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