wwpn == port->wwpn_iid[i].wwpn) {
1600                                 iid = i;
1601                                 break;
1602                         }
1603                         if (name != NULL && port->wwpn_iid[i].name != NULL &&
1604                             strcmp(name, port->wwpn_iid[i].name) == 0) {
1605                                 iid = i;
1606                                 break;
1607                         }
1608                 }
1609         }
1610
1611         if (iid < 0) {
1612                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1613                         if (port->wwpn_iid[i].in_use == 0 &&
1614                             port->wwpn_iid[i].wwpn == 0 &&
1615                             port->wwpn_iid[i].name == NULL) {
1616                                 iid = i;
1617                                 break;
1618                         }
1619                 }
1620         }
1621
1622         if (iid < 0) {
1623                 best = -1;
1624                 best_time = INT32_MAX;
1625                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1626                         if (port->wwpn_iid[i].in_use == 0) {
1627                                 if (port->wwpn_iid[i].last_use < best_time) {
1628                                         best = i;
1629                                         best_time = port->wwpn_iid[i].last_use;
1630                                 }
1631                         }
1632                 }
1633                 iid = best;
1634         }
1635
1636         if (iid < 0) {
1637                 mtx_unlock(&softc->ctl_lock);
1638                 free(name, M_CTL);
1639                 return (-2);
1640         }
1641
1642         if (port->wwpn_iid[iid].in_use > 0 && (wwpn != 0 || name != NULL)) {
1643                 /*
1644                  * This is not an error yet.
1645                  */
1646                 if (wwpn != 0 && wwpn == port->wwpn_iid[iid].wwpn) {
1647 #if 0
1648                         printf("%s: port %d iid %u WWPN %#jx arrived"
1649                             " again\n", __func__, port->targ_port,
1650                             iid, (uintmax_t)wwpn);
1651 #endif
1652                         goto take;
1653                 }
1654                 if (name != NULL && port->wwpn_iid[iid].name != NULL &&
1655                     strcmp(name, port->wwpn_iid[iid].name) == 0) {
1656 #if 0
1657                         printf("%s: port %d iid %u name '%s' arrived"
1658                             " again\n", __func__, port->targ_port,
1659                             iid, name);
1660 #endif
1661                         goto take;
1662                 }
1663
1664                 /*
1665                  * This is an error, but what do we do about it?  The
1666                  * driver is telling us we have a new WWPN for this
1667                  * initiator ID, so we pretty much need to use it.
1668                  */
1669                 printf("%s: port %d iid %u WWPN %#jx '%s' arrived,"
1670                     " but WWPN %#jx '%s' is still at that address\n",
1671                     __func__, port->targ_port, iid, wwpn, name,
1672                     (uintmax_t)port->wwpn_iid[iid].wwpn,
1673                     port->wwpn_iid[iid].name);
1674
1675                 /*
1676                  * XXX KDM clear have_ca and ua_pending on each LUN for
1677                  * this initiator.
1678                  */
1679         }
1680 take:
1681         free(port->wwpn_iid[iid].name, M_CTL);
1682         port->wwpn_iid[iid].name = name;
1683         port->wwpn_iid[iid].wwpn = wwpn;
1684         port->wwpn_iid[iid].in_use++;
1685         mtx_unlock(&softc->ctl_lock);
1686
1687         return (iid);
1688 }
1689
1690 static int
1691 ctl_create_iid(struct ctl_port *port, int iid, uint8_t *buf)
1692 {
1693         int len;
1694
1695         switch (port->port_type) {
1696         case CTL_PORT_FC:
1697         {
1698                 struct scsi_transportid_fcp *id =
1699                     (struct scsi_transportid_fcp *)buf;
1700                 if (port->wwpn_iid[iid].wwpn == 0)
1701                         return (0);
1702                 memset(id, 0, sizeof(*id));
1703                 id->format_protocol = SCSI_PROTO_FC;
1704                 scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->n_port_name);
1705                 return (sizeof(*id));
1706         }
1707         case CTL_PORT_ISCSI:
1708         {
1709                 struct scsi_transportid_iscsi_port *id =
1710                     (struct scsi_transportid_iscsi_port *)buf;
1711                 if (port->wwpn_iid[iid].name == NULL)
1712                         return (0);
1713                 memset(id, 0, 256);
1714                 id->format_protocol = SCSI_TRN_ISCSI_FORMAT_PORT |
1715                     SCSI_PROTO_ISCSI;
1716                 len = strlcpy(id->iscsi_name, port->wwpn_iid[iid].name, 252) + 1;
1717                 len = roundup2(min(len, 252), 4);
1718                 scsi_ulto2b(len, id->additional_length);
1719                 return (sizeof(*id) + len);
1720         }
1721         case CTL_PORT_SAS:
1722         {
1723                 struct scsi_transportid_sas *id =
1724                     (struct scsi_transportid_sas *)buf;
1725                 if (port->wwpn_iid[iid].wwpn == 0)
1726                         return (0);
1727                 memset(id, 0, sizeof(*id));
1728                 id->format_protocol = SCSI_PROTO_SAS;
1729                 scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->sas_address);
1730                 return (sizeof(*id));
1731         }
1732         default:
1733         {
1734                 struct scsi_transportid_spi *id =
1735                     (struct scsi_transportid_spi *)buf;
1736                 memset(id, 0, sizeof(*id));
1737                 id->format_protocol = SCSI_PROTO_SPI;
1738                 scsi_ulto2b(iid, id->scsi_addr);
1739                 scsi_ulto2b(port->targ_port, id->rel_trgt_port_id);
1740                 return (sizeof(*id));
1741         }
1742         }
1743 }
1744
1745 /*
1746  * Serialize a command that went down the "wrong" side, and so was sent to
1747  * this controller for execution.  The logic is a little different than the
1748  * standard case in ctl_scsiio_precheck().  Errors in this case need to get
1749  * sent back to the other side, but in the success case, we execute the
1750  * command on this side (XFER mode) or tell the other side to execute it
1751  * (SER_ONLY mode).
1752  */
1753 static int
1754 ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio)
1755 {
1756         struct ctl_softc *softc;
1757         union ctl_ha_msg msg_info;
1758         struct ctl_lun *lun;
1759         const struct ctl_cmd_entry *entry;
1760         int retval = 0;
1761         uint32_t targ_lun;
1762
1763         softc = control_softc;
1764
1765         targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
1766         mtx_lock(&softc->ctl_lock);
1767         if ((targ_lun < CTL_MAX_LUNS) &&
1768             ((lun = softc->ctl_luns[targ_lun]) != NULL)) {
1769                 mtx_lock(&lun->lun_lock);
1770                 mtx_unlock(&softc->ctl_lock);
1771                 /*
1772                  * If the LUN is invalid, pretend that it doesn't exist.
1773                  * It will go away as soon as all pending I/O has been
1774                  * completed.
1775                  */
1776                 if (lun->flags & CTL_LUN_DISABLED) {
1777                         mtx_unlock(&lun->lun_lock);
1778                         lun = NULL;
1779                 }
1780         } else {
1781                 mtx_unlock(&softc->ctl_lock);
1782                 lun = NULL;
1783         }
1784         if (lun == NULL) {
1785                 /*
1786                  * The other node would not send this request to us unless
1787                  * received announce that we are primary node for this LUN.
1788                  * If this LUN does not exist now, it is probably result of
1789                  * a race, so respond to initiator in the most opaque way.
1790                  */
1791                 ctl_set_busy(ctsio);
1792                 ctl_copy_sense_data_back((union ctl_io *)ctsio, &msg_info);
1793                 msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1794                 msg_info.hdr.serializing_sc = NULL;
1795                 msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1796                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1797                     sizeof(msg_info.scsi), M_WAITOK);
1798                 return(1);
1799         }
1800
1801         entry = ctl_get_cmd_entry(ctsio, NULL);
1802         if (ctl_scsiio_lun_check(lun, entry, ctsio) != 0) {
1803                 mtx_unlock(&lun->lun_lock);
1804                 ctl_copy_sense_data_back((union ctl_io *)ctsio, &msg_info);
1805                 msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1806                 msg_info.hdr.serializing_sc = NULL;
1807                 msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1808                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1809                     sizeof(msg_info.scsi), M_WAITOK);
1810                 return(1);
1811         }
1812
1813         ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = lun;
1814         ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = lun->be_lun;
1815
1816         /*
1817          * Every I/O goes into the OOA queue for a
1818          * particular LUN, and stays there until completion.
1819          */
1820 #ifdef CTL_TIME_IO
1821         if (TAILQ_EMPTY(&lun->ooa_queue))
1822                 lun->idle_time += getsbinuptime() - lun->last_busy;
1823 #endif
1824         TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1825
1826         switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
1827                 (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr, ctl_ooaq,
1828                  ooa_links))) {
1829         case CTL_ACTION_BLOCK:
1830                 ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
1831                 TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
1832                                   blocked_links);
1833                 mtx_unlock(&lun->lun_lock);
1834                 break;
1835         case CTL_ACTION_PASS:
1836         case CTL_ACTION_SKIP:
1837                 if (softc->ha_mode == CTL_HA_MODE_XFER) {
1838                         ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
1839                         ctl_enqueue_rtr((union ctl_io *)ctsio);
1840                         mtx_unlock(&lun->lun_lock);
1841                 } else {
1842                         ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
1843                         mtx_unlock(&lun->lun_lock);
1844
1845                         /* send msg back to other side */
1846                         msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1847                         msg_info.hdr.serializing_sc = (union ctl_io *)ctsio;
1848                         msg_info.hdr.msg_type = CTL_MSG_R2R;
1849                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1850                             sizeof(msg_info.hdr), M_WAITOK);
1851                 }
1852                 break;
1853         case CTL_ACTION_OVERLAP:
1854                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1855                 mtx_unlock(&lun->lun_lock);
1856                 retval = 1;
1857
1858                 ctl_set_overlapped_cmd(ctsio);
1859                 ctl_copy_sense_data_back((union ctl_io *)ctsio, &msg_info);
1860                 msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1861                 msg_info.hdr.serializing_sc = NULL;
1862                 msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1863                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1864                     sizeof(msg_info.scsi), M_WAITOK);
1865                 break;
1866         case CTL_ACTION_OVERLAP_TAG:
1867                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1868                 mtx_unlock(&lun->lun_lock);
1869                 retval = 1;
1870                 ctl_set_overlapped_tag(ctsio, ctsio->tag_num);
1871                 ctl_copy_sense_data_back((union ctl_io *)ctsio, &msg_info);
1872                 msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1873                 msg_info.hdr.serializing_sc = NULL;
1874                 msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1875                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1876                     sizeof(msg_info.scsi), M_WAITOK);
1877                 break;
1878         case CTL_ACTION_ERROR:
1879         default:
1880                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
1881                 mtx_unlock(&lun->lun_lock);
1882                 retval = 1;
1883
1884                 ctl_set_internal_failure(ctsio, /*sks_valid*/ 0,
1885                                          /*retry_count*/ 0);
1886                 ctl_copy_sense_data_back((union ctl_io *)ctsio, &msg_info);
1887                 msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
1888                 msg_info.hdr.serializing_sc = NULL;
1889                 msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
1890                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
1891                     sizeof(msg_info.scsi), M_WAITOK);
1892                 break;
1893         }
1894         return (retval);
1895 }
1896
1897 /*
1898  * Returns 0 for success, errno for failure.
1899  */
1900 static int
1901 ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
1902                    struct ctl_ooa *ooa_hdr, struct ctl_ooa_entry *kern_entries)
1903 {
1904         union ctl_io *io;
1905         int retval;
1906
1907         retval = 0;
1908
1909         mtx_lock(&lun->lun_lock);
1910         for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); (io != NULL);
1911              (*cur_fill_num)++, io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
1912              ooa_links)) {
1913                 struct ctl_ooa_entry *entry;
1914
1915                 /*
1916                  * If we've got more than we can fit, just count the
1917                  * remaining entries.
1918                  */
1919                 if (*cur_fill_num >= ooa_hdr->alloc_num)
1920                         continue;
1921
1922                 entry = &kern_entries[*cur_fill_num];
1923
1924                 entry->tag_num = io->scsiio.tag_num;
1925                 entry->lun_num = lun->lun;
1926 #ifdef CTL_TIME_IO
1927                 entry->start_bt = io->io_hdr.start_bt;
1928 #endif
1929                 bcopy(io->scsiio.cdb, entry->cdb, io->scsiio.cdb_len);
1930                 entry->cdb_len = io->scsiio.cdb_len;
1931                 if (io->io_hdr.flags & CTL_FLAG_BLOCKED)
1932                         entry->cmd_flags |= CTL_OOACMD_FLAG_BLOCKED;
1933
1934                 if (io->io_hdr.flags & CTL_FLAG_DMA_INPROG)
1935                         entry->cmd_flags |= CTL_OOACMD_FLAG_DMA;
1936
1937                 if (io->io_hdr.flags & CTL_FLAG_ABORT)
1938                         entry->cmd_flags |= CTL_OOACMD_FLAG_ABORT;
1939
1940                 if (io->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR)
1941                         entry->cmd_flags |= CTL_OOACMD_FLAG_RTR;
1942
1943                 if (io->io_hdr.flags & CTL_FLAG_DMA_QUEUED)
1944                         entry->cmd_flags |= CTL_OOACMD_FLAG_DMA_QUEUED;
1945         }
1946         mtx_unlock(&lun->lun_lock);
1947
1948         return (retval);
1949 }
1950
1951 static void *
1952 ctl_copyin_alloc(void *user_addr, int len, char *error_str,
1953                  size_t error_str_len)
1954 {
1955         void *kptr;
1956
1957         kptr = malloc(len, M_CTL, M_WAITOK | M_ZERO);
1958
1959         if (copyin(user_addr, kptr, len) != 0) {
1960                 snprintf(error_str, error_str_len, "Error copying %d bytes "
1961                          "from user address %p to kernel address %p", len,
1962                          user_addr, kptr);
1963                 free(kptr, M_CTL);
1964                 return (NULL);
1965         }
1966
1967         return (kptr);
1968 }
1969
1970 static void
1971 ctl_free_args(int num_args, struct ctl_be_arg *args)
1972 {
1973         int i;
1974
1975         if (args == NULL)
1976                 return;
1977
1978         for (i = 0; i < num_args; i++) {
1979                 free(args[i].kname, M_CTL);
1980                 free(args[i].kvalue, M_CTL);
1981         }
1982
1983         free(args, M_CTL);
1984 }
1985
1986 static struct ctl_be_arg *
1987 ctl_copyin_args(int num_args, struct ctl_be_arg *uargs,
1988                 char *error_str, size_t error_str_len)
1989 {
1990         struct ctl_be_arg *args;
1991         int i;
1992
1993         args = ctl_copyin_alloc(uargs, num_args * sizeof(*args),
1994                                 error_str, error_str_len);
1995
1996         if (args == NULL)
1997                 goto bailout;
1998
1999         for (i = 0; i < num_args; i++) {
2000                 args[i].kname = NULL;
2001                 args[i].kvalue = NULL;
2002         }
2003
2004         for (i = 0; i < num_args; i++) {
2005                 uint8_t *tmpptr;
2006
2007                 args[i].kname = ctl_copyin_alloc(args[i].name,
2008                         args[i].namelen, error_str, error_str_len);
2009                 if (args[i].kname == NULL)
2010                         goto bailout;
2011
2012                 if (args[i].kname[args[i].namelen - 1] != '\0') {
2013                         snprintf(error_str, error_str_len, "Argument %d "
2014                                  "name is not NUL-terminated", i);
2015                         goto bailout;
2016                 }
2017
2018                 if (args[i].flags & CTL_BEARG_RD) {
2019                         tmpptr = ctl_copyin_alloc(args[i].value,
2020                                 args[i].vallen, error_str, error_str_len);
2021                         if (tmpptr == NULL)
2022                                 goto bailout;
2023                         if ((args[i].flags & CTL_BEARG_ASCII)
2024                          && (tmpptr[args[i].vallen - 1] != '\0')) {
2025                                 snprintf(error_str, error_str_len, "Argument "
2026                                     "%d value is not NUL-terminated", i);
2027                                 goto bailout;
2028                         }
2029                         args[i].kvalue = tmpptr;
2030                 } else {
2031                         args[i].kvalue = malloc(args[i].vallen,
2032                             M_CTL, M_WAITOK | M_ZERO);
2033                 }
2034         }
2035
2036         return (args);
2037 bailout:
2038
2039         ctl_free_args(num_args, args);
2040
2041         return (NULL);
2042 }
2043
2044 static void
2045 ctl_copyout_args(int num_args, struct ctl_be_arg *args)
2046 {
2047         int i;
2048
2049         for (i = 0; i < num_args; i++) {
2050                 if (args[i].flags & CTL_BEARG_WR)
2051                         copyout(args[i].kvalue, args[i].value, args[i].vallen);
2052         }
2053 }
2054
2055 /*
2056  * Escape characters that are illegal or not recommended in XML.
2057  */
2058 int
2059 ctl_sbuf_printf_esc(struct sbuf *sb, char *str, int size)
2060 {
2061         char *end = str + size;
2062         int retval;
2063
2064         retval = 0;
2065
2066         for (; *str && str < end; str++) {
2067                 switch (*str) {
2068                 case '&':
2069                         retval = sbuf_printf(sb, "&amp;");
2070                         break;
2071                 case '>':
2072                         retval = sbuf_printf(sb, "&gt;");
2073                         break;
2074                 case '<':
2075                         retval = sbuf_printf(sb, "&lt;");
2076                         break;
2077                 default:
2078                         retval = sbuf_putc(sb, *str);
2079                         break;
2080                 }
2081
2082                 if (retval != 0)
2083                         break;
2084
2085         }
2086
2087         return (retval);
2088 }
2089
2090 static void
2091 ctl_id_sbuf(struct ctl_devid *id, struct sbuf *sb)
2092 {
2093         struct scsi_vpd_id_descriptor *desc;
2094         int i;
2095
2096         if (id == NULL || id->len < 4)
2097                 return;
2098         desc = (struct scsi_vpd_id_descriptor *)id->data;
2099         switch (desc->id_type & SVPD_ID_TYPE_MASK) {
2100         case SVPD_ID_TYPE_T10:
2101                 sbuf_printf(sb, "t10.");
2102                 break;
2103         case SVPD_ID_TYPE_EUI64:
2104                 sbuf_printf(sb, "eui.");
2105                 break;
2106         case SVPD_ID_TYPE_NAA:
2107                 sbuf_printf(sb, "naa.");
2108                 break;
2109         case SVPD_ID_TYPE_SCSI_NAME:
2110                 break;
2111         }
2112         switch (desc->proto_codeset & SVPD_ID_CODESET_MASK) {
2113         case SVPD_ID_CODESET_BINARY:
2114                 for (i = 0; i < desc->length; i++)
2115                         sbuf_printf(sb, "%02x", desc->identifier[i]);
2116                 break;
2117         case SVPD_ID_CODESET_ASCII:
2118                 sbuf_printf(sb, "%.*s", (int)desc->length,
2119                     (char *)desc->identifier);
2120                 break;
2121         case SVPD_ID_CODESET_UTF8:
2122                 sbuf_printf(sb, "%s", (char *)desc->identifier);
2123                 break;
2124         }
2125 }
2126
2127 static int
2128 ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
2129           struct thread *td)
2130 {
2131         struct ctl_softc *softc;
2132         struct ctl_lun *lun;
2133         int retval;
2134
2135         softc = control_softc;
2136
2137         retval = 0;
2138
2139         switch (cmd) {
2140         case CTL_IO:
2141                 retval = ctl_ioctl_io(dev, cmd, addr, flag, td);
2142                 break;
2143         case CTL_ENABLE_PORT:
2144         case CTL_DISABLE_PORT:
2145         case CTL_SET_PORT_WWNS: {
2146                 struct ctl_port *port;
2147                 struct ctl_port_entry *entry;
2148
2149                 entry = (struct ctl_port_entry *)addr;
2150
2151                 mtx_lock(&softc->ctl_lock);
2152                 STAILQ_FOREACH(port, &softc->port_list, links) {
2153                         int action, done;
2154
2155                         if (port->targ_port < softc->port_min ||
2156                             port->targ_port >= softc->port_max)
2157                                 continue;
2158
2159                         action = 0;
2160                         done = 0;
2161                         if ((entry->port_type == CTL_PORT_NONE)
2162                          && (entry->targ_port == port->targ_port)) {
2163                                 /*
2164                                  * If the user only wants to enable or
2165                                  * disable or set WWNs on a specific port,
2166                                  * do the operation and we're done.
2167                                  */
2168                                 action = 1;
2169                                 done = 1;
2170                         } else if (entry->port_type & port->port_type) {
2171                                 /*
2172                                  * Compare the user's type mask with the
2173                                  * particular frontend type to see if we
2174                                  * have a match.
2175                                  */
2176                                 action = 1;
2177                                 done = 0;
2178
2179                                 /*
2180                                  * Make sure the user isn't trying to set
2181                                  * WWNs on multiple ports at the same time.
2182                                  */
2183                                 if (cmd == CTL_SET_PORT_WWNS) {
2184                                         printf("%s: Can't set WWNs on "
2185                                                "multiple ports\n", __func__);
2186                                         retval = EINVAL;
2187                                         break;
2188                                 }
2189                         }
2190                         if (action == 0)
2191                                 continue;
2192
2193                         /*
2194                          * XXX KDM we have to drop the lock here, because
2195                          * the online/offline operations can potentially
2196                          * block.  We need to reference count the frontends
2197                          * so they can't go away,
2198                          */
2199                         if (cmd == CTL_ENABLE_PORT) {
2200                                 mtx_unlock(&softc->ctl_lock);
2201                                 ctl_port_online(port);
2202                                 mtx_lock(&softc->ctl_lock);
2203                         } else if (cmd == CTL_DISABLE_PORT) {
2204                                 mtx_unlock(&softc->ctl_lock);
2205                                 ctl_port_offline(port);
2206                                 mtx_lock(&softc->ctl_lock);
2207                         } else if (cmd == CTL_SET_PORT_WWNS) {
2208                                 ctl_port_set_wwns(port,
2209                                     (entry->flags & CTL_PORT_WWNN_VALID) ?
2210                                     1 : 0, entry->wwnn,
2211                                     (entry->flags & CTL_PORT_WWPN_VALID) ?
2212                                     1 : 0, entry->wwpn);
2213                         }
2214                         if (done != 0)
2215                                 break;
2216                 }
2217                 mtx_unlock(&softc->ctl_lock);
2218                 break;
2219         }
2220         case CTL_GET_PORT_LIST: {
2221                 struct ctl_port *port;
2222                 struct ctl_port_list *list;
2223                 int i;
2224
2225                 list = (struct ctl_port_list *)addr;
2226
2227                 if (list->alloc_len != (list->alloc_num *
2228                     sizeof(struct ctl_port_entry))) {
2229                         printf("%s: CTL_GET_PORT_LIST: alloc_len %u != "
2230                                "alloc_num %u * sizeof(struct ctl_port_entry) "
2231                                "%zu\n", __func__, list->alloc_len,
2232                                list->alloc_num, sizeof(struct ctl_port_entry));
2233                         retval = EINVAL;
2234                         break;
2235                 }
2236                 list->fill_len = 0;
2237                 list->fill_num = 0;
2238                 list->dropped_num = 0;
2239                 i = 0;
2240                 mtx_lock(&softc->ctl_lock);
2241                 STAILQ_FOREACH(port, &softc->port_list, links) {
2242                         struct ctl_port_entry entry, *list_entry;
2243
2244                         if (list->fill_num >= list->alloc_num) {
2245                                 list->dropped_num++;
2246                                 continue;
2247                         }
2248
2249                         entry.port_type = port->port_type;
2250                         strlcpy(entry.port_name, port->port_name,
2251                                 sizeof(entry.port_name));
2252                         entry.targ_port = port->targ_port;
2253                         entry.physical_port = port->physical_port;
2254                         entry.virtual_port = port->virtual_port;
2255                         entry.wwnn = port->wwnn;
2256                         entry.wwpn = port->wwpn;
2257                         if (port->status & CTL_PORT_STATUS_ONLINE)
2258                                 entry.online = 1;
2259                         else
2260                                 entry.online = 0;
2261
2262                         list_entry = &list->entries[i];
2263
2264                         retval = copyout(&entry, list_entry, sizeof(entry));
2265                         if (retval != 0) {
2266                                 printf("%s: CTL_GET_PORT_LIST: copyout "
2267                                        "returned %d\n", __func__, retval);
2268                                 break;
2269                         }
2270                         i++;
2271                         list->fill_num++;
2272                         list->fill_len += sizeof(entry);
2273                 }
2274                 mtx_unlock(&softc->ctl_lock);
2275
2276                 /*
2277                  * If this is non-zero, we had a copyout fault, so there's
2278                  * probably no point in attempting to set the status inside
2279                  * the structure.
2280                  */
2281                 if (retval != 0)
2282                         break;
2283
2284                 if (list->dropped_num > 0)
2285                         list->status = CTL_PORT_LIST_NEED_MORE_SPACE;
2286                 else
2287                         list->status = CTL_PORT_LIST_OK;
2288                 break;
2289         }
2290         case CTL_DUMP_OOA: {
2291                 union ctl_io *io;
2292                 char printbuf[128];
2293                 struct sbuf sb;
2294
2295                 mtx_lock(&softc->ctl_lock);
2296                 printf("Dumping OOA queues:\n");
2297                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
2298                         mtx_lock(&lun->lun_lock);
2299                         for (io = (union ctl_io *)TAILQ_FIRST(
2300                              &lun->ooa_queue); io != NULL;
2301                              io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2302                              ooa_links)) {
2303                                 sbuf_new(&sb, printbuf, sizeof(printbuf),
2304                                          SBUF_FIXEDLEN);
2305                                 sbuf_printf(&sb, "LUN %jd tag 0x%04x%s%s%s%s: ",
2306                                             (intmax_t)lun->lun,
2307                                             io->scsiio.tag_num,
2308                                             (io->io_hdr.flags &
2309                                             CTL_FLAG_BLOCKED) ? "" : " BLOCKED",
2310                                             (io->io_hdr.flags &
2311                                             CTL_FLAG_DMA_INPROG) ? " DMA" : "",
2312                                             (io->io_hdr.flags &
2313                                             CTL_FLAG_ABORT) ? " ABORT" : "",
2314                                             (io->io_hdr.flags &
2315                                         CTL_FLAG_IS_WAS_ON_RTR) ? " RTR" : "");
2316                                 ctl_scsi_command_string(&io->scsiio, NULL, &sb);
2317                                 sbuf_finish(&sb);
2318                                 printf("%s\n", sbuf_data(&sb));
2319                         }
2320                         mtx_unlock(&lun->lun_lock);
2321                 }
2322                 printf("OOA queues dump done\n");
2323                 mtx_unlock(&softc->ctl_lock);
2324                 break;
2325         }
2326         case CTL_GET_OOA: {
2327                 struct ctl_ooa *ooa_hdr;
2328                 struct ctl_ooa_entry *entries;
2329                 uint32_t cur_fill_num;
2330
2331                 ooa_hdr = (struct ctl_ooa *)addr;
2332
2333                 if ((ooa_hdr->alloc_len == 0)
2334                  || (ooa_hdr->alloc_num == 0)) {
2335                         printf("%s: CTL_GET_OOA: alloc len %u and alloc num %u "
2336                                "must be non-zero\n", __func__,
2337                                ooa_hdr->alloc_len, ooa_hdr->alloc_num);
2338                         retval = EINVAL;
2339                         break;
2340                 }
2341
2342                 if (ooa_hdr->alloc_len != (ooa_hdr->alloc_num *
2343                     sizeof(struct ctl_ooa_entry))) {
2344                         printf("%s: CTL_GET_OOA: alloc len %u must be alloc "
2345                                "num %d * sizeof(struct ctl_ooa_entry) %zd\n",
2346                                __func__, ooa_hdr->alloc_len,
2347                                ooa_hdr->alloc_num,sizeof(struct ctl_ooa_entry));
2348                         retval = EINVAL;
2349                         break;
2350                 }
2351
2352                 entries = malloc(ooa_hdr->alloc_len, M_CTL, M_WAITOK | M_ZERO);
2353                 if (entries == NULL) {
2354                         printf("%s: could not allocate %d bytes for OOA "
2355                                "dump\n", __func__, ooa_hdr->alloc_len);
2356                         retval = ENOMEM;
2357                         break;
2358                 }
2359
2360                 mtx_lock(&softc->ctl_lock);
2361                 if (((ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) == 0)
2362                  && ((ooa_hdr->lun_num >= CTL_MAX_LUNS)
2363                   || (softc->ctl_luns[ooa_hdr->lun_num] == NULL))) {
2364                         mtx_unlock(&softc->ctl_lock);
2365                         free(entries, M_CTL);
2366                         printf("%s: CTL_GET_OOA: invalid LUN %ju\n",
2367                                __func__, (uintmax_t)ooa_hdr->lun_num);
2368                         retval = EINVAL;
2369                         break;
2370                 }
2371
2372                 cur_fill_num = 0;
2373
2374                 if (ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) {
2375                         STAILQ_FOREACH(lun, &softc->lun_list, links) {
2376                                 retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,
2377                                         ooa_hdr, entries);
2378                                 if (retval != 0)
2379                                         break;
2380                         }
2381                         if (retval != 0) {
2382                                 mtx_unlock(&softc->ctl_lock);
2383                                 free(entries, M_CTL);
2384                                 break;
2385                         }
2386                 } else {
2387                         lun = softc->ctl_luns[ooa_hdr->lun_num];
2388
2389                         retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num,ooa_hdr,
2390                                                     entries);
2391                 }
2392                 mtx_unlock(&softc->ctl_lock);
2393
2394                 ooa_hdr->fill_num = min(cur_fill_num, ooa_hdr->alloc_num);
2395                 ooa_hdr->fill_len = ooa_hdr->fill_num *
2396                         sizeof(struct ctl_ooa_entry);
2397                 retval = copyout(entries, ooa_hdr->entries, ooa_hdr->fill_len);
2398                 if (retval != 0) {
2399                         printf("%s: error copying out %d bytes for OOA dump\n",
2400                                __func__, ooa_hdr->fill_len);
2401                 }
2402
2403                 getbintime(&ooa_hdr->cur_bt);
2404
2405                 if (cur_fill_num > ooa_hdr->alloc_num) {
2406                         ooa_hdr->dropped_num = cur_fill_num -ooa_hdr->alloc_num;
2407                         ooa_hdr->status = CTL_OOA_NEED_MORE_SPACE;
2408                 } else {
2409                         ooa_hdr->dropped_num = 0;
2410                         ooa_hdr->status = CTL_OOA_OK;
2411                 }
2412
2413                 free(entries, M_CTL);
2414                 break;
2415         }
2416         case CTL_CHECK_OOA: {
2417                 union ctl_io *io;
2418                 struct ctl_ooa_info *ooa_info;
2419
2420
2421                 ooa_info = (struct ctl_ooa_info *)addr;
2422
2423                 if (ooa_info->lun_id >= CTL_MAX_LUNS) {
2424                         ooa_info->status = CTL_OOA_INVALID_LUN;
2425                         break;
2426                 }
2427                 mtx_lock(&softc->ctl_lock);
2428                 lun = softc->ctl_luns[ooa_info->lun_id];
2429                 if (lun == NULL) {
2430                         mtx_unlock(&softc->ctl_lock);
2431                         ooa_info->status = CTL_OOA_INVALID_LUN;
2432                         break;
2433                 }
2434                 mtx_lock(&lun->lun_lock);
2435                 mtx_unlock(&softc->ctl_lock);
2436                 ooa_info->num_entries = 0;
2437                 for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue);
2438                      io != NULL; io = (union ctl_io *)TAILQ_NEXT(
2439                      &io->io_hdr, ooa_links)) {
2440                         ooa_info->num_entries++;
2441                 }
2442                 mtx_unlock(&lun->lun_lock);
2443
2444                 ooa_info->status = CTL_OOA_SUCCESS;
2445
2446                 break;
2447         }
2448         case CTL_DELAY_IO: {
2449                 struct ctl_io_delay_info *delay_info;
2450
2451                 delay_info = (struct ctl_io_delay_info *)addr;
2452
2453 #ifdef CTL_IO_DELAY
2454                 mtx_lock(&softc->ctl_lock);
2455
2456                 if ((delay_info->lun_id >= CTL_MAX_LUNS)
2457                  || (softc->ctl_luns[delay_info->lun_id] == NULL)) {
2458                         delay_info->status = CTL_DELAY_STATUS_INVALID_LUN;
2459                 } else {
2460                         lun = softc->ctl_luns[delay_info->lun_id];
2461                         mtx_lock(&lun->lun_lock);
2462
2463                         delay_info->status = CTL_DELAY_STATUS_OK;
2464
2465                         switch (delay_info->delay_type) {
2466                         case CTL_DELAY_TYPE_CONT:
2467                                 break;
2468                         case CTL_DELAY_TYPE_ONESHOT:
2469                                 break;
2470                         default:
2471                                 delay_info->status =
2472                                         CTL_DELAY_STATUS_INVALID_TYPE;
2473                                 break;
2474                         }
2475
2476                         switch (delay_info->delay_loc) {
2477                         case CTL_DELAY_LOC_DATAMOVE:
2478                                 lun->delay_info.datamove_type =
2479                                         delay_info->delay_type;
2480                                 lun->delay_info.datamove_delay =
2481                                         delay_info->delay_secs;
2482                                 break;
2483                         case CTL_DELAY_LOC_DONE:
2484                                 lun->delay_info.done_type =
2485                                         delay_info->delay_type;
2486                                 lun->delay_info.done_delay =
2487                                         delay_info->delay_secs;
2488                                 break;
2489                         default:
2490                                 delay_info->status =
2491                                         CTL_DELAY_STATUS_INVALID_LOC;
2492                                 break;
2493                         }
2494                         mtx_unlock(&lun->lun_lock);
2495                 }
2496
2497                 mtx_unlock(&softc->ctl_lock);
2498 #else
2499                 delay_info->status = CTL_DELAY_STATUS_NOT_IMPLEMENTED;
2500 #endif /* CTL_IO_DELAY */
2501                 break;
2502         }
2503         case CTL_REALSYNC_SET: {
2504                 int *syncstate;
2505
2506                 syncstate = (int *)addr;
2507
2508                 mtx_lock(&softc->ctl_lock);
2509                 switch (*syncstate) {
2510                 case 0:
2511                         softc->flags &= ~CTL_FLAG_REAL_SYNC;
2512                         break;
2513                 case 1:
2514                         softc->flags |= CTL_FLAG_REAL_SYNC;
2515                         break;
2516                 default:
2517                         retval = EINVAL;
2518                         break;
2519                 }
2520                 mtx_unlock(&softc->ctl_lock);
2521                 break;
2522         }
2523         case CTL_REALSYNC_GET: {
2524                 int *syncstate;
2525
2526                 syncstate = (int*)addr;
2527
2528                 mtx_lock(&softc->ctl_lock);
2529                 if (softc->flags & CTL_FLAG_REAL_SYNC)
2530                         *syncstate = 1;
2531                 else
2532                         *syncstate = 0;
2533                 mtx_unlock(&softc->ctl_lock);
2534
2535                 break;
2536         }
2537         case CTL_SETSYNC:
2538         case CTL_GETSYNC: {
2539                 struct ctl_sync_info *sync_info;
2540
2541                 sync_info = (struct ctl_sync_info *)addr;
2542
2543                 mtx_lock(&softc->ctl_lock);
2544                 lun = softc->ctl_luns[sync_info->lun_id];
2545                 if (lun == NULL) {
2546                         mtx_unlock(&softc->ctl_lock);
2547                         sync_info->status = CTL_GS_SYNC_NO_LUN;
2548                         break;
2549                 }
2550                 /*
2551                  * Get or set the sync interval.  We're not bounds checking
2552                  * in the set case, hopefully the user won't do something
2553                  * silly.
2554                  */
2555                 mtx_lock(&lun->lun_lock);
2556                 mtx_unlock(&softc->ctl_lock);
2557                 if (cmd == CTL_GETSYNC)
2558                         sync_info->sync_interval = lun->sync_interval;
2559                 else
2560                         lun->sync_interval = sync_info->sync_interval;
2561                 mtx_unlock(&lun->lun_lock);
2562
2563                 sync_info->status = CTL_GS_SYNC_OK;
2564
2565                 break;
2566         }
2567         case CTL_GETSTATS: {
2568                 struct ctl_stats *stats;
2569                 int i;
2570
2571                 stats = (struct ctl_stats *)addr;
2572
2573                 if ((sizeof(struct ctl_lun_io_stats) * softc->num_luns) >
2574                      stats->alloc_len) {
2575                         stats->status = CTL_SS_NEED_MORE_SPACE;
2576                         stats->num_luns = softc->num_luns;
2577                         break;
2578                 }
2579                 /*
2580                  * XXX KDM no locking here.  If the LUN list changes,
2581                  * things can blow up.
2582                  */
2583                 for (i = 0, lun = STAILQ_FIRST(&softc->lun_list); lun != NULL;
2584                      i++, lun = STAILQ_NEXT(lun, links)) {
2585                         retval = copyout(&lun->stats, &stats->lun_stats[i],
2586                                          sizeof(lun->stats));
2587                         if (retval != 0)
2588                                 break;
2589                 }
2590                 stats->num_luns = softc->num_luns;
2591                 stats->fill_len = sizeof(struct ctl_lun_io_stats) *
2592                                  softc->num_luns;
2593                 stats->status = CTL_SS_OK;
2594 #ifdef CTL_TIME_IO
2595                 stats->flags = CTL_STATS_FLAG_TIME_VALID;
2596 #else
2597                 stats->flags = CTL_STATS_FLAG_NONE;
2598 #endif
2599                 getnanouptime(&stats->timestamp);
2600                 break;
2601         }
2602         case CTL_ERROR_INJECT: {
2603                 struct ctl_error_desc *err_desc, *new_err_desc;
2604
2605                 err_desc = (struct ctl_error_desc *)addr;
2606
2607                 new_err_desc = malloc(sizeof(*new_err_desc), M_CTL,
2608                                       M_WAITOK | M_ZERO);
2609                 bcopy(err_desc, new_err_desc, sizeof(*new_err_desc));
2610
2611                 mtx_lock(&softc->ctl_lock);
2612                 lun = softc->ctl_luns[err_desc->lun_id];
2613                 if (lun == NULL) {
2614                         mtx_unlock(&softc->ctl_lock);
2615                         free(new_err_desc, M_CTL);
2616                         printf("%s: CTL_ERROR_INJECT: invalid LUN %ju\n",
2617                                __func__, (uintmax_t)err_desc->lun_id);
2618                         retval = EINVAL;
2619                         break;
2620                 }
2621                 mtx_lock(&lun->lun_lock);
2622                 mtx_unlock(&softc->ctl_lock);
2623
2624                 /*
2625                  * We could do some checking here to verify the validity
2626                  * of the request, but given the complexity of error
2627                  * injection requests, the checking logic would be fairly
2628                  * complex.
2629                  *
2630                  * For now, if the request is invalid, it just won't get
2631                  * executed and might get deleted.
2632                  */
2633                 STAILQ_INSERT_TAIL(&lun->error_list, new_err_desc, links);
2634
2635                 /*
2636                  * XXX KDM check to make sure the serial number is unique,
2637                  * in case we somehow manage to wrap.  That shouldn't
2638                  * happen for a very long time, but it's the right thing to
2639                  * do.
2640                  */
2641                 new_err_desc->serial = lun->error_serial;
2642                 err_desc->serial = lun->error_serial;
2643                 lun->error_serial++;
2644
2645                 mtx_unlock(&lun->lun_lock);
2646                 break;
2647         }
2648         case CTL_ERROR_INJECT_DELETE: {
2649                 struct ctl_error_desc *delete_desc, *desc, *desc2;
2650                 int delete_done;
2651
2652                 delete_desc = (struct ctl_error_desc *)addr;
2653                 delete_done = 0;
2654
2655                 mtx_lock(&softc->ctl_lock);
2656                 lun = softc->ctl_luns[delete_desc->lun_id];
2657                 if (lun == NULL) {
2658                         mtx_unlock(&softc->ctl_lock);
2659                         printf("%s: CTL_ERROR_INJECT_DELETE: invalid LUN %ju\n",
2660                                __func__, (uintmax_t)delete_desc->lun_id);
2661                         retval = EINVAL;
2662                         break;
2663                 }
2664                 mtx_lock(&lun->lun_lock);
2665                 mtx_unlock(&softc->ctl_lock);
2666                 STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
2667                         if (desc->serial != delete_desc->serial)
2668                                 continue;
2669
2670                         STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc,
2671                                       links);
2672                         free(desc, M_CTL);
2673                         delete_done = 1;
2674                 }
2675                 mtx_unlock(&lun->lun_lock);
2676                 if (delete_done == 0) {
2677                         printf("%s: CTL_ERROR_INJECT_DELETE: can't find "
2678                                "error serial %ju on LUN %u\n", __func__,
2679                                delete_desc->serial, delete_desc->lun_id);
2680                         retval = EINVAL;
2681                         break;
2682                 }
2683                 break;
2684         }
2685         case CTL_DUMP_STRUCTS: {
2686                 int i, j, k;
2687                 struct ctl_port *port;
2688                 struct ctl_frontend *fe;
2689
2690                 mtx_lock(&softc->ctl_lock);
2691                 printf("CTL Persistent Reservation information start:\n");
2692                 for (i = 0; i < CTL_MAX_LUNS; i++) {
2693                         lun = softc->ctl_luns[i];
2694
2695                         if ((lun == NULL)
2696                          || ((lun->flags & CTL_LUN_DISABLED) != 0))
2697                                 continue;
2698
2699                         for (j = 0; j < CTL_MAX_PORTS; j++) {
2700                                 if (lun->pr_keys[j] == NULL)
2701                                         continue;
2702                                 for (k = 0; k < CTL_MAX_INIT_PER_PORT; k++){
2703                                         if (lun->pr_keys[j][k] == 0)
2704                                                 continue;
2705                                         printf("  LUN %d port %d iid %d key "
2706                                                "%#jx\n", i, j, k,
2707                                                (uintmax_t)lun->pr_keys[j][k]);
2708                                 }
2709                         }
2710                 }
2711                 printf("CTL Persistent Reservation information end\n");
2712                 printf("CTL Ports:\n");
2713                 STAILQ_FOREACH(port, &softc->port_list, links) {
2714                         printf("  Port %d '%s' Frontend '%s' Type %u pp %d vp %d WWNN "
2715                                "%#jx WWPN %#jx\n", port->targ_port, port->port_name,
2716                                port->frontend->name, port->port_type,
2717                                port->physical_port, port->virtual_port,
2718                                (uintmax_t)port->wwnn, (uintmax_t)port->wwpn);
2719                         for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
2720                                 if (port->wwpn_iid[j].in_use == 0 &&
2721                                     port->wwpn_iid[j].wwpn == 0 &&
2722                                     port->wwpn_iid[j].name == NULL)
2723                                         continue;
2724
2725                                 printf("    iid %u use %d WWPN %#jx '%s'\n",
2726                                     j, port->wwpn_iid[j].in_use,
2727                                     (uintmax_t)port->wwpn_iid[j].wwpn,
2728                                     port->wwpn_iid[j].name);
2729                         }
2730                 }
2731                 printf("CTL Port information end\n");
2732                 mtx_unlock(&softc->ctl_lock);
2733                 /*
2734                  * XXX KDM calling this without a lock.  We'd likely want
2735                  * to drop the lock before calling the frontend's dump
2736                  * routine anyway.
2737                  */
2738                 printf("CTL Frontends:\n");
2739                 STAILQ_FOREACH(fe, &softc->fe_list, links) {
2740                         printf("  Frontend '%s'\n", fe->name);
2741                         if (fe->fe_dump != NULL)
2742                                 fe->fe_dump();
2743                 }
2744                 printf("CTL Frontend information end\n");
2745                 break;
2746         }
2747         case CTL_LUN_REQ: {
2748                 struct ctl_lun_req *lun_req;
2749                 struct ctl_backend_driver *backend;
2750
2751                 lun_req = (struct ctl_lun_req *)addr;
2752
2753                 backend = ctl_backend_find(lun_req->backend);
2754                 if (backend == NULL) {
2755                         lun_req->status = CTL_LUN_ERROR;
2756                         snprintf(lun_req->error_str,
2757                                  sizeof(lun_req->error_str),
2758                                  "Backend \"%s\" not found.",
2759                                  lun_req->backend);
2760                         break;
2761                 }
2762                 if (lun_req->num_be_args > 0) {
2763                         lun_req->kern_be_args = ctl_copyin_args(
2764                                 lun_req->num_be_args,
2765                                 lun_req->be_args,
2766                                 lun_req->error_str,
2767                                 sizeof(lun_req->error_str));
2768                         if (lun_req->kern_be_args == NULL) {
2769                                 lun_req->status = CTL_LUN_ERROR;
2770                                 break;
2771                         }
2772                 }
2773
2774                 retval = backend->ioctl(dev, cmd, addr, flag, td);
2775
2776                 if (lun_req->num_be_args > 0) {
2777                         ctl_copyout_args(lun_req->num_be_args,
2778                                       lun_req->kern_be_args);
2779                         ctl_free_args(lun_req->num_be_args,
2780                                       lun_req->kern_be_args);
2781                 }
2782                 break;
2783         }
2784         case CTL_LUN_LIST: {
2785                 struct sbuf *sb;
2786                 struct ctl_lun_list *list;
2787                 struct ctl_option *opt;
2788
2789                 list = (struct ctl_lun_list *)addr;
2790
2791                 /*
2792                  * Allocate a fixed length sbuf here, based on the length
2793                  * of the user's buffer.  We could allocate an auto-extending
2794                  * buffer, and then tell the user how much larger our
2795                  * amount of data is than his buffer, but that presents
2796                  * some problems:
2797                  *
2798                  * 1.  The sbuf(9) routines use a blocking malloc, and so
2799                  *     we can't hold a lock while calling them with an
2800                  *     auto-extending buffer.
2801                  *
2802                  * 2.  There is not currently a LUN reference counting
2803                  *     mechanism, outside of outstanding transactions on
2804                  *     the LUN's OOA queue.  So a LUN could go away on us
2805                  *     while we're getting the LUN number, backend-specific
2806                  *     information, etc.  Thus, given the way things
2807                  *     currently work, we need to hold the CTL lock while
2808                  *     grabbing LUN information.
2809                  *
2810                  * So, from the user's standpoint, the best thing to do is
2811                  * allocate what he thinks is a reasonable buffer length,
2812                  * and then if he gets a CTL_LUN_LIST_NEED_MORE_SPACE error,
2813                  * double the buffer length and try again.  (And repeat
2814                  * that until he succeeds.)
2815                  */
2816                 sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
2817                 if (sb == NULL) {
2818                         list->status = CTL_LUN_LIST_ERROR;
2819                         snprintf(list->error_str, sizeof(list->error_str),
2820                                  "Unable to allocate %d bytes for LUN list",
2821                                  list->alloc_len);
2822                         break;
2823                 }
2824
2825                 sbuf_printf(sb, "<ctllunlist>\n");
2826
2827                 mtx_lock(&softc->ctl_lock);
2828                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
2829                         mtx_lock(&lun->lun_lock);
2830                         retval = sbuf_printf(sb, "<lun id=\"%ju\">\n",
2831                                              (uintmax_t)lun->lun);
2832
2833                         /*
2834                          * Bail out as soon as we see that we've overfilled
2835                          * the buffer.
2836                          */
2837                         if (retval != 0)
2838                                 break;
2839
2840                         retval = sbuf_printf(sb, "\t<backend_type>%s"
2841                                              "</backend_type>\n",
2842                                              (lun->backend == NULL) ?  "none" :
2843                                              lun->backend->name);
2844
2845                         if (retval != 0)
2846                                 break;
2847
2848                         retval = sbuf_printf(sb, "\t<lun_type>%d</lun_type>\n",
2849                                              lun->be_lun->lun_type);
2850
2851                         if (retval != 0)
2852                                 break;
2853
2854                         if (lun->backend == NULL) {
2855                                 retval = sbuf_printf(sb, "</lun>\n");
2856                                 if (retval != 0)
2857                                         break;
2858                                 continue;
2859                         }
2860
2861                         retval = sbuf_printf(sb, "\t<size>%ju</size>\n",
2862                                              (lun->be_lun->maxlba > 0) ?
2863                                              lun->be_lun->maxlba + 1 : 0);
2864
2865                         if (retval != 0)
2866                                 break;
2867
2868                         retval = sbuf_printf(sb, "\t<blocksize>%u</blocksize>\n",
2869                                              lun->be_lun->blocksize);
2870
2871                         if (retval != 0)
2872                                 break;
2873
2874                         retval = sbuf_printf(sb, "\t<serial_number>");
2875
2876                         if (retval != 0)
2877                                 break;
2878
2879                         retval = ctl_sbuf_printf_esc(sb,
2880                             lun->be_lun->serial_num,
2881                             sizeof(lun->be_lun->serial_num));
2882
2883                         if (retval != 0)
2884                                 break;
2885
2886                         retval = sbuf_printf(sb, "</serial_number>\n");
2887
2888                         if (retval != 0)
2889                                 break;
2890
2891                         retval = sbuf_printf(sb, "\t<device_id>");
2892
2893                         if (retval != 0)
2894                                 break;
2895
2896                         retval = ctl_sbuf_printf_esc(sb,
2897                             lun->be_lun->device_id,
2898                             sizeof(lun->be_lun->device_id));
2899
2900                         if (retval != 0)
2901                                 break;
2902
2903                         retval = sbuf_printf(sb, "</device_id>\n");
2904
2905                         if (retval != 0)
2906                                 break;
2907
2908                         if (lun->backend->lun_info != NULL) {
2909                                 retval = lun->backend->lun_info(lun->be_lun->be_lun, sb);
2910                                 if (retval != 0)
2911                                         break;
2912                         }
2913                         STAILQ_FOREACH(opt, &lun->be_lun->options, links) {
2914                                 retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
2915                                     opt->name, opt->value, opt->name);
2916                                 if (retval != 0)
2917                                         break;
2918                         }
2919
2920                         retval = sbuf_printf(sb, "</lun>\n");
2921
2922                         if (retval != 0)
2923                                 break;
2924                         mtx_unlock(&lun->lun_lock);
2925                 }
2926                 if (lun != NULL)
2927                         mtx_unlock(&lun->lun_lock);
2928                 mtx_unlock(&softc->ctl_lock);
2929
2930                 if ((retval != 0)
2931                  || ((retval = sbuf_printf(sb, "</ctllunlist>\n")) != 0)) {
2932                         retval = 0;
2933                         sbuf_delete(sb);
2934                         list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
2935                         snprintf(list->error_str, sizeof(list->error_str),
2936                                  "Out of space, %d bytes is too small",
2937                                  list->alloc_len);
2938                         break;
2939                 }
2940
2941                 sbuf_finish(sb);
2942
2943                 retval = copyout(sbuf_data(sb), list->lun_xml,
2944                                  sbuf_len(sb) + 1);
2945
2946                 list->fill_len = sbuf_len(sb) + 1;
2947                 list->status = CTL_LUN_LIST_OK;
2948                 sbuf_delete(sb);
2949                 break;
2950         }
2951         case CTL_ISCSI: {
2952                 struct ctl_iscsi *ci;
2953                 struct ctl_frontend *fe;
2954
2955                 ci = (struct ctl_iscsi *)addr;
2956
2957                 fe = ctl_frontend_find("iscsi");
2958                 if (fe == NULL) {
2959                         ci->status = CTL_ISCSI_ERROR;
2960                         snprintf(ci->error_str, sizeof(ci->error_str),
2961                             "Frontend \"iscsi\" not found.");
2962                         break;
2963                 }
2964
2965                 retval = fe->ioctl(dev, cmd, addr, flag, td);
2966                 break;
2967         }
2968         case CTL_PORT_REQ: {
2969                 struct ctl_req *req;
2970                 struct ctl_frontend *fe;
2971
2972                 req = (struct ctl_req *)addr;
2973
2974                 fe = ctl_frontend_find(req->driver);
2975                 if (fe == NULL) {
2976                         req->status = CTL_LUN_ERROR;
2977                         snprintf(req->error_str, sizeof(req->error_str),
2978                             "Frontend \"%s\" not found.", req->driver);
2979                         break;
2980                 }
2981                 if (req->num_args > 0) {
2982                         req->kern_args = ctl_copyin_args(req->num_args,
2983                             req->args, req->error_str, sizeof(req->error_str));
2984                         if (req->kern_args == NULL) {
2985                                 req->status = CTL_LUN_ERROR;
2986                                 break;
2987                         }
2988                 }
2989
2990                 if (fe->ioctl)
2991                         retval = fe->ioctl(dev, cmd, addr, flag, td);
2992                 else
2993                         retval = ENODEV;
2994
2995                 if (req->num_args > 0) {
2996                         ctl_copyout_args(req->num_args, req->kern_args);
2997                         ctl_free_args(req->num_args, req->kern_args);
2998                 }
2999                 break;
3000         }
3001         case CTL_PORT_LIST: {
3002                 struct sbuf *sb;
3003                 struct ctl_port *port;
3004                 struct ctl_lun_list *list;
3005                 struct ctl_option *opt;
3006                 int j;
3007                 uint32_t plun;
3008
3009                 list = (struct ctl_lun_list *)addr;
3010
3011                 sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3012                 if (sb == NULL) {
3013                         list->status = CTL_LUN_LIST_ERROR;
3014                         snprintf(list->error_str, sizeof(list->error_str),
3015                                  "Unable to allocate %d bytes for LUN list",
3016                                  list->alloc_len);
3017                         break;
3018                 }
3019
3020                 sbuf_printf(sb, "<ctlportlist>\n");
3021
3022                 mtx_lock(&softc->ctl_lock);
3023                 STAILQ_FOREACH(port, &softc->port_list, links) {
3024                         retval = sbuf_printf(sb, "<targ_port id=\"%ju\">\n",
3025                                              (uintmax_t)port->targ_port);
3026
3027                         /*
3028                          * Bail out as soon as we see that we've overfilled
3029                          * the buffer.
3030                          */
3031                         if (retval != 0)
3032                                 break;
3033
3034                         retval = sbuf_printf(sb, "\t<frontend_type>%s"
3035                             "</frontend_type>\n", port->frontend->name);
3036                         if (retval != 0)
3037                                 break;
3038
3039                         retval = sbuf_printf(sb, "\t<port_type>%d</port_type>\n",
3040                                              port->port_type);
3041                         if (retval != 0)
3042                                 break;
3043
3044                         retval = sbuf_printf(sb, "\t<online>%s</online>\n",
3045                             (port->status & CTL_PORT_STATUS_ONLINE) ? "YES" : "NO");
3046                         if (retval != 0)
3047                                 break;
3048
3049                         retval = sbuf_printf(sb, "\t<port_name>%s</port_name>\n",
3050                             port->port_name);
3051                         if (retval != 0)
3052                                 break;
3053
3054                         retval = sbuf_printf(sb, "\t<physical_port>%d</physical_port>\n",
3055                             port->physical_port);
3056                         if (retval != 0)
3057                                 break;
3058
3059                         retval = sbuf_printf(sb, "\t<virtual_port>%d</virtual_port>\n",
3060                             port->virtual_port);
3061                         if (retval != 0)
3062                                 break;
3063
3064                         if (port->target_devid != NULL) {
3065                                 sbuf_printf(sb, "\t<target>");
3066                                 ctl_id_sbuf(port->target_devid, sb);
3067                                 sbuf_printf(sb, "</target>\n");
3068                         }
3069
3070                         if (port->port_devid != NULL) {
3071                                 sbuf_printf(sb, "\t<port>");
3072                                 ctl_id_sbuf(port->port_devid, sb);
3073                                 sbuf_printf(sb, "</port>\n");
3074                         }
3075
3076                         if (port->port_info != NULL) {
3077                                 retval = port->port_info(port->onoff_arg, sb);
3078                                 if (retval != 0)
3079                                         break;
3080                         }
3081                         STAILQ_FOREACH(opt, &port->options, links) {
3082                                 retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3083                                     opt->name, opt->value, opt->name);
3084                                 if (retval != 0)
3085                                         break;
3086                         }
3087
3088                         if (port->lun_map != NULL) {
3089                                 sbuf_printf(sb, "\t<lun_map>on</lun_map>\n");
3090                                 for (j = 0; j < CTL_MAX_LUNS; j++) {
3091                                         plun = ctl_lun_map_from_port(port, j);
3092                                         if (plun >= CTL_MAX_LUNS)
3093                                                 continue;
3094                                         sbuf_printf(sb,
3095                                             "\t<lun id=\"%u\">%u</lun>\n",
3096                                             j, plun);
3097                                 }
3098                         }
3099
3100                         for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
3101                                 if (port->wwpn_iid[j].in_use == 0 ||
3102                                     (port->wwpn_iid[j].wwpn == 0 &&
3103                                      port->wwpn_iid[j].name == NULL))
3104                                         continue;
3105
3106                                 if (port->wwpn_iid[j].name != NULL)
3107                                         retval = sbuf_printf(sb,
3108                                             "\t<initiator id=\"%u\">%s</initiator>\n",
3109                                             j, port->wwpn_iid[j].name);
3110                                 else
3111                                         retval = sbuf_printf(sb,
3112                                             "\t<initiator id=\"%u\">naa.%08jx</initiator>\n",
3113                                             j, port->wwpn_iid[j].wwpn);
3114                                 if (retval != 0)
3115                                         break;
3116                         }
3117                         if (retval != 0)
3118                                 break;
3119
3120                         retval = sbuf_printf(sb, "</targ_port>\n");
3121                         if (retval != 0)
3122                                 break;
3123                 }
3124                 mtx_unlock(&softc->ctl_lock);
3125
3126                 if ((retval != 0)
3127                  || ((retval = sbuf_printf(sb, "</ctlportlist>\n")) != 0)) {
3128                         retval = 0;
3129                         sbuf_delete(sb);
3130                         list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3131                         snprintf(list->error_str, sizeof(list->error_str),
3132                                  "Out of space, %d bytes is too small",
3133                                  list->alloc_len);
3134                         break;
3135                 }
3136
3137                 sbuf_finish(sb);
3138
3139                 retval = copyout(sbuf_data(sb), list->lun_xml,
3140                                  sbuf_len(sb) + 1);
3141
3142                 list->fill_len = sbuf_len(sb) + 1;
3143                 list->status = CTL_LUN_LIST_OK;
3144                 sbuf_delete(sb);
3145                 break;
3146         }
3147         case CTL_LUN_MAP: {
3148                 struct ctl_lun_map *lm  = (struct ctl_lun_map *)addr;
3149                 struct ctl_port *port;
3150
3151                 mtx_lock(&softc->ctl_lock);
3152                 if (lm->port < softc->port_min ||
3153                     lm->port >= softc->port_max ||
3154                     (port = softc->ctl_ports[lm->port]) == NULL) {
3155                         mtx_unlock(&softc->ctl_lock);
3156                         return (ENXIO);
3157                 }
3158                 if (port->status & CTL_PORT_STATUS_ONLINE) {
3159                         STAILQ_FOREACH(lun, &softc->lun_list, links) {
3160                                 if (ctl_lun_map_to_port(port, lun->lun) >=
3161                                     CTL_MAX_LUNS)
3162                                         continue;
3163                                 mtx_lock(&lun->lun_lock);
3164                                 ctl_est_ua_port(lun, lm->port, -1,
3165                                     CTL_UA_LUN_CHANGE);
3166                                 mtx_unlock(&lun->lun_lock);
3167                         }
3168                 }
3169                 mtx_unlock(&softc->ctl_lock); // XXX: port_enable sleeps
3170                 if (lm->plun < CTL_MAX_LUNS) {
3171                         if (lm->lun == UINT32_MAX)
3172                                 retval = ctl_lun_map_unset(port, lm->plun);
3173                         else if (lm->lun < CTL_MAX_LUNS &&
3174                             softc->ctl_luns[lm->lun] != NULL)
3175                                 retval = ctl_lun_map_set(port, lm->plun, lm->lun);
3176                         else
3177                                 return (ENXIO);
3178                 } else if (lm->plun == UINT32_MAX) {
3179                         if (lm->lun == UINT32_MAX)
3180                                 retval = ctl_lun_map_deinit(port);
3181                         else
3182                                 retval = ctl_lun_map_init(port);
3183                 } else
3184                         return (ENXIO);
3185                 if (port->status & CTL_PORT_STATUS_ONLINE)
3186                         ctl_isc_announce_port(port);
3187                 break;
3188         }
3189         default: {
3190                 /* XXX KDM should we fix this? */
3191 #if 0
3192                 struct ctl_backend_driver *backend;
3193                 unsigned int type;
3194                 int found;
3195
3196                 found = 0;
3197
3198                 /*
3199                  * We encode the backend type as the ioctl type for backend
3200                  * ioctls.  So parse it out here, and then search for a
3201                  * backend of this type.
3202                  */
3203                 type = _IOC_TYPE(cmd);
3204
3205                 STAILQ_FOREACH(backend, &softc->be_list, links) {
3206                         if (backend->type == type) {
3207                                 found = 1;
3208                                 break;
3209                         }
3210                 }
3211                 if (found == 0) {
3212                         printf("ctl: unknown ioctl command %#lx or backend "
3213                                "%d\n", cmd, type);
3214                         retval = EINVAL;
3215                         break;
3216                 }
3217                 retval = backend->ioctl(dev, cmd, addr, flag, td);
3218 #endif
3219                 retval = ENOTTY;
3220                 break;
3221         }
3222         }
3223         return (retval);
3224 }
3225
3226 uint32_t
3227 ctl_get_initindex(struct ctl_nexus *nexus)
3228 {
3229         return (nexus->initid + (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3230 }
3231
3232 int
3233 ctl_lun_map_init(struct ctl_port *port)
3234 {
3235         struct ctl_softc *softc = control_softc;
3236         struct ctl_lun *lun;
3237         uint32_t i;
3238
3239         if (port->lun_map == NULL)
3240                 port->lun_map = malloc(sizeof(uint32_t) * CTL_MAX_LUNS,
3241                     M_CTL, M_NOWAIT);
3242         if (port->lun_map == NULL)
3243                 return (ENOMEM);
3244         for (i = 0; i < CTL_MAX_LUNS; i++)
3245                 port->lun_map[i] = UINT32_MAX;
3246         if (port->status & CTL_PORT_STATUS_ONLINE) {
3247                 if (port->lun_disable != NULL) {
3248                         STAILQ_FOREACH(lun, &softc->lun_list, links)
3249                                 port->lun_disable(port->targ_lun_arg, lun->lun);
3250                 }
3251                 ctl_isc_announce_port(port);
3252         }
3253         return (0);
3254 }
3255
3256 int
3257 ctl_lun_map_deinit(struct ctl_port *port)
3258 {
3259         struct ctl_softc *softc = control_softc;
3260         struct ctl_lun *lun;
3261
3262         if (port->lun_map == NULL)
3263                 return (0);
3264         free(port->lun_map, M_CTL);
3265         port->lun_map = NULL;
3266         if (port->status & CTL_PORT_STATUS_ONLINE) {
3267                 if (port->lun_enable != NULL) {
3268                         STAILQ_FOREACH(lun, &softc->lun_list, links)
3269                                 port->lun_enable(port->targ_lun_arg, lun->lun);
3270                 }
3271                 ctl_isc_announce_port(port);
3272         }
3273         return (0);
3274 }
3275
3276 int
3277 ctl_lun_map_set(struct ctl_port *port, uint32_t plun, uint32_t glun)
3278 {
3279         int status;
3280         uint32_t old;
3281
3282         if (port->lun_map == NULL) {
3283                 status = ctl_lun_map_init(port);
3284                 if (status != 0)
3285                         return (status);
3286         }
3287         old = port->lun_map[plun];
3288         port->lun_map[plun] = glun;
3289         if ((port->status & CTL_PORT_STATUS_ONLINE) && old >= CTL_MAX_LUNS) {
3290                 if (port->lun_enable != NULL)
3291                         port->lun_enable(port->targ_lun_arg, plun);
3292                 ctl_isc_announce_port(port);
3293         }
3294         return (0);
3295 }
3296
3297 int
3298 ctl_lun_map_unset(struct ctl_port *port, uint32_t plun)
3299 {
3300         uint32_t old;
3301
3302         if (port->lun_map == NULL)
3303                 return (0);
3304         old = port->lun_map[plun];
3305         port->lun_map[plun] = UINT32_MAX;
3306         if ((port->status & CTL_PORT_STATUS_ONLINE) && old < CTL_MAX_LUNS) {
3307                 if (port->lun_disable != NULL)
3308                         port->lun_disable(port->targ_lun_arg, plun);
3309                 ctl_isc_announce_port(port);
3310         }
3311         return (0);
3312 }
3313
3314 uint32_t
3315 ctl_lun_map_from_port(struct ctl_port *port, uint32_t lun_id)
3316 {
3317
3318         if (port == NULL)
3319                 return (UINT32_MAX);
3320         if (port->lun_map == NULL || lun_id >= CTL_MAX_LUNS)
3321                 return (lun_id);
3322         return (port->lun_map[lun_id]);
3323 }
3324
3325 uint32_t
3326 ctl_lun_map_to_port(struct ctl_port *port, uint32_t lun_id)
3327 {
3328         uint32_t i;
3329
3330         if (port == NULL)
3331                 return (UINT32_MAX);
3332         if (port->lun_map == NULL)
3333                 return (lun_id);
3334         for (i = 0; i < CTL_MAX_LUNS; i++) {
3335                 if (port->lun_map[i] == lun_id)
3336                         return (i);
3337         }
3338         return (UINT32_MAX);
3339 }
3340
3341 static struct ctl_port *
3342 ctl_io_port(struct ctl_io_hdr *io_hdr)
3343 {
3344
3345         return (control_softc->ctl_ports[io_hdr->nexus.targ_port]);
3346 }
3347
3348 int
3349 ctl_ffz(uint32_t *mask, uint32_t first, uint32_t last)
3350 {
3351         int i;
3352
3353         for (i = first; i < last; i++) {
3354                 if ((mask[i / 32] & (1 << (i % 32))) == 0)
3355                         return (i);
3356         }
3357         return (-1);
3358 }
3359
3360 int
3361 ctl_set_mask(uint32_t *mask, uint32_t bit)
3362 {
3363         uint32_t chunk, piece;
3364
3365         chunk = bit >> 5;
3366         piece = bit % (sizeof(uint32_t) * 8);
3367
3368         if ((mask[chunk] & (1 << piece)) != 0)
3369                 return (-1);
3370         else
3371                 mask[chunk] |= (1 << piece);
3372
3373         return (0);
3374 }
3375
3376 int
3377 ctl_clear_mask(uint32_t *mask, uint32_t bit)
3378 {
3379         uint32_t chunk, piece;
3380
3381         chunk = bit >> 5;
3382         piece = bit % (sizeof(uint32_t) * 8);
3383
3384         if ((mask[chunk] & (1 << piece)) == 0)
3385                 return (-1);
3386         else
3387                 mask[chunk] &= ~(1 << piece);
3388
3389         return (0);
3390 }
3391
3392 int
3393 ctl_is_set(uint32_t *mask, uint32_t bit)
3394 {
3395         uint32_t chunk, piece;
3396
3397         chunk = bit >> 5;
3398         piece = bit % (sizeof(uint32_t) * 8);
3399
3400         if ((mask[chunk] & (1 << piece)) == 0)
3401                 return (0);
3402         else
3403                 return (1);
3404 }
3405
3406 static uint64_t
3407 ctl_get_prkey(struct ctl_lun *lun, uint32_t residx)
3408 {
3409         uint64_t *t;
3410
3411         t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3412         if (t == NULL)
3413                 return (0);
3414         return (t[residx % CTL_MAX_INIT_PER_PORT]);
3415 }
3416
3417 static void
3418 ctl_clr_prkey(struct ctl_lun *lun, uint32_t residx)
3419 {
3420         uint64_t *t;
3421
3422         t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3423         if (t == NULL)
3424                 return;
3425         t[residx % CTL_MAX_INIT_PER_PORT] = 0;
3426 }
3427
3428 static void
3429 ctl_alloc_prkey(struct ctl_lun *lun, uint32_t residx)
3430 {
3431         uint64_t *p;
3432         u_int i;
3433
3434         i = residx/CTL_MAX_INIT_PER_PORT;
3435         if (lun->pr_keys[i] != NULL)
3436                 return;
3437         mtx_unlock(&lun->lun_lock);
3438         p = malloc(sizeof(uint64_t) * CTL_MAX_INIT_PER_PORT, M_CTL,
3439             M_WAITOK | M_ZERO);
3440         mtx_lock(&lun->lun_lock);
3441         if (lun->pr_keys[i] == NULL)
3442                 lun->pr_keys[i] = p;
3443         else
3444                 free(p, M_CTL);
3445 }
3446
3447 static void
3448 ctl_set_prkey(struct ctl_lun *lun, uint32_t residx, uint64_t key)
3449 {
3450         uint64_t *t;
3451
3452         t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3453         KASSERT(t != NULL, ("prkey %d is not allocated", residx));
3454         t[residx % CTL_MAX_INIT_PER_PORT] = key;
3455 }
3456
3457 /*
3458  * ctl_softc, pool_name, total_ctl_io are passed in.
3459  * npool is passed out.
3460  */
3461 int
3462 ctl_pool_create(struct ctl_softc *ctl_softc, const char *pool_name,
3463                 uint32_t total_ctl_io, void **npool)
3464 {
3465 #ifdef IO_POOLS
3466         struct ctl_io_pool *pool;
3467
3468         pool = (struct ctl_io_pool *)malloc(sizeof(*pool), M_CTL,
3469                                             M_NOWAIT | M_ZERO);
3470         if (pool == NULL)
3471                 return (ENOMEM);
3472
3473         snprintf(pool->name, sizeof(pool->name), "CTL IO %s", pool_name);
3474         pool->ctl_softc = ctl_softc;
3475         pool->zone = uma_zsecond_create(pool->name, NULL,
3476             NULL, NULL, NULL, ctl_softc->io_zone);
3477         /* uma_prealloc(pool->zone, total_ctl_io); */
3478
3479         *npool = pool;
3480 #else
3481         *npool = ctl_softc->io_zone;
3482 #endif
3483         return (0);
3484 }
3485
3486 void
3487 ctl_pool_free(struct ctl_io_pool *pool)
3488 {
3489
3490         if (pool == NULL)
3491                 return;
3492
3493 #ifdef IO_POOLS
3494         uma_zdestroy(pool->zone);
3495         free(pool, M_CTL);
3496 #endif
3497 }
3498
3499 union ctl_io *
3500 ctl_alloc_io(void *pool_ref)
3501 {
3502         union ctl_io *io;
3503 #ifdef IO_POOLS
3504         struct ctl_io_pool *pool = (struct ctl_io_pool *)pool_ref;
3505
3506         io = uma_zalloc(pool->zone, M_WAITOK);
3507 #else
3508         io = uma_zalloc((uma_zone_t)pool_ref, M_WAITOK);
3509 #endif
3510         if (io != NULL)
3511                 io->io_hdr.pool = pool_ref;
3512         return (io);
3513 }
3514
3515 union ctl_io *
3516 ctl_alloc_io_nowait(void *pool_ref)
3517 {
3518         union ctl_io *io;
3519 #ifdef IO_POOLS
3520         struct ctl_io_pool *pool = (struct ctl_io_pool *)pool_ref;
3521
3522         io = uma_zalloc(pool->zone, M_NOWAIT);
3523 #else
3524         io = uma_zalloc((uma_zone_t)pool_ref, M_NOWAIT);
3525 #endif
3526         if (io != NULL)
3527                 io->io_hdr.pool = pool_ref;
3528         return (io);
3529 }
3530
3531 void
3532 ctl_free_io(union ctl_io *io)
3533 {
3534 #ifdef IO_POOLS
3535         struct ctl_io_pool *pool;
3536 #endif
3537
3538         if (io == NULL)
3539                 return;
3540
3541 #ifdef IO_POOLS
3542         pool = (struct ctl_io_pool *)io->io_hdr.pool;
3543         uma_zfree(pool->zone, io);
3544 #else
3545         uma_zfree((uma_zone_t)io->io_hdr.pool, io);
3546 #endif
3547 }
3548
3549 void
3550 ctl_zero_io(union ctl_io *io)
3551 {
3552         void *pool_ref;
3553
3554         if (io == NULL)
3555                 return;
3556
3557         /*
3558          * May need to preserve linked list pointers at some point too.
3559          */
3560         pool_ref = io->io_hdr.pool;
3561         memset(io, 0, sizeof(*io));
3562         io->io_hdr.pool = pool_ref;
3563 }
3564
3565 /*
3566  * This routine is currently used for internal copies of ctl_ios that need
3567  * to persist for some reason after we've already returned status to the
3568  * FETD.  (Thus the flag set.)
3569  *
3570  * XXX XXX
3571  * Note that this makes a blind copy of all fields in the ctl_io, except
3572  * for the pool reference.  This includes any memory that has been
3573  * allocated!  That memory will no longer be valid after done has been
3574  * called, so this would be VERY DANGEROUS for command that actually does
3575  * any reads or writes.  Right now (11/7/2005), this is only used for immediate
3576  * start and stop commands, which don't transfer any data, so this is not a
3577  * problem.  If it is used for anything else, the caller would also need to
3578  * allocate data buffer space and this routine would need to be modified to
3579  * copy the data buffer(s) as well.
3580  */
3581 void
3582 ctl_copy_io(union ctl_io *src, union ctl_io *dest)
3583 {
3584         void *pool_ref;
3585
3586         if ((src == NULL)
3587          || (dest == NULL))
3588                 return;
3589
3590         /*
3591          * May need to preserve linked list pointers at some point too.
3592          */
3593         pool_ref = dest->io_hdr.pool;
3594
3595         memcpy(dest, src, MIN(sizeof(*src), sizeof(*dest)));
3596
3597         dest->io_hdr.pool = pool_ref;
3598         /*
3599          * We need to know that this is an internal copy, and doesn't need
3600          * to get passed back to the FETD that allocated it.
3601          */
3602         dest->io_hdr.flags |= CTL_FLAG_INT_COPY;
3603 }
3604
3605 int
3606 ctl_expand_number(const char *buf, uint64_t *num)
3607 {
3608         char *endptr;
3609         uint64_t number;
3610         unsigned shift;
3611
3612         number = strtoq(buf, &endptr, 0);
3613
3614         switch (tolower((unsigned char)*endptr)) {
3615         case 'e':
3616                 shift = 60;
3617                 break;
3618         case 'p':
3619                 shift = 50;
3620                 break;
3621         case 't':
3622                 shift = 40;
3623                 break;
3624         case 'g':
3625                 shift = 30;
3626                 break;
3627         case 'm':
3628                 shift = 20;
3629                 break;
3630         case 'k':
3631                 shift = 10;
3632                 break;
3633         case 'b':
3634         case '\0': /* No unit. */
3635                 *num = number;
3636                 return (0);
3637         default:
3638                 /* Unrecognized unit. */
3639                 return (-1);
3640         }
3641
3642         if ((number << shift) >> shift != number) {
3643                 /* Overflow */
3644                 return (-1);
3645         }
3646         *num = number << shift;
3647         return (0);
3648 }
3649
3650
3651 /*
3652  * This routine could be used in the future to load default and/or saved
3653  * mode page parameters for a particuar lun.
3654  */
3655 static int
3656 ctl_init_page_index(struct ctl_lun *lun)
3657 {
3658         int i;
3659         struct ctl_page_index *page_index;
3660         const char *value;
3661         uint64_t ival;
3662
3663         memcpy(&lun->mode_pages.index, page_index_template,
3664                sizeof(page_index_template));
3665
3666         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
3667
3668                 page_index = &lun->mode_pages.index[i];
3669                 /*
3670                  * If this is a disk-only mode page, there's no point in
3671                  * setting it up.  For some pages, we have to have some
3672                  * basic information about the disk in order to calculate the
3673                  * mode page data.
3674                  */
3675                 if ((lun->be_lun->lun_type != T_DIRECT)
3676                  && (page_index->page_flags & CTL_PAGE_FLAG_DISK_ONLY))
3677                         continue;
3678
3679                 switch (page_index->page_code & SMPH_PC_MASK) {
3680                 case SMS_RW_ERROR_RECOVERY_PAGE: {
3681                         if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3682                                 panic("subpage is incorrect!");
3683                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_CURRENT],
3684                                &rw_er_page_default,
3685                                sizeof(rw_er_page_default));
3686                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_CHANGEABLE],
3687                                &rw_er_page_changeable,
3688                                sizeof(rw_er_page_changeable));
3689                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_DEFAULT],
3690                                &rw_er_page_default,
3691                                sizeof(rw_er_page_default));
3692                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_SAVED],
3693                                &rw_er_page_default,
3694                                sizeof(rw_er_page_default));
3695                         page_index->page_data =
3696                                 (uint8_t *)lun->mode_pages.rw_er_page;
3697                         break;
3698                 }
3699                 case SMS_FORMAT_DEVICE_PAGE: {
3700                         struct scsi_format_page *format_page;
3701
3702                         if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3703                                 panic("subpage is incorrect!");
3704
3705                         /*
3706                          * Sectors per track are set above.  Bytes per
3707                          * sector need to be set here on a per-LUN basis.
3708                          */
3709                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_CURRENT],
3710                                &format_page_default,
3711                                sizeof(format_page_default));
3712                         memcpy(&lun->mode_pages.format_page[
3713                                CTL_PAGE_CHANGEABLE], &format_page_changeable,
3714                                sizeof(format_page_changeable));
3715                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_DEFAULT],
3716                                &format_page_default,
3717                                sizeof(format_page_default));
3718                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_SAVED],
3719                                &format_page_default,
3720                                sizeof(format_page_default));
3721
3722                         format_page = &lun->mode_pages.format_page[
3723                                 CTL_PAGE_CURRENT];
3724                         scsi_ulto2b(lun->be_lun->blocksize,
3725                                     format_page->bytes_per_sector);
3726
3727                         format_page = &lun->mode_pages.format_page[
3728                                 CTL_PAGE_DEFAULT];
3729                         scsi_ulto2b(lun->be_lun->blocksize,
3730                                     format_page->bytes_per_sector);
3731
3732                         format_page = &lun->mode_pages.format_page[
3733                                 CTL_PAGE_SAVED];
3734                         scsi_ulto2b(lun->be_lun->blocksize,
3735                                     format_page->bytes_per_sector);
3736
3737                         page_index->page_data =
3738                                 (uint8_t *)lun->mode_pages.format_page;
3739                         break;
3740                 }
3741                 case SMS_RIGID_DISK_PAGE: {
3742                         struct scsi_rigid_disk_page *rigid_disk_page;
3743                         uint32_t sectors_per_cylinder;
3744                         uint64_t cylinders;
3745 #ifndef __XSCALE__
3746                         int shift;
3747 #endif /* !__XSCALE__ */
3748
3749                         if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3750                                 panic("invalid subpage value %d",
3751                                       page_index->subpage);
3752
3753                         /*
3754                          * Rotation rate and sectors per track are set
3755                          * above.  We calculate the cylinders here based on
3756                          * capacity.  Due to the number of heads and
3757                          * sectors per track we're using, smaller arrays
3758                          * may turn out to have 0 cylinders.  Linux and
3759                          * FreeBSD don't pay attention to these mode pages
3760                          * to figure out capacity, but Solaris does.  It
3761                          * seems to deal with 0 cylinders just fine, and
3762                          * works out a fake geometry based on the capacity.
3763                          */
3764                         memcpy(&lun->mode_pages.rigid_disk_page[
3765                                CTL_PAGE_DEFAULT], &rigid_disk_page_default,
3766                                sizeof(rigid_disk_page_default));
3767                         memcpy(&lun->mode_pages.rigid_disk_page[
3768                                CTL_PAGE_CHANGEABLE],&rigid_disk_page_changeable,
3769                                sizeof(rigid_disk_page_changeable));
3770
3771                         sectors_per_cylinder = CTL_DEFAULT_SECTORS_PER_TRACK *
3772                                 CTL_DEFAULT_HEADS;
3773
3774                         /*
3775                          * The divide method here will be more accurate,
3776                          * probably, but results in floating point being
3777                          * used in the kernel on i386 (__udivdi3()).  On the
3778                          * XScale, though, __udivdi3() is implemented in
3779                          * software.
3780                          *
3781                          * The shift method for cylinder calculation is
3782                          * accurate if sectors_per_cylinder is a power of
3783                          * 2.  Otherwise it might be slightly off -- you
3784                          * might have a bit of a truncation problem.
3785                          */
3786 #ifdef  __XSCALE__
3787                         cylinders = (lun->be_lun->maxlba + 1) /
3788                                 sectors_per_cylinder;
3789 #else
3790                         for (shift = 31; shift > 0; shift--) {
3791                                 if (sectors_per_cylinder & (1 << shift))
3792                                         break;
3793                         }
3794                         cylinders = (lun->be_lun->maxlba + 1) >> shift;
3795 #endif
3796
3797                         /*
3798                          * We've basically got 3 bytes, or 24 bits for the
3799                          * cylinder size in the mode page.  If we're over,
3800                          * just round down to 2^24.
3801                          */
3802                         if (cylinders > 0xffffff)
3803                                 cylinders = 0xffffff;
3804
3805                         rigid_disk_page = &lun->mode_pages.rigid_disk_page[
3806                                 CTL_PAGE_DEFAULT];
3807                         scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
3808
3809                         if ((value = ctl_get_opt(&lun->be_lun->options,
3810                             "rpm")) != NULL) {
3811                                 scsi_ulto2b(strtol(value, NULL, 0),
3812                                      rigid_disk_page->rotation_rate);
3813                         }
3814
3815                         memcpy(&lun->mode_pages.rigid_disk_page[CTL_PAGE_CURRENT],
3816                                &lun->mode_pages.rigid_disk_page[CTL_PAGE_DEFAULT],
3817                                sizeof(rigid_disk_page_default));
3818                         memcpy(&lun->mode_pages.rigid_disk_page[CTL_PAGE_SAVED],
3819                                &lun->mode_pages.rigid_disk_page[CTL_PAGE_DEFAULT],
3820                                sizeof(rigid_disk_page_default));
3821
3822                         page_index->page_data =
3823                                 (uint8_t *)lun->mode_pages.rigid_disk_page;
3824                         break;
3825                 }
3826                 case SMS_CACHING_PAGE: {
3827                         struct scsi_caching_page *caching_page;
3828
3829                         if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3830                                 panic("invalid subpage value %d",
3831                                       page_index->subpage);
3832                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_DEFAULT],
3833                                &caching_page_default,
3834                                sizeof(caching_page_default));
3835                         memcpy(&lun->mode_pages.caching_page[
3836                                CTL_PAGE_CHANGEABLE], &caching_page_changeable,
3837                                sizeof(caching_page_changeable));
3838                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_SAVED],
3839                                &caching_page_default,
3840                                sizeof(caching_page_default));
3841                         caching_page = &lun->mode_pages.caching_page[
3842                             CTL_PAGE_SAVED];
3843                         value = ctl_get_opt(&lun->be_lun->options, "writecache");
3844                         if (value != NULL && strcmp(value, "off") == 0)
3845                                 caching_page->flags1 &= ~SCP_WCE;
3846                         value = ctl_get_opt(&lun->be_lun->options, "readcache");
3847                         if (value != NULL && strcmp(value, "off") == 0)
3848                                 caching_page->flags1 |= SCP_RCD;
3849                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_CURRENT],
3850                                &lun->mode_pages.caching_page[CTL_PAGE_SAVED],
3851                                sizeof(caching_page_default));
3852                         page_index->page_data =
3853                                 (uint8_t *)lun->mode_pages.caching_page;
3854                         break;
3855                 }
3856                 case SMS_CONTROL_MODE_PAGE: {
3857                         struct scsi_control_page *control_page;
3858
3859                         if (page_index->subpage != SMS_SUBPAGE_PAGE_0)
3860                                 panic("invalid subpage value %d",
3861                                       page_index->subpage);
3862
3863                         memcpy(&lun->mode_pages.control_page[CTL_PAGE_DEFAULT],
3864                                &control_page_default,
3865                                sizeof(control_page_default));
3866                         memcpy(&lun->mode_pages.control_page[
3867                                CTL_PAGE_CHANGEABLE], &control_page_changeable,
3868                                sizeof(control_page_changeable));
3869                         memcpy(&lun->mode_pages.control_page[CTL_PAGE_SAVED],
3870                                &control_page_default,
3871                                sizeof(control_page_default));
3872                         control_page = &lun->mode_pages.control_page[
3873                             CTL_PAGE_SAVED];
3874                         value = ctl_get_opt(&lun->be_lun->options, "reordering");
3875                         if (value != NULL && strcmp(value, "unrestricted") == 0) {
3876                                 control_page->queue_flags &= ~SCP_QUEUE_ALG_MASK;
3877                                 control_page->queue_flags |= SCP_QUEUE_ALG_UNRESTRICTED;
3878                         }
3879                         memcpy(&lun->mode_pages.control_page[CTL_PAGE_CURRENT],
3880                                &lun->mode_pages.control_page[CTL_PAGE_SAVED],
3881                                sizeof(control_page_default));
3882                         page_index->page_data =
3883                                 (uint8_t *)lun->mode_pages.control_page;
3884                         break;
3885
3886                 }
3887                 case SMS_INFO_EXCEPTIONS_PAGE: {
3888                         switch (page_index->subpage) {
3889                         case SMS_SUBPAGE_PAGE_0:
3890                                 memcpy(&lun->mode_pages.ie_page[CTL_PAGE_CURRENT],
3891                                        &ie_page_default,
3892                                        sizeof(ie_page_default));
3893                                 memcpy(&lun->mode_pages.ie_page[
3894                                        CTL_PAGE_CHANGEABLE], &ie_page_changeable,
3895                                        sizeof(ie_page_changeable));
3896                                 memcpy(&lun->mode_pages.ie_page[CTL_PAGE_DEFAULT],
3897                                        &ie_page_default,
3898                                        sizeof(ie_page_default));
3899                                 memcpy(&lun->mode_pages.ie_page[CTL_PAGE_SAVED],
3900                                        &ie_page_default,
3901                                        sizeof(ie_page_default));
3902                                 page_index->page_data =
3903                                         (uint8_t *)lun->mode_pages.ie_page;
3904                                 break;
3905                         case 0x02: {
3906                                 struct ctl_logical_block_provisioning_page *page;
3907
3908                                 memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_DEFAULT],
3909                                        &lbp_page_default,
3910                                        sizeof(lbp_page_default));
3911                                 memcpy(&lun->mode_pages.lbp_page[
3912                                        CTL_PAGE_CHANGEABLE], &lbp_page_changeable,
3913                                        sizeof(lbp_page_changeable));
3914                                 memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_SAVED],
3915                                        &lbp_page_default,
3916                                        sizeof(lbp_page_default));
3917                                 page = &lun->mode_pages.lbp_page[CTL_PAGE_SAVED];
3918                                 value = ctl_get_opt(&lun->be_lun->options,
3919                                     "avail-threshold");
3920                                 if (value != NULL &&
3921                                     ctl_expand_number(value, &ival) == 0) {
3922                                         page->descr[0].flags |= SLBPPD_ENABLED |
3923                                             SLBPPD_ARMING_DEC;
3924                                         if (lun->be_lun->blocksize)
3925                                                 ival /= lun->be_lun->blocksize;
3926                                         else
3927                                                 ival /= 512;
3928                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
3929                                             page->descr[0].count);
3930                                 }
3931                                 value = ctl_get_opt(&lun->be_lun->options,
3932                                     "used-threshold");
3933                                 if (value != NULL &&
3934                                     ctl_expand_number(value, &ival) == 0) {
3935                                         page->descr[1].flags |= SLBPPD_ENABLED |
3936                                             SLBPPD_ARMING_INC;
3937                                         if (lun->be_lun->blocksize)
3938                                                 ival /= lun->be_lun->blocksize;
3939                                         else
3940                                                 ival /= 512;
3941                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
3942                                             page->descr[1].count);
3943                                 }
3944                                 value = ctl_get_opt(&lun->be_lun->options,
3945                                     "pool-avail-threshold");
3946                                 if (value != NULL &&
3947                                     ctl_expand_number(value, &ival) == 0) {
3948                                         page->descr[2].flags |= SLBPPD_ENABLED |
3949                                             SLBPPD_ARMING_DEC;
3950                                         if (lun->be_lun->blocksize)
3951                                                 ival /= lun->be_lun->blocksize;
3952                                         else
3953                                                 ival /= 512;
3954                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
3955                                             page->descr[2].count);
3956                                 }
3957                                 value = ctl_get_opt(&lun->be_lun->options,
3958                                     "pool-used-threshold");
3959                                 if (value != NULL &&
3960                                     ctl_expand_number(value, &ival) == 0) {
3961                                         page->descr[3].flags |= SLBPPD_ENABLED |
3962                                             SLBPPD_ARMING_INC;
3963                                         if (lun->be_lun->blocksize)
3964                                                 ival /= lun->be_lun->blocksize;
3965                                         else
3966                                                 ival /= 512;
3967                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
3968                                             page->descr[3].count);
3969                                 }
3970                                 memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_CURRENT],
3971                                        &lun->mode_pages.lbp_page[CTL_PAGE_SAVED],
3972                                        sizeof(lbp_page_default));
3973                                 page_index->page_data =
3974                                         (uint8_t *)lun->mode_pages.lbp_page;
3975                         }}
3976                         break;
3977                 }
3978                 case SMS_VENDOR_SPECIFIC_PAGE:{
3979                         switch (page_index->subpage) {
3980                         case DBGCNF_SUBPAGE_CODE: {
3981                                 struct copan_debugconf_subpage *current_page,
3982                                                                *saved_page;
3983
3984                                 memcpy(&lun->mode_pages.debugconf_subpage[
3985                                        CTL_PAGE_CURRENT],
3986                                        &debugconf_page_default,
3987                                        sizeof(debugconf_page_default));
3988                                 memcpy(&lun->mode_pages.debugconf_subpage[
3989                                        CTL_PAGE_CHANGEABLE],
3990                                        &debugconf_page_changeable,
3991                                        sizeof(debugconf_page_changeable));
3992                                 memcpy(&lun->mode_pages.debugconf_subpage[
3993                                        CTL_PAGE_DEFAULT],
3994                                        &debugconf_page_default,
3995                                        sizeof(debugconf_page_default));
3996                                 memcpy(&lun->mode_pages.debugconf_subpage[
3997                                        CTL_PAGE_SAVED],
3998                                        &debugconf_page_default,
3999                                        sizeof(debugconf_page_default));
4000                                 page_index->page_data =
4001                                         (uint8_t *)lun->mode_pages.debugconf_subpage;
4002
4003                                 current_page = (struct copan_debugconf_subpage *)
4004                                         (page_index->page_data +
4005                                          (page_index->page_len *
4006                                           CTL_PAGE_CURRENT));
4007                                 saved_page = (struct copan_debugconf_subpage *)
4008                                         (page_index->page_data +
4009                                          (page_index->page_len *
4010                                           CTL_PAGE_SAVED));
4011                                 break;
4012                         }
4013                         default:
4014                                 panic("invalid subpage value %d",
4015                                       page_index->subpage);
4016                                 break;
4017                         }
4018                         break;
4019                 }
4020                 default:
4021                         panic("invalid page value %d",
4022                               page_index->page_code & SMPH_PC_MASK);
4023                         break;
4024         }
4025         }
4026
4027         return (CTL_RETVAL_COMPLETE);
4028 }
4029
4030 static int
4031 ctl_init_log_page_index(struct ctl_lun *lun)
4032 {
4033         struct ctl_page_index *page_index;
4034         int i, j, k, prev;
4035
4036         memcpy(&lun->log_pages.index, log_page_index_template,
4037                sizeof(log_page_index_template));
4038
4039         prev = -1;
4040         for (i = 0, j = 0, k = 0; i < CTL_NUM_LOG_PAGES; i++) {
4041
4042                 page_index = &lun->log_pages.index[i];
4043                 /*
4044                  * If this is a disk-only mode page, there's no point in
4045                  * setting it up.  For some pages, we have to have some
4046                  * basic information about the disk in order to calculate the
4047                  * mode page data.
4048                  */
4049                 if ((lun->be_lun->lun_type != T_DIRECT)
4050                  && (page_index->page_flags & CTL_PAGE_FLAG_DISK_ONLY))
4051                         continue;
4052
4053                 if (page_index->page_code == SLS_LOGICAL_BLOCK_PROVISIONING &&
4054                      lun->backend->lun_attr == NULL)
4055                         continue;
4056
4057                 if (page_index->page_code != prev) {
4058                         lun->log_pages.pages_page[j] = page_index->page_code;
4059                         prev = page_index->page_code;
4060                         j++;
4061                 }
4062                 lun->log_pages.subpages_page[k*2] = page_index->page_code;
4063                 lun->log_pages.subpages_page[k*2+1] = page_index->subpage;
4064                 k++;
4065         }
4066         lun->log_pages.index[0].page_data = &lun->log_pages.pages_page[0];
4067         lun->log_pages.index[0].page_len = j;
4068         lun->log_pages.index[1].page_data = &lun->log_pages.subpages_page[0];
4069         lun->log_pages.index[1].page_len = k * 2;
4070         lun->log_pages.index[2].page_data = &lun->log_pages.lbp_page[0];
4071         lun->log_pages.index[2].page_len = 12*CTL_NUM_LBP_PARAMS;
4072         lun->log_pages.index[3].page_data = (uint8_t *)&lun->log_pages.stat_page;
4073         lun->log_pages.index[3].page_len = sizeof(lun->log_pages.stat_page);
4074
4075         return (CTL_RETVAL_COMPLETE);
4076 }
4077
4078 static int
4079 hex2bin(const char *str, uint8_t *buf, int buf_size)
4080 {
4081         int i;
4082         u_char c;
4083
4084         memset(buf, 0, buf_size);
4085         while (isspace(str[0]))
4086                 str++;
4087         if (str[0] == '0' && (str[1] == 'x' || str[1] == 'X'))
4088                 str += 2;
4089         buf_size *= 2;
4090         for (i = 0; str[i] != 0 && i < buf_size; i++) {
4091                 c = str[i];
4092                 if (isdigit(c))
4093                         c -= '0';
4094                 else if (isalpha(c))
4095                         c -= isupper(c) ? 'A' - 10 : 'a' - 10;
4096                 else
4097                         break;
4098                 if (c >= 16)
4099                         break;
4100                 if ((i & 1) == 0)
4101                         buf[i / 2] |= (c << 4);
4102                 else
4103                         buf[i / 2] |= c;
4104         }
4105         return ((i + 1) / 2);
4106 }
4107
4108 /*
4109  * LUN allocation.
4110  *
4111  * Requirements:
4112  * - caller allocates and zeros LUN storage, or passes in a NULL LUN if he
4113  *   wants us to allocate the LUN and he can block.
4114  * - ctl_softc is always set
4115  * - be_lun is set if the LUN has a backend (needed for disk LUNs)
4116  *
4117  * Returns 0 for success, non-zero (errno) for failure.
4118  */
4119 static int
4120 ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *ctl_lun,
4121               struct ctl_be_lun *const be_lun)
4122 {
4123         struct ctl_lun *nlun, *lun;
4124         struct scsi_vpd_id_descriptor *desc;
4125         struct scsi_vpd_id_t10 *t10id;
4126         const char *eui, *naa, *scsiname, *vendor;
4127         int lun_number, i, lun_malloced;
4128         int devidlen, idlen1, idlen2 = 0, len;
4129
4130         if (be_lun == NULL)
4131                 return (EINVAL);
4132
4133         /*
4134          * We currently only support Direct Access or Processor LUN types.
4135          */
4136         switch (be_lun->lun_type) {
4137         case T_DIRECT:
4138                 break;
4139         case T_PROCESSOR:
4140                 break;
4141         case T_SEQUENTIAL:
4142         case T_CHANGER:
4143         default:
4144                 be_lun->lun_config_status(be_lun->be_lun,
4145                                           CTL_LUN_CONFIG_FAILURE);
4146                 break;
4147         }
4148         if (ctl_lun == NULL) {
4149                 lun = malloc(sizeof(*lun), M_CTL, M_WAITOK);
4150                 lun_malloced = 1;
4151         } else {
4152                 lun_malloced = 0;
4153                 lun = ctl_lun;
4154         }
4155
4156         memset(lun, 0, sizeof(*lun));
4157         if (lun_malloced)
4158                 lun->flags = CTL_LUN_MALLOCED;
4159
4160         /* Generate LUN ID. */
4161         devidlen = max(CTL_DEVID_MIN_LEN,
4162             strnlen(be_lun->device_id, CTL_DEVID_LEN));
4163         idlen1 = sizeof(*t10id) + devidlen;
4164         len = sizeof(struct scsi_vpd_id_descriptor) + idlen1;
4165         scsiname = ctl_get_opt(&be_lun->options, "scsiname");
4166         if (scsiname != NULL) {
4167                 idlen2 = roundup2(strlen(scsiname) + 1, 4);
4168                 len += sizeof(struct scsi_vpd_id_descriptor) + idlen2;
4169         }
4170         eui = ctl_get_opt(&be_lun->options, "eui");
4171         if (eui != NULL) {
4172                 len += sizeof(struct scsi_vpd_id_descriptor) + 16;
4173         }
4174         naa = ctl_get_opt(&be_lun->options, "naa");
4175         if (naa != NULL) {
4176                 len += sizeof(struct scsi_vpd_id_descriptor) + 16;
4177         }
4178         lun->lun_devid = malloc(sizeof(struct ctl_devid) + len,
4179             M_CTL, M_WAITOK | M_ZERO);
4180         desc = (struct scsi_vpd_id_descriptor *)lun->lun_devid->data;
4181         desc->proto_codeset = SVPD_ID_CODESET_ASCII;
4182         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_T10;
4183         desc->length = idlen1;
4184         t10id = (struct scsi_vpd_id_t10 *)&desc->identifier[0];
4185         memset(t10id->vendor, ' ', sizeof(t10id->vendor));
4186         if ((vendor = ctl_get_opt(&be_lun->options, "vendor")) == NULL) {
4187                 strncpy((char *)t10id->vendor, CTL_VENDOR, sizeof(t10id->vendor));
4188         } else {
4189                 strncpy(t10id->vendor, vendor,
4190                     min(sizeof(t10id->vendor), strlen(vendor)));
4191         }
4192         strncpy((char *)t10id->vendor_spec_id,
4193             (char *)be_lun->device_id, devidlen);
4194         if (scsiname != NULL) {
4195                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4196                     desc->length);
4197                 desc->proto_codeset = SVPD_ID_CODESET_UTF8;
4198                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4199                     SVPD_ID_TYPE_SCSI_NAME;
4200                 desc->length = idlen2;
4201                 strlcpy(desc->identifier, scsiname, idlen2);
4202         }
4203         if (eui != NULL) {
4204                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4205                     desc->length);
4206                 desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4207                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4208                     SVPD_ID_TYPE_EUI64;
4209                 desc->length = hex2bin(eui, desc->identifier, 16);
4210                 desc->length = desc->length > 12 ? 16 :
4211                     (desc->length > 8 ? 12 : 8);
4212                 len -= 16 - desc->length;
4213         }
4214         if (naa != NULL) {
4215                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4216                     desc->length);
4217                 desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4218                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4219                     SVPD_ID_TYPE_NAA;
4220                 desc->length = hex2bin(naa, desc->identifier, 16);
4221                 desc->length = desc->length > 8 ? 16 : 8;
4222                 len -= 16 - desc->length;
4223         }
4224         lun->lun_devid->len = len;
4225
4226         mtx_lock(&ctl_softc->ctl_lock);
4227         /*
4228          * See if the caller requested a particular LUN number.  If so, see
4229          * if it is available.  Otherwise, allocate the first available LUN.
4230          */
4231         if (be_lun->flags & CTL_LUN_FLAG_ID_REQ) {
4232                 if ((be_lun->req_lun_id > (CTL_MAX_LUNS - 1))
4233                  || (ctl_is_set(ctl_softc->ctl_lun_mask, be_lun->req_lun_id))) {
4234                         mtx_unlock(&ctl_softc->ctl_lock);
4235                         if (be_lun->req_lun_id > (CTL_MAX_LUNS - 1)) {
4236                                 printf("ctl: requested LUN ID %d is higher "
4237                                        "than CTL_MAX_LUNS - 1 (%d)\n",
4238                                        be_lun->req_lun_id, CTL_MAX_LUNS - 1);
4239                         } else {
4240                                 /*
4241                                  * XXX KDM return an error, or just assign
4242                                  * another LUN ID in this case??
4243                                  */
4244                                 printf("ctl: requested LUN ID %d is already "
4245                                        "in use\n", be_lun->req_lun_id);
4246                         }
4247                         if (lun->flags & CTL_LUN_MALLOCED)
4248                                 free(lun, M_CTL);
4249                         be_lun->lun_config_status(be_lun->be_lun,
4250                                                   CTL_LUN_CONFIG_FAILURE);
4251                         return (ENOSPC);
4252                 }
4253                 lun_number = be_lun->req_lun_id;
4254         } else {
4255                 lun_number = ctl_ffz(ctl_softc->ctl_lun_mask, 0, CTL_MAX_LUNS);
4256                 if (lun_number == -1) {
4257                         mtx_unlock(&ctl_softc->ctl_lock);
4258                         printf("ctl: can't allocate LUN, out of LUNs\n");
4259                         if (lun->flags & CTL_LUN_MALLOCED)
4260                                 free(lun, M_CTL);
4261                         be_lun->lun_config_status(be_lun->be_lun,
4262                                                   CTL_LUN_CONFIG_FAILURE);
4263                         return (ENOSPC);
4264                 }
4265         }
4266         ctl_set_mask(ctl_softc->ctl_lun_mask, lun_number);
4267
4268         mtx_init(&lun->lun_lock, "CTL LUN", NULL, MTX_DEF);
4269         lun->lun = lun_number;
4270         lun->be_lun = be_lun;
4271         /*
4272          * The processor LUN is always enabled.  Disk LUNs come on line
4273          * disabled, and must be enabled by the backend.
4274          */
4275         lun->flags |= CTL_LUN_DISABLED;
4276         lun->backend = be_lun->be;
4277         be_lun->ctl_lun = lun;
4278         be_lun->lun_id = lun_number;
4279         atomic_add_int(&be_lun->be->num_luns, 1);
4280         if (be_lun->flags & CTL_LUN_FLAG_OFFLINE)
4281                 lun->flags |= CTL_LUN_OFFLINE;
4282
4283         if (be_lun->flags & CTL_LUN_FLAG_POWERED_OFF)
4284                 lun->flags |= CTL_LUN_STOPPED;
4285
4286         if (be_lun->flags & CTL_LUN_FLAG_INOPERABLE)
4287                 lun->flags |= CTL_LUN_INOPERABLE;
4288
4289         if (be_lun->flags & CTL_LUN_FLAG_PRIMARY)
4290                 lun->flags |= CTL_LUN_PRIMARY_SC;
4291
4292         lun->ctl_softc = ctl_softc;
4293 #ifdef CTL_TIME_IO
4294         lun->last_busy = getsbinuptime();
4295 #endif
4296         TAILQ_INIT(&lun->ooa_queue);
4297         TAILQ_INIT(&lun->blocked_queue);
4298         STAILQ_INIT(&lun->error_list);
4299         ctl_tpc_lun_init(lun);
4300
4301         /*
4302          * Initialize the mode and log page index.
4303          */
4304         ctl_init_page_index(lun);
4305         ctl_init_log_page_index(lun);
4306
4307         /*
4308          * Now, before we insert this lun on the lun list, set the lun
4309          * inventory changed UA for all other luns.
4310          */
4311         STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) {
4312                 mtx_lock(&nlun->lun_lock);
4313                 ctl_est_ua_all(nlun, -1, CTL_UA_LUN_CHANGE);
4314                 mtx_unlock(&nlun->lun_lock);
4315         }
4316
4317         STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links);
4318
4319         ctl_softc->ctl_luns[lun_number] = lun;
4320
4321         ctl_softc->num_luns++;
4322
4323         /* Setup statistics gathering */
4324         lun->stats.device_type = be_lun->lun_type;
4325         lun->stats.lun_number = lun_number;
4326         if (lun->stats.device_type == T_DIRECT)
4327                 lun->stats.blocksize = be_lun->blocksize;
4328         else
4329                 lun->stats.flags = CTL_LUN_STATS_NO_BLOCKSIZE;
4330         for (i = 0;i < CTL_MAX_PORTS;i++)
4331                 lun->stats.ports[i].targ_port = i;
4332
4333         mtx_unlock(&ctl_softc->ctl_lock);
4334
4335         lun->be_lun->lun_config_status(lun->be_lun->be_lun, CTL_LUN_CONFIG_OK);
4336         return (0);
4337 }
4338
4339 /*
4340  * Delete a LUN.
4341  * Assumptions:
4342  * - LUN has already been marked invalid and any pending I/O has been taken
4343  *   care of.
4344  */
4345 static int
4346 ctl_free_lun(struct ctl_lun *lun)
4347 {
4348         struct ctl_softc *softc;
4349         struct ctl_lun *nlun;
4350         int i;
4351
4352         softc = lun->ctl_softc;
4353
4354         mtx_assert(&softc->ctl_lock, MA_OWNED);
4355
4356         STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links);
4357
4358         ctl_clear_mask(softc->ctl_lun_mask, lun->lun);
4359
4360         softc->ctl_luns[lun->lun] = NULL;
4361
4362         if (!TAILQ_EMPTY(&lun->ooa_queue))
4363                 panic("Freeing a LUN %p with outstanding I/O!!\n", lun);
4364
4365         softc->num_luns--;
4366
4367         /*
4368          * Tell the backend to free resources, if this LUN has a backend.
4369          */
4370         atomic_subtract_int(&lun->be_lun->be->num_luns, 1);
4371         lun->be_lun->lun_shutdown(lun->be_lun->be_lun);
4372
4373         ctl_tpc_lun_shutdown(lun);
4374         mtx_destroy(&lun->lun_lock);
4375         free(lun->lun_devid, M_CTL);
4376         for (i = 0; i < CTL_MAX_PORTS; i++)
4377                 free(lun->pending_ua[i], M_CTL);
4378         for (i = 0; i < CTL_MAX_PORTS; i++)
4379                 free(lun->pr_keys[i], M_CTL);
4380         free(lun->write_buffer, M_CTL);
4381         if (lun->flags & CTL_LUN_MALLOCED)
4382                 free(lun, M_CTL);
4383
4384         STAILQ_FOREACH(nlun, &softc->lun_list, links) {
4385                 mtx_lock(&nlun->lun_lock);
4386                 ctl_est_ua_all(nlun, -1, CTL_UA_LUN_CHANGE);
4387                 mtx_unlock(&nlun->lun_lock);
4388         }
4389
4390         return (0);
4391 }
4392
4393 static void
4394 ctl_create_lun(struct ctl_be_lun *be_lun)
4395 {
4396         struct ctl_softc *softc;
4397
4398         softc = control_softc;
4399
4400         /*
4401          * ctl_alloc_lun() should handle all potential failure cases.
4402          */
4403         ctl_alloc_lun(softc, NULL, be_lun);
4404 }
4405
4406 int
4407 ctl_add_lun(struct ctl_be_lun *be_lun)
4408 {
4409         struct ctl_softc *softc = control_softc;
4410
4411         mtx_lock(&softc->ctl_lock);
4412         STAILQ_INSERT_TAIL(&softc->pending_lun_queue, be_lun, links);
4413         mtx_unlock(&softc->ctl_lock);
4414         wakeup(&softc->pending_lun_queue);
4415
4416         return (0);
4417 }
4418
4419 int
4420 ctl_enable_lun(struct ctl_be_lun *be_lun)
4421 {
4422         struct ctl_softc *softc;
4423         struct ctl_port *port, *nport;
4424         struct ctl_lun *lun;
4425         int retval;
4426
4427         lun = (struct ctl_lun *)be_lun->ctl_lun;
4428         softc = lun->ctl_softc;
4429
4430         mtx_lock(&softc->ctl_lock);
4431         mtx_lock(&lun->lun_lock);
4432         if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4433                 /*
4434                  * eh?  Why did we get called if the LUN is already
4435                  * enabled?
4436                  */
4437                 mtx_unlock(&lun->lun_lock);
4438                 mtx_unlock(&softc->ctl_lock);
4439                 return (0);
4440         }
4441         lun->flags &= ~CTL_LUN_DISABLED;
4442         mtx_unlock(&lun->lun_lock);
4443
4444         for (port = STAILQ_FIRST(&softc->port_list); port != NULL; port = nport) {
4445                 nport = STAILQ_NEXT(port, links);
4446                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0 ||
4447                     port->lun_map != NULL || port->lun_enable == NULL)
4448                         continue;
4449
4450                 /*
4451                  * Drop the lock while we call the FETD's enable routine.
4452                  * This can lead to a callback into CTL (at least in the
4453                  * case of the internal initiator frontend.
4454                  */
4455                 mtx_unlock(&softc->ctl_lock);
4456                 retval = port->lun_enable(port->targ_lun_arg, lun->lun);
4457                 mtx_lock(&softc->ctl_lock);
4458                 if (retval != 0) {
4459                         printf("%s: FETD %s port %d returned error "
4460                                "%d for lun_enable on lun %jd\n",
4461                                __func__, port->port_name, port->targ_port,
4462                                retval, (intmax_t)lun->lun);
4463                 }
4464         }
4465
4466         mtx_unlock(&softc->ctl_lock);
4467         ctl_isc_announce_lun(lun);
4468
4469         return (0);
4470 }
4471
4472 int
4473 ctl_disable_lun(struct ctl_be_lun *be_lun)
4474 {
4475         struct ctl_softc *softc;
4476         struct ctl_port *port;
4477         struct ctl_lun *lun;
4478         int retval;
4479
4480         lun = (struct ctl_lun *)be_lun->ctl_lun;
4481         softc = lun->ctl_softc;
4482
4483         mtx_lock(&softc->ctl_lock);
4484         mtx_lock(&lun->lun_lock);
4485         if (lun->flags & CTL_LUN_DISABLED) {
4486                 mtx_unlock(&lun->lun_lock);
4487                 mtx_unlock(&softc->ctl_lock);
4488                 return (0);
4489         }
4490         lun->flags |= CTL_LUN_DISABLED;
4491         mtx_unlock(&lun->lun_lock);
4492
4493         STAILQ_FOREACH(port, &softc->port_list, links) {
4494                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0 ||
4495                     port->lun_map != NULL || port->lun_disable == NULL)
4496                         continue;
4497
4498                 /*
4499                  * Drop the lock before we call the frontend's disable
4500                  * routine, to avoid lock order reversals.
4501                  *
4502                  * XXX KDM what happens if the frontend list changes while
4503                  * we're traversing it?  It's unlikely, but should be handled.
4504                  */
4505                 mtx_unlock(&softc->ctl_lock);
4506                 retval = port->lun_disable(port->targ_lun_arg, lun->lun);
4507                 mtx_lock(&softc->ctl_lock);
4508                 if (retval != 0) {
4509                         printf("%s: FETD %s port %d returned error "
4510                                "%d for lun_disable on lun %jd\n",
4511                                __func__, port->port_name, port->targ_port,
4512                                retval, (intmax_t)lun->lun);
4513                 }
4514         }
4515
4516         mtx_unlock(&softc->ctl_lock);
4517         ctl_isc_announce_lun(lun);
4518
4519         return (0);
4520 }
4521
4522 int
4523 ctl_start_lun(struct ctl_be_lun *be_lun)
4524 {
4525         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4526
4527         mtx_lock(&lun->lun_lock);
4528         lun->flags &= ~CTL_LUN_STOPPED;
4529         mtx_unlock(&lun->lun_lock);
4530         return (0);
4531 }
4532
4533 int
4534 ctl_stop_lun(struct ctl_be_lun *be_lun)
4535 {
4536         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4537
4538         mtx_lock(&lun->lun_lock);
4539         lun->flags |= CTL_LUN_STOPPED;
4540         mtx_unlock(&lun->lun_lock);
4541         return (0);
4542 }
4543
4544 int
4545 ctl_lun_offline(struct ctl_be_lun *be_lun)
4546 {
4547         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4548
4549         mtx_lock(&lun->lun_lock);
4550         lun->flags |= CTL_LUN_OFFLINE;
4551         mtx_unlock(&lun->lun_lock);
4552         return (0);
4553 }
4554
4555 int
4556 ctl_lun_online(struct ctl_be_lun *be_lun)
4557 {
4558         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4559
4560         mtx_lock(&lun->lun_lock);
4561         lun->flags &= ~CTL_LUN_OFFLINE;
4562         mtx_unlock(&lun->lun_lock);
4563         return (0);
4564 }
4565
4566 int
4567 ctl_lun_primary(struct ctl_be_lun *be_lun)
4568 {
4569         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4570
4571         mtx_lock(&lun->lun_lock);
4572         lun->flags |= CTL_LUN_PRIMARY_SC;
4573         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
4574         mtx_unlock(&lun->lun_lock);
4575         ctl_isc_announce_lun(lun);
4576         return (0);
4577 }
4578
4579 int
4580 ctl_lun_secondary(struct ctl_be_lun *be_lun)
4581 {
4582         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4583
4584         mtx_lock(&lun->lun_lock);
4585         lun->flags &= ~CTL_LUN_PRIMARY_SC;
4586         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
4587         mtx_unlock(&lun->lun_lock);
4588         ctl_isc_announce_lun(lun);
4589         return (0);
4590 }
4591
4592 int
4593 ctl_invalidate_lun(struct ctl_be_lun *be_lun)
4594 {
4595         struct ctl_softc *softc;
4596         struct ctl_lun *lun;
4597
4598         lun = (struct ctl_lun *)be_lun->ctl_lun;
4599         softc = lun->ctl_softc;
4600
4601         mtx_lock(&lun->lun_lock);
4602
4603         /*
4604          * The LUN needs to be disabled before it can be marked invalid.
4605          */
4606         if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4607                 mtx_unlock(&lun->lun_lock);
4608                 return (-1);
4609         }
4610         /*
4611          * Mark the LUN invalid.
4612          */
4613         lun->flags |= CTL_LUN_INVALID;
4614
4615         /*
4616          * If there is nothing in the OOA queue, go ahead and free the LUN.
4617          * If we have something in the OOA queue, we'll free it when the
4618          * last I/O completes.
4619          */
4620         if (TAILQ_EMPTY(&lun->ooa_queue)) {
4621                 mtx_unlock(&lun->lun_lock);
4622                 mtx_lock(&softc->ctl_lock);
4623                 ctl_free_lun(lun);
4624                 mtx_unlock(&softc->ctl_lock);
4625         } else
4626                 mtx_unlock(&lun->lun_lock);
4627
4628         return (0);
4629 }
4630
4631 int
4632 ctl_lun_inoperable(struct ctl_be_lun *be_lun)
4633 {
4634         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4635
4636         mtx_lock(&lun->lun_lock);
4637         lun->flags |= CTL_LUN_INOPERABLE;
4638         mtx_unlock(&lun->lun_lock);
4639         return (0);
4640 }
4641
4642 int
4643 ctl_lun_operable(struct ctl_be_lun *be_lun)
4644 {
4645         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4646
4647         mtx_lock(&lun->lun_lock);
4648         lun->flags &= ~CTL_LUN_INOPERABLE;
4649         mtx_unlock(&lun->lun_lock);
4650         return (0);
4651 }
4652
4653 void
4654 ctl_lun_capacity_changed(struct ctl_be_lun *be_lun)
4655 {
4656         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4657         union ctl_ha_msg msg;
4658
4659         mtx_lock(&lun->lun_lock);
4660         ctl_est_ua_all(lun, -1, CTL_UA_CAPACITY_CHANGED);
4661         mtx_unlock(&lun->lun_lock);
4662         if (lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
4663                 /* Send msg to other side. */
4664                 bzero(&msg.ua, sizeof(msg.ua));
4665                 msg.hdr.msg_type = CTL_MSG_UA;
4666                 msg.hdr.nexus.initid = -1;
4667                 msg.hdr.nexus.targ_port = -1;
4668                 msg.hdr.nexus.targ_lun = lun->lun;
4669                 msg.hdr.nexus.targ_mapped_lun = lun->lun;
4670                 msg.ua.ua_all = 1;
4671                 msg.ua.ua_set = 1;
4672                 msg.ua.ua_type = CTL_UA_CAPACITY_CHANGED;
4673                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg.ua),
4674                     M_WAITOK);
4675         }
4676 }
4677
4678 /*
4679  * Backend "memory move is complete" callback for requests that never
4680  * make it down to say RAIDCore's configuration code.
4681  */
4682 int
4683 ctl_config_move_done(union ctl_io *io)
4684 {
4685         int retval;
4686
4687         CTL_DEBUG_PRINT(("ctl_config_move_done\n"));
4688         KASSERT(io->io_hdr.io_type == CTL_IO_SCSI,
4689             ("Config I/O type isn't CTL_IO_SCSI (%d)!", io->io_hdr.io_type));
4690
4691         if ((io->io_hdr.port_status != 0) &&
4692             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
4693              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
4694                 /*
4695                  * For hardware error sense keys, the sense key
4696                  * specific value is defined to be a retry count,
4697                  * but we use it to pass back an internal FETD
4698                  * error code.  XXX KDM  Hopefully the FETD is only
4699                  * using 16 bits for an error code, since that's
4700                  * all the space we have in the sks field.
4701                  */
4702                 ctl_set_internal_failure(&io->scsiio,
4703                                          /*sks_valid*/ 1,
4704                                          /*retry_count*/
4705                                          io->io_hdr.port_status);
4706         }
4707
4708         if (ctl_debug & CTL_DEBUG_CDB_DATA)
4709                 ctl_data_print(io);
4710         if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN) ||
4711             ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
4712              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS) ||
4713             ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) {
4714                 /*
4715                  * XXX KDM just assuming a single pointer here, and not a
4716                  * S/G list.  If we start using S/G lists for config data,
4717                  * we'll need to know how to clean them up here as well.
4718                  */
4719                 if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
4720                         free(io->scsiio.kern_data_ptr, M_CTL);
4721                 ctl_done(io);
4722                 retval = CTL_RETVAL_COMPLETE;
4723         } else {
4724                 /*
4725                  * XXX KDM now we need to continue data movement.  Some
4726                  * options:
4727                  * - call ctl_scsiio() again?  We don't do this for data
4728                  *   writes, because for those at least we know ahead of
4729                  *   time where the write will go and how long it is.  For
4730                  *   config writes, though, that information is largely
4731                  *   contained within the write itself, thus we need to
4732                  *   parse out the data again.
4733                  *
4734                  * - Call some other function once the data is in?
4735                  */
4736
4737                 /*
4738                  * XXX KDM call ctl_scsiio() again for now, and check flag
4739                  * bits to see whether we're allocated or not.
4740                  */
4741                 retval = ctl_scsiio(&io->scsiio);
4742         }
4743         return (retval);
4744 }
4745
4746 /*
4747  * This gets called by a backend driver when it is done with a
4748  * data_submit method.
4749  */
4750 void
4751 ctl_data_submit_done(union ctl_io *io)
4752 {
4753         /*
4754          * If the IO_CONT flag is set, we need to call the supplied
4755          * function to continue processing the I/O, instead of completing
4756          * the I/O just yet.
4757          *
4758          * If there is an error, though, we don't want to keep processing.
4759          * Instead, just send status back to the initiator.
4760          */
4761         if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
4762             (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
4763             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
4764              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
4765                 io->scsiio.io_cont(io);
4766                 return;
4767         }
4768         ctl_done(io);
4769 }
4770
4771 /*
4772  * This gets called by a backend driver when it is done with a
4773  * configuration write.
4774  */
4775 void
4776 ctl_config_write_done(union ctl_io *io)
4777 {
4778         uint8_t *buf;
4779
4780         /*
4781          * If the IO_CONT flag is set, we need to call the supplied
4782          * function to continue processing the I/O, instead of completing
4783          * the I/O just yet.
4784          *
4785          * If there is an error, though, we don't want to keep processing.
4786          * Instead, just send status back to the initiator.
4787          */
4788         if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
4789             (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
4790             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
4791              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
4792                 io->scsiio.io_cont(io);
4793                 return;
4794         }
4795         /*
4796          * Since a configuration write can be done for commands that actually
4797          * have data allocated, like write buffer, and commands that have
4798          * no data, like start/stop unit, we need to check here.
4799          */
4800         if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
4801                 buf = io->scsiio.kern_data_ptr;
4802         else
4803                 buf = NULL;
4804         ctl_done(io);
4805         if (buf)
4806                 free(buf, M_CTL);
4807 }
4808
4809 void
4810 ctl_config_read_done(union ctl_io *io)
4811 {
4812         uint8_t *buf;
4813
4814         /*
4815          * If there is some error -- we are done, skip data transfer.
4816          */
4817         if ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0 ||
4818             ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
4819              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) {
4820                 if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
4821                         buf = io->scsiio.kern_data_ptr;
4822                 else
4823                         buf = NULL;
4824                 ctl_done(io);
4825                 if (buf)
4826                         free(buf, M_CTL);
4827                 return;
4828         }
4829
4830         /*
4831          * If the IO_CONT flag is set, we need to call the supplied
4832          * function to continue processing the I/O, instead of completing
4833          * the I/O just yet.
4834          */
4835         if (io->io_hdr.flags & CTL_FLAG_IO_CONT) {
4836                 io->scsiio.io_cont(io);
4837                 return;
4838         }
4839
4840         ctl_datamove(io);
4841 }
4842
4843 /*
4844  * SCSI release command.
4845  */
4846 int
4847 ctl_scsi_release(struct ctl_scsiio *ctsio)
4848 {
4849         int length, longid, thirdparty_id, resv_id;
4850         struct ctl_lun *lun;
4851         uint32_t residx;
4852
4853         length = 0;
4854         resv_id = 0;
4855
4856         CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
4857
4858         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
4859         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
4860
4861         switch (ctsio->cdb[0]) {
4862         case RELEASE_10: {
4863                 struct scsi_release_10 *cdb;
4864
4865                 cdb = (struct scsi_release_10 *)ctsio->cdb;
4866
4867                 if (cdb->byte2 & SR10_LONGID)
4868                         longid = 1;
4869                 else
4870                         thirdparty_id = cdb->thirdparty_id;
4871
4872                 resv_id = cdb->resv_id;
4873                 length = scsi_2btoul(cdb->length);
4874                 break;
4875         }
4876         }
4877
4878
4879         /*
4880          * XXX KDM right now, we only support LUN reservation.  We don't
4881          * support 3rd party reservations, or extent reservations, which
4882          * might actually need the parameter list.  If we've gotten this
4883          * far, we've got a LUN reservation.  Anything else got kicked out
4884          * above.  So, according to SPC, ignore the length.
4885          */
4886         length = 0;
4887
4888         if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
4889          && (length > 0)) {
4890                 ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
4891                 ctsio->kern_data_len = length;
4892                 ctsio->kern_total_len = length;
4893                 ctsio->kern_data_resid = 0;
4894                 ctsio->kern_rel_offset = 0;
4895                 ctsio->kern_sg_entries = 0;
4896                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
4897                 ctsio->be_move_done = ctl_config_move_done;
4898                 ctl_datamove((union ctl_io *)ctsio);
4899
4900                 return (CTL_RETVAL_COMPLETE);
4901         }
4902
4903         if (length > 0)
4904                 thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
4905
4906         mtx_lock(&lun->lun_lock);
4907
4908         /*
4909          * According to SPC, it is not an error for an intiator to attempt
4910          * to release a reservation on a LUN that isn't reserved, or that
4911          * is reserved by another initiator.  The reservation can only be
4912          * released, though, by the initiator who made it or by one of
4913          * several reset type events.
4914          */
4915         if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == residx))
4916                         lun->flags &= ~CTL_LUN_RESERVED;
4917
4918         mtx_unlock(&lun->lun_lock);
4919
4920         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
4921                 free(ctsio->kern_data_ptr, M_CTL);
4922                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
4923         }
4924
4925         ctl_set_success(ctsio);
4926         ctl_done((union ctl_io *)ctsio);
4927         return (CTL_RETVAL_COMPLETE);
4928 }
4929
4930 int
4931 ctl_scsi_reserve(struct ctl_scsiio *ctsio)
4932 {
4933         int extent, thirdparty, longid;
4934         int resv_id, length;
4935         uint64_t thirdparty_id;
4936         struct ctl_lun *lun;
4937         uint32_t residx;
4938
4939         extent = 0;
4940         thirdparty = 0;
4941         longid = 0;
4942         resv_id = 0;
4943         length = 0;
4944         thirdparty_id = 0;
4945
4946         CTL_DEBUG_PRINT(("ctl_reserve\n"));
4947
4948         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
4949         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
4950
4951         switch (ctsio->cdb[0]) {
4952         case RESERVE_10: {
4953                 struct scsi_reserve_10 *cdb;
4954
4955                 cdb = (struct scsi_reserve_10 *)ctsio->cdb;
4956
4957                 if (cdb->byte2 & SR10_LONGID)
4958                         longid = 1;
4959                 else
4960                         thirdparty_id = cdb->thirdparty_id;
4961
4962                 resv_id = cdb->resv_id;
4963                 length = scsi_2btoul(cdb->length);
4964                 break;
4965         }
4966         }
4967
4968         /*
4969          * XXX KDM right now, we only support LUN reservation.  We don't
4970          * support 3rd party reservations, or extent reservations, which
4971          * might actually need the parameter list.  If we've gotten this
4972          * far, we've got a LUN reservation.  Anything else got kicked out
4973          * above.  So, according to SPC, ignore the length.
4974          */
4975         length = 0;
4976
4977         if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
4978          && (length > 0)) {
4979                 ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
4980                 ctsio->kern_data_len = length;
4981                 ctsio->kern_total_len = length;
4982                 ctsio->kern_data_resid = 0;
4983                 ctsio->kern_rel_offset = 0;
4984                 ctsio->kern_sg_entries = 0;
4985                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
4986                 ctsio->be_move_done = ctl_config_move_done;
4987                 ctl_datamove((union ctl_io *)ctsio);
4988
4989                 return (CTL_RETVAL_COMPLETE);
4990         }
4991
4992         if (length > 0)
4993                 thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr);
4994
4995         mtx_lock(&lun->lun_lock);
4996         if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx != residx)) {
4997                 ctl_set_reservation_conflict(ctsio);
4998                 goto bailout;
4999         }
5000
5001         lun->flags |= CTL_LUN_RESERVED;
5002         lun->res_idx = residx;
5003
5004         ctl_set_success(ctsio);
5005
5006 bailout:
5007         mtx_unlock(&lun->lun_lock);
5008
5009         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5010                 free(ctsio->kern_data_ptr, M_CTL);
5011                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5012         }
5013
5014         ctl_done((union ctl_io *)ctsio);
5015         return (CTL_RETVAL_COMPLETE);
5016 }
5017
5018 int
5019 ctl_start_stop(struct ctl_scsiio *ctsio)
5020 {
5021         struct scsi_start_stop_unit *cdb;
5022         struct ctl_lun *lun;
5023         int retval;
5024
5025         CTL_DEBUG_PRINT(("ctl_start_stop\n"));
5026
5027         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5028         retval = 0;
5029
5030         cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
5031
5032         /*
5033          * XXX KDM
5034          * We don't support the immediate bit on a stop unit.  In order to
5035          * do that, we would need to code up a way to know that a stop is
5036          * pending, and hold off any new commands until it completes, one
5037          * way or another.  Then we could accept or reject those commands
5038          * depending on its status.  We would almost need to do the reverse
5039          * of what we do below for an immediate start -- return the copy of
5040          * the ctl_io to the FETD with status to send to the host (and to
5041          * free the copy!) and then free the original I/O once the stop
5042          * actually completes.  That way, the OOA queue mechanism can work
5043          * to block commands that shouldn't proceed.  Another alternative
5044          * would be to put the copy in the queue in place of the original,
5045          * and return the original back to the caller.  That could be
5046          * slightly safer..
5047          */
5048         if ((cdb->byte2 & SSS_IMMED)
5049          && ((cdb->how & SSS_START) == 0)) {
5050                 ctl_set_invalid_field(ctsio,
5051                                       /*sks_valid*/ 1,
5052                                       /*command*/ 1,
5053                                       /*field*/ 1,
5054                                       /*bit_valid*/ 1,
5055                                       /*bit*/ 0);
5056                 ctl_done((union ctl_io *)ctsio);
5057                 return (CTL_RETVAL_COMPLETE);
5058         }
5059
5060         if ((lun->flags & CTL_LUN_PR_RESERVED)
5061          && ((cdb->how & SSS_START)==0)) {
5062                 uint32_t residx;
5063
5064                 residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5065                 if (ctl_get_prkey(lun, residx) == 0
5066                  || (lun->pr_res_idx!=residx && lun->res_type < 4)) {
5067
5068                         ctl_set_reservation_conflict(ctsio);
5069                         ctl_done((union ctl_io *)ctsio);
5070                         return (CTL_RETVAL_COMPLETE);
5071                 }
5072         }
5073
5074         /*
5075          * If there is no backend on this device, we can't start or stop
5076          * it.  In theory we shouldn't get any start/stop commands in the
5077          * first place at this level if the LUN doesn't have a backend.
5078          * That should get stopped by the command decode code.
5079          */
5080         if (lun->backend == NULL) {
5081                 ctl_set_invalid_opcode(ctsio);
5082                 ctl_done((union ctl_io *)ctsio);
5083                 return (CTL_RETVAL_COMPLETE);
5084         }
5085
5086         /*
5087          * XXX KDM Copan-specific offline behavior.
5088          * Figure out a reasonable way to port this?
5089          */
5090 #ifdef NEEDTOPORT
5091         mtx_lock(&lun->lun_lock);
5092
5093         if (((cdb->byte2 & SSS_ONOFFLINE) == 0)
5094          && (lun->flags & CTL_LUN_OFFLINE)) {
5095                 /*
5096                  * If the LUN is offline, and the on/offline bit isn't set,
5097                  * reject the start or stop.  Otherwise, let it through.
5098                  */
5099                 mtx_unlock(&lun->lun_lock);
5100                 ctl_set_lun_not_ready(ctsio);
5101                 ctl_done((union ctl_io *)ctsio);
5102         } else {
5103                 mtx_unlock(&lun->lun_lock);
5104 #endif /* NEEDTOPORT */
5105                 /*
5106                  * This could be a start or a stop when we're online,
5107                  * or a stop/offline or start/online.  A start or stop when
5108                  * we're offline is covered in the case above.
5109                  */
5110                 /*
5111                  * In the non-immediate case, we send the request to
5112                  * the backend and return status to the user when
5113                  * it is done.
5114                  *
5115                  * In the immediate case, we allocate a new ctl_io
5116                  * to hold a copy of the request, and send that to
5117                  * the backend.  We then set good status on the
5118                  * user's request and return it immediately.
5119                  */
5120                 if (cdb->byte2 & SSS_IMMED) {
5121                         union ctl_io *new_io;
5122
5123                         new_io = ctl_alloc_io(ctsio->io_hdr.pool);
5124                         ctl_copy_io((union ctl_io *)ctsio, new_io);
5125                         retval = lun->backend->config_write(new_io);
5126                         ctl_set_success(ctsio);
5127                         ctl_done((union ctl_io *)ctsio);
5128                 } else {
5129                         retval = lun->backend->config_write(
5130                                 (union ctl_io *)ctsio);
5131                 }
5132 #ifdef NEEDTOPORT
5133         }
5134 #endif
5135         return (retval);
5136 }
5137
5138 /*
5139  * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
5140  * we don't really do anything with the LBA and length fields if the user
5141  * passes them in.  Instead we'll just flush out the cache for the entire
5142  * LUN.
5143  */
5144 int
5145 ctl_sync_cache(struct ctl_scsiio *ctsio)
5146 {
5147         struct ctl_lun *lun;
5148         struct ctl_softc *softc;
5149         struct ctl_lba_len_flags *lbalen;
5150         uint64_t starting_lba;
5151         uint32_t block_count;
5152         int retval;
5153         uint8_t byte2;
5154
5155         CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
5156
5157         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5158         softc = lun->ctl_softc;
5159         retval = 0;
5160
5161         switch (ctsio->cdb[0]) {
5162         case SYNCHRONIZE_CACHE: {
5163                 struct scsi_sync_cache *cdb;
5164                 cdb = (struct scsi_sync_cache *)ctsio->cdb;
5165
5166                 starting_lba = scsi_4btoul(cdb->begin_lba);
5167                 block_count = scsi_2btoul(cdb->lb_count);
5168                 byte2 = cdb->byte2;
5169                 break;
5170         }
5171         case SYNCHRONIZE_CACHE_16: {
5172                 struct scsi_sync_cache_16 *cdb;
5173                 cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5174
5175                 starting_lba = scsi_8btou64(cdb->begin_lba);
5176                 block_count = scsi_4btoul(cdb->lb_count);
5177                 byte2 = cdb->byte2;
5178                 break;
5179         }
5180         default:
5181                 ctl_set_invalid_opcode(ctsio);
5182                 ctl_done((union ctl_io *)ctsio);
5183                 goto bailout;
5184                 break; /* NOTREACHED */
5185         }
5186
5187         /*
5188          * We check the LBA and length, but don't do anything with them.
5189          * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5190          * get flushed.  This check will just help satisfy anyone who wants
5191          * to see an error for an out of range LBA.
5192          */
5193         if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5194                 ctl_set_lba_out_of_range(ctsio);
5195                 ctl_done((union ctl_io *)ctsio);
5196                 goto bailout;
5197         }
5198
5199         /*
5200          * If this LUN has no backend, we can't flush the cache anyway.
5201          */
5202         if (lun->backend == NULL) {
5203                 ctl_set_invalid_opcode(ctsio);
5204                 ctl_done((union ctl_io *)ctsio);
5205                 goto bailout;
5206         }
5207
5208         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5209         lbalen->lba = starting_lba;
5210         lbalen->len = block_count;
5211         lbalen->flags = byte2;
5212
5213         /*
5214          * Check to see whether we're configured to send the SYNCHRONIZE
5215          * CACHE command directly to the back end.
5216          */
5217         mtx_lock(&lun->lun_lock);
5218         if ((softc->flags & CTL_FLAG_REAL_SYNC)
5219          && (++(lun->sync_count) >= lun->sync_interval)) {
5220                 lun->sync_count = 0;
5221                 mtx_unlock(&lun->lun_lock);
5222                 retval = lun->backend->config_write((union ctl_io *)ctsio);
5223         } else {
5224                 mtx_unlock(&lun->lun_lock);
5225                 ctl_set_success(ctsio);
5226                 ctl_done((union ctl_io *)ctsio);
5227         }
5228
5229 bailout:
5230
5231         return (retval);
5232 }
5233
5234 int
5235 ctl_format(struct ctl_scsiio *ctsio)
5236 {
5237         struct scsi_format *cdb;
5238         struct ctl_lun *lun;
5239         int length, defect_list_len;
5240
5241         CTL_DEBUG_PRINT(("ctl_format\n"));
5242
5243         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5244
5245         cdb = (struct scsi_format *)ctsio->cdb;
5246
5247         length = 0;
5248         if (cdb->byte2 & SF_FMTDATA) {
5249                 if (cdb->byte2 & SF_LONGLIST)
5250                         length = sizeof(struct scsi_format_header_long);
5251                 else
5252                         length = sizeof(struct scsi_format_header_short);
5253         }
5254
5255         if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5256          && (length > 0)) {
5257                 ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5258                 ctsio->kern_data_len = length;
5259                 ctsio->kern_total_len = length;
5260                 ctsio->kern_data_resid = 0;
5261                 ctsio->kern_rel_offset = 0;
5262                 ctsio->kern_sg_entries = 0;
5263                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5264                 ctsio->be_move_done = ctl_config_move_done;
5265                 ctl_datamove((union ctl_io *)ctsio);
5266
5267                 return (CTL_RETVAL_COMPLETE);
5268         }
5269
5270         defect_list_len = 0;
5271
5272         if (cdb->byte2 & SF_FMTDATA) {
5273                 if (cdb->byte2 & SF_LONGLIST) {
5274                         struct scsi_format_header_long *header;
5275
5276                         header = (struct scsi_format_header_long *)
5277                                 ctsio->kern_data_ptr;
5278
5279                         defect_list_len = scsi_4btoul(header->defect_list_len);
5280                         if (defect_list_len != 0) {
5281                                 ctl_set_invalid_field(ctsio,
5282                                                       /*sks_valid*/ 1,
5283                                                       /*command*/ 0,
5284                                                       /*field*/ 2,
5285                                                       /*bit_valid*/ 0,
5286                                                       /*bit*/ 0);
5287                                 goto bailout;
5288                         }
5289                 } else {
5290                         struct scsi_format_header_short *header;
5291
5292                         header = (struct scsi_format_header_short *)
5293                                 ctsio->kern_data_ptr;
5294
5295                         defect_list_len = scsi_2btoul(header->defect_list_len);
5296                         if (defect_list_len != 0) {
5297                                 ctl_set_invalid_field(ctsio,
5298                                                       /*sks_valid*/ 1,
5299                                                       /*command*/ 0,
5300                                                       /*field*/ 2,
5301                                                       /*bit_valid*/ 0,
5302                                                       /*bit*/ 0);
5303                                 goto bailout;
5304                         }
5305                 }
5306         }
5307
5308         /*
5309          * The format command will clear out the "Medium format corrupted"
5310          * status if set by the configuration code.  That status is really
5311          * just a way to notify the host that we have lost the media, and
5312          * get them to issue a command that will basically make them think
5313          * they're blowing away the media.
5314          */
5315         mtx_lock(&lun->lun_lock);
5316         lun->flags &= ~CTL_LUN_INOPERABLE;
5317         mtx_unlock(&lun->lun_lock);
5318
5319         ctl_set_success(ctsio);
5320 bailout:
5321
5322         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5323                 free(ctsio->kern_data_ptr, M_CTL);
5324                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5325         }
5326
5327         ctl_done((union ctl_io *)ctsio);
5328         return (CTL_RETVAL_COMPLETE);
5329 }
5330
5331 int
5332 ctl_read_buffer(struct ctl_scsiio *ctsio)
5333 {
5334         struct scsi_read_buffer *cdb;
5335         struct ctl_lun *lun;
5336         int buffer_offset, len;
5337         static uint8_t descr[4];
5338         static uint8_t echo_descr[4] = { 0 };
5339
5340         CTL_DEBUG_PRINT(("ctl_read_buffer\n"));
5341
5342         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5343         cdb = (struct scsi_read_buffer *)ctsio->cdb;
5344
5345         if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA &&
5346             (cdb->byte2 & RWB_MODE) != RWB_MODE_ECHO_DESCR &&
5347             (cdb->byte2 & RWB_MODE) != RWB_MODE_DESCR) {
5348                 ctl_set_invalid_field(ctsio,
5349                                       /*sks_valid*/ 1,
5350                                       /*command*/ 1,
5351                                       /*field*/ 1,
5352                                       /*bit_valid*/ 1,
5353                                       /*bit*/ 4);
5354                 ctl_done((union ctl_io *)ctsio);
5355                 return (CTL_RETVAL_COMPLETE);
5356         }
5357
5358         len = scsi_3btoul(cdb->length);
5359         buffer_offset = scsi_3btoul(cdb->offset);
5360
5361         if (buffer_offset + len > CTL_WRITE_BUFFER_SIZE) {
5362                 ctl_set_invalid_field(ctsio,
5363                                       /*sks_valid*/ 1,
5364                                       /*command*/ 1,
5365                                       /*field*/ 6,
5366                                       /*bit_valid*/ 0,
5367                                       /*bit*/ 0);
5368                 ctl_done((union ctl_io *)ctsio);
5369                 return (CTL_RETVAL_COMPLETE);
5370         }
5371
5372         if ((cdb->byte2 & RWB_MODE) == RWB_MODE_DESCR) {
5373                 descr[0] = 0;
5374                 scsi_ulto3b(CTL_WRITE_BUFFER_SIZE, &descr[1]);
5375                 ctsio->kern_data_ptr = descr;
5376                 len = min(len, sizeof(descr));
5377         } else if ((cdb->byte2 & RWB_MODE) == RWB_MODE_ECHO_DESCR) {
5378                 ctsio->kern_data_ptr = echo_descr;
5379                 len = min(len, sizeof(echo_descr));
5380         } else {
5381                 if (lun->write_buffer == NULL) {
5382                         lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE,
5383                             M_CTL, M_WAITOK);
5384                 }
5385                 ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5386         }
5387         ctsio->kern_data_len = len;
5388         ctsio->kern_total_len = len;
5389         ctsio->kern_data_resid = 0;
5390         ctsio->kern_rel_offset = 0;
5391         ctsio->kern_sg_entries = 0;
5392         ctl_set_success(ctsio);
5393         ctsio->be_move_done = ctl_config_move_done;
5394         ctl_datamove((union ctl_io *)ctsio);
5395         return (CTL_RETVAL_COMPLETE);
5396 }
5397
5398 int
5399 ctl_write_buffer(struct ctl_scsiio *ctsio)
5400 {
5401         struct scsi_write_buffer *cdb;
5402         struct ctl_lun *lun;
5403         int buffer_offset, len;
5404
5405         CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5406
5407         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5408         cdb = (struct scsi_write_buffer *)ctsio->cdb;
5409
5410         if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA) {
5411                 ctl_set_invalid_field(ctsio,
5412                                       /*sks_valid*/ 1,
5413                                       /*command*/ 1,
5414                                       /*field*/ 1,
5415                                       /*bit_valid*/ 1,
5416                                       /*bit*/ 4);
5417                 ctl_done((union ctl_io *)ctsio);
5418                 return (CTL_RETVAL_COMPLETE);
5419         }
5420
5421         len = scsi_3btoul(cdb->length);
5422         buffer_offset = scsi_3btoul(cdb->offset);
5423
5424         if (buffer_offset + len > CTL_WRITE_BUFFER_SIZE) {
5425                 ctl_set_invalid_field(ctsio,
5426                                       /*sks_valid*/ 1,
5427                                       /*command*/ 1,
5428                                       /*field*/ 6,
5429                                       /*bit_valid*/ 0,
5430                                       /*bit*/ 0);
5431                 ctl_done((union ctl_io *)ctsio);
5432                 return (CTL_RETVAL_COMPLETE);
5433         }
5434
5435         /*
5436          * If we've got a kernel request that hasn't been malloced yet,
5437          * malloc it and tell the caller the data buffer is here.
5438          */
5439         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5440                 if (lun->write_buffer == NULL) {
5441                         lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE,
5442                             M_CTL, M_WAITOK);
5443                 }
5444                 ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5445                 ctsio->kern_data_len = len;
5446                 ctsio->kern_total_len = len;
5447                 ctsio->kern_data_resid = 0;
5448                 ctsio->kern_rel_offset = 0;
5449                 ctsio->kern_sg_entries = 0;
5450                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5451                 ctsio->be_move_done = ctl_config_move_done;
5452                 ctl_datamove((union ctl_io *)ctsio);
5453
5454                 return (CTL_RETVAL_COMPLETE);
5455         }
5456
5457         ctl_set_success(ctsio);
5458         ctl_done((union ctl_io *)ctsio);
5459         return (CTL_RETVAL_COMPLETE);
5460 }
5461
5462 int
5463 ctl_write_same(struct ctl_scsiio *ctsio)
5464 {
5465         struct ctl_lun *lun;
5466         struct ctl_lba_len_flags *lbalen;
5467         uint64_t lba;
5468         uint32_t num_blocks;
5469         int len, retval;
5470         uint8_t byte2;
5471
5472         retval = CTL_RETVAL_COMPLETE;
5473
5474         CTL_DEBUG_PRINT(("ctl_write_same\n"));
5475
5476         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5477
5478         switch (ctsio->cdb[0]) {
5479         case WRITE_SAME_10: {
5480                 struct scsi_write_same_10 *cdb;
5481
5482                 cdb = (struct scsi_write_same_10 *)ctsio->cdb;
5483
5484                 lba = scsi_4btoul(cdb->addr);
5485                 num_blocks = scsi_2btoul(cdb->length);
5486                 byte2 = cdb->byte2;
5487                 break;
5488         }
5489         case WRITE_SAME_16: {
5490                 struct scsi_write_same_16 *cdb;
5491
5492                 cdb = (struct scsi_write_same_16 *)ctsio->cdb;
5493
5494                 lba = scsi_8btou64(cdb->addr);
5495                 num_blocks = scsi_4btoul(cdb->length);
5496                 byte2 = cdb->byte2;
5497                 break;
5498         }
5499         default:
5500                 /*
5501                  * We got a command we don't support.  This shouldn't
5502                  * happen, commands should be filtered out above us.
5503                  */
5504                 ctl_set_invalid_opcode(ctsio);
5505                 ctl_done((union ctl_io *)ctsio);
5506
5507                 return (CTL_RETVAL_COMPLETE);
5508                 break; /* NOTREACHED */
5509         }
5510
5511         /* NDOB and ANCHOR flags can be used only together with UNMAP */
5512         if ((byte2 & SWS_UNMAP) == 0 &&
5513             (byte2 & (SWS_NDOB | SWS_ANCHOR)) != 0) {
5514                 ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
5515                     /*command*/ 1, /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 0);
5516                 ctl_done((union ctl_io *)ctsio);
5517                 return (CTL_RETVAL_COMPLETE);
5518         }
5519
5520         /*
5521          * The first check is to make sure we're in bounds, the second
5522          * check is to catch wrap-around problems.  If the lba + num blocks
5523          * is less than the lba, then we've wrapped around and the block
5524          * range is invalid anyway.
5525          */
5526         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5527          || ((lba + num_blocks) < lba)) {
5528                 ctl_set_lba_out_of_range(ctsio);
5529                 ctl_done((union ctl_io *)ctsio);
5530                 return (CTL_RETVAL_COMPLETE);
5531         }
5532
5533         /* Zero number of blocks means "to the last logical block" */
5534         if (num_blocks == 0) {
5535                 if ((lun->be_lun->maxlba + 1) - lba > UINT32_MAX) {
5536                         ctl_set_invalid_field(ctsio,
5537                                               /*sks_valid*/ 0,
5538                                               /*command*/ 1,
5539                                               /*field*/ 0,
5540                                               /*bit_valid*/ 0,
5541                                               /*bit*/ 0);
5542                         ctl_done((union ctl_io *)ctsio);
5543                         return (CTL_RETVAL_COMPLETE);
5544                 }
5545                 num_blocks = (lun->be_lun->maxlba + 1) - lba;
5546         }
5547
5548         len = lun->be_lun->blocksize;
5549
5550         /*
5551          * If we've got a kernel request that hasn't been malloced yet,
5552          * malloc it and tell the caller the data buffer is here.
5553          */
5554         if ((byte2 & SWS_NDOB) == 0 &&
5555             (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5556                 ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
5557                 ctsio->kern_data_len = len;
5558                 ctsio->kern_total_len = len;
5559                 ctsio->kern_data_resid = 0;
5560                 ctsio->kern_rel_offset = 0;
5561                 ctsio->kern_sg_entries = 0;
5562                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5563                 ctsio->be_move_done = ctl_config_move_done;
5564                 ctl_datamove((union ctl_io *)ctsio);
5565
5566                 return (CTL_RETVAL_COMPLETE);
5567         }
5568
5569         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5570         lbalen->lba = lba;
5571         lbalen->len = num_blocks;
5572         lbalen->flags = byte2;
5573         retval = lun->backend->config_write((union ctl_io *)ctsio);
5574
5575         return (retval);
5576 }
5577
5578 int
5579 ctl_unmap(struct ctl_scsiio *ctsio)
5580 {
5581         struct ctl_lun *lun;
5582         struct scsi_unmap *cdb;
5583         struct ctl_ptr_len_flags *ptrlen;
5584         struct scsi_unmap_header *hdr;
5585         struct scsi_unmap_desc *buf, *end, *endnz, *range;
5586         uint64_t lba;
5587         uint32_t num_blocks;
5588         int len, retval;
5589         uint8_t byte2;
5590
5591         retval = CTL_RETVAL_COMPLETE;
5592
5593         CTL_DEBUG_PRINT(("ctl_unmap\n"));
5594
5595         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5596         cdb = (struct scsi_unmap *)ctsio->cdb;
5597
5598         len = scsi_2btoul(cdb->length);
5599         byte2 = cdb->byte2;
5600
5601         /*
5602          * If we've got a kernel request that hasn't been malloced yet,
5603          * malloc it and tell the caller the data buffer is here.
5604          */
5605         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5606                 ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);;
5607                 ctsio->kern_data_len = len;
5608                 ctsio->kern_total_len = len;
5609                 ctsio->kern_data_resid = 0;
5610                 ctsio->kern_rel_offset = 0;
5611                 ctsio->kern_sg_entries = 0;
5612                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5613                 ctsio->be_move_done = ctl_config_move_done;
5614                 ctl_datamove((union ctl_io *)ctsio);
5615
5616                 return (CTL_RETVAL_COMPLETE);
5617         }
5618
5619         len = ctsio->kern_total_len - ctsio->kern_data_resid;
5620         hdr = (struct scsi_unmap_header *)ctsio->kern_data_ptr;
5621         if (len < sizeof (*hdr) ||
5622             len < (scsi_2btoul(hdr->length) + sizeof(hdr->length)) ||
5623             len < (scsi_2btoul(hdr->desc_length) + sizeof (*hdr)) ||
5624             scsi_2btoul(hdr->desc_length) % sizeof(*buf) != 0) {
5625                 ctl_set_invalid_field(ctsio,
5626                                       /*sks_valid*/ 0,
5627                                       /*command*/ 0,
5628                                       /*field*/ 0,
5629                                       /*bit_valid*/ 0,
5630                                       /*bit*/ 0);
5631                 goto done;
5632         }
5633         len = scsi_2btoul(hdr->desc_length);
5634         buf = (struct scsi_unmap_desc *)(hdr + 1);
5635         end = buf + len / sizeof(*buf);
5636
5637         endnz = buf;
5638         for (range = buf; range < end; range++) {
5639                 lba = scsi_8btou64(range->lba);
5640                 num_blocks = scsi_4btoul(range->length);
5641                 if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5642                  || ((lba + num_blocks) < lba)) {
5643                         ctl_set_lba_out_of_range(ctsio);
5644                         ctl_done((union ctl_io *)ctsio);
5645                         return (CTL_RETVAL_COMPLETE);
5646                 }
5647                 if (num_blocks != 0)
5648                         endnz = range + 1;
5649         }
5650
5651         /*
5652          * Block backend can not handle zero last range.
5653          * Filter it out and return if there is nothing left.
5654          */
5655         len = (uint8_t *)endnz - (uint8_t *)buf;
5656         if (len == 0) {
5657                 ctl_set_success(ctsio);
5658                 goto done;
5659         }
5660
5661         mtx_lock(&lun->lun_lock);
5662         ptrlen = (struct ctl_ptr_len_flags *)
5663             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5664         ptrlen->ptr = (void *)buf;
5665         ptrlen->len = len;
5666         ptrlen->flags = byte2;
5667         ctl_check_blocked(lun);
5668         mtx_unlock(&lun->lun_lock);
5669
5670         retval = lun->backend->config_write((union ctl_io *)ctsio);
5671         return (retval);
5672
5673 done:
5674         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5675                 free(ctsio->kern_data_ptr, M_CTL);
5676                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5677         }
5678         ctl_done((union ctl_io *)ctsio);
5679         return (CTL_RETVAL_COMPLETE);
5680 }
5681
5682 /*
5683  * Note that this function currently doesn't actually do anything inside
5684  * CTL to enforce things if the DQue bit is turned on.
5685  *
5686  * Also note that this function can't be used in the default case, because
5687  * the DQue bit isn't set in the changeable mask for the control mode page
5688  * anyway.  This is just here as an example for how to implement a page
5689  * handler, and a placeholder in case we want to allow the user to turn
5690  * tagged queueing on and off.
5691  *
5692  * The D_SENSE bit handling is functional, however, and will turn
5693  * descriptor sense on and off for a given LUN.
5694  */
5695 int
5696 ctl_control_page_handler(struct ctl_scsiio *ctsio,
5697                          struct ctl_page_index *page_index, uint8_t *page_ptr)
5698 {
5699         struct scsi_control_page *current_cp, *saved_cp, *user_cp;
5700         struct ctl_lun *lun;
5701         int set_ua;
5702         uint32_t initidx;
5703
5704         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5705         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5706         set_ua = 0;
5707
5708         user_cp = (struct scsi_control_page *)page_ptr;
5709         current_cp = (struct scsi_control_page *)
5710                 (page_index->page_data + (page_index->page_len *
5711                 CTL_PAGE_CURRENT));
5712         saved_cp = (struct scsi_control_page *)
5713                 (page_index->page_data + (page_index->page_len *
5714                 CTL_PAGE_SAVED));
5715
5716         mtx_lock(&lun->lun_lock);
5717         if (((current_cp->rlec & SCP_DSENSE) == 0)
5718          && ((user_cp->rlec & SCP_DSENSE) != 0)) {
5719                 /*
5720                  * Descriptor sense is currently turned off and the user
5721                  * wants to turn it on.
5722                  */
5723                 current_cp->rlec |= SCP_DSENSE;
5724                 saved_cp->rlec |= SCP_DSENSE;
5725                 lun->flags |= CTL_LUN_SENSE_DESC;
5726                 set_ua = 1;
5727         } else if (((current_cp->rlec & SCP_DSENSE) != 0)
5728                 && ((user_cp->rlec & SCP_DSENSE) == 0)) {
5729                 /*
5730                  * Descriptor sense is currently turned on, and the user
5731                  * wants to turn it off.
5732                  */
5733                 current_cp->rlec &= ~SCP_DSENSE;
5734                 saved_cp->rlec &= ~SCP_DSENSE;
5735                 lun->flags &= ~CTL_LUN_SENSE_DESC;
5736                 set_ua = 1;
5737         }
5738         if ((current_cp->queue_flags & SCP_QUEUE_ALG_MASK) !=
5739             (user_cp->queue_flags & SCP_QUEUE_ALG_MASK)) {
5740                 current_cp->queue_flags &= ~SCP_QUEUE_ALG_MASK;
5741                 current_cp->queue_flags |= user_cp->queue_flags & SCP_QUEUE_ALG_MASK;
5742                 saved_cp->queue_flags &= ~SCP_QUEUE_ALG_MASK;
5743                 saved_cp->queue_flags |= user_cp->queue_flags & SCP_QUEUE_ALG_MASK;
5744                 set_ua = 1;
5745         }
5746         if ((current_cp->eca_and_aen & SCP_SWP) !=
5747             (user_cp->eca_and_aen & SCP_SWP)) {
5748                 current_cp->eca_and_aen &= ~SCP_SWP;
5749                 current_cp->eca_and_aen |= user_cp->eca_and_aen & SCP_SWP;
5750                 saved_cp->eca_and_aen &= ~SCP_SWP;
5751                 saved_cp->eca_and_aen |= user_cp->eca_and_aen & SCP_SWP;
5752                 set_ua = 1;
5753         }
5754         if (set_ua != 0)
5755                 ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
5756         mtx_unlock(&lun->lun_lock);
5757
5758         return (0);
5759 }
5760
5761 int
5762 ctl_caching_sp_handler(struct ctl_scsiio *ctsio,
5763                      struct ctl_page_index *page_index, uint8_t *page_ptr)
5764 {
5765         struct scsi_caching_page *current_cp, *saved_cp, *user_cp;
5766         struct ctl_lun *lun;
5767         int set_ua;
5768         uint32_t initidx;
5769
5770         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5771         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5772         set_ua = 0;
5773
5774         user_cp = (struct scsi_caching_page *)page_ptr;
5775         current_cp = (struct scsi_caching_page *)
5776                 (page_index->page_data + (page_index->page_len *
5777                 CTL_PAGE_CURRENT));
5778         saved_cp = (struct scsi_caching_page *)
5779                 (page_index->page_data + (page_index->page_len *
5780                 CTL_PAGE_SAVED));
5781
5782         mtx_lock(&lun->lun_lock);
5783         if ((current_cp->flags1 & (SCP_WCE | SCP_RCD)) !=
5784             (user_cp->flags1 & (SCP_WCE | SCP_RCD))) {
5785                 current_cp->flags1 &= ~(SCP_WCE | SCP_RCD);
5786                 current_cp->flags1 |= user_cp->flags1 & (SCP_WCE | SCP_RCD);
5787                 saved_cp->flags1 &= ~(SCP_WCE | SCP_RCD);
5788                 saved_cp->flags1 |= user_cp->flags1 & (SCP_WCE | SCP_RCD);
5789                 set_ua = 1;
5790         }
5791         if (set_ua != 0)
5792                 ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
5793         mtx_unlock(&lun->lun_lock);
5794
5795         return (0);
5796 }
5797
5798 int
5799 ctl_debugconf_sp_select_handler(struct ctl_scsiio *ctsio,
5800                                 struct ctl_page_index *page_index,
5801                                 uint8_t *page_ptr)
5802 {
5803         uint8_t *c;
5804         int i;
5805
5806         c = ((struct copan_debugconf_subpage *)page_ptr)->ctl_time_io_secs;
5807         ctl_time_io_secs =
5808                 (c[0] << 8) |
5809                 (c[1] << 0) |
5810                 0;
5811         CTL_DEBUG_PRINT(("set ctl_time_io_secs to %d\n", ctl_time_io_secs));
5812         printf("set ctl_time_io_secs to %d\n", ctl_time_io_secs);
5813         printf("page data:");
5814         for (i=0; i<8; i++)
5815                 printf(" %.2x",page_ptr[i]);
5816         printf("\n");
5817         return (0);
5818 }
5819
5820 int
5821 ctl_debugconf_sp_sense_handler(struct ctl_scsiio *ctsio,
5822                                struct ctl_page_index *page_index,
5823                                int pc)
5824 {
5825         struct copan_debugconf_subpage *page;
5826
5827         page = (struct copan_debugconf_subpage *)page_index->page_data +
5828                 (page_index->page_len * pc);
5829
5830         switch (pc) {
5831         case SMS_PAGE_CTRL_CHANGEABLE >> 6:
5832         case SMS_PAGE_CTRL_DEFAULT >> 6:
5833         case SMS_PAGE_CTRL_SAVED >> 6:
5834                 /*
5835                  * We don't update the changable or default bits for this page.
5836                  */
5837                 break;
5838         case SMS_PAGE_CTRL_CURRENT >> 6:
5839                 page->ctl_time_io_secs[0] = ctl_time_io_secs >> 8;
5840                 page->ctl_time_io_secs[1] = ctl_time_io_secs >> 0;
5841                 break;
5842         default:
5843 #ifdef NEEDTOPORT
5844                 EPRINT(0, "Invalid PC %d!!", pc);
5845 #endif /* NEEDTOPORT */
5846                 break;
5847         }
5848         return (0);
5849 }
5850
5851
5852 static int
5853 ctl_do_mode_select(union ctl_io *io)
5854 {
5855         struct scsi_mode_page_header *page_header;
5856         struct ctl_page_index *page_index;
5857         struct ctl_scsiio *ctsio;
5858         int control_dev, page_len;
5859         int page_len_offset, page_len_size;
5860         union ctl_modepage_info *modepage_info;
5861         struct ctl_lun *lun;
5862         int *len_left, *len_used;
5863         int retval, i;
5864
5865         ctsio = &io->scsiio;
5866         page_index = NULL;
5867         page_len = 0;
5868         retval = CTL_RETVAL_COMPLETE;
5869
5870         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5871
5872         if (lun->be_lun->lun_type != T_DIRECT)
5873                 control_dev = 1;
5874         else
5875                 control_dev = 0;
5876
5877         modepage_info = (union ctl_modepage_info *)
5878                 ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
5879         len_left = &modepage_info->header.len_left;
5880         len_used = &modepage_info->header.len_used;
5881
5882 do_next_page:
5883
5884         page_header = (struct scsi_mode_page_header *)
5885                 (ctsio->kern_data_ptr + *len_used);
5886
5887         if (*len_left == 0) {
5888                 free(ctsio->kern_data_ptr, M_CTL);
5889                 ctl_set_success(ctsio);
5890                 ctl_done((union ctl_io *)ctsio);
5891                 return (CTL_RETVAL_COMPLETE);
5892         } else if (*len_left < sizeof(struct scsi_mode_page_header)) {
5893
5894                 free(ctsio->kern_data_ptr, M_CTL);
5895                 ctl_set_param_len_error(ctsio);
5896                 ctl_done((union ctl_io *)ctsio);
5897                 return (CTL_RETVAL_COMPLETE);
5898
5899         } else if ((page_header->page_code & SMPH_SPF)
5900                 && (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
5901
5902                 free(ctsio->kern_data_ptr, M_CTL);
5903                 ctl_set_param_len_error(ctsio);
5904                 ctl_done((union ctl_io *)ctsio);
5905                 return (CTL_RETVAL_COMPLETE);
5906         }
5907
5908
5909         /*
5910          * XXX KDM should we do something with the block descriptor?
5911          */
5912         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
5913
5914                 if ((control_dev != 0)
5915                  && (lun->mode_pages.index[i].page_flags &
5916                      CTL_PAGE_FLAG_DISK_ONLY))
5917                         continue;
5918
5919                 if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) !=
5920                     (page_header->page_code & SMPH_PC_MASK))
5921                         continue;
5922
5923                 /*
5924                  * If neither page has a subpage code, then we've got a
5925                  * match.
5926                  */
5927                 if (((lun->mode_pages.index[i].page_code & SMPH_SPF) == 0)
5928                  && ((page_header->page_code & SMPH_SPF) == 0)) {
5929                         page_index = &lun->mode_pages.index[i];
5930                         page_len = page_header->page_length;
5931                         break;
5932                 }
5933
5934                 /*
5935                  * If both pages have subpages, then the subpage numbers
5936                  * have to match.
5937                  */
5938                 if ((lun->mode_pages.index[i].page_code & SMPH_SPF)
5939                   && (page_header->page_code & SMPH_SPF)) {
5940                         struct scsi_mode_page_header_sp *sph;
5941
5942                         sph = (struct scsi_mode_page_header_sp *)page_header;
5943
5944                         if (lun->mode_pages.index[i].subpage ==
5945                             sph->subpage) {
5946                                 page_index = &lun->mode_pages.index[i];
5947                                 page_len = scsi_2btoul(sph->page_length);
5948                                 break;
5949                         }
5950                 }
5951         }
5952
5953         /*
5954          * If we couldn't find the page, or if we don't have a mode select
5955          * handler for it, send back an error to the user.
5956          */
5957         if ((page_index == NULL)
5958          || (page_index->select_handler == NULL)) {
5959                 ctl_set_invalid_field(ctsio,
5960                                       /*sks_valid*/ 1,
5961                                       /*command*/ 0,
5962                                       /*field*/ *len_used,
5963                                       /*bit_valid*/ 0,
5964                                       /*bit*/ 0);
5965                 free(ctsio->kern_data_ptr, M_CTL);
5966                 ctl_done((union ctl_io *)ctsio);
5967                 return (CTL_RETVAL_COMPLETE);
5968         }
5969
5970         if (page_index->page_code & SMPH_SPF) {
5971                 page_len_offset = 2;
5972                 page_len_size = 2;
5973         } else {
5974                 page_len_size = 1;
5975                 page_len_offset = 1;
5976         }
5977
5978         /*
5979          * If the length the initiator gives us isn't the one we specify in
5980          * the mode page header, or if they didn't specify enough data in
5981          * the CDB to avoid truncating this page, kick out the request.
5982          */
5983         if ((page_len != (page_index->page_len - page_len_offset -
5984                           page_len_size))
5985          || (*len_left < page_index->page_len)) {
5986
5987
5988                 ctl_set_invalid_field(ctsio,
5989                                       /*sks_valid*/ 1,
5990                                       /*command*/ 0,
5991                                       /*field*/ *len_used + page_len_offset,
5992                                       /*bit_valid*/ 0,
5993                                       /*bit*/ 0);
5994                 free(ctsio->kern_data_ptr, M_CTL);
5995                 ctl_done((union ctl_io *)ctsio);
5996                 return (CTL_RETVAL_COMPLETE);
5997         }
5998
5999         /*
6000          * Run through the mode page, checking to make sure that the bits
6001          * the user changed are actually legal for him to change.
6002          */
6003         for (i = 0; i < page_index->page_len; i++) {
6004                 uint8_t *user_byte, *change_mask, *current_byte;
6005                 int bad_bit;
6006                 int j;
6007
6008                 user_byte = (uint8_t *)page_header + i;
6009                 change_mask = page_index->page_data +
6010                               (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
6011                 current_byte = page_index->page_data +
6012                                (page_index->page_len * CTL_PAGE_CURRENT) + i;
6013
6014                 /*
6015                  * Check to see whether the user set any bits in this byte
6016                  * that he is not allowed to set.
6017                  */
6018                 if ((*user_byte & ~(*change_mask)) ==
6019                     (*current_byte & ~(*change_mask)))
6020                         continue;
6021
6022                 /*
6023                  * Go through bit by bit to determine which one is illegal.
6024                  */
6025                 bad_bit = 0;
6026                 for (j = 7; j >= 0; j--) {
6027                         if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
6028                             (((1 << i) & ~(*change_mask)) & *current_byte)) {
6029                                 bad_bit = i;
6030                                 break;
6031                         }
6032                 }
6033                 ctl_set_invalid_field(ctsio,
6034                                       /*sks_valid*/ 1,
6035                                       /*command*/ 0,
6036                                       /*field*/ *len_used + i,
6037                                       /*bit_valid*/ 1,
6038                                       /*bit*/ bad_bit);
6039                 free(ctsio->kern_data_ptr, M_CTL);
6040                 ctl_done((union ctl_io *)ctsio);
6041                 return (CTL_RETVAL_COMPLETE);
6042         }
6043
6044         /*
6045          * Decrement these before we call the page handler, since we may
6046          * end up getting called back one way or another before the handler
6047          * returns to this context.
6048          */
6049         *len_left -= page_index->page_len;
6050         *len_used += page_index->page_len;
6051
6052         retval = page_index->select_handler(ctsio, page_index,
6053                                             (uint8_t *)page_header);
6054
6055         /*
6056          * If the page handler returns CTL_RETVAL_QUEUED, then we need to
6057          * wait until this queued command completes to finish processing
6058          * the mode page.  If it returns anything other than
6059          * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
6060          * already set the sense information, freed the data pointer, and
6061          * completed the io for us.
6062          */
6063         if (retval != CTL_RETVAL_COMPLETE)
6064                 goto bailout_no_done;
6065
6066         /*
6067          * If the initiator sent us more than one page, parse the next one.
6068          */
6069         if (*len_left > 0)
6070                 goto do_next_page;
6071
6072         ctl_set_success(ctsio);
6073         free(ctsio->kern_data_ptr, M_CTL);
6074         ctl_done((union ctl_io *)ctsio);
6075
6076 bailout_no_done:
6077
6078         return (CTL_RETVAL_COMPLETE);
6079
6080 }
6081
6082 int
6083 ctl_mode_select(struct ctl_scsiio *ctsio)
6084 {
6085         int param_len, pf, sp;
6086         int header_size, bd_len;
6087         int len_left, len_used;
6088         struct ctl_page_index *page_index;
6089         struct ctl_lun *lun;
6090         int control_dev, page_len;
6091         union ctl_modepage_info *modepage_info;
6092         int retval;
6093
6094         pf = 0;
6095         sp = 0;
6096         page_len = 0;
6097         len_used = 0;
6098         len_left = 0;
6099         retval = 0;
6100         bd_len = 0;
6101         page_index = NULL;
6102
6103         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6104
6105         if (lun->be_lun->lun_type != T_DIRECT)
6106                 control_dev = 1;
6107         else
6108                 control_dev = 0;
6109
6110         switch (ctsio->cdb[0]) {
6111         case MODE_SELECT_6: {
6112                 struct scsi_mode_select_6 *cdb;
6113
6114                 cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
6115
6116                 pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6117                 sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6118
6119                 param_len = cdb->length;
6120                 header_size = sizeof(struct scsi_mode_header_6);
6121                 break;
6122         }
6123         case MODE_SELECT_10: {
6124                 struct scsi_mode_select_10 *cdb;
6125
6126                 cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
6127
6128                 pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6129                 sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6130
6131                 param_len = scsi_2btoul(cdb->length);
6132                 header_size = sizeof(struct scsi_mode_header_10);
6133                 break;
6134         }
6135         default:
6136                 ctl_set_invalid_opcode(ctsio);
6137                 ctl_done((union ctl_io *)ctsio);
6138                 return (CTL_RETVAL_COMPLETE);
6139                 break; /* NOTREACHED */
6140         }
6141
6142         /*
6143          * From SPC-3:
6144          * "A parameter list length of zero indicates that the Data-Out Buffer
6145          * shall be empty. This condition shall not be considered as an error."
6146          */
6147         if (param_len == 0) {
6148                 ctl_set_success(ctsio);
6149                 ctl_done((union ctl_io *)ctsio);
6150                 return (CTL_RETVAL_COMPLETE);
6151         }
6152
6153         /*
6154          * Since we'll hit this the first time through, prior to
6155          * allocation, we don't need to free a data buffer here.
6156          */
6157         if (param_len < header_size) {
6158                 ctl_set_param_len_error(ctsio);
6159                 ctl_done((union ctl_io *)ctsio);
6160                 return (CTL_RETVAL_COMPLETE);
6161         }
6162
6163         /*
6164          * Allocate the data buffer and grab the user's data.  In theory,
6165          * we shouldn't have to sanity check the parameter list length here
6166          * because the maximum size is 64K.  We should be able to malloc
6167          * that much without too many problems.
6168          */
6169         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6170                 ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6171                 ctsio->kern_data_len = param_len;
6172                 ctsio->kern_total_len = param_len;
6173                 ctsio->kern_data_resid = 0;
6174                 ctsio->kern_rel_offset = 0;
6175                 ctsio->kern_sg_entries = 0;
6176                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6177                 ctsio->be_move_done = ctl_config_move_done;
6178                 ctl_datamove((union ctl_io *)ctsio);
6179
6180                 return (CTL_RETVAL_COMPLETE);
6181         }
6182
6183         switch (ctsio->cdb[0]) {
6184         case MODE_SELECT_6: {
6185                 struct scsi_mode_header_6 *mh6;
6186
6187                 mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6188                 bd_len = mh6->blk_desc_len;
6189                 break;
6190         }
6191         case MODE_SELECT_10: {
6192                 struct scsi_mode_header_10 *mh10;
6193
6194                 mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6195                 bd_len = scsi_2btoul(mh10->blk_desc_len);
6196                 break;
6197         }
6198         default:
6199                 panic("Invalid CDB type %#x", ctsio->cdb[0]);
6200                 break;
6201         }
6202
6203         if (param_len < (header_size + bd_len)) {
6204                 free(ctsio->kern_data_ptr, M_CTL);
6205                 ctl_set_param_len_error(ctsio);
6206                 ctl_done((union ctl_io *)ctsio);
6207                 return (CTL_RETVAL_COMPLETE);
6208         }
6209
6210         /*
6211          * Set the IO_CONT flag, so that if this I/O gets passed to
6212          * ctl_config_write_done(), it'll get passed back to
6213          * ctl_do_mode_select() for further processing, or completion if
6214          * we're all done.
6215          */
6216         ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6217         ctsio->io_cont = ctl_do_mode_select;
6218
6219         modepage_info = (union ctl_modepage_info *)
6220                 ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6221
6222         memset(modepage_info, 0, sizeof(*modepage_info));
6223
6224         len_left = param_len - header_size - bd_len;
6225         len_used = header_size + bd_len;
6226
6227         modepage_info->header.len_left = len_left;
6228         modepage_info->header.len_used = len_used;
6229
6230         return (ctl_do_mode_select((union ctl_io *)ctsio));
6231 }
6232
6233 int
6234 ctl_mode_sense(struct ctl_scsiio *ctsio)
6235 {
6236         struct ctl_lun *lun;
6237         int pc, page_code, dbd, llba, subpage;
6238         int alloc_len, page_len, header_len, total_len;
6239         struct scsi_mode_block_descr *block_desc;
6240         struct ctl_page_index *page_index;
6241         int control_dev;
6242
6243         dbd = 0;
6244         llba = 0;
6245         block_desc = NULL;
6246         page_index = NULL;
6247
6248         CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6249
6250         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6251
6252         if (lun->be_lun->lun_type != T_DIRECT)
6253                 control_dev = 1;
6254         else
6255                 control_dev = 0;
6256
6257         switch (ctsio->cdb[0]) {
6258         case MODE_SENSE_6: {
6259                 struct scsi_mode_sense_6 *cdb;
6260
6261                 cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6262
6263                 header_len = sizeof(struct scsi_mode_hdr_6);
6264                 if (cdb->byte2 & SMS_DBD)
6265                         dbd = 1;
6266                 else
6267                         header_len += sizeof(struct scsi_mode_block_descr);
6268
6269                 pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6270                 page_code = cdb->page & SMS_PAGE_CODE;
6271                 subpage = cdb->subpage;
6272                 alloc_len = cdb->length;
6273                 break;
6274         }
6275         case MODE_SENSE_10: {
6276                 struct scsi_mode_sense_10 *cdb;
6277
6278                 cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6279
6280                 header_len = sizeof(struct scsi_mode_hdr_10);
6281
6282                 if (cdb->byte2 & SMS_DBD)
6283                         dbd = 1;
6284                 else
6285                         header_len += sizeof(struct scsi_mode_block_descr);
6286                 if (cdb->byte2 & SMS10_LLBAA)
6287                         llba = 1;
6288                 pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6289                 page_code = cdb->page & SMS_PAGE_CODE;
6290                 subpage = cdb->subpage;
6291                 alloc_len = scsi_2btoul(cdb->length);
6292                 break;
6293         }
6294         default:
6295                 ctl_set_invalid_opcode(ctsio);
6296                 ctl_done((union ctl_io *)ctsio);
6297                 return (CTL_RETVAL_COMPLETE);
6298                 break; /* NOTREACHED */
6299         }
6300
6301         /*
6302          * We have to make a first pass through to calculate the size of
6303          * the pages that match the user's query.  Then we allocate enough
6304          * memory to hold it, and actually copy the data into the buffer.
6305          */
6306         switch (page_code) {
6307         case SMS_ALL_PAGES_PAGE: {
6308                 int i;
6309
6310                 page_len = 0;
6311
6312                 /*
6313                  * At the moment, values other than 0 and 0xff here are
6314                  * reserved according to SPC-3.
6315                  */
6316                 if ((subpage != SMS_SUBPAGE_PAGE_0)
6317                  && (subpage != SMS_SUBPAGE_ALL)) {
6318                         ctl_set_invalid_field(ctsio,
6319                                               /*sks_valid*/ 1,
6320                                               /*command*/ 1,
6321                                               /*field*/ 3,
6322                                               /*bit_valid*/ 0,
6323                                               /*bit*/ 0);
6324                         ctl_done((union ctl_io *)ctsio);
6325                         return (CTL_RETVAL_COMPLETE);
6326                 }
6327
6328                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6329                         if ((control_dev != 0)
6330                          && (lun->mode_pages.index[i].page_flags &
6331                              CTL_PAGE_FLAG_DISK_ONLY))
6332                                 continue;
6333
6334                         /*
6335                          * We don't use this subpage if the user didn't
6336                          * request all subpages.
6337                          */
6338                         if ((lun->mode_pages.index[i].subpage != 0)
6339                          && (subpage == SMS_SUBPAGE_PAGE_0))
6340                                 continue;
6341
6342 #if 0
6343                         printf("found page %#x len %d\n",
6344                                lun->mode_pages.index[i].page_code &
6345                                SMPH_PC_MASK,
6346                                lun->mode_pages.index[i].page_len);
6347 #endif
6348                         page_len += lun->mode_pages.index[i].page_len;
6349                 }
6350                 break;
6351         }
6352         default: {
6353                 int i;
6354
6355                 page_len = 0;
6356
6357                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6358                         /* Look for the right page code */
6359                         if ((lun->mode_pages.index[i].page_code &
6360                              SMPH_PC_MASK) != page_code)
6361                                 continue;
6362
6363                         /* Look for the right subpage or the subpage wildcard*/
6364                         if ((lun->mode_pages.index[i].subpage != subpage)
6365                          && (subpage != SMS_SUBPAGE_ALL))
6366                                 continue;
6367
6368                         /* Make sure the page is supported for this dev type */
6369                         if ((control_dev != 0)
6370                          && (lun->mode_pages.index[i].page_flags &
6371                              CTL_PAGE_FLAG_DISK_ONLY))
6372                                 continue;
6373
6374 #if 0
6375                         printf("found page %#x len %d\n",
6376                                lun->mode_pages.index[i].page_code &
6377                                SMPH_PC_MASK,
6378                                lun->mode_pages.index[i].page_len);
6379 #endif
6380
6381                         page_len += lun->mode_pages.index[i].page_len;
6382                 }
6383
6384                 if (page_len == 0) {
6385                         ctl_set_invalid_field(ctsio,
6386                                               /*sks_valid*/ 1,
6387                                               /*command*/ 1,
6388                                               /*field*/ 2,
6389                                               /*bit_valid*/ 1,
6390                                               /*bit*/ 5);
6391                         ctl_done((union ctl_io *)ctsio);
6392                         return (CTL_RETVAL_COMPLETE);
6393                 }
6394                 break;
6395         }
6396         }
6397
6398         total_len = header_len + page_len;
6399 #if 0
6400         printf("header_len = %d, page_len = %d, total_len = %d\n",
6401                header_len, page_len, total_len);
6402 #endif
6403
6404         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6405         ctsio->kern_sg_entries = 0;
6406         ctsio->kern_data_resid = 0;
6407         ctsio->kern_rel_offset = 0;
6408         if (total_len < alloc_len) {
6409                 ctsio->residual = alloc_len - total_len;
6410                 ctsio->kern_data_len = total_len;
6411                 ctsio->kern_total_len = total_len;
6412         } else {
6413                 ctsio->residual = 0;
6414                 ctsio->kern_data_len = alloc_len;
6415                 ctsio->kern_total_len = alloc_len;
6416         }
6417
6418         switch (ctsio->cdb[0]) {
6419         case MODE_SENSE_6: {
6420                 struct scsi_mode_hdr_6 *header;
6421
6422                 header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
6423
6424                 header->datalen = MIN(total_len - 1, 254);
6425                 if (control_dev == 0) {
6426                         header->dev_specific = 0x10; /* DPOFUA */
6427                         if ((lun->be_lun->flags & CTL_LUN_FLAG_READONLY) ||
6428                             (lun->mode_pages.control_page[CTL_PAGE_CURRENT]
6429                             .eca_and_aen & SCP_SWP) != 0)
6430                                     header->dev_specific |= 0x80; /* WP */
6431                 }
6432                 if (dbd)
6433                         header->block_descr_len = 0;
6434                 else
6435                         header->block_descr_len =
6436                                 sizeof(struct scsi_mode_block_descr);
6437                 block_desc = (struct scsi_mode_block_descr *)&header[1];
6438                 break;
6439         }
6440         case MODE_SENSE_10: {
6441                 struct scsi_mode_hdr_10 *header;
6442                 int datalen;
6443
6444                 header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
6445
6446                 datalen = MIN(total_len - 2, 65533);
6447                 scsi_ulto2b(datalen, header->datalen);
6448                 if (control_dev == 0) {
6449                         header->dev_specific = 0x10; /* DPOFUA */
6450                         if ((lun->be_lun->flags & CTL_LUN_FLAG_READONLY) ||
6451                             (lun->mode_pages.control_page[CTL_PAGE_CURRENT]
6452                             .eca_and_aen & SCP_SWP) != 0)
6453                                     header->dev_specific |= 0x80; /* WP */
6454                 }
6455                 if (dbd)
6456                         scsi_ulto2b(0, header->block_descr_len);
6457                 else
6458                         scsi_ulto2b(sizeof(struct scsi_mode_block_descr),
6459                                     header->block_descr_len);
6460                 block_desc = (struct scsi_mode_block_descr *)&header[1];
6461                 break;
6462         }
6463         default:
6464                 panic("invalid CDB type %#x", ctsio->cdb[0]);
6465                 break; /* NOTREACHED */
6466         }
6467
6468         /*
6469          * If we've got a disk, use its blocksize in the block
6470          * descriptor.  Otherwise, just set it to 0.
6471          */
6472         if (dbd == 0) {
6473                 if (control_dev == 0)
6474                         scsi_ulto3b(lun->be_lun->blocksize,
6475                                     block_desc->block_len);
6476                 else
6477                         scsi_ulto3b(0, block_desc->block_len);
6478         }
6479
6480         switch (page_code) {
6481         case SMS_ALL_PAGES_PAGE: {
6482                 int i, data_used;
6483
6484                 data_used = header_len;
6485                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6486                         struct ctl_page_index *page_index;
6487
6488                         page_index = &lun->mode_pages.index[i];
6489
6490                         if ((control_dev != 0)
6491                          && (page_index->page_flags &
6492                             CTL_PAGE_FLAG_DISK_ONLY))
6493                                 continue;
6494
6495                         /*
6496                          * We don't use this subpage if the user didn't
6497                          * request all subpages.  We already checked (above)
6498                          * to make sure the user only specified a subpage
6499                          * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case.
6500                          */
6501                         if ((page_index->subpage != 0)
6502                          && (subpage == SMS_SUBPAGE_PAGE_0))
6503                                 continue;
6504
6505                         /*
6506                          * Call the handler, if it exists, to update the
6507                          * page to the latest values.
6508                          */
6509                         if (page_index->sense_handler != NULL)
6510                                 page_index->sense_handler(ctsio, page_index,pc);
6511
6512                         memcpy(ctsio->kern_data_ptr + data_used,
6513                                page_index->page_data +
6514                                (page_index->page_len * pc),
6515                                page_index->page_len);
6516                         data_used += page_index->page_len;
6517                 }
6518                 break;
6519         }
6520         default: {
6521                 int i, data_used;
6522
6523                 data_used = header_len;
6524
6525                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6526                         struct ctl_page_index *page_index;
6527
6528                         page_index = &lun->mode_pages.index[i];
6529
6530                         /* Look for the right page code */
6531                         if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6532                                 continue;
6533
6534                         /* Look for the right subpage or the subpage wildcard*/
6535                         if ((page_index->subpage != subpage)
6536                          && (subpage != SMS_SUBPAGE_ALL))
6537                                 continue;
6538
6539                         /* Make sure the page is supported for this dev type */
6540                         if ((control_dev != 0)
6541                          && (page_index->page_flags &
6542                              CTL_PAGE_FLAG_DISK_ONLY))
6543                                 continue;
6544
6545                         /*
6546                          * Call the handler, if it exists, to update the
6547                          * page to the latest values.
6548                          */
6549                         if (page_index->sense_handler != NULL)
6550                                 page_index->sense_handler(ctsio, page_index,pc);
6551
6552                         memcpy(ctsio->kern_data_ptr + data_used,
6553                                page_index->page_data +
6554                                (page_index->page_len * pc),
6555                                page_index->page_len);
6556                         data_used += page_index->page_len;
6557                 }
6558                 break;
6559         }
6560         }
6561
6562         ctl_set_success(ctsio);
6563         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6564         ctsio->be_move_done = ctl_config_move_done;
6565         ctl_datamove((union ctl_io *)ctsio);
6566         return (CTL_RETVAL_COMPLETE);
6567 }
6568
6569 int
6570 ctl_lbp_log_sense_handler(struct ctl_scsiio *ctsio,
6571                                struct ctl_page_index *page_index,
6572                                int pc)
6573 {
6574         struct ctl_lun *lun;
6575         struct scsi_log_param_header *phdr;
6576         uint8_t *data;
6577         uint64_t val;
6578
6579         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6580         data = page_index->page_data;
6581
6582         if (lun->backend->lun_attr != NULL &&
6583             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "blocksavail"))
6584              != UINT64_MAX) {
6585                 phdr = (struct scsi_log_param_header *)data;
6586                 scsi_ulto2b(0x0001, phdr->param_code);
6587                 phdr->param_control = SLP_LBIN | SLP_LP;
6588                 phdr->param_len = 8;
6589                 data = (uint8_t *)(phdr + 1);
6590                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6591                 data[4] = 0x02; /* per-pool */
6592                 data += phdr->param_len;
6593         }
6594
6595         if (lun->backend->lun_attr != NULL &&
6596             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "blocksused"))
6597              != UINT64_MAX) {
6598                 phdr = (struct scsi_log_param_header *)data;
6599                 scsi_ulto2b(0x0002, phdr->param_code);
6600                 phdr->param_control = SLP_LBIN | SLP_LP;
6601                 phdr->param_len = 8;
6602                 data = (uint8_t *)(phdr + 1);
6603                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6604                 data[4] = 0x01; /* per-LUN */
6605                 data += phdr->param_len;
6606         }
6607
6608         if (lun->backend->lun_attr != NULL &&
6609             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "poolblocksavail"))
6610              != UINT64_MAX) {
6611                 phdr = (struct scsi_log_param_header *)data;
6612                 scsi_ulto2b(0x00f1, phdr->param_code);
6613                 phdr->param_control = SLP_LBIN | SLP_LP;
6614                 phdr->param_len = 8;
6615                 data = (uint8_t *)(phdr + 1);
6616                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6617                 data[4] = 0x02; /* per-pool */
6618                 data += phdr->param_len;
6619         }
6620
6621         if (lun->backend->lun_attr != NULL &&
6622             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "poolblocksused"))
6623              != UINT64_MAX) {
6624                 phdr = (struct scsi_log_param_header *)data;
6625                 scsi_ulto2b(0x00f2, phdr->param_code);
6626                 phdr->param_control = SLP_LBIN | SLP_LP;
6627                 phdr->param_len = 8;
6628                 data = (uint8_t *)(phdr + 1);
6629                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6630                 data[4] = 0x02; /* per-pool */
6631                 data += phdr->param_len;
6632         }
6633
6634         page_index->page_len = data - page_index->page_data;
6635         return (0);
6636 }
6637
6638 int
6639 ctl_sap_log_sense_handler(struct ctl_scsiio *ctsio,
6640                                struct ctl_page_index *page_index,
6641                                int pc)
6642 {
6643         struct ctl_lun *lun;
6644         struct stat_page *data;
6645         uint64_t rn, wn, rb, wb;
6646         struct bintime rt, wt;
6647         int i;
6648
6649         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6650         data = (struct stat_page *)page_index->page_data;
6651
6652         scsi_ulto2b(SLP_SAP, data->sap.hdr.param_code);
6653         data->sap.hdr.param_control = SLP_LBIN;
6654         data->sap.hdr.param_len = sizeof(struct scsi_log_stat_and_perf) -
6655             sizeof(struct scsi_log_param_header);
6656         rn = wn = rb = wb = 0;
6657         bintime_clear(&rt);
6658         bintime_clear(&wt);
6659         for (i = 0; i < CTL_MAX_PORTS; i++) {
6660                 rn += lun->stats.ports[i].operations[CTL_STATS_READ];
6661                 wn += lun->stats.ports[i].operations[CTL_STATS_WRITE];
6662                 rb += lun->stats.ports[i].bytes[CTL_STATS_READ];
6663                 wb += lun->stats.ports[i].bytes[CTL_STATS_WRITE];
6664                 bintime_add(&rt, &lun->stats.ports[i].time[CTL_STATS_READ]);
6665                 bintime_add(&wt, &lun->stats.ports[i].time[CTL_STATS_WRITE]);
6666         }
6667         scsi_u64to8b(rn, data->sap.read_num);
6668         scsi_u64to8b(wn, data->sap.write_num);
6669         if (lun->stats.blocksize > 0) {
6670                 scsi_u64to8b(wb / lun->stats.blocksize,
6671                     data->sap.recvieved_lba);
6672                 scsi_u64to8b(rb / lun->stats.blocksize,
6673                     data->sap.transmitted_lba);
6674         }
6675         scsi_u64to8b((uint64_t)rt.sec * 1000 + rt.frac / (UINT64_MAX / 1000),
6676             data->sap.read_int);
6677         scsi_u64to8b((uint64_t)wt.sec * 1000 + wt.frac / (UINT64_MAX / 1000),
6678             data->sap.write_int);
6679         scsi_u64to8b(0, data->sap.weighted_num);
6680         scsi_u64to8b(0, data->sap.weighted_int);
6681         scsi_ulto2b(SLP_IT, data->it.hdr.param_code);
6682         data->it.hdr.param_control = SLP_LBIN;
6683         data->it.hdr.param_len = sizeof(struct scsi_log_idle_time) -
6684             sizeof(struct scsi_log_param_header);
6685 #ifdef CTL_TIME_IO
6686         scsi_u64to8b(lun->idle_time / SBT_1MS, data->it.idle_int);
6687 #endif
6688         scsi_ulto2b(SLP_TI, data->ti.hdr.param_code);
6689         data->it.hdr.param_control = SLP_LBIN;
6690         data->ti.hdr.param_len = sizeof(struct scsi_log_time_interval) -
6691             sizeof(struct scsi_log_param_header);
6692         scsi_ulto4b(3, data->ti.exponent);
6693         scsi_ulto4b(1, data->ti.integer);
6694
6695         page_index->page_len = sizeof(*data);
6696         return (0);
6697 }
6698
6699 int
6700 ctl_log_sense(struct ctl_scsiio *ctsio)
6701 {
6702         struct ctl_lun *lun;
6703         int i, pc, page_code, subpage;
6704         int alloc_len, total_len;
6705         struct ctl_page_index *page_index;
6706         struct scsi_log_sense *cdb;
6707         struct scsi_log_header *header;
6708
6709         CTL_DEBUG_PRINT(("ctl_log_sense\n"));
6710
6711         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6712         cdb = (struct scsi_log_sense *)ctsio->cdb;
6713         pc = (cdb->page & SLS_PAGE_CTRL_MASK) >> 6;
6714         page_code = cdb->page & SLS_PAGE_CODE;
6715         subpage = cdb->subpage;
6716         alloc_len = scsi_2btoul(cdb->length);
6717
6718         page_index = NULL;
6719         for (i = 0; i < CTL_NUM_LOG_PAGES; i++) {
6720                 page_index = &lun->log_pages.index[i];
6721
6722                 /* Look for the right page code */
6723                 if ((page_index->page_code & SL_PAGE_CODE) != page_code)
6724                         continue;
6725
6726                 /* Look for the right subpage or the subpage wildcard*/
6727                 if (page_index->subpage != subpage)
6728                         continue;
6729
6730                 break;
6731         }
6732         if (i >= CTL_NUM_LOG_PAGES) {
6733                 ctl_set_invalid_field(ctsio,
6734                                       /*sks_valid*/ 1,
6735                                       /*command*/ 1,
6736                                       /*field*/ 2,
6737                                       /*bit_valid*/ 0,
6738                                       /*bit*/ 0);
6739                 ctl_done((union ctl_io *)ctsio);
6740                 return (CTL_RETVAL_COMPLETE);
6741         }
6742
6743         total_len = sizeof(struct scsi_log_header) + page_index->page_len;
6744
6745         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6746         ctsio->kern_sg_entries = 0;
6747         ctsio->kern_data_resid = 0;
6748         ctsio->kern_rel_offset = 0;
6749         if (total_len < alloc_len) {
6750                 ctsio->residual = alloc_len - total_len;
6751                 ctsio->kern_data_len = total_len;
6752                 ctsio->kern_total_len = total_len;
6753         } else {
6754                 ctsio->residual = 0;
6755                 ctsio->kern_data_len = alloc_len;
6756                 ctsio->kern_total_len = alloc_len;
6757         }
6758
6759         header = (struct scsi_log_header *)ctsio->kern_data_ptr;
6760         header->page = page_index->page_code;
6761         if (page_index->subpage) {
6762                 header->page |= SL_SPF;
6763                 header->subpage = page_index->subpage;
6764         }
6765         scsi_ulto2b(page_index->page_len, header->datalen);
6766
6767         /*
6768          * Call the handler, if it exists, to update the
6769          * page to the latest values.
6770          */
6771         if (page_index->sense_handler != NULL)
6772                 page_index->sense_handler(ctsio, page_index, pc);
6773
6774         memcpy(header + 1, page_index->page_data, page_index->page_len);
6775
6776         ctl_set_success(ctsio);
6777         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6778         ctsio->be_move_done = ctl_config_move_done;
6779         ctl_datamove((union ctl_io *)ctsio);
6780         return (CTL_RETVAL_COMPLETE);
6781 }
6782
6783 int
6784 ctl_read_capacity(struct ctl_scsiio *ctsio)
6785 {
6786         struct scsi_read_capacity *cdb;
6787         struct scsi_read_capacity_data *data;
6788         struct ctl_lun *lun;
6789         uint32_t lba;
6790
6791         CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
6792
6793         cdb = (struct scsi_read_capacity *)ctsio->cdb;
6794
6795         lba = scsi_4btoul(cdb->addr);
6796         if (((cdb->pmi & SRC_PMI) == 0)
6797          && (lba != 0)) {
6798                 ctl_set_invalid_field(/*ctsio*/ ctsio,
6799                                       /*sks_valid*/ 1,
6800                                       /*command*/ 1,
6801                                       /*field*/ 2,
6802                                       /*bit_valid*/ 0,
6803                                       /*bit*/ 0);
6804                 ctl_done((union ctl_io *)ctsio);
6805                 return (CTL_RETVAL_COMPLETE);
6806         }
6807
6808         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6809
6810         ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
6811         data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
6812         ctsio->residual = 0;
6813         ctsio->kern_data_len = sizeof(*data);
6814         ctsio->kern_total_len = sizeof(*data);
6815         ctsio->kern_data_resid = 0;
6816         ctsio->kern_rel_offset = 0;
6817         ctsio->kern_sg_entries = 0;
6818
6819         /*
6820          * If the maximum LBA is greater than 0xfffffffe, the user must
6821          * issue a SERVICE ACTION IN (16) command, with the read capacity
6822          * serivce action set.
6823          */
6824         if (lun->be_lun->maxlba > 0xfffffffe)
6825                 scsi_ulto4b(0xffffffff, data->addr);
6826         else
6827                 scsi_ulto4b(lun->be_lun->maxlba, data->addr);
6828
6829         /*
6830          * XXX KDM this may not be 512 bytes...
6831          */
6832         scsi_ulto4b(lun->be_lun->blocksize, data->length);
6833
6834         ctl_set_success(ctsio);
6835         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6836         ctsio->be_move_done = ctl_config_move_done;
6837         ctl_datamove((union ctl_io *)ctsio);
6838         return (CTL_RETVAL_COMPLETE);
6839 }
6840
6841 int
6842 ctl_read_capacity_16(struct ctl_scsiio *ctsio)
6843 {
6844         struct scsi_read_capacity_16 *cdb;
6845         struct scsi_read_capacity_data_long *data;
6846         struct ctl_lun *lun;
6847         uint64_t lba;
6848         uint32_t alloc_len;
6849
6850         CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
6851
6852         cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
6853
6854         alloc_len = scsi_4btoul(cdb->alloc_len);
6855         lba = scsi_8btou64(cdb->addr);
6856
6857         if ((cdb->reladr & SRC16_PMI)
6858          && (lba != 0)) {
6859                 ctl_set_invalid_field(/*ctsio*/ ctsio,
6860                                       /*sks_valid*/ 1,
6861                                       /*command*/ 1,
6862                                       /*field*/ 2,
6863                                       /*bit_valid*/ 0,
6864                                       /*bit*/ 0);
6865                 ctl_done((union ctl_io *)ctsio);
6866                 return (CTL_RETVAL_COMPLETE);
6867         }
6868
6869         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6870
6871         ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
6872         data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
6873
6874         if (sizeof(*data) < alloc_len) {
6875                 ctsio->residual = alloc_len - sizeof(*data);
6876                 ctsio->kern_data_len = sizeof(*data);
6877                 ctsio->kern_total_len = sizeof(*data);
6878         } else {
6879                 ctsio->residual = 0;
6880                 ctsio->kern_data_len = alloc_len;
6881                 ctsio->kern_total_len = alloc_len;
6882         }
6883         ctsio->kern_data_resid = 0;
6884         ctsio->kern_rel_offset = 0;
6885         ctsio->kern_sg_entries = 0;
6886
6887         scsi_u64to8b(lun->be_lun->maxlba, data->addr);
6888         /* XXX KDM this may not be 512 bytes... */
6889         scsi_ulto4b(lun->be_lun->blocksize, data->length);
6890         data->prot_lbppbe = lun->be_lun->pblockexp & SRC16_LBPPBE;
6891         scsi_ulto2b(lun->be_lun->pblockoff & SRC16_LALBA_A, data->lalba_lbp);
6892         if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
6893                 data->lalba_lbp[0] |= SRC16_LBPME | SRC16_LBPRZ;
6894
6895         ctl_set_success(ctsio);
6896         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6897         ctsio->be_move_done = ctl_config_move_done;
6898         ctl_datamove((union ctl_io *)ctsio);
6899         return (CTL_RETVAL_COMPLETE);
6900 }
6901
6902 int
6903 ctl_get_lba_status(struct ctl_scsiio *ctsio)
6904 {
6905         struct scsi_get_lba_status *cdb;
6906         struct scsi_get_lba_status_data *data;
6907         struct ctl_lun *lun;
6908         struct ctl_lba_len_flags *lbalen;
6909         uint64_t lba;
6910         uint32_t alloc_len, total_len;
6911         int retval;
6912
6913         CTL_DEBUG_PRINT(("ctl_get_lba_status\n"));
6914
6915         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6916         cdb = (struct scsi_get_lba_status *)ctsio->cdb;
6917         lba = scsi_8btou64(cdb->addr);
6918         alloc_len = scsi_4btoul(cdb->alloc_len);
6919
6920         if (lba > lun->be_lun->maxlba) {
6921                 ctl_set_lba_out_of_range(ctsio);
6922                 ctl_done((union ctl_io *)ctsio);
6923                 return (CTL_RETVAL_COMPLETE);
6924         }
6925
6926         total_len = sizeof(*data) + sizeof(data->descr[0]);
6927         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6928         data = (struct scsi_get_lba_status_data *)ctsio->kern_data_ptr;
6929
6930         if (total_len < alloc_len) {
6931                 ctsio->residual = alloc_len - total_len;
6932                 ctsio->kern_data_len = total_len;
6933                 ctsio->kern_total_len = total_len;
6934         } else {
6935                 ctsio->residual = 0;
6936                 ctsio->kern_data_len = alloc_len;
6937                 ctsio->kern_total_len = alloc_len;
6938         }
6939         ctsio->kern_data_resid = 0;
6940         ctsio->kern_rel_offset = 0;
6941         ctsio->kern_sg_entries = 0;
6942
6943         /* Fill dummy data in case backend can't tell anything. */
6944         scsi_ulto4b(4 + sizeof(data->descr[0]), data->length);
6945         scsi_u64to8b(lba, data->descr[0].addr);
6946         scsi_ulto4b(MIN(UINT32_MAX, lun->be_lun->maxlba + 1 - lba),
6947             data->descr[0].length);
6948         data->descr[0].status = 0; /* Mapped or unknown. */
6949
6950         ctl_set_success(ctsio);
6951         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6952         ctsio->be_move_done = ctl_config_move_done;
6953
6954         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
6955         lbalen->lba = lba;
6956         lbalen->len = total_len;
6957         lbalen->flags = 0;
6958         retval = lun->backend->config_read((union ctl_io *)ctsio);
6959         return (CTL_RETVAL_COMPLETE);
6960 }
6961
6962 int
6963 ctl_read_defect(struct ctl_scsiio *ctsio)
6964 {
6965         struct scsi_read_defect_data_10 *ccb10;
6966         struct scsi_read_defect_data_12 *ccb12;
6967         struct scsi_read_defect_data_hdr_10 *data10;
6968         struct scsi_read_defect_data_hdr_12 *data12;
6969         uint32_t alloc_len, data_len;
6970         uint8_t format;
6971
6972         CTL_DEBUG_PRINT(("ctl_read_defect\n"));
6973
6974         if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
6975                 ccb10 = (struct scsi_read_defect_data_10 *)&ctsio->cdb;
6976                 format = ccb10->format;
6977                 alloc_len = scsi_2btoul(ccb10->alloc_length);
6978                 data_len = sizeof(*data10);
6979         } else {
6980                 ccb12 = (struct scsi_read_defect_data_12 *)&ctsio->cdb;
6981                 format = ccb12->format;
6982                 alloc_len = scsi_4btoul(ccb12->alloc_length);
6983                 data_len = sizeof(*data12);
6984         }
6985         if (alloc_len == 0) {
6986                 ctl_set_success(ctsio);
6987                 ctl_done((union ctl_io *)ctsio);
6988                 return (CTL_RETVAL_COMPLETE);
6989         }
6990
6991         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
6992         if (data_len < alloc_len) {
6993                 ctsio->residual = alloc_len - data_len;
6994                 ctsio->kern_data_len = data_len;
6995                 ctsio->kern_total_len = data_len;
6996         } else {
6997                 ctsio->residual = 0;
6998                 ctsio->kern_data_len = alloc_len;
6999                 ctsio->kern_total_len = alloc_len;
7000         }
7001         ctsio->kern_data_resid = 0;
7002         ctsio->kern_rel_offset = 0;
7003         ctsio->kern_sg_entries = 0;
7004
7005         if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
7006                 data10 = (struct scsi_read_defect_data_hdr_10 *)
7007                     ctsio->kern_data_ptr;
7008                 data10->format = format;
7009                 scsi_ulto2b(0, data10->length);
7010         } else {
7011                 data12 = (struct scsi_read_defect_data_hdr_12 *)
7012                     ctsio->kern_data_ptr;
7013                 data12->format = format;
7014                 scsi_ulto2b(0, data12->generation);
7015                 scsi_ulto4b(0, data12->length);
7016         }
7017
7018         ctl_set_success(ctsio);
7019         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7020         ctsio->be_move_done = ctl_config_move_done;
7021         ctl_datamove((union ctl_io *)ctsio);
7022         return (CTL_RETVAL_COMPLETE);
7023 }
7024
7025 int
7026 ctl_report_tagret_port_groups(struct ctl_scsiio *ctsio)
7027 {
7028         struct scsi_maintenance_in *cdb;
7029         int retval;
7030         int alloc_len, ext, total_len = 0, g, pc, pg, gs, os;
7031         int num_target_port_groups, num_target_ports;
7032         struct ctl_lun *lun;
7033         struct ctl_softc *softc;
7034         struct ctl_port *port;
7035         struct scsi_target_group_data *rtg_ptr;
7036         struct scsi_target_group_data_extended *rtg_ext_ptr;
7037         struct scsi_target_port_group_descriptor *tpg_desc;
7038
7039         CTL_DEBUG_PRINT(("ctl_report_tagret_port_groups\n"));
7040
7041         cdb = (struct scsi_maintenance_in *)ctsio->cdb;
7042         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7043         softc = lun->ctl_softc;
7044
7045         retval = CTL_RETVAL_COMPLETE;
7046
7047         switch (cdb->byte2 & STG_PDF_MASK) {
7048         case STG_PDF_LENGTH:
7049                 ext = 0;
7050                 break;
7051         case STG_PDF_EXTENDED:
7052                 ext = 1;
7053                 break;
7054         default:
7055                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7056                                       /*sks_valid*/ 1,
7057                                       /*command*/ 1,
7058                                       /*field*/ 2,
7059                                       /*bit_valid*/ 1,
7060                                       /*bit*/ 5);
7061                 ctl_done((union ctl_io *)ctsio);
7062                 return(retval);
7063         }
7064
7065         if (softc->is_single)
7066                 num_target_port_groups = 1;
7067         else
7068                 num_target_port_groups = NUM_TARGET_PORT_GROUPS;
7069         num_target_ports = 0;
7070         mtx_lock(&softc->ctl_lock);
7071         STAILQ_FOREACH(port, &softc->port_list, links) {
7072                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7073                         continue;
7074                 if (ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
7075                         continue;
7076                 num_target_ports++;
7077         }
7078         mtx_unlock(&softc->ctl_lock);
7079
7080         if (ext)
7081                 total_len = sizeof(struct scsi_target_group_data_extended);
7082         else
7083                 total_len = sizeof(struct scsi_target_group_data);
7084         total_len += sizeof(struct scsi_target_port_group_descriptor) *
7085                 num_target_port_groups +
7086             sizeof(struct scsi_target_port_descriptor) * num_target_ports;
7087
7088         alloc_len = scsi_4btoul(cdb->length);
7089
7090         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7091
7092         ctsio->kern_sg_entries = 0;
7093
7094         if (total_len < alloc_len) {
7095                 ctsio->residual = alloc_len - total_len;
7096                 ctsio->kern_data_len = total_len;
7097                 ctsio->kern_total_len = total_len;
7098         } else {
7099                 ctsio->residual = 0;
7100                 ctsio->kern_data_len = alloc_len;
7101                 ctsio->kern_total_len = alloc_len;
7102         }
7103         ctsio->kern_data_resid = 0;
7104         ctsio->kern_rel_offset = 0;
7105
7106         if (ext) {
7107                 rtg_ext_ptr = (struct scsi_target_group_data_extended *)
7108                     ctsio->kern_data_ptr;
7109                 scsi_ulto4b(total_len - 4, rtg_ext_ptr->length);
7110                 rtg_ext_ptr->format_type = 0x10;
7111                 rtg_ext_ptr->implicit_transition_time = 0;
7112                 tpg_desc = &rtg_ext_ptr->groups[0];
7113         } else {
7114                 rtg_ptr = (struct scsi_target_group_data *)
7115                     ctsio->kern_data_ptr;
7116                 scsi_ulto4b(total_len - 4, rtg_ptr->length);
7117                 tpg_desc = &rtg_ptr->groups[0];
7118         }
7119
7120         mtx_lock(&softc->ctl_lock);
7121         pg = softc->port_min / softc->port_cnt;
7122         if (softc->ha_link == CTL_HA_LINK_OFFLINE)
7123                 gs = TPG_ASYMMETRIC_ACCESS_UNAVAILABLE;
7124         else if (softc->ha_link == CTL_HA_LINK_UNKNOWN)
7125                 gs = TPG_ASYMMETRIC_ACCESS_TRANSITIONING;
7126         else if (softc->ha_mode == CTL_HA_MODE_ACT_STBY)
7127                 gs = TPG_ASYMMETRIC_ACCESS_STANDBY;
7128         else
7129                 gs = TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7130         if (lun->flags & CTL_LUN_PRIMARY_SC) {
7131                 os = gs;
7132                 gs = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7133         } else
7134                 os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7135         for (g = 0; g < num_target_port_groups; g++) {
7136                 tpg_desc->pref_state = (g == pg) ? gs : os;
7137                 tpg_desc->support = TPG_AO_SUP | TPG_AN_SUP | TPG_S_SUP |
7138                     TPG_U_SUP | TPG_T_SUP;
7139                 scsi_ulto2b(g + 1, tpg_desc->target_port_group);
7140                 tpg_desc->status = TPG_IMPLICIT;
7141                 pc = 0;
7142                 STAILQ_FOREACH(port, &softc->port_list, links) {
7143                         if (port->targ_port < g * softc->port_cnt ||
7144                             port->targ_port >= (g + 1) * softc->port_cnt)
7145                                 continue;
7146                         if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7147                                 continue;
7148                         if (ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
7149                                 continue;
7150                         scsi_ulto2b(port->targ_port, tpg_desc->descriptors[pc].
7151                             relative_target_port_identifier);
7152                         pc++;
7153                 }
7154                 tpg_desc->target_port_count = pc;
7155                 tpg_desc = (struct scsi_target_port_group_descriptor *)
7156                     &tpg_desc->descriptors[pc];
7157         }
7158         mtx_unlock(&softc->ctl_lock);
7159
7160         ctl_set_success(ctsio);
7161         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7162         ctsio->be_move_done = ctl_config_move_done;
7163         ctl_datamove((union ctl_io *)ctsio);
7164         return(retval);
7165 }
7166
7167 int
7168 ctl_report_supported_opcodes(struct ctl_scsiio *ctsio)
7169 {
7170         struct ctl_lun *lun;
7171         struct scsi_report_supported_opcodes *cdb;
7172         const struct ctl_cmd_entry *entry, *sentry;
7173         struct scsi_report_supported_opcodes_all *all;
7174         struct scsi_report_supported_opcodes_descr *descr;
7175         struct scsi_report_supported_opcodes_one *one;
7176         int retval;
7177         int alloc_len, total_len;
7178         int opcode, service_action, i, j, num;
7179
7180         CTL_DEBUG_PRINT(("ctl_report_supported_opcodes\n"));
7181
7182         cdb = (struct scsi_report_supported_opcodes *)ctsio->cdb;
7183         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7184
7185         retval = CTL_RETVAL_COMPLETE;
7186
7187         opcode = cdb->requested_opcode;
7188         service_action = scsi_2btoul(cdb->requested_service_action);
7189         switch (cdb->options & RSO_OPTIONS_MASK) {
7190         case RSO_OPTIONS_ALL:
7191                 num = 0;
7192                 for (i = 0; i < 256; i++) {
7193                         entry = &ctl_cmd_table[i];
7194                         if (entry->flags & CTL_CMD_FLAG_SA5) {
7195                                 for (j = 0; j < 32; j++) {
7196                                         sentry = &((const struct ctl_cmd_entry *)
7197                                             entry->execute)[j];
7198                                         if (ctl_cmd_applicable(
7199                                             lun->be_lun->lun_type, sentry))
7200                                                 num++;
7201                                 }
7202                         } else {
7203                                 if (ctl_cmd_applicable(lun->be_lun->lun_type,
7204                                     entry))
7205                                         num++;
7206                         }
7207                 }
7208                 total_len = sizeof(struct scsi_report_supported_opcodes_all) +
7209                     num * sizeof(struct scsi_report_supported_opcodes_descr);
7210                 break;
7211         case RSO_OPTIONS_OC:
7212                 if (ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) {
7213                         ctl_set_invalid_field(/*ctsio*/ ctsio,
7214                                               /*sks_valid*/ 1,
7215                                               /*command*/ 1,
7216                                               /*field*/ 2,
7217                                               /*bit_valid*/ 1,
7218                                               /*bit*/ 2);
7219                         ctl_done((union ctl_io *)ctsio);
7220                         return (CTL_RETVAL_COMPLETE);
7221                 }
7222                 total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7223                 break;
7224         case RSO_OPTIONS_OC_SA:
7225                 if ((ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) == 0 ||
7226                     service_action >= 32) {
7227                         ctl_set_invalid_field(/*ctsio*/ ctsio,
7228                                               /*sks_valid*/ 1,
7229                                               /*command*/ 1,
7230                                               /*field*/ 2,
7231                                               /*bit_valid*/ 1,
7232                                               /*bit*/ 2);
7233                         ctl_done((union ctl_io *)ctsio);
7234                         return (CTL_RETVAL_COMPLETE);
7235                 }
7236                 total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7237                 break;
7238         default:
7239                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7240                                       /*sks_valid*/ 1,
7241                                       /*command*/ 1,
7242                                       /*field*/ 2,
7243                                       /*bit_valid*/ 1,
7244                                       /*bit*/ 2);
7245                 ctl_done((union ctl_io *)ctsio);
7246                 return (CTL_RETVAL_COMPLETE);
7247         }
7248
7249         alloc_len = scsi_4btoul(cdb->length);
7250
7251         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7252
7253         ctsio->kern_sg_entries = 0;
7254
7255         if (total_len < alloc_len) {
7256                 ctsio->residual = alloc_len - total_len;
7257                 ctsio->kern_data_len = total_len;
7258                 ctsio->kern_total_len = total_len;
7259         } else {
7260                 ctsio->residual = 0;
7261                 ctsio->kern_data_len = alloc_len;
7262                 ctsio->kern_total_len = alloc_len;
7263         }
7264         ctsio->kern_data_resid = 0;
7265         ctsio->kern_rel_offset = 0;
7266
7267         switch (cdb->options & RSO_OPTIONS_MASK) {
7268         case RSO_OPTIONS_ALL:
7269                 all = (struct scsi_report_supported_opcodes_all *)
7270                     ctsio->kern_data_ptr;
7271                 num = 0;
7272                 for (i = 0; i < 256; i++) {
7273                         entry = &ctl_cmd_table[i];
7274                         if (entry->flags & CTL_CMD_FLAG_SA5) {
7275                                 for (j = 0; j < 32; j++) {
7276                                         sentry = &((const struct ctl_cmd_entry *)
7277                                             entry->execute)[j];
7278                                         if (!ctl_cmd_applicable(
7279                                             lun->be_lun->lun_type, sentry))
7280                                                 continue;
7281                                         descr = &all->descr[num++];
7282                                         descr->opcode = i;
7283                                         scsi_ulto2b(j, descr->service_action);
7284                                         descr->flags = RSO_SERVACTV;
7285                                         scsi_ulto2b(sentry->length,
7286                                             descr->cdb_length);
7287                                 }
7288                         } else {
7289                                 if (!ctl_cmd_applicable(lun->be_lun->lun_type,
7290                                     entry))
7291                                         continue;
7292                                 descr = &all->descr[num++];
7293                                 descr->opcode = i;
7294                                 scsi_ulto2b(0, descr->service_action);
7295                                 descr->flags = 0;
7296                                 scsi_ulto2b(entry->length, descr->cdb_length);
7297                         }
7298                 }
7299                 scsi_ulto4b(
7300                     num * sizeof(struct scsi_report_supported_opcodes_descr),
7301                     all->length);
7302                 break;
7303         case RSO_OPTIONS_OC:
7304                 one = (struct scsi_report_supported_opcodes_one *)
7305                     ctsio->kern_data_ptr;
7306                 entry = &ctl_cmd_table[opcode];
7307                 goto fill_one;
7308         case RSO_OPTIONS_OC_SA:
7309                 one = (struct scsi_report_supported_opcodes_one *)
7310                     ctsio->kern_data_ptr;
7311                 entry = &ctl_cmd_table[opcode];
7312                 entry = &((const struct ctl_cmd_entry *)
7313                     entry->execute)[service_action];
7314 fill_one:
7315                 if (ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
7316                         one->support = 3;
7317                         scsi_ulto2b(entry->length, one->cdb_length);
7318                         one->cdb_usage[0] = opcode;
7319                         memcpy(&one->cdb_usage[1], entry->usage,
7320                             entry->length - 1);
7321                 } else
7322                         one->support = 1;
7323                 break;
7324         }
7325
7326         ctl_set_success(ctsio);
7327         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7328         ctsio->be_move_done = ctl_config_move_done;
7329         ctl_datamove((union ctl_io *)ctsio);
7330         return(retval);
7331 }
7332
7333 int
7334 ctl_report_supported_tmf(struct ctl_scsiio *ctsio)
7335 {
7336         struct scsi_report_supported_tmf *cdb;
7337         struct scsi_report_supported_tmf_data *data;
7338         int retval;
7339         int alloc_len, total_len;
7340
7341         CTL_DEBUG_PRINT(("ctl_report_supported_tmf\n"));
7342
7343         cdb = (struct scsi_report_supported_tmf *)ctsio->cdb;
7344
7345         retval = CTL_RETVAL_COMPLETE;
7346
7347         total_len = sizeof(struct scsi_report_supported_tmf_data);
7348         alloc_len = scsi_4btoul(cdb->length);
7349
7350         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7351
7352         ctsio->kern_sg_entries = 0;
7353
7354         if (total_len < alloc_len) {
7355                 ctsio->residual = alloc_len - total_len;
7356                 ctsio->kern_data_len = total_len;
7357                 ctsio->kern_total_len = total_len;
7358         } else {
7359                 ctsio->residual = 0;
7360                 ctsio->kern_data_len = alloc_len;
7361                 ctsio->kern_total_len = alloc_len;
7362         }
7363         ctsio->kern_data_resid = 0;
7364         ctsio->kern_rel_offset = 0;
7365
7366         data = (struct scsi_report_supported_tmf_data *)ctsio->kern_data_ptr;
7367         data->byte1 |= RST_ATS | RST_ATSS | RST_CTSS | RST_LURS | RST_TRS;
7368         data->byte2 |= RST_ITNRS;
7369
7370         ctl_set_success(ctsio);
7371         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7372         ctsio->be_move_done = ctl_config_move_done;
7373         ctl_datamove((union ctl_io *)ctsio);
7374         return (retval);
7375 }
7376
7377 int
7378 ctl_report_timestamp(struct ctl_scsiio *ctsio)
7379 {
7380         struct scsi_report_timestamp *cdb;
7381         struct scsi_report_timestamp_data *data;
7382         struct timeval tv;
7383         int64_t timestamp;
7384         int retval;
7385         int alloc_len, total_len;
7386
7387         CTL_DEBUG_PRINT(("ctl_report_timestamp\n"));
7388
7389         cdb = (struct scsi_report_timestamp *)ctsio->cdb;
7390
7391         retval = CTL_RETVAL_COMPLETE;
7392
7393         total_len = sizeof(struct scsi_report_timestamp_data);
7394         alloc_len = scsi_4btoul(cdb->length);
7395
7396         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7397
7398         ctsio->kern_sg_entries = 0;
7399
7400         if (total_len < alloc_len) {
7401                 ctsio->residual = alloc_len - total_len;
7402                 ctsio->kern_data_len = total_len;
7403                 ctsio->kern_total_len = total_len;
7404         } else {
7405                 ctsio->residual = 0;
7406                 ctsio->kern_data_len = alloc_len;
7407                 ctsio->kern_total_len = alloc_len;
7408         }
7409         ctsio->kern_data_resid = 0;
7410         ctsio->kern_rel_offset = 0;
7411
7412         data = (struct scsi_report_timestamp_data *)ctsio->kern_data_ptr;
7413         scsi_ulto2b(sizeof(*data) - 2, data->length);
7414         data->origin = RTS_ORIG_OUTSIDE;
7415         getmicrotime(&tv);
7416         timestamp = (int64_t)tv.tv_sec * 1000 + tv.tv_usec / 1000;
7417         scsi_ulto4b(timestamp >> 16, data->timestamp);
7418         scsi_ulto2b(timestamp & 0xffff, &data->timestamp[4]);
7419
7420         ctl_set_success(ctsio);
7421         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7422         ctsio->be_move_done = ctl_config_move_done;
7423         ctl_datamove((union ctl_io *)ctsio);
7424         return (retval);
7425 }
7426
7427 int
7428 ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7429 {
7430         struct scsi_per_res_in *cdb;
7431         int alloc_len, total_len = 0;
7432         /* struct scsi_per_res_in_rsrv in_data; */
7433         struct ctl_lun *lun;
7434         struct ctl_softc *softc;
7435         uint64_t key;
7436
7437         CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7438
7439         cdb = (struct scsi_per_res_in *)ctsio->cdb;
7440
7441         alloc_len = scsi_2btoul(cdb->length);
7442
7443         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7444         softc = lun->ctl_softc;
7445
7446 retry:
7447         mtx_lock(&lun->lun_lock);
7448         switch (cdb->action) {
7449         case SPRI_RK: /* read keys */
7450                 total_len = sizeof(struct scsi_per_res_in_keys) +
7451                         lun->pr_key_count *
7452                         sizeof(struct scsi_per_res_key);
7453                 break;
7454         case SPRI_RR: /* read reservation */
7455                 if (lun->flags & CTL_LUN_PR_RESERVED)
7456                         total_len = sizeof(struct scsi_per_res_in_rsrv);
7457                 else
7458                         total_len = sizeof(struct scsi_per_res_in_header);
7459                 break;
7460         case SPRI_RC: /* report capabilities */
7461                 total_len = sizeof(struct scsi_per_res_cap);
7462                 break;
7463         case SPRI_RS: /* read full status */
7464                 total_len = sizeof(struct scsi_per_res_in_header) +
7465                     (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7466                     lun->pr_key_count;
7467                 break;
7468         default:
7469                 panic("Invalid PR type %x", cdb->action);
7470         }
7471         mtx_unlock(&lun->lun_lock);
7472
7473         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7474
7475         if (total_len < alloc_len) {
7476                 ctsio->residual = alloc_len - total_len;
7477                 ctsio->kern_data_len = total_len;
7478                 ctsio->kern_total_len = total_len;
7479         } else {
7480                 ctsio->residual = 0;
7481                 ctsio->kern_data_len = alloc_len;
7482                 ctsio->kern_total_len = alloc_len;
7483         }
7484
7485         ctsio->kern_data_resid = 0;
7486         ctsio->kern_rel_offset = 0;
7487         ctsio->kern_sg_entries = 0;
7488
7489         mtx_lock(&lun->lun_lock);
7490         switch (cdb->action) {
7491         case SPRI_RK: { // read keys
7492         struct scsi_per_res_in_keys *res_keys;
7493                 int i, key_count;
7494
7495                 res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7496
7497                 /*
7498                  * We had to drop the lock to allocate our buffer, which
7499                  * leaves time for someone to come in with another
7500                  * persistent reservation.  (That is unlikely, though,
7501                  * since this should be the only persistent reservation
7502                  * command active right now.)
7503                  */
7504                 if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7505                     (lun->pr_key_count *
7506                      sizeof(struct scsi_per_res_key)))){
7507                         mtx_unlock(&lun->lun_lock);
7508                         free(ctsio->kern_data_ptr, M_CTL);
7509                         printf("%s: reservation length changed, retrying\n",
7510                                __func__);
7511                         goto retry;
7512                 }
7513
7514                 scsi_ulto4b(lun->PRGeneration, res_keys->header.generation);
7515
7516                 scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7517                              lun->pr_key_count, res_keys->header.length);
7518
7519                 for (i = 0, key_count = 0; i < CTL_MAX_INITIATORS; i++) {
7520                         if ((key = ctl_get_prkey(lun, i)) == 0)
7521                                 continue;
7522
7523                         /*
7524                          * We used lun->pr_key_count to calculate the
7525                          * size to allocate.  If it turns out the number of
7526                          * initiators with the registered flag set is
7527                          * larger than that (i.e. they haven't been kept in
7528                          * sync), we've got a problem.
7529                          */
7530                         if (key_count >= lun->pr_key_count) {
7531 #ifdef NEEDTOPORT
7532                                 csevent_log(CSC_CTL | CSC_SHELF_SW |
7533                                             CTL_PR_ERROR,
7534                                             csevent_LogType_Fault,
7535                                             csevent_AlertLevel_Yellow,
7536                                             csevent_FRU_ShelfController,
7537                                             csevent_FRU_Firmware,
7538                                         csevent_FRU_Unknown,
7539                                             "registered keys %d >= key "
7540                                             "count %d", key_count,
7541                                             lun->pr_key_count);
7542 #endif
7543                                 key_count++;
7544                                 continue;
7545                         }
7546                         scsi_u64to8b(key, res_keys->keys[key_count].key);
7547                         key_count++;
7548                 }
7549                 break;
7550         }
7551         case SPRI_RR: { // read reservation
7552                 struct scsi_per_res_in_rsrv *res;
7553                 int tmp_len, header_only;
7554
7555                 res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7556
7557                 scsi_ulto4b(lun->PRGeneration, res->header.generation);
7558
7559                 if (lun->flags & CTL_LUN_PR_RESERVED)
7560                 {
7561                         tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7562                         scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7563                                     res->header.length);
7564                         header_only = 0;
7565                 } else {
7566                         tmp_len = sizeof(struct scsi_per_res_in_header);
7567                         scsi_ulto4b(0, res->header.length);
7568                         header_only = 1;
7569                 }
7570
7571                 /*
7572                  * We had to drop the lock to allocate our buffer, which
7573                  * leaves time for someone to come in with another
7574                  * persistent reservation.  (That is unlikely, though,
7575                  * since this should be the only persistent reservation
7576                  * command active right now.)
7577                  */
7578                 if (tmp_len != total_len) {
7579                         mtx_unlock(&lun->lun_lock);
7580                         free(ctsio->kern_data_ptr, M_CTL);
7581                         printf("%s: reservation status changed, retrying\n",
7582                                __func__);
7583                         goto retry;
7584                 }
7585
7586                 /*
7587                  * No reservation held, so we're done.
7588                  */
7589                 if (header_only != 0)
7590                         break;
7591
7592                 /*
7593                  * If the registration is an All Registrants type, the key
7594                  * is 0, since it doesn't really matter.
7595                  */
7596                 if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7597                         scsi_u64to8b(ctl_get_prkey(lun, lun->pr_res_idx),
7598                             res->data.reservation);
7599                 }
7600                 res->data.scopetype = lun->res_type;
7601                 break;
7602         }
7603         case SPRI_RC:     //report capabilities
7604         {
7605                 struct scsi_per_res_cap *res_cap;
7606                 uint16_t type_mask;
7607
7608                 res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7609                 scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7610                 res_cap->flags2 |= SPRI_TMV | SPRI_ALLOW_5;
7611                 type_mask = SPRI_TM_WR_EX_AR |
7612                             SPRI_TM_EX_AC_RO |
7613                             SPRI_TM_WR_EX_RO |
7614                             SPRI_TM_EX_AC |
7615                             SPRI_TM_WR_EX |
7616                             SPRI_TM_EX_AC_AR;
7617                 scsi_ulto2b(type_mask, res_cap->type_mask);
7618                 break;
7619         }
7620         case SPRI_RS: { // read full status
7621                 struct scsi_per_res_in_full *res_status;
7622                 struct scsi_per_res_in_full_desc *res_desc;
7623                 struct ctl_port *port;
7624                 int i, len;
7625
7626                 res_status = (struct scsi_per_res_in_full*)ctsio->kern_data_ptr;
7627
7628                 /*
7629                  * We had to drop the lock to allocate our buffer, which
7630                  * leaves time for someone to come in with another
7631                  * persistent reservation.  (That is unlikely, though,
7632                  * since this should be the only persistent reservation
7633                  * command active right now.)
7634                  */
7635                 if (total_len < (sizeof(struct scsi_per_res_in_header) +
7636                     (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7637                      lun->pr_key_count)){
7638                         mtx_unlock(&lun->lun_lock);
7639                         free(ctsio->kern_data_ptr, M_CTL);
7640                         printf("%s: reservation length changed, retrying\n",
7641                                __func__);
7642                         goto retry;
7643                 }
7644
7645                 scsi_ulto4b(lun->PRGeneration, res_status->header.generation);
7646
7647                 res_desc = &res_status->desc[0];
7648                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
7649                         if ((key = ctl_get_prkey(lun, i)) == 0)
7650                                 continue;
7651
7652                         scsi_u64to8b(key, res_desc->res_key.key);
7653                         if ((lun->flags & CTL_LUN_PR_RESERVED) &&
7654                             (lun->pr_res_idx == i ||
7655                              lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS)) {
7656                                 res_desc->flags = SPRI_FULL_R_HOLDER;
7657                                 res_desc->scopetype = lun->res_type;
7658                         }
7659                         scsi_ulto2b(i / CTL_MAX_INIT_PER_PORT,
7660                             res_desc->rel_trgt_port_id);
7661                         len = 0;
7662                         port = softc->ctl_ports[i / CTL_MAX_INIT_PER_PORT];
7663                         if (port != NULL)
7664                                 len = ctl_create_iid(port,
7665                                     i % CTL_MAX_INIT_PER_PORT,
7666                                     res_desc->transport_id);
7667                         scsi_ulto4b(len, res_desc->additional_length);
7668                         res_desc = (struct scsi_per_res_in_full_desc *)
7669                             &res_desc->transport_id[len];
7670                 }
7671                 scsi_ulto4b((uint8_t *)res_desc - (uint8_t *)&res_status->desc[0],
7672                     res_status->header.length);
7673                 break;
7674         }
7675         default:
7676                 /*
7677                  * This is a bug, because we just checked for this above,
7678                  * and should have returned an error.
7679                  */
7680                 panic("Invalid PR type %x", cdb->action);
7681                 break; /* NOTREACHED */
7682         }
7683         mtx_unlock(&lun->lun_lock);
7684
7685         ctl_set_success(ctsio);
7686         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7687         ctsio->be_move_done = ctl_config_move_done;
7688         ctl_datamove((union ctl_io *)ctsio);
7689         return (CTL_RETVAL_COMPLETE);
7690 }
7691
7692 /*
7693  * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
7694  * it should return.
7695  */
7696 static int
7697 ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
7698                 uint64_t sa_res_key, uint8_t type, uint32_t residx,
7699                 struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
7700                 struct scsi_per_res_out_parms* param)
7701 {
7702         union ctl_ha_msg persis_io;
7703         int i;
7704
7705         mtx_lock(&lun->lun_lock);
7706         if (sa_res_key == 0) {
7707                 if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
7708                         /* validate scope and type */
7709                         if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7710                              SPR_LU_SCOPE) {
7711                                 mtx_unlock(&lun->lun_lock);
7712                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7713                                                       /*sks_valid*/ 1,
7714                                                       /*command*/ 1,
7715                                                       /*field*/ 2,
7716                                                       /*bit_valid*/ 1,
7717                                                       /*bit*/ 4);
7718                                 ctl_done((union ctl_io *)ctsio);
7719                                 return (1);
7720                         }
7721
7722                         if (type>8 || type==2 || type==4 || type==0) {
7723                                 mtx_unlock(&lun->lun_lock);
7724                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7725                                                       /*sks_valid*/ 1,
7726                                                       /*command*/ 1,
7727                                                       /*field*/ 2,
7728                                                       /*bit_valid*/ 1,
7729                                                       /*bit*/ 0);
7730                                 ctl_done((union ctl_io *)ctsio);
7731                                 return (1);
7732                         }
7733
7734                         /*
7735                          * Unregister everybody else and build UA for
7736                          * them
7737                          */
7738                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
7739                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
7740                                         continue;
7741
7742                                 ctl_clr_prkey(lun, i);
7743                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7744                         }
7745                         lun->pr_key_count = 1;
7746                         lun->res_type = type;
7747                         if (lun->res_type != SPR_TYPE_WR_EX_AR
7748                          && lun->res_type != SPR_TYPE_EX_AC_AR)
7749                                 lun->pr_res_idx = residx;
7750                         lun->PRGeneration++;
7751                         mtx_unlock(&lun->lun_lock);
7752
7753                         /* send msg to other side */
7754                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7755                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7756                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7757                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
7758                         persis_io.pr.pr_info.res_type = type;
7759                         memcpy(persis_io.pr.pr_info.sa_res_key,
7760                                param->serv_act_res_key,
7761                                sizeof(param->serv_act_res_key));
7762                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
7763                             sizeof(persis_io.pr), M_WAITOK);
7764                 } else {
7765                         /* not all registrants */
7766                         mtx_unlock(&lun->lun_lock);
7767                         free(ctsio->kern_data_ptr, M_CTL);
7768                         ctl_set_invalid_field(ctsio,
7769                                               /*sks_valid*/ 1,
7770                                               /*command*/ 0,
7771                                               /*field*/ 8,
7772                                               /*bit_valid*/ 0,
7773                                               /*bit*/ 0);
7774                         ctl_done((union ctl_io *)ctsio);
7775                         return (1);
7776                 }
7777         } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
7778                 || !(lun->flags & CTL_LUN_PR_RESERVED)) {
7779                 int found = 0;
7780
7781                 if (res_key == sa_res_key) {
7782                         /* special case */
7783                         /*
7784                          * The spec implies this is not good but doesn't
7785                          * say what to do. There are two choices either
7786                          * generate a res conflict or check condition
7787                          * with illegal field in parameter data. Since
7788                          * that is what is done when the sa_res_key is
7789                          * zero I'll take that approach since this has
7790                          * to do with the sa_res_key.
7791                          */
7792                         mtx_unlock(&lun->lun_lock);
7793                         free(ctsio->kern_data_ptr, M_CTL);
7794                         ctl_set_invalid_field(ctsio,
7795                                               /*sks_valid*/ 1,
7796                                               /*command*/ 0,
7797                                               /*field*/ 8,
7798                                               /*bit_valid*/ 0,
7799                                               /*bit*/ 0);
7800                         ctl_done((union ctl_io *)ctsio);
7801                         return (1);
7802                 }
7803
7804                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
7805                         if (ctl_get_prkey(lun, i) != sa_res_key)
7806                                 continue;
7807
7808                         found = 1;
7809                         ctl_clr_prkey(lun, i);
7810                         lun->pr_key_count--;
7811                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7812                 }
7813                 if (!found) {
7814                         mtx_unlock(&lun->lun_lock);
7815                         free(ctsio->kern_data_ptr, M_CTL);
7816                         ctl_set_reservation_conflict(ctsio);
7817                         ctl_done((union ctl_io *)ctsio);
7818                         return (CTL_RETVAL_COMPLETE);
7819                 }
7820                 lun->PRGeneration++;
7821                 mtx_unlock(&lun->lun_lock);
7822
7823                 /* send msg to other side */
7824                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7825                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7826                 persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7827                 persis_io.pr.pr_info.residx = lun->pr_res_idx;
7828                 persis_io.pr.pr_info.res_type = type;
7829                 memcpy(persis_io.pr.pr_info.sa_res_key,
7830                        param->serv_act_res_key,
7831                        sizeof(param->serv_act_res_key));
7832                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
7833                     sizeof(persis_io.pr), M_WAITOK);
7834         } else {
7835                 /* Reserved but not all registrants */
7836                 /* sa_res_key is res holder */
7837                 if (sa_res_key == ctl_get_prkey(lun, lun->pr_res_idx)) {
7838                         /* validate scope and type */
7839                         if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7840                              SPR_LU_SCOPE) {
7841                                 mtx_unlock(&lun->lun_lock);
7842                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7843                                                       /*sks_valid*/ 1,
7844                                                       /*command*/ 1,
7845                                                       /*field*/ 2,
7846                                                       /*bit_valid*/ 1,
7847                                                       /*bit*/ 4);
7848                                 ctl_done((union ctl_io *)ctsio);
7849                                 return (1);
7850                         }
7851
7852                         if (type>8 || type==2 || type==4 || type==0) {
7853                                 mtx_unlock(&lun->lun_lock);
7854                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7855                                                       /*sks_valid*/ 1,
7856                                                       /*command*/ 1,
7857                                                       /*field*/ 2,
7858                                                       /*bit_valid*/ 1,
7859                                                       /*bit*/ 0);
7860                                 ctl_done((union ctl_io *)ctsio);
7861                                 return (1);
7862                         }
7863
7864                         /*
7865                          * Do the following:
7866                          * if sa_res_key != res_key remove all
7867                          * registrants w/sa_res_key and generate UA
7868                          * for these registrants(Registrations
7869                          * Preempted) if it wasn't an exclusive
7870                          * reservation generate UA(Reservations
7871                          * Preempted) for all other registered nexuses
7872                          * if the type has changed. Establish the new
7873                          * reservation and holder. If res_key and
7874                          * sa_res_key are the same do the above
7875                          * except don't unregister the res holder.
7876                          */
7877
7878                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
7879                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
7880                                         continue;
7881
7882                                 if (sa_res_key == ctl_get_prkey(lun, i)) {
7883                                         ctl_clr_prkey(lun, i);
7884                                         lun->pr_key_count--;
7885                                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7886                                 } else if (type != lun->res_type
7887                                         && (lun->res_type == SPR_TYPE_WR_EX_RO
7888                                          || lun->res_type ==SPR_TYPE_EX_AC_RO)){
7889                                         ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
7890                                 }
7891                         }
7892                         lun->res_type = type;
7893                         if (lun->res_type != SPR_TYPE_WR_EX_AR
7894                          && lun->res_type != SPR_TYPE_EX_AC_AR)
7895                                 lun->pr_res_idx = residx;
7896                         else
7897                                 lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
7898                         lun->PRGeneration++;
7899                         mtx_unlock(&lun->lun_lock);
7900
7901                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7902                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7903                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7904                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
7905                         persis_io.pr.pr_info.res_type = type;
7906                         memcpy(persis_io.pr.pr_info.sa_res_key,
7907                                param->serv_act_res_key,
7908                                sizeof(param->serv_act_res_key));
7909                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
7910                             sizeof(persis_io.pr), M_WAITOK);
7911                 } else {
7912                         /*
7913                          * sa_res_key is not the res holder just
7914                          * remove registrants
7915                          */
7916                         int found=0;
7917
7918                         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
7919                                 if (sa_res_key != ctl_get_prkey(lun, i))
7920                                         continue;
7921
7922                                 found = 1;
7923                                 ctl_clr_prkey(lun, i);
7924                                 lun->pr_key_count--;
7925                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7926                         }
7927
7928                         if (!found) {
7929                                 mtx_unlock(&lun->lun_lock);
7930                                 free(ctsio->kern_data_ptr, M_CTL);
7931                                 ctl_set_reservation_conflict(ctsio);
7932                                 ctl_done((union ctl_io *)ctsio);
7933                                 return (1);
7934                         }
7935                         lun->PRGeneration++;
7936                         mtx_unlock(&lun->lun_lock);
7937
7938                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7939                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7940                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7941                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
7942                         persis_io.pr.pr_info.res_type = type;
7943                         memcpy(persis_io.pr.pr_info.sa_res_key,
7944                                param->serv_act_res_key,
7945                                sizeof(param->serv_act_res_key));
7946                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
7947                             sizeof(persis_io.pr), M_WAITOK);
7948                 }
7949         }
7950         return (0);
7951 }
7952
7953 static void
7954 ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
7955 {
7956         uint64_t sa_res_key;
7957         int i;
7958
7959         sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
7960
7961         if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
7962          || lun->pr_res_idx == CTL_PR_NO_RESERVATION
7963          || sa_res_key != ctl_get_prkey(lun, lun->pr_res_idx)) {
7964                 if (sa_res_key == 0) {
7965                         /*
7966                          * Unregister everybody else and build UA for
7967                          * them
7968                          */
7969                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
7970                                 if (i == msg->pr.pr_info.residx ||
7971                                     ctl_get_prkey(lun, i) == 0)
7972                                         continue;
7973
7974                                 ctl_clr_prkey(lun, i);
7975                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7976                         }
7977
7978                         lun->pr_key_count = 1;
7979                         lun->res_type = msg->pr.pr_info.res_type;
7980                         if (lun->res_type != SPR_TYPE_WR_EX_AR
7981                          && lun->res_type != SPR_TYPE_EX_AC_AR)
7982                                 lun->pr_res_idx = msg->pr.pr_info.residx;
7983                 } else {
7984                         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
7985                                 if (sa_res_key == ctl_get_prkey(lun, i))
7986                                         continue;
7987
7988                                 ctl_clr_prkey(lun, i);
7989                                 lun->pr_key_count--;
7990                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7991                         }
7992                 }
7993         } else {
7994                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
7995                         if (i == msg->pr.pr_info.residx ||
7996                             ctl_get_prkey(lun, i) == 0)
7997                                 continue;
7998
7999                         if (sa_res_key == ctl_get_prkey(lun, i)) {
8000                                 ctl_clr_prkey(lun, i);
8001                                 lun->pr_key_count--;
8002                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8003                         } else if (msg->pr.pr_info.res_type != lun->res_type
8004                                 && (lun->res_type == SPR_TYPE_WR_EX_RO
8005                                  || lun->res_type == SPR_TYPE_EX_AC_RO)) {
8006                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8007                         }
8008                 }
8009                 lun->res_type = msg->pr.pr_info.res_type;
8010                 if (lun->res_type != SPR_TYPE_WR_EX_AR
8011                  && lun->res_type != SPR_TYPE_EX_AC_AR)
8012                         lun->pr_res_idx = msg->pr.pr_info.residx;
8013                 else
8014                         lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8015         }
8016         lun->PRGeneration++;
8017
8018 }
8019
8020
8021 int
8022 ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
8023 {
8024         int retval;
8025         u_int32_t param_len;
8026         struct scsi_per_res_out *cdb;
8027         struct ctl_lun *lun;
8028         struct scsi_per_res_out_parms* param;
8029         struct ctl_softc *softc;
8030         uint32_t residx;
8031         uint64_t res_key, sa_res_key, key;
8032         uint8_t type;
8033         union ctl_ha_msg persis_io;
8034         int    i;
8035
8036         CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n"));
8037
8038         retval = CTL_RETVAL_COMPLETE;
8039
8040         cdb = (struct scsi_per_res_out *)ctsio->cdb;
8041         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8042         softc = lun->ctl_softc;
8043
8044         /*
8045          * We only support whole-LUN scope.  The scope & type are ignored for
8046          * register, register and ignore existing key and clear.
8047          * We sometimes ignore scope and type on preempts too!!
8048          * Verify reservation type here as well.
8049          */
8050         type = cdb->scope_type & SPR_TYPE_MASK;
8051         if ((cdb->action == SPRO_RESERVE)
8052          || (cdb->action == SPRO_RELEASE)) {
8053                 if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
8054                         ctl_set_invalid_field(/*ctsio*/ ctsio,
8055                                               /*sks_valid*/ 1,
8056                                               /*command*/ 1,
8057                                               /*field*/ 2,
8058                                               /*bit_valid*/ 1,
8059                                               /*bit*/ 4);
8060                         ctl_done((union ctl_io *)ctsio);
8061                         return (CTL_RETVAL_COMPLETE);
8062                 }
8063
8064                 if (type>8 || type==2 || type==4 || type==0) {
8065                         ctl_set_invalid_field(/*ctsio*/ ctsio,
8066                                               /*sks_valid*/ 1,
8067                                               /*command*/ 1,
8068                                               /*field*/ 2,
8069                                               /*bit_valid*/ 1,
8070                                               /*bit*/ 0);
8071                         ctl_done((union ctl_io *)ctsio);
8072                         return (CTL_RETVAL_COMPLETE);
8073                 }
8074         }
8075
8076         param_len = scsi_4btoul(cdb->length);
8077
8078         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
8079                 ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
8080                 ctsio->kern_data_len = param_len;
8081                 ctsio->kern_total_len = param_len;
8082                 ctsio->kern_data_resid = 0;
8083                 ctsio->kern_rel_offset = 0;
8084                 ctsio->kern_sg_entries = 0;
8085                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
8086                 ctsio->be_move_done = ctl_config_move_done;
8087                 ctl_datamove((union ctl_io *)ctsio);
8088
8089                 return (CTL_RETVAL_COMPLETE);
8090         }
8091
8092         param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
8093
8094         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
8095         res_key = scsi_8btou64(param->res_key.key);
8096         sa_res_key = scsi_8btou64(param->serv_act_res_key);
8097
8098         /*
8099          * Validate the reservation key here except for SPRO_REG_IGNO
8100          * This must be done for all other service actions
8101          */
8102         if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
8103                 mtx_lock(&lun->lun_lock);
8104                 if ((key = ctl_get_prkey(lun, residx)) != 0) {
8105                         if (res_key != key) {
8106                                 /*
8107                                  * The current key passed in doesn't match
8108                                  * the one the initiator previously
8109                                  * registered.
8110                                  */
8111                                 mtx_unlock(&lun->lun_lock);
8112                                 free(ctsio->kern_data_ptr, M_CTL);
8113                                 ctl_set_reservation_conflict(ctsio);
8114                                 ctl_done((union ctl_io *)ctsio);
8115                                 return (CTL_RETVAL_COMPLETE);
8116                         }
8117                 } else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
8118                         /*
8119                          * We are not registered
8120                          */
8121                         mtx_unlock(&lun->lun_lock);
8122                         free(ctsio->kern_data_ptr, M_CTL);
8123                         ctl_set_reservation_conflict(ctsio);
8124                         ctl_done((union ctl_io *)ctsio);
8125                         return (CTL_RETVAL_COMPLETE);
8126                 } else if (res_key != 0) {
8127                         /*
8128                          * We are not registered and trying to register but
8129                          * the register key isn't zero.
8130                          */
8131                         mtx_unlock(&lun->lun_lock);
8132                         free(ctsio->kern_data_ptr, M_CTL);
8133                         ctl_set_reservation_conflict(ctsio);
8134                         ctl_done((union ctl_io *)ctsio);
8135                         return (CTL_RETVAL_COMPLETE);
8136                 }
8137                 mtx_unlock(&lun->lun_lock);
8138         }
8139
8140         switch (cdb->action & SPRO_ACTION_MASK) {
8141         case SPRO_REGISTER:
8142         case SPRO_REG_IGNO: {
8143
8144 #if 0
8145                 printf("Registration received\n");
8146 #endif
8147
8148                 /*
8149                  * We don't support any of these options, as we report in
8150                  * the read capabilities request (see
8151                  * ctl_persistent_reserve_in(), above).
8152                  */
8153                 if ((param->flags & SPR_SPEC_I_PT)
8154                  || (param->flags & SPR_ALL_TG_PT)
8155                  || (param->flags & SPR_APTPL)) {
8156                         int bit_ptr;
8157
8158                         if (param->flags & SPR_APTPL)
8159                                 bit_ptr = 0;
8160                         else if (param->flags & SPR_ALL_TG_PT)
8161                                 bit_ptr = 2;
8162                         else /* SPR_SPEC_I_PT */
8163                                 bit_ptr = 3;
8164
8165                         free(ctsio->kern_data_ptr, M_CTL);
8166                         ctl_set_invalid_field(ctsio,
8167                                               /*sks_valid*/ 1,
8168                                               /*command*/ 0,
8169                                               /*field*/ 20,
8170                                               /*bit_valid*/ 1,
8171                                               /*bit*/ bit_ptr);
8172                         ctl_done((union ctl_io *)ctsio);
8173                         return (CTL_RETVAL_COMPLETE);
8174                 }
8175
8176                 mtx_lock(&lun->lun_lock);
8177
8178                 /*
8179                  * The initiator wants to clear the
8180                  * key/unregister.
8181                  */
8182                 if (sa_res_key == 0) {
8183                         if ((res_key == 0
8184                           && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
8185                          || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
8186                           && ctl_get_prkey(lun, residx) == 0)) {
8187                                 mtx_unlock(&lun->lun_lock);
8188                                 goto done;
8189                         }
8190
8191                         ctl_clr_prkey(lun, residx);
8192                         lun->pr_key_count--;
8193
8194                         if (residx == lun->pr_res_idx) {
8195                                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8196                                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8197
8198                                 if ((lun->res_type == SPR_TYPE_WR_EX_RO
8199                                   || lun->res_type == SPR_TYPE_EX_AC_RO)
8200                                  && lun->pr_key_count) {
8201                                         /*
8202                                          * If the reservation is a registrants
8203                                          * only type we need to generate a UA
8204                                          * for other registered inits.  The
8205                                          * sense code should be RESERVATIONS
8206                                          * RELEASED
8207                                          */
8208
8209                                         for (i = softc->init_min; i < softc->init_max; i++){
8210                                                 if (ctl_get_prkey(lun, i) == 0)
8211                                                         continue;
8212                                                 ctl_est_ua(lun, i,
8213                                                     CTL_UA_RES_RELEASE);
8214                                         }
8215                                 }
8216                                 lun->res_type = 0;
8217                         } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8218                                 if (lun->pr_key_count==0) {
8219                                         lun->flags &= ~CTL_LUN_PR_RESERVED;
8220                                         lun->res_type = 0;
8221                                         lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8222                                 }
8223                         }
8224                         lun->PRGeneration++;
8225                         mtx_unlock(&lun->lun_lock);
8226
8227                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8228                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8229                         persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
8230                         persis_io.pr.pr_info.residx = residx;
8231                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8232                             sizeof(persis_io.pr), M_WAITOK);
8233                 } else /* sa_res_key != 0 */ {
8234
8235                         /*
8236                          * If we aren't registered currently then increment
8237                          * the key count and set the registered flag.
8238                          */
8239                         ctl_alloc_prkey(lun, residx);
8240                         if (ctl_get_prkey(lun, residx) == 0)
8241                                 lun->pr_key_count++;
8242                         ctl_set_prkey(lun, residx, sa_res_key);
8243                         lun->PRGeneration++;
8244                         mtx_unlock(&lun->lun_lock);
8245
8246                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8247                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8248                         persis_io.pr.pr_info.action = CTL_PR_REG_KEY;
8249                         persis_io.pr.pr_info.residx = residx;
8250                         memcpy(persis_io.pr.pr_info.sa_res_key,
8251                                param->serv_act_res_key,
8252                                sizeof(param->serv_act_res_key));
8253                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8254                             sizeof(persis_io.pr), M_WAITOK);
8255                 }
8256
8257                 break;
8258         }
8259         case SPRO_RESERVE:
8260 #if 0
8261                 printf("Reserve executed type %d\n", type);
8262 #endif
8263                 mtx_lock(&lun->lun_lock);
8264                 if (lun->flags & CTL_LUN_PR_RESERVED) {
8265                         /*
8266                          * if this isn't the reservation holder and it's
8267                          * not a "all registrants" type or if the type is
8268                          * different then we have a conflict
8269                          */
8270                         if ((lun->pr_res_idx != residx
8271                           && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8272                          || lun->res_type != type) {
8273                                 mtx_unlock(&lun->lun_lock);
8274                                 free(ctsio->kern_data_ptr, M_CTL);
8275                                 ctl_set_reservation_conflict(ctsio);
8276                                 ctl_done((union ctl_io *)ctsio);
8277                                 return (CTL_RETVAL_COMPLETE);
8278                         }
8279                         mtx_unlock(&lun->lun_lock);
8280                 } else /* create a reservation */ {
8281                         /*
8282                          * If it's not an "all registrants" type record
8283                          * reservation holder
8284                          */
8285                         if (type != SPR_TYPE_WR_EX_AR
8286                          && type != SPR_TYPE_EX_AC_AR)
8287                                 lun->pr_res_idx = residx; /* Res holder */
8288                         else
8289                                 lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8290
8291                         lun->flags |= CTL_LUN_PR_RESERVED;
8292                         lun->res_type = type;
8293
8294                         mtx_unlock(&lun->lun_lock);
8295
8296                         /* send msg to other side */
8297                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8298                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8299                         persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8300                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8301                         persis_io.pr.pr_info.res_type = type;
8302                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8303                             sizeof(persis_io.pr), M_WAITOK);
8304                 }
8305                 break;
8306
8307         case SPRO_RELEASE:
8308                 mtx_lock(&lun->lun_lock);
8309                 if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8310                         /* No reservation exists return good status */
8311                         mtx_unlock(&lun->lun_lock);
8312                         goto done;
8313                 }
8314                 /*
8315                  * Is this nexus a reservation holder?
8316                  */
8317                 if (lun->pr_res_idx != residx
8318                  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8319                         /*
8320                          * not a res holder return good status but
8321                          * do nothing
8322                          */
8323                         mtx_unlock(&lun->lun_lock);
8324                         goto done;
8325                 }
8326
8327                 if (lun->res_type != type) {
8328                         mtx_unlock(&lun->lun_lock);
8329                         free(ctsio->kern_data_ptr, M_CTL);
8330                         ctl_set_illegal_pr_release(ctsio);
8331                         ctl_done((union ctl_io *)ctsio);
8332                         return (CTL_RETVAL_COMPLETE);
8333                 }
8334
8335                 /* okay to release */
8336                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8337                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8338                 lun->res_type = 0;
8339
8340                 /*
8341                  * if this isn't an exclusive access
8342                  * res generate UA for all other
8343                  * registrants.
8344                  */
8345                 if (type != SPR_TYPE_EX_AC
8346                  && type != SPR_TYPE_WR_EX) {
8347                         for (i = softc->init_min; i < softc->init_max; i++) {
8348                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
8349                                         continue;
8350                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8351                         }
8352                 }
8353                 mtx_unlock(&lun->lun_lock);
8354
8355                 /* Send msg to other side */
8356                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8357                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8358                 persis_io.pr.pr_info.action = CTL_PR_RELEASE;
8359                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8360                      sizeof(persis_io.pr), M_WAITOK);
8361                 break;
8362
8363         case SPRO_CLEAR:
8364                 /* send msg to other side */
8365
8366                 mtx_lock(&lun->lun_lock);
8367                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8368                 lun->res_type = 0;
8369                 lun->pr_key_count = 0;
8370                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8371
8372                 ctl_clr_prkey(lun, residx);
8373                 for (i = 0; i < CTL_MAX_INITIATORS; i++)
8374                         if (ctl_get_prkey(lun, i) != 0) {
8375                                 ctl_clr_prkey(lun, i);
8376                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8377                         }
8378                 lun->PRGeneration++;
8379                 mtx_unlock(&lun->lun_lock);
8380
8381                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8382                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8383                 persis_io.pr.pr_info.action = CTL_PR_CLEAR;
8384                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8385                      sizeof(persis_io.pr), M_WAITOK);
8386                 break;
8387
8388         case SPRO_PREEMPT:
8389         case SPRO_PRE_ABO: {
8390                 int nretval;
8391
8392                 nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8393                                           residx, ctsio, cdb, param);
8394                 if (nretval != 0)
8395                         return (CTL_RETVAL_COMPLETE);
8396                 break;
8397         }
8398         default:
8399                 panic("Invalid PR type %x", cdb->action);
8400         }
8401
8402 done:
8403         free(ctsio->kern_data_ptr, M_CTL);
8404         ctl_set_success(ctsio);
8405         ctl_done((union ctl_io *)ctsio);
8406
8407         return (retval);
8408 }
8409
8410 /*
8411  * This routine is for handling a message from the other SC pertaining to
8412  * persistent reserve out. All the error checking will have been done
8413  * so only perorming the action need be done here to keep the two
8414  * in sync.
8415  */
8416 static void
8417 ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg)
8418 {
8419         struct ctl_lun *lun;
8420         struct ctl_softc *softc;
8421         int i;
8422         uint32_t residx, targ_lun;
8423
8424         softc = control_softc;
8425         targ_lun = msg->hdr.nexus.targ_mapped_lun;
8426         mtx_lock(&softc->ctl_lock);
8427         if ((targ_lun >= CTL_MAX_LUNS) ||
8428             ((lun = softc->ctl_luns[targ_lun]) == NULL)) {
8429                 mtx_unlock(&softc->ctl_lock);
8430                 return;
8431         }
8432         mtx_lock(&lun->lun_lock);
8433         mtx_unlock(&softc->ctl_lock);
8434         if (lun->flags & CTL_LUN_DISABLED) {
8435                 mtx_unlock(&lun->lun_lock);
8436                 return;
8437         }
8438         residx = ctl_get_initindex(&msg->hdr.nexus);
8439         switch(msg->pr.pr_info.action) {
8440         case CTL_PR_REG_KEY:
8441                 ctl_alloc_prkey(lun, msg->pr.pr_info.residx);
8442                 if (ctl_get_prkey(lun, msg->pr.pr_info.residx) == 0)
8443                         lun->pr_key_count++;
8444                 ctl_set_prkey(lun, msg->pr.pr_info.residx,
8445                     scsi_8btou64(msg->pr.pr_info.sa_res_key));
8446                 lun->PRGeneration++;
8447                 break;
8448
8449         case CTL_PR_UNREG_KEY:
8450                 ctl_clr_prkey(lun, msg->pr.pr_info.residx);
8451                 lun->pr_key_count--;
8452
8453                 /* XXX Need to see if the reservation has been released */
8454                 /* if so do we need to generate UA? */
8455                 if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8456                         lun->flags &= ~CTL_LUN_PR_RESERVED;
8457                         lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8458
8459                         if ((lun->res_type == SPR_TYPE_WR_EX_RO
8460                           || lun->res_type == SPR_TYPE_EX_AC_RO)
8461                          && lun->pr_key_count) {
8462                                 /*
8463                                  * If the reservation is a registrants
8464                                  * only type we need to generate a UA
8465                                  * for other registered inits.  The
8466                                  * sense code should be RESERVATIONS
8467                                  * RELEASED
8468                                  */
8469
8470                                 for (i = softc->init_min; i < softc->init_max; i++) {
8471                                         if (ctl_get_prkey(lun, i) == 0)
8472                                                 continue;
8473
8474                                         ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8475                                 }
8476                         }
8477                         lun->res_type = 0;
8478                 } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8479                         if (lun->pr_key_count==0) {
8480                                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8481                                 lun->res_type = 0;
8482                                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8483                         }
8484                 }
8485                 lun->PRGeneration++;
8486                 break;
8487
8488         case CTL_PR_RESERVE:
8489                 lun->flags |= CTL_LUN_PR_RESERVED;
8490                 lun->res_type = msg->pr.pr_info.res_type;
8491                 lun->pr_res_idx = msg->pr.pr_info.residx;
8492
8493                 break;
8494
8495         case CTL_PR_RELEASE:
8496                 /*
8497                  * if this isn't an exclusive access res generate UA for all
8498                  * other registrants.
8499                  */
8500                 if (lun->res_type != SPR_TYPE_EX_AC
8501                  && lun->res_type != SPR_TYPE_WR_EX) {
8502                         for (i = softc->init_min; i < softc->init_max; i++)
8503                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
8504                                         continue;
8505                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8506                 }
8507
8508                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8509                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8510                 lun->res_type = 0;
8511                 break;
8512
8513         case CTL_PR_PREEMPT:
8514                 ctl_pro_preempt_other(lun, msg);
8515                 break;
8516         case CTL_PR_CLEAR:
8517                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8518                 lun->res_type = 0;
8519                 lun->pr_key_count = 0;
8520                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8521
8522                 for (i=0; i < CTL_MAX_INITIATORS; i++) {
8523                         if (ctl_get_prkey(lun, i) == 0)
8524                                 continue;
8525                         ctl_clr_prkey(lun, i);
8526                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8527                 }
8528                 lun->PRGeneration++;
8529                 break;
8530         }
8531
8532         mtx_unlock(&lun->lun_lock);
8533 }
8534
8535 int
8536 ctl_read_write(struct ctl_scsiio *ctsio)
8537 {
8538         struct ctl_lun *lun;
8539         struct ctl_lba_len_flags *lbalen;
8540         uint64_t lba;
8541         uint32_t num_blocks;
8542         int flags, retval;
8543         int isread;
8544
8545         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8546
8547         CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
8548
8549         flags = 0;
8550         retval = CTL_RETVAL_COMPLETE;
8551
8552         isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
8553               || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
8554         switch (ctsio->cdb[0]) {
8555         case READ_6:
8556         case WRITE_6: {
8557                 struct scsi_rw_6 *cdb;
8558
8559                 cdb = (struct scsi_rw_6 *)ctsio->cdb;
8560
8561                 lba = scsi_3btoul(cdb->addr);
8562                 /* only 5 bits are valid in the most significant address byte */
8563                 lba &= 0x1fffff;
8564                 num_blocks = cdb->length;
8565                 /*
8566                  * This is correct according to SBC-2.
8567                  */
8568                 if (num_blocks == 0)
8569                         num_blocks = 256;
8570                 break;
8571         }
8572         case READ_10:
8573         case WRITE_10: {
8574                 struct scsi_rw_10 *cdb;
8575
8576                 cdb = (struct scsi_rw_10 *)ctsio->cdb;
8577                 if (cdb->byte2 & SRW10_FUA)
8578                         flags |= CTL_LLF_FUA;
8579                 if (cdb->byte2 & SRW10_DPO)
8580                         flags |= CTL_LLF_DPO;
8581                 lba = scsi_4btoul(cdb->addr);
8582                 num_blocks = scsi_2btoul(cdb->length);
8583                 break;
8584         }
8585         case WRITE_VERIFY_10: {
8586                 struct scsi_write_verify_10 *cdb;
8587
8588                 cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
8589                 flags |= CTL_LLF_FUA;
8590                 if (cdb->byte2 & SWV_DPO)
8591                         flags |= CTL_LLF_DPO;
8592                 lba = scsi_4btoul(cdb->addr);
8593                 num_blocks = scsi_2btoul(cdb->length);
8594                 break;
8595         }
8596         case READ_12:
8597         case WRITE_12: {
8598                 struct scsi_rw_12 *cdb;
8599
8600                 cdb = (struct scsi_rw_12 *)ctsio->cdb;
8601                 if (cdb->byte2 & SRW12_FUA)
8602                         flags |= CTL_LLF_FUA;
8603                 if (cdb->byte2 & SRW12_DPO)
8604                         flags |= CTL_LLF_DPO;
8605                 lba = scsi_4btoul(cdb->addr);
8606                 num_blocks = scsi_4btoul(cdb->length);
8607                 break;
8608         }
8609         case WRITE_VERIFY_12: {
8610                 struct scsi_write_verify_12 *cdb;
8611
8612                 cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
8613                 flags |= CTL_LLF_FUA;
8614                 if (cdb->byte2 & SWV_DPO)
8615                         flags |= CTL_LLF_DPO;
8616                 lba = scsi_4btoul(cdb->addr);
8617                 num_blocks = scsi_4btoul(cdb->length);
8618                 break;
8619         }
8620         case READ_16:
8621         case WRITE_16: {
8622                 struct scsi_rw_16 *cdb;
8623
8624                 cdb = (struct scsi_rw_16 *)ctsio->cdb;
8625                 if (cdb->byte2 & SRW12_FUA)
8626                         flags |= CTL_LLF_FUA;
8627                 if (cdb->byte2 & SRW12_DPO)
8628                         flags |= CTL_LLF_DPO;
8629                 lba = scsi_8btou64(cdb->addr);
8630                 num_blocks = scsi_4btoul(cdb->length);
8631                 break;
8632         }
8633         case WRITE_ATOMIC_16: {
8634                 struct scsi_rw_16 *cdb;
8635
8636                 if (lun->be_lun->atomicblock == 0) {
8637                         ctl_set_invalid_opcode(ctsio);
8638                         ctl_done((union ctl_io *)ctsio);
8639                         return (CTL_RETVAL_COMPLETE);
8640                 }
8641
8642                 cdb = (struct scsi_rw_16 *)ctsio->cdb;
8643                 if (cdb->byte2 & SRW12_FUA)
8644                         flags |= CTL_LLF_FUA;
8645                 if (cdb->byte2 & SRW12_DPO)
8646                         flags |= CTL_LLF_DPO;
8647                 lba = scsi_8btou64(cdb->addr);
8648                 num_blocks = scsi_4btoul(cdb->length);
8649                 if (num_blocks > lun->be_lun->atomicblock) {
8650                         ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
8651                             /*command*/ 1, /*field*/ 12, /*bit_valid*/ 0,
8652                             /*bit*/ 0);
8653                         ctl_done((union ctl_io *)ctsio);
8654                         return (CTL_RETVAL_COMPLETE);
8655                 }
8656                 break;
8657         }
8658         case WRITE_VERIFY_16: {
8659                 struct scsi_write_verify_16 *cdb;
8660
8661                 cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
8662                 flags |= CTL_LLF_FUA;
8663                 if (cdb->byte2 & SWV_DPO)
8664                         flags |= CTL_LLF_DPO;
8665                 lba = scsi_8btou64(cdb->addr);
8666                 num_blocks = scsi_4btoul(cdb->length);
8667                 break;
8668         }
8669         default:
8670                 /*
8671                  * We got a command we don't support.  This shouldn't
8672                  * happen, commands should be filtered out above us.
8673                  */
8674                 ctl_set_invalid_opcode(ctsio);
8675                 ctl_done((union ctl_io *)ctsio);
8676
8677                 return (CTL_RETVAL_COMPLETE);
8678                 break; /* NOTREACHED */
8679         }
8680
8681         /*
8682          * The first check is to make sure we're in bounds, the second
8683          * check is to catch wrap-around problems.  If the lba + num blocks
8684          * is less than the lba, then we've wrapped around and the block
8685          * range is invalid anyway.
8686          */
8687         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8688          || ((lba + num_blocks) < lba)) {
8689                 ctl_set_lba_out_of_range(ctsio);
8690                 ctl_done((union ctl_io *)ctsio);
8691                 return (CTL_RETVAL_COMPLETE);
8692         }
8693
8694         /*
8695          * According to SBC-3, a transfer length of 0 is not an error.
8696          * Note that this cannot happen with WRITE(6) or READ(6), since 0
8697          * translates to 256 blocks for those commands.
8698          */
8699         if (num_blocks == 0) {
8700                 ctl_set_success(ctsio);
8701                 ctl_done((union ctl_io *)ctsio);
8702                 return (CTL_RETVAL_COMPLETE);
8703         }
8704
8705         /* Set FUA and/or DPO if caches are disabled. */
8706         if (isread) {
8707                 if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
8708                     SCP_RCD) != 0)
8709                         flags |= CTL_LLF_FUA | CTL_LLF_DPO;
8710         } else {
8711                 if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
8712                     SCP_WCE) == 0)
8713                         flags |= CTL_LLF_FUA;
8714         }
8715
8716         lbalen = (struct ctl_lba_len_flags *)
8717             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8718         lbalen->lba = lba;
8719         lbalen->len = num_blocks;
8720         lbalen->flags = (isread ? CTL_LLF_READ : CTL_LLF_WRITE) | flags;
8721
8722         ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
8723         ctsio->kern_rel_offset = 0;
8724
8725         CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
8726
8727         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8728
8729         return (retval);
8730 }
8731
8732 static int
8733 ctl_cnw_cont(union ctl_io *io)
8734 {
8735         struct ctl_scsiio *ctsio;
8736         struct ctl_lun *lun;
8737         struct ctl_lba_len_flags *lbalen;
8738         int retval;
8739
8740         ctsio = &io->scsiio;
8741         ctsio->io_hdr.status = CTL_STATUS_NONE;
8742         ctsio->io_hdr.flags &= ~CTL_FLAG_IO_CONT;
8743         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8744         lbalen = (struct ctl_lba_len_flags *)
8745             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8746         lbalen->flags &= ~CTL_LLF_COMPARE;
8747         lbalen->flags |= CTL_LLF_WRITE;
8748
8749         CTL_DEBUG_PRINT(("ctl_cnw_cont: calling data_submit()\n"));
8750         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8751         return (retval);
8752 }
8753
8754 int
8755 ctl_cnw(struct ctl_scsiio *ctsio)
8756 {
8757         struct ctl_lun *lun;
8758         struct ctl_lba_len_flags *lbalen;
8759         uint64_t lba;
8760         uint32_t num_blocks;
8761         int flags, retval;
8762
8763         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8764
8765         CTL_DEBUG_PRINT(("ctl_cnw: command: %#x\n", ctsio->cdb[0]));
8766
8767         flags = 0;
8768         retval = CTL_RETVAL_COMPLETE;
8769
8770         switch (ctsio->cdb[0]) {
8771         case COMPARE_AND_WRITE: {
8772                 struct scsi_compare_and_write *cdb;
8773
8774                 cdb = (struct scsi_compare_and_write *)ctsio->cdb;
8775                 if (cdb->byte2 & SRW10_FUA)
8776                         flags |= CTL_LLF_FUA;
8777                 if (cdb->byte2 & SRW10_DPO)
8778                         flags |= CTL_LLF_DPO;
8779                 lba = scsi_8btou64(cdb->addr);
8780                 num_blocks = cdb->length;
8781                 break;
8782         }
8783         default:
8784                 /*
8785                  * We got a command we don't support.  This shouldn't
8786                  * happen, commands should be filtered out above us.
8787                  */
8788                 ctl_set_invalid_opcode(ctsio);
8789                 ctl_done((union ctl_io *)ctsio);
8790
8791                 return (CTL_RETVAL_COMPLETE);
8792                 break; /* NOTREACHED */
8793         }
8794
8795         /*
8796          * The first check is to make sure we're in bounds, the second
8797          * check is to catch wrap-around problems.  If the lba + num blocks
8798          * is less than the lba, then we've wrapped around and the block
8799          * range is invalid anyway.
8800          */
8801         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8802          || ((lba + num_blocks) < lba)) {
8803                 ctl_set_lba_out_of_range(ctsio);
8804                 ctl_done((union ctl_io *)ctsio);
8805                 return (CTL_RETVAL_COMPLETE);
8806         }
8807
8808         /*
8809          * According to SBC-3, a transfer length of 0 is not an error.
8810          */
8811         if (num_blocks == 0) {
8812                 ctl_set_success(ctsio);
8813                 ctl_done((union ctl_io *)ctsio);
8814                 return (CTL_RETVAL_COMPLETE);
8815         }
8816
8817         /* Set FUA if write cache is disabled. */
8818         if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
8819             SCP_WCE) == 0)
8820                 flags |= CTL_LLF_FUA;
8821
8822         ctsio->kern_total_len = 2 * num_blocks * lun->be_lun->blocksize;
8823         ctsio->kern_rel_offset = 0;
8824
8825         /*
8826          * Set the IO_CONT flag, so that if this I/O gets passed to
8827          * ctl_data_submit_done(), it'll get passed back to
8828          * ctl_ctl_cnw_cont() for further processing.
8829          */
8830         ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
8831         ctsio->io_cont = ctl_cnw_cont;
8832
8833         lbalen = (struct ctl_lba_len_flags *)
8834             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8835         lbalen->lba = lba;
8836         lbalen->len = num_blocks;
8837         lbalen->flags = CTL_LLF_COMPARE | flags;
8838
8839         CTL_DEBUG_PRINT(("ctl_cnw: calling data_submit()\n"));
8840         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8841         return (retval);
8842 }
8843
8844 int
8845 ctl_verify(struct ctl_scsiio *ctsio)
8846 {
8847         struct ctl_lun *lun;
8848         struct ctl_lba_len_flags *lbalen;
8849         uint64_t lba;
8850         uint32_t num_blocks;
8851         int bytchk, flags;
8852         int retval;
8853
8854         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8855
8856         CTL_DEBUG_PRINT(("ctl_verify: command: %#x\n", ctsio->cdb[0]));
8857
8858         bytchk = 0;
8859         flags = CTL_LLF_FUA;
8860         retval = CTL_RETVAL_COMPLETE;
8861
8862         switch (ctsio->cdb[0]) {
8863         case VERIFY_10: {
8864                 struct scsi_verify_10 *cdb;
8865
8866                 cdb = (struct scsi_verify_10 *)ctsio->cdb;
8867                 if (cdb->byte2 & SVFY_BYTCHK)
8868                         bytchk = 1;
8869                 if (cdb->byte2 & SVFY_DPO)
8870                         flags |= CTL_LLF_DPO;
8871                 lba = scsi_4btoul(cdb->addr);
8872                 num_blocks = scsi_2btoul(cdb->length);
8873                 break;
8874         }
8875         case VERIFY_12: {
8876                 struct scsi_verify_12 *cdb;
8877
8878                 cdb = (struct scsi_verify_12 *)ctsio->cdb;
8879                 if (cdb->byte2 & SVFY_BYTCHK)
8880                         bytchk = 1;
8881                 if (cdb->byte2 & SVFY_DPO)
8882                         flags |= CTL_LLF_DPO;
8883                 lba = scsi_4btoul(cdb->addr);
8884                 num_blocks = scsi_4btoul(cdb->length);
8885                 break;
8886         }
8887         case VERIFY_16: {
8888                 struct scsi_rw_16 *cdb;
8889
8890                 cdb = (struct scsi_rw_16 *)ctsio->cdb;
8891                 if (cdb->byte2 & SVFY_BYTCHK)
8892                         bytchk = 1;
8893                 if (cdb->byte2 & SVFY_DPO)
8894                         flags |= CTL_LLF_DPO;
8895                 lba = scsi_8btou64(cdb->addr);
8896                 num_blocks = scsi_4btoul(cdb->length);
8897                 break;
8898         }
8899         default:
8900                 /*
8901                  * We got a command we don't support.  This shouldn't
8902                  * happen, commands should be filtered out above us.
8903                  */
8904                 ctl_set_invalid_opcode(ctsio);
8905                 ctl_done((union ctl_io *)ctsio);
8906                 return (CTL_RETVAL_COMPLETE);
8907         }
8908
8909         /*
8910          * The first check is to make sure we're in bounds, the second
8911          * check is to catch wrap-around problems.  If the lba + num blocks
8912          * is less than the lba, then we've wrapped around and the block
8913          * range is invalid anyway.
8914          */
8915         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8916          || ((lba + num_blocks) < lba)) {
8917                 ctl_set_lba_out_of_range(ctsio);
8918                 ctl_done((union ctl_io *)ctsio);
8919                 return (CTL_RETVAL_COMPLETE);
8920         }
8921
8922         /*
8923          * According to SBC-3, a transfer length of 0 is not an error.
8924          */
8925         if (num_blocks == 0) {
8926                 ctl_set_success(ctsio);
8927                 ctl_done((union ctl_io *)ctsio);
8928                 return (CTL_RETVAL_COMPLETE);
8929         }
8930
8931         lbalen = (struct ctl_lba_len_flags *)
8932             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8933         lbalen->lba = lba;
8934         lbalen->len = num_blocks;
8935         if (bytchk) {
8936                 lbalen->flags = CTL_LLF_COMPARE | flags;
8937                 ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
8938         } else {
8939                 lbalen->flags = CTL_LLF_VERIFY | flags;
8940                 ctsio->kern_total_len = 0;
8941         }
8942         ctsio->kern_rel_offset = 0;
8943
8944         CTL_DEBUG_PRINT(("ctl_verify: calling data_submit()\n"));
8945         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8946         return (retval);
8947 }
8948
8949 int
8950 ctl_report_luns(struct ctl_scsiio *ctsio)
8951 {
8952         struct ctl_softc *softc = control_softc;
8953         struct scsi_report_luns *cdb;
8954         struct scsi_report_luns_data *lun_data;
8955         struct ctl_lun *lun, *request_lun;
8956         struct ctl_port *port;
8957         int num_luns, retval;
8958         uint32_t alloc_len, lun_datalen;
8959         int num_filled, well_known;
8960         uint32_t initidx, targ_lun_id, lun_id;
8961
8962         retval = CTL_RETVAL_COMPLETE;
8963         well_known = 0;
8964
8965         cdb = (struct scsi_report_luns *)ctsio->cdb;
8966         port = ctl_io_port(&ctsio->io_hdr);
8967
8968         CTL_DEBUG_PRINT(("ctl_report_luns\n"));
8969
8970         mtx_lock(&softc->ctl_lock);
8971         num_luns = 0;
8972         for (targ_lun_id = 0; targ_lun_id < CTL_MAX_LUNS; targ_lun_id++) {
8973                 if (ctl_lun_map_from_port(port, targ_lun_id) < CTL_MAX_LUNS)
8974                         num_luns++;
8975         }
8976         mtx_unlock(&softc->ctl_lock);
8977
8978         switch (cdb->select_report) {
8979         case RPL_REPORT_DEFAULT:
8980         case RPL_REPORT_ALL:
8981                 break;
8982         case RPL_REPORT_WELLKNOWN:
8983                 well_known = 1;
8984                 num_luns = 0;
8985                 break;
8986         default:
8987                 ctl_set_invalid_field(ctsio,
8988                                       /*sks_valid*/ 1,
8989                                       /*command*/ 1,
8990                                       /*field*/ 2,
8991                                       /*bit_valid*/ 0,
8992                                       /*bit*/ 0);
8993                 ctl_done((union ctl_io *)ctsio);
8994                 return (retval);
8995                 break; /* NOTREACHED */
8996         }
8997
8998         alloc_len = scsi_4btoul(cdb->length);
8999         /*
9000          * The initiator has to allocate at least 16 bytes for this request,
9001          * so he can at least get the header and the first LUN.  Otherwise
9002          * we reject the request (per SPC-3 rev 14, section 6.21).
9003          */
9004         if (alloc_len < (sizeof(struct scsi_report_luns_data) +
9005             sizeof(struct scsi_report_luns_lundata))) {
9006                 ctl_set_invalid_field(ctsio,
9007                                       /*sks_valid*/ 1,
9008                                       /*command*/ 1,
9009                                       /*field*/ 6,
9010                                       /*bit_valid*/ 0,
9011                                       /*bit*/ 0);
9012                 ctl_done((union ctl_io *)ctsio);
9013                 return (retval);
9014         }
9015
9016         request_lun = (struct ctl_lun *)
9017                 ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9018
9019         lun_datalen = sizeof(*lun_data) +
9020                 (num_luns * sizeof(struct scsi_report_luns_lundata));
9021
9022         ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO);
9023         lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
9024         ctsio->kern_sg_entries = 0;
9025
9026         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9027
9028         mtx_lock(&softc->ctl_lock);
9029         for (targ_lun_id = 0, num_filled = 0; targ_lun_id < CTL_MAX_LUNS && num_filled < num_luns; targ_lun_id++) {
9030                 lun_id = ctl_lun_map_from_port(port, targ_lun_id);
9031                 if (lun_id >= CTL_MAX_LUNS)
9032                         continue;
9033                 lun = softc->ctl_luns[lun_id];
9034                 if (lun == NULL)
9035                         continue;
9036
9037                 if (targ_lun_id <= 0xff) {
9038                         /*
9039                          * Peripheral addressing method, bus number 0.
9040                          */
9041                         lun_data->luns[num_filled].lundata[0] =
9042                                 RPL_LUNDATA_ATYP_PERIPH;
9043                         lun_data->luns[num_filled].lundata[1] = targ_lun_id;
9044                         num_filled++;
9045                 } else if (targ_lun_id <= 0x3fff) {
9046                         /*
9047                          * Flat addressing method.
9048                          */
9049                         lun_data->luns[num_filled].lundata[0] =
9050                                 RPL_LUNDATA_ATYP_FLAT | (targ_lun_id >> 8);
9051                         lun_data->luns[num_filled].lundata[1] =
9052                                 (targ_lun_id & 0xff);
9053                         num_filled++;
9054                 } else if (targ_lun_id <= 0xffffff) {
9055                         /*
9056                          * Extended flat addressing method.
9057                          */
9058                         lun_data->luns[num_filled].lundata[0] =
9059                             RPL_LUNDATA_ATYP_EXTLUN | 0x12;
9060                         scsi_ulto3b(targ_lun_id,
9061                             &lun_data->luns[num_filled].lundata[1]);
9062                         num_filled++;
9063                 } else {
9064                         printf("ctl_report_luns: bogus LUN number %jd, "
9065                                "skipping\n", (intmax_t)targ_lun_id);
9066                 }
9067                 /*
9068                  * According to SPC-3, rev 14 section 6.21:
9069                  *
9070                  * "The execution of a REPORT LUNS command to any valid and
9071                  * installed logical unit shall clear the REPORTED LUNS DATA
9072                  * HAS CHANGED unit attention condition for all logical
9073                  * units of that target with respect to the requesting
9074                  * initiator. A valid and installed logical unit is one
9075                  * having a PERIPHERAL QUALIFIER of 000b in the standard
9076                  * INQUIRY data (see 6.4.2)."
9077                  *
9078                  * If request_lun is NULL, the LUN this report luns command
9079                  * was issued to is either disabled or doesn't exist. In that
9080                  * case, we shouldn't clear any pending lun change unit
9081                  * attention.
9082                  */
9083                 if (request_lun != NULL) {
9084                         mtx_lock(&lun->lun_lock);
9085                         ctl_clr_ua(lun, initidx, CTL_UA_LUN_CHANGE);
9086                         mtx_unlock(&lun->lun_lock);
9087                 }
9088         }
9089         mtx_unlock(&softc->ctl_lock);
9090
9091         /*
9092          * It's quite possible that we've returned fewer LUNs than we allocated
9093          * space for.  Trim it.
9094          */
9095         lun_datalen = sizeof(*lun_data) +
9096                 (num_filled * sizeof(struct scsi_report_luns_lundata));
9097
9098         if (lun_datalen < alloc_len) {
9099                 ctsio->residual = alloc_len - lun_datalen;
9100                 ctsio->kern_data_len = lun_datalen;
9101                 ctsio->kern_total_len = lun_datalen;
9102         } else {
9103                 ctsio->residual = 0;
9104                 ctsio->kern_data_len = alloc_len;
9105                 ctsio->kern_total_len = alloc_len;
9106         }
9107         ctsio->kern_data_resid = 0;
9108         ctsio->kern_rel_offset = 0;
9109         ctsio->kern_sg_entries = 0;
9110
9111         /*
9112          * We set this to the actual data length, regardless of how much
9113          * space we actually have to return results.  If the user looks at
9114          * this value, he'll know whether or not he allocated enough space
9115          * and reissue the command if necessary.  We don't support well
9116          * known logical units, so if the user asks for that, return none.
9117          */
9118         scsi_ulto4b(lun_datalen - 8, lun_data->length);
9119
9120         /*
9121          * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
9122          * this request.
9123          */
9124         ctl_set_success(ctsio);
9125         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9126         ctsio->be_move_done = ctl_config_move_done;
9127         ctl_datamove((union ctl_io *)ctsio);
9128         return (retval);
9129 }
9130
9131 int
9132 ctl_request_sense(struct ctl_scsiio *ctsio)
9133 {
9134         struct scsi_request_sense *cdb;
9135         struct scsi_sense_data *sense_ptr;
9136         struct ctl_softc *ctl_softc;
9137         struct ctl_lun *lun;
9138         uint32_t initidx;
9139         int have_error;
9140         scsi_sense_data_type sense_format;
9141         ctl_ua_type ua_type;
9142
9143         cdb = (struct scsi_request_sense *)ctsio->cdb;
9144
9145         ctl_softc = control_softc;
9146         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9147
9148         CTL_DEBUG_PRINT(("ctl_request_sense\n"));
9149
9150         /*
9151          * Determine which sense format the user wants.
9152          */
9153         if (cdb->byte2 & SRS_DESC)
9154                 sense_format = SSD_TYPE_DESC;
9155         else
9156                 sense_format = SSD_TYPE_FIXED;
9157
9158         ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
9159         sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
9160         ctsio->kern_sg_entries = 0;
9161
9162         /*
9163          * struct scsi_sense_data, which is currently set to 256 bytes, is
9164          * larger than the largest allowed value for the length field in the
9165          * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
9166          */
9167         ctsio->residual = 0;
9168         ctsio->kern_data_len = cdb->length;
9169         ctsio->kern_total_len = cdb->length;
9170
9171         ctsio->kern_data_resid = 0;
9172         ctsio->kern_rel_offset = 0;
9173         ctsio->kern_sg_entries = 0;
9174
9175         /*
9176          * If we don't have a LUN, we don't have any pending sense.
9177          */
9178         if (lun == NULL)
9179                 goto no_sense;
9180
9181         have_error = 0;
9182         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9183         /*
9184          * Check for pending sense, and then for pending unit attentions.
9185          * Pending sense gets returned first, then pending unit attentions.
9186          */
9187         mtx_lock(&lun->lun_lock);
9188 #ifdef CTL_WITH_CA
9189         if (ctl_is_set(lun->have_ca, initidx)) {
9190                 scsi_sense_data_type stored_format;
9191
9192                 /*
9193                  * Check to see which sense format was used for the stored
9194                  * sense data.
9195                  */
9196                 stored_format = scsi_sense_type(&lun->pending_sense[initidx]);
9197
9198                 /*
9199                  * If the user requested a different sense format than the
9200                  * one we stored, then we need to convert it to the other
9201                  * format.  If we're going from descriptor to fixed format
9202                  * sense data, we may lose things in translation, depending
9203                  * on what options were used.
9204                  *
9205                  * If the stored format is SSD_TYPE_NONE (i.e. invalid),
9206                  * for some reason we'll just copy it out as-is.
9207                  */
9208                 if ((stored_format == SSD_TYPE_FIXED)
9209                  && (sense_format == SSD_TYPE_DESC))
9210                         ctl_sense_to_desc((struct scsi_sense_data_fixed *)
9211                             &lun->pending_sense[initidx],
9212                             (struct scsi_sense_data_desc *)sense_ptr);
9213                 else if ((stored_format == SSD_TYPE_DESC)
9214                       && (sense_format == SSD_TYPE_FIXED))
9215                         ctl_sense_to_fixed((struct scsi_sense_data_desc *)
9216                             &lun->pending_sense[initidx],
9217                             (struct scsi_sense_data_fixed *)sense_ptr);
9218                 else
9219                         memcpy(sense_ptr, &lun->pending_sense[initidx],
9220                                MIN(sizeof(*sense_ptr),
9221                                sizeof(lun->pending_sense[initidx])));
9222
9223                 ctl_clear_mask(lun->have_ca, initidx);
9224                 have_error = 1;
9225         } else
9226 #endif
9227         {
9228                 ua_type = ctl_build_ua(lun, initidx, sense_ptr, sense_format);
9229                 if (ua_type != CTL_UA_NONE)
9230                         have_error = 1;
9231                 if (ua_type == CTL_UA_LUN_CHANGE) {
9232                         mtx_unlock(&lun->lun_lock);
9233                         mtx_lock(&ctl_softc->ctl_lock);
9234                         ctl_clr_ua_allluns(ctl_softc, initidx, ua_type);
9235                         mtx_unlock(&ctl_softc->ctl_lock);
9236                         mtx_lock(&lun->lun_lock);
9237                 }
9238
9239         }
9240         mtx_unlock(&lun->lun_lock);
9241
9242         /*
9243          * We already have a pending error, return it.
9244          */
9245         if (have_error != 0) {
9246                 /*
9247                  * We report the SCSI status as OK, since the status of the
9248                  * request sense command itself is OK.
9249                  * We report 0 for the sense length, because we aren't doing
9250                  * autosense in this case.  We're reporting sense as
9251                  * parameter data.
9252                  */
9253                 ctl_set_success(ctsio);
9254                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9255                 ctsio->be_move_done = ctl_config_move_done;
9256                 ctl_datamove((union ctl_io *)ctsio);
9257                 return (CTL_RETVAL_COMPLETE);
9258         }
9259
9260 no_sense:
9261
9262         /*
9263          * No sense information to report, so we report that everything is
9264          * okay.
9265          */
9266         ctl_set_sense_data(sense_ptr,
9267                            lun,
9268                            sense_format,
9269                            /*current_error*/ 1,
9270                            /*sense_key*/ SSD_KEY_NO_SENSE,
9271                            /*asc*/ 0x00,
9272                            /*ascq*/ 0x00,
9273                            SSD_ELEM_NONE);
9274
9275         /*
9276          * We report 0 for the sense length, because we aren't doing
9277          * autosense in this case.  We're reporting sense as parameter data.
9278          */
9279         ctl_set_success(ctsio);
9280         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9281         ctsio->be_move_done = ctl_config_move_done;
9282         ctl_datamove((union ctl_io *)ctsio);
9283         return (CTL_RETVAL_COMPLETE);
9284 }
9285
9286 int
9287 ctl_tur(struct ctl_scsiio *ctsio)
9288 {
9289
9290         CTL_DEBUG_PRINT(("ctl_tur\n"));
9291
9292         ctl_set_success(ctsio);
9293         ctl_done((union ctl_io *)ctsio);
9294
9295         return (CTL_RETVAL_COMPLETE);
9296 }
9297
9298 /*
9299  * SCSI VPD page 0x00, the Supported VPD Pages page.
9300  */
9301 static int
9302 ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
9303 {
9304         struct scsi_vpd_supported_pages *pages;
9305         int sup_page_size;
9306         struct ctl_lun *lun;
9307         int p;
9308
9309         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9310
9311         sup_page_size = sizeof(struct scsi_vpd_supported_pages) *
9312             SCSI_EVPD_NUM_SUPPORTED_PAGES;
9313         ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO);
9314         pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
9315         ctsio->kern_sg_entries = 0;
9316
9317         if (sup_page_size < alloc_len) {
9318                 ctsio->residual = alloc_len - sup_page_size;
9319                 ctsio->kern_data_len = sup_page_size;
9320                 ctsio->kern_total_len = sup_page_size;
9321         } else {
9322                 ctsio->residual = 0;
9323                 ctsio->kern_data_len = alloc_len;
9324                 ctsio->kern_total_len = alloc_len;
9325         }
9326         ctsio->kern_data_resid = 0;
9327         ctsio->kern_rel_offset = 0;
9328         ctsio->kern_sg_entries = 0;
9329
9330         /*
9331          * The control device is always connected.  The disk device, on the
9332          * other hand, may not be online all the time.  Need to change this
9333          * to figure out whether the disk device is actually online or not.
9334          */
9335         if (lun != NULL)
9336                 pages->device = (SID_QUAL_LU_CONNECTED << 5) |
9337                                 lun->be_lun->lun_type;
9338         else
9339                 pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9340
9341         p = 0;
9342         /* Supported VPD pages */
9343         pages->page_list[p++] = SVPD_SUPPORTED_PAGES;
9344         /* Serial Number */
9345         pages->page_list[p++] = SVPD_UNIT_SERIAL_NUMBER;
9346         /* Device Identification */
9347         pages->page_list[p++] = SVPD_DEVICE_ID;
9348         /* Extended INQUIRY Data */
9349         pages->page_list[p++] = SVPD_EXTENDED_INQUIRY_DATA;
9350         /* Mode Page Policy */
9351         pages->page_list[p++] = SVPD_MODE_PAGE_POLICY;
9352         /* SCSI Ports */
9353         pages->page_list[p++] = SVPD_SCSI_PORTS;
9354         /* Third-party Copy */
9355         pages->page_list[p++] = SVPD_SCSI_TPC;
9356         if (lun != NULL && lun->be_lun->lun_type == T_DIRECT) {
9357                 /* Block limits */
9358                 pages->page_list[p++] = SVPD_BLOCK_LIMITS;
9359                 /* Block Device Characteristics */
9360                 pages->page_list[p++] = SVPD_BDC;
9361                 /* Logical Block Provisioning */
9362                 pages->page_list[p++] = SVPD_LBP;
9363         }
9364         pages->length = p;
9365
9366         ctl_set_success(ctsio);
9367         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9368         ctsio->be_move_done = ctl_config_move_done;
9369         ctl_datamove((union ctl_io *)ctsio);
9370         return (CTL_RETVAL_COMPLETE);
9371 }
9372
9373 /*
9374  * SCSI VPD page 0x80, the Unit Serial Number page.
9375  */
9376 static int
9377 ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
9378 {
9379         struct scsi_vpd_unit_serial_number *sn_ptr;
9380         struct ctl_lun *lun;
9381         int data_len;
9382
9383         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9384
9385         data_len = 4 + CTL_SN_LEN;
9386         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9387         sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
9388         if (data_len < alloc_len) {
9389                 ctsio->residual = alloc_len - data_len;
9390                 ctsio->kern_data_len = data_len;
9391                 ctsio->kern_total_len = data_len;
9392         } else {
9393                 ctsio->residual = 0;
9394                 ctsio->kern_data_len = alloc_len;
9395                 ctsio->kern_total_len = alloc_len;
9396         }
9397         ctsio->kern_data_resid = 0;
9398         ctsio->kern_rel_offset = 0;
9399         ctsio->kern_sg_entries = 0;
9400
9401         /*
9402          * The control device is always connected.  The disk device, on the
9403          * other hand, may not be online all the time.  Need to change this
9404          * to figure out whether the disk device is actually online or not.
9405          */
9406         if (lun != NULL)
9407                 sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9408                                   lun->be_lun->lun_type;
9409         else
9410                 sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9411
9412         sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9413         sn_ptr->length = CTL_SN_LEN;
9414         /*
9415          * If we don't have a LUN, we just leave the serial number as
9416          * all spaces.
9417          */
9418         if (lun != NULL) {
9419                 strncpy((char *)sn_ptr->serial_num,
9420                         (char *)lun->be_lun->serial_num, CTL_SN_LEN);
9421         } else
9422                 memset(sn_ptr->serial_num, 0x20, CTL_SN_LEN);
9423
9424         ctl_set_success(ctsio);
9425         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9426         ctsio->be_move_done = ctl_config_move_done;
9427         ctl_datamove((union ctl_io *)ctsio);
9428         return (CTL_RETVAL_COMPLETE);
9429 }
9430
9431
9432 /*
9433  * SCSI VPD page 0x86, the Extended INQUIRY Data page.
9434  */
9435 static int
9436 ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len)
9437 {
9438         struct scsi_vpd_extended_inquiry_data *eid_ptr;
9439         struct ctl_lun *lun;
9440         int data_len;
9441
9442         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9443
9444         data_len = sizeof(struct scsi_vpd_extended_inquiry_data);
9445         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9446         eid_ptr = (struct scsi_vpd_extended_inquiry_data *)ctsio->kern_data_ptr;
9447         ctsio->kern_sg_entries = 0;
9448
9449         if (data_len < alloc_len) {
9450                 ctsio->residual = alloc_len - data_len;
9451                 ctsio->kern_data_len = data_len;
9452                 ctsio->kern_total_len = data_len;
9453         } else {
9454                 ctsio->residual = 0;
9455                 ctsio->kern_data_len = alloc_len;
9456                 ctsio->kern_total_len = alloc_len;
9457         }
9458         ctsio->kern_data_resid = 0;
9459         ctsio->kern_rel_offset = 0;
9460         ctsio->kern_sg_entries = 0;
9461
9462         /*
9463          * The control device is always connected.  The disk device, on the
9464          * other hand, may not be online all the time.
9465          */
9466         if (lun != NULL)
9467                 eid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9468                                      lun->be_lun->lun_type;
9469         else
9470                 eid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9471         eid_ptr->page_code = SVPD_EXTENDED_INQUIRY_DATA;
9472         scsi_ulto2b(data_len - 4, eid_ptr->page_length);
9473         /*
9474          * We support head of queue, ordered and simple tags.
9475          */
9476         eid_ptr->flags2 = SVPD_EID_HEADSUP | SVPD_EID_ORDSUP | SVPD_EID_SIMPSUP;
9477         /*
9478          * Volatile cache supported.
9479          */
9480         eid_ptr->flags3 = SVPD_EID_V_SUP;
9481
9482         /*
9483          * This means that we clear the REPORTED LUNS DATA HAS CHANGED unit
9484          * attention for a particular IT nexus on all LUNs once we report
9485          * it to that nexus once.  This bit is required as of SPC-4.
9486          */
9487         eid_ptr->flags4 = SVPD_EID_LUICLT;
9488
9489         /*
9490          * XXX KDM in order to correctly answer this, we would need
9491          * information from the SIM to determine how much sense data it
9492          * can send.  So this would really be a path inquiry field, most
9493          * likely.  This can be set to a maximum of 252 according to SPC-4,
9494          * but the hardware may or may not be able to support that much.
9495          * 0 just means that the maximum sense data length is not reported.
9496          */
9497         eid_ptr->max_sense_length = 0;
9498
9499         ctl_set_success(ctsio);
9500         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9501         ctsio->be_move_done = ctl_config_move_done;
9502         ctl_datamove((union ctl_io *)ctsio);
9503         return (CTL_RETVAL_COMPLETE);
9504 }
9505
9506 static int
9507 ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len)
9508 {
9509         struct scsi_vpd_mode_page_policy *mpp_ptr;
9510         struct ctl_lun *lun;
9511         int data_len;
9512
9513         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9514
9515         data_len = sizeof(struct scsi_vpd_mode_page_policy) +
9516             sizeof(struct scsi_vpd_mode_page_policy_descr);
9517
9518         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9519         mpp_ptr = (struct scsi_vpd_mode_page_policy *)ctsio->kern_data_ptr;
9520         ctsio->kern_sg_entries = 0;
9521
9522         if (data_len < alloc_len) {
9523                 ctsio->residual = alloc_len - data_len;
9524                 ctsio->kern_data_len = data_len;
9525                 ctsio->kern_total_len = data_len;
9526         } else {
9527                 ctsio->residual = 0;
9528                 ctsio->kern_data_len = alloc_len;
9529                 ctsio->kern_total_len = alloc_len;
9530         }
9531         ctsio->kern_data_resid = 0;
9532         ctsio->kern_rel_offset = 0;
9533         ctsio->kern_sg_entries = 0;
9534
9535         /*
9536          * The control device is always connected.  The disk device, on the
9537          * other hand, may not be online all the time.
9538          */
9539         if (lun != NULL)
9540                 mpp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9541                                      lun->be_lun->lun_type;
9542         else
9543                 mpp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9544         mpp_ptr->page_code = SVPD_MODE_PAGE_POLICY;
9545         scsi_ulto2b(data_len - 4, mpp_ptr->page_length);
9546         mpp_ptr->descr[0].page_code = 0x3f;
9547         mpp_ptr->descr[0].subpage_code = 0xff;
9548         mpp_ptr->descr[0].policy = SVPD_MPP_SHARED;
9549
9550         ctl_set_success(ctsio);
9551         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9552         ctsio->be_move_done = ctl_config_move_done;
9553         ctl_datamove((union ctl_io *)ctsio);
9554         return (CTL_RETVAL_COMPLETE);
9555 }
9556
9557 /*
9558  * SCSI VPD page 0x83, the Device Identification page.
9559  */
9560 static int
9561 ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
9562 {
9563         struct scsi_vpd_device_id *devid_ptr;
9564         struct scsi_vpd_id_descriptor *desc;
9565         struct ctl_softc *softc;
9566         struct ctl_lun *lun;
9567         struct ctl_port *port;
9568         int data_len;
9569         uint8_t proto;
9570
9571         softc = control_softc;
9572
9573         port = ctl_io_port(&ctsio->io_hdr);
9574         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9575
9576         data_len = sizeof(struct scsi_vpd_device_id) +
9577             sizeof(struct scsi_vpd_id_descriptor) +
9578                 sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
9579             sizeof(struct scsi_vpd_id_descriptor) +
9580                 sizeof(struct scsi_vpd_id_trgt_port_grp_id);
9581         if (lun && lun->lun_devid)
9582                 data_len += lun->lun_devid->len;
9583         if (port && port->port_devid)
9584                 data_len += port->port_devid->len;
9585         if (port && port->target_devid)
9586                 data_len += port->target_devid->len;
9587
9588         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9589         devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
9590         ctsio->kern_sg_entries = 0;
9591
9592         if (data_len < alloc_len) {
9593                 ctsio->residual = alloc_len - data_len;
9594                 ctsio->kern_data_len = data_len;
9595                 ctsio->kern_total_len = data_len;
9596         } else {
9597                 ctsio->residual = 0;
9598                 ctsio->kern_data_len = alloc_len;
9599                 ctsio->kern_total_len = alloc_len;
9600         }
9601         ctsio->kern_data_resid = 0;
9602         ctsio->kern_rel_offset = 0;
9603         ctsio->kern_sg_entries = 0;
9604
9605         /*
9606          * The control device is always connected.  The disk device, on the
9607          * other hand, may not be online all the time.
9608          */
9609         if (lun != NULL)
9610                 devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9611                                      lun->be_lun->lun_type;
9612         else
9613                 devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9614         devid_ptr->page_code = SVPD_DEVICE_ID;
9615         scsi_ulto2b(data_len - 4, devid_ptr->length);
9616
9617         if (port && port->port_type == CTL_PORT_FC)
9618                 proto = SCSI_PROTO_FC << 4;
9619         else if (port && port->port_type == CTL_PORT_ISCSI)
9620                 proto = SCSI_PROTO_ISCSI << 4;
9621         else
9622                 proto = SCSI_PROTO_SPI << 4;
9623         desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
9624
9625         /*
9626          * We're using a LUN association here.  i.e., this device ID is a
9627          * per-LUN identifier.
9628          */
9629         if (lun && lun->lun_devid) {
9630                 memcpy(desc, lun->lun_devid->data, lun->lun_devid->len);
9631                 desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9632                     lun->lun_devid->len);
9633         }
9634
9635         /*
9636          * This is for the WWPN which is a port association.
9637          */
9638         if (port && port->port_devid) {
9639                 memcpy(desc, port->port_devid->data, port->port_devid->len);
9640                 desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9641                     port->port_devid->len);
9642         }
9643
9644         /*
9645          * This is for the Relative Target Port(type 4h) identifier
9646          */
9647         desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9648         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9649             SVPD_ID_TYPE_RELTARG;
9650         desc->length = 4;
9651         scsi_ulto2b(ctsio->io_hdr.nexus.targ_port, &desc->identifier[2]);
9652         desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9653             sizeof(struct scsi_vpd_id_rel_trgt_port_id));
9654
9655         /*
9656          * This is for the Target Port Group(type 5h) identifier
9657          */
9658         desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9659         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9660             SVPD_ID_TYPE_TPORTGRP;
9661         desc->length = 4;
9662         scsi_ulto2b(ctsio->io_hdr.nexus.targ_port / softc->port_cnt + 1,
9663             &desc->identifier[2]);
9664         desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9665             sizeof(struct scsi_vpd_id_trgt_port_grp_id));
9666
9667         /*
9668          * This is for the Target identifier
9669          */
9670         if (port && port->target_devid) {
9671                 memcpy(desc, port->target_devid->data, port->target_devid->len);
9672         }
9673
9674         ctl_set_success(ctsio);
9675         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9676         ctsio->be_move_done = ctl_config_move_done;
9677         ctl_datamove((union ctl_io *)ctsio);
9678         return (CTL_RETVAL_COMPLETE);
9679 }
9680
9681 static int
9682 ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio, int alloc_len)
9683 {
9684         struct ctl_softc *softc = control_softc;
9685         struct scsi_vpd_scsi_ports *sp;
9686         struct scsi_vpd_port_designation *pd;
9687         struct scsi_vpd_port_designation_cont *pdc;
9688         struct ctl_lun *lun;
9689         struct ctl_port *port;
9690         int data_len, num_target_ports, iid_len, id_len;
9691
9692         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9693
9694         num_target_ports = 0;
9695         iid_len = 0;
9696         id_len = 0;
9697         mtx_lock(&softc->ctl_lock);
9698         STAILQ_FOREACH(port, &softc->port_list, links) {
9699                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
9700                         continue;
9701                 if (lun != NULL &&
9702                     ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
9703                         continue;
9704                 num_target_ports++;
9705                 if (port->init_devid)
9706                         iid_len += port->init_devid->len;
9707                 if (port->port_devid)
9708                         id_len += port->port_devid->len;
9709         }
9710         mtx_unlock(&softc->ctl_lock);
9711
9712         data_len = sizeof(struct scsi_vpd_scsi_ports) +
9713             num_target_ports * (sizeof(struct scsi_vpd_port_designation) +
9714              sizeof(struct scsi_vpd_port_designation_cont)) + iid_len + id_len;
9715         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9716         sp = (struct scsi_vpd_scsi_ports *)ctsio->kern_data_ptr;
9717         ctsio->kern_sg_entries = 0;
9718
9719         if (data_len < alloc_len) {
9720                 ctsio->residual = alloc_len - data_len;
9721                 ctsio->kern_data_len = data_len;
9722                 ctsio->kern_total_len = data_len;
9723         } else {
9724                 ctsio->residual = 0;
9725                 ctsio->kern_data_len = alloc_len;
9726                 ctsio->kern_total_len = alloc_len;
9727         }
9728         ctsio->kern_data_resid = 0;
9729         ctsio->kern_rel_offset = 0;
9730         ctsio->kern_sg_entries = 0;
9731
9732         /*
9733          * The control device is always connected.  The disk device, on the
9734          * other hand, may not be online all the time.  Need to change this
9735          * to figure out whether the disk device is actually online or not.
9736          */
9737         if (lun != NULL)
9738                 sp->device = (SID_QUAL_LU_CONNECTED << 5) |
9739                                   lun->be_lun->lun_type;
9740         else
9741                 sp->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9742
9743         sp->page_code = SVPD_SCSI_PORTS;
9744         scsi_ulto2b(data_len - sizeof(struct scsi_vpd_scsi_ports),
9745             sp->page_length);
9746         pd = &sp->design[0];
9747
9748         mtx_lock(&softc->ctl_lock);
9749         STAILQ_FOREACH(port, &softc->port_list, links) {
9750                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
9751                         continue;
9752                 if (lun != NULL &&
9753                     ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
9754                         continue;
9755                 scsi_ulto2b(port->targ_port, pd->relative_port_id);
9756                 if (port->init_devid) {
9757                         iid_len = port->init_devid->len;
9758                         memcpy(pd->initiator_transportid,
9759                             port->init_devid->data, port->init_devid->len);
9760                 } else
9761                         iid_len = 0;
9762                 scsi_ulto2b(iid_len, pd->initiator_transportid_length);
9763                 pdc = (struct scsi_vpd_port_designation_cont *)
9764                     (&pd->initiator_transportid[iid_len]);
9765                 if (port->port_devid) {
9766                         id_len = port->port_devid->len;
9767                         memcpy(pdc->target_port_descriptors,
9768                             port->port_devid->data, port->port_devid->len);
9769                 } else
9770                         id_len = 0;
9771                 scsi_ulto2b(id_len, pdc->target_port_descriptors_length);
9772                 pd = (struct scsi_vpd_port_designation *)
9773                     ((uint8_t *)pdc->target_port_descriptors + id_len);
9774         }
9775         mtx_unlock(&softc->ctl_lock);
9776
9777         ctl_set_success(ctsio);
9778         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9779         ctsio->be_move_done = ctl_config_move_done;
9780         ctl_datamove((union ctl_io *)ctsio);
9781         return (CTL_RETVAL_COMPLETE);
9782 }
9783
9784 static int
9785 ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio, int alloc_len)
9786 {
9787         struct scsi_vpd_block_limits *bl_ptr;
9788         struct ctl_lun *lun;
9789         int bs;
9790
9791         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9792
9793         ctsio->kern_data_ptr = malloc(sizeof(*bl_ptr), M_CTL, M_WAITOK | M_ZERO);
9794         bl_ptr = (struct scsi_vpd_block_limits *)ctsio->kern_data_ptr;
9795         ctsio->kern_sg_entries = 0;
9796
9797         if (sizeof(*bl_ptr) < alloc_len) {
9798                 ctsio->residual = alloc_len - sizeof(*bl_ptr);
9799                 ctsio->kern_data_len = sizeof(*bl_ptr);
9800                 ctsio->kern_total_len = sizeof(*bl_ptr);
9801         } else {
9802                 ctsio->residual = 0;
9803                 ctsio->kern_data_len = alloc_len;
9804                 ctsio->kern_total_len = alloc_len;
9805         }
9806         ctsio->kern_data_resid = 0;
9807         ctsio->kern_rel_offset = 0;
9808         ctsio->kern_sg_entries = 0;
9809
9810         /*
9811          * The control device is always connected.  The disk device, on the
9812          * other hand, may not be online all the time.  Need to change this
9813          * to figure out whether the disk device is actually online or not.
9814          */
9815         if (lun != NULL)
9816                 bl_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9817                                   lun->be_lun->lun_type;
9818         else
9819                 bl_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9820
9821         bl_ptr->page_code = SVPD_BLOCK_LIMITS;
9822         scsi_ulto2b(sizeof(*bl_ptr) - 4, bl_ptr->page_length);
9823         bl_ptr->max_cmp_write_len = 0xff;
9824         scsi_ulto4b(0xffffffff, bl_ptr->max_txfer_len);
9825         if (lun != NULL) {
9826                 bs = lun->be_lun->blocksize;
9827                 scsi_ulto4b(lun->be_lun->opttxferlen, bl_ptr->opt_txfer_len);
9828                 if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
9829                         scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_lba_cnt);
9830                         scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_blk_cnt);
9831                         if (lun->be_lun->ublockexp != 0) {
9832                                 scsi_ulto4b((1 << lun->be_lun->ublockexp),
9833                                     bl_ptr->opt_unmap_grain);
9834                                 scsi_ulto4b(0x80000000 | lun->be_lun->ublockoff,
9835                                     bl_ptr->unmap_grain_align);
9836                         }
9837                 }
9838                 scsi_ulto4b(lun->be_lun->atomicblock,
9839                     bl_ptr->max_atomic_transfer_length);
9840                 scsi_ulto4b(0, bl_ptr->atomic_alignment);
9841                 scsi_ulto4b(0, bl_ptr->atomic_transfer_length_granularity);
9842         }
9843         scsi_u64to8b(UINT64_MAX, bl_ptr->max_write_same_length);
9844
9845         ctl_set_success(ctsio);
9846         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9847         ctsio->be_move_done = ctl_config_move_done;
9848         ctl_datamove((union ctl_io *)ctsio);
9849         return (CTL_RETVAL_COMPLETE);
9850 }
9851
9852 static int
9853 ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len)
9854 {
9855         struct scsi_vpd_block_device_characteristics *bdc_ptr;
9856         struct ctl_lun *lun;
9857         const char *value;
9858         u_int i;
9859
9860         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9861
9862         ctsio->kern_data_ptr = malloc(sizeof(*bdc_ptr), M_CTL, M_WAITOK | M_ZERO);
9863         bdc_ptr = (struct scsi_vpd_block_device_characteristics *)ctsio->kern_data_ptr;
9864         ctsio->kern_sg_entries = 0;
9865
9866         if (sizeof(*bdc_ptr) < alloc_len) {
9867                 ctsio->residual = alloc_len - sizeof(*bdc_ptr);
9868                 ctsio->kern_data_len = sizeof(*bdc_ptr);
9869                 ctsio->kern_total_len = sizeof(*bdc_ptr);
9870         } else {
9871                 ctsio->residual = 0;
9872                 ctsio->kern_data_len = alloc_len;
9873                 ctsio->kern_total_len = alloc_len;
9874         }
9875         ctsio->kern_data_resid = 0;
9876         ctsio->kern_rel_offset = 0;
9877         ctsio->kern_sg_entries = 0;
9878
9879         /*
9880          * The control device is always connected.  The disk device, on the
9881          * other hand, may not be online all the time.  Need to change this
9882          * to figure out whether the disk device is actually online or not.
9883          */
9884         if (lun != NULL)
9885                 bdc_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9886                                   lun->be_lun->lun_type;
9887         else
9888                 bdc_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9889         bdc_ptr->page_code = SVPD_BDC;
9890         scsi_ulto2b(sizeof(*bdc_ptr) - 4, bdc_ptr->page_length);
9891         if (lun != NULL &&
9892             (value = ctl_get_opt(&lun->be_lun->options, "rpm")) != NULL)
9893                 i = strtol(value, NULL, 0);
9894         else
9895                 i = CTL_DEFAULT_ROTATION_RATE;
9896         scsi_ulto2b(i, bdc_ptr->medium_rotation_rate);
9897         if (lun != NULL &&
9898             (value = ctl_get_opt(&lun->be_lun->options, "formfactor")) != NULL)
9899                 i = strtol(value, NULL, 0);
9900         else
9901                 i = 0;
9902         bdc_ptr->wab_wac_ff = (i & 0x0f);
9903         bdc_ptr->flags = SVPD_FUAB | SVPD_VBULS;
9904
9905         ctl_set_success(ctsio);
9906         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9907         ctsio->be_move_done = ctl_config_move_done;
9908         ctl_datamove((union ctl_io *)ctsio);
9909         return (CTL_RETVAL_COMPLETE);
9910 }
9911
9912 static int
9913 ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len)
9914 {
9915         struct scsi_vpd_logical_block_prov *lbp_ptr;
9916         struct ctl_lun *lun;
9917
9918         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9919
9920         ctsio->kern_data_ptr = malloc(sizeof(*lbp_ptr), M_CTL, M_WAITOK | M_ZERO);
9921         lbp_ptr = (struct scsi_vpd_logical_block_prov *)ctsio->kern_data_ptr;
9922         ctsio->kern_sg_entries = 0;
9923
9924         if (sizeof(*lbp_ptr) < alloc_len) {
9925                 ctsio->residual = alloc_len - sizeof(*lbp_ptr);
9926                 ctsio->kern_data_len = sizeof(*lbp_ptr);
9927                 ctsio->kern_total_len = sizeof(*lbp_ptr);
9928         } else {
9929                 ctsio->residual = 0;
9930                 ctsio->kern_data_len = alloc_len;
9931                 ctsio->kern_total_len = alloc_len;
9932         }
9933         ctsio->kern_data_resid = 0;
9934         ctsio->kern_rel_offset = 0;
9935         ctsio->kern_sg_entries = 0;
9936
9937         /*
9938          * The control device is always connected.  The disk device, on the
9939          * other hand, may not be online all the time.  Need to change this
9940          * to figure out whether the disk device is actually online or not.
9941          */
9942         if (lun != NULL)
9943                 lbp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9944                                   lun->be_lun->lun_type;
9945         else
9946                 lbp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9947
9948         lbp_ptr->page_code = SVPD_LBP;
9949         scsi_ulto2b(sizeof(*lbp_ptr) - 4, lbp_ptr->page_length);
9950         lbp_ptr->threshold_exponent = CTL_LBP_EXPONENT;
9951         if (lun != NULL && lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
9952                 lbp_ptr->flags = SVPD_LBP_UNMAP | SVPD_LBP_WS16 |
9953                     SVPD_LBP_WS10 | SVPD_LBP_RZ | SVPD_LBP_ANC_SUP;
9954                 lbp_ptr->prov_type = SVPD_LBP_THIN;
9955         }
9956
9957         ctl_set_success(ctsio);
9958         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9959         ctsio->be_move_done = ctl_config_move_done;
9960         ctl_datamove((union ctl_io *)ctsio);
9961         return (CTL_RETVAL_COMPLETE);
9962 }
9963
9964 /*
9965  * INQUIRY with the EVPD bit set.
9966  */
9967 static int
9968 ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
9969 {
9970         struct ctl_lun *lun;
9971         struct scsi_inquiry *cdb;
9972         int alloc_len, retval;
9973
9974         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9975         cdb = (struct scsi_inquiry *)ctsio->cdb;
9976         alloc_len = scsi_2btoul(cdb->length);
9977
9978         switch (cdb->page_code) {
9979         case SVPD_SUPPORTED_PAGES:
9980                 retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
9981                 break;
9982         case SVPD_UNIT_SERIAL_NUMBER:
9983                 retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
9984                 break;
9985         case SVPD_DEVICE_ID:
9986                 retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
9987                 break;
9988         case SVPD_EXTENDED_INQUIRY_DATA:
9989                 retval = ctl_inquiry_evpd_eid(ctsio, alloc_len);
9990                 break;
9991         case SVPD_MODE_PAGE_POLICY:
9992                 retval = ctl_inquiry_evpd_mpp(ctsio, alloc_len);
9993                 break;
9994         case SVPD_SCSI_PORTS:
9995                 retval = ctl_inquiry_evpd_scsi_ports(ctsio, alloc_len);
9996                 break;
9997         case SVPD_SCSI_TPC:
9998                 retval = ctl_inquiry_evpd_tpc(ctsio, alloc_len);
9999                 break;
10000         case SVPD_BLOCK_LIMITS:
10001                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
10002                         goto err;
10003                 retval = ctl_inquiry_evpd_block_limits(ctsio, alloc_len);
10004                 break;
10005         case SVPD_BDC:
10006                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
10007                         goto err;
10008                 retval = ctl_inquiry_evpd_bdc(ctsio, alloc_len);
10009                 break;
10010         case SVPD_LBP:
10011                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
10012                         goto err;
10013                 retval = ctl_inquiry_evpd_lbp(ctsio, alloc_len);
10014                 break;
10015         default:
10016 err:
10017                 ctl_set_invalid_field(ctsio,
10018                                       /*sks_valid*/ 1,
10019                                       /*command*/ 1,
10020                                       /*field*/ 2,
10021                                       /*bit_valid*/ 0,
10022                                       /*bit*/ 0);
10023                 ctl_done((union ctl_io *)ctsio);
10024                 retval = CTL_RETVAL_COMPLETE;
10025                 break;
10026         }
10027
10028         return (retval);
10029 }
10030
10031 /*
10032  * Standard INQUIRY data.
10033  */
10034 static int
10035 ctl_inquiry_std(struct ctl_scsiio *ctsio)
10036 {
10037         struct scsi_inquiry_data *inq_ptr;
10038         struct scsi_inquiry *cdb;
10039         struct ctl_softc *softc;
10040         struct ctl_port *port;
10041         struct ctl_lun *lun;
10042         char *val;
10043         uint32_t alloc_len, data_len;
10044         ctl_port_type port_type;
10045
10046         softc = control_softc;
10047
10048         /*
10049          * Figure out whether we're talking to a Fibre Channel port or not.
10050          * We treat the ioctl front end, and any SCSI adapters, as packetized
10051          * SCSI front ends.
10052          */
10053         port = ctl_io_port(&ctsio->io_hdr);
10054         if (port != NULL)
10055                 port_type = port->port_type;
10056         else
10057                 port_type = CTL_PORT_SCSI;
10058         if (port_type == CTL_PORT_IOCTL || port_type == CTL_PORT_INTERNAL)
10059                 port_type = CTL_PORT_SCSI;
10060
10061         lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10062         cdb = (struct scsi_inquiry *)ctsio->cdb;
10063         alloc_len = scsi_2btoul(cdb->length);
10064
10065         /*
10066          * We malloc the full inquiry data size here and fill it
10067          * in.  If the user only asks for less, we'll give him
10068          * that much.
10069          */
10070         data_len = offsetof(struct scsi_inquiry_data, vendor_specific1);
10071         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10072         inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
10073         ctsio->kern_sg_entries = 0;
10074         ctsio->kern_data_resid = 0;
10075         ctsio->kern_rel_offset = 0;
10076
10077         if (data_len < alloc_len) {
10078                 ctsio->residual = alloc_len - data_len;
10079                 ctsio->kern_data_len = data_len;
10080                 ctsio->kern_total_len = data_len;
10081         } else {
10082                 ctsio->residual = 0;
10083                 ctsio->kern_data_len = alloc_len;
10084                 ctsio->kern_total_len = alloc_len;
10085         }
10086
10087         if (lun != NULL) {
10088                 if ((lun->flags & CTL_LUN_PRIMARY_SC) ||
10089                     softc->ha_link >= CTL_HA_LINK_UNKNOWN) {
10090                         inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10091                             lun->be_lun->lun_type;
10092                 } else {
10093                         inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) |
10094                             lun->be_lun->lun_type;
10095                 }
10096         } else
10097                 inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE;
10098
10099         /* RMB in byte 2 is 0 */
10100         inq_ptr->version = SCSI_REV_SPC4;
10101
10102         /*
10103          * According to SAM-3, even if a device only supports a single
10104          * level of LUN addressing, it should still set the HISUP bit:
10105          *
10106          * 4.9.1 Logical unit numbers overview
10107          *
10108          * All logical unit number formats described in this standard are
10109          * hierarchical in structure even when only a single level in that
10110          * hierarchy is used. The HISUP bit shall be set to one in the
10111          * standard INQUIRY data (see SPC-2) when any logical unit number
10112          * format described in this standard is used.  Non-hierarchical
10113          * formats are outside the scope of this standard.
10114          *
10115          * Therefore we set the HiSup bit here.
10116          *
10117          * The reponse format is 2, per SPC-3.
10118          */
10119         inq_ptr->response_format = SID_HiSup | 2;
10120
10121         inq_ptr->additional_length = data_len -
10122             (offsetof(struct scsi_inquiry_data, additional_length) + 1);
10123         CTL_DEBUG_PRINT(("additional_length = %d\n",
10124                          inq_ptr->additional_length));
10125
10126         inq_ptr->spc3_flags = SPC3_SID_3PC | SPC3_SID_TPGS_IMPLICIT;
10127         /* 16 bit addressing */
10128         if (port_type == CTL_PORT_SCSI)
10129                 inq_ptr->spc2_flags = SPC2_SID_ADDR16;
10130         /* XXX set the SID_MultiP bit here if we're actually going to
10131            respond on multiple ports */
10132         inq_ptr->spc2_flags |= SPC2_SID_MultiP;
10133
10134         /* 16 bit data bus, synchronous transfers */
10135         if (port_type == CTL_PORT_SCSI)
10136                 inq_ptr->flags = SID_WBus16 | SID_Sync;
10137         /*
10138          * XXX KDM do we want to support tagged queueing on the control
10139          * device at all?
10140          */
10141         if ((lun == NULL)
10142          || (lun->be_lun->lun_type != T_PROCESSOR))
10143                 inq_ptr->flags |= SID_CmdQue;
10144         /*
10145          * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
10146          * We have 8 bytes for the vendor name, and 16 bytes for the device
10147          * name and 4 bytes for the revision.
10148          */
10149         if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10150             "vendor")) == NULL) {
10151                 strncpy(inq_ptr->vendor, CTL_VENDOR, sizeof(inq_ptr->vendor));
10152         } else {
10153                 memset(inq_ptr->vendor, ' ', sizeof(inq_ptr->vendor));
10154                 strncpy(inq_ptr->vendor, val,
10155                     min(sizeof(inq_ptr->vendor), strlen(val)));
10156         }
10157         if (lun == NULL) {
10158                 strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10159                     sizeof(inq_ptr->product));
10160         } else if ((val = ctl_get_opt(&lun->be_lun->options, "product")) == NULL) {
10161                 switch (lun->be_lun->lun_type) {
10162                 case T_DIRECT:
10163                         strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10164                             sizeof(inq_ptr->product));
10165                         break;
10166                 case T_PROCESSOR:
10167                         strncpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT,
10168                             sizeof(inq_ptr->product));
10169                         break;
10170                 default:
10171                         strncpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT,
10172                             sizeof(inq_ptr->product));
10173                         break;
10174                 }
10175         } else {
10176                 memset(inq_ptr->product, ' ', sizeof(inq_ptr->product));
10177                 strncpy(inq_ptr->product, val,
10178                     min(sizeof(inq_ptr->product), strlen(val)));
10179         }
10180
10181         /*
10182          * XXX make this a macro somewhere so it automatically gets
10183          * incremented when we make changes.
10184          */
10185         if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10186             "revision")) == NULL) {
10187                 strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
10188         } else {
10189                 memset(inq_ptr->revision, ' ', sizeof(inq_ptr->revision));
10190                 strncpy(inq_ptr->revision, val,
10191                     min(sizeof(inq_ptr->revision), strlen(val)));
10192         }
10193
10194         /*
10195          * For parallel SCSI, we support double transition and single
10196          * transition clocking.  We also support QAS (Quick Arbitration
10197          * and Selection) and Information Unit transfers on both the
10198          * control and array devices.
10199          */
10200         if (port_type == CTL_PORT_SCSI)
10201                 inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
10202                                     SID_SPI_IUS;
10203
10204         /* SAM-5 (no version claimed) */
10205         scsi_ulto2b(0x00A0, inq_ptr->version1);
10206         /* SPC-4 (no version claimed) */
10207         scsi_ulto2b(0x0460, inq_ptr->version2);
10208         if (port_type == CTL_PORT_FC) {
10209                 /* FCP-2 ANSI INCITS.350:2003 */
10210                 scsi_ulto2b(0x0917, inq_ptr->version3);
10211         } else if (port_type == CTL_PORT_SCSI) {
10212                 /* SPI-4 ANSI INCITS.362:200x */
10213                 scsi_ulto2b(0x0B56, inq_ptr->version3);
10214         } else if (port_type == CTL_PORT_ISCSI) {
10215                 /* iSCSI (no version claimed) */
10216                 scsi_ulto2b(0x0960, inq_ptr->version3);
10217         } else if (port_type == CTL_PORT_SAS) {
10218                 /* SAS (no version claimed) */
10219                 scsi_ulto2b(0x0BE0, inq_ptr->version3);
10220         }
10221
10222         if (lun == NULL) {
10223                 /* SBC-4 (no version claimed) */
10224                 scsi_ulto2b(0x0600, inq_ptr->version4);
10225         } else {
10226                 switch (lun->be_lun->lun_type) {
10227                 case T_DIRECT:
10228                         /* SBC-4 (no version claimed) */
10229                         scsi_ulto2b(0x0600, inq_ptr->version4);
10230                         break;
10231                 case T_PROCESSOR:
10232                 default:
10233                         break;
10234                 }
10235         }
10236
10237         ctl_set_success(ctsio);
10238         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10239         ctsio->be_move_done = ctl_config_move_done;
10240         ctl_datamove((union ctl_io *)ctsio);
10241         return (CTL_RETVAL_COMPLETE);
10242 }
10243
10244 int
10245 ctl_inquiry(struct ctl_scsiio *ctsio)
10246 {
10247         struct scsi_inquiry *cdb;
10248         int retval;
10249
10250         CTL_DEBUG_PRINT(("ctl_inquiry\n"));
10251
10252         cdb = (struct scsi_inquiry *)ctsio->cdb;
10253         if (cdb->byte2 & SI_EVPD)
10254                 retval = ctl_inquiry_evpd(ctsio);
10255         else if (cdb->page_code == 0)
10256                 retval = ctl_inquiry_std(ctsio);
10257         else {
10258                 ctl_set_invalid_field(ctsio,
10259                                       /*sks_valid*/ 1,
10260                                       /*command*/ 1,
10261                                       /*field*/ 2,
10262                                       /*bit_valid*/ 0,
10263                                       /*bit*/ 0);
10264                 ctl_done((union ctl_io *)ctsio);
10265                 return (CTL_RETVAL_COMPLETE);
10266         }
10267
10268         return (retval);
10269 }
10270
10271 /*
10272  * For known CDB types, parse the LBA and length.
10273  */
10274 static int
10275 ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len)
10276 {
10277         if (io->io_hdr.io_type != CTL_IO_SCSI)
10278                 return (1);
10279
10280         switch (io->scsiio.cdb[0]) {
10281         case COMPARE_AND_WRITE: {
10282                 struct scsi_compare_and_write *cdb;
10283
10284                 cdb = (struct scsi_compare_and_write *)io->scsiio.cdb;
10285
10286                 *lba = scsi_8btou64(cdb->addr);
10287                 *len = cdb->length;
10288                 break;
10289         }
10290         case READ_6:
10291         case WRITE_6: {
10292                 struct scsi_rw_6 *cdb;
10293
10294                 cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
10295
10296                 *lba = scsi_3btoul(cdb->addr);
10297                 /* only 5 bits are valid in the most significant address byte */
10298                 *lba &= 0x1fffff;
10299                 *len = cdb->length;
10300                 break;
10301         }
10302         case READ_10:
10303         case WRITE_10: {
10304                 struct scsi_rw_10 *cdb;
10305
10306                 cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
10307
10308                 *lba = scsi_4btoul(cdb->addr);
10309                 *len = scsi_2btoul(cdb->length);
10310                 break;
10311         }
10312         case WRITE_VERIFY_10: {
10313                 struct scsi_write_verify_10 *cdb;
10314
10315                 cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
10316
10317                 *lba = scsi_4btoul(cdb->addr);
10318                 *len = scsi_2btoul(cdb->length);
10319                 break;
10320         }
10321         case READ_12:
10322         case WRITE_12: {
10323                 struct scsi_rw_12 *cdb;
10324
10325                 cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
10326
10327                 *lba = scsi_4btoul(cdb->addr);
10328                 *len = scsi_4btoul(cdb->length);
10329                 break;
10330         }
10331         case WRITE_VERIFY_12: {
10332                 struct scsi_write_verify_12 *cdb;
10333
10334                 cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
10335
10336                 *lba = scsi_4btoul(cdb->addr);
10337                 *len = scsi_4btoul(cdb->length);
10338                 break;
10339         }
10340         case READ_16:
10341         case WRITE_16:
10342         case WRITE_ATOMIC_16: {
10343                 struct scsi_rw_16 *cdb;
10344
10345                 cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
10346
10347                 *lba = scsi_8btou64(cdb->addr);
10348                 *len = scsi_4btoul(cdb->length);
10349                 break;
10350         }
10351         case WRITE_VERIFY_16: {
10352                 struct scsi_write_verify_16 *cdb;
10353
10354                 cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
10355
10356                 *lba = scsi_8btou64(cdb->addr);
10357                 *len = scsi_4btoul(cdb->length);
10358                 break;
10359         }
10360         case WRITE_SAME_10: {
10361                 struct scsi_write_same_10 *cdb;
10362
10363                 cdb = (struct scsi_write_same_10 *)io->scsiio.cdb;
10364
10365                 *lba = scsi_4btoul(cdb->addr);
10366                 *len = scsi_2btoul(cdb->length);
10367                 break;
10368         }
10369         case WRITE_SAME_16: {
10370                 struct scsi_write_same_16 *cdb;
10371
10372                 cdb = (struct scsi_write_same_16 *)io->scsiio.cdb;
10373
10374                 *lba = scsi_8btou64(cdb->addr);
10375                 *len = scsi_4btoul(cdb->length);
10376                 break;
10377         }
10378         case VERIFY_10: {
10379                 struct scsi_verify_10 *cdb;
10380
10381                 cdb = (struct scsi_verify_10 *)io->scsiio.cdb;
10382
10383                 *lba = scsi_4btoul(cdb->addr);
10384                 *len = scsi_2btoul(cdb->length);
10385                 break;
10386         }
10387         case VERIFY_12: {
10388                 struct scsi_verify_12 *cdb;
10389
10390                 cdb = (struct scsi_verify_12 *)io->scsiio.cdb;
10391
10392                 *lba = scsi_4btoul(cdb->addr);
10393                 *len = scsi_4btoul(cdb->length);
10394                 break;
10395         }
10396         case VERIFY_16: {
10397                 struct scsi_verify_16 *cdb;
10398
10399                 cdb = (struct scsi_verify_16 *)io->scsiio.cdb;
10400
10401                 *lba = scsi_8btou64(cdb->addr);
10402                 *len = scsi_4btoul(cdb->length);
10403                 break;
10404         }
10405         case UNMAP: {
10406                 *lba = 0;
10407                 *len = UINT64_MAX;
10408                 break;
10409         }
10410         case SERVICE_ACTION_IN: {       /* GET LBA STATUS */
10411                 struct scsi_get_lba_status *cdb;
10412
10413                 cdb = (struct scsi_get_lba_status *)io->scsiio.cdb;
10414                 *lba = scsi_8btou64(cdb->addr);
10415                 *len = UINT32_MAX;
10416                 break;
10417         }
10418         default:
10419                 return (1);
10420                 break; /* NOTREACHED */
10421         }
10422
10423         return (0);
10424 }
10425
10426 static ctl_action
10427 ctl_extent_check_lba(uint64_t lba1, uint64_t len1, uint64_t lba2, uint64_t len2,
10428     bool seq)
10429 {
10430         uint64_t endlba1, endlba2;
10431
10432         endlba1 = lba1 + len1 - (seq ? 0 : 1);
10433         endlba2 = lba2 + len2 - 1;
10434
10435         if ((endlba1 < lba2) || (endlba2 < lba1))
10436                 return (CTL_ACTION_PASS);
10437         else
10438                 return (CTL_ACTION_BLOCK);
10439 }
10440
10441 static int
10442 ctl_extent_check_unmap(union ctl_io *io, uint64_t lba2, uint64_t len2)
10443 {
10444         struct ctl_ptr_len_flags *ptrlen;
10445         struct scsi_unmap_desc *buf, *end, *range;
10446         uint64_t lba;
10447         uint32_t len;
10448
10449         /* If not UNMAP -- go other way. */
10450         if (io->io_hdr.io_type != CTL_IO_SCSI ||
10451             io->scsiio.cdb[0] != UNMAP)
10452                 return (CTL_ACTION_ERROR);
10453
10454         /* If UNMAP without data -- block and wait for data. */
10455         ptrlen = (struct ctl_ptr_len_flags *)
10456             &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
10457         if ((io->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0 ||
10458             ptrlen->ptr == NULL)
10459                 return (CTL_ACTION_BLOCK);
10460
10461         /* UNMAP with data -- check for collision. */
10462         buf = (struct scsi_unmap_desc *)ptrlen->ptr;
10463         end = buf + ptrlen->len / sizeof(*buf);
10464         for (range = buf; range < end; range++) {
10465                 lba = scsi_8btou64(range->lba);
10466                 len = scsi_4btoul(range->length);
10467                 if ((lba < lba2 + len2) && (lba + len > lba2))
10468                         return (CTL_ACTION_BLOCK);
10469         }
10470         return (CTL_ACTION_PASS);
10471 }
10472
10473 static ctl_action
10474 ctl_extent_check(union ctl_io *io1, union ctl_io *io2, bool seq)
10475 {
10476         uint64_t lba1, lba2;
10477         uint64_t len1, len2;
10478         int retval;
10479
10480         if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
10481                 return (CTL_ACTION_ERROR);
10482
10483         retval = ctl_extent_check_unmap(io1, lba2, len2);
10484         if (retval != CTL_ACTION_ERROR)
10485                 return (retval);
10486
10487         if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
10488                 return (CTL_ACTION_ERROR);
10489
10490         return (ctl_extent_check_lba(lba1, len1, lba2, len2, seq));
10491 }
10492
10493 static ctl_action
10494 ctl_extent_check_seq(union ctl_io *io1, union ctl_io *io2)
10495 {
10496         uint64_t lba1, lba2;
10497         uint64_t len1, len2;
10498
10499         if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
10500                 return (CTL_ACTION_ERROR);
10501         if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
10502                 return (CTL_ACTION_ERROR);
10503
10504         if (lba1 + len1 == lba2)
10505                 return (CTL_ACTION_BLOCK);
10506         return (CTL_ACTION_PASS);
10507 }
10508
10509 static ctl_action
10510 ctl_check_for_blockage(struct ctl_lun *lun, union ctl_io *pending_io,
10511     union ctl_io *ooa_io)
10512 {
10513         const struct ctl_cmd_entry *pending_entry, *ooa_entry;
10514         ctl_serialize_action *serialize_row;
10515
10516         /*
10517          * The initiator attempted multiple untagged commands at the same
10518          * time.  Can't do that.
10519          */
10520         if ((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10521          && (ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10522          && ((pending_io->io_hdr.nexus.targ_port ==
10523               ooa_io->io_hdr.nexus.targ_port)
10524           && (pending_io->io_hdr.nexus.initid ==
10525               ooa_io->io_hdr.nexus.initid))
10526          && ((ooa_io->io_hdr.flags & (CTL_FLAG_ABORT |
10527               CTL_FLAG_STATUS_SENT)) == 0))
10528                 return (CTL_ACTION_OVERLAP);
10529
10530         /*
10531          * The initiator attempted to send multiple tagged commands with
10532          * the same ID.  (It's fine if different initiators have the same
10533          * tag ID.)
10534          *
10535          * Even if all of those conditions are true, we don't kill the I/O
10536          * if the command ahead of us has been aborted.  We won't end up
10537          * sending it to the FETD, and it's perfectly legal to resend a
10538          * command with the same tag number as long as the previous
10539          * instance of this tag number has been aborted somehow.
10540          */
10541         if ((pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10542          && (ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
10543          && (pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
10544          && ((pending_io->io_hdr.nexus.targ_port ==
10545               ooa_io->io_hdr.nexus.targ_port)
10546           && (pending_io->io_hdr.nexus.initid ==
10547               ooa_io->io_hdr.nexus.initid))
10548          && ((ooa_io->io_hdr.flags & (CTL_FLAG_ABORT |
10549               CTL_FLAG_STATUS_SENT)) == 0))
10550                 return (CTL_ACTION_OVERLAP_TAG);
10551
10552         /*
10553          * If we get a head of queue tag, SAM-3 says that we should
10554          * immediately execute it.
10555          *
10556          * What happens if this command would normally block for some other
10557          * reason?  e.g. a request sense with a head of queue tag
10558          * immediately after a write.  Normally that would block, but this
10559          * will result in its getting executed immediately...
10560          *
10561          * We currently return "pass" instead of "skip", so we'll end up
10562          * going through the rest of the queue to check for overlapped tags.
10563          *
10564          * XXX KDM check for other types of blockage first??
10565          */
10566         if (pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10567                 return (CTL_ACTION_PASS);
10568
10569         /*
10570          * Ordered tags have to block until all items ahead of them
10571          * have completed.  If we get called with an ordered tag, we always
10572          * block, if something else is ahead of us in the queue.
10573          */
10574         if (pending_io->scsiio.tag_type == CTL_TAG_ORDERED)
10575                 return (CTL_ACTION_BLOCK);
10576
10577         /*
10578          * Simple tags get blocked until all head of queue and ordered tags
10579          * ahead of them have completed.  I'm lumping untagged commands in
10580          * with simple tags here.  XXX KDM is that the right thing to do?
10581          */
10582         if (((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10583           || (pending_io->scsiio.tag_type == CTL_TAG_SIMPLE))
10584          && ((ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
10585           || (ooa_io->scsiio.tag_type == CTL_TAG_ORDERED)))
10586                 return (CTL_ACTION_BLOCK);
10587
10588         pending_entry = ctl_get_cmd_entry(&pending_io->scsiio, NULL);
10589         ooa_entry = ctl_get_cmd_entry(&ooa_io->scsiio, NULL);
10590
10591         serialize_row = ctl_serialize_table[ooa_entry->seridx];
10592
10593         switch (serialize_row[pending_entry->seridx]) {
10594         case CTL_SER_BLOCK:
10595                 return (CTL_ACTION_BLOCK);
10596         case CTL_SER_EXTENT:
10597                 return (ctl_extent_check(ooa_io, pending_io,
10598                     (lun->be_lun && lun->be_lun->serseq == CTL_LUN_SERSEQ_ON)));
10599         case CTL_SER_EXTENTOPT:
10600                 if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT].queue_flags
10601                     & SCP_QUEUE_ALG_MASK) != SCP_QUEUE_ALG_UNRESTRICTED)
10602                         return (ctl_extent_check(ooa_io, pending_io,
10603                             (lun->be_lun &&
10604                              lun->be_lun->serseq == CTL_LUN_SERSEQ_ON)));
10605                 return (CTL_ACTION_PASS);
10606         case CTL_SER_EXTENTSEQ:
10607                 if (lun->be_lun && lun->be_lun->serseq != CTL_LUN_SERSEQ_OFF)
10608                         return (ctl_extent_check_seq(ooa_io, pending_io));
10609                 return (CTL_ACTION_PASS);
10610         case CTL_SER_PASS:
10611                 return (CTL_ACTION_PASS);
10612         case CTL_SER_BLOCKOPT:
10613                 if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT].queue_flags
10614                     & SCP_QUEUE_ALG_MASK) != SCP_QUEUE_ALG_UNRESTRICTED)
10615                         return (CTL_ACTION_BLOCK);
10616                 return (CTL_ACTION_PASS);
10617         case CTL_SER_SKIP:
10618                 return (CTL_ACTION_SKIP);
10619         default:
10620                 panic("invalid serialization value %d",
10621                       serialize_row[pending_entry->seridx]);
10622         }
10623
10624         return (CTL_ACTION_ERROR);
10625 }
10626
10627 /*
10628  * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
10629  * Assumptions:
10630  * - pending_io is generally either incoming, or on the blocked queue
10631  * - starting I/O is the I/O we want to start the check with.
10632  */
10633 static ctl_action
10634 ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
10635               union ctl_io *starting_io)
10636 {
10637         union ctl_io *ooa_io;
10638         ctl_action action;
10639
10640         mtx_assert(&lun->lun_lock, MA_OWNED);
10641
10642         /*
10643          * Run back along the OOA queue, starting with the current
10644          * blocked I/O and going through every I/O before it on the
10645          * queue.  If starting_io is NULL, we'll just end up returning
10646          * CTL_ACTION_PASS.
10647          */
10648         for (ooa_io = starting_io; ooa_io != NULL;
10649              ooa_io = (union ctl_io *)TAILQ_PREV(&ooa_io->io_hdr, ctl_ooaq,
10650              ooa_links)){
10651
10652                 /*
10653                  * This routine just checks to see whether
10654                  * cur_blocked is blocked by ooa_io, which is ahead
10655                  * of it in the queue.  It doesn't queue/dequeue
10656                  * cur_blocked.
10657                  */
10658                 action = ctl_check_for_blockage(lun, pending_io, ooa_io);
10659                 switch (action) {
10660                 case CTL_ACTION_BLOCK:
10661                 case CTL_ACTION_OVERLAP:
10662                 case CTL_ACTION_OVERLAP_TAG:
10663                 case CTL_ACTION_SKIP:
10664                 case CTL_ACTION_ERROR:
10665                         return (action);
10666                         break; /* NOTREACHED */
10667                 case CTL_ACTION_PASS:
10668                         break;
10669                 default:
10670                         panic("invalid action %d", action);
10671                         break;  /* NOTREACHED */
10672                 }
10673         }
10674
10675         return (CTL_ACTION_PASS);
10676 }
10677
10678 /*
10679  * Assumptions:
10680  * - An I/O has just completed, and has been removed from the per-LUN OOA
10681  *   queue, so some items on the blocked queue may now be unblocked.
10682  */
10683 static int
10684 ctl_check_blocked(struct ctl_lun *lun)
10685 {
10686         struct ctl_softc *softc = lun->ctl_softc;
10687         union ctl_io *cur_blocked, *next_blocked;
10688
10689         mtx_assert(&lun->lun_lock, MA_OWNED);
10690
10691         /*
10692          * Run forward from the head of the blocked queue, checking each
10693          * entry against the I/Os prior to it on the OOA queue to see if
10694          * there is still any blockage.
10695          *
10696          * We cannot use the TAILQ_FOREACH() macro, because it can't deal
10697          * with our removing a variable on it while it is traversing the
10698          * list.
10699          */
10700         for (cur_blocked = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue);
10701              cur_blocked != NULL; cur_blocked = next_blocked) {
10702                 union ctl_io *prev_ooa;
10703                 ctl_action action;
10704
10705                 next_blocked = (union ctl_io *)TAILQ_NEXT(&cur_blocked->io_hdr,
10706                                                           blocked_links);
10707
10708                 prev_ooa = (union ctl_io *)TAILQ_PREV(&cur_blocked->io_hdr,
10709                                                       ctl_ooaq, ooa_links);
10710
10711                 /*
10712                  * If cur_blocked happens to be the first item in the OOA
10713                  * queue now, prev_ooa will be NULL, and the action
10714                  * returned will just be CTL_ACTION_PASS.
10715                  */
10716                 action = ctl_check_ooa(lun, cur_blocked, prev_ooa);
10717
10718                 switch (action) {
10719                 case CTL_ACTION_BLOCK:
10720                         /* Nothing to do here, still blocked */
10721                         break;
10722                 case CTL_ACTION_OVERLAP:
10723                 case CTL_ACTION_OVERLAP_TAG:
10724                         /*
10725                          * This shouldn't happen!  In theory we've already
10726                          * checked this command for overlap...
10727                          */
10728                         break;
10729                 case CTL_ACTION_PASS:
10730                 case CTL_ACTION_SKIP: {
10731                         const struct ctl_cmd_entry *entry;
10732
10733                         /*
10734                          * The skip case shouldn't happen, this transaction
10735                          * should have never made it onto the blocked queue.
10736                          */
10737                         /*
10738                          * This I/O is no longer blocked, we can remove it
10739                          * from the blocked queue.  Since this is a TAILQ
10740                          * (doubly linked list), we can do O(1) removals
10741                          * from any place on the list.
10742                          */
10743                         TAILQ_REMOVE(&lun->blocked_queue, &cur_blocked->io_hdr,
10744                                      blocked_links);
10745                         cur_blocked->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
10746
10747                         if ((softc->ha_mode != CTL_HA_MODE_XFER) &&
10748                             (cur_blocked->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)){
10749                                 /*
10750                                  * Need to send IO back to original side to
10751                                  * run
10752                                  */
10753                                 union ctl_ha_msg msg_info;
10754
10755                                 cur_blocked->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
10756                                 msg_info.hdr.original_sc =
10757                                         cur_blocked->io_hdr.original_sc;
10758                                 msg_info.hdr.serializing_sc = cur_blocked;
10759                                 msg_info.hdr.msg_type = CTL_MSG_R2R;
10760                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
10761                                     sizeof(msg_info.hdr), M_NOWAIT);
10762                                 break;
10763                         }
10764                         entry = ctl_get_cmd_entry(&cur_blocked->scsiio, NULL);
10765
10766                         /*
10767                          * Check this I/O for LUN state changes that may
10768                          * have happened while this command was blocked.
10769                          * The LUN state may have been changed by a command
10770                          * ahead of us in the queue, so we need to re-check
10771                          * for any states that can be caused by SCSI
10772                          * commands.
10773                          */
10774                         if (ctl_scsiio_lun_check(lun, entry,
10775                                                  &cur_blocked->scsiio) == 0) {
10776                                 cur_blocked->io_hdr.flags |=
10777                                                       CTL_FLAG_IS_WAS_ON_RTR;
10778                                 ctl_enqueue_rtr(cur_blocked);
10779                         } else
10780                                 ctl_done(cur_blocked);
10781                         break;
10782                 }
10783                 default:
10784                         /*
10785                          * This probably shouldn't happen -- we shouldn't
10786                          * get CTL_ACTION_ERROR, or anything else.
10787                          */
10788                         break;
10789                 }
10790         }
10791
10792         return (CTL_RETVAL_COMPLETE);
10793 }
10794
10795 /*
10796  * This routine (with one exception) checks LUN flags that can be set by
10797  * commands ahead of us in the OOA queue.  These flags have to be checked
10798  * when a command initially comes in, and when we pull a command off the
10799  * blocked queue and are preparing to execute it.  The reason we have to
10800  * check these flags for commands on the blocked queue is that the LUN
10801  * state may have been changed by a command ahead of us while we're on the
10802  * blocked queue.
10803  *
10804  * Ordering is somewhat important with these checks, so please pay
10805  * careful attention to the placement of any new checks.
10806  */
10807 static int
10808 ctl_scsiio_lun_check(struct ctl_lun *lun,
10809     const struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
10810 {
10811         struct ctl_softc *softc = lun->ctl_softc;
10812         int retval;
10813         uint32_t residx;
10814
10815         retval = 0;
10816
10817         mtx_assert(&lun->lun_lock, MA_OWNED);
10818
10819         /*
10820          * If this shelf is a secondary shelf controller, we may have to
10821          * reject some commands disallowed by HA mode and link state.
10822          */
10823         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
10824                 if (softc->ha_link == CTL_HA_LINK_OFFLINE &&
10825                     (entry->flags & CTL_CMD_FLAG_OK_ON_UNAVAIL) == 0) {
10826                         ctl_set_lun_unavail(ctsio);
10827                         retval = 1;
10828                         goto bailout;
10829                 }
10830                 if ((lun->flags & CTL_LUN_PEER_SC_PRIMARY) == 0 &&
10831                     (entry->flags & CTL_CMD_FLAG_OK_ON_UNAVAIL) == 0) {
10832                         ctl_set_lun_transit(ctsio);
10833                         retval = 1;
10834                         goto bailout;
10835                 }
10836                 if (softc->ha_mode == CTL_HA_MODE_ACT_STBY &&
10837                     (entry->flags & CTL_CMD_FLAG_OK_ON_STANDBY) == 0) {
10838                         ctl_set_lun_standby(ctsio);
10839                         retval = 1;
10840                         goto bailout;
10841                 }
10842
10843                 /* The rest of checks are only done on executing side */
10844                 if (softc->ha_mode == CTL_HA_MODE_XFER)
10845                         goto bailout;
10846         }
10847
10848         if (entry->pattern & CTL_LUN_PAT_WRITE) {
10849                 if (lun->be_lun &&
10850                     lun->be_lun->flags & CTL_LUN_FLAG_READONLY) {
10851                         ctl_set_hw_write_protected(ctsio);
10852                         retval = 1;
10853                         goto bailout;
10854                 }
10855                 if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT]
10856                     .eca_and_aen & SCP_SWP) != 0) {
10857                         ctl_set_sense(ctsio, /*current_error*/ 1,
10858                             /*sense_key*/ SSD_KEY_DATA_PROTECT,
10859                             /*asc*/ 0x27, /*ascq*/ 0x02, SSD_ELEM_NONE);
10860                         retval = 1;
10861                         goto bailout;
10862                 }
10863         }
10864
10865         /*
10866          * Check for a reservation conflict.  If this command isn't allowed
10867          * even on reserved LUNs, and if this initiator isn't the one who
10868          * reserved us, reject the command with a reservation conflict.
10869          */
10870         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
10871         if ((lun->flags & CTL_LUN_RESERVED)
10872          && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
10873                 if (lun->res_idx != residx) {
10874                         ctl_set_reservation_conflict(ctsio);
10875                         retval = 1;
10876                         goto bailout;
10877                 }
10878         }
10879
10880         if ((lun->flags & CTL_LUN_PR_RESERVED) == 0 ||
10881             (entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV)) {
10882                 /* No reservation or command is allowed. */;
10883         } else if ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_WRESV) &&
10884             (lun->res_type == SPR_TYPE_WR_EX ||
10885              lun->res_type == SPR_TYPE_WR_EX_RO ||
10886              lun->res_type == SPR_TYPE_WR_EX_AR)) {
10887                 /* The command is allowed for Write Exclusive resv. */;
10888         } else {
10889                 /*
10890                  * if we aren't registered or it's a res holder type
10891                  * reservation and this isn't the res holder then set a
10892                  * conflict.
10893                  */
10894                 if (ctl_get_prkey(lun, residx) == 0
10895                  || (residx != lun->pr_res_idx && lun->res_type < 4)) {
10896                         ctl_set_reservation_conflict(ctsio);
10897                         retval = 1;
10898                         goto bailout;
10899                 }
10900         }
10901
10902         if ((lun->flags & CTL_LUN_OFFLINE)
10903          && ((entry->flags & CTL_CMD_FLAG_OK_ON_STANDBY) == 0)) {
10904                 ctl_set_lun_not_ready(ctsio);
10905                 retval = 1;
10906                 goto bailout;
10907         }
10908
10909         if ((lun->flags & CTL_LUN_STOPPED)
10910          && ((entry->flags & CTL_CMD_FLAG_OK_ON_STOPPED) == 0)) {
10911                 /* "Logical unit not ready, initializing cmd. required" */
10912                 ctl_set_lun_stopped(ctsio);
10913                 retval = 1;
10914                 goto bailout;
10915         }
10916
10917         if ((lun->flags & CTL_LUN_INOPERABLE)
10918          && ((entry->flags & CTL_CMD_FLAG_OK_ON_INOPERABLE) == 0)) {
10919                 /* "Medium format corrupted" */
10920                 ctl_set_medium_format_corrupted(ctsio);
10921                 retval = 1;
10922                 goto bailout;
10923         }
10924
10925 bailout:
10926         return (retval);
10927 }
10928
10929 static void
10930 ctl_failover_io(union ctl_io *io, int have_lock)
10931 {
10932         ctl_set_busy(&io->scsiio);
10933         ctl_done(io);
10934 }
10935
10936 static void
10937 ctl_failover_lun(struct ctl_lun *lun)
10938 {
10939         struct ctl_softc *softc = lun->ctl_softc;
10940         struct ctl_io_hdr *io, *next_io;
10941
10942         CTL_DEBUG_PRINT(("FAILOVER for lun %ju\n", lun->lun));
10943         if (softc->ha_mode == CTL_HA_MODE_XFER) {
10944                 TAILQ_FOREACH_SAFE(io, &lun->ooa_queue, ooa_links, next_io) {
10945                         /* We are master */
10946                         if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
10947                                 if (io->flags & CTL_FLAG_IO_ACTIVE) {
10948                                         io->flags |= CTL_FLAG_ABORT;
10949                                 } else { /* This can be only due to DATAMOVE */
10950                                         io->msg_type = CTL_MSG_DATAMOVE_DONE;
10951                                         io->flags |= CTL_FLAG_IO_ACTIVE;
10952                                         io->port_status = 31340;
10953                                         ctl_enqueue_isc((union ctl_io *)io);
10954                                 }
10955                         }
10956                         /* We are slave */
10957                         if (io->flags & CTL_FLAG_SENT_2OTHER_SC) {
10958                                 io->flags &= ~CTL_FLAG_SENT_2OTHER_SC;
10959                                 if (io->flags & CTL_FLAG_IO_ACTIVE) {
10960                                         io->flags |= CTL_FLAG_FAILOVER;
10961                                 } else {
10962                                         ctl_set_busy(&((union ctl_io *)io)->
10963                                             scsiio);
10964                                         ctl_done((union ctl_io *)io);
10965                                 }
10966                         }
10967                 }
10968         } else { /* SERIALIZE modes */
10969                 TAILQ_FOREACH_SAFE(io, &lun->blocked_queue, blocked_links,
10970                     next_io) {
10971                         /* We are master */
10972                         if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
10973                                 TAILQ_REMOVE(&lun->blocked_queue, io,
10974                                     blocked_links);
10975                                 io->flags &= ~CTL_FLAG_BLOCKED;
10976                                 TAILQ_REMOVE(&lun->ooa_queue, io, ooa_links);
10977                                 ctl_free_io((union ctl_io *)io);
10978                         }
10979                 }
10980                 TAILQ_FOREACH_SAFE(io, &lun->ooa_queue, ooa_links, next_io) {
10981                         /* We are master */
10982                         if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
10983                                 TAILQ_REMOVE(&lun->ooa_queue, io, ooa_links);
10984                                 ctl_free_io((union ctl_io *)io);
10985                         }
10986                         /* We are slave */
10987                         if (io->flags & CTL_FLAG_SENT_2OTHER_SC) {
10988                                 io->flags &= ~CTL_FLAG_SENT_2OTHER_SC;
10989                                 if (!(io->flags & CTL_FLAG_IO_ACTIVE)) {
10990                                         ctl_set_busy(&((union ctl_io *)io)->
10991                                             scsiio);
10992                                         ctl_done((union ctl_io *)io);
10993                                 }
10994                         }
10995                 }
10996                 ctl_check_blocked(lun);
10997         }
10998 }
10999
11000 static int
11001 ctl_scsiio_precheck(struct ctl_softc *softc, struct ctl_scsiio *ctsio)
11002 {
11003         struct ctl_lun *lun;
11004         const struct ctl_cmd_entry *entry;
11005         uint32_t initidx, targ_lun;
11006         int retval;
11007
11008         retval = 0;
11009
11010         lun = NULL;
11011
11012         targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
11013         if ((targ_lun < CTL_MAX_LUNS)
11014          && ((lun = softc->ctl_luns[targ_lun]) != NULL)) {
11015                 /*
11016                  * If the LUN is invalid, pretend that it doesn't exist.
11017                  * It will go away as soon as all pending I/O has been
11018                  * completed.
11019                  */
11020                 mtx_lock(&lun->lun_lock);
11021                 if (lun->flags & CTL_LUN_DISABLED) {
11022                         mtx_unlock(&lun->lun_lock);
11023                         lun = NULL;
11024                         ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
11025                         ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
11026                 } else {
11027                         ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = lun;
11028                         ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr =
11029                                 lun->be_lun;
11030
11031                         /*
11032                          * Every I/O goes into the OOA queue for a
11033                          * particular LUN, and stays there until completion.
11034                          */
11035 #ifdef CTL_TIME_IO
11036                         if (TAILQ_EMPTY(&lun->ooa_queue)) {
11037                                 lun->idle_time += getsbinuptime() -
11038                                     lun->last_busy;
11039                         }
11040 #endif
11041                         TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr,
11042                             ooa_links);
11043                 }
11044         } else {
11045                 ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
11046                 ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
11047         }
11048
11049         /* Get command entry and return error if it is unsuppotyed. */
11050         entry = ctl_validate_command(ctsio);
11051         if (entry == NULL) {
11052                 if (lun)
11053                         mtx_unlock(&lun->lun_lock);
11054                 return (retval);
11055         }
11056
11057         ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
11058         ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
11059
11060         /*
11061          * Check to see whether we can send this command to LUNs that don't
11062          * exist.  This should pretty much only be the case for inquiry
11063          * and request sense.  Further checks, below, really require having
11064          * a LUN, so we can't really check the command anymore.  Just put
11065          * it on the rtr queue.
11066          */
11067         if (lun == NULL) {
11068                 if (entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) {
11069                         ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11070                         ctl_enqueue_rtr((union ctl_io *)ctsio);
11071                         return (retval);
11072                 }
11073
11074                 ctl_set_unsupported_lun(ctsio);
11075                 ctl_done((union ctl_io *)ctsio);
11076                 CTL_DEBUG_PRINT(("ctl_scsiio_precheck: bailing out due to invalid LUN\n"));
11077                 return (retval);
11078         } else {
11079                 /*
11080                  * Make sure we support this particular command on this LUN.
11081                  * e.g., we don't support writes to the control LUN.
11082                  */
11083                 if (!ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
11084                         mtx_unlock(&lun->lun_lock);
11085                         ctl_set_invalid_opcode(ctsio);
11086                         ctl_done((union ctl_io *)ctsio);
11087                         return (retval);
11088                 }
11089         }
11090
11091         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11092
11093 #ifdef CTL_WITH_CA
11094         /*
11095          * If we've got a request sense, it'll clear the contingent
11096          * allegiance condition.  Otherwise, if we have a CA condition for
11097          * this initiator, clear it, because it sent down a command other
11098          * than request sense.
11099          */
11100         if ((ctsio->cdb[0] != REQUEST_SENSE)
11101          && (ctl_is_set(lun->have_ca, initidx)))
11102                 ctl_clear_mask(lun->have_ca, initidx);
11103 #endif
11104
11105         /*
11106          * If the command has this flag set, it handles its own unit
11107          * attention reporting, we shouldn't do anything.  Otherwise we
11108          * check for any pending unit attentions, and send them back to the
11109          * initiator.  We only do this when a command initially comes in,
11110          * not when we pull it off the blocked queue.
11111          *
11112          * According to SAM-3, section 5.3.2, the order that things get
11113          * presented back to the host is basically unit attentions caused
11114          * by some sort of reset event, busy status, reservation conflicts
11115          * or task set full, and finally any other status.
11116          *
11117          * One issue here is that some of the unit attentions we report
11118          * don't fall into the "reset" category (e.g. "reported luns data
11119          * has changed").  So reporting it here, before the reservation
11120          * check, may be technically wrong.  I guess the only thing to do
11121          * would be to check for and report the reset events here, and then
11122          * check for the other unit attention types after we check for a
11123          * reservation conflict.
11124          *
11125          * XXX KDM need to fix this
11126          */
11127         if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
11128                 ctl_ua_type ua_type;
11129                 scsi_sense_data_type sense_format;
11130
11131                 if (lun->flags & CTL_LUN_SENSE_DESC)
11132                         sense_format = SSD_TYPE_DESC;
11133                 else
11134                         sense_format = SSD_TYPE_FIXED;
11135
11136                 ua_type = ctl_build_ua(lun, initidx, &ctsio->sense_data,
11137                     sense_format);
11138                 if (ua_type != CTL_UA_NONE) {
11139                         mtx_unlock(&lun->lun_lock);
11140                         ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
11141                         ctsio->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
11142                         ctsio->sense_len = SSD_FULL_SIZE;
11143                         ctl_done((union ctl_io *)ctsio);
11144                         return (retval);
11145                 }
11146         }
11147
11148
11149         if (ctl_scsiio_lun_check(lun, entry, ctsio) != 0) {
11150                 mtx_unlock(&lun->lun_lock);
11151                 ctl_done((union ctl_io *)ctsio);
11152                 return (retval);
11153         }
11154
11155         /*
11156          * XXX CHD this is where we want to send IO to other side if
11157          * this LUN is secondary on this SC. We will need to make a copy
11158          * of the IO and flag the IO on this side as SENT_2OTHER and the flag
11159          * the copy we send as FROM_OTHER.
11160          * We also need to stuff the address of the original IO so we can
11161          * find it easily. Something similar will need be done on the other
11162          * side so when we are done we can find the copy.
11163          */
11164         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
11165             (lun->flags & CTL_LUN_PEER_SC_PRIMARY) != 0) {
11166                 union ctl_ha_msg msg_info;
11167                 int isc_retval;
11168
11169                 ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11170                 ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11171                 mtx_unlock(&lun->lun_lock);
11172
11173                 msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
11174                 msg_info.hdr.original_sc = (union ctl_io *)ctsio;
11175                 msg_info.hdr.serializing_sc = NULL;
11176                 msg_info.hdr.nexus = ctsio->io_hdr.nexus;
11177                 msg_info.scsi.tag_num = ctsio->tag_num;
11178                 msg_info.scsi.tag_type = ctsio->tag_type;
11179                 msg_info.scsi.cdb_len = ctsio->cdb_len;
11180                 memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
11181
11182                 if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11183                     sizeof(msg_info.scsi) - sizeof(msg_info.scsi.sense_data),
11184                     M_WAITOK)) > CTL_HA_STATUS_SUCCESS) {
11185                         ctl_set_busy(ctsio);
11186                         ctl_done((union ctl_io *)ctsio);
11187                         return (retval);
11188                 }
11189                 return (retval);
11190         }
11191
11192         switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
11193                               (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr,
11194                               ctl_ooaq, ooa_links))) {
11195         case CTL_ACTION_BLOCK:
11196                 ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
11197                 TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
11198                                   blocked_links);
11199                 mtx_unlock(&lun->lun_lock);
11200                 return (retval);
11201         case CTL_ACTION_PASS:
11202         case CTL_ACTION_SKIP:
11203                 ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11204                 mtx_unlock(&lun->lun_lock);
11205                 ctl_enqueue_rtr((union ctl_io *)ctsio);
11206                 break;
11207         case CTL_ACTION_OVERLAP:
11208                 mtx_unlock(&lun->lun_lock);
11209                 ctl_set_overlapped_cmd(ctsio);
11210                 ctl_done((union ctl_io *)ctsio);
11211                 break;
11212         case CTL_ACTION_OVERLAP_TAG:
11213                 mtx_unlock(&lun->lun_lock);
11214                 ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
11215                 ctl_done((union ctl_io *)ctsio);
11216                 break;
11217         case CTL_ACTION_ERROR:
11218         default:
11219                 mtx_unlock(&lun->lun_lock);
11220                 ctl_set_internal_failure(ctsio,
11221                                          /*sks_valid*/ 0,
11222                                          /*retry_count*/ 0);
11223                 ctl_done((union ctl_io *)ctsio);
11224                 break;
11225         }
11226         return (retval);
11227 }
11228
11229 const struct ctl_cmd_entry *
11230 ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa)
11231 {
11232         const struct ctl_cmd_entry *entry;
11233         int service_action;
11234
11235         entry = &ctl_cmd_table[ctsio->cdb[0]];
11236         if (sa)
11237                 *sa = ((entry->flags & CTL_CMD_FLAG_SA5) != 0);
11238         if (entry->flags & CTL_CMD_FLAG_SA5) {
11239                 service_action = ctsio->cdb[1] & SERVICE_ACTION_MASK;
11240                 entry = &((const struct ctl_cmd_entry *)
11241                     entry->execute)[service_action];
11242         }
11243         return (entry);
11244 }
11245
11246 const struct ctl_cmd_entry *
11247 ctl_validate_command(struct ctl_scsiio *ctsio)
11248 {
11249         const struct ctl_cmd_entry *entry;
11250         int i, sa;
11251         uint8_t diff;
11252
11253         entry = ctl_get_cmd_entry(ctsio, &sa);
11254         if (entry->execute == NULL) {
11255                 if (sa)
11256                         ctl_set_invalid_field(ctsio,
11257                                               /*sks_valid*/ 1,
11258                                               /*command*/ 1,
11259                                               /*field*/ 1,
11260                                               /*bit_valid*/ 1,
11261                                               /*bit*/ 4);
11262                 else
11263                         ctl_set_invalid_opcode(ctsio);
11264                 ctl_done((union ctl_io *)ctsio);
11265                 return (NULL);
11266         }
11267         KASSERT(entry->length > 0,
11268             ("Not defined length for command 0x%02x/0x%02x",
11269              ctsio->cdb[0], ctsio->cdb[1]));
11270         for (i = 1; i < entry->length; i++) {
11271                 diff = ctsio->cdb[i] & ~entry->usage[i - 1];
11272                 if (diff == 0)
11273                         continue;
11274                 ctl_set_invalid_field(ctsio,
11275                                       /*sks_valid*/ 1,
11276                                       /*command*/ 1,
11277                                       /*field*/ i,
11278                                       /*bit_valid*/ 1,
11279                                       /*bit*/ fls(diff) - 1);
11280                 ctl_done((union ctl_io *)ctsio);
11281                 return (NULL);
11282         }
11283         return (entry);
11284 }
11285
11286 static int
11287 ctl_cmd_applicable(uint8_t lun_type, const struct ctl_cmd_entry *entry)
11288 {
11289
11290         switch (lun_type) {
11291         case T_PROCESSOR:
11292                 if (((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0) &&
11293                     ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) == 0))
11294                         return (0);
11295                 break;
11296         case T_DIRECT:
11297                 if (((entry->flags & CTL_CMD_FLAG_OK_ON_SLUN) == 0) &&
11298                     ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) == 0))
11299                         return (0);
11300                 break;
11301         default:
11302                 return (0);
11303         }
11304         return (1);
11305 }
11306
11307 static int
11308 ctl_scsiio(struct ctl_scsiio *ctsio)
11309 {
11310         int retval;
11311         const struct ctl_cmd_entry *entry;
11312
11313         retval = CTL_RETVAL_COMPLETE;
11314
11315         CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
11316
11317         entry = ctl_get_cmd_entry(ctsio, NULL);
11318
11319         /*
11320          * If this I/O has been aborted, just send it straight to
11321          * ctl_done() without executing it.
11322          */
11323         if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
11324                 ctl_done((union ctl_io *)ctsio);
11325                 goto bailout;
11326         }
11327
11328         /*
11329          * All the checks should have been handled by ctl_scsiio_precheck().
11330          * We should be clear now to just execute the I/O.
11331          */
11332         retval = entry->execute(ctsio);
11333
11334 bailout:
11335         return (retval);
11336 }
11337
11338 /*
11339  * Since we only implement one target right now, a bus reset simply resets
11340  * our single target.
11341  */
11342 static int
11343 ctl_bus_reset(struct ctl_softc *softc, union ctl_io *io)
11344 {
11345         return(ctl_target_reset(softc, io, CTL_UA_BUS_RESET));
11346 }
11347
11348 static int
11349 ctl_target_reset(struct ctl_softc *softc, union ctl_io *io,
11350                  ctl_ua_type ua_type)
11351 {
11352         struct ctl_port *port;
11353         struct ctl_lun *lun;
11354         int retval;
11355
11356         if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11357                 union ctl_ha_msg msg_info;
11358
11359                 msg_info.hdr.nexus = io->io_hdr.nexus;
11360                 if (ua_type==CTL_UA_TARG_RESET)
11361                         msg_info.task.task_action = CTL_TASK_TARGET_RESET;
11362                 else
11363                         msg_info.task.task_action = CTL_TASK_BUS_RESET;
11364                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11365                 msg_info.hdr.original_sc = NULL;
11366                 msg_info.hdr.serializing_sc = NULL;
11367                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11368                     sizeof(msg_info.task), M_WAITOK);
11369         }
11370         retval = 0;
11371
11372         mtx_lock(&softc->ctl_lock);
11373         port = softc->ctl_ports[io->io_hdr.nexus.targ_port];
11374         STAILQ_FOREACH(lun, &softc->lun_list, links) {
11375                 if (port != NULL &&
11376                     ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
11377                         continue;
11378                 retval += ctl_lun_reset(lun, io, ua_type);
11379         }
11380         mtx_unlock(&softc->ctl_lock);
11381
11382         return (retval);
11383 }
11384
11385 /*
11386  * The LUN should always be set.  The I/O is optional, and is used to
11387  * distinguish between I/Os sent by this initiator, and by other
11388  * initiators.  We set unit attention for initiators other than this one.
11389  * SAM-3 is vague on this point.  It does say that a unit attention should
11390  * be established for other initiators when a LUN is reset (see section
11391  * 5.7.3), but it doesn't specifically say that the unit attention should
11392  * be established for this particular initiator when a LUN is reset.  Here
11393  * is the relevant text, from SAM-3 rev 8:
11394  *
11395  * 5.7.2 When a SCSI initiator port aborts its own tasks
11396  *
11397  * When a SCSI initiator port causes its own task(s) to be aborted, no
11398  * notification that the task(s) have been aborted shall be returned to
11399  * the SCSI initiator port other than the completion response for the
11400  * command or task management function action that caused the task(s) to
11401  * be aborted and notification(s) associated with related effects of the
11402  * action (e.g., a reset unit attention condition).
11403  *
11404  * XXX KDM for now, we're setting unit attention for all initiators.
11405  */
11406 static int
11407 ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io, ctl_ua_type ua_type)
11408 {
11409         union ctl_io *xio;
11410 #if 0
11411         uint32_t initidx;
11412 #endif
11413 #ifdef CTL_WITH_CA
11414         int i;
11415 #endif
11416
11417         mtx_lock(&lun->lun_lock);
11418         /*
11419          * Run through the OOA queue and abort each I/O.
11420          */
11421         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11422              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11423                 xio->io_hdr.flags |= CTL_FLAG_ABORT | CTL_FLAG_ABORT_STATUS;
11424         }
11425
11426         /*
11427          * This version sets unit attention for every
11428          */
11429 #if 0
11430         initidx = ctl_get_initindex(&io->io_hdr.nexus);
11431         ctl_est_ua_all(lun, initidx, ua_type);
11432 #else
11433         ctl_est_ua_all(lun, -1, ua_type);
11434 #endif
11435
11436         /*
11437          * A reset (any kind, really) clears reservations established with
11438          * RESERVE/RELEASE.  It does not clear reservations established
11439          * with PERSISTENT RESERVE OUT, but we don't support that at the
11440          * moment anyway.  See SPC-2, section 5.6.  SPC-3 doesn't address
11441          * reservations made with the RESERVE/RELEASE commands, because
11442          * those commands are obsolete in SPC-3.
11443          */
11444         lun->flags &= ~CTL_LUN_RESERVED;
11445
11446 #ifdef CTL_WITH_CA
11447         for (i = 0; i < CTL_MAX_INITIATORS; i++)
11448                 ctl_clear_mask(lun->have_ca, i);
11449 #endif
11450         mtx_unlock(&lun->lun_lock);
11451
11452         return (0);
11453 }
11454
11455 static void
11456 ctl_abort_tasks_lun(struct ctl_lun *lun, uint32_t targ_port, uint32_t init_id,
11457     int other_sc)
11458 {
11459         union ctl_io *xio;
11460
11461         mtx_assert(&lun->lun_lock, MA_OWNED);
11462
11463         /*
11464          * Run through the OOA queue and attempt to find the given I/O.
11465          * The target port, initiator ID, tag type and tag number have to
11466          * match the values that we got from the initiator.  If we have an
11467          * untagged command to abort, simply abort the first untagged command
11468          * we come to.  We only allow one untagged command at a time of course.
11469          */
11470         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11471              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11472
11473                 if ((targ_port == UINT32_MAX ||
11474                      targ_port == xio->io_hdr.nexus.targ_port) &&
11475                     (init_id == UINT32_MAX ||
11476                      init_id == xio->io_hdr.nexus.initid)) {
11477                         if (targ_port != xio->io_hdr.nexus.targ_port ||
11478                             init_id != xio->io_hdr.nexus.initid)
11479                                 xio->io_hdr.flags |= CTL_FLAG_ABORT_STATUS;
11480                         xio->io_hdr.flags |= CTL_FLAG_ABORT;
11481                         if (!other_sc && !(lun->flags & CTL_LUN_PRIMARY_SC)) {
11482                                 union ctl_ha_msg msg_info;
11483
11484                                 msg_info.hdr.nexus = xio->io_hdr.nexus;
11485                                 msg_info.task.task_action = CTL_TASK_ABORT_TASK;
11486                                 msg_info.task.tag_num = xio->scsiio.tag_num;
11487                                 msg_info.task.tag_type = xio->scsiio.tag_type;
11488                                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11489                                 msg_info.hdr.original_sc = NULL;
11490                                 msg_info.hdr.serializing_sc = NULL;
11491                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11492                                     sizeof(msg_info.task), M_NOWAIT);
11493                         }
11494                 }
11495         }
11496 }
11497
11498 static int
11499 ctl_abort_task_set(union ctl_io *io)
11500 {
11501         struct ctl_softc *softc = control_softc;
11502         struct ctl_lun *lun;
11503         uint32_t targ_lun;
11504
11505         /*
11506          * Look up the LUN.
11507          */
11508         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11509         mtx_lock(&softc->ctl_lock);
11510         if ((targ_lun < CTL_MAX_LUNS) && (softc->ctl_luns[targ_lun] != NULL))
11511                 lun = softc->ctl_luns[targ_lun];
11512         else {
11513                 mtx_unlock(&softc->ctl_lock);
11514                 return (1);
11515         }
11516
11517         mtx_lock(&lun->lun_lock);
11518         mtx_unlock(&softc->ctl_lock);
11519         if (io->taskio.task_action == CTL_TASK_ABORT_TASK_SET) {
11520                 ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
11521                     io->io_hdr.nexus.initid,
11522                     (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
11523         } else { /* CTL_TASK_CLEAR_TASK_SET */
11524                 ctl_abort_tasks_lun(lun, UINT32_MAX, UINT32_MAX,
11525                     (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
11526         }
11527         mtx_unlock(&lun->lun_lock);
11528         return (0);
11529 }
11530
11531 static int
11532 ctl_i_t_nexus_reset(union ctl_io *io)
11533 {
11534         struct ctl_softc *softc = control_softc;
11535         struct ctl_lun *lun;
11536         uint32_t initidx;
11537
11538         if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11539                 union ctl_ha_msg msg_info;
11540
11541                 msg_info.hdr.nexus = io->io_hdr.nexus;
11542                 msg_info.task.task_action = CTL_TASK_I_T_NEXUS_RESET;
11543                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11544                 msg_info.hdr.original_sc = NULL;
11545                 msg_info.hdr.serializing_sc = NULL;
11546                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11547                     sizeof(msg_info.task), M_WAITOK);
11548         }
11549
11550         initidx = ctl_get_initindex(&io->io_hdr.nexus);
11551         mtx_lock(&softc->ctl_lock);
11552         STAILQ_FOREACH(lun, &softc->lun_list, links) {
11553                 mtx_lock(&lun->lun_lock);
11554                 ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
11555                     io->io_hdr.nexus.initid, 1);
11556 #ifdef CTL_WITH_CA
11557                 ctl_clear_mask(lun->have_ca, initidx);
11558 #endif
11559                 if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == initidx))
11560                         lun->flags &= ~CTL_LUN_RESERVED;
11561                 ctl_est_ua(lun, initidx, CTL_UA_I_T_NEXUS_LOSS);
11562                 mtx_unlock(&lun->lun_lock);
11563         }
11564         mtx_unlock(&softc->ctl_lock);
11565         return (0);
11566 }
11567
11568 static int
11569 ctl_abort_task(union ctl_io *io)
11570 {
11571         union ctl_io *xio;
11572         struct ctl_lun *lun;
11573         struct ctl_softc *softc;
11574 #if 0
11575         struct sbuf sb;
11576         char printbuf[128];
11577 #endif
11578         int found;
11579         uint32_t targ_lun;
11580
11581         softc = control_softc;
11582         found = 0;
11583
11584         /*
11585          * Look up the LUN.
11586          */
11587         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11588         mtx_lock(&softc->ctl_lock);
11589         if ((targ_lun < CTL_MAX_LUNS)
11590          && (softc->ctl_luns[targ_lun] != NULL))
11591                 lun = softc->ctl_luns[targ_lun];
11592         else {
11593                 mtx_unlock(&softc->ctl_lock);
11594                 return (1);
11595         }
11596
11597 #if 0
11598         printf("ctl_abort_task: called for lun %lld, tag %d type %d\n",
11599                lun->lun, io->taskio.tag_num, io->taskio.tag_type);
11600 #endif
11601
11602         mtx_lock(&lun->lun_lock);
11603         mtx_unlock(&softc->ctl_lock);
11604         /*
11605          * Run through the OOA queue and attempt to find the given I/O.
11606          * The target port, initiator ID, tag type and tag number have to
11607          * match the values that we got from the initiator.  If we have an
11608          * untagged command to abort, simply abort the first untagged command
11609          * we come to.  We only allow one untagged command at a time of course.
11610          */
11611         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11612              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11613 #if 0
11614                 sbuf_new(&sb, printbuf, sizeof(printbuf), SBUF_FIXEDLEN);
11615
11616                 sbuf_printf(&sb, "LUN %lld tag %d type %d%s%s%s%s: ",
11617                             lun->lun, xio->scsiio.tag_num,
11618                             xio->scsiio.tag_type,
11619                             (xio->io_hdr.blocked_links.tqe_prev
11620                             == NULL) ? "" : " BLOCKED",
11621                             (xio->io_hdr.flags &
11622                             CTL_FLAG_DMA_INPROG) ? " DMA" : "",
11623                             (xio->io_hdr.flags &
11624                             CTL_FLAG_ABORT) ? " ABORT" : "",
11625                             (xio->io_hdr.flags &
11626                             CTL_FLAG_IS_WAS_ON_RTR ? " RTR" : ""));
11627                 ctl_scsi_command_string(&xio->scsiio, NULL, &sb);
11628                 sbuf_finish(&sb);
11629                 printf("%s\n", sbuf_data(&sb));
11630 #endif
11631
11632                 if ((xio->io_hdr.nexus.targ_port != io->io_hdr.nexus.targ_port)
11633                  || (xio->io_hdr.nexus.initid != io->io_hdr.nexus.initid)
11634                  || (xio->io_hdr.flags & CTL_FLAG_ABORT))
11635                         continue;
11636
11637                 /*
11638                  * If the abort says that the task is untagged, the
11639                  * task in the queue must be untagged.  Otherwise,
11640                  * we just check to see whether the tag numbers
11641                  * match.  This is because the QLogic firmware
11642                  * doesn't pass back the tag type in an abort
11643                  * request.
11644                  */
11645 #if 0
11646                 if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
11647                   && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
11648                  || (xio->scsiio.tag_num == io->taskio.tag_num))
11649 #endif
11650                 /*
11651                  * XXX KDM we've got problems with FC, because it
11652                  * doesn't send down a tag type with aborts.  So we
11653                  * can only really go by the tag number...
11654                  * This may cause problems with parallel SCSI.
11655                  * Need to figure that out!!
11656                  */
11657                 if (xio->scsiio.tag_num == io->taskio.tag_num) {
11658                         xio->io_hdr.flags |= CTL_FLAG_ABORT;
11659                         found = 1;
11660                         if ((io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) == 0 &&
11661                             !(lun->flags & CTL_LUN_PRIMARY_SC)) {
11662                                 union ctl_ha_msg msg_info;
11663
11664                                 msg_info.hdr.nexus = io->io_hdr.nexus;
11665                                 msg_info.task.task_action = CTL_TASK_ABORT_TASK;
11666                                 msg_info.task.tag_num = io->taskio.tag_num;
11667                                 msg_info.task.tag_type = io->taskio.tag_type;
11668                                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11669                                 msg_info.hdr.original_sc = NULL;
11670                                 msg_info.hdr.serializing_sc = NULL;
11671 #if 0
11672                                 printf("Sent Abort to other side\n");
11673 #endif
11674                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11675                                     sizeof(msg_info.task), M_NOWAIT);
11676                         }
11677 #if 0
11678                         printf("ctl_abort_task: found I/O to abort\n");
11679 #endif
11680                 }
11681         }
11682         mtx_unlock(&lun->lun_lock);
11683
11684         if (found == 0) {
11685                 /*
11686                  * This isn't really an error.  It's entirely possible for
11687                  * the abort and command completion to cross on the wire.
11688                  * This is more of an informative/diagnostic error.
11689                  */
11690 #if 0
11691                 printf("ctl_abort_task: ABORT sent for nonexistent I/O: "
11692                        "%u:%u:%u tag %d type %d\n",
11693                        io->io_hdr.nexus.initid,
11694                        io->io_hdr.nexus.targ_port,
11695                        io->io_hdr.nexus.targ_lun, io->taskio.tag_num,
11696                        io->taskio.tag_type);
11697 #endif
11698         }
11699         return (0);
11700 }
11701
11702 static void
11703 ctl_run_task(union ctl_io *io)
11704 {
11705         struct ctl_softc *softc = control_softc;
11706         int retval = 1;
11707         const char *task_desc;
11708
11709         CTL_DEBUG_PRINT(("ctl_run_task\n"));
11710
11711         KASSERT(io->io_hdr.io_type == CTL_IO_TASK,
11712             ("ctl_run_task: Unextected io_type %d\n",
11713              io->io_hdr.io_type));
11714
11715         task_desc = ctl_scsi_task_string(&io->taskio);
11716         if (task_desc != NULL) {
11717 #ifdef NEEDTOPORT
11718                 csevent_log(CSC_CTL | CSC_SHELF_SW |
11719                             CTL_TASK_REPORT,
11720                             csevent_LogType_Trace,
11721                             csevent_Severity_Information,
11722                             csevent_AlertLevel_Green,
11723                             csevent_FRU_Firmware,
11724                             csevent_FRU_Unknown,
11725                             "CTL: received task: %s",task_desc);
11726 #endif
11727         } else {
11728 #ifdef NEEDTOPORT
11729                 csevent_log(CSC_CTL | CSC_SHELF_SW |
11730                             CTL_TASK_REPORT,
11731                             csevent_LogType_Trace,
11732                             csevent_Severity_Information,
11733                             csevent_AlertLevel_Green,
11734                             csevent_FRU_Firmware,
11735                             csevent_FRU_Unknown,
11736                             "CTL: received unknown task "
11737                             "type: %d (%#x)",
11738                             io->taskio.task_action,
11739                             io->taskio.task_action);
11740 #endif
11741         }
11742         switch (io->taskio.task_action) {
11743         case CTL_TASK_ABORT_TASK:
11744                 retval = ctl_abort_task(io);
11745                 break;
11746         case CTL_TASK_ABORT_TASK_SET:
11747         case CTL_TASK_CLEAR_TASK_SET:
11748                 retval = ctl_abort_task_set(io);
11749                 break;
11750         case CTL_TASK_CLEAR_ACA:
11751                 break;
11752         case CTL_TASK_I_T_NEXUS_RESET:
11753                 retval = ctl_i_t_nexus_reset(io);
11754                 break;
11755         case CTL_TASK_LUN_RESET: {
11756                 struct ctl_lun *lun;
11757                 uint32_t targ_lun;
11758
11759                 targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11760                 mtx_lock(&softc->ctl_lock);
11761                 if ((targ_lun < CTL_MAX_LUNS)
11762                  && (softc->ctl_luns[targ_lun] != NULL))
11763                         lun = softc->ctl_luns[targ_lun];
11764                 else {
11765                         mtx_unlock(&softc->ctl_lock);
11766                         retval = 1;
11767                         break;
11768                 }
11769                 retval = ctl_lun_reset(lun, io, CTL_UA_LUN_RESET);
11770                 mtx_unlock(&softc->ctl_lock);
11771
11772                 if ((io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) == 0) {
11773                         union ctl_ha_msg msg_info;
11774
11775                         msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11776                         msg_info.hdr.nexus = io->io_hdr.nexus;
11777                         msg_info.task.task_action = CTL_TASK_LUN_RESET;
11778                         msg_info.hdr.original_sc = NULL;
11779                         msg_info.hdr.serializing_sc = NULL;
11780                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11781                             sizeof(msg_info.task), M_WAITOK);
11782                 }
11783                 break;
11784         }
11785         case CTL_TASK_TARGET_RESET:
11786                 retval = ctl_target_reset(softc, io, CTL_UA_TARG_RESET);
11787                 break;
11788         case CTL_TASK_BUS_RESET:
11789                 retval = ctl_bus_reset(softc, io);
11790                 break;
11791         case CTL_TASK_PORT_LOGIN:
11792                 break;
11793         case CTL_TASK_PORT_LOGOUT:
11794                 break;
11795         default:
11796                 printf("ctl_run_task: got unknown task management event %d\n",
11797                        io->taskio.task_action);
11798                 break;
11799         }
11800         if (retval == 0)
11801                 io->io_hdr.status = CTL_SUCCESS;
11802         else
11803                 io->io_hdr.status = CTL_ERROR;
11804         ctl_done(io);
11805 }
11806
11807 /*
11808  * For HA operation.  Handle commands that come in from the other
11809  * controller.
11810  */
11811 static void
11812 ctl_handle_isc(union ctl_io *io)
11813 {
11814         int free_io;
11815         struct ctl_lun *lun;
11816         struct ctl_softc *softc;
11817         uint32_t targ_lun;
11818
11819         softc = control_softc;
11820
11821         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11822         lun = softc->ctl_luns[targ_lun];
11823
11824         switch (io->io_hdr.msg_type) {
11825         case CTL_MSG_SERIALIZE:
11826                 free_io = ctl_serialize_other_sc_cmd(&io->scsiio);
11827                 break;
11828         case CTL_MSG_R2R: {
11829                 const struct ctl_cmd_entry *entry;
11830
11831                 /*
11832                  * This is only used in SER_ONLY mode.
11833                  */
11834                 free_io = 0;
11835                 entry = ctl_get_cmd_entry(&io->scsiio, NULL);
11836                 mtx_lock(&lun->lun_lock);
11837                 if (ctl_scsiio_lun_check(lun,
11838                     entry, (struct ctl_scsiio *)io) != 0) {
11839                         mtx_unlock(&lun->lun_lock);
11840                         ctl_done(io);
11841                         break;
11842                 }
11843                 io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11844                 mtx_unlock(&lun->lun_lock);
11845                 ctl_enqueue_rtr(io);
11846                 break;
11847         }
11848         case CTL_MSG_FINISH_IO:
11849                 if (softc->ha_mode == CTL_HA_MODE_XFER) {
11850                         free_io = 0;
11851                         ctl_done(io);
11852                 } else {
11853                         free_io = 1;
11854                         mtx_lock(&lun->lun_lock);
11855                         TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr,
11856                                      ooa_links);
11857                         ctl_check_blocked(lun);
11858                         mtx_unlock(&lun->lun_lock);
11859                 }
11860                 break;
11861         case CTL_MSG_PERS_ACTION:
11862                 ctl_hndl_per_res_out_on_other_sc(
11863                         (union ctl_ha_msg *)&io->presio.pr_msg);
11864                 free_io = 1;
11865                 break;
11866         case CTL_MSG_BAD_JUJU:
11867                 free_io = 0;
11868                 ctl_done(io);
11869                 break;
11870         case CTL_MSG_DATAMOVE:
11871                 /* Only used in XFER mode */
11872                 free_io = 0;
11873                 ctl_datamove_remote(io);
11874                 break;
11875         case CTL_MSG_DATAMOVE_DONE:
11876                 /* Only used in XFER mode */
11877                 free_io = 0;
11878                 io->scsiio.be_move_done(io);
11879                 break;
11880         case CTL_MSG_FAILOVER:
11881                 mtx_lock(&lun->lun_lock);
11882                 ctl_failover_lun(lun);
11883                 mtx_unlock(&lun->lun_lock);
11884                 free_io = 1;
11885                 break;
11886         default:
11887                 free_io = 1;
11888                 printf("%s: Invalid message type %d\n",
11889                        __func__, io->io_hdr.msg_type);
11890                 break;
11891         }
11892         if (free_io)
11893                 ctl_free_io(io);
11894
11895 }
11896
11897
11898 /*
11899  * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
11900  * there is no match.
11901  */
11902 static ctl_lun_error_pattern
11903 ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
11904 {
11905         const struct ctl_cmd_entry *entry;
11906         ctl_lun_error_pattern filtered_pattern, pattern;
11907
11908         pattern = desc->error_pattern;
11909
11910         /*
11911          * XXX KDM we need more data passed into this function to match a
11912          * custom pattern, and we actually need to implement custom pattern
11913          * matching.
11914          */
11915         if (pattern & CTL_LUN_PAT_CMD)
11916                 return (CTL_LUN_PAT_CMD);
11917
11918         if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
11919                 return (CTL_LUN_PAT_ANY);
11920
11921         entry = ctl_get_cmd_entry(ctsio, NULL);
11922
11923         filtered_pattern = entry->pattern & pattern;
11924
11925         /*
11926          * If the user requested specific flags in the pattern (e.g.
11927          * CTL_LUN_PAT_RANGE), make sure the command supports all of those
11928          * flags.
11929          *
11930          * If the user did not specify any flags, it doesn't matter whether
11931          * or not the command supports the flags.
11932          */
11933         if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
11934              (pattern & ~CTL_LUN_PAT_MASK))
11935                 return (CTL_LUN_PAT_NONE);
11936
11937         /*
11938          * If the user asked for a range check, see if the requested LBA
11939          * range overlaps with this command's LBA range.
11940          */
11941         if (filtered_pattern & CTL_LUN_PAT_RANGE) {
11942                 uint64_t lba1;
11943                 uint64_t len1;
11944                 ctl_action action;
11945                 int retval;
11946
11947                 retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
11948                 if (retval != 0)
11949                         return (CTL_LUN_PAT_NONE);
11950
11951                 action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
11952                                               desc->lba_range.len, FALSE);
11953                 /*
11954                  * A "pass" means that the LBA ranges don't overlap, so
11955                  * this doesn't match the user's range criteria.
11956                  */
11957                 if (action == CTL_ACTION_PASS)
11958                         return (CTL_LUN_PAT_NONE);
11959         }
11960
11961         return (filtered_pattern);
11962 }
11963
11964 static void
11965 ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
11966 {
11967         struct ctl_error_desc *desc, *desc2;
11968
11969         mtx_assert(&lun->lun_lock, MA_OWNED);
11970
11971         STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
11972                 ctl_lun_error_pattern pattern;
11973                 /*
11974                  * Check to see whether this particular command matches
11975                  * the pattern in the descriptor.
11976                  */
11977                 pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
11978                 if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
11979                         continue;
11980
11981                 switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
11982                 case CTL_LUN_INJ_ABORTED:
11983                         ctl_set_aborted(&io->scsiio);
11984                         break;
11985                 case CTL_LUN_INJ_MEDIUM_ERR:
11986                         ctl_set_medium_error(&io->scsiio);
11987                         break;
11988                 case CTL_LUN_INJ_UA:
11989                         /* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
11990                          * OCCURRED */
11991                         ctl_set_ua(&io->scsiio, 0x29, 0x00);
11992                         break;
11993                 case CTL_LUN_INJ_CUSTOM:
11994                         /*
11995                          * We're assuming the user knows what he is doing.
11996                          * Just copy the sense information without doing
11997                          * checks.
11998                          */
11999                         bcopy(&desc->custom_sense, &io->scsiio.sense_data,
12000                               MIN(sizeof(desc->custom_sense),
12001                                   sizeof(io->scsiio.sense_data)));
12002                         io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
12003                         io->scsiio.sense_len = SSD_FULL_SIZE;
12004                         io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
12005                         break;
12006                 case CTL_LUN_INJ_NONE:
12007                 default:
12008                         /*
12009                          * If this is an error injection type we don't know
12010                          * about, clear the continuous flag (if it is set)
12011                          * so it will get deleted below.
12012                          */
12013                         desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
12014                         break;
12015                 }
12016                 /*
12017                  * By default, each error injection action is a one-shot
12018                  */
12019                 if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
12020                         continue;
12021
12022                 STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
12023
12024                 free(desc, M_CTL);
12025         }
12026 }
12027
12028 #ifdef CTL_IO_DELAY
12029 static void
12030 ctl_datamove_timer_wakeup(void *arg)
12031 {
12032         union ctl_io *io;
12033
12034         io = (union ctl_io *)arg;
12035
12036         ctl_datamove(io);
12037 }
12038 #endif /* CTL_IO_DELAY */
12039
12040 void
12041 ctl_datamove(union ctl_io *io)
12042 {
12043         void (*fe_datamove)(union ctl_io *io);
12044
12045         mtx_assert(&control_softc->ctl_lock, MA_NOTOWNED);
12046
12047         CTL_DEBUG_PRINT(("ctl_datamove\n"));
12048
12049 #ifdef CTL_TIME_IO
12050         if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12051                 char str[256];
12052                 char path_str[64];
12053                 struct sbuf sb;
12054
12055                 ctl_scsi_path_string(io, path_str, sizeof(path_str));
12056                 sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12057
12058                 sbuf_cat(&sb, path_str);
12059                 switch (io->io_hdr.io_type) {
12060                 case CTL_IO_SCSI:
12061                         ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12062                         sbuf_printf(&sb, "\n");
12063                         sbuf_cat(&sb, path_str);
12064                         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12065                                     io->scsiio.tag_num, io->scsiio.tag_type);
12066                         break;
12067                 case CTL_IO_TASK:
12068                         sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12069                                     "Tag Type: %d\n", io->taskio.task_action,
12070                                     io->taskio.tag_num, io->taskio.tag_type);
12071                         break;
12072                 default:
12073                         printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12074                         panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12075                         break;
12076                 }
12077                 sbuf_cat(&sb, path_str);
12078                 sbuf_printf(&sb, "ctl_datamove: %jd seconds\n",
12079                             (intmax_t)time_uptime - io->io_hdr.start_time);
12080                 sbuf_finish(&sb);
12081                 printf("%s", sbuf_data(&sb));
12082         }
12083 #endif /* CTL_TIME_IO */
12084
12085 #ifdef CTL_IO_DELAY
12086         if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
12087                 io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
12088         } else {
12089                 struct ctl_lun *lun;
12090
12091                 lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12092                 if ((lun != NULL)
12093                  && (lun->delay_info.datamove_delay > 0)) {
12094
12095                         callout_init(&io->io_hdr.delay_callout, /*mpsafe*/ 1);
12096                         io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
12097                         callout_reset(&io->io_hdr.delay_callout,
12098                                       lun->delay_info.datamove_delay * hz,
12099                                       ctl_datamove_timer_wakeup, io);
12100                         if (lun->delay_info.datamove_type ==
12101                             CTL_DELAY_TYPE_ONESHOT)
12102                                 lun->delay_info.datamove_delay = 0;
12103                         return;
12104                 }
12105         }
12106 #endif
12107
12108         /*
12109          * This command has been aborted.  Set the port status, so we fail
12110          * the data move.
12111          */
12112         if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12113                 printf("ctl_datamove: tag 0x%04x on (%u:%u:%u) aborted\n",
12114                        io->scsiio.tag_num, io->io_hdr.nexus.initid,
12115                        io->io_hdr.nexus.targ_port,
12116                        io->io_hdr.nexus.targ_lun);
12117                 io->io_hdr.port_status = 31337;
12118                 /*
12119                  * Note that the backend, in this case, will get the
12120                  * callback in its context.  In other cases it may get
12121                  * called in the frontend's interrupt thread context.
12122                  */
12123                 io->scsiio.be_move_done(io);
12124                 return;
12125         }
12126
12127         /* Don't confuse frontend with zero length data move. */
12128         if (io->scsiio.kern_data_len == 0) {
12129                 io->scsiio.be_move_done(io);
12130                 return;
12131         }
12132
12133         /*
12134          * If we're in XFER mode and this I/O is from the other shelf
12135          * controller, we need to send the DMA to the other side to
12136          * actually transfer the data to/from the host.  In serialize only
12137          * mode the transfer happens below CTL and ctl_datamove() is only
12138          * called on the machine that originally received the I/O.
12139          */
12140         if ((control_softc->ha_mode == CTL_HA_MODE_XFER)
12141          && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
12142                 union ctl_ha_msg msg;
12143                 uint32_t sg_entries_sent;
12144                 int do_sg_copy;
12145                 int i;
12146
12147                 memset(&msg, 0, sizeof(msg));
12148                 msg.hdr.msg_type = CTL_MSG_DATAMOVE;
12149                 msg.hdr.original_sc = io->io_hdr.original_sc;
12150                 msg.hdr.serializing_sc = io;
12151                 msg.hdr.nexus = io->io_hdr.nexus;
12152                 msg.dt.flags = io->io_hdr.flags;
12153                 /*
12154                  * We convert everything into a S/G list here.  We can't
12155                  * pass by reference, only by value between controllers.
12156                  * So we can't pass a pointer to the S/G list, only as many
12157                  * S/G entries as we can fit in here.  If it's possible for
12158                  * us to get more than CTL_HA_MAX_SG_ENTRIES S/G entries,
12159                  * then we need to break this up into multiple transfers.
12160                  */
12161                 if (io->scsiio.kern_sg_entries == 0) {
12162                         msg.dt.kern_sg_entries = 1;
12163 #if 0
12164                         /*
12165                          * Convert to a physical address if this is a
12166                          * virtual address.
12167                          */
12168                         if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
12169                                 msg.dt.sg_list[0].addr =
12170                                         io->scsiio.kern_data_ptr;
12171                         } else {
12172                                 /*
12173                                  * XXX KDM use busdma here!
12174                                  */
12175                                 msg.dt.sg_list[0].addr = (void *)
12176                                         vtophys(io->scsiio.kern_data_ptr);
12177                         }
12178 #else
12179                         KASSERT((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0,
12180                             ("HA does not support BUS_ADDR"));
12181                         msg.dt.sg_list[0].addr = io->scsiio.kern_data_ptr;
12182 #endif
12183
12184                         msg.dt.sg_list[0].len = io->scsiio.kern_data_len;
12185                         do_sg_copy = 0;
12186                 } else {
12187                         msg.dt.kern_sg_entries = io->scsiio.kern_sg_entries;
12188                         do_sg_copy = 1;
12189                 }
12190
12191                 msg.dt.kern_data_len = io->scsiio.kern_data_len;
12192                 msg.dt.kern_total_len = io->scsiio.kern_total_len;
12193                 msg.dt.kern_data_resid = io->scsiio.kern_data_resid;
12194                 msg.dt.kern_rel_offset = io->scsiio.kern_rel_offset;
12195                 msg.dt.sg_sequence = 0;
12196
12197                 /*
12198                  * Loop until we've sent all of the S/G entries.  On the
12199                  * other end, we'll recompose these S/G entries into one
12200                  * contiguous list before passing it to the
12201                  */
12202                 for (sg_entries_sent = 0; sg_entries_sent <
12203                      msg.dt.kern_sg_entries; msg.dt.sg_sequence++) {
12204                         msg.dt.cur_sg_entries = MIN((sizeof(msg.dt.sg_list)/
12205                                 sizeof(msg.dt.sg_list[0])),
12206                                 msg.dt.kern_sg_entries - sg_entries_sent);
12207
12208                         if (do_sg_copy != 0) {
12209                                 struct ctl_sg_entry *sgl;
12210                                 int j;
12211
12212                                 sgl = (struct ctl_sg_entry *)
12213                                         io->scsiio.kern_data_ptr;
12214                                 /*
12215                                  * If this is in cached memory, flush the cache
12216                                  * before we send the DMA request to the other
12217                                  * controller.  We want to do this in either
12218                                  * the * read or the write case.  The read
12219                                  * case is straightforward.  In the write
12220                                  * case, we want to make sure nothing is
12221                                  * in the local cache that could overwrite
12222                                  * the DMAed data.
12223                                  */
12224
12225                                 for (i = sg_entries_sent, j = 0;
12226                                      i < msg.dt.cur_sg_entries; i++, j++) {
12227 #if 0
12228                                         if ((io->io_hdr.flags &
12229                                              CTL_FLAG_BUS_ADDR) == 0) {
12230                                                 /*
12231                                                  * XXX KDM use busdma.
12232                                                  */
12233                                                 msg.dt.sg_list[j].addr =(void *)
12234                                                        vtophys(sgl[i].addr);
12235                                         } else {
12236                                                 msg.dt.sg_list[j].addr =
12237                                                         sgl[i].addr;
12238                                         }
12239 #else
12240                                         KASSERT((io->io_hdr.flags &
12241                                             CTL_FLAG_BUS_ADDR) == 0,
12242                                             ("HA does not support BUS_ADDR"));
12243                                         msg.dt.sg_list[j].addr = sgl[i].addr;
12244 #endif
12245                                         msg.dt.sg_list[j].len = sgl[i].len;
12246                                 }
12247                         }
12248
12249                         sg_entries_sent += msg.dt.cur_sg_entries;
12250                         if (sg_entries_sent >= msg.dt.kern_sg_entries)
12251                                 msg.dt.sg_last = 1;
12252                         else
12253                                 msg.dt.sg_last = 0;
12254
12255                         if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12256                             sizeof(msg.dt) - sizeof(msg.dt.sg_list) +
12257                             sizeof(struct ctl_sg_entry)*msg.dt.cur_sg_entries,
12258                             M_WAITOK) > CTL_HA_STATUS_SUCCESS) {
12259                                 io->io_hdr.port_status = 31341;
12260                                 io->scsiio.be_move_done(io);
12261                                 return;
12262                         }
12263
12264                         msg.dt.sent_sg_entries = sg_entries_sent;
12265                 }
12266                 io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12267         } else {
12268
12269                 /*
12270                  * Lookup the fe_datamove() function for this particular
12271                  * front end.
12272                  */
12273                 fe_datamove = ctl_io_port(&io->io_hdr)->fe_datamove;
12274
12275                 fe_datamove(io);
12276         }
12277 }
12278
12279 static void
12280 ctl_send_datamove_done(union ctl_io *io, int have_lock)
12281 {
12282         union ctl_ha_msg msg;
12283
12284         memset(&msg, 0, sizeof(msg));
12285
12286         msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
12287         msg.hdr.original_sc = io;
12288         msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
12289         msg.hdr.nexus = io->io_hdr.nexus;
12290         msg.hdr.status = io->io_hdr.status;
12291         msg.scsi.tag_num = io->scsiio.tag_num;
12292         msg.scsi.tag_type = io->scsiio.tag_type;
12293         msg.scsi.scsi_status = io->scsiio.scsi_status;
12294         memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
12295                io->scsiio.sense_len);
12296         msg.scsi.sense_len = io->scsiio.sense_len;
12297         msg.scsi.sense_residual = io->scsiio.sense_residual;
12298         msg.scsi.fetd_status = io->io_hdr.port_status;
12299         msg.scsi.residual = io->scsiio.residual;
12300         io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12301
12302         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12303                 ctl_failover_io(io, /*have_lock*/ have_lock);
12304                 return;
12305         }
12306
12307         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12308             sizeof(msg.scsi) - sizeof(msg.scsi.sense_data) +
12309             msg.scsi.sense_len, M_WAITOK);
12310 }
12311
12312 /*
12313  * The DMA to the remote side is done, now we need to tell the other side
12314  * we're done so it can continue with its data movement.
12315  */
12316 static void
12317 ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
12318 {
12319         union ctl_io *io;
12320         int i;
12321
12322         io = rq->context;
12323
12324         if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12325                 printf("%s: ISC DMA write failed with error %d", __func__,
12326                        rq->ret);
12327                 ctl_set_internal_failure(&io->scsiio,
12328                                          /*sks_valid*/ 1,
12329                                          /*retry_count*/ rq->ret);
12330         }
12331
12332         ctl_dt_req_free(rq);
12333
12334         for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12335                 free(io->io_hdr.local_sglist[i].addr, M_CTL);
12336         free(io->io_hdr.remote_sglist, M_CTL);
12337         io->io_hdr.remote_sglist = NULL;
12338         io->io_hdr.local_sglist = NULL;
12339
12340         /*
12341          * The data is in local and remote memory, so now we need to send
12342          * status (good or back) back to the other side.
12343          */
12344         ctl_send_datamove_done(io, /*have_lock*/ 0);
12345 }
12346
12347 /*
12348  * We've moved the data from the host/controller into local memory.  Now we
12349  * need to push it over to the remote controller's memory.
12350  */
12351 static int
12352 ctl_datamove_remote_dm_write_cb(union ctl_io *io)
12353 {
12354         int retval;
12355
12356         retval = 0;
12357
12358         retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
12359                                           ctl_datamove_remote_write_cb);
12360
12361         return (retval);
12362 }
12363
12364 static void
12365 ctl_datamove_remote_write(union ctl_io *io)
12366 {
12367         int retval;
12368         void (*fe_datamove)(union ctl_io *io);
12369
12370         /*
12371          * - Get the data from the host/HBA into local memory.
12372          * - DMA memory from the local controller to the remote controller.
12373          * - Send status back to the remote controller.
12374          */
12375
12376         retval = ctl_datamove_remote_sgl_setup(io);
12377         if (retval != 0)
12378                 return;
12379
12380         /* Switch the pointer over so the FETD knows what to do */
12381         io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12382
12383         /*
12384          * Use a custom move done callback, since we need to send completion
12385          * back to the other controller, not to the backend on this side.
12386          */
12387         io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
12388
12389         fe_datamove = ctl_io_port(&io->io_hdr)->fe_datamove;
12390
12391         fe_datamove(io);
12392
12393         return;
12394
12395 }
12396
12397 static int
12398 ctl_datamove_remote_dm_read_cb(union ctl_io *io)
12399 {
12400 #if 0
12401         char str[256];
12402         char path_str[64];
12403         struct sbuf sb;
12404 #endif
12405         int i;
12406
12407         for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12408                 free(io->io_hdr.local_sglist[i].addr, M_CTL);
12409         free(io->io_hdr.remote_sglist, M_CTL);
12410         io->io_hdr.remote_sglist = NULL;
12411         io->io_hdr.local_sglist = NULL;
12412
12413 #if 0
12414         scsi_path_string(io, path_str, sizeof(path_str));
12415         sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12416         sbuf_cat(&sb, path_str);
12417         scsi_command_string(&io->scsiio, NULL, &sb);
12418         sbuf_printf(&sb, "\n");
12419         sbuf_cat(&sb, path_str);
12420         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12421                     io->scsiio.tag_num, io->scsiio.tag_type);
12422         sbuf_cat(&sb, path_str);
12423         sbuf_printf(&sb, "%s: flags %#x, status %#x\n", __func__,
12424                     io->io_hdr.flags, io->io_hdr.status);
12425         sbuf_finish(&sb);
12426         printk("%s", sbuf_data(&sb));
12427 #endif
12428
12429
12430         /*
12431          * The read is done, now we need to send status (good or bad) back
12432          * to the other side.
12433          */
12434         ctl_send_datamove_done(io, /*have_lock*/ 0);
12435
12436         return (0);
12437 }
12438
12439 static void
12440 ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
12441 {
12442         union ctl_io *io;
12443         void (*fe_datamove)(union ctl_io *io);
12444
12445         io = rq->context;
12446
12447         if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12448                 printf("%s: ISC DMA read failed with error %d\n", __func__,
12449                        rq->ret);
12450                 ctl_set_internal_failure(&io->scsiio,
12451                                          /*sks_valid*/ 1,
12452                                          /*retry_count*/ rq->ret);
12453         }
12454
12455         ctl_dt_req_free(rq);
12456
12457         /* Switch the pointer over so the FETD knows what to do */
12458         io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12459
12460         /*
12461          * Use a custom move done callback, since we need to send completion
12462          * back to the other controller, not to the backend on this side.
12463          */
12464         io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
12465
12466         /* XXX KDM add checks like the ones in ctl_datamove? */
12467
12468         fe_datamove = ctl_io_port(&io->io_hdr)->fe_datamove;
12469
12470         fe_datamove(io);
12471 }
12472
12473 static int
12474 ctl_datamove_remote_sgl_setup(union ctl_io *io)
12475 {
12476         struct ctl_sg_entry *local_sglist, *remote_sglist;
12477         struct ctl_softc *softc;
12478         uint32_t len_to_go;
12479         int retval;
12480         int i;
12481
12482         retval = 0;
12483         softc = control_softc;
12484         local_sglist = io->io_hdr.local_sglist;
12485         remote_sglist = io->io_hdr.remote_sglist;
12486         len_to_go = io->scsiio.kern_data_len;
12487
12488         /*
12489          * The difficult thing here is that the size of the various
12490          * S/G segments may be different than the size from the
12491          * remote controller.  That'll make it harder when DMAing
12492          * the data back to the other side.
12493          */
12494         for (i = 0; len_to_go > 0; i++) {
12495                 local_sglist[i].len = MIN(len_to_go, CTL_HA_DATAMOVE_SEGMENT);
12496                 local_sglist[i].addr =
12497                     malloc(local_sglist[i].len, M_CTL, M_WAITOK);
12498
12499                 len_to_go -= local_sglist[i].len;
12500         }
12501         /*
12502          * Reset the number of S/G entries accordingly.  The original
12503          * number of S/G entries is available in rem_sg_entries.
12504          */
12505         io->scsiio.kern_sg_entries = i;
12506
12507 #if 0
12508         printf("%s: kern_sg_entries = %d\n", __func__,
12509                io->scsiio.kern_sg_entries);
12510         for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12511                 printf("%s: sg[%d] = %p, %d\n", __func__, i,
12512                        local_sglist[i].addr, local_sglist[i].len);
12513 #endif
12514
12515         return (retval);
12516 }
12517
12518 static int
12519 ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
12520                          ctl_ha_dt_cb callback)
12521 {
12522         struct ctl_ha_dt_req *rq;
12523         struct ctl_sg_entry *remote_sglist, *local_sglist;
12524         uint32_t local_used, remote_used, total_used;
12525         int i, j, isc_ret;
12526
12527         rq = ctl_dt_req_alloc();
12528
12529         /*
12530          * If we failed to allocate the request, and if the DMA didn't fail
12531          * anyway, set busy status.  This is just a resource allocation
12532          * failure.
12533          */
12534         if ((rq == NULL)
12535          && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE))
12536                 ctl_set_busy(&io->scsiio);
12537
12538         if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE) {
12539
12540                 if (rq != NULL)
12541                         ctl_dt_req_free(rq);
12542
12543                 /*
12544                  * The data move failed.  We need to return status back
12545                  * to the other controller.  No point in trying to DMA
12546                  * data to the remote controller.
12547                  */
12548
12549                 ctl_send_datamove_done(io, /*have_lock*/ 0);
12550
12551                 return (1);
12552         }
12553
12554         local_sglist = io->io_hdr.local_sglist;
12555         remote_sglist = io->io_hdr.remote_sglist;
12556         local_used = 0;
12557         remote_used = 0;
12558         total_used = 0;
12559
12560         /*
12561          * Pull/push the data over the wire from/to the other controller.
12562          * This takes into account the possibility that the local and
12563          * remote sglists may not be identical in terms of the size of
12564          * the elements and the number of elements.
12565          *
12566          * One fundamental assumption here is that the length allocated for
12567          * both the local and remote sglists is identical.  Otherwise, we've
12568          * essentially got a coding error of some sort.
12569          */
12570         isc_ret = CTL_HA_STATUS_SUCCESS;
12571         for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
12572                 uint32_t cur_len;
12573                 uint8_t *tmp_ptr;
12574
12575                 rq->command = command;
12576                 rq->context = io;
12577
12578                 /*
12579                  * Both pointers should be aligned.  But it is possible
12580                  * that the allocation length is not.  They should both
12581                  * also have enough slack left over at the end, though,
12582                  * to round up to the next 8 byte boundary.
12583                  */
12584                 cur_len = MIN(local_sglist[i].len - local_used,
12585                               remote_sglist[j].len - remote_used);
12586                 rq->size = cur_len;
12587
12588                 tmp_ptr = (uint8_t *)local_sglist[i].addr;
12589                 tmp_ptr += local_used;
12590
12591 #if 0
12592                 /* Use physical addresses when talking to ISC hardware */
12593                 if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
12594                         /* XXX KDM use busdma */
12595                         rq->local = vtophys(tmp_ptr);
12596                 } else
12597                         rq->local = tmp_ptr;
12598 #else
12599                 KASSERT((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0,
12600                     ("HA does not support BUS_ADDR"));
12601                 rq->local = tmp_ptr;
12602 #endif
12603
12604                 tmp_ptr = (uint8_t *)remote_sglist[j].addr;
12605                 tmp_ptr += remote_used;
12606                 rq->remote = tmp_ptr;
12607
12608                 rq->callback = NULL;
12609
12610                 local_used += cur_len;
12611                 if (local_used >= local_sglist[i].len) {
12612                         i++;
12613                         local_used = 0;
12614                 }
12615
12616                 remote_used += cur_len;
12617                 if (remote_used >= remote_sglist[j].len) {
12618                         j++;
12619                         remote_used = 0;
12620                 }
12621                 total_used += cur_len;
12622
12623                 if (total_used >= io->scsiio.kern_data_len)
12624                         rq->callback = callback;
12625
12626 #if 0
12627                 printf("%s: %s: local %#x remote %#x size %d\n", __func__,
12628                        (command == CTL_HA_DT_CMD_WRITE) ? "WRITE" : "READ",
12629                        rq->local, rq->remote, rq->size);
12630 #endif
12631
12632                 isc_ret = ctl_dt_single(rq);
12633                 if (isc_ret > CTL_HA_STATUS_SUCCESS)
12634                         break;
12635         }
12636         if (isc_ret != CTL_HA_STATUS_WAIT) {
12637                 rq->ret = isc_ret;
12638                 callback(rq);
12639         }
12640
12641         return (0);
12642 }
12643
12644 static void
12645 ctl_datamove_remote_read(union ctl_io *io)
12646 {
12647         int retval;
12648         int i;
12649
12650         /*
12651          * This will send an error to the other controller in the case of a
12652          * failure.
12653          */
12654         retval = ctl_datamove_remote_sgl_setup(io);
12655         if (retval != 0)
12656                 return;
12657
12658         retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
12659                                           ctl_datamove_remote_read_cb);
12660         if (retval != 0) {
12661                 /*
12662                  * Make sure we free memory if there was an error..  The
12663                  * ctl_datamove_remote_xfer() function will send the
12664                  * datamove done message, or call the callback with an
12665                  * error if there is a problem.
12666                  */
12667                 for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12668                         free(io->io_hdr.local_sglist[i].addr, M_CTL);
12669                 free(io->io_hdr.remote_sglist, M_CTL);
12670                 io->io_hdr.remote_sglist = NULL;
12671                 io->io_hdr.local_sglist = NULL;
12672         }
12673
12674         return;
12675 }
12676
12677 /*
12678  * Process a datamove request from the other controller.  This is used for
12679  * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
12680  * first.  Once that is complete, the data gets DMAed into the remote
12681  * controller's memory.  For reads, we DMA from the remote controller's
12682  * memory into our memory first, and then move it out to the FETD.
12683  */
12684 static void
12685 ctl_datamove_remote(union ctl_io *io)
12686 {
12687
12688         mtx_assert(&control_softc->ctl_lock, MA_NOTOWNED);
12689
12690         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12691                 ctl_failover_io(io, /*have_lock*/ 0);
12692                 return;
12693         }
12694
12695         /*
12696          * Note that we look for an aborted I/O here, but don't do some of
12697          * the other checks that ctl_datamove() normally does.
12698          * We don't need to run the datamove delay code, since that should
12699          * have been done if need be on the other controller.
12700          */
12701         if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12702                 printf("%s: tag 0x%04x on (%u:%u:%u) aborted\n", __func__,
12703                        io->scsiio.tag_num, io->io_hdr.nexus.initid,
12704                        io->io_hdr.nexus.targ_port,
12705                        io->io_hdr.nexus.targ_lun);
12706                 io->io_hdr.port_status = 31338;
12707                 ctl_send_datamove_done(io, /*have_lock*/ 0);
12708                 return;
12709         }
12710
12711         if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT)
12712                 ctl_datamove_remote_write(io);
12713         else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN)
12714                 ctl_datamove_remote_read(io);
12715         else {
12716                 io->io_hdr.port_status = 31339;
12717                 ctl_send_datamove_done(io, /*have_lock*/ 0);
12718         }
12719 }
12720
12721 static int
12722 ctl_process_done(union ctl_io *io)
12723 {
12724         struct ctl_lun *lun;
12725         struct ctl_softc *softc = control_softc;
12726         void (*fe_done)(union ctl_io *io);
12727         union ctl_ha_msg msg;
12728         uint32_t targ_port = io->io_hdr.nexus.targ_port;
12729
12730         CTL_DEBUG_PRINT(("ctl_process_done\n"));
12731
12732         if ((io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) == 0)
12733                 fe_done = softc->ctl_ports[targ_port]->fe_done;
12734         else
12735                 fe_done = NULL;
12736
12737 #ifdef CTL_TIME_IO
12738         if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12739                 char str[256];
12740                 char path_str[64];
12741                 struct sbuf sb;
12742
12743                 ctl_scsi_path_string(io, path_str, sizeof(path_str));
12744                 sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12745
12746                 sbuf_cat(&sb, path_str);
12747                 switch (io->io_hdr.io_type) {
12748                 case CTL_IO_SCSI:
12749                         ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12750                         sbuf_printf(&sb, "\n");
12751                         sbuf_cat(&sb, path_str);
12752                         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12753                                     io->scsiio.tag_num, io->scsiio.tag_type);
12754                         break;
12755                 case CTL_IO_TASK:
12756                         sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12757                                     "Tag Type: %d\n", io->taskio.task_action,
12758                                     io->taskio.tag_num, io->taskio.tag_type);
12759                         break;
12760                 default:
12761                         printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12762                         panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type);
12763                         break;
12764                 }
12765                 sbuf_cat(&sb, path_str);
12766                 sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
12767                             (intmax_t)time_uptime - io->io_hdr.start_time);
12768                 sbuf_finish(&sb);
12769                 printf("%s", sbuf_data(&sb));
12770         }
12771 #endif /* CTL_TIME_IO */
12772
12773         switch (io->io_hdr.io_type) {
12774         case CTL_IO_SCSI:
12775                 break;
12776         case CTL_IO_TASK:
12777                 if (ctl_debug & CTL_DEBUG_INFO)
12778                         ctl_io_error_print(io, NULL);
12779                 if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)
12780                         ctl_free_io(io);
12781                 else
12782                         fe_done(io);
12783                 return (CTL_RETVAL_COMPLETE);
12784         default:
12785                 panic("ctl_process_done: invalid io type %d\n",
12786                       io->io_hdr.io_type);
12787                 break; /* NOTREACHED */
12788         }
12789
12790         lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12791         if (lun == NULL) {
12792                 CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
12793                                  io->io_hdr.nexus.targ_mapped_lun));
12794                 goto bailout;
12795         }
12796
12797         mtx_lock(&lun->lun_lock);
12798
12799         /*
12800          * Check to see if we have any errors to inject here.  We only
12801          * inject errors for commands that don't already have errors set.
12802          */
12803         if ((STAILQ_FIRST(&lun->error_list) != NULL) &&
12804             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS) &&
12805             ((io->io_hdr.flags & CTL_FLAG_STATUS_SENT) == 0))
12806                 ctl_inject_error(lun, io);
12807
12808         /*
12809          * XXX KDM how do we treat commands that aren't completed
12810          * successfully?
12811          *
12812          * XXX KDM should we also track I/O latency?
12813          */
12814         if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS &&
12815             io->io_hdr.io_type == CTL_IO_SCSI) {
12816 #ifdef CTL_TIME_IO
12817                 struct bintime cur_bt;
12818 #endif
12819                 int type;
12820
12821                 if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
12822                     CTL_FLAG_DATA_IN)
12823                         type = CTL_STATS_READ;
12824                 else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
12825                     CTL_FLAG_DATA_OUT)
12826                         type = CTL_STATS_WRITE;
12827                 else
12828                         type = CTL_STATS_NO_IO;
12829
12830                 lun->stats.ports[targ_port].bytes[type] +=
12831                     io->scsiio.kern_total_len;
12832                 lun->stats.ports[targ_port].operations[type]++;
12833 #ifdef CTL_TIME_IO
12834                 bintime_add(&lun->stats.ports[targ_port].dma_time[type],
12835                    &io->io_hdr.dma_bt);
12836                 lun->stats.ports[targ_port].num_dmas[type] +=
12837                     io->io_hdr.num_dmas;
12838                 getbintime(&cur_bt);
12839                 bintime_sub(&cur_bt, &io->io_hdr.start_bt);
12840                 bintime_add(&lun->stats.ports[targ_port].time[type], &cur_bt);
12841 #endif
12842         }
12843
12844         /*
12845          * Remove this from the OOA queue.
12846          */
12847         TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
12848 #ifdef CTL_TIME_IO
12849         if (TAILQ_EMPTY(&lun->ooa_queue))
12850                 lun->last_busy = getsbinuptime();
12851 #endif
12852
12853         /*
12854          * Run through the blocked queue on this LUN and see if anything
12855          * has become unblocked, now that this transaction is done.
12856          */
12857         ctl_check_blocked(lun);
12858
12859         /*
12860          * If the LUN has been invalidated, free it if there is nothing
12861          * left on its OOA queue.
12862          */
12863         if ((lun->flags & CTL_LUN_INVALID)
12864          && TAILQ_EMPTY(&lun->ooa_queue)) {
12865                 mtx_unlock(&lun->lun_lock);
12866                 mtx_lock(&softc->ctl_lock);
12867                 ctl_free_lun(lun);
12868                 mtx_unlock(&softc->ctl_lock);
12869         } else
12870                 mtx_unlock(&lun->lun_lock);
12871
12872 bailout:
12873
12874         /*
12875          * If this command has been aborted, make sure we set the status
12876          * properly.  The FETD is responsible for freeing the I/O and doing
12877          * whatever it needs to do to clean up its state.
12878          */
12879         if (io->io_hdr.flags & CTL_FLAG_ABORT)
12880                 ctl_set_task_aborted(&io->scsiio);
12881
12882         /*
12883          * If enabled, print command error status.
12884          */
12885         if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS &&
12886             (ctl_debug & CTL_DEBUG_INFO) != 0)
12887                 ctl_io_error_print(io, NULL);
12888
12889         /*
12890          * Tell the FETD or the other shelf controller we're done with this
12891          * command.  Note that only SCSI commands get to this point.  Task
12892          * management commands are completed above.
12893          */
12894         if ((softc->ha_mode != CTL_HA_MODE_XFER) &&
12895             (io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)) {
12896                 memset(&msg, 0, sizeof(msg));
12897                 msg.hdr.msg_type = CTL_MSG_FINISH_IO;
12898                 msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
12899                 msg.hdr.nexus = io->io_hdr.nexus;
12900                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12901                     sizeof(msg.scsi) - sizeof(msg.scsi.sense_data),
12902                     M_WAITOK);
12903         }
12904         if ((softc->ha_mode == CTL_HA_MODE_XFER)
12905          && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
12906                 memset(&msg, 0, sizeof(msg));
12907                 msg.hdr.msg_type = CTL_MSG_FINISH_IO;
12908                 msg.hdr.original_sc = io->io_hdr.original_sc;
12909                 msg.hdr.nexus = io->io_hdr.nexus;
12910                 msg.hdr.status = io->io_hdr.status;
12911                 msg.scsi.scsi_status = io->scsiio.scsi_status;
12912                 msg.scsi.tag_num = io->scsiio.tag_num;
12913                 msg.scsi.tag_type = io->scsiio.tag_type;
12914                 msg.scsi.sense_len = io->scsiio.sense_len;
12915                 msg.scsi.sense_residual = io->scsiio.sense_residual;
12916                 msg.scsi.residual = io->scsiio.residual;
12917                 memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
12918                        io->scsiio.sense_len);
12919                 /*
12920                  * We copy this whether or not this is an I/O-related
12921                  * command.  Otherwise, we'd have to go and check to see
12922                  * whether it's a read/write command, and it really isn't
12923                  * worth it.
12924                  */
12925                 memcpy(&msg.scsi.lbalen,
12926                        &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes,
12927                        sizeof(msg.scsi.lbalen));
12928
12929                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12930                     sizeof(msg.scsi) - sizeof(msg.scsi.sense_data) +
12931                     msg.scsi.sense_len, M_WAITOK);
12932                 ctl_free_io(io);
12933         } else
12934                 fe_done(io);
12935
12936         return (CTL_RETVAL_COMPLETE);
12937 }
12938
12939 #ifdef CTL_WITH_CA
12940 /*
12941  * Front end should call this if it doesn't do autosense.  When the request
12942  * sense comes back in from the initiator, we'll dequeue this and send it.
12943  */
12944 int
12945 ctl_queue_sense(union ctl_io *io)
12946 {
12947         struct ctl_lun *lun;
12948         struct ctl_port *port;
12949         struct ctl_softc *softc;
12950         uint32_t initidx, targ_lun;
12951
12952         softc = control_softc;
12953
12954         CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
12955
12956         /*
12957          * LUN lookup will likely move to the ctl_work_thread() once we
12958          * have our new queueing infrastructure (that doesn't put things on
12959          * a per-LUN queue initially).  That is so that we can handle
12960          * things like an INQUIRY to a LUN that we don't have enabled.  We
12961          * can't deal with that right now.
12962          */
12963         mtx_lock(&softc->ctl_lock);
12964
12965         /*
12966          * If we don't have a LUN for this, just toss the sense
12967          * information.
12968          */
12969         port = ctl_io_port(&ctsio->io_hdr);
12970         targ_lun = ctl_lun_map_from_port(port, io->io_hdr.nexus.targ_lun);
12971         if ((targ_lun < CTL_MAX_LUNS)
12972          && (softc->ctl_luns[targ_lun] != NULL))
12973                 lun = softc->ctl_luns[targ_lun];
12974         else
12975                 goto bailout;
12976
12977         initidx = ctl_get_initindex(&io->io_hdr.nexus);
12978
12979         mtx_lock(&lun->lun_lock);
12980         /*
12981          * Already have CA set for this LUN...toss the sense information.
12982          */
12983         if (ctl_is_set(lun->have_ca, initidx)) {
12984                 mtx_unlock(&lun->lun_lock);
12985                 goto bailout;
12986         }
12987
12988         memcpy(&lun->pending_sense[initidx], &io->scsiio.sense_data,
12989                MIN(sizeof(lun->pending_sense[initidx]),
12990                sizeof(io->scsiio.sense_data)));
12991         ctl_set_mask(lun->have_ca, initidx);
12992         mtx_unlock(&lun->lun_lock);
12993
12994 bailout:
12995         mtx_unlock(&softc->ctl_lock);
12996
12997         ctl_free_io(io);
12998
12999         return (CTL_RETVAL_COMPLETE);
13000 }
13001 #endif
13002
13003 /*
13004  * Primary command inlet from frontend ports.  All SCSI and task I/O
13005  * requests must go through this function.
13006  */
13007 int
13008 ctl_queue(union ctl_io *io)
13009 {
13010         struct ctl_port *port;
13011
13012         CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
13013
13014 #ifdef CTL_TIME_IO
13015         io->io_hdr.start_time = time_uptime;
13016         getbintime(&io->io_hdr.start_bt);
13017 #endif /* CTL_TIME_IO */
13018
13019         /* Map FE-specific LUN ID into global one. */
13020         port = ctl_io_port(&io->io_hdr);
13021         io->io_hdr.nexus.targ_mapped_lun =
13022             ctl_lun_map_from_port(port, io->io_hdr.nexus.targ_lun);
13023
13024         switch (io->io_hdr.io_type) {
13025         case CTL_IO_SCSI:
13026         case CTL_IO_TASK:
13027                 if (ctl_debug & CTL_DEBUG_CDB)
13028                         ctl_io_print(io);
13029                 ctl_enqueue_incoming(io);
13030                 break;
13031         default:
13032                 printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
13033                 return (EINVAL);
13034         }
13035
13036         return (CTL_RETVAL_COMPLETE);
13037 }
13038
13039 #ifdef CTL_IO_DELAY
13040 static void
13041 ctl_done_timer_wakeup(void *arg)
13042 {
13043         union ctl_io *io;
13044
13045         io = (union ctl_io *)arg;
13046         ctl_done(io);
13047 }
13048 #endif /* CTL_IO_DELAY */
13049
13050 void
13051 ctl_done(union ctl_io *io)
13052 {
13053
13054         /*
13055          * Enable this to catch duplicate completion issues.
13056          */
13057 #if 0
13058         if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
13059                 printf("%s: type %d msg %d cdb %x iptl: "
13060                        "%u:%u:%u tag 0x%04x "
13061                        "flag %#x status %x\n",
13062                         __func__,
13063                         io->io_hdr.io_type,
13064                         io->io_hdr.msg_type,
13065                         io->scsiio.cdb[0],
13066                         io->io_hdr.nexus.initid,
13067                         io->io_hdr.nexus.targ_port,
13068                         io->io_hdr.nexus.targ_lun,
13069                         (io->io_hdr.io_type ==
13070                         CTL_IO_TASK) ?
13071                         io->taskio.tag_num :
13072                         io->scsiio.tag_num,
13073                         io->io_hdr.flags,
13074                         io->io_hdr.status);
13075         } else
13076                 io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
13077 #endif
13078
13079         /*
13080          * This is an internal copy of an I/O, and should not go through
13081          * the normal done processing logic.
13082          */
13083         if (io->io_hdr.flags & CTL_FLAG_INT_COPY)
13084                 return;
13085
13086 #ifdef CTL_IO_DELAY
13087         if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
13088                 struct ctl_lun *lun;
13089
13090                 lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13091
13092                 io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
13093         } else {
13094                 struct ctl_lun *lun;
13095
13096                 lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13097
13098                 if ((lun != NULL)
13099                  && (lun->delay_info.done_delay > 0)) {
13100
13101                         callout_init(&io->io_hdr.delay_callout, /*mpsafe*/ 1);
13102                         io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
13103                         callout_reset(&io->io_hdr.delay_callout,
13104                                       lun->delay_info.done_delay * hz,
13105                                       ctl_done_timer_wakeup, io);
13106                         if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
13107                                 lun->delay_info.done_delay = 0;
13108                         return;
13109                 }
13110         }
13111 #endif /* CTL_IO_DELAY */
13112
13113         ctl_enqueue_done(io);
13114 }
13115
13116 static void
13117 ctl_work_thread(void *arg)
13118 {
13119         struct ctl_thread *thr = (struct ctl_thread *)arg;
13120         struct ctl_softc *softc = thr->ctl_softc;
13121         union ctl_io *io;
13122         int retval;
13123
13124         CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
13125
13126         for (;;) {
13127                 retval = 0;
13128
13129                 /*
13130                  * We handle the queues in this order:
13131                  * - ISC
13132                  * - done queue (to free up resources, unblock other commands)
13133                  * - RtR queue
13134                  * - incoming queue
13135                  *
13136                  * If those queues are empty, we break out of the loop and
13137                  * go to sleep.
13138                  */
13139                 mtx_lock(&thr->queue_lock);
13140                 io = (union ctl_io *)STAILQ_FIRST(&thr->isc_queue);
13141                 if (io != NULL) {
13142                         STAILQ_REMOVE_HEAD(&thr->isc_queue, links);
13143                         mtx_unlock(&thr->queue_lock);
13144                         ctl_handle_isc(io);
13145                         continue;
13146                 }
13147                 io = (union ctl_io *)STAILQ_FIRST(&thr->done_queue);
13148                 if (io != NULL) {
13149                         STAILQ_REMOVE_HEAD(&thr->done_queue, links);
13150                         /* clear any blocked commands, call fe_done */
13151                         mtx_unlock(&thr->queue_lock);
13152                         retval = ctl_process_done(io);
13153                         continue;
13154                 }
13155                 io = (union ctl_io *)STAILQ_FIRST(&thr->incoming_queue);
13156                 if (io != NULL) {
13157                         STAILQ_REMOVE_HEAD(&thr->incoming_queue, links);
13158                         mtx_unlock(&thr->queue_lock);
13159                         if (io->io_hdr.io_type == CTL_IO_TASK)
13160                                 ctl_run_task(io);
13161                         else
13162                                 ctl_scsiio_precheck(softc, &io->scsiio);
13163                         continue;
13164                 }
13165                 io = (union ctl_io *)STAILQ_FIRST(&thr->rtr_queue);
13166                 if (io != NULL) {
13167                         STAILQ_REMOVE_HEAD(&thr->rtr_queue, links);
13168                         mtx_unlock(&thr->queue_lock);
13169                         retval = ctl_scsiio(&io->scsiio);
13170                         if (retval != CTL_RETVAL_COMPLETE)
13171                                 CTL_DEBUG_PRINT(("ctl_scsiio failed\n"));
13172                         continue;
13173                 }
13174
13175                 /* Sleep until we have something to do. */
13176                 mtx_sleep(thr, &thr->queue_lock, PDROP | PRIBIO, "-", 0);
13177         }
13178 }
13179
13180 static void
13181 ctl_lun_thread(void *arg)
13182 {
13183         struct ctl_softc *softc = (struct ctl_softc *)arg;
13184         struct ctl_be_lun *be_lun;
13185         int retval;
13186
13187         CTL_DEBUG_PRINT(("ctl_lun_thread starting\n"));
13188
13189         for (;;) {
13190                 retval = 0;
13191                 mtx_lock(&softc->ctl_lock);
13192                 be_lun = STAILQ_FIRST(&softc->pending_lun_queue);
13193                 if (be_lun != NULL) {
13194                         STAILQ_REMOVE_HEAD(&softc->pending_lun_queue, links);
13195                         mtx_unlock(&softc->ctl_lock);
13196                         ctl_create_lun(be_lun);
13197                         continue;
13198                 }
13199
13200                 /* Sleep until we have something to do. */
13201                 mtx_sleep(&softc->pending_lun_queue, &softc->ctl_lock,
13202                     PDROP | PRIBIO, "-", 0);
13203         }
13204 }
13205
13206 static void
13207 ctl_thresh_thread(void *arg)
13208 {
13209         struct ctl_softc *softc = (struct ctl_softc *)arg;
13210         struct ctl_lun *lun;
13211         struct ctl_be_lun *be_lun;
13212         struct scsi_da_rw_recovery_page *rwpage;
13213         struct ctl_logical_block_provisioning_page *page;
13214         const char *attr;
13215         union ctl_ha_msg msg;
13216         uint64_t thres, val;
13217         int i, e, set;
13218
13219         CTL_DEBUG_PRINT(("ctl_thresh_thread starting\n"));
13220
13221         for (;;) {
13222                 mtx_lock(&softc->ctl_lock);
13223                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
13224                         be_lun = lun->be_lun;
13225                         if ((lun->flags & CTL_LUN_DISABLED) ||
13226                             (lun->flags & CTL_LUN_OFFLINE) ||
13227                             lun->backend->lun_attr == NULL)
13228                                 continue;
13229                         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
13230                             softc->ha_mode == CTL_HA_MODE_XFER)
13231                                 continue;
13232                         rwpage = &lun->mode_pages.rw_er_page[CTL_PAGE_CURRENT];
13233                         if ((rwpage->byte8 & SMS_RWER_LBPERE) == 0)
13234                                 continue;
13235                         e = 0;
13236                         page = &lun->mode_pages.lbp_page[CTL_PAGE_CURRENT];
13237                         for (i = 0; i < CTL_NUM_LBP_THRESH; i++) {
13238                                 if ((page->descr[i].flags & SLBPPD_ENABLED) == 0)
13239                                         continue;
13240                                 thres = scsi_4btoul(page->descr[i].count);
13241                                 thres <<= CTL_LBP_EXPONENT;
13242                                 switch (page->descr[i].resource) {
13243                                 case 0x01:
13244                                         attr = "blocksavail";
13245                                         break;
13246                                 case 0x02:
13247                                         attr = "blocksused";
13248                                         break;
13249                                 case 0xf1:
13250                                         attr = "poolblocksavail";
13251                                         break;
13252                                 case 0xf2:
13253                                         attr = "poolblocksused";
13254                                         break;
13255                                 default:
13256                                         continue;
13257                                 }
13258                                 mtx_unlock(&softc->ctl_lock); // XXX
13259                                 val = lun->backend->lun_attr(
13260                                     lun->be_lun->be_lun, attr);
13261                                 mtx_lock(&softc->ctl_lock);
13262                                 if (val == UINT64_MAX)
13263                                         continue;
13264                                 if ((page->descr[i].flags & SLBPPD_ARMING_MASK)
13265                                     == SLBPPD_ARMING_INC)
13266                                         e |= (val >= thres);
13267                                 else
13268                                         e |= (val <= thres);
13269                         }
13270                         mtx_lock(&lun->lun_lock);
13271                         if (e) {
13272                                 if (lun->lasttpt == 0 ||
13273                                     time_uptime - lun->lasttpt >= CTL_LBP_UA_PERIOD) {
13274                                         lun->lasttpt = time_uptime;
13275                                         ctl_est_ua_all(lun, -1, CTL_UA_THIN_PROV_THRES);
13276                                         set = 1;
13277                                 } else
13278                                         set = 0;
13279                         } else {
13280                                 lun->lasttpt = 0;
13281                                 ctl_clr_ua_all(lun, -1, CTL_UA_THIN_PROV_THRES);
13282                                 set = -1;
13283                         }
13284                         mtx_unlock(&lun->lun_lock);
13285                         if (set != 0 &&
13286                             lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
13287                                 /* Send msg to other side. */
13288                                 bzero(&msg.ua, sizeof(msg.ua));
13289                                 msg.hdr.msg_type = CTL_MSG_UA;
13290                                 msg.hdr.nexus.initid = -1;
13291                                 msg.hdr.nexus.targ_port = -1;
13292                                 msg.hdr.nexus.targ_lun = lun->lun;
13293                                 msg.hdr.nexus.targ_mapped_lun = lun->lun;
13294                                 msg.ua.ua_all = 1;
13295                                 msg.ua.ua_set = (set > 0);
13296                                 msg.ua.ua_type = CTL_UA_THIN_PROV_THRES;
13297                                 mtx_unlock(&softc->ctl_lock); // XXX
13298                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13299                                     sizeof(msg.ua), M_WAITOK);
13300                                 mtx_lock(&softc->ctl_lock);
13301                         }
13302                 }
13303                 mtx_unlock(&softc->ctl_lock);
13304                 pause("-", CTL_LBP_PERIOD * hz);
13305         }
13306 }
13307
13308 static void
13309 ctl_enqueue_incoming(union ctl_io *io)
13310 {
13311         struct ctl_softc *softc = control_softc;
13312         struct ctl_thread *thr;
13313         u_int idx;
13314
13315         idx = (io->io_hdr.nexus.targ_port * 127 +
13316                io->io_hdr.nexus.initid) % worker_threads;
13317         thr = &softc->threads[idx];
13318         mtx_lock(&thr->queue_lock);
13319         STAILQ_INSERT_TAIL(&thr->incoming_queue, &io->io_hdr, links);
13320         mtx_unlock(&thr->queue_lock);
13321         wakeup(thr);
13322 }
13323
13324 static void
13325 ctl_enqueue_rtr(union ctl_io *io)
13326 {
13327         struct ctl_softc *softc = control_softc;
13328         struct ctl_thread *thr;
13329
13330         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13331         mtx_lock(&thr->queue_lock);
13332         STAILQ_INSERT_TAIL(&thr->rtr_queue, &io->io_hdr, links);
13333         mtx_unlock(&thr->queue_lock);
13334         wakeup(thr);
13335 }
13336
13337 static void
13338 ctl_enqueue_done(union ctl_io *io)
13339 {
13340         struct ctl_softc *softc = control_softc;
13341         struct ctl_thread *thr;
13342
13343         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13344         mtx_lock(&thr->queue_lock);
13345         STAILQ_INSERT_TAIL(&thr->done_queue, &io->io_hdr, links);
13346         mtx_unlock(&thr->queue_lock);
13347         wakeup(thr);
13348 }
13349
13350 static void
13351 ctl_enqueue_isc(union ctl_io *io)
13352 {
13353         struct ctl_softc *softc = control_softc;
13354         struct ctl_thread *thr;
13355
13356         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13357         mtx_lock(&thr->queue_lock);
13358         STAILQ_INSERT_TAIL(&thr->isc_queue, &io->io_hdr, links);
13359         mtx_unlock(&thr->queue_lock);
13360         wakeup(thr);
13361 }
13362
13363 /*
13364  *  vim: ts=8
13365  */